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                 dprintk("%s: client (clientid %08x/%08x) already expired\n",
1026                         __func__,
1027                         clp->cl_clientid.cl_boot,
1028                         clp->cl_clientid.cl_id);
1029                 return;
1030         }
1031
1032         dprintk("renewing client (clientid %08x/%08x)\n",
1033                         clp->cl_clientid.cl_boot,
1034                         clp->cl_clientid.cl_id);
1035         list_move_tail(&clp->cl_lru, &client_lru);
1036         clp->cl_time = get_seconds();
1037 }
1038
1039 static inline void
1040 renew_client(struct nfs4_client *clp)
1041 {
1042         spin_lock(&client_lock);
1043         renew_client_locked(clp);
1044         spin_unlock(&client_lock);
1045 }
1046
1047 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1048 static int
1049 STALE_CLIENTID(clientid_t *clid)
1050 {
1051         if (clid->cl_boot == boot_time)
1052                 return 0;
1053         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1054                 clid->cl_boot, clid->cl_id, boot_time);
1055         return 1;
1056 }
1057
1058 /*
1059  * XXX Should we use a slab cache ?
1060  * This type of memory management is somewhat inefficient, but we use it
1061  * anyway since SETCLIENTID is not a common operation.
1062  */
1063 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1064 {
1065         struct nfs4_client *clp;
1066
1067         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1068         if (clp == NULL)
1069                 return NULL;
1070         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1071         if (clp->cl_name.data == NULL) {
1072                 kfree(clp);
1073                 return NULL;
1074         }
1075         clp->cl_name.len = name.len;
1076         return clp;
1077 }
1078
1079 static inline void
1080 free_client(struct nfs4_client *clp)
1081 {
1082         BUG_ON(!spin_is_locked(&client_lock));
1083         while (!list_empty(&clp->cl_sessions)) {
1084                 struct nfsd4_session *ses;
1085                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1086                                 se_perclnt);
1087                 list_del(&ses->se_perclnt);
1088                 nfsd4_put_session_locked(ses);
1089         }
1090         free_svc_cred(&clp->cl_cred);
1091         kfree(clp->cl_name.data);
1092         kfree(clp);
1093 }
1094
1095 void
1096 release_session_client(struct nfsd4_session *session)
1097 {
1098         struct nfs4_client *clp = session->se_client;
1099
1100         if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
1101                 return;
1102         if (is_client_expired(clp)) {
1103                 free_client(clp);
1104                 session->se_client = NULL;
1105         } else
1106                 renew_client_locked(clp);
1107         spin_unlock(&client_lock);
1108 }
1109
1110 /* must be called under the client_lock */
1111 static inline void
1112 unhash_client_locked(struct nfs4_client *clp)
1113 {
1114         struct nfsd4_session *ses;
1115
1116         mark_client_expired(clp);
1117         list_del(&clp->cl_lru);
1118         spin_lock(&clp->cl_lock);
1119         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1120                 list_del_init(&ses->se_hash);
1121         spin_unlock(&clp->cl_lock);
1122 }
1123
1124 static void
1125 expire_client(struct nfs4_client *clp)
1126 {
1127         struct nfs4_openowner *oo;
1128         struct nfs4_delegation *dp;
1129         struct list_head reaplist;
1130
1131         INIT_LIST_HEAD(&reaplist);
1132         spin_lock(&recall_lock);
1133         while (!list_empty(&clp->cl_delegations)) {
1134                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1135                 list_del_init(&dp->dl_perclnt);
1136                 list_move(&dp->dl_recall_lru, &reaplist);
1137         }
1138         spin_unlock(&recall_lock);
1139         while (!list_empty(&reaplist)) {
1140                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1141                 unhash_delegation(dp);
1142         }
1143         while (!list_empty(&clp->cl_openowners)) {
1144                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1145                 release_openowner(oo);
1146         }
1147         nfsd4_shutdown_callback(clp);
1148         if (clp->cl_cb_conn.cb_xprt)
1149                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1150         list_del(&clp->cl_idhash);
1151         list_del(&clp->cl_strhash);
1152         spin_lock(&client_lock);
1153         unhash_client_locked(clp);
1154         if (atomic_read(&clp->cl_refcount) == 0)
1155                 free_client(clp);
1156         spin_unlock(&client_lock);
1157 }
1158
1159 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1160 {
1161         memcpy(target->cl_verifier.data, source->data,
1162                         sizeof(target->cl_verifier.data));
1163 }
1164
1165 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1166 {
1167         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1168         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1169 }
1170
1171 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1172 {
1173         if (source->cr_principal) {
1174                 target->cr_principal =
1175                                 kstrdup(source->cr_principal, GFP_KERNEL);
1176                 if (target->cr_principal == NULL)
1177                         return -ENOMEM;
1178         } else
1179                 target->cr_principal = NULL;
1180         target->cr_flavor = source->cr_flavor;
1181         target->cr_uid = source->cr_uid;
1182         target->cr_gid = source->cr_gid;
1183         target->cr_group_info = source->cr_group_info;
1184         get_group_info(target->cr_group_info);
1185         return 0;
1186 }
1187
1188 static int same_name(const char *n1, const char *n2)
1189 {
1190         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1191 }
1192
1193 static int
1194 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1195 {
1196         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1197 }
1198
1199 static int
1200 same_clid(clientid_t *cl1, clientid_t *cl2)
1201 {
1202         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1203 }
1204
1205 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1206 {
1207         int i;
1208
1209         if (g1->ngroups != g2->ngroups)
1210                 return false;
1211         for (i=0; i<g1->ngroups; i++)
1212                 if (GROUP_AT(g1, i) != GROUP_AT(g2, i))
1213                         return false;
1214         return true;
1215 }
1216
1217 static int
1218 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1219 {
1220         if ((cr1->cr_flavor != cr2->cr_flavor)
1221                 || (cr1->cr_uid != cr2->cr_uid)
1222                 || (cr1->cr_gid != cr2->cr_gid)
1223                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1224                 return false;
1225         if (cr1->cr_principal == cr2->cr_principal)
1226                 return true;
1227         if (!cr1->cr_principal || !cr2->cr_principal)
1228                 return false;
1229         return 0 == strcmp(cr1->cr_principal, cr1->cr_principal);
1230 }
1231
1232 static void gen_clid(struct nfs4_client *clp)
1233 {
1234         static u32 current_clientid = 1;
1235
1236         clp->cl_clientid.cl_boot = boot_time;
1237         clp->cl_clientid.cl_id = current_clientid++;
1238 }
1239
1240 static void gen_confirm(struct nfs4_client *clp)
1241 {
1242         __be32 verf[2];
1243         static u32 i;
1244
1245         verf[0] = (__be32)get_seconds();
1246         verf[1] = (__be32)i++;
1247         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1248 }
1249
1250 static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1251 {
1252         return idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1253 }
1254
1255 static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1256 {
1257         struct nfs4_stid *s;
1258
1259         s = find_stateid(cl, t);
1260         if (!s)
1261                 return NULL;
1262         if (typemask & s->sc_type)
1263                 return s;
1264         return NULL;
1265 }
1266
1267 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1268                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1269 {
1270         struct nfs4_client *clp;
1271         struct sockaddr *sa = svc_addr(rqstp);
1272         int ret;
1273
1274         clp = alloc_client(name);
1275         if (clp == NULL)
1276                 return NULL;
1277
1278         INIT_LIST_HEAD(&clp->cl_sessions);
1279         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1280         if (ret) {
1281                 spin_lock(&client_lock);
1282                 free_client(clp);
1283                 spin_unlock(&client_lock);
1284                 return NULL;
1285         }
1286         idr_init(&clp->cl_stateids);
1287         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1288         atomic_set(&clp->cl_refcount, 0);
1289         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1290         INIT_LIST_HEAD(&clp->cl_idhash);
1291         INIT_LIST_HEAD(&clp->cl_strhash);
1292         INIT_LIST_HEAD(&clp->cl_openowners);
1293         INIT_LIST_HEAD(&clp->cl_delegations);
1294         INIT_LIST_HEAD(&clp->cl_lru);
1295         INIT_LIST_HEAD(&clp->cl_callbacks);
1296         spin_lock_init(&clp->cl_lock);
1297         INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1298         clp->cl_time = get_seconds();
1299         clear_bit(0, &clp->cl_cb_slot_busy);
1300         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1301         copy_verf(clp, verf);
1302         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1303         gen_confirm(clp);
1304         clp->cl_cb_session = NULL;
1305         return clp;
1306 }
1307
1308 static void
1309 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1310 {
1311         unsigned int idhashval;
1312
1313         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1314         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1315         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1316         renew_client(clp);
1317 }
1318
1319 static void
1320 move_to_confirmed(struct nfs4_client *clp)
1321 {
1322         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1323         unsigned int strhashval;
1324
1325         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1326         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1327         strhashval = clientstr_hashval(clp->cl_recdir);
1328         list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1329         renew_client(clp);
1330 }
1331
1332 static struct nfs4_client *
1333 find_confirmed_client(clientid_t *clid)
1334 {
1335         struct nfs4_client *clp;
1336         unsigned int idhashval = clientid_hashval(clid->cl_id);
1337
1338         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1339                 if (same_clid(&clp->cl_clientid, clid)) {
1340                         renew_client(clp);
1341                         return clp;
1342                 }
1343         }
1344         return NULL;
1345 }
1346
1347 static struct nfs4_client *
1348 find_unconfirmed_client(clientid_t *clid)
1349 {
1350         struct nfs4_client *clp;
1351         unsigned int idhashval = clientid_hashval(clid->cl_id);
1352
1353         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1354                 if (same_clid(&clp->cl_clientid, clid))
1355                         return clp;
1356         }
1357         return NULL;
1358 }
1359
1360 static bool clp_used_exchangeid(struct nfs4_client *clp)
1361 {
1362         return clp->cl_exchange_flags != 0;
1363 }
1364
1365 static struct nfs4_client *
1366 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
1367 {
1368         struct nfs4_client *clp;
1369
1370         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1371                 if (same_name(clp->cl_recdir, dname))
1372                         return clp;
1373         }
1374         return NULL;
1375 }
1376
1377 static struct nfs4_client *
1378 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
1379 {
1380         struct nfs4_client *clp;
1381
1382         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1383                 if (same_name(clp->cl_recdir, dname))
1384                         return clp;
1385         }
1386         return NULL;
1387 }
1388
1389 static void
1390 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1391 {
1392         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1393         struct sockaddr *sa = svc_addr(rqstp);
1394         u32 scopeid = rpc_get_scope_id(sa);
1395         unsigned short expected_family;
1396
1397         /* Currently, we only support tcp and tcp6 for the callback channel */
1398         if (se->se_callback_netid_len == 3 &&
1399             !memcmp(se->se_callback_netid_val, "tcp", 3))
1400                 expected_family = AF_INET;
1401         else if (se->se_callback_netid_len == 4 &&
1402                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1403                 expected_family = AF_INET6;
1404         else
1405                 goto out_err;
1406
1407         conn->cb_addrlen = rpc_uaddr2sockaddr(&init_net, se->se_callback_addr_val,
1408                                             se->se_callback_addr_len,
1409                                             (struct sockaddr *)&conn->cb_addr,
1410                                             sizeof(conn->cb_addr));
1411
1412         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1413                 goto out_err;
1414
1415         if (conn->cb_addr.ss_family == AF_INET6)
1416                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1417
1418         conn->cb_prog = se->se_callback_prog;
1419         conn->cb_ident = se->se_callback_ident;
1420         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1421         return;
1422 out_err:
1423         conn->cb_addr.ss_family = AF_UNSPEC;
1424         conn->cb_addrlen = 0;
1425         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1426                 "will not receive delegations\n",
1427                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1428
1429         return;
1430 }
1431
1432 /*
1433  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1434  */
1435 void
1436 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1437 {
1438         struct nfsd4_slot *slot = resp->cstate.slot;
1439         unsigned int base;
1440
1441         dprintk("--> %s slot %p\n", __func__, slot);
1442
1443         slot->sl_opcnt = resp->opcnt;
1444         slot->sl_status = resp->cstate.status;
1445
1446         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
1447         if (nfsd4_not_cached(resp)) {
1448                 slot->sl_datalen = 0;
1449                 return;
1450         }
1451         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1452         base = (char *)resp->cstate.datap -
1453                                         (char *)resp->xbuf->head[0].iov_base;
1454         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1455                                     slot->sl_datalen))
1456                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1457         return;
1458 }
1459
1460 /*
1461  * Encode the replay sequence operation from the slot values.
1462  * If cachethis is FALSE encode the uncached rep error on the next
1463  * operation which sets resp->p and increments resp->opcnt for
1464  * nfs4svc_encode_compoundres.
1465  *
1466  */
1467 static __be32
1468 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1469                           struct nfsd4_compoundres *resp)
1470 {
1471         struct nfsd4_op *op;
1472         struct nfsd4_slot *slot = resp->cstate.slot;
1473
1474         /* Encode the replayed sequence operation */
1475         op = &args->ops[resp->opcnt - 1];
1476         nfsd4_encode_operation(resp, op);
1477
1478         /* Return nfserr_retry_uncached_rep in next operation. */
1479         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
1480                 op = &args->ops[resp->opcnt++];
1481                 op->status = nfserr_retry_uncached_rep;
1482                 nfsd4_encode_operation(resp, op);
1483         }
1484         return op->status;
1485 }
1486
1487 /*
1488  * The sequence operation is not cached because we can use the slot and
1489  * session values.
1490  */
1491 __be32
1492 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1493                          struct nfsd4_sequence *seq)
1494 {
1495         struct nfsd4_slot *slot = resp->cstate.slot;
1496         __be32 status;
1497
1498         dprintk("--> %s slot %p\n", __func__, slot);
1499
1500         /* Either returns 0 or nfserr_retry_uncached */
1501         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1502         if (status == nfserr_retry_uncached_rep)
1503                 return status;
1504
1505         /* The sequence operation has been encoded, cstate->datap set. */
1506         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1507
1508         resp->opcnt = slot->sl_opcnt;
1509         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1510         status = slot->sl_status;
1511
1512         return status;
1513 }
1514
1515 /*
1516  * Set the exchange_id flags returned by the server.
1517  */
1518 static void
1519 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1520 {
1521         /* pNFS is not supported */
1522         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1523
1524         /* Referrals are supported, Migration is not. */
1525         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1526
1527         /* set the wire flags to return to client. */
1528         clid->flags = new->cl_exchange_flags;
1529 }
1530
1531 static bool client_has_state(struct nfs4_client *clp)
1532 {
1533         /*
1534          * Note clp->cl_openowners check isn't quite right: there's no
1535          * need to count owners without stateid's.
1536          *
1537          * Also note we should probably be using this in 4.0 case too.
1538          */
1539         return list_empty(&clp->cl_openowners)
1540                 && list_empty(&clp->cl_delegations)
1541                 && list_empty(&clp->cl_sessions);
1542 }
1543
1544 __be32
1545 nfsd4_exchange_id(struct svc_rqst *rqstp,
1546                   struct nfsd4_compound_state *cstate,
1547                   struct nfsd4_exchange_id *exid)
1548 {
1549         struct nfs4_client *unconf, *conf, *new;
1550         __be32 status;
1551         unsigned int            strhashval;
1552         char                    dname[HEXDIR_LEN];
1553         char                    addr_str[INET6_ADDRSTRLEN];
1554         nfs4_verifier           verf = exid->verifier;
1555         struct sockaddr         *sa = svc_addr(rqstp);
1556         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
1557
1558         rpc_ntop(sa, addr_str, sizeof(addr_str));
1559         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1560                 "ip_addr=%s flags %x, spa_how %d\n",
1561                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1562                 addr_str, exid->flags, exid->spa_how);
1563
1564         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1565                 return nfserr_inval;
1566
1567         /* Currently only support SP4_NONE */
1568         switch (exid->spa_how) {
1569         case SP4_NONE:
1570                 break;
1571         case SP4_SSV:
1572                 return nfserr_serverfault;
1573         default:
1574                 BUG();                          /* checked by xdr code */
1575         case SP4_MACH_CRED:
1576                 return nfserr_serverfault;      /* no excuse :-/ */
1577         }
1578
1579         status = nfs4_make_rec_clidname(dname, &exid->clname);
1580
1581         if (status)
1582                 return status;
1583
1584         strhashval = clientstr_hashval(dname);
1585
1586         /* Cases below refer to rfc 5661 section 18.35.4: */
1587         nfs4_lock_state();
1588         conf = find_confirmed_client_by_str(dname, strhashval);
1589         if (conf) {
1590                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
1591                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
1592
1593                 if (update) {
1594                         if (!clp_used_exchangeid(conf)) { /* buggy client */
1595                                 status = nfserr_inval;
1596                                 goto out;
1597                         }
1598                         if (!creds_match) { /* case 9 */
1599                                 status = nfserr_perm;
1600                                 goto out;
1601                         }
1602                         if (!verfs_match) { /* case 8 */
1603                                 status = nfserr_not_same;
1604                                 goto out;
1605                         }
1606                         /* case 6 */
1607                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1608                         new = conf;
1609                         goto out_copy;
1610                 }
1611                 if (!creds_match) { /* case 3 */
1612                         if (client_has_state(conf)) {
1613                                 status = nfserr_clid_inuse;
1614                                 goto out;
1615                         }
1616                         goto expire_client;
1617                 }
1618                 if (verfs_match) { /* case 2 */
1619                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1620                         new = conf;
1621                         goto out_copy;
1622                 }
1623                 /* case 5, client reboot */
1624 expire_client:
1625                 expire_client(conf);
1626                 goto out_new;
1627         }
1628
1629         if (update) { /* case 7 */
1630                 status = nfserr_noent;
1631                 goto out;
1632         }
1633
1634         unconf  = find_unconfirmed_client_by_str(dname, strhashval);
1635         if (unconf) /* case 4, possible retry or client restart */
1636                 expire_client(unconf);
1637
1638         /* case 1 (normal case) */
1639 out_new:
1640         new = create_client(exid->clname, dname, rqstp, &verf);
1641         if (new == NULL) {
1642                 status = nfserr_jukebox;
1643                 goto out;
1644         }
1645
1646         gen_clid(new);
1647         add_to_unconfirmed(new, strhashval);
1648 out_copy:
1649         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1650         exid->clientid.cl_id = new->cl_clientid.cl_id;
1651
1652         exid->seqid = 1;
1653         nfsd4_set_ex_flags(new, exid);
1654
1655         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1656                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1657         status = nfs_ok;
1658
1659 out:
1660         nfs4_unlock_state();
1661         return status;
1662 }
1663
1664 static __be32
1665 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1666 {
1667         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1668                 slot_seqid);
1669
1670         /* The slot is in use, and no response has been sent. */
1671         if (slot_inuse) {
1672                 if (seqid == slot_seqid)
1673                         return nfserr_jukebox;
1674                 else
1675                         return nfserr_seq_misordered;
1676         }
1677         /* Note unsigned 32-bit arithmetic handles wraparound: */
1678         if (likely(seqid == slot_seqid + 1))
1679                 return nfs_ok;
1680         if (seqid == slot_seqid)
1681                 return nfserr_replay_cache;
1682         return nfserr_seq_misordered;
1683 }
1684
1685 /*
1686  * Cache the create session result into the create session single DRC
1687  * slot cache by saving the xdr structure. sl_seqid has been set.
1688  * Do this for solo or embedded create session operations.
1689  */
1690 static void
1691 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1692                            struct nfsd4_clid_slot *slot, __be32 nfserr)
1693 {
1694         slot->sl_status = nfserr;
1695         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1696 }
1697
1698 static __be32
1699 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1700                             struct nfsd4_clid_slot *slot)
1701 {
1702         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1703         return slot->sl_status;
1704 }
1705
1706 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1707                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1708                         1 +     /* MIN tag is length with zero, only length */ \
1709                         3 +     /* version, opcount, opcode */ \
1710                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1711                                 /* seqid, slotID, slotID, cache */ \
1712                         4 ) * sizeof(__be32))
1713
1714 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1715                         2 +     /* verifier: AUTH_NULL, length 0 */\
1716                         1 +     /* status */ \
1717                         1 +     /* MIN tag is length with zero, only length */ \
1718                         3 +     /* opcount, opcode, opstatus*/ \
1719                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1720                                 /* seqid, slotID, slotID, slotID, status */ \
1721                         5 ) * sizeof(__be32))
1722
1723 static bool check_forechannel_attrs(struct nfsd4_channel_attrs fchannel)
1724 {
1725         return fchannel.maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ
1726                 || fchannel.maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ;
1727 }
1728
1729 __be32
1730 nfsd4_create_session(struct svc_rqst *rqstp,
1731                      struct nfsd4_compound_state *cstate,
1732                      struct nfsd4_create_session *cr_ses)
1733 {
1734         struct sockaddr *sa = svc_addr(rqstp);
1735         struct nfs4_client *conf, *unconf;
1736         struct nfsd4_session *new;
1737         struct nfsd4_clid_slot *cs_slot = NULL;
1738         bool confirm_me = false;
1739         __be32 status = 0;
1740
1741         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1742                 return nfserr_inval;
1743
1744         nfs4_lock_state();
1745         unconf = find_unconfirmed_client(&cr_ses->clientid);
1746         conf = find_confirmed_client(&cr_ses->clientid);
1747
1748         if (conf) {
1749                 cs_slot = &conf->cl_cs_slot;
1750                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1751                 if (status == nfserr_replay_cache) {
1752                         dprintk("Got a create_session replay! seqid= %d\n",
1753                                 cs_slot->sl_seqid);
1754                         /* Return the cached reply status */
1755                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1756                         goto out;
1757                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1758                         status = nfserr_seq_misordered;
1759                         dprintk("Sequence misordered!\n");
1760                         dprintk("Expected seqid= %d but got seqid= %d\n",
1761                                 cs_slot->sl_seqid, cr_ses->seqid);
1762                         goto out;
1763                 }
1764         } else if (unconf) {
1765                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1766                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1767                         status = nfserr_clid_inuse;
1768                         goto out;
1769                 }
1770
1771                 cs_slot = &unconf->cl_cs_slot;
1772                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1773                 if (status) {
1774                         /* an unconfirmed replay returns misordered */
1775                         status = nfserr_seq_misordered;
1776                         goto out;
1777                 }
1778
1779                 confirm_me = true;
1780                 conf = unconf;
1781         } else {
1782                 status = nfserr_stale_clientid;
1783                 goto out;
1784         }
1785
1786         /*
1787          * XXX: we should probably set this at creation time, and check
1788          * for consistent minorversion use throughout:
1789          */
1790         conf->cl_minorversion = 1;
1791         /*
1792          * We do not support RDMA or persistent sessions
1793          */
1794         cr_ses->flags &= ~SESSION4_PERSIST;
1795         cr_ses->flags &= ~SESSION4_RDMA;
1796
1797         status = nfserr_toosmall;
1798         if (check_forechannel_attrs(cr_ses->fore_channel))
1799                 goto out;
1800
1801         status = nfserr_jukebox;
1802         new = alloc_init_session(rqstp, conf, cr_ses);
1803         if (!new)
1804                 goto out;
1805         status = nfs_ok;
1806         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1807                NFS4_MAX_SESSIONID_LEN);
1808         memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1809                 sizeof(struct nfsd4_channel_attrs));
1810         cs_slot->sl_seqid++;
1811         cr_ses->seqid = cs_slot->sl_seqid;
1812
1813         /* cache solo and embedded create sessions under the state lock */
1814         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1815         if (confirm_me)
1816                 move_to_confirmed(conf);
1817 out:
1818         nfs4_unlock_state();
1819         dprintk("%s returns %d\n", __func__, ntohl(status));
1820         return status;
1821 }
1822
1823 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1824 {
1825         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1826         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1827
1828         return argp->opcnt == resp->opcnt;
1829 }
1830
1831 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1832 {
1833         switch (*dir) {
1834         case NFS4_CDFC4_FORE:
1835         case NFS4_CDFC4_BACK:
1836                 return nfs_ok;
1837         case NFS4_CDFC4_FORE_OR_BOTH:
1838         case NFS4_CDFC4_BACK_OR_BOTH:
1839                 *dir = NFS4_CDFC4_BOTH;
1840                 return nfs_ok;
1841         };
1842         return nfserr_inval;
1843 }
1844
1845 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1846                      struct nfsd4_compound_state *cstate,
1847                      struct nfsd4_bind_conn_to_session *bcts)
1848 {
1849         __be32 status;
1850
1851         if (!nfsd4_last_compound_op(rqstp))
1852                 return nfserr_not_only_op;
1853         spin_lock(&client_lock);
1854         cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid);
1855         /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1856          * client_lock iself: */
1857         if (cstate->session) {
1858                 nfsd4_get_session(cstate->session);
1859                 atomic_inc(&cstate->session->se_client->cl_refcount);
1860         }
1861         spin_unlock(&client_lock);
1862         if (!cstate->session)
1863                 return nfserr_badsession;
1864
1865         status = nfsd4_map_bcts_dir(&bcts->dir);
1866         if (!status)
1867                 nfsd4_new_conn(rqstp, cstate->session, bcts->dir);
1868         return status;
1869 }
1870
1871 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1872 {
1873         if (!session)
1874                 return 0;
1875         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1876 }
1877
1878 __be32
1879 nfsd4_destroy_session(struct svc_rqst *r,
1880                       struct nfsd4_compound_state *cstate,
1881                       struct nfsd4_destroy_session *sessionid)
1882 {
1883         struct nfsd4_session *ses;
1884         __be32 status = nfserr_badsession;
1885
1886         /* Notes:
1887          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1888          * - Should we return nfserr_back_chan_busy if waiting for
1889          *   callbacks on to-be-destroyed session?
1890          * - Do we need to clear any callback info from previous session?
1891          */
1892
1893         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1894                 if (!nfsd4_last_compound_op(r))
1895                         return nfserr_not_only_op;
1896         }
1897         dump_sessionid(__func__, &sessionid->sessionid);
1898         spin_lock(&client_lock);
1899         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1900         if (!ses) {
1901                 spin_unlock(&client_lock);
1902                 goto out;
1903         }
1904
1905         unhash_session(ses);
1906         spin_unlock(&client_lock);
1907
1908         nfs4_lock_state();
1909         nfsd4_probe_callback_sync(ses->se_client);
1910         nfs4_unlock_state();
1911
1912         spin_lock(&client_lock);
1913         nfsd4_del_conns(ses);
1914         nfsd4_put_session_locked(ses);
1915         spin_unlock(&client_lock);
1916         status = nfs_ok;
1917 out:
1918         dprintk("%s returns %d\n", __func__, ntohl(status));
1919         return status;
1920 }
1921
1922 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1923 {
1924         struct nfsd4_conn *c;
1925
1926         list_for_each_entry(c, &s->se_conns, cn_persession) {
1927                 if (c->cn_xprt == xpt) {
1928                         return c;
1929                 }
1930         }
1931         return NULL;
1932 }
1933
1934 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1935 {
1936         struct nfs4_client *clp = ses->se_client;
1937         struct nfsd4_conn *c;
1938         int ret;
1939
1940         spin_lock(&clp->cl_lock);
1941         c = __nfsd4_find_conn(new->cn_xprt, ses);
1942         if (c) {
1943                 spin_unlock(&clp->cl_lock);
1944                 free_conn(new);
1945                 return;
1946         }
1947         __nfsd4_hash_conn(new, ses);
1948         spin_unlock(&clp->cl_lock);
1949         ret = nfsd4_register_conn(new);
1950         if (ret)
1951                 /* oops; xprt is already down: */
1952                 nfsd4_conn_lost(&new->cn_xpt_user);
1953         return;
1954 }
1955
1956 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
1957 {
1958         struct nfsd4_compoundargs *args = rqstp->rq_argp;
1959
1960         return args->opcnt > session->se_fchannel.maxops;
1961 }
1962
1963 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
1964                                   struct nfsd4_session *session)
1965 {
1966         struct xdr_buf *xb = &rqstp->rq_arg;
1967
1968         return xb->len > session->se_fchannel.maxreq_sz;
1969 }
1970
1971 __be32
1972 nfsd4_sequence(struct svc_rqst *rqstp,
1973                struct nfsd4_compound_state *cstate,
1974                struct nfsd4_sequence *seq)
1975 {
1976         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1977         struct nfsd4_session *session;
1978         struct nfsd4_slot *slot;
1979         struct nfsd4_conn *conn;
1980         __be32 status;
1981
1982         if (resp->opcnt != 1)
1983                 return nfserr_sequence_pos;
1984
1985         /*
1986          * Will be either used or freed by nfsd4_sequence_check_conn
1987          * below.
1988          */
1989         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
1990         if (!conn)
1991                 return nfserr_jukebox;
1992
1993         spin_lock(&client_lock);
1994         status = nfserr_badsession;
1995         session = find_in_sessionid_hashtbl(&seq->sessionid);
1996         if (!session)
1997                 goto out;
1998
1999         status = nfserr_too_many_ops;
2000         if (nfsd4_session_too_many_ops(rqstp, session))
2001                 goto out;
2002
2003         status = nfserr_req_too_big;
2004         if (nfsd4_request_too_big(rqstp, session))
2005                 goto out;
2006
2007         status = nfserr_badslot;
2008         if (seq->slotid >= session->se_fchannel.maxreqs)
2009                 goto out;
2010
2011         slot = session->se_slots[seq->slotid];
2012         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2013
2014         /* We do not negotiate the number of slots yet, so set the
2015          * maxslots to the session maxreqs which is used to encode
2016          * sr_highest_slotid and the sr_target_slot id to maxslots */
2017         seq->maxslots = session->se_fchannel.maxreqs;
2018
2019         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2020                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2021         if (status == nfserr_replay_cache) {
2022                 status = nfserr_seq_misordered;
2023                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2024                         goto out;
2025                 cstate->slot = slot;
2026                 cstate->session = session;
2027                 /* Return the cached reply status and set cstate->status
2028                  * for nfsd4_proc_compound processing */
2029                 status = nfsd4_replay_cache_entry(resp, seq);
2030                 cstate->status = nfserr_replay_cache;
2031                 goto out;
2032         }
2033         if (status)
2034                 goto out;
2035
2036         nfsd4_sequence_check_conn(conn, session);
2037         conn = NULL;
2038
2039         /* Success! bump slot seqid */
2040         slot->sl_seqid = seq->seqid;
2041         slot->sl_flags |= NFSD4_SLOT_INUSE;
2042         if (seq->cachethis)
2043                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2044         else
2045                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2046
2047         cstate->slot = slot;
2048         cstate->session = session;
2049
2050 out:
2051         /* Hold a session reference until done processing the compound. */
2052         if (cstate->session) {
2053                 struct nfs4_client *clp = session->se_client;
2054
2055                 nfsd4_get_session(cstate->session);
2056                 atomic_inc(&clp->cl_refcount);
2057                 switch (clp->cl_cb_state) {
2058                 case NFSD4_CB_DOWN:
2059                         seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2060                         break;
2061                 case NFSD4_CB_FAULT:
2062                         seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2063                         break;
2064                 default:
2065                         seq->status_flags = 0;
2066                 }
2067         }
2068         kfree(conn);
2069         spin_unlock(&client_lock);
2070         dprintk("%s: return %d\n", __func__, ntohl(status));
2071         return status;
2072 }
2073
2074 static inline bool has_resources(struct nfs4_client *clp)
2075 {
2076         return !list_empty(&clp->cl_openowners)
2077                 || !list_empty(&clp->cl_delegations)
2078                 || !list_empty(&clp->cl_sessions);
2079 }
2080
2081 __be32
2082 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2083 {
2084         struct nfs4_client *conf, *unconf, *clp;
2085         __be32 status = 0;
2086
2087         nfs4_lock_state();
2088         unconf = find_unconfirmed_client(&dc->clientid);
2089         conf = find_confirmed_client(&dc->clientid);
2090
2091         if (conf) {
2092                 clp = conf;
2093
2094                 if (!is_client_expired(conf) && has_resources(conf)) {
2095                         status = nfserr_clientid_busy;
2096                         goto out;
2097                 }
2098
2099                 /* rfc5661 18.50.3 */
2100                 if (cstate->session && conf == cstate->session->se_client) {
2101                         status = nfserr_clientid_busy;
2102                         goto out;
2103                 }
2104         } else if (unconf)
2105                 clp = unconf;
2106         else {
2107                 status = nfserr_stale_clientid;
2108                 goto out;
2109         }
2110
2111         expire_client(clp);
2112 out:
2113         nfs4_unlock_state();
2114         dprintk("%s return %d\n", __func__, ntohl(status));
2115         return status;
2116 }
2117
2118 __be32
2119 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2120 {
2121         __be32 status = 0;
2122
2123         if (rc->rca_one_fs) {
2124                 if (!cstate->current_fh.fh_dentry)
2125                         return nfserr_nofilehandle;
2126                 /*
2127                  * We don't take advantage of the rca_one_fs case.
2128                  * That's OK, it's optional, we can safely ignore it.
2129                  */
2130                  return nfs_ok;
2131         }
2132
2133         nfs4_lock_state();
2134         status = nfserr_complete_already;
2135         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2136                              &cstate->session->se_client->cl_flags))
2137                 goto out;
2138
2139         status = nfserr_stale_clientid;
2140         if (is_client_expired(cstate->session->se_client))
2141                 /*
2142                  * The following error isn't really legal.
2143                  * But we only get here if the client just explicitly
2144                  * destroyed the client.  Surely it no longer cares what
2145                  * error it gets back on an operation for the dead
2146                  * client.
2147                  */
2148                 goto out;
2149
2150         status = nfs_ok;
2151         nfsd4_client_record_create(cstate->session->se_client);
2152 out:
2153         nfs4_unlock_state();
2154         return status;
2155 }
2156
2157 __be32
2158 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2159                   struct nfsd4_setclientid *setclid)
2160 {
2161         struct xdr_netobj       clname = setclid->se_name;
2162         nfs4_verifier           clverifier = setclid->se_verf;
2163         unsigned int            strhashval;
2164         struct nfs4_client      *conf, *unconf, *new;
2165         __be32                  status;
2166         char                    dname[HEXDIR_LEN];
2167
2168         status = nfs4_make_rec_clidname(dname, &clname);
2169         if (status)
2170                 return status;
2171
2172         /*
2173          * XXX The Duplicate Request Cache (DRC) has been checked (??)
2174          * We get here on a DRC miss.
2175          */
2176
2177         strhashval = clientstr_hashval(dname);
2178
2179         nfs4_lock_state();
2180         conf = find_confirmed_client_by_str(dname, strhashval);
2181         if (conf) {
2182                 /* RFC 3530 14.2.33 CASE 0: */
2183                 status = nfserr_clid_inuse;
2184                 if (clp_used_exchangeid(conf))
2185                         goto out;
2186                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2187                         char addr_str[INET6_ADDRSTRLEN];
2188                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2189                                  sizeof(addr_str));
2190                         dprintk("NFSD: setclientid: string in use by client "
2191                                 "at %s\n", addr_str);
2192                         goto out;
2193                 }
2194         }
2195         /*
2196          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
2197          * has a description of SETCLIENTID request processing consisting
2198          * of 5 bullet points, labeled as CASE0 - CASE4 below.
2199          */
2200         unconf = find_unconfirmed_client_by_str(dname, strhashval);
2201         status = nfserr_jukebox;
2202         if (!conf) {
2203                 /*
2204                  * RFC 3530 14.2.33 CASE 4:
2205                  * placed first, because it is the normal case
2206                  */
2207                 if (unconf)
2208                         expire_client(unconf);
2209                 new = create_client(clname, dname, rqstp, &clverifier);
2210                 if (new == NULL)
2211                         goto out;
2212                 gen_clid(new);
2213         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
2214                 /*
2215                  * RFC 3530 14.2.33 CASE 1:
2216                  * probable callback update
2217                  */
2218                 if (unconf) {
2219                         /* Note this is removing unconfirmed {*x***},
2220                          * which is stronger than RFC recommended {vxc**}.
2221                          * This has the advantage that there is at most
2222                          * one {*x***} in either list at any time.
2223                          */
2224                         expire_client(unconf);
2225                 }
2226                 new = create_client(clname, dname, rqstp, &clverifier);
2227                 if (new == NULL)
2228                         goto out;
2229                 copy_clid(new, conf);
2230         } else if (!unconf) {
2231                 /*
2232                  * RFC 3530 14.2.33 CASE 2:
2233                  * probable client reboot; state will be removed if
2234                  * confirmed.
2235                  */
2236                 new = create_client(clname, dname, rqstp, &clverifier);
2237                 if (new == NULL)
2238                         goto out;
2239                 gen_clid(new);
2240         } else {
2241                 /*
2242                  * RFC 3530 14.2.33 CASE 3:
2243                  * probable client reboot; state will be removed if
2244                  * confirmed.
2245                  */
2246                 expire_client(unconf);
2247                 new = create_client(clname, dname, rqstp, &clverifier);
2248                 if (new == NULL)
2249                         goto out;
2250                 gen_clid(new);
2251         }
2252         /*
2253          * XXX: we should probably set this at creation time, and check
2254          * for consistent minorversion use throughout:
2255          */
2256         new->cl_minorversion = 0;
2257         gen_callback(new, setclid, rqstp);
2258         add_to_unconfirmed(new, strhashval);
2259         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2260         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2261         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2262         status = nfs_ok;
2263 out:
2264         nfs4_unlock_state();
2265         return status;
2266 }
2267
2268
2269 /*
2270  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
2271  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
2272  * bullets, labeled as CASE1 - CASE4 below.
2273  */
2274 __be32
2275 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2276                          struct nfsd4_compound_state *cstate,
2277                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2278 {
2279         struct nfs4_client *conf, *unconf;
2280         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2281         clientid_t * clid = &setclientid_confirm->sc_clientid;
2282         __be32 status;
2283
2284         if (STALE_CLIENTID(clid))
2285                 return nfserr_stale_clientid;
2286         /*
2287          * XXX The Duplicate Request Cache (DRC) has been checked (??)
2288          * We get here on a DRC miss.
2289          */
2290
2291         nfs4_lock_state();
2292
2293         conf = find_confirmed_client(clid);
2294         unconf = find_unconfirmed_client(clid);
2295
2296         /*
2297          * section 14.2.34 of RFC 3530 has a description of
2298          * SETCLIENTID_CONFIRM request processing consisting
2299          * of 4 bullet points, labeled as CASE1 - CASE4 below.
2300          */
2301         status = nfserr_clid_inuse;
2302         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
2303                 /*
2304                  * RFC 3530 14.2.34 CASE 1:
2305                  * callback update
2306                  */
2307                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
2308                         status = nfserr_clid_inuse;
2309                 else {
2310                         nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2311                         nfsd4_probe_callback(conf);
2312                         expire_client(unconf);
2313                         status = nfs_ok;
2314                 }
2315         } else if (conf && !unconf) {
2316                 /*
2317                  * RFC 3530 14.2.34 CASE 2:
2318                  * probable retransmitted request; play it safe and
2319                  * do nothing.
2320                  */
2321                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
2322                         status = nfserr_clid_inuse;
2323                 else
2324                         status = nfs_ok;
2325         } else if (!conf && unconf
2326                         && same_verf(&unconf->cl_confirm, &confirm)) {
2327                 /*
2328                  * RFC 3530 14.2.34 CASE 3:
2329                  * Normal case; new or rebooted client:
2330                  */
2331                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
2332                         status = nfserr_clid_inuse;
2333                 } else {
2334                         unsigned int hash =
2335                                 clientstr_hashval(unconf->cl_recdir);
2336                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
2337                                                             hash);
2338                         if (conf) {
2339                                 nfsd4_client_record_remove(conf);
2340                                 expire_client(conf);
2341                         }
2342                         move_to_confirmed(unconf);
2343                         conf = unconf;
2344                         nfsd4_probe_callback(conf);
2345                         status = nfs_ok;
2346                 }
2347         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
2348             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
2349                                                                 &confirm)))) {
2350                 /*
2351                  * RFC 3530 14.2.34 CASE 4:
2352                  * Client probably hasn't noticed that we rebooted yet.
2353                  */
2354                 status = nfserr_stale_clientid;
2355         }
2356
2357         nfs4_unlock_state();
2358         return status;
2359 }
2360
2361 static struct nfs4_file *nfsd4_alloc_file(void)
2362 {
2363         return kmem_cache_alloc(file_slab, GFP_KERNEL);
2364 }
2365
2366 /* OPEN Share state helper functions */
2367 static void nfsd4_init_file(struct nfs4_file *fp, struct inode *ino)
2368 {
2369         unsigned int hashval = file_hashval(ino);
2370
2371         atomic_set(&fp->fi_ref, 1);
2372         INIT_LIST_HEAD(&fp->fi_hash);
2373         INIT_LIST_HEAD(&fp->fi_stateids);
2374         INIT_LIST_HEAD(&fp->fi_delegations);
2375         fp->fi_inode = igrab(ino);
2376         fp->fi_had_conflict = false;
2377         fp->fi_lease = NULL;
2378         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2379         memset(fp->fi_access, 0, sizeof(fp->fi_access));
2380         spin_lock(&recall_lock);
2381         list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2382         spin_unlock(&recall_lock);
2383 }
2384
2385 static void
2386 nfsd4_free_slab(struct kmem_cache **slab)
2387 {
2388         if (*slab == NULL)
2389                 return;
2390         kmem_cache_destroy(*slab);
2391         *slab = NULL;
2392 }
2393
2394 void
2395 nfsd4_free_slabs(void)
2396 {
2397         nfsd4_free_slab(&openowner_slab);
2398         nfsd4_free_slab(&lockowner_slab);
2399         nfsd4_free_slab(&file_slab);
2400         nfsd4_free_slab(&stateid_slab);
2401         nfsd4_free_slab(&deleg_slab);
2402 }
2403
2404 int
2405 nfsd4_init_slabs(void)
2406 {
2407         openowner_slab = kmem_cache_create("nfsd4_openowners",
2408                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2409         if (openowner_slab == NULL)
2410                 goto out_nomem;
2411         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2412                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2413         if (lockowner_slab == NULL)
2414                 goto out_nomem;
2415         file_slab = kmem_cache_create("nfsd4_files",
2416                         sizeof(struct nfs4_file), 0, 0, NULL);
2417         if (file_slab == NULL)
2418                 goto out_nomem;
2419         stateid_slab = kmem_cache_create("nfsd4_stateids",
2420                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2421         if (stateid_slab == NULL)
2422                 goto out_nomem;
2423         deleg_slab = kmem_cache_create("nfsd4_delegations",
2424                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2425         if (deleg_slab == NULL)
2426                 goto out_nomem;
2427         return 0;
2428 out_nomem:
2429         nfsd4_free_slabs();
2430         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2431         return -ENOMEM;
2432 }
2433
2434 void nfs4_free_openowner(struct nfs4_openowner *oo)
2435 {
2436         kfree(oo->oo_owner.so_owner.data);
2437         kmem_cache_free(openowner_slab, oo);
2438 }
2439
2440 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2441 {
2442         kfree(lo->lo_owner.so_owner.data);
2443         kmem_cache_free(lockowner_slab, lo);
2444 }
2445
2446 static void init_nfs4_replay(struct nfs4_replay *rp)
2447 {
2448         rp->rp_status = nfserr_serverfault;
2449         rp->rp_buflen = 0;
2450         rp->rp_buf = rp->rp_ibuf;
2451 }
2452
2453 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2454 {
2455         struct nfs4_stateowner *sop;
2456
2457         sop = kmem_cache_alloc(slab, GFP_KERNEL);
2458         if (!sop)
2459                 return NULL;
2460
2461         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2462         if (!sop->so_owner.data) {
2463                 kmem_cache_free(slab, sop);
2464                 return NULL;
2465         }
2466         sop->so_owner.len = owner->len;
2467
2468         INIT_LIST_HEAD(&sop->so_stateids);
2469         sop->so_client = clp;
2470         init_nfs4_replay(&sop->so_replay);
2471         return sop;
2472 }
2473
2474 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2475 {
2476         list_add(&oo->oo_owner.so_strhash, &ownerstr_hashtbl[strhashval]);
2477         list_add(&oo->oo_perclient, &clp->cl_openowners);
2478 }
2479
2480 static struct nfs4_openowner *
2481 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2482         struct nfs4_openowner *oo;
2483
2484         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2485         if (!oo)
2486                 return NULL;
2487         oo->oo_owner.so_is_open_owner = 1;
2488         oo->oo_owner.so_seqid = open->op_seqid;
2489         oo->oo_flags = NFS4_OO_NEW;
2490         oo->oo_time = 0;
2491         oo->oo_last_closed_stid = NULL;
2492         INIT_LIST_HEAD(&oo->oo_close_lru);
2493         hash_openowner(oo, clp, strhashval);
2494         return oo;
2495 }
2496
2497 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2498         struct nfs4_openowner *oo = open->op_openowner;
2499         struct nfs4_client *clp = oo->oo_owner.so_client;
2500
2501         init_stid(&stp->st_stid, clp, NFS4_OPEN_STID);
2502         INIT_LIST_HEAD(&stp->st_lockowners);
2503         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2504         list_add(&stp->st_perfile, &fp->fi_stateids);
2505         stp->st_stateowner = &oo->oo_owner;
2506         get_nfs4_file(fp);
2507         stp->st_file = fp;
2508         stp->st_access_bmap = 0;
2509         stp->st_deny_bmap = 0;
2510         set_access(open->op_share_access, stp);
2511         set_deny(open->op_share_deny, stp);
2512         stp->st_openstp = NULL;
2513 }
2514
2515 static void
2516 move_to_close_lru(struct nfs4_openowner *oo)
2517 {
2518         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2519
2520         list_move_tail(&oo->oo_close_lru, &close_lru);
2521         oo->oo_time = get_seconds();
2522 }
2523
2524 static int
2525 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2526                                                         clientid_t *clid)
2527 {
2528         return (sop->so_owner.len == owner->len) &&
2529                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2530                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2531 }
2532
2533 static struct nfs4_openowner *
2534 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2535 {
2536         struct nfs4_stateowner *so;
2537         struct nfs4_openowner *oo;
2538
2539         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2540                 if (!so->so_is_open_owner)
2541                         continue;
2542                 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2543                         oo = openowner(so);
2544                         renew_client(oo->oo_owner.so_client);
2545                         return oo;
2546                 }
2547         }
2548         return NULL;
2549 }
2550
2551 /* search file_hashtbl[] for file */
2552 static struct nfs4_file *
2553 find_file(struct inode *ino)
2554 {
2555         unsigned int hashval = file_hashval(ino);
2556         struct nfs4_file *fp;
2557
2558         spin_lock(&recall_lock);
2559         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2560                 if (fp->fi_inode == ino) {
2561                         get_nfs4_file(fp);
2562                         spin_unlock(&recall_lock);
2563                         return fp;
2564                 }
2565         }
2566         spin_unlock(&recall_lock);
2567         return NULL;
2568 }
2569
2570 /*
2571  * Called to check deny when READ with all zero stateid or
2572  * WRITE with all zero or all one stateid
2573  */
2574 static __be32
2575 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2576 {
2577         struct inode *ino = current_fh->fh_dentry->d_inode;
2578         struct nfs4_file *fp;
2579         struct nfs4_ol_stateid *stp;
2580         __be32 ret;
2581
2582         dprintk("NFSD: nfs4_share_conflict\n");
2583
2584         fp = find_file(ino);
2585         if (!fp)
2586                 return nfs_ok;
2587         ret = nfserr_locked;
2588         /* Search for conflicting share reservations */
2589         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2590                 if (test_deny(deny_type, stp) ||
2591                     test_deny(NFS4_SHARE_DENY_BOTH, stp))
2592                         goto out;
2593         }
2594         ret = nfs_ok;
2595 out:
2596         put_nfs4_file(fp);
2597         return ret;
2598 }
2599
2600 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2601 {
2602         /* We're assuming the state code never drops its reference
2603          * without first removing the lease.  Since we're in this lease
2604          * callback (and since the lease code is serialized by the kernel
2605          * lock) we know the server hasn't removed the lease yet, we know
2606          * it's safe to take a reference: */
2607         atomic_inc(&dp->dl_count);
2608
2609         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2610
2611         /* only place dl_time is set. protected by lock_flocks*/
2612         dp->dl_time = get_seconds();
2613
2614         nfsd4_cb_recall(dp);
2615 }
2616
2617 /* Called from break_lease() with lock_flocks() held. */
2618 static void nfsd_break_deleg_cb(struct file_lock *fl)
2619 {
2620         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2621         struct nfs4_delegation *dp;
2622
2623         BUG_ON(!fp);
2624         /* We assume break_lease is only called once per lease: */
2625         BUG_ON(fp->fi_had_conflict);
2626         /*
2627          * We don't want the locks code to timeout the lease for us;
2628          * we'll remove it ourself if a delegation isn't returned
2629          * in time:
2630          */
2631         fl->fl_break_time = 0;
2632
2633         spin_lock(&recall_lock);
2634         fp->fi_had_conflict = true;
2635         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2636                 nfsd_break_one_deleg(dp);
2637         spin_unlock(&recall_lock);
2638 }
2639
2640 static
2641 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2642 {
2643         if (arg & F_UNLCK)
2644                 return lease_modify(onlist, arg);
2645         else
2646                 return -EAGAIN;
2647 }
2648
2649 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2650         .lm_break = nfsd_break_deleg_cb,
2651         .lm_change = nfsd_change_deleg_cb,
2652 };
2653
2654 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2655 {
2656         if (nfsd4_has_session(cstate))
2657                 return nfs_ok;
2658         if (seqid == so->so_seqid - 1)
2659                 return nfserr_replay_me;
2660         if (seqid == so->so_seqid)
2661                 return nfs_ok;
2662         return nfserr_bad_seqid;
2663 }
2664
2665 __be32
2666 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2667                     struct nfsd4_open *open)
2668 {
2669         clientid_t *clientid = &open->op_clientid;
2670         struct nfs4_client *clp = NULL;
2671         unsigned int strhashval;
2672         struct nfs4_openowner *oo = NULL;
2673         __be32 status;
2674
2675         if (STALE_CLIENTID(&open->op_clientid))
2676                 return nfserr_stale_clientid;
2677         /*
2678          * In case we need it later, after we've already created the
2679          * file and don't want to risk a further failure:
2680          */
2681         open->op_file = nfsd4_alloc_file();
2682         if (open->op_file == NULL)
2683                 return nfserr_jukebox;
2684
2685         strhashval = ownerstr_hashval(clientid->cl_id, &open->op_owner);
2686         oo = find_openstateowner_str(strhashval, open);
2687         open->op_openowner = oo;
2688         if (!oo) {
2689                 clp = find_confirmed_client(clientid);
2690                 if (clp == NULL)
2691                         return nfserr_expired;
2692                 goto new_owner;
2693         }
2694         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2695                 /* Replace unconfirmed owners without checking for replay. */
2696                 clp = oo->oo_owner.so_client;
2697                 release_openowner(oo);
2698                 open->op_openowner = NULL;
2699                 goto new_owner;
2700         }
2701         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2702         if (status)
2703                 return status;
2704         clp = oo->oo_owner.so_client;
2705         goto alloc_stateid;
2706 new_owner:
2707         oo = alloc_init_open_stateowner(strhashval, clp, open);
2708         if (oo == NULL)
2709                 return nfserr_jukebox;
2710         open->op_openowner = oo;
2711 alloc_stateid:
2712         open->op_stp = nfs4_alloc_stateid(clp);
2713         if (!open->op_stp)
2714                 return nfserr_jukebox;
2715         return nfs_ok;
2716 }
2717
2718 static inline __be32
2719 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2720 {
2721         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2722                 return nfserr_openmode;
2723         else
2724                 return nfs_ok;
2725 }
2726
2727 static int share_access_to_flags(u32 share_access)
2728 {
2729         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2730 }
2731
2732 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2733 {
2734         struct nfs4_stid *ret;
2735
2736         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2737         if (!ret)
2738                 return NULL;
2739         return delegstateid(ret);
2740 }
2741
2742 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
2743 {
2744         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
2745                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
2746 }
2747
2748 static __be32
2749 nfs4_check_deleg(struct nfs4_client *cl, struct nfs4_file *fp, struct nfsd4_open *open,
2750                 struct nfs4_delegation **dp)
2751 {
2752         int flags;
2753         __be32 status = nfserr_bad_stateid;
2754
2755         *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2756         if (*dp == NULL)
2757                 goto out;
2758         flags = share_access_to_flags(open->op_share_access);
2759         status = nfs4_check_delegmode(*dp, flags);
2760         if (status)
2761                 *dp = NULL;
2762 out:
2763         if (!nfsd4_is_deleg_cur(open))
2764                 return nfs_ok;
2765         if (status)
2766                 return status;
2767         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2768         return nfs_ok;
2769 }
2770
2771 static __be32
2772 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2773 {
2774         struct nfs4_ol_stateid *local;
2775         struct nfs4_openowner *oo = open->op_openowner;
2776
2777         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2778                 /* ignore lock owners */
2779                 if (local->st_stateowner->so_is_open_owner == 0)
2780                         continue;
2781                 /* remember if we have seen this open owner */
2782                 if (local->st_stateowner == &oo->oo_owner)
2783                         *stpp = local;
2784                 /* check for conflicting share reservations */
2785                 if (!test_share(local, open))
2786                         return nfserr_share_denied;
2787         }
2788         return nfs_ok;
2789 }
2790
2791 static void nfs4_free_stateid(struct nfs4_ol_stateid *s)
2792 {
2793         kmem_cache_free(stateid_slab, s);
2794 }
2795
2796 static inline int nfs4_access_to_access(u32 nfs4_access)
2797 {
2798         int flags = 0;
2799
2800         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2801                 flags |= NFSD_MAY_READ;
2802         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2803                 flags |= NFSD_MAY_WRITE;
2804         return flags;
2805 }
2806
2807 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2808                 struct svc_fh *cur_fh, struct nfsd4_open *open)
2809 {
2810         __be32 status;
2811         int oflag = nfs4_access_to_omode(open->op_share_access);
2812         int access = nfs4_access_to_access(open->op_share_access);
2813
2814         if (!fp->fi_fds[oflag]) {
2815                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2816                         &fp->fi_fds[oflag]);
2817                 if (status)
2818                         return status;
2819         }
2820         nfs4_file_get_access(fp, oflag);
2821
2822         return nfs_ok;
2823 }
2824
2825 static inline __be32
2826 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2827                 struct nfsd4_open *open)
2828 {
2829         struct iattr iattr = {
2830                 .ia_valid = ATTR_SIZE,
2831                 .ia_size = 0,
2832         };
2833         if (!open->op_truncate)
2834                 return 0;
2835         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2836                 return nfserr_inval;
2837         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2838 }
2839
2840 static __be32
2841 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)
2842 {
2843         u32 op_share_access = open->op_share_access;
2844         bool new_access;
2845         __be32 status;
2846
2847         new_access = !test_access(op_share_access, stp);
2848         if (new_access) {
2849                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2850                 if (status)
2851                         return status;
2852         }
2853         status = nfsd4_truncate(rqstp, cur_fh, open);
2854         if (status) {
2855                 if (new_access) {
2856                         int oflag = nfs4_access_to_omode(op_share_access);
2857                         nfs4_file_put_access(fp, oflag);
2858                 }
2859                 return status;
2860         }
2861         /* remember the open */
2862         set_access(op_share_access, stp);
2863         set_deny(open->op_share_deny, stp);
2864
2865         return nfs_ok;
2866 }
2867
2868
2869 static void
2870 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
2871 {
2872         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2873 }
2874
2875 /* Should we give out recallable state?: */
2876 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2877 {
2878         if (clp->cl_cb_state == NFSD4_CB_UP)
2879                 return true;
2880         /*
2881          * In the sessions case, since we don't have to establish a
2882          * separate connection for callbacks, we assume it's OK
2883          * until we hear otherwise:
2884          */
2885         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2886 }
2887
2888 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2889 {
2890         struct file_lock *fl;
2891
2892         fl = locks_alloc_lock();
2893         if (!fl)
2894                 return NULL;
2895         locks_init_lock(fl);
2896         fl->fl_lmops = &nfsd_lease_mng_ops;
2897         fl->fl_flags = FL_LEASE;
2898         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2899         fl->fl_end = OFFSET_MAX;
2900         fl->fl_owner = (fl_owner_t)(dp->dl_file);
2901         fl->fl_pid = current->tgid;
2902         return fl;
2903 }
2904
2905 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2906 {
2907         struct nfs4_file *fp = dp->dl_file;
2908         struct file_lock *fl;
2909         int status;
2910
2911         fl = nfs4_alloc_init_lease(dp, flag);
2912         if (!fl)
2913                 return -ENOMEM;
2914         fl->fl_file = find_readable_file(fp);
2915         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2916         status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2917         if (status) {
2918                 list_del_init(&dp->dl_perclnt);
2919                 locks_free_lock(fl);
2920                 return -ENOMEM;
2921         }
2922         fp->fi_lease = fl;
2923         fp->fi_deleg_file = fl->fl_file;
2924         get_file(fp->fi_deleg_file);
2925         atomic_set(&fp->fi_delegees, 1);
2926         list_add(&dp->dl_perfile, &fp->fi_delegations);
2927         return 0;
2928 }
2929
2930 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2931 {
2932         struct nfs4_file *fp = dp->dl_file;
2933
2934         if (!fp->fi_lease)
2935                 return nfs4_setlease(dp, flag);
2936         spin_lock(&recall_lock);
2937         if (fp->fi_had_conflict) {
2938                 spin_unlock(&recall_lock);
2939                 return -EAGAIN;
2940         }
2941         atomic_inc(&fp->fi_delegees);
2942         list_add(&dp->dl_perfile, &fp->fi_delegations);
2943         spin_unlock(&recall_lock);
2944         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2945         return 0;
2946 }
2947
2948 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
2949 {
2950         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2951         if (status == -EAGAIN)
2952                 open->op_why_no_deleg = WND4_CONTENTION;
2953         else {
2954                 open->op_why_no_deleg = WND4_RESOURCE;
2955                 switch (open->op_deleg_want) {
2956                 case NFS4_SHARE_WANT_READ_DELEG:
2957                 case NFS4_SHARE_WANT_WRITE_DELEG:
2958                 case NFS4_SHARE_WANT_ANY_DELEG:
2959                         break;
2960                 case NFS4_SHARE_WANT_CANCEL:
2961                         open->op_why_no_deleg = WND4_CANCELLED;
2962                         break;
2963                 case NFS4_SHARE_WANT_NO_DELEG:
2964                         BUG();  /* not supposed to get here */
2965                 }
2966         }
2967 }
2968
2969 /*
2970  * Attempt to hand out a delegation.
2971  */
2972 static void
2973 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
2974 {
2975         struct nfs4_delegation *dp;
2976         struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
2977         int cb_up;
2978         int status = 0, flag = 0;
2979
2980         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
2981         flag = NFS4_OPEN_DELEGATE_NONE;
2982         open->op_recall = 0;
2983         switch (open->op_claim_type) {
2984                 case NFS4_OPEN_CLAIM_PREVIOUS:
2985                         if (!cb_up)
2986                                 open->op_recall = 1;
2987                         flag = open->op_delegate_type;
2988                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2989                                 goto out;
2990                         break;
2991                 case NFS4_OPEN_CLAIM_NULL:
2992                         /* Let's not give out any delegations till everyone's
2993                          * had the chance to reclaim theirs.... */
2994                         if (locks_in_grace())
2995                                 goto out;
2996                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
2997                                 goto out;
2998                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2999                                 flag = NFS4_OPEN_DELEGATE_WRITE;
3000                         else
3001                                 flag = NFS4_OPEN_DELEGATE_READ;
3002                         break;
3003                 default:
3004                         goto out;
3005         }
3006
3007         dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh, flag);
3008         if (dp == NULL)
3009                 goto out_no_deleg;
3010         status = nfs4_set_delegation(dp, flag);
3011         if (status)
3012                 goto out_free;
3013
3014         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
3015
3016         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
3017                 STATEID_VAL(&dp->dl_stid.sc_stateid));
3018 out:
3019         open->op_delegate_type = flag;
3020         if (flag == NFS4_OPEN_DELEGATE_NONE) {
3021                 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
3022                     open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
3023                         dprintk("NFSD: WARNING: refusing delegation reclaim\n");
3024
3025                 /* 4.1 client asking for a delegation? */
3026                 if (open->op_deleg_want)
3027                         nfsd4_open_deleg_none_ext(open, status);
3028         }
3029         return;
3030 out_free:
3031         nfs4_put_delegation(dp);
3032 out_no_deleg:
3033         flag = NFS4_OPEN_DELEGATE_NONE;
3034         goto out;
3035 }
3036
3037 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
3038                                         struct nfs4_delegation *dp)
3039 {
3040         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
3041             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3042                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3043                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
3044         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
3045                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3046                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3047                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
3048         }
3049         /* Otherwise the client must be confused wanting a delegation
3050          * it already has, therefore we don't return
3051          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
3052          */
3053 }
3054
3055 /*
3056  * called with nfs4_lock_state() held.
3057  */
3058 __be32
3059 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
3060 {
3061         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3062         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
3063         struct nfs4_file *fp = NULL;
3064         struct inode *ino = current_fh->fh_dentry->d_inode;
3065         struct nfs4_ol_stateid *stp = NULL;
3066         struct nfs4_delegation *dp = NULL;
3067         __be32 status;
3068
3069         /*
3070          * Lookup file; if found, lookup stateid and check open request,
3071          * and check for delegations in the process of being recalled.
3072          * If not found, create the nfs4_file struct
3073          */
3074         fp = find_file(ino);
3075         if (fp) {
3076                 if ((status = nfs4_check_open(fp, open, &stp)))
3077                         goto out;
3078                 status = nfs4_check_deleg(cl, fp, open, &dp);
3079                 if (status)
3080                         goto out;
3081         } else {
3082                 status = nfserr_bad_stateid;
3083                 if (nfsd4_is_deleg_cur(open))
3084                         goto out;
3085                 status = nfserr_jukebox;
3086                 fp = open->op_file;
3087                 open->op_file = NULL;
3088                 nfsd4_init_file(fp, ino);
3089         }
3090
3091         /*
3092          * OPEN the file, or upgrade an existing OPEN.
3093          * If truncate fails, the OPEN fails.
3094          */
3095         if (stp) {
3096                 /* Stateid was found, this is an OPEN upgrade */
3097                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
3098                 if (status)
3099                         goto out;
3100         } else {
3101                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, open);
3102                 if (status)
3103                         goto out;
3104                 stp = open->op_stp;
3105                 open->op_stp = NULL;
3106                 init_open_stateid(stp, fp, open);
3107                 status = nfsd4_truncate(rqstp, current_fh, open);
3108                 if (status) {
3109                         release_open_stateid(stp);
3110                         goto out;
3111                 }
3112         }
3113         update_stateid(&stp->st_stid.sc_stateid);
3114         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3115
3116         if (nfsd4_has_session(&resp->cstate)) {
3117                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3118
3119                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
3120                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3121                         open->op_why_no_deleg = WND4_NOT_WANTED;
3122                         goto nodeleg;
3123                 }
3124         }
3125
3126         /*
3127         * Attempt to hand out a delegation. No error return, because the
3128         * OPEN succeeds even if we fail.
3129         */
3130         nfs4_open_delegation(current_fh, open, stp);
3131 nodeleg:
3132         status = nfs_ok;
3133
3134         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
3135                 STATEID_VAL(&stp->st_stid.sc_stateid));
3136 out:
3137         /* 4.1 client trying to upgrade/downgrade delegation? */
3138         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
3139             open->op_deleg_want)
3140                 nfsd4_deleg_xgrade_none_ext(open, dp);
3141
3142         if (fp)
3143                 put_nfs4_file(fp);
3144         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
3145                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
3146         /*
3147         * To finish the open response, we just need to set the rflags.
3148         */
3149         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
3150         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
3151             !nfsd4_has_session(&resp->cstate))
3152                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
3153
3154         return status;
3155 }
3156
3157 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
3158 {
3159         if (open->op_openowner) {
3160                 struct nfs4_openowner *oo = open->op_openowner;
3161
3162                 if (!list_empty(&oo->oo_owner.so_stateids))
3163                         list_del_init(&oo->oo_close_lru);
3164                 if (oo->oo_flags & NFS4_OO_NEW) {
3165                         if (status) {
3166                                 release_openowner(oo);
3167                                 open->op_openowner = NULL;
3168                         } else
3169                                 oo->oo_flags &= ~NFS4_OO_NEW;
3170                 }
3171         }
3172         if (open->op_file)
3173                 nfsd4_free_file(open->op_file);
3174         if (open->op_stp)
3175                 nfs4_free_stateid(open->op_stp);
3176 }
3177
3178 __be32
3179 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3180             clientid_t *clid)
3181 {
3182         struct nfs4_client *clp;
3183         __be32 status;
3184
3185         nfs4_lock_state();
3186         dprintk("process_renew(%08x/%08x): starting\n",
3187                         clid->cl_boot, clid->cl_id);
3188         status = nfserr_stale_clientid;
3189         if (STALE_CLIENTID(clid))
3190                 goto out;
3191         clp = find_confirmed_client(clid);
3192         status = nfserr_expired;
3193         if (clp == NULL) {
3194                 /* We assume the client took too long to RENEW. */
3195                 dprintk("nfsd4_renew: clientid not found!\n");
3196                 goto out;
3197         }
3198         status = nfserr_cb_path_down;
3199         if (!list_empty(&clp->cl_delegations)
3200                         && clp->cl_cb_state != NFSD4_CB_UP)
3201                 goto out;
3202         status = nfs_ok;
3203 out:
3204         nfs4_unlock_state();
3205         return status;
3206 }
3207
3208 static struct lock_manager nfsd4_manager = {
3209 };
3210
3211 static bool grace_ended;
3212
3213 static void
3214 nfsd4_end_grace(void)
3215 {
3216         /* do nothing if grace period already ended */
3217         if (grace_ended)
3218                 return;
3219
3220         dprintk("NFSD: end of grace period\n");
3221         grace_ended = true;
3222         nfsd4_record_grace_done(&init_net, boot_time);
3223         locks_end_grace(&nfsd4_manager);
3224         /*
3225          * Now that every NFSv4 client has had the chance to recover and
3226          * to see the (possibly new, possibly shorter) lease time, we
3227          * can safely set the next grace time to the current lease time:
3228          */
3229         nfsd4_grace = nfsd4_lease;
3230 }
3231
3232 static time_t
3233 nfs4_laundromat(void)
3234 {
3235         struct nfs4_client *clp;
3236         struct nfs4_openowner *oo;
3237         struct nfs4_delegation *dp;
3238         struct list_head *pos, *next, reaplist;
3239         time_t cutoff = get_seconds() - nfsd4_lease;
3240         time_t t, clientid_val = nfsd4_lease;
3241         time_t u, test_val = nfsd4_lease;
3242
3243         nfs4_lock_state();
3244
3245         dprintk("NFSD: laundromat service - starting\n");
3246         nfsd4_end_grace();
3247         INIT_LIST_HEAD(&reaplist);
3248         spin_lock(&client_lock);
3249         list_for_each_safe(pos, next, &client_lru) {
3250                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3251                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3252                         t = clp->cl_time - cutoff;
3253                         if (clientid_val > t)
3254                                 clientid_val = t;
3255                         break;
3256                 }
3257                 if (atomic_read(&clp->cl_refcount)) {
3258                         dprintk("NFSD: client in use (clientid %08x)\n",
3259                                 clp->cl_clientid.cl_id);
3260                         continue;
3261                 }
3262                 unhash_client_locked(clp);
3263                 list_add(&clp->cl_lru, &reaplist);
3264         }
3265         spin_unlock(&client_lock);
3266         list_for_each_safe(pos, next, &reaplist) {
3267                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3268                 dprintk("NFSD: purging unused client (clientid %08x)\n",
3269                         clp->cl_clientid.cl_id);
3270                 nfsd4_client_record_remove(clp);
3271                 expire_client(clp);
3272         }
3273         spin_lock(&recall_lock);
3274         list_for_each_safe(pos, next, &del_recall_lru) {
3275                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3276                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3277                         u = dp->dl_time - cutoff;
3278                         if (test_val > u)
3279                                 test_val = u;
3280                         break;
3281                 }
3282                 list_move(&dp->dl_recall_lru, &reaplist);
3283         }
3284         spin_unlock(&recall_lock);
3285         list_for_each_safe(pos, next, &reaplist) {
3286                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3287                 unhash_delegation(dp);
3288         }
3289         test_val = nfsd4_lease;
3290         list_for_each_safe(pos, next, &close_lru) {
3291                 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3292                 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3293                         u = oo->oo_time - cutoff;
3294                         if (test_val > u)
3295                                 test_val = u;
3296                         break;
3297                 }
3298                 release_openowner(oo);
3299         }
3300         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3301                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3302         nfs4_unlock_state();
3303         return clientid_val;
3304 }
3305
3306 static struct workqueue_struct *laundry_wq;
3307 static void laundromat_main(struct work_struct *);
3308 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
3309
3310 static void
3311 laundromat_main(struct work_struct *not_used)
3312 {
3313         time_t t;
3314
3315         t = nfs4_laundromat();
3316         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3317         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
3318 }
3319
3320 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3321 {
3322         if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3323                 return nfserr_bad_stateid;
3324         return nfs_ok;
3325 }
3326
3327 static int
3328 STALE_STATEID(stateid_t *stateid)
3329 {
3330         if (stateid->si_opaque.so_clid.cl_boot == boot_time)
3331                 return 0;
3332         dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
3333                 STATEID_VAL(stateid));
3334         return 1;
3335 }
3336
3337 static inline int
3338 access_permit_read(struct nfs4_ol_stateid *stp)
3339 {
3340         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
3341                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
3342                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
3343 }
3344
3345 static inline int
3346 access_permit_write(struct nfs4_ol_stateid *stp)
3347 {
3348         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
3349                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
3350 }
3351
3352 static
3353 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3354 {
3355         __be32 status = nfserr_openmode;
3356
3357         /* For lock stateid's, we test the parent open, not the lock: */
3358         if (stp->st_openstp)
3359                 stp = stp->st_openstp;
3360         if ((flags & WR_STATE) && !access_permit_write(stp))
3361                 goto out;
3362         if ((flags & RD_STATE) && !access_permit_read(stp))
3363                 goto out;
3364         status = nfs_ok;
3365 out:
3366         return status;
3367 }
3368
3369 static inline __be32
3370 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
3371 {
3372         if (ONE_STATEID(stateid) && (flags & RD_STATE))
3373                 return nfs_ok;
3374         else if (locks_in_grace()) {
3375                 /* Answer in remaining cases depends on existence of
3376                  * conflicting state; so we must wait out the grace period. */
3377                 return nfserr_grace;
3378         } else if (flags & WR_STATE)
3379                 return nfs4_share_conflict(current_fh,
3380                                 NFS4_SHARE_DENY_WRITE);
3381         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3382                 return nfs4_share_conflict(current_fh,
3383                                 NFS4_SHARE_DENY_READ);
3384 }
3385
3386 /*
3387  * Allow READ/WRITE during grace period on recovered state only for files
3388  * that are not able to provide mandatory locking.
3389  */
3390 static inline int
3391 grace_disallows_io(struct inode *inode)
3392 {
3393         return locks_in_grace() && mandatory_lock(inode);
3394 }
3395
3396 /* Returns true iff a is later than b: */
3397 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3398 {
3399         return (s32)a->si_generation - (s32)b->si_generation > 0;
3400 }
3401
3402 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3403 {
3404         /*
3405          * When sessions are used the stateid generation number is ignored
3406          * when it is zero.
3407          */
3408         if (has_session && in->si_generation == 0)
3409                 return nfs_ok;
3410
3411         if (in->si_generation == ref->si_generation)
3412                 return nfs_ok;
3413
3414         /* If the client sends us a stateid from the future, it's buggy: */
3415         if (stateid_generation_after(in, ref))
3416                 return nfserr_bad_stateid;
3417         /*
3418          * However, we could see a stateid from the past, even from a
3419          * non-buggy client.  For example, if the client sends a lock
3420          * while some IO is outstanding, the lock may bump si_generation
3421          * while the IO is still in flight.  The client could avoid that
3422          * situation by waiting for responses on all the IO requests,
3423          * but better performance may result in retrying IO that
3424          * receives an old_stateid error if requests are rarely
3425          * reordered in flight:
3426          */
3427         return nfserr_old_stateid;
3428 }
3429
3430 __be32 nfs4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
3431 {
3432         struct nfs4_stid *s;
3433         struct nfs4_ol_stateid *ols;
3434         __be32 status;
3435
3436         if (STALE_STATEID(stateid))
3437                 return nfserr_stale_stateid;
3438
3439         s = find_stateid(cl, stateid);
3440         if (!s)
3441                  return nfserr_stale_stateid;
3442         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3443         if (status)
3444                 return status;
3445         if (!(s->sc_type & (NFS4_OPEN_STID | NFS4_LOCK_STID)))
3446                 return nfs_ok;
3447         ols = openlockstateid(s);
3448         if (ols->st_stateowner->so_is_open_owner
3449             && !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3450                 return nfserr_bad_stateid;
3451         return nfs_ok;
3452 }
3453
3454 static __be32 nfsd4_lookup_stateid(stateid_t *stateid, unsigned char typemask, struct nfs4_stid **s)
3455 {
3456         struct nfs4_client *cl;
3457
3458         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3459                 return nfserr_bad_stateid;
3460         if (STALE_STATEID(stateid))
3461                 return nfserr_stale_stateid;
3462         cl = find_confirmed_client(&stateid->si_opaque.so_clid);
3463         if (!cl)
3464                 return nfserr_expired;
3465         *s = find_stateid_by_type(cl, stateid, typemask);
3466         if (!*s)
3467                 return nfserr_bad_stateid;
3468         return nfs_ok;
3469
3470 }
3471
3472 /*
3473 * Checks for stateid operations
3474 */
3475 __be32
3476 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
3477                            stateid_t *stateid, int flags, struct file **filpp)
3478 {
3479         struct nfs4_stid *s;
3480         struct nfs4_ol_stateid *stp = NULL;
3481         struct nfs4_delegation *dp = NULL;
3482         struct svc_fh *current_fh = &cstate->current_fh;
3483         struct inode *ino = current_fh->fh_dentry->d_inode;
3484         __be32 status;
3485
3486         if (filpp)
3487                 *filpp = NULL;
3488
3489         if (grace_disallows_io(ino))
3490                 return nfserr_grace;
3491
3492         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3493                 return check_special_stateids(current_fh, stateid, flags);
3494
3495         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID, &s);
3496         if (status)
3497                 return status;
3498         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
3499         if (status)
3500                 goto out;
3501         switch (s->sc_type) {
3502         case NFS4_DELEG_STID:
3503                 dp = delegstateid(s);
3504                 status = nfs4_check_delegmode(dp, flags);
3505                 if (status)
3506                         goto out;
3507                 if (filpp) {
3508                         *filpp = dp->dl_file->fi_deleg_file;
3509                         BUG_ON(!*filpp);
3510                 }
3511                 break;
3512         case NFS4_OPEN_STID:
3513         case NFS4_LOCK_STID:
3514                 stp = openlockstateid(s);
3515                 status = nfs4_check_fh(current_fh, stp);
3516                 if (status)
3517                         goto out;
3518                 if (stp->st_stateowner->so_is_open_owner
3519                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3520                         goto out;
3521                 status = nfs4_check_openmode(stp, flags);
3522                 if (status)
3523                         goto out;
3524                 if (filpp) {
3525                         if (flags & RD_STATE)
3526                                 *filpp = find_readable_file(stp->st_file);
3527                         else
3528                                 *filpp = find_writeable_file(stp->st_file);
3529                 }
3530                 break;
3531         default:
3532                 return nfserr_bad_stateid;
3533         }
3534         status = nfs_ok;
3535 out:
3536         return status;
3537 }
3538
3539 static __be32
3540 nfsd4_free_lock_stateid(struct nfs4_ol_stateid *stp)
3541 {
3542         if (check_for_locks(stp->st_file, lockowner(stp->st_stateowner)))
3543                 return nfserr_locks_held;
3544         release_lock_stateid(stp);
3545         return nfs_ok;
3546 }
3547
3548 /*
3549  * Test if the stateid is valid
3550  */
3551 __be32
3552 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3553                    struct nfsd4_test_stateid *test_stateid)
3554 {
3555         struct nfsd4_test_stateid_id *stateid;
3556         struct nfs4_client *cl = cstate->session->se_client;
3557
3558         nfs4_lock_state();
3559         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
3560                 stateid->ts_id_status = nfs4_validate_stateid(cl, &stateid->ts_id_stateid);
3561         nfs4_unlock_state();
3562
3563         return nfs_ok;
3564 }
3565
3566 __be32
3567 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3568                    struct nfsd4_free_stateid *free_stateid)
3569 {
3570         stateid_t *stateid = &free_stateid->fr_stateid;
3571         struct nfs4_stid *s;
3572         struct nfs4_client *cl = cstate->session->se_client;
3573         __be32 ret = nfserr_bad_stateid;
3574
3575         nfs4_lock_state();
3576         s = find_stateid(cl, stateid);
3577         if (!s)
3578                 goto out;
3579         switch (s->sc_type) {
3580         case NFS4_DELEG_STID:
3581                 ret = nfserr_locks_held;
3582                 goto out;
3583         case NFS4_OPEN_STID:
3584         case NFS4_LOCK_STID:
3585                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
3586                 if (ret)
3587                         goto out;
3588                 if (s->sc_type == NFS4_LOCK_STID)
3589                         ret = nfsd4_free_lock_stateid(openlockstateid(s));
3590                 else
3591                         ret = nfserr_locks_held;
3592                 break;
3593         default:
3594                 ret = nfserr_bad_stateid;
3595         }
3596 out:
3597         nfs4_unlock_state();
3598         return ret;
3599 }
3600
3601 static inline int
3602 setlkflg (int type)
3603 {
3604         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3605                 RD_STATE : WR_STATE;
3606 }
3607
3608 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
3609 {
3610         struct svc_fh *current_fh = &cstate->current_fh;
3611         struct nfs4_stateowner *sop = stp->st_stateowner;
3612         __be32 status;
3613
3614         status = nfsd4_check_seqid(cstate, sop, seqid);
3615         if (status)
3616                 return status;
3617         if (stp->st_stid.sc_type == NFS4_CLOSED_STID)
3618                 /*
3619                  * "Closed" stateid's exist *only* to return
3620                  * nfserr_replay_me from the previous step.
3621                  */
3622                 return nfserr_bad_stateid;
3623         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
3624         if (status)
3625                 return status;
3626         return nfs4_check_fh(current_fh, stp);
3627 }
3628
3629 /*
3630  * Checks for sequence id mutating operations.
3631  */
3632 static __be32
3633 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3634                          stateid_t *stateid, char typemask,
3635                          struct nfs4_ol_stateid **stpp)
3636 {
3637         __be32 status;
3638         struct nfs4_stid *s;
3639
3640         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3641                 seqid, STATEID_VAL(stateid));
3642
3643         *stpp = NULL;
3644         status = nfsd4_lookup_stateid(stateid, typemask, &s);
3645         if (status)
3646                 return status;
3647         *stpp = openlockstateid(s);
3648         cstate->replay_owner = (*stpp)->st_stateowner;
3649
3650         return nfs4_seqid_op_checks(cstate, stateid, seqid, *stpp);
3651 }
3652
3653 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid, stateid_t *stateid, struct nfs4_ol_stateid **stpp)
3654 {
3655         __be32 status;
3656         struct nfs4_openowner *oo;
3657
3658         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
3659                                                 NFS4_OPEN_STID, stpp);
3660         if (status)
3661                 return status;
3662         oo = openowner((*stpp)->st_stateowner);
3663         if (!(oo->oo_flags & NFS4_OO_CONFIRMED))
3664                 return nfserr_bad_stateid;
3665         return nfs_ok;
3666 }
3667
3668 __be32
3669 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3670                    struct nfsd4_open_confirm *oc)
3671 {
3672         __be32 status;
3673         struct nfs4_openowner *oo;
3674         struct nfs4_ol_stateid *stp;
3675
3676         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3677                         (int)cstate->current_fh.fh_dentry->d_name.len,
3678                         cstate->current_fh.fh_dentry->d_name.name);
3679
3680         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3681         if (status)
3682                 return status;
3683
3684         nfs4_lock_state();
3685
3686         status = nfs4_preprocess_seqid_op(cstate,
3687                                         oc->oc_seqid, &oc->oc_req_stateid,
3688                                         NFS4_OPEN_STID, &stp);
3689         if (status)
3690                 goto out;
3691         oo = openowner(stp->st_stateowner);
3692         status = nfserr_bad_stateid;
3693         if (oo->oo_flags & NFS4_OO_CONFIRMED)
3694                 goto out;
3695         oo->oo_flags |= NFS4_OO_CONFIRMED;
3696         update_stateid(&stp->st_stid.sc_stateid);
3697         memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3698         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3699                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
3700
3701         nfsd4_client_record_create(oo->oo_owner.so_client);
3702         status = nfs_ok;
3703 out:
3704         if (!cstate->replay_owner)
3705                 nfs4_unlock_state();
3706         return status;
3707 }
3708
3709 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
3710 {
3711         if (!test_access(access, stp))
3712                 return;
3713         nfs4_file_put_access(stp->st_file, nfs4_access_to_omode(access));
3714         clear_access(access, stp);
3715 }
3716
3717 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
3718 {
3719         switch (to_access) {
3720         case NFS4_SHARE_ACCESS_READ:
3721                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
3722                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3723                 break;
3724         case NFS4_SHARE_ACCESS_WRITE:
3725                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
3726                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3727                 break;
3728         case NFS4_SHARE_ACCESS_BOTH:
3729                 break;
3730         default:
3731                 BUG();
3732         }
3733 }
3734
3735 static void
3736 reset_union_bmap_deny(unsigned long deny, struct nfs4_ol_stateid *stp)
3737 {
3738         int i;
3739         for (i = 0; i < 4; i++) {
3740                 if ((i & deny) != i)
3741                         clear_deny(i, stp);
3742         }
3743 }
3744
3745 __be32
3746 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3747                      struct nfsd4_compound_state *cstate,
3748                      struct nfsd4_open_downgrade *od)
3749 {
3750         __be32 status;
3751         struct nfs4_ol_stateid *stp;
3752
3753         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3754                         (int)cstate->current_fh.fh_dentry->d_name.len,
3755                         cstate->current_fh.fh_dentry->d_name.name);
3756
3757         /* We don't yet support WANT bits: */
3758         if (od->od_deleg_want)
3759                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
3760                         od->od_deleg_want);
3761
3762         nfs4_lock_state();
3763         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
3764                                         &od->od_stateid, &stp);
3765         if (status)
3766                 goto out;
3767         status = nfserr_inval;
3768         if (!test_access(od->od_share_access, stp)) {
3769                 dprintk("NFSD: access not a subset current bitmap: 0x%lx, input access=%08x\n",
3770                         stp->st_access_bmap, od->od_share_access);
3771                 goto out;
3772         }
3773         if (!test_deny(od->od_share_deny, stp)) {
3774                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3775                         stp->st_deny_bmap, od->od_share_deny);
3776                 goto out;
3777         }
3778         nfs4_stateid_downgrade(stp, od->od_share_access);
3779
3780         reset_union_bmap_deny(od->od_share_deny, stp);
3781
3782         update_stateid(&stp->st_stid.sc_stateid);
3783         memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3784         status = nfs_ok;
3785 out:
3786         if (!cstate->replay_owner)
3787                 nfs4_unlock_state();
3788         return status;
3789 }
3790
3791 void nfsd4_purge_closed_stateid(struct nfs4_stateowner *so)
3792 {
3793         struct nfs4_openowner *oo;
3794         struct nfs4_ol_stateid *s;
3795
3796         if (!so->so_is_open_owner)
3797                 return;
3798         oo = openowner(so);
3799         s = oo->oo_last_closed_stid;
3800         if (!s)
3801                 return;
3802         if (!(oo->oo_flags & NFS4_OO_PURGE_CLOSE)) {
3803                 /* Release the last_closed_stid on the next seqid bump: */
3804                 oo->oo_flags |= NFS4_OO_PURGE_CLOSE;
3805                 return;
3806         }
3807         oo->oo_flags &= ~NFS4_OO_PURGE_CLOSE;
3808         release_last_closed_stateid(oo);
3809 }
3810
3811 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
3812 {
3813         unhash_open_stateid(s);
3814         s->st_stid.sc_type = NFS4_CLOSED_STID;
3815 }
3816
3817 /*
3818  * nfs4_unlock_state() called after encode
3819  */
3820 __be32
3821 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3822             struct nfsd4_close *close)
3823 {
3824         __be32 status;
3825         struct nfs4_openowner *oo;
3826         struct nfs4_ol_stateid *stp;
3827
3828         dprintk("NFSD: nfsd4_close on file %.*s\n",
3829                         (int)cstate->current_fh.fh_dentry->d_name.len,
3830                         cstate->current_fh.fh_dentry->d_name.name);
3831
3832         nfs4_lock_state();
3833         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
3834                                         &close->cl_stateid,
3835                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
3836                                         &stp);
3837         if (status)
3838                 goto out;
3839         oo = openowner(stp->st_stateowner);
3840         status = nfs_ok;
3841         update_stateid(&stp->st_stid.sc_stateid);
3842         memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3843
3844         nfsd4_close_open_stateid(stp);
3845         oo->oo_last_closed_stid = stp;
3846
3847         /* place unused nfs4_stateowners on so_close_lru list to be
3848          * released by the laundromat service after the lease period
3849          * to enable us to handle CLOSE replay
3850          */
3851         if (list_empty(&oo->oo_owner.so_stateids))
3852                 move_to_close_lru(oo);
3853 out:
3854         if (!cstate->replay_owner)
3855                 nfs4_unlock_state();
3856         return status;
3857 }
3858
3859 __be32
3860 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3861                   struct nfsd4_delegreturn *dr)
3862 {
3863         struct nfs4_delegation *dp;
3864         stateid_t *stateid = &dr->dr_stateid;
3865         struct nfs4_stid *s;
3866         struct inode *inode;
3867         __be32 status;
3868
3869         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3870                 return status;
3871         inode = cstate->current_fh.fh_dentry->d_inode;
3872
3873         nfs4_lock_state();
3874         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID, &s);
3875         if (status)
3876                 goto out;
3877         dp = delegstateid(s);
3878         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
3879         if (status)
3880                 goto out;
3881
3882         unhash_delegation(dp);
3883 out:
3884         nfs4_unlock_state();
3885
3886         return status;
3887 }
3888
3889
3890 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3891
3892 #define LOCKOWNER_INO_HASH_BITS 8
3893 #define LOCKOWNER_INO_HASH_SIZE (1 << LOCKOWNER_INO_HASH_BITS)
3894 #define LOCKOWNER_INO_HASH_MASK (LOCKOWNER_INO_HASH_SIZE - 1)
3895
3896 static inline u64
3897 end_offset(u64 start, u64 len)
3898 {
3899         u64 end;
3900
3901         end = start + len;
3902         return end >= start ? end: NFS4_MAX_UINT64;
3903 }
3904
3905 /* last octet in a range */
3906 static inline u64
3907 last_byte_offset(u64 start, u64 len)
3908 {
3909         u64 end;
3910
3911         BUG_ON(!len);
3912         end = start + len;
3913         return end > start ? end - 1: NFS4_MAX_UINT64;
3914 }
3915
3916 static unsigned int lockowner_ino_hashval(struct inode *inode, u32 cl_id, struct xdr_netobj *ownername)
3917 {
3918         return (file_hashval(inode) + cl_id
3919                         + opaque_hashval(ownername->data, ownername->len))
3920                 & LOCKOWNER_INO_HASH_MASK;
3921 }
3922
3923 static struct list_head lockowner_ino_hashtbl[LOCKOWNER_INO_HASH_SIZE];
3924
3925 /*
3926  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3927  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3928  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3929  * locking, this prevents us from being completely protocol-compliant.  The
3930  * real solution to this problem is to start using unsigned file offsets in
3931  * the VFS, but this is a very deep change!
3932  */
3933 static inline void
3934 nfs4_transform_lock_offset(struct file_lock *lock)
3935 {
3936         if (lock->fl_start < 0)
3937                 lock->fl_start = OFFSET_MAX;
3938         if (lock->fl_end < 0)
3939                 lock->fl_end = OFFSET_MAX;
3940 }
3941
3942 /* Hack!: For now, we're defining this just so we can use a pointer to it
3943  * as a unique cookie to identify our (NFSv4's) posix locks. */
3944 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
3945 };
3946
3947 static inline void
3948 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3949 {
3950         struct nfs4_lockowner *lo;
3951
3952         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3953                 lo = (struct nfs4_lockowner *) fl->fl_owner;
3954                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
3955                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
3956                 if (!deny->ld_owner.data)
3957                         /* We just don't care that much */
3958                         goto nevermind;
3959                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
3960                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
3961         } else {
3962 nevermind:
3963                 deny->ld_owner.len = 0;
3964                 deny->ld_owner.data = NULL;
3965                 deny->ld_clientid.cl_boot = 0;
3966                 deny->ld_clientid.cl_id = 0;
3967         }
3968         deny->ld_start = fl->fl_start;
3969         deny->ld_length = NFS4_MAX_UINT64;
3970         if (fl->fl_end != NFS4_MAX_UINT64)
3971                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3972         deny->ld_type = NFS4_READ_LT;
3973         if (fl->fl_type != F_RDLCK)
3974                 deny->ld_type = NFS4_WRITE_LT;
3975 }
3976
3977 static bool same_lockowner_ino(struct nfs4_lockowner *lo, struct inode *inode, clientid_t *clid, struct xdr_netobj *owner)
3978 {
3979         struct nfs4_ol_stateid *lst;
3980
3981         if (!same_owner_str(&lo->lo_owner, owner, clid))
3982                 return false;
3983         lst = list_first_entry(&lo->lo_owner.so_stateids,
3984                                struct nfs4_ol_stateid, st_perstateowner);
3985         return lst->st_file->fi_inode == inode;
3986 }
3987
3988 static struct nfs4_lockowner *
3989 find_lockowner_str(struct inode *inode, clientid_t *clid,
3990                 struct xdr_netobj *owner)
3991 {
3992         unsigned int hashval = lockowner_ino_hashval(inode, clid->cl_id, owner);
3993         struct nfs4_lockowner *lo;
3994
3995         list_for_each_entry(lo, &lockowner_ino_hashtbl[hashval], lo_owner_ino_hash) {
3996                 if (same_lockowner_ino(lo, inode, clid, owner))
3997                         return lo;
3998         }
3999         return NULL;
4000 }
4001
4002 static void hash_lockowner(struct nfs4_lockowner *lo, unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp)
4003 {
4004         struct inode *inode = open_stp->st_file->fi_inode;
4005         unsigned int inohash = lockowner_ino_hashval(inode,
4006                         clp->cl_clientid.cl_id, &lo->lo_owner.so_owner);
4007
4008         list_add(&lo->lo_owner.so_strhash, &ownerstr_hashtbl[strhashval]);
4009         list_add(&lo->lo_owner_ino_hash, &lockowner_ino_hashtbl[inohash]);
4010         list_add(&lo->lo_perstateid, &open_stp->st_lockowners);
4011 }
4012
4013 /*
4014  * Alloc a lock owner structure.
4015  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
4016  * occurred.
4017  *
4018  * strhashval = ownerstr_hashval
4019  */
4020
4021 static struct nfs4_lockowner *
4022 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp, struct nfsd4_lock *lock) {
4023         struct nfs4_lockowner *lo;
4024
4025         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
4026         if (!lo)
4027                 return NULL;
4028         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
4029         lo->lo_owner.so_is_open_owner = 0;
4030         /* It is the openowner seqid that will be incremented in encode in the
4031          * case of new lockowners; so increment the lock seqid manually: */
4032         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid + 1;
4033         hash_lockowner(lo, strhashval, clp, open_stp);
4034         return lo;
4035 }
4036
4037 static struct nfs4_ol_stateid *
4038 alloc_init_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp, struct nfs4_ol_stateid *open_stp)
4039 {
4040         struct nfs4_ol_stateid *stp;
4041         struct nfs4_client *clp = lo->lo_owner.so_client;
4042
4043         stp = nfs4_alloc_stateid(clp);
4044         if (stp == NULL)
4045                 return NULL;
4046         init_stid(&stp->st_stid, clp, NFS4_LOCK_STID);
4047         list_add(&stp->st_perfile, &fp->fi_stateids);
4048         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
4049         stp->st_stateowner = &lo->lo_owner;
4050         get_nfs4_file(fp);
4051         stp->st_file = fp;
4052         stp->st_access_bmap = 0;
4053         stp->st_deny_bmap = open_stp->st_deny_bmap;
4054         stp->st_openstp = open_stp;
4055         return stp;
4056 }
4057
4058 static int
4059 check_lock_length(u64 offset, u64 length)
4060 {
4061         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
4062              LOFF_OVERFLOW(offset, length)));
4063 }
4064
4065 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
4066 {
4067         struct nfs4_file *fp = lock_stp->st_file;
4068         int oflag = nfs4_access_to_omode(access);
4069
4070         if (test_access(access, lock_stp))
4071                 return;
4072         nfs4_file_get_access(fp, oflag);
4073         set_access(access, lock_stp);
4074 }
4075
4076 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)
4077 {
4078         struct nfs4_file *fi = ost->st_file;
4079         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
4080         struct nfs4_client *cl = oo->oo_owner.so_client;
4081         struct nfs4_lockowner *lo;
4082         unsigned int strhashval;
4083
4084         lo = find_lockowner_str(fi->fi_inode, &cl->cl_clientid, &lock->v.new.owner);
4085         if (lo) {
4086                 if (!cstate->minorversion)
4087                         return nfserr_bad_seqid;
4088                 /* XXX: a lockowner always has exactly one stateid: */
4089                 *lst = list_first_entry(&lo->lo_owner.so_stateids,
4090                                 struct nfs4_ol_stateid, st_perstateowner);
4091                 return nfs_ok;
4092         }
4093         strhashval = ownerstr_hashval(cl->cl_clientid.cl_id,
4094                         &lock->v.new.owner);
4095         lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
4096         if (lo == NULL)
4097                 return nfserr_jukebox;
4098         *lst = alloc_init_lock_stateid(lo, fi, ost);
4099         if (*lst == NULL) {
4100                 release_lockowner(lo);
4101                 return nfserr_jukebox;
4102         }
4103         *new = true;
4104         return nfs_ok;
4105 }
4106
4107 /*
4108  *  LOCK operation
4109  */
4110 __be32
4111 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4112            struct nfsd4_lock *lock)
4113 {
4114         struct nfs4_openowner *open_sop = NULL;
4115         struct nfs4_lockowner *lock_sop = NULL;
4116         struct nfs4_ol_stateid *lock_stp;
4117         struct file *filp = NULL;
4118         struct file_lock file_lock;
4119         struct file_lock conflock;
4120         __be32 status = 0;
4121         bool new_state = false;
4122         int lkflg;
4123         int err;
4124
4125         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
4126                 (long long) lock->lk_offset,
4127                 (long long) lock->lk_length);
4128
4129         if (check_lock_length(lock->lk_offset, lock->lk_length))
4130                  return nfserr_inval;
4131
4132         if ((status = fh_verify(rqstp, &cstate->current_fh,
4133                                 S_IFREG, NFSD_MAY_LOCK))) {
4134                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
4135                 return status;
4136         }
4137
4138         nfs4_lock_state();
4139
4140         if (lock->lk_is_new) {
4141                 /*
4142                  * Client indicates that this is a new lockowner.
4143                  * Use open owner and open stateid to create lock owner and
4144                  * lock stateid.
4145                  */
4146                 struct nfs4_ol_stateid *open_stp = NULL;
4147
4148                 if (nfsd4_has_session(cstate))
4149                         /* See rfc 5661 18.10.3: given clientid is ignored: */
4150                         memcpy(&lock->v.new.clientid,
4151                                 &cstate->session->se_client->cl_clientid,
4152                                 sizeof(clientid_t));
4153
4154                 status = nfserr_stale_clientid;
4155                 if (STALE_CLIENTID(&lock->lk_new_clientid))
4156                         goto out;
4157
4158                 /* validate and update open stateid and open seqid */
4159                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
4160                                         lock->lk_new_open_seqid,
4161                                         &lock->lk_new_open_stateid,
4162                                         &open_stp);
4163                 if (status)
4164                         goto out;
4165                 open_sop = openowner(open_stp->st_stateowner);
4166                 status = nfserr_bad_stateid;
4167                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
4168                                                 &lock->v.new.clientid))
4169                         goto out;
4170                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
4171                                                         &lock_stp, &new_state);
4172                 if (status)
4173                         goto out;
4174         } else {
4175                 /* lock (lock owner + lock stateid) already exists */
4176                 status = nfs4_preprocess_seqid_op(cstate,
4177                                        lock->lk_old_lock_seqid,
4178                                        &lock->lk_old_lock_stateid,
4179                                        NFS4_LOCK_STID, &lock_stp);
4180                 if (status)
4181                         goto out;
4182         }
4183         lock_sop = lockowner(lock_stp->st_stateowner);
4184
4185         lkflg = setlkflg(lock->lk_type);
4186         status = nfs4_check_openmode(lock_stp, lkflg);
4187         if (status)
4188                 goto out;
4189
4190         status = nfserr_grace;
4191         if (locks_in_grace() && !lock->lk_reclaim)
4192                 goto out;
4193         status = nfserr_no_grace;
4194         if (!locks_in_grace() && lock->lk_reclaim)
4195                 goto out;
4196
4197         locks_init_lock(&file_lock);
4198         switch (lock->lk_type) {
4199                 case NFS4_READ_LT:
4200                 case NFS4_READW_LT:
4201                         filp = find_readable_file(lock_stp->st_file);
4202                         if (filp)
4203                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
4204                         file_lock.fl_type = F_RDLCK;
4205                         break;
4206                 case NFS4_WRITE_LT:
4207                 case NFS4_WRITEW_LT:
4208                         filp = find_writeable_file(lock_stp->st_file);
4209                         if (filp)
4210                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
4211                         file_lock.fl_type = F_WRLCK;
4212                         break;
4213                 default:
4214                         status = nfserr_inval;
4215                 goto out;
4216         }
4217         if (!filp) {
4218                 status = nfserr_openmode;
4219                 goto out;
4220         }
4221         file_lock.fl_owner = (fl_owner_t)lock_sop;
4222         file_lock.fl_pid = current->tgid;
4223         file_lock.fl_file = filp;
4224         file_lock.fl_flags = FL_POSIX;
4225         file_lock.fl_lmops = &nfsd_posix_mng_ops;
4226
4227         file_lock.fl_start = lock->lk_offset;
4228         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
4229         nfs4_transform_lock_offset(&file_lock);
4230
4231         /*
4232         * Try to lock the file in the VFS.
4233         * Note: locks.c uses the BKL to protect the inode's lock list.
4234         */
4235
4236         err = vfs_lock_file(filp, F_SETLK, &file_lock, &conflock);
4237         switch (-err) {
4238         case 0: /* success! */
4239                 update_stateid(&lock_stp->st_stid.sc_stateid);
4240                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
4241                                 sizeof(stateid_t));
4242                 status = 0;
4243                 break;
4244         case (EAGAIN):          /* conflock holds conflicting lock */
4245                 status = nfserr_denied;
4246                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
4247                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
4248                 break;
4249         case (EDEADLK):
4250                 status = nfserr_deadlock;
4251                 break;
4252         default:
4253                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4254                 status = nfserrno(err);
4255                 break;
4256         }
4257 out:
4258         if (status && new_state)
4259                 release_lockowner(lock_sop);
4260         if (!cstate->replay_owner)
4261                 nfs4_unlock_state();
4262         return status;
4263 }
4264
4265 /*
4266  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4267  * so we do a temporary open here just to get an open file to pass to
4268  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
4269  * inode operation.)
4270  */
4271 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4272 {
4273         struct file *file;
4274         int err;
4275
4276         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4277         if (err)
4278                 return err;
4279         err = vfs_test_lock(file, lock);
4280         nfsd_close(file);
4281         return err;
4282 }
4283
4284 /*
4285  * LOCKT operation
4286  */
4287 __be32
4288 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4289             struct nfsd4_lockt *lockt)
4290 {
4291         struct inode *inode;
4292         struct file_lock file_lock;
4293         struct nfs4_lockowner *lo;
4294         int error;
4295         __be32 status;
4296
4297         if (locks_in_grace())
4298                 return nfserr_grace;
4299
4300         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4301                  return nfserr_inval;
4302
4303         nfs4_lock_state();
4304
4305         status = nfserr_stale_clientid;
4306         if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
4307                 goto out;
4308
4309         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4310                 goto out;
4311
4312         inode = cstate->current_fh.fh_dentry->d_inode;
4313         locks_init_lock(&file_lock);
4314         switch (lockt->lt_type) {
4315                 case NFS4_READ_LT:
4316                 case NFS4_READW_LT:
4317                         file_lock.fl_type = F_RDLCK;
4318                 break;
4319                 case NFS4_WRITE_LT:
4320                 case NFS4_WRITEW_LT:
4321                         file_lock.fl_type = F_WRLCK;
4322                 break;
4323                 default:
4324                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4325                         status = nfserr_inval;
4326                 goto out;
4327         }
4328
4329         lo = find_lockowner_str(inode, &lockt->lt_clientid, &lockt->lt_owner);
4330         if (lo)
4331                 file_lock.fl_owner = (fl_owner_t)lo;
4332         file_lock.fl_pid = current->tgid;
4333         file_lock.fl_flags = FL_POSIX;
4334
4335         file_lock.fl_start = lockt->lt_offset;
4336         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4337
4338         nfs4_transform_lock_offset(&file_lock);
4339
4340         status = nfs_ok;
4341         error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
4342         if (error) {
4343                 status = nfserrno(error);
4344                 goto out;
4345         }
4346         if (file_lock.fl_type != F_UNLCK) {
4347                 status = nfserr_denied;
4348                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
4349         }
4350 out:
4351         nfs4_unlock_state();
4352         return status;
4353 }
4354
4355 __be32
4356 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4357             struct nfsd4_locku *locku)
4358 {
4359         struct nfs4_ol_stateid *stp;
4360         struct file *filp = NULL;
4361         struct file_lock file_lock;
4362         __be32 status;
4363         int err;
4364
4365         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4366                 (long long) locku->lu_offset,
4367                 (long long) locku->lu_length);
4368
4369         if (check_lock_length(locku->lu_offset, locku->lu_length))
4370                  return nfserr_inval;
4371
4372         nfs4_lock_state();
4373
4374         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
4375                                         &locku->lu_stateid, NFS4_LOCK_STID, &stp);
4376         if (status)
4377                 goto out;
4378         filp = find_any_file(stp->st_file);
4379         if (!filp) {
4380                 status = nfserr_lock_range;
4381                 goto out;
4382         }
4383         BUG_ON(!filp);
4384         locks_init_lock(&file_lock);
4385         file_lock.fl_type = F_UNLCK;
4386         file_lock.fl_owner = (fl_owner_t)lockowner(stp->st_stateowner);
4387         file_lock.fl_pid = current->tgid;
4388         file_lock.fl_file = filp;
4389         file_lock.fl_flags = FL_POSIX;
4390         file_lock.fl_lmops = &nfsd_posix_mng_ops;
4391         file_lock.fl_start = locku->lu_offset;
4392
4393         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
4394         nfs4_transform_lock_offset(&file_lock);
4395
4396         /*
4397         *  Try to unlock the file in the VFS.
4398         */
4399         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
4400         if (err) {
4401                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4402                 goto out_nfserr;
4403         }
4404         /*
4405         * OK, unlock succeeded; the only thing left to do is update the stateid.
4406         */
4407         update_stateid(&stp->st_stid.sc_stateid);
4408         memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4409
4410 out:
4411         if (!cstate->replay_owner)
4412                 nfs4_unlock_state();
4413         return status;
4414
4415 out_nfserr:
4416         status = nfserrno(err);
4417         goto out;
4418 }
4419
4420 /*
4421  * returns
4422  *      1: locks held by lockowner
4423  *      0: no locks held by lockowner
4424  */
4425 static int
4426 check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner)
4427 {
4428         struct file_lock **flpp;
4429         struct inode *inode = filp->fi_inode;
4430         int status = 0;
4431
4432         lock_flocks();
4433         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4434                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4435                         status = 1;
4436                         goto out;
4437                 }
4438         }
4439 out:
4440         unlock_flocks();
4441         return status;
4442 }
4443
4444 __be32
4445 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4446                         struct nfsd4_compound_state *cstate,
4447                         struct nfsd4_release_lockowner *rlockowner)
4448 {
4449         clientid_t *clid = &rlockowner->rl_clientid;
4450         struct nfs4_stateowner *sop;
4451         struct nfs4_lockowner *lo;
4452         struct nfs4_ol_stateid *stp;
4453         struct xdr_netobj *owner = &rlockowner->rl_owner;
4454         struct list_head matches;
4455         unsigned int hashval = ownerstr_hashval(clid->cl_id, owner);
4456         __be32 status;
4457
4458         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4459                 clid->cl_boot, clid->cl_id);
4460
4461         /* XXX check for lease expiration */
4462
4463         status = nfserr_stale_clientid;
4464         if (STALE_CLIENTID(clid))
4465                 return status;
4466
4467         nfs4_lock_state();
4468
4469         status = nfserr_locks_held;
4470         INIT_LIST_HEAD(&matches);
4471
4472         list_for_each_entry(sop, &ownerstr_hashtbl[hashval], so_strhash) {
4473                 if (sop->so_is_open_owner)
4474                         continue;
4475                 if (!same_owner_str(sop, owner, clid))
4476                         continue;
4477                 list_for_each_entry(stp, &sop->so_stateids,
4478                                 st_perstateowner) {
4479                         lo = lockowner(sop);
4480                         if (check_for_locks(stp->st_file, lo))
4481                                 goto out;
4482                         list_add(&lo->lo_list, &matches);
4483                 }
4484         }
4485         /* Clients probably won't expect us to return with some (but not all)
4486          * of the lockowner state released; so don't release any until all
4487          * have been checked. */
4488         status = nfs_ok;
4489         while (!list_empty(&matches)) {
4490                 lo = list_entry(matches.next, struct nfs4_lockowner,
4491                                                                 lo_list);
4492                 /* unhash_stateowner deletes so_perclient only
4493                  * for openowners. */
4494                 list_del(&lo->lo_list);
4495                 release_lockowner(lo);
4496         }
4497 out:
4498         nfs4_unlock_state();
4499         return status;
4500 }
4501
4502 static inline struct nfs4_client_reclaim *
4503 alloc_reclaim(void)
4504 {
4505         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4506 }
4507
4508 int
4509 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
4510 {
4511         unsigned int strhashval = clientstr_hashval(name);
4512         struct nfs4_client *clp;
4513
4514         clp = find_confirmed_client_by_str(name, strhashval);
4515         if (!clp)
4516                 return 0;
4517         return test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
4518 }
4519
4520 /*
4521  * failure => all reset bets are off, nfserr_no_grace...
4522  */
4523 int
4524 nfs4_client_to_reclaim(const char *name)
4525 {
4526         unsigned int strhashval;
4527         struct nfs4_client_reclaim *crp = NULL;
4528
4529         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4530         crp = alloc_reclaim();
4531         if (!crp)
4532                 return 0;
4533         strhashval = clientstr_hashval(name);
4534         INIT_LIST_HEAD(&crp->cr_strhash);
4535         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4536         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4537         reclaim_str_hashtbl_size++;
4538         return 1;
4539 }
4540
4541 void
4542 nfs4_release_reclaim(void)
4543 {
4544         struct nfs4_client_reclaim *crp = NULL;
4545         int i;
4546
4547         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4548                 while (!list_empty(&reclaim_str_hashtbl[i])) {
4549                         crp = list_entry(reclaim_str_hashtbl[i].next,
4550                                         struct nfs4_client_reclaim, cr_strhash);
4551                         list_del(&crp->cr_strhash);
4552                         kfree(crp);
4553                         reclaim_str_hashtbl_size--;
4554                 }
4555         }
4556         BUG_ON(reclaim_str_hashtbl_size);
4557 }
4558
4559 /*
4560  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4561 struct nfs4_client_reclaim *
4562 nfsd4_find_reclaim_client(struct nfs4_client *clp)
4563 {
4564         unsigned int strhashval;
4565         struct nfs4_client_reclaim *crp = NULL;
4566
4567         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4568                             clp->cl_name.len, clp->cl_name.data,
4569                             clp->cl_recdir);
4570
4571         /* find clp->cl_name in reclaim_str_hashtbl */
4572         strhashval = clientstr_hashval(clp->cl_recdir);
4573         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4574                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4575                         return crp;
4576                 }
4577         }
4578         return NULL;
4579 }
4580
4581 /*
4582 * Called from OPEN. Look for clientid in reclaim list.
4583 */
4584 __be32
4585 nfs4_check_open_reclaim(clientid_t *clid)
4586 {
4587         struct nfs4_client *clp;
4588
4589         /* find clientid in conf_id_hashtbl */
4590         clp = find_confirmed_client(clid);
4591         if (clp == NULL)
4592                 return nfserr_reclaim_bad;
4593
4594         return nfsd4_client_record_check(clp) ? nfserr_reclaim_bad : nfs_ok;
4595 }
4596
4597 #ifdef CONFIG_NFSD_FAULT_INJECTION
4598
4599 void nfsd_forget_clients(u64 num)
4600 {
4601         struct nfs4_client *clp, *next;
4602         int count = 0;
4603
4604         nfs4_lock_state();
4605         list_for_each_entry_safe(clp, next, &client_lru, cl_lru) {
4606                 nfsd4_client_record_remove(clp);
4607                 expire_client(clp);
4608                 if (++count == num)
4609                         break;
4610         }
4611         nfs4_unlock_state();
4612
4613         printk(KERN_INFO "NFSD: Forgot %d clients", count);
4614 }
4615
4616 static void release_lockowner_sop(struct nfs4_stateowner *sop)
4617 {
4618         release_lockowner(lockowner(sop));
4619 }
4620
4621 static void release_openowner_sop(struct nfs4_stateowner *sop)
4622 {
4623         release_openowner(openowner(sop));
4624 }
4625
4626 static int nfsd_release_n_owners(u64 num, bool is_open_owner,
4627                                 void (*release_sop)(struct nfs4_stateowner *))
4628 {
4629         int i, count = 0;
4630         struct nfs4_stateowner *sop, *next;
4631
4632         for (i = 0; i < OWNER_HASH_SIZE; i++) {
4633                 list_for_each_entry_safe(sop, next, &ownerstr_hashtbl[i], so_strhash) {
4634                         if (sop->so_is_open_owner != is_open_owner)
4635                                 continue;
4636                         release_sop(sop);
4637                         if (++count == num)
4638                                 return count;
4639                 }
4640         }
4641         return count;
4642 }
4643
4644 void nfsd_forget_locks(u64 num)
4645 {
4646         int count;
4647
4648         nfs4_lock_state();
4649         count = nfsd_release_n_owners(num, false, release_lockowner_sop);
4650         nfs4_unlock_state();
4651
4652         printk(KERN_INFO "NFSD: Forgot %d locks", count);
4653 }
4654
4655 void nfsd_forget_openowners(u64 num)
4656 {
4657         int count;
4658
4659         nfs4_lock_state();
4660         count = nfsd_release_n_owners(num, true, release_openowner_sop);
4661         nfs4_unlock_state();
4662
4663         printk(KERN_INFO "NFSD: Forgot %d open owners", count);
4664 }
4665
4666 int nfsd_process_n_delegations(u64 num, void (*deleg_func)(struct nfs4_delegation *))
4667 {
4668         int i, count = 0;
4669         struct nfs4_file *fp, *fnext;
4670         struct nfs4_delegation *dp, *dnext;
4671
4672         for (i = 0; i < FILE_HASH_SIZE; i++) {
4673                 list_for_each_entry_safe(fp, fnext, &file_hashtbl[i], fi_hash) {
4674                         list_for_each_entry_safe(dp, dnext, &fp->fi_delegations, dl_perfile) {
4675                                 deleg_func(dp);
4676                                 if (++count == num)
4677                                         return count;
4678                         }
4679                 }
4680         }
4681
4682         return count;
4683 }
4684
4685 void nfsd_forget_delegations(u64 num)
4686 {
4687         unsigned int count;
4688
4689         nfs4_lock_state();
4690         count = nfsd_process_n_delegations(num, unhash_delegation);
4691         nfs4_unlock_state();
4692
4693         printk(KERN_INFO "NFSD: Forgot %d delegations", count);
4694 }
4695
4696 void nfsd_recall_delegations(u64 num)
4697 {
4698         unsigned int count;
4699
4700         nfs4_lock_state();
4701         spin_lock(&recall_lock);
4702         count = nfsd_process_n_delegations(num, nfsd_break_one_deleg);
4703         spin_unlock(&recall_lock);
4704         nfs4_unlock_state();
4705
4706         printk(KERN_INFO "NFSD: Recalled %d delegations", count);
4707 }
4708
4709 #endif /* CONFIG_NFSD_FAULT_INJECTION */
4710
4711 /* initialization to perform at module load time: */
4712
4713 void
4714 nfs4_state_init(void)
4715 {
4716         int i;
4717
4718         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4719                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4720                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4721                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4722                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4723                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4724         }
4725         for (i = 0; i < SESSION_HASH_SIZE; i++)
4726                 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4727         for (i = 0; i < FILE_HASH_SIZE; i++) {
4728                 INIT_LIST_HEAD(&file_hashtbl[i]);
4729         }
4730         for (i = 0; i < OWNER_HASH_SIZE; i++) {
4731                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4732         }
4733         for (i = 0; i < LOCKOWNER_INO_HASH_SIZE; i++)
4734                 INIT_LIST_HEAD(&lockowner_ino_hashtbl[i]);
4735         INIT_LIST_HEAD(&close_lru);
4736         INIT_LIST_HEAD(&client_lru);
4737         INIT_LIST_HEAD(&del_recall_lru);
4738         reclaim_str_hashtbl_size = 0;
4739 }
4740
4741 /*
4742  * Since the lifetime of a delegation isn't limited to that of an open, a
4743  * client may quite reasonably hang on to a delegation as long as it has
4744  * the inode cached.  This becomes an obvious problem the first time a
4745  * client's inode cache approaches the size of the server's total memory.
4746  *
4747  * For now we avoid this problem by imposing a hard limit on the number
4748  * of delegations, which varies according to the server's memory size.
4749  */
4750 static void
4751 set_max_delegations(void)
4752 {
4753         /*
4754          * Allow at most 4 delegations per megabyte of RAM.  Quick
4755          * estimates suggest that in the worst case (where every delegation
4756          * is for a different inode), a delegation could take about 1.5K,
4757          * giving a worst case usage of about 6% of memory.
4758          */
4759         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4760 }
4761
4762 /* initialization to perform when the nfsd service is started: */
4763
4764 int
4765 nfs4_state_start(void)
4766 {
4767         int ret;
4768
4769         /*
4770          * FIXME: For now, we hang most of the pernet global stuff off of
4771          * init_net until nfsd is fully containerized. Eventually, we'll
4772          * need to pass a net pointer into this function, take a reference
4773          * to that instead and then do most of the rest of this on a per-net
4774          * basis.
4775          */
4776         get_net(&init_net);
4777         nfsd4_client_tracking_init(&init_net);
4778         boot_time = get_seconds();
4779         locks_start_grace(&nfsd4_manager);
4780         grace_ended = false;
4781         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4782                nfsd4_grace);
4783         ret = set_callback_cred();
4784         if (ret) {
4785                 ret = -ENOMEM;
4786                 goto out_recovery;
4787         }
4788         laundry_wq = create_singlethread_workqueue("nfsd4");
4789         if (laundry_wq == NULL) {
4790                 ret = -ENOMEM;
4791                 goto out_recovery;
4792         }
4793         ret = nfsd4_create_callback_queue();
4794         if (ret)
4795                 goto out_free_laundry;
4796         queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4797         set_max_delegations();
4798         return 0;
4799 out_free_laundry:
4800         destroy_workqueue(laundry_wq);
4801 out_recovery:
4802         nfsd4_client_tracking_exit(&init_net);
4803         put_net(&init_net);
4804         return ret;
4805 }
4806
4807 static void
4808 __nfs4_state_shutdown(void)
4809 {
4810         int i;
4811         struct nfs4_client *clp = NULL;
4812         struct nfs4_delegation *dp = NULL;
4813         struct list_head *pos, *next, reaplist;
4814
4815         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4816                 while (!list_empty(&conf_id_hashtbl[i])) {
4817                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4818                         expire_client(clp);
4819                 }
4820                 while (!list_empty(&unconf_str_hashtbl[i])) {
4821                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4822                         expire_client(clp);
4823                 }
4824         }
4825         INIT_LIST_HEAD(&reaplist);
4826         spin_lock(&recall_lock);
4827         list_for_each_safe(pos, next, &del_recall_lru) {
4828                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4829                 list_move(&dp->dl_recall_lru, &reaplist);
4830         }
4831         spin_unlock(&recall_lock);
4832         list_for_each_safe(pos, next, &reaplist) {
4833                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4834                 unhash_delegation(dp);
4835         }
4836
4837         nfsd4_client_tracking_exit(&init_net);
4838         put_net(&init_net);
4839 }
4840
4841 void
4842 nfs4_state_shutdown(void)
4843 {
4844         cancel_delayed_work_sync(&laundromat_work);
4845         destroy_workqueue(laundry_wq);
4846         locks_end_grace(&nfsd4_manager);
4847         nfs4_lock_state();
4848         __nfs4_state_shutdown();
4849         nfs4_unlock_state();
4850         nfsd4_destroy_callback_queue();
4851 }
4852
4853 static void
4854 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
4855 {
4856         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
4857                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
4858 }
4859
4860 static void
4861 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
4862 {
4863         if (cstate->minorversion) {
4864                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
4865                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
4866         }
4867 }
4868
4869 void
4870 clear_current_stateid(struct nfsd4_compound_state *cstate)
4871 {
4872         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
4873 }
4874
4875 /*
4876  * functions to set current state id
4877  */
4878 void
4879 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
4880 {
4881         put_stateid(cstate, &odp->od_stateid);
4882 }
4883
4884 void
4885 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
4886 {
4887         put_stateid(cstate, &open->op_stateid);
4888 }
4889
4890 void
4891 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
4892 {
4893         put_stateid(cstate, &close->cl_stateid);
4894 }
4895
4896 void
4897 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
4898 {
4899         put_stateid(cstate, &lock->lk_resp_stateid);
4900 }
4901
4902 /*
4903  * functions to consume current state id
4904  */
4905
4906 void
4907 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
4908 {
4909         get_stateid(cstate, &odp->od_stateid);
4910 }
4911
4912 void
4913 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
4914 {
4915         get_stateid(cstate, &drp->dr_stateid);
4916 }
4917
4918 void
4919 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
4920 {
4921         get_stateid(cstate, &fsp->fr_stateid);
4922 }
4923
4924 void
4925 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
4926 {
4927         get_stateid(cstate, &setattr->sa_stateid);
4928 }
4929
4930 void
4931 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
4932 {
4933         get_stateid(cstate, &close->cl_stateid);
4934 }
4935
4936 void
4937 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
4938 {
4939         get_stateid(cstate, &locku->lu_stateid);
4940 }
4941
4942 void
4943 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
4944 {
4945         get_stateid(cstate, &read->rd_stateid);
4946 }
4947
4948 void
4949 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
4950 {
4951         get_stateid(cstate, &write->wr_stateid);
4952 }