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         if (clp->cl_cred.cr_group_info)
1091                 put_group_info(clp->cl_cred.cr_group_info);
1092         kfree(clp->cl_principal);
1093         kfree(clp->cl_name.data);
1094         kfree(clp);
1095 }
1096
1097 void
1098 release_session_client(struct nfsd4_session *session)
1099 {
1100         struct nfs4_client *clp = session->se_client;
1101
1102         if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
1103                 return;
1104         if (is_client_expired(clp)) {
1105                 free_client(clp);
1106                 session->se_client = NULL;
1107         } else
1108                 renew_client_locked(clp);
1109         spin_unlock(&client_lock);
1110 }
1111
1112 /* must be called under the client_lock */
1113 static inline void
1114 unhash_client_locked(struct nfs4_client *clp)
1115 {
1116         struct nfsd4_session *ses;
1117
1118         mark_client_expired(clp);
1119         list_del(&clp->cl_lru);
1120         spin_lock(&clp->cl_lock);
1121         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1122                 list_del_init(&ses->se_hash);
1123         spin_unlock(&clp->cl_lock);
1124 }
1125
1126 static void
1127 expire_client(struct nfs4_client *clp)
1128 {
1129         struct nfs4_openowner *oo;
1130         struct nfs4_delegation *dp;
1131         struct list_head reaplist;
1132
1133         INIT_LIST_HEAD(&reaplist);
1134         spin_lock(&recall_lock);
1135         while (!list_empty(&clp->cl_delegations)) {
1136                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1137                 list_del_init(&dp->dl_perclnt);
1138                 list_move(&dp->dl_recall_lru, &reaplist);
1139         }
1140         spin_unlock(&recall_lock);
1141         while (!list_empty(&reaplist)) {
1142                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1143                 unhash_delegation(dp);
1144         }
1145         while (!list_empty(&clp->cl_openowners)) {
1146                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1147                 release_openowner(oo);
1148         }
1149         nfsd4_shutdown_callback(clp);
1150         if (clp->cl_cb_conn.cb_xprt)
1151                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1152         list_del(&clp->cl_idhash);
1153         list_del(&clp->cl_strhash);
1154         spin_lock(&client_lock);
1155         unhash_client_locked(clp);
1156         if (atomic_read(&clp->cl_refcount) == 0)
1157                 free_client(clp);
1158         spin_unlock(&client_lock);
1159 }
1160
1161 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1162 {
1163         memcpy(target->cl_verifier.data, source->data,
1164                         sizeof(target->cl_verifier.data));
1165 }
1166
1167 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1168 {
1169         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1170         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1171 }
1172
1173 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
1174 {
1175         target->cr_uid = source->cr_uid;
1176         target->cr_gid = source->cr_gid;
1177         target->cr_group_info = source->cr_group_info;
1178         get_group_info(target->cr_group_info);
1179 }
1180
1181 static int same_name(const char *n1, const char *n2)
1182 {
1183         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1184 }
1185
1186 static int
1187 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1188 {
1189         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1190 }
1191
1192 static int
1193 same_clid(clientid_t *cl1, clientid_t *cl2)
1194 {
1195         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1196 }
1197
1198 /* XXX what about NGROUP */
1199 static int
1200 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1201 {
1202         return cr1->cr_uid == cr2->cr_uid;
1203 }
1204
1205 static void gen_clid(struct nfs4_client *clp)
1206 {
1207         static u32 current_clientid = 1;
1208
1209         clp->cl_clientid.cl_boot = boot_time;
1210         clp->cl_clientid.cl_id = current_clientid++;
1211 }
1212
1213 static void gen_confirm(struct nfs4_client *clp)
1214 {
1215         __be32 verf[2];
1216         static u32 i;
1217
1218         verf[0] = (__be32)get_seconds();
1219         verf[1] = (__be32)i++;
1220         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1221 }
1222
1223 static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1224 {
1225         return idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1226 }
1227
1228 static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1229 {
1230         struct nfs4_stid *s;
1231
1232         s = find_stateid(cl, t);
1233         if (!s)
1234                 return NULL;
1235         if (typemask & s->sc_type)
1236                 return s;
1237         return NULL;
1238 }
1239
1240 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1241                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1242 {
1243         struct nfs4_client *clp;
1244         struct sockaddr *sa = svc_addr(rqstp);
1245         char *princ;
1246
1247         clp = alloc_client(name);
1248         if (clp == NULL)
1249                 return NULL;
1250
1251         INIT_LIST_HEAD(&clp->cl_sessions);
1252
1253         princ = svc_gss_principal(rqstp);
1254         if (princ) {
1255                 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
1256                 if (clp->cl_principal == NULL) {
1257                         spin_lock(&client_lock);
1258                         free_client(clp);
1259                         spin_unlock(&client_lock);
1260                         return NULL;
1261                 }
1262         }
1263
1264         idr_init(&clp->cl_stateids);
1265         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1266         atomic_set(&clp->cl_refcount, 0);
1267         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1268         INIT_LIST_HEAD(&clp->cl_idhash);
1269         INIT_LIST_HEAD(&clp->cl_strhash);
1270         INIT_LIST_HEAD(&clp->cl_openowners);
1271         INIT_LIST_HEAD(&clp->cl_delegations);
1272         INIT_LIST_HEAD(&clp->cl_lru);
1273         INIT_LIST_HEAD(&clp->cl_callbacks);
1274         spin_lock_init(&clp->cl_lock);
1275         INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1276         clp->cl_time = get_seconds();
1277         clear_bit(0, &clp->cl_cb_slot_busy);
1278         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1279         copy_verf(clp, verf);
1280         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1281         clp->cl_flavor = rqstp->rq_flavor;
1282         copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1283         gen_confirm(clp);
1284         clp->cl_cb_session = NULL;
1285         return clp;
1286 }
1287
1288 static void
1289 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1290 {
1291         unsigned int idhashval;
1292
1293         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1294         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1295         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1296         renew_client(clp);
1297 }
1298
1299 static void
1300 move_to_confirmed(struct nfs4_client *clp)
1301 {
1302         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1303         unsigned int strhashval;
1304
1305         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1306         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1307         strhashval = clientstr_hashval(clp->cl_recdir);
1308         list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1309         renew_client(clp);
1310 }
1311
1312 static struct nfs4_client *
1313 find_confirmed_client(clientid_t *clid)
1314 {
1315         struct nfs4_client *clp;
1316         unsigned int idhashval = clientid_hashval(clid->cl_id);
1317
1318         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1319                 if (same_clid(&clp->cl_clientid, clid)) {
1320                         renew_client(clp);
1321                         return clp;
1322                 }
1323         }
1324         return NULL;
1325 }
1326
1327 static struct nfs4_client *
1328 find_unconfirmed_client(clientid_t *clid)
1329 {
1330         struct nfs4_client *clp;
1331         unsigned int idhashval = clientid_hashval(clid->cl_id);
1332
1333         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1334                 if (same_clid(&clp->cl_clientid, clid))
1335                         return clp;
1336         }
1337         return NULL;
1338 }
1339
1340 static bool clp_used_exchangeid(struct nfs4_client *clp)
1341 {
1342         return clp->cl_exchange_flags != 0;
1343 }
1344
1345 static struct nfs4_client *
1346 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
1347 {
1348         struct nfs4_client *clp;
1349
1350         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1351                 if (same_name(clp->cl_recdir, dname))
1352                         return clp;
1353         }
1354         return NULL;
1355 }
1356
1357 static struct nfs4_client *
1358 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
1359 {
1360         struct nfs4_client *clp;
1361
1362         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1363                 if (same_name(clp->cl_recdir, dname))
1364                         return clp;
1365         }
1366         return NULL;
1367 }
1368
1369 static void
1370 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1371 {
1372         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1373         struct sockaddr *sa = svc_addr(rqstp);
1374         u32 scopeid = rpc_get_scope_id(sa);
1375         unsigned short expected_family;
1376
1377         /* Currently, we only support tcp and tcp6 for the callback channel */
1378         if (se->se_callback_netid_len == 3 &&
1379             !memcmp(se->se_callback_netid_val, "tcp", 3))
1380                 expected_family = AF_INET;
1381         else if (se->se_callback_netid_len == 4 &&
1382                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1383                 expected_family = AF_INET6;
1384         else
1385                 goto out_err;
1386
1387         conn->cb_addrlen = rpc_uaddr2sockaddr(&init_net, se->se_callback_addr_val,
1388                                             se->se_callback_addr_len,
1389                                             (struct sockaddr *)&conn->cb_addr,
1390                                             sizeof(conn->cb_addr));
1391
1392         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1393                 goto out_err;
1394
1395         if (conn->cb_addr.ss_family == AF_INET6)
1396                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1397
1398         conn->cb_prog = se->se_callback_prog;
1399         conn->cb_ident = se->se_callback_ident;
1400         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1401         return;
1402 out_err:
1403         conn->cb_addr.ss_family = AF_UNSPEC;
1404         conn->cb_addrlen = 0;
1405         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1406                 "will not receive delegations\n",
1407                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1408
1409         return;
1410 }
1411
1412 /*
1413  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1414  */
1415 void
1416 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1417 {
1418         struct nfsd4_slot *slot = resp->cstate.slot;
1419         unsigned int base;
1420
1421         dprintk("--> %s slot %p\n", __func__, slot);
1422
1423         slot->sl_opcnt = resp->opcnt;
1424         slot->sl_status = resp->cstate.status;
1425
1426         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
1427         if (nfsd4_not_cached(resp)) {
1428                 slot->sl_datalen = 0;
1429                 return;
1430         }
1431         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1432         base = (char *)resp->cstate.datap -
1433                                         (char *)resp->xbuf->head[0].iov_base;
1434         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1435                                     slot->sl_datalen))
1436                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1437         return;
1438 }
1439
1440 /*
1441  * Encode the replay sequence operation from the slot values.
1442  * If cachethis is FALSE encode the uncached rep error on the next
1443  * operation which sets resp->p and increments resp->opcnt for
1444  * nfs4svc_encode_compoundres.
1445  *
1446  */
1447 static __be32
1448 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1449                           struct nfsd4_compoundres *resp)
1450 {
1451         struct nfsd4_op *op;
1452         struct nfsd4_slot *slot = resp->cstate.slot;
1453
1454         /* Encode the replayed sequence operation */
1455         op = &args->ops[resp->opcnt - 1];
1456         nfsd4_encode_operation(resp, op);
1457
1458         /* Return nfserr_retry_uncached_rep in next operation. */
1459         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
1460                 op = &args->ops[resp->opcnt++];
1461                 op->status = nfserr_retry_uncached_rep;
1462                 nfsd4_encode_operation(resp, op);
1463         }
1464         return op->status;
1465 }
1466
1467 /*
1468  * The sequence operation is not cached because we can use the slot and
1469  * session values.
1470  */
1471 __be32
1472 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1473                          struct nfsd4_sequence *seq)
1474 {
1475         struct nfsd4_slot *slot = resp->cstate.slot;
1476         __be32 status;
1477
1478         dprintk("--> %s slot %p\n", __func__, slot);
1479
1480         /* Either returns 0 or nfserr_retry_uncached */
1481         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1482         if (status == nfserr_retry_uncached_rep)
1483                 return status;
1484
1485         /* The sequence operation has been encoded, cstate->datap set. */
1486         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1487
1488         resp->opcnt = slot->sl_opcnt;
1489         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1490         status = slot->sl_status;
1491
1492         return status;
1493 }
1494
1495 /*
1496  * Set the exchange_id flags returned by the server.
1497  */
1498 static void
1499 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1500 {
1501         /* pNFS is not supported */
1502         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1503
1504         /* Referrals are supported, Migration is not. */
1505         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1506
1507         /* set the wire flags to return to client. */
1508         clid->flags = new->cl_exchange_flags;
1509 }
1510
1511 __be32
1512 nfsd4_exchange_id(struct svc_rqst *rqstp,
1513                   struct nfsd4_compound_state *cstate,
1514                   struct nfsd4_exchange_id *exid)
1515 {
1516         struct nfs4_client *unconf, *conf, *new;
1517         __be32 status;
1518         unsigned int            strhashval;
1519         char                    dname[HEXDIR_LEN];
1520         char                    addr_str[INET6_ADDRSTRLEN];
1521         nfs4_verifier           verf = exid->verifier;
1522         struct sockaddr         *sa = svc_addr(rqstp);
1523
1524         rpc_ntop(sa, addr_str, sizeof(addr_str));
1525         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1526                 "ip_addr=%s flags %x, spa_how %d\n",
1527                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1528                 addr_str, exid->flags, exid->spa_how);
1529
1530         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1531                 return nfserr_inval;
1532
1533         /* Currently only support SP4_NONE */
1534         switch (exid->spa_how) {
1535         case SP4_NONE:
1536                 break;
1537         case SP4_SSV:
1538                 return nfserr_serverfault;
1539         default:
1540                 BUG();                          /* checked by xdr code */
1541         case SP4_MACH_CRED:
1542                 return nfserr_serverfault;      /* no excuse :-/ */
1543         }
1544
1545         status = nfs4_make_rec_clidname(dname, &exid->clname);
1546
1547         if (status)
1548                 return status;
1549
1550         strhashval = clientstr_hashval(dname);
1551
1552         nfs4_lock_state();
1553         conf = find_confirmed_client_by_str(dname, strhashval);
1554         if (conf) {
1555                 if (!clp_used_exchangeid(conf)) {
1556                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1557                                 status = nfserr_inval; /* buggy client */
1558                                 goto out;
1559                         }
1560                 }
1561                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1562                         /* 18.35.4 case 9 */
1563                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A)
1564                                 status = nfserr_perm;
1565                         else /* case 3 */
1566                                 status = nfserr_clid_inuse;
1567                         goto out;
1568                 }
1569                 if (!same_verf(&verf, &conf->cl_verifier)) {
1570                         /* 18.35.4 case 8 */
1571                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1572                                 status = nfserr_not_same;
1573                                 goto out;
1574                         }
1575                         /* Client reboot: destroy old state */
1576                         expire_client(conf);
1577                         goto out_new;
1578                 }
1579                 if (!clp_used_exchangeid(conf)) {
1580                         status = nfserr_inval;
1581                         goto out;
1582                 }
1583                 /*
1584                  * Set bit when the owner id and verifier map to an already
1585                  * confirmed client id (18.35.3).
1586                  */
1587                 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1588
1589                 /*
1590                  * Falling into 18.35.4 case 2, possible router replay.
1591                  * Leave confirmed record intact and return same result.
1592                  */
1593                 new = conf;
1594                 goto out_copy;
1595         }
1596
1597         /* 18.35.4 case 7 */
1598         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1599                 status = nfserr_noent;
1600                 goto out;
1601         }
1602
1603         unconf  = find_unconfirmed_client_by_str(dname, strhashval);
1604         if (unconf) {
1605                 /*
1606                  * Possible retry or client restart.  Per 18.35.4 case 4,
1607                  * a new unconfirmed record should be generated regardless
1608                  * of whether any properties have changed.
1609                  */
1610                 expire_client(unconf);
1611         }
1612
1613 out_new:
1614         /* Normal case */
1615         new = create_client(exid->clname, dname, rqstp, &verf);
1616         if (new == NULL) {
1617                 status = nfserr_jukebox;
1618                 goto out;
1619         }
1620
1621         gen_clid(new);
1622         add_to_unconfirmed(new, strhashval);
1623 out_copy:
1624         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1625         exid->clientid.cl_id = new->cl_clientid.cl_id;
1626
1627         exid->seqid = 1;
1628         nfsd4_set_ex_flags(new, exid);
1629
1630         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1631                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1632         status = nfs_ok;
1633
1634 out:
1635         nfs4_unlock_state();
1636         return status;
1637 }
1638
1639 static __be32
1640 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1641 {
1642         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1643                 slot_seqid);
1644
1645         /* The slot is in use, and no response has been sent. */
1646         if (slot_inuse) {
1647                 if (seqid == slot_seqid)
1648                         return nfserr_jukebox;
1649                 else
1650                         return nfserr_seq_misordered;
1651         }
1652         /* Note unsigned 32-bit arithmetic handles wraparound: */
1653         if (likely(seqid == slot_seqid + 1))
1654                 return nfs_ok;
1655         if (seqid == slot_seqid)
1656                 return nfserr_replay_cache;
1657         return nfserr_seq_misordered;
1658 }
1659
1660 /*
1661  * Cache the create session result into the create session single DRC
1662  * slot cache by saving the xdr structure. sl_seqid has been set.
1663  * Do this for solo or embedded create session operations.
1664  */
1665 static void
1666 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1667                            struct nfsd4_clid_slot *slot, __be32 nfserr)
1668 {
1669         slot->sl_status = nfserr;
1670         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1671 }
1672
1673 static __be32
1674 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1675                             struct nfsd4_clid_slot *slot)
1676 {
1677         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1678         return slot->sl_status;
1679 }
1680
1681 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1682                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1683                         1 +     /* MIN tag is length with zero, only length */ \
1684                         3 +     /* version, opcount, opcode */ \
1685                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1686                                 /* seqid, slotID, slotID, cache */ \
1687                         4 ) * sizeof(__be32))
1688
1689 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1690                         2 +     /* verifier: AUTH_NULL, length 0 */\
1691                         1 +     /* status */ \
1692                         1 +     /* MIN tag is length with zero, only length */ \
1693                         3 +     /* opcount, opcode, opstatus*/ \
1694                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1695                                 /* seqid, slotID, slotID, slotID, status */ \
1696                         5 ) * sizeof(__be32))
1697
1698 static bool check_forechannel_attrs(struct nfsd4_channel_attrs fchannel)
1699 {
1700         return fchannel.maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ
1701                 || fchannel.maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ;
1702 }
1703
1704 __be32
1705 nfsd4_create_session(struct svc_rqst *rqstp,
1706                      struct nfsd4_compound_state *cstate,
1707                      struct nfsd4_create_session *cr_ses)
1708 {
1709         struct sockaddr *sa = svc_addr(rqstp);
1710         struct nfs4_client *conf, *unconf;
1711         struct nfsd4_session *new;
1712         struct nfsd4_clid_slot *cs_slot = NULL;
1713         bool confirm_me = false;
1714         __be32 status = 0;
1715
1716         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1717                 return nfserr_inval;
1718
1719         nfs4_lock_state();
1720         unconf = find_unconfirmed_client(&cr_ses->clientid);
1721         conf = find_confirmed_client(&cr_ses->clientid);
1722
1723         if (conf) {
1724                 cs_slot = &conf->cl_cs_slot;
1725                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1726                 if (status == nfserr_replay_cache) {
1727                         dprintk("Got a create_session replay! seqid= %d\n",
1728                                 cs_slot->sl_seqid);
1729                         /* Return the cached reply status */
1730                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1731                         goto out;
1732                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1733                         status = nfserr_seq_misordered;
1734                         dprintk("Sequence misordered!\n");
1735                         dprintk("Expected seqid= %d but got seqid= %d\n",
1736                                 cs_slot->sl_seqid, cr_ses->seqid);
1737                         goto out;
1738                 }
1739         } else if (unconf) {
1740                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1741                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1742                         status = nfserr_clid_inuse;
1743                         goto out;
1744                 }
1745
1746                 cs_slot = &unconf->cl_cs_slot;
1747                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1748                 if (status) {
1749                         /* an unconfirmed replay returns misordered */
1750                         status = nfserr_seq_misordered;
1751                         goto out;
1752                 }
1753
1754                 confirm_me = true;
1755                 conf = unconf;
1756         } else {
1757                 status = nfserr_stale_clientid;
1758                 goto out;
1759         }
1760
1761         /*
1762          * XXX: we should probably set this at creation time, and check
1763          * for consistent minorversion use throughout:
1764          */
1765         conf->cl_minorversion = 1;
1766         /*
1767          * We do not support RDMA or persistent sessions
1768          */
1769         cr_ses->flags &= ~SESSION4_PERSIST;
1770         cr_ses->flags &= ~SESSION4_RDMA;
1771
1772         status = nfserr_toosmall;
1773         if (check_forechannel_attrs(cr_ses->fore_channel))
1774                 goto out;
1775
1776         status = nfserr_jukebox;
1777         new = alloc_init_session(rqstp, conf, cr_ses);
1778         if (!new)
1779                 goto out;
1780         status = nfs_ok;
1781         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1782                NFS4_MAX_SESSIONID_LEN);
1783         memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1784                 sizeof(struct nfsd4_channel_attrs));
1785         cs_slot->sl_seqid++;
1786         cr_ses->seqid = cs_slot->sl_seqid;
1787
1788         /* cache solo and embedded create sessions under the state lock */
1789         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1790         if (confirm_me)
1791                 move_to_confirmed(conf);
1792 out:
1793         nfs4_unlock_state();
1794         dprintk("%s returns %d\n", __func__, ntohl(status));
1795         return status;
1796 }
1797
1798 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1799 {
1800         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1801         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1802
1803         return argp->opcnt == resp->opcnt;
1804 }
1805
1806 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1807 {
1808         switch (*dir) {
1809         case NFS4_CDFC4_FORE:
1810         case NFS4_CDFC4_BACK:
1811                 return nfs_ok;
1812         case NFS4_CDFC4_FORE_OR_BOTH:
1813         case NFS4_CDFC4_BACK_OR_BOTH:
1814                 *dir = NFS4_CDFC4_BOTH;
1815                 return nfs_ok;
1816         };
1817         return nfserr_inval;
1818 }
1819
1820 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1821                      struct nfsd4_compound_state *cstate,
1822                      struct nfsd4_bind_conn_to_session *bcts)
1823 {
1824         __be32 status;
1825
1826         if (!nfsd4_last_compound_op(rqstp))
1827                 return nfserr_not_only_op;
1828         spin_lock(&client_lock);
1829         cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid);
1830         /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1831          * client_lock iself: */
1832         if (cstate->session) {
1833                 nfsd4_get_session(cstate->session);
1834                 atomic_inc(&cstate->session->se_client->cl_refcount);
1835         }
1836         spin_unlock(&client_lock);
1837         if (!cstate->session)
1838                 return nfserr_badsession;
1839
1840         status = nfsd4_map_bcts_dir(&bcts->dir);
1841         if (!status)
1842                 nfsd4_new_conn(rqstp, cstate->session, bcts->dir);
1843         return status;
1844 }
1845
1846 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1847 {
1848         if (!session)
1849                 return 0;
1850         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1851 }
1852
1853 __be32
1854 nfsd4_destroy_session(struct svc_rqst *r,
1855                       struct nfsd4_compound_state *cstate,
1856                       struct nfsd4_destroy_session *sessionid)
1857 {
1858         struct nfsd4_session *ses;
1859         __be32 status = nfserr_badsession;
1860
1861         /* Notes:
1862          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1863          * - Should we return nfserr_back_chan_busy if waiting for
1864          *   callbacks on to-be-destroyed session?
1865          * - Do we need to clear any callback info from previous session?
1866          */
1867
1868         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1869                 if (!nfsd4_last_compound_op(r))
1870                         return nfserr_not_only_op;
1871         }
1872         dump_sessionid(__func__, &sessionid->sessionid);
1873         spin_lock(&client_lock);
1874         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1875         if (!ses) {
1876                 spin_unlock(&client_lock);
1877                 goto out;
1878         }
1879
1880         unhash_session(ses);
1881         spin_unlock(&client_lock);
1882
1883         nfs4_lock_state();
1884         nfsd4_probe_callback_sync(ses->se_client);
1885         nfs4_unlock_state();
1886
1887         spin_lock(&client_lock);
1888         nfsd4_del_conns(ses);
1889         nfsd4_put_session_locked(ses);
1890         spin_unlock(&client_lock);
1891         status = nfs_ok;
1892 out:
1893         dprintk("%s returns %d\n", __func__, ntohl(status));
1894         return status;
1895 }
1896
1897 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1898 {
1899         struct nfsd4_conn *c;
1900
1901         list_for_each_entry(c, &s->se_conns, cn_persession) {
1902                 if (c->cn_xprt == xpt) {
1903                         return c;
1904                 }
1905         }
1906         return NULL;
1907 }
1908
1909 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1910 {
1911         struct nfs4_client *clp = ses->se_client;
1912         struct nfsd4_conn *c;
1913         int ret;
1914
1915         spin_lock(&clp->cl_lock);
1916         c = __nfsd4_find_conn(new->cn_xprt, ses);
1917         if (c) {
1918                 spin_unlock(&clp->cl_lock);
1919                 free_conn(new);
1920                 return;
1921         }
1922         __nfsd4_hash_conn(new, ses);
1923         spin_unlock(&clp->cl_lock);
1924         ret = nfsd4_register_conn(new);
1925         if (ret)
1926                 /* oops; xprt is already down: */
1927                 nfsd4_conn_lost(&new->cn_xpt_user);
1928         return;
1929 }
1930
1931 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
1932 {
1933         struct nfsd4_compoundargs *args = rqstp->rq_argp;
1934
1935         return args->opcnt > session->se_fchannel.maxops;
1936 }
1937
1938 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
1939                                   struct nfsd4_session *session)
1940 {
1941         struct xdr_buf *xb = &rqstp->rq_arg;
1942
1943         return xb->len > session->se_fchannel.maxreq_sz;
1944 }
1945
1946 __be32
1947 nfsd4_sequence(struct svc_rqst *rqstp,
1948                struct nfsd4_compound_state *cstate,
1949                struct nfsd4_sequence *seq)
1950 {
1951         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1952         struct nfsd4_session *session;
1953         struct nfsd4_slot *slot;
1954         struct nfsd4_conn *conn;
1955         __be32 status;
1956
1957         if (resp->opcnt != 1)
1958                 return nfserr_sequence_pos;
1959
1960         /*
1961          * Will be either used or freed by nfsd4_sequence_check_conn
1962          * below.
1963          */
1964         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
1965         if (!conn)
1966                 return nfserr_jukebox;
1967
1968         spin_lock(&client_lock);
1969         status = nfserr_badsession;
1970         session = find_in_sessionid_hashtbl(&seq->sessionid);
1971         if (!session)
1972                 goto out;
1973
1974         status = nfserr_too_many_ops;
1975         if (nfsd4_session_too_many_ops(rqstp, session))
1976                 goto out;
1977
1978         status = nfserr_req_too_big;
1979         if (nfsd4_request_too_big(rqstp, session))
1980                 goto out;
1981
1982         status = nfserr_badslot;
1983         if (seq->slotid >= session->se_fchannel.maxreqs)
1984                 goto out;
1985
1986         slot = session->se_slots[seq->slotid];
1987         dprintk("%s: slotid %d\n", __func__, seq->slotid);
1988
1989         /* We do not negotiate the number of slots yet, so set the
1990          * maxslots to the session maxreqs which is used to encode
1991          * sr_highest_slotid and the sr_target_slot id to maxslots */
1992         seq->maxslots = session->se_fchannel.maxreqs;
1993
1994         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
1995                                         slot->sl_flags & NFSD4_SLOT_INUSE);
1996         if (status == nfserr_replay_cache) {
1997                 status = nfserr_seq_misordered;
1998                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
1999                         goto out;
2000                 cstate->slot = slot;
2001                 cstate->session = session;
2002                 /* Return the cached reply status and set cstate->status
2003                  * for nfsd4_proc_compound processing */
2004                 status = nfsd4_replay_cache_entry(resp, seq);
2005                 cstate->status = nfserr_replay_cache;
2006                 goto out;
2007         }
2008         if (status)
2009                 goto out;
2010
2011         nfsd4_sequence_check_conn(conn, session);
2012         conn = NULL;
2013
2014         /* Success! bump slot seqid */
2015         slot->sl_seqid = seq->seqid;
2016         slot->sl_flags |= NFSD4_SLOT_INUSE;
2017         if (seq->cachethis)
2018                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2019         else
2020                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2021
2022         cstate->slot = slot;
2023         cstate->session = session;
2024
2025 out:
2026         /* Hold a session reference until done processing the compound. */
2027         if (cstate->session) {
2028                 struct nfs4_client *clp = session->se_client;
2029
2030                 nfsd4_get_session(cstate->session);
2031                 atomic_inc(&clp->cl_refcount);
2032                 switch (clp->cl_cb_state) {
2033                 case NFSD4_CB_DOWN:
2034                         seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2035                         break;
2036                 case NFSD4_CB_FAULT:
2037                         seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2038                         break;
2039                 default:
2040                         seq->status_flags = 0;
2041                 }
2042         }
2043         kfree(conn);
2044         spin_unlock(&client_lock);
2045         dprintk("%s: return %d\n", __func__, ntohl(status));
2046         return status;
2047 }
2048
2049 static inline bool has_resources(struct nfs4_client *clp)
2050 {
2051         return !list_empty(&clp->cl_openowners)
2052                 || !list_empty(&clp->cl_delegations)
2053                 || !list_empty(&clp->cl_sessions);
2054 }
2055
2056 __be32
2057 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2058 {
2059         struct nfs4_client *conf, *unconf, *clp;
2060         __be32 status = 0;
2061
2062         nfs4_lock_state();
2063         unconf = find_unconfirmed_client(&dc->clientid);
2064         conf = find_confirmed_client(&dc->clientid);
2065
2066         if (conf) {
2067                 clp = conf;
2068
2069                 if (!is_client_expired(conf) && has_resources(conf)) {
2070                         status = nfserr_clientid_busy;
2071                         goto out;
2072                 }
2073
2074                 /* rfc5661 18.50.3 */
2075                 if (cstate->session && conf == cstate->session->se_client) {
2076                         status = nfserr_clientid_busy;
2077                         goto out;
2078                 }
2079         } else if (unconf)
2080                 clp = unconf;
2081         else {
2082                 status = nfserr_stale_clientid;
2083                 goto out;
2084         }
2085
2086         expire_client(clp);
2087 out:
2088         nfs4_unlock_state();
2089         dprintk("%s return %d\n", __func__, ntohl(status));
2090         return status;
2091 }
2092
2093 __be32
2094 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2095 {
2096         __be32 status = 0;
2097
2098         if (rc->rca_one_fs) {
2099                 if (!cstate->current_fh.fh_dentry)
2100                         return nfserr_nofilehandle;
2101                 /*
2102                  * We don't take advantage of the rca_one_fs case.
2103                  * That's OK, it's optional, we can safely ignore it.
2104                  */
2105                  return nfs_ok;
2106         }
2107
2108         nfs4_lock_state();
2109         status = nfserr_complete_already;
2110         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2111                              &cstate->session->se_client->cl_flags))
2112                 goto out;
2113
2114         status = nfserr_stale_clientid;
2115         if (is_client_expired(cstate->session->se_client))
2116                 /*
2117                  * The following error isn't really legal.
2118                  * But we only get here if the client just explicitly
2119                  * destroyed the client.  Surely it no longer cares what
2120                  * error it gets back on an operation for the dead
2121                  * client.
2122                  */
2123                 goto out;
2124
2125         status = nfs_ok;
2126         nfsd4_client_record_create(cstate->session->se_client);
2127 out:
2128         nfs4_unlock_state();
2129         return status;
2130 }
2131
2132 __be32
2133 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2134                   struct nfsd4_setclientid *setclid)
2135 {
2136         struct xdr_netobj       clname = setclid->se_name;
2137         nfs4_verifier           clverifier = setclid->se_verf;
2138         unsigned int            strhashval;
2139         struct nfs4_client      *conf, *unconf, *new;
2140         __be32                  status;
2141         char                    dname[HEXDIR_LEN];
2142
2143         status = nfs4_make_rec_clidname(dname, &clname);
2144         if (status)
2145                 return status;
2146
2147         /*
2148          * XXX The Duplicate Request Cache (DRC) has been checked (??)
2149          * We get here on a DRC miss.
2150          */
2151
2152         strhashval = clientstr_hashval(dname);
2153
2154         nfs4_lock_state();
2155         conf = find_confirmed_client_by_str(dname, strhashval);
2156         if (conf) {
2157                 /* RFC 3530 14.2.33 CASE 0: */
2158                 status = nfserr_clid_inuse;
2159                 if (clp_used_exchangeid(conf))
2160                         goto out;
2161                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2162                         char addr_str[INET6_ADDRSTRLEN];
2163                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2164                                  sizeof(addr_str));
2165                         dprintk("NFSD: setclientid: string in use by client "
2166                                 "at %s\n", addr_str);
2167                         goto out;
2168                 }
2169         }
2170         /*
2171          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
2172          * has a description of SETCLIENTID request processing consisting
2173          * of 5 bullet points, labeled as CASE0 - CASE4 below.
2174          */
2175         unconf = find_unconfirmed_client_by_str(dname, strhashval);
2176         status = nfserr_jukebox;
2177         if (!conf) {
2178                 /*
2179                  * RFC 3530 14.2.33 CASE 4:
2180                  * placed first, because it is the normal case
2181                  */
2182                 if (unconf)
2183                         expire_client(unconf);
2184                 new = create_client(clname, dname, rqstp, &clverifier);
2185                 if (new == NULL)
2186                         goto out;
2187                 gen_clid(new);
2188         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
2189                 /*
2190                  * RFC 3530 14.2.33 CASE 1:
2191                  * probable callback update
2192                  */
2193                 if (unconf) {
2194                         /* Note this is removing unconfirmed {*x***},
2195                          * which is stronger than RFC recommended {vxc**}.
2196                          * This has the advantage that there is at most
2197                          * one {*x***} in either list at any time.
2198                          */
2199                         expire_client(unconf);
2200                 }
2201                 new = create_client(clname, dname, rqstp, &clverifier);
2202                 if (new == NULL)
2203                         goto out;
2204                 copy_clid(new, conf);
2205         } else if (!unconf) {
2206                 /*
2207                  * RFC 3530 14.2.33 CASE 2:
2208                  * probable client reboot; state will be removed if
2209                  * confirmed.
2210                  */
2211                 new = create_client(clname, dname, rqstp, &clverifier);
2212                 if (new == NULL)
2213                         goto out;
2214                 gen_clid(new);
2215         } else {
2216                 /*
2217                  * RFC 3530 14.2.33 CASE 3:
2218                  * probable client reboot; state will be removed if
2219                  * confirmed.
2220                  */
2221                 expire_client(unconf);
2222                 new = create_client(clname, dname, rqstp, &clverifier);
2223                 if (new == NULL)
2224                         goto out;
2225                 gen_clid(new);
2226         }
2227         /*
2228          * XXX: we should probably set this at creation time, and check
2229          * for consistent minorversion use throughout:
2230          */
2231         new->cl_minorversion = 0;
2232         gen_callback(new, setclid, rqstp);
2233         add_to_unconfirmed(new, strhashval);
2234         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2235         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2236         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2237         status = nfs_ok;
2238 out:
2239         nfs4_unlock_state();
2240         return status;
2241 }
2242
2243
2244 /*
2245  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
2246  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
2247  * bullets, labeled as CASE1 - CASE4 below.
2248  */
2249 __be32
2250 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2251                          struct nfsd4_compound_state *cstate,
2252                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2253 {
2254         struct nfs4_client *conf, *unconf;
2255         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2256         clientid_t * clid = &setclientid_confirm->sc_clientid;
2257         __be32 status;
2258
2259         if (STALE_CLIENTID(clid))
2260                 return nfserr_stale_clientid;
2261         /*
2262          * XXX The Duplicate Request Cache (DRC) has been checked (??)
2263          * We get here on a DRC miss.
2264          */
2265
2266         nfs4_lock_state();
2267
2268         conf = find_confirmed_client(clid);
2269         unconf = find_unconfirmed_client(clid);
2270
2271         /*
2272          * section 14.2.34 of RFC 3530 has a description of
2273          * SETCLIENTID_CONFIRM request processing consisting
2274          * of 4 bullet points, labeled as CASE1 - CASE4 below.
2275          */
2276         status = nfserr_clid_inuse;
2277         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
2278                 /*
2279                  * RFC 3530 14.2.34 CASE 1:
2280                  * callback update
2281                  */
2282                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
2283                         status = nfserr_clid_inuse;
2284                 else {
2285                         nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2286                         nfsd4_probe_callback(conf);
2287                         expire_client(unconf);
2288                         status = nfs_ok;
2289                 }
2290         } else if (conf && !unconf) {
2291                 /*
2292                  * RFC 3530 14.2.34 CASE 2:
2293                  * probable retransmitted request; play it safe and
2294                  * do nothing.
2295                  */
2296                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
2297                         status = nfserr_clid_inuse;
2298                 else
2299                         status = nfs_ok;
2300         } else if (!conf && unconf
2301                         && same_verf(&unconf->cl_confirm, &confirm)) {
2302                 /*
2303                  * RFC 3530 14.2.34 CASE 3:
2304                  * Normal case; new or rebooted client:
2305                  */
2306                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
2307                         status = nfserr_clid_inuse;
2308                 } else {
2309                         unsigned int hash =
2310                                 clientstr_hashval(unconf->cl_recdir);
2311                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
2312                                                             hash);
2313                         if (conf) {
2314                                 nfsd4_client_record_remove(conf);
2315                                 expire_client(conf);
2316                         }
2317                         move_to_confirmed(unconf);
2318                         conf = unconf;
2319                         nfsd4_probe_callback(conf);
2320                         status = nfs_ok;
2321                 }
2322         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
2323             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
2324                                                                 &confirm)))) {
2325                 /*
2326                  * RFC 3530 14.2.34 CASE 4:
2327                  * Client probably hasn't noticed that we rebooted yet.
2328                  */
2329                 status = nfserr_stale_clientid;
2330         }
2331
2332         nfs4_unlock_state();
2333         return status;
2334 }
2335
2336 static struct nfs4_file *nfsd4_alloc_file(void)
2337 {
2338         return kmem_cache_alloc(file_slab, GFP_KERNEL);
2339 }
2340
2341 /* OPEN Share state helper functions */
2342 static void nfsd4_init_file(struct nfs4_file *fp, struct inode *ino)
2343 {
2344         unsigned int hashval = file_hashval(ino);
2345
2346         atomic_set(&fp->fi_ref, 1);
2347         INIT_LIST_HEAD(&fp->fi_hash);
2348         INIT_LIST_HEAD(&fp->fi_stateids);
2349         INIT_LIST_HEAD(&fp->fi_delegations);
2350         fp->fi_inode = igrab(ino);
2351         fp->fi_had_conflict = false;
2352         fp->fi_lease = NULL;
2353         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2354         memset(fp->fi_access, 0, sizeof(fp->fi_access));
2355         spin_lock(&recall_lock);
2356         list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2357         spin_unlock(&recall_lock);
2358 }
2359
2360 static void
2361 nfsd4_free_slab(struct kmem_cache **slab)
2362 {
2363         if (*slab == NULL)
2364                 return;
2365         kmem_cache_destroy(*slab);
2366         *slab = NULL;
2367 }
2368
2369 void
2370 nfsd4_free_slabs(void)
2371 {
2372         nfsd4_free_slab(&openowner_slab);
2373         nfsd4_free_slab(&lockowner_slab);
2374         nfsd4_free_slab(&file_slab);
2375         nfsd4_free_slab(&stateid_slab);
2376         nfsd4_free_slab(&deleg_slab);
2377 }
2378
2379 int
2380 nfsd4_init_slabs(void)
2381 {
2382         openowner_slab = kmem_cache_create("nfsd4_openowners",
2383                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2384         if (openowner_slab == NULL)
2385                 goto out_nomem;
2386         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2387                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2388         if (lockowner_slab == NULL)
2389                 goto out_nomem;
2390         file_slab = kmem_cache_create("nfsd4_files",
2391                         sizeof(struct nfs4_file), 0, 0, NULL);
2392         if (file_slab == NULL)
2393                 goto out_nomem;
2394         stateid_slab = kmem_cache_create("nfsd4_stateids",
2395                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2396         if (stateid_slab == NULL)
2397                 goto out_nomem;
2398         deleg_slab = kmem_cache_create("nfsd4_delegations",
2399                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2400         if (deleg_slab == NULL)
2401                 goto out_nomem;
2402         return 0;
2403 out_nomem:
2404         nfsd4_free_slabs();
2405         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2406         return -ENOMEM;
2407 }
2408
2409 void nfs4_free_openowner(struct nfs4_openowner *oo)
2410 {
2411         kfree(oo->oo_owner.so_owner.data);
2412         kmem_cache_free(openowner_slab, oo);
2413 }
2414
2415 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2416 {
2417         kfree(lo->lo_owner.so_owner.data);
2418         kmem_cache_free(lockowner_slab, lo);
2419 }
2420
2421 static void init_nfs4_replay(struct nfs4_replay *rp)
2422 {
2423         rp->rp_status = nfserr_serverfault;
2424         rp->rp_buflen = 0;
2425         rp->rp_buf = rp->rp_ibuf;
2426 }
2427
2428 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2429 {
2430         struct nfs4_stateowner *sop;
2431
2432         sop = kmem_cache_alloc(slab, GFP_KERNEL);
2433         if (!sop)
2434                 return NULL;
2435
2436         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2437         if (!sop->so_owner.data) {
2438                 kmem_cache_free(slab, sop);
2439                 return NULL;
2440         }
2441         sop->so_owner.len = owner->len;
2442
2443         INIT_LIST_HEAD(&sop->so_stateids);
2444         sop->so_client = clp;
2445         init_nfs4_replay(&sop->so_replay);
2446         return sop;
2447 }
2448
2449 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2450 {
2451         list_add(&oo->oo_owner.so_strhash, &ownerstr_hashtbl[strhashval]);
2452         list_add(&oo->oo_perclient, &clp->cl_openowners);
2453 }
2454
2455 static struct nfs4_openowner *
2456 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2457         struct nfs4_openowner *oo;
2458
2459         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2460         if (!oo)
2461                 return NULL;
2462         oo->oo_owner.so_is_open_owner = 1;
2463         oo->oo_owner.so_seqid = open->op_seqid;
2464         oo->oo_flags = NFS4_OO_NEW;
2465         oo->oo_time = 0;
2466         oo->oo_last_closed_stid = NULL;
2467         INIT_LIST_HEAD(&oo->oo_close_lru);
2468         hash_openowner(oo, clp, strhashval);
2469         return oo;
2470 }
2471
2472 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2473         struct nfs4_openowner *oo = open->op_openowner;
2474         struct nfs4_client *clp = oo->oo_owner.so_client;
2475
2476         init_stid(&stp->st_stid, clp, NFS4_OPEN_STID);
2477         INIT_LIST_HEAD(&stp->st_lockowners);
2478         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2479         list_add(&stp->st_perfile, &fp->fi_stateids);
2480         stp->st_stateowner = &oo->oo_owner;
2481         get_nfs4_file(fp);
2482         stp->st_file = fp;
2483         stp->st_access_bmap = 0;
2484         stp->st_deny_bmap = 0;
2485         set_access(open->op_share_access, stp);
2486         set_deny(open->op_share_deny, stp);
2487         stp->st_openstp = NULL;
2488 }
2489
2490 static void
2491 move_to_close_lru(struct nfs4_openowner *oo)
2492 {
2493         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2494
2495         list_move_tail(&oo->oo_close_lru, &close_lru);
2496         oo->oo_time = get_seconds();
2497 }
2498
2499 static int
2500 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2501                                                         clientid_t *clid)
2502 {
2503         return (sop->so_owner.len == owner->len) &&
2504                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2505                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2506 }
2507
2508 static struct nfs4_openowner *
2509 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2510 {
2511         struct nfs4_stateowner *so;
2512         struct nfs4_openowner *oo;
2513
2514         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2515                 if (!so->so_is_open_owner)
2516                         continue;
2517                 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2518                         oo = openowner(so);
2519                         renew_client(oo->oo_owner.so_client);
2520                         return oo;
2521                 }
2522         }
2523         return NULL;
2524 }
2525
2526 /* search file_hashtbl[] for file */
2527 static struct nfs4_file *
2528 find_file(struct inode *ino)
2529 {
2530         unsigned int hashval = file_hashval(ino);
2531         struct nfs4_file *fp;
2532
2533         spin_lock(&recall_lock);
2534         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2535                 if (fp->fi_inode == ino) {
2536                         get_nfs4_file(fp);
2537                         spin_unlock(&recall_lock);
2538                         return fp;
2539                 }
2540         }
2541         spin_unlock(&recall_lock);
2542         return NULL;
2543 }
2544
2545 /*
2546  * Called to check deny when READ with all zero stateid or
2547  * WRITE with all zero or all one stateid
2548  */
2549 static __be32
2550 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2551 {
2552         struct inode *ino = current_fh->fh_dentry->d_inode;
2553         struct nfs4_file *fp;
2554         struct nfs4_ol_stateid *stp;
2555         __be32 ret;
2556
2557         dprintk("NFSD: nfs4_share_conflict\n");
2558
2559         fp = find_file(ino);
2560         if (!fp)
2561                 return nfs_ok;
2562         ret = nfserr_locked;
2563         /* Search for conflicting share reservations */
2564         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2565                 if (test_deny(deny_type, stp) ||
2566                     test_deny(NFS4_SHARE_DENY_BOTH, stp))
2567                         goto out;
2568         }
2569         ret = nfs_ok;
2570 out:
2571         put_nfs4_file(fp);
2572         return ret;
2573 }
2574
2575 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2576 {
2577         /* We're assuming the state code never drops its reference
2578          * without first removing the lease.  Since we're in this lease
2579          * callback (and since the lease code is serialized by the kernel
2580          * lock) we know the server hasn't removed the lease yet, we know
2581          * it's safe to take a reference: */
2582         atomic_inc(&dp->dl_count);
2583
2584         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2585
2586         /* only place dl_time is set. protected by lock_flocks*/
2587         dp->dl_time = get_seconds();
2588
2589         nfsd4_cb_recall(dp);
2590 }
2591
2592 /* Called from break_lease() with lock_flocks() held. */
2593 static void nfsd_break_deleg_cb(struct file_lock *fl)
2594 {
2595         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2596         struct nfs4_delegation *dp;
2597
2598         BUG_ON(!fp);
2599         /* We assume break_lease is only called once per lease: */
2600         BUG_ON(fp->fi_had_conflict);
2601         /*
2602          * We don't want the locks code to timeout the lease for us;
2603          * we'll remove it ourself if a delegation isn't returned
2604          * in time:
2605          */
2606         fl->fl_break_time = 0;
2607
2608         spin_lock(&recall_lock);
2609         fp->fi_had_conflict = true;
2610         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2611                 nfsd_break_one_deleg(dp);
2612         spin_unlock(&recall_lock);
2613 }
2614
2615 static
2616 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2617 {
2618         if (arg & F_UNLCK)
2619                 return lease_modify(onlist, arg);
2620         else
2621                 return -EAGAIN;
2622 }
2623
2624 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2625         .lm_break = nfsd_break_deleg_cb,
2626         .lm_change = nfsd_change_deleg_cb,
2627 };
2628
2629 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2630 {
2631         if (nfsd4_has_session(cstate))
2632                 return nfs_ok;
2633         if (seqid == so->so_seqid - 1)
2634                 return nfserr_replay_me;
2635         if (seqid == so->so_seqid)
2636                 return nfs_ok;
2637         return nfserr_bad_seqid;
2638 }
2639
2640 __be32
2641 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2642                     struct nfsd4_open *open)
2643 {
2644         clientid_t *clientid = &open->op_clientid;
2645         struct nfs4_client *clp = NULL;
2646         unsigned int strhashval;
2647         struct nfs4_openowner *oo = NULL;
2648         __be32 status;
2649
2650         if (STALE_CLIENTID(&open->op_clientid))
2651                 return nfserr_stale_clientid;
2652         /*
2653          * In case we need it later, after we've already created the
2654          * file and don't want to risk a further failure:
2655          */
2656         open->op_file = nfsd4_alloc_file();
2657         if (open->op_file == NULL)
2658                 return nfserr_jukebox;
2659
2660         strhashval = ownerstr_hashval(clientid->cl_id, &open->op_owner);
2661         oo = find_openstateowner_str(strhashval, open);
2662         open->op_openowner = oo;
2663         if (!oo) {
2664                 clp = find_confirmed_client(clientid);
2665                 if (clp == NULL)
2666                         return nfserr_expired;
2667                 goto new_owner;
2668         }
2669         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2670                 /* Replace unconfirmed owners without checking for replay. */
2671                 clp = oo->oo_owner.so_client;
2672                 release_openowner(oo);
2673                 open->op_openowner = NULL;
2674                 goto new_owner;
2675         }
2676         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2677         if (status)
2678                 return status;
2679         clp = oo->oo_owner.so_client;
2680         goto alloc_stateid;
2681 new_owner:
2682         oo = alloc_init_open_stateowner(strhashval, clp, open);
2683         if (oo == NULL)
2684                 return nfserr_jukebox;
2685         open->op_openowner = oo;
2686 alloc_stateid:
2687         open->op_stp = nfs4_alloc_stateid(clp);
2688         if (!open->op_stp)
2689                 return nfserr_jukebox;
2690         return nfs_ok;
2691 }
2692
2693 static inline __be32
2694 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2695 {
2696         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2697                 return nfserr_openmode;
2698         else
2699                 return nfs_ok;
2700 }
2701
2702 static int share_access_to_flags(u32 share_access)
2703 {
2704         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2705 }
2706
2707 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2708 {
2709         struct nfs4_stid *ret;
2710
2711         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2712         if (!ret)
2713                 return NULL;
2714         return delegstateid(ret);
2715 }
2716
2717 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
2718 {
2719         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
2720                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
2721 }
2722
2723 static __be32
2724 nfs4_check_deleg(struct nfs4_client *cl, struct nfs4_file *fp, struct nfsd4_open *open,
2725                 struct nfs4_delegation **dp)
2726 {
2727         int flags;
2728         __be32 status = nfserr_bad_stateid;
2729
2730         *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2731         if (*dp == NULL)
2732                 goto out;
2733         flags = share_access_to_flags(open->op_share_access);
2734         status = nfs4_check_delegmode(*dp, flags);
2735         if (status)
2736                 *dp = NULL;
2737 out:
2738         if (!nfsd4_is_deleg_cur(open))
2739                 return nfs_ok;
2740         if (status)
2741                 return status;
2742         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2743         return nfs_ok;
2744 }
2745
2746 static __be32
2747 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2748 {
2749         struct nfs4_ol_stateid *local;
2750         struct nfs4_openowner *oo = open->op_openowner;
2751
2752         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2753                 /* ignore lock owners */
2754                 if (local->st_stateowner->so_is_open_owner == 0)
2755                         continue;
2756                 /* remember if we have seen this open owner */
2757                 if (local->st_stateowner == &oo->oo_owner)
2758                         *stpp = local;
2759                 /* check for conflicting share reservations */
2760                 if (!test_share(local, open))
2761                         return nfserr_share_denied;
2762         }
2763         return nfs_ok;
2764 }
2765
2766 static void nfs4_free_stateid(struct nfs4_ol_stateid *s)
2767 {
2768         kmem_cache_free(stateid_slab, s);
2769 }
2770
2771 static inline int nfs4_access_to_access(u32 nfs4_access)
2772 {
2773         int flags = 0;
2774
2775         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2776                 flags |= NFSD_MAY_READ;
2777         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2778                 flags |= NFSD_MAY_WRITE;
2779         return flags;
2780 }
2781
2782 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2783                 struct svc_fh *cur_fh, struct nfsd4_open *open)
2784 {
2785         __be32 status;
2786         int oflag = nfs4_access_to_omode(open->op_share_access);
2787         int access = nfs4_access_to_access(open->op_share_access);
2788
2789         if (!fp->fi_fds[oflag]) {
2790                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2791                         &fp->fi_fds[oflag]);
2792                 if (status)
2793                         return status;
2794         }
2795         nfs4_file_get_access(fp, oflag);
2796
2797         return nfs_ok;
2798 }
2799
2800 static inline __be32
2801 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2802                 struct nfsd4_open *open)
2803 {
2804         struct iattr iattr = {
2805                 .ia_valid = ATTR_SIZE,
2806                 .ia_size = 0,
2807         };
2808         if (!open->op_truncate)
2809                 return 0;
2810         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2811                 return nfserr_inval;
2812         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2813 }
2814
2815 static __be32
2816 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)
2817 {
2818         u32 op_share_access = open->op_share_access;
2819         bool new_access;
2820         __be32 status;
2821
2822         new_access = !test_access(op_share_access, stp);
2823         if (new_access) {
2824                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2825                 if (status)
2826                         return status;
2827         }
2828         status = nfsd4_truncate(rqstp, cur_fh, open);
2829         if (status) {
2830                 if (new_access) {
2831                         int oflag = nfs4_access_to_omode(op_share_access);
2832                         nfs4_file_put_access(fp, oflag);
2833                 }
2834                 return status;
2835         }
2836         /* remember the open */
2837         set_access(op_share_access, stp);
2838         set_deny(open->op_share_deny, stp);
2839
2840         return nfs_ok;
2841 }
2842
2843
2844 static void
2845 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
2846 {
2847         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2848 }
2849
2850 /* Should we give out recallable state?: */
2851 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2852 {
2853         if (clp->cl_cb_state == NFSD4_CB_UP)
2854                 return true;
2855         /*
2856          * In the sessions case, since we don't have to establish a
2857          * separate connection for callbacks, we assume it's OK
2858          * until we hear otherwise:
2859          */
2860         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2861 }
2862
2863 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2864 {
2865         struct file_lock *fl;
2866
2867         fl = locks_alloc_lock();
2868         if (!fl)
2869                 return NULL;
2870         locks_init_lock(fl);
2871         fl->fl_lmops = &nfsd_lease_mng_ops;
2872         fl->fl_flags = FL_LEASE;
2873         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2874         fl->fl_end = OFFSET_MAX;
2875         fl->fl_owner = (fl_owner_t)(dp->dl_file);
2876         fl->fl_pid = current->tgid;
2877         return fl;
2878 }
2879
2880 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2881 {
2882         struct nfs4_file *fp = dp->dl_file;
2883         struct file_lock *fl;
2884         int status;
2885
2886         fl = nfs4_alloc_init_lease(dp, flag);
2887         if (!fl)
2888                 return -ENOMEM;
2889         fl->fl_file = find_readable_file(fp);
2890         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2891         status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2892         if (status) {
2893                 list_del_init(&dp->dl_perclnt);
2894                 locks_free_lock(fl);
2895                 return -ENOMEM;
2896         }
2897         fp->fi_lease = fl;
2898         fp->fi_deleg_file = fl->fl_file;
2899         get_file(fp->fi_deleg_file);
2900         atomic_set(&fp->fi_delegees, 1);
2901         list_add(&dp->dl_perfile, &fp->fi_delegations);
2902         return 0;
2903 }
2904
2905 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2906 {
2907         struct nfs4_file *fp = dp->dl_file;
2908
2909         if (!fp->fi_lease)
2910                 return nfs4_setlease(dp, flag);
2911         spin_lock(&recall_lock);
2912         if (fp->fi_had_conflict) {
2913                 spin_unlock(&recall_lock);
2914                 return -EAGAIN;
2915         }
2916         atomic_inc(&fp->fi_delegees);
2917         list_add(&dp->dl_perfile, &fp->fi_delegations);
2918         spin_unlock(&recall_lock);
2919         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2920         return 0;
2921 }
2922
2923 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
2924 {
2925         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2926         if (status == -EAGAIN)
2927                 open->op_why_no_deleg = WND4_CONTENTION;
2928         else {
2929                 open->op_why_no_deleg = WND4_RESOURCE;
2930                 switch (open->op_deleg_want) {
2931                 case NFS4_SHARE_WANT_READ_DELEG:
2932                 case NFS4_SHARE_WANT_WRITE_DELEG:
2933                 case NFS4_SHARE_WANT_ANY_DELEG:
2934                         break;
2935                 case NFS4_SHARE_WANT_CANCEL:
2936                         open->op_why_no_deleg = WND4_CANCELLED;
2937                         break;
2938                 case NFS4_SHARE_WANT_NO_DELEG:
2939                         BUG();  /* not supposed to get here */
2940                 }
2941         }
2942 }
2943
2944 /*
2945  * Attempt to hand out a delegation.
2946  */
2947 static void
2948 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
2949 {
2950         struct nfs4_delegation *dp;
2951         struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
2952         int cb_up;
2953         int status = 0, flag = 0;
2954
2955         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
2956         flag = NFS4_OPEN_DELEGATE_NONE;
2957         open->op_recall = 0;
2958         switch (open->op_claim_type) {
2959                 case NFS4_OPEN_CLAIM_PREVIOUS:
2960                         if (!cb_up)
2961                                 open->op_recall = 1;
2962                         flag = open->op_delegate_type;
2963                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2964                                 goto out;
2965                         break;
2966                 case NFS4_OPEN_CLAIM_NULL:
2967                         /* Let's not give out any delegations till everyone's
2968                          * had the chance to reclaim theirs.... */
2969                         if (locks_in_grace())
2970                                 goto out;
2971                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
2972                                 goto out;
2973                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2974                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2975                         else
2976                                 flag = NFS4_OPEN_DELEGATE_READ;
2977                         break;
2978                 default:
2979                         goto out;
2980         }
2981
2982         dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh, flag);
2983         if (dp == NULL)
2984                 goto out_no_deleg;
2985         status = nfs4_set_delegation(dp, flag);
2986         if (status)
2987                 goto out_free;
2988
2989         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
2990
2991         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2992                 STATEID_VAL(&dp->dl_stid.sc_stateid));
2993 out:
2994         open->op_delegate_type = flag;
2995         if (flag == NFS4_OPEN_DELEGATE_NONE) {
2996                 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
2997                     open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2998                         dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2999
3000                 /* 4.1 client asking for a delegation? */
3001                 if (open->op_deleg_want)
3002                         nfsd4_open_deleg_none_ext(open, status);
3003         }
3004         return;
3005 out_free:
3006         nfs4_put_delegation(dp);
3007 out_no_deleg:
3008         flag = NFS4_OPEN_DELEGATE_NONE;
3009         goto out;
3010 }
3011
3012 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
3013                                         struct nfs4_delegation *dp)
3014 {
3015         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
3016             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3017                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3018                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
3019         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
3020                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3021                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3022                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
3023         }
3024         /* Otherwise the client must be confused wanting a delegation
3025          * it already has, therefore we don't return
3026          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
3027          */
3028 }
3029
3030 /*
3031  * called with nfs4_lock_state() held.
3032  */
3033 __be32
3034 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
3035 {
3036         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3037         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
3038         struct nfs4_file *fp = NULL;
3039         struct inode *ino = current_fh->fh_dentry->d_inode;
3040         struct nfs4_ol_stateid *stp = NULL;
3041         struct nfs4_delegation *dp = NULL;
3042         __be32 status;
3043
3044         /*
3045          * Lookup file; if found, lookup stateid and check open request,
3046          * and check for delegations in the process of being recalled.
3047          * If not found, create the nfs4_file struct
3048          */
3049         fp = find_file(ino);
3050         if (fp) {
3051                 if ((status = nfs4_check_open(fp, open, &stp)))
3052                         goto out;
3053                 status = nfs4_check_deleg(cl, fp, open, &dp);
3054                 if (status)
3055                         goto out;
3056         } else {
3057                 status = nfserr_bad_stateid;
3058                 if (nfsd4_is_deleg_cur(open))
3059                         goto out;
3060                 status = nfserr_jukebox;
3061                 fp = open->op_file;
3062                 open->op_file = NULL;
3063                 nfsd4_init_file(fp, ino);
3064         }
3065
3066         /*
3067          * OPEN the file, or upgrade an existing OPEN.
3068          * If truncate fails, the OPEN fails.
3069          */
3070         if (stp) {
3071                 /* Stateid was found, this is an OPEN upgrade */
3072                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
3073                 if (status)
3074                         goto out;
3075         } else {
3076                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, open);
3077                 if (status)
3078                         goto out;
3079                 stp = open->op_stp;
3080                 open->op_stp = NULL;
3081                 init_open_stateid(stp, fp, open);
3082                 status = nfsd4_truncate(rqstp, current_fh, open);
3083                 if (status) {
3084                         release_open_stateid(stp);
3085                         goto out;
3086                 }
3087         }
3088         update_stateid(&stp->st_stid.sc_stateid);
3089         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3090
3091         if (nfsd4_has_session(&resp->cstate)) {
3092                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3093
3094                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
3095                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3096                         open->op_why_no_deleg = WND4_NOT_WANTED;
3097                         goto nodeleg;
3098                 }
3099         }
3100
3101         /*
3102         * Attempt to hand out a delegation. No error return, because the
3103         * OPEN succeeds even if we fail.
3104         */
3105         nfs4_open_delegation(current_fh, open, stp);
3106 nodeleg:
3107         status = nfs_ok;
3108
3109         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
3110                 STATEID_VAL(&stp->st_stid.sc_stateid));
3111 out:
3112         /* 4.1 client trying to upgrade/downgrade delegation? */
3113         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
3114             open->op_deleg_want)
3115                 nfsd4_deleg_xgrade_none_ext(open, dp);
3116
3117         if (fp)
3118                 put_nfs4_file(fp);
3119         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
3120                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
3121         /*
3122         * To finish the open response, we just need to set the rflags.
3123         */
3124         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
3125         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
3126             !nfsd4_has_session(&resp->cstate))
3127                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
3128
3129         return status;
3130 }
3131
3132 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
3133 {
3134         if (open->op_openowner) {
3135                 struct nfs4_openowner *oo = open->op_openowner;
3136
3137                 if (!list_empty(&oo->oo_owner.so_stateids))
3138                         list_del_init(&oo->oo_close_lru);
3139                 if (oo->oo_flags & NFS4_OO_NEW) {
3140                         if (status) {
3141                                 release_openowner(oo);
3142                                 open->op_openowner = NULL;
3143                         } else
3144                                 oo->oo_flags &= ~NFS4_OO_NEW;
3145                 }
3146         }
3147         if (open->op_file)
3148                 nfsd4_free_file(open->op_file);
3149         if (open->op_stp)
3150                 nfs4_free_stateid(open->op_stp);
3151 }
3152
3153 __be32
3154 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3155             clientid_t *clid)
3156 {
3157         struct nfs4_client *clp;
3158         __be32 status;
3159
3160         nfs4_lock_state();
3161         dprintk("process_renew(%08x/%08x): starting\n",
3162                         clid->cl_boot, clid->cl_id);
3163         status = nfserr_stale_clientid;
3164         if (STALE_CLIENTID(clid))
3165                 goto out;
3166         clp = find_confirmed_client(clid);
3167         status = nfserr_expired;
3168         if (clp == NULL) {
3169                 /* We assume the client took too long to RENEW. */
3170                 dprintk("nfsd4_renew: clientid not found!\n");
3171                 goto out;
3172         }
3173         status = nfserr_cb_path_down;
3174         if (!list_empty(&clp->cl_delegations)
3175                         && clp->cl_cb_state != NFSD4_CB_UP)
3176                 goto out;
3177         status = nfs_ok;
3178 out:
3179         nfs4_unlock_state();
3180         return status;
3181 }
3182
3183 static struct lock_manager nfsd4_manager = {
3184 };
3185
3186 static bool grace_ended;
3187
3188 static void
3189 nfsd4_end_grace(void)
3190 {
3191         /* do nothing if grace period already ended */
3192         if (grace_ended)
3193                 return;
3194
3195         dprintk("NFSD: end of grace period\n");
3196         grace_ended = true;
3197         nfsd4_record_grace_done(&init_net, boot_time);
3198         locks_end_grace(&nfsd4_manager);
3199         /*
3200          * Now that every NFSv4 client has had the chance to recover and
3201          * to see the (possibly new, possibly shorter) lease time, we
3202          * can safely set the next grace time to the current lease time:
3203          */
3204         nfsd4_grace = nfsd4_lease;
3205 }
3206
3207 static time_t
3208 nfs4_laundromat(void)
3209 {
3210         struct nfs4_client *clp;
3211         struct nfs4_openowner *oo;
3212         struct nfs4_delegation *dp;
3213         struct list_head *pos, *next, reaplist;
3214         time_t cutoff = get_seconds() - nfsd4_lease;
3215         time_t t, clientid_val = nfsd4_lease;
3216         time_t u, test_val = nfsd4_lease;
3217
3218         nfs4_lock_state();
3219
3220         dprintk("NFSD: laundromat service - starting\n");
3221         nfsd4_end_grace();
3222         INIT_LIST_HEAD(&reaplist);
3223         spin_lock(&client_lock);
3224         list_for_each_safe(pos, next, &client_lru) {
3225                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3226                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3227                         t = clp->cl_time - cutoff;
3228                         if (clientid_val > t)
3229                                 clientid_val = t;
3230                         break;
3231                 }
3232                 if (atomic_read(&clp->cl_refcount)) {
3233                         dprintk("NFSD: client in use (clientid %08x)\n",
3234                                 clp->cl_clientid.cl_id);
3235                         continue;
3236                 }
3237                 unhash_client_locked(clp);
3238                 list_add(&clp->cl_lru, &reaplist);
3239         }
3240         spin_unlock(&client_lock);
3241         list_for_each_safe(pos, next, &reaplist) {
3242                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3243                 dprintk("NFSD: purging unused client (clientid %08x)\n",
3244                         clp->cl_clientid.cl_id);
3245                 nfsd4_client_record_remove(clp);
3246                 expire_client(clp);
3247         }
3248         spin_lock(&recall_lock);
3249         list_for_each_safe(pos, next, &del_recall_lru) {
3250                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3251                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3252                         u = dp->dl_time - cutoff;
3253                         if (test_val > u)
3254                                 test_val = u;
3255                         break;
3256                 }
3257                 list_move(&dp->dl_recall_lru, &reaplist);
3258         }
3259         spin_unlock(&recall_lock);
3260         list_for_each_safe(pos, next, &reaplist) {
3261                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3262                 unhash_delegation(dp);
3263         }
3264         test_val = nfsd4_lease;
3265         list_for_each_safe(pos, next, &close_lru) {
3266                 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3267                 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3268                         u = oo->oo_time - cutoff;
3269                         if (test_val > u)
3270                                 test_val = u;
3271                         break;
3272                 }
3273                 release_openowner(oo);
3274         }
3275         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3276                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3277         nfs4_unlock_state();
3278         return clientid_val;
3279 }
3280
3281 static struct workqueue_struct *laundry_wq;
3282 static void laundromat_main(struct work_struct *);
3283 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
3284
3285 static void
3286 laundromat_main(struct work_struct *not_used)
3287 {
3288         time_t t;
3289
3290         t = nfs4_laundromat();
3291         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3292         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
3293 }
3294
3295 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3296 {
3297         if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3298                 return nfserr_bad_stateid;
3299         return nfs_ok;
3300 }
3301
3302 static int
3303 STALE_STATEID(stateid_t *stateid)
3304 {
3305         if (stateid->si_opaque.so_clid.cl_boot == boot_time)
3306                 return 0;
3307         dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
3308                 STATEID_VAL(stateid));
3309         return 1;
3310 }
3311
3312 static inline int
3313 access_permit_read(struct nfs4_ol_stateid *stp)
3314 {
3315         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
3316                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
3317                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
3318 }
3319
3320 static inline int
3321 access_permit_write(struct nfs4_ol_stateid *stp)
3322 {
3323         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
3324                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
3325 }
3326
3327 static
3328 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3329 {
3330         __be32 status = nfserr_openmode;
3331
3332         /* For lock stateid's, we test the parent open, not the lock: */
3333         if (stp->st_openstp)
3334                 stp = stp->st_openstp;
3335         if ((flags & WR_STATE) && !access_permit_write(stp))
3336                 goto out;
3337         if ((flags & RD_STATE) && !access_permit_read(stp))
3338                 goto out;
3339         status = nfs_ok;
3340 out:
3341         return status;
3342 }
3343
3344 static inline __be32
3345 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
3346 {
3347         if (ONE_STATEID(stateid) && (flags & RD_STATE))
3348                 return nfs_ok;
3349         else if (locks_in_grace()) {
3350                 /* Answer in remaining cases depends on existence of
3351                  * conflicting state; so we must wait out the grace period. */
3352                 return nfserr_grace;
3353         } else if (flags & WR_STATE)
3354                 return nfs4_share_conflict(current_fh,
3355                                 NFS4_SHARE_DENY_WRITE);
3356         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3357                 return nfs4_share_conflict(current_fh,
3358                                 NFS4_SHARE_DENY_READ);
3359 }
3360
3361 /*
3362  * Allow READ/WRITE during grace period on recovered state only for files
3363  * that are not able to provide mandatory locking.
3364  */
3365 static inline int
3366 grace_disallows_io(struct inode *inode)
3367 {
3368         return locks_in_grace() && mandatory_lock(inode);
3369 }
3370
3371 /* Returns true iff a is later than b: */
3372 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3373 {
3374         return (s32)a->si_generation - (s32)b->si_generation > 0;
3375 }
3376
3377 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3378 {
3379         /*
3380          * When sessions are used the stateid generation number is ignored
3381          * when it is zero.
3382          */
3383         if (has_session && in->si_generation == 0)
3384                 return nfs_ok;
3385
3386         if (in->si_generation == ref->si_generation)
3387                 return nfs_ok;
3388
3389         /* If the client sends us a stateid from the future, it's buggy: */
3390         if (stateid_generation_after(in, ref))
3391                 return nfserr_bad_stateid;
3392         /*
3393          * However, we could see a stateid from the past, even from a
3394          * non-buggy client.  For example, if the client sends a lock
3395          * while some IO is outstanding, the lock may bump si_generation
3396          * while the IO is still in flight.  The client could avoid that
3397          * situation by waiting for responses on all the IO requests,
3398          * but better performance may result in retrying IO that
3399          * receives an old_stateid error if requests are rarely
3400          * reordered in flight:
3401          */
3402         return nfserr_old_stateid;
3403 }
3404
3405 __be32 nfs4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
3406 {
3407         struct nfs4_stid *s;
3408         struct nfs4_ol_stateid *ols;
3409         __be32 status;
3410
3411         if (STALE_STATEID(stateid))
3412                 return nfserr_stale_stateid;
3413
3414         s = find_stateid(cl, stateid);
3415         if (!s)
3416                  return nfserr_stale_stateid;
3417         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3418         if (status)
3419                 return status;
3420         if (!(s->sc_type & (NFS4_OPEN_STID | NFS4_LOCK_STID)))
3421                 return nfs_ok;
3422         ols = openlockstateid(s);
3423         if (ols->st_stateowner->so_is_open_owner
3424             && !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3425                 return nfserr_bad_stateid;
3426         return nfs_ok;
3427 }
3428
3429 static __be32 nfsd4_lookup_stateid(stateid_t *stateid, unsigned char typemask, struct nfs4_stid **s)
3430 {
3431         struct nfs4_client *cl;
3432
3433         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3434                 return nfserr_bad_stateid;
3435         if (STALE_STATEID(stateid))
3436                 return nfserr_stale_stateid;
3437         cl = find_confirmed_client(&stateid->si_opaque.so_clid);
3438         if (!cl)
3439                 return nfserr_expired;
3440         *s = find_stateid_by_type(cl, stateid, typemask);
3441         if (!*s)
3442                 return nfserr_bad_stateid;
3443         return nfs_ok;
3444
3445 }
3446
3447 /*
3448 * Checks for stateid operations
3449 */
3450 __be32
3451 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
3452                            stateid_t *stateid, int flags, struct file **filpp)
3453 {
3454         struct nfs4_stid *s;
3455         struct nfs4_ol_stateid *stp = NULL;
3456         struct nfs4_delegation *dp = NULL;
3457         struct svc_fh *current_fh = &cstate->current_fh;
3458         struct inode *ino = current_fh->fh_dentry->d_inode;
3459         __be32 status;
3460
3461         if (filpp)
3462                 *filpp = NULL;
3463
3464         if (grace_disallows_io(ino))
3465                 return nfserr_grace;
3466
3467         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3468                 return check_special_stateids(current_fh, stateid, flags);
3469
3470         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID, &s);
3471         if (status)
3472                 return status;
3473         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
3474         if (status)
3475                 goto out;
3476         switch (s->sc_type) {
3477         case NFS4_DELEG_STID:
3478                 dp = delegstateid(s);
3479                 status = nfs4_check_delegmode(dp, flags);
3480                 if (status)
3481                         goto out;
3482                 if (filpp) {
3483                         *filpp = dp->dl_file->fi_deleg_file;
3484                         BUG_ON(!*filpp);
3485                 }
3486                 break;
3487         case NFS4_OPEN_STID:
3488         case NFS4_LOCK_STID:
3489                 stp = openlockstateid(s);
3490                 status = nfs4_check_fh(current_fh, stp);
3491                 if (status)
3492                         goto out;
3493                 if (stp->st_stateowner->so_is_open_owner
3494                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3495                         goto out;
3496                 status = nfs4_check_openmode(stp, flags);
3497                 if (status)
3498                         goto out;
3499                 if (filpp) {
3500                         if (flags & RD_STATE)
3501                                 *filpp = find_readable_file(stp->st_file);
3502                         else
3503                                 *filpp = find_writeable_file(stp->st_file);
3504                 }
3505                 break;
3506         default:
3507                 return nfserr_bad_stateid;
3508         }
3509         status = nfs_ok;
3510 out:
3511         return status;
3512 }
3513
3514 static __be32
3515 nfsd4_free_lock_stateid(struct nfs4_ol_stateid *stp)
3516 {
3517         if (check_for_locks(stp->st_file, lockowner(stp->st_stateowner)))
3518                 return nfserr_locks_held;
3519         release_lock_stateid(stp);
3520         return nfs_ok;
3521 }
3522
3523 /*
3524  * Test if the stateid is valid
3525  */
3526 __be32
3527 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3528                    struct nfsd4_test_stateid *test_stateid)
3529 {
3530         struct nfsd4_test_stateid_id *stateid;
3531         struct nfs4_client *cl = cstate->session->se_client;
3532
3533         nfs4_lock_state();
3534         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
3535                 stateid->ts_id_status = nfs4_validate_stateid(cl, &stateid->ts_id_stateid);
3536         nfs4_unlock_state();
3537
3538         return nfs_ok;
3539 }
3540
3541 __be32
3542 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3543                    struct nfsd4_free_stateid *free_stateid)
3544 {
3545         stateid_t *stateid = &free_stateid->fr_stateid;
3546         struct nfs4_stid *s;
3547         struct nfs4_client *cl = cstate->session->se_client;
3548         __be32 ret = nfserr_bad_stateid;
3549
3550         nfs4_lock_state();
3551         s = find_stateid(cl, stateid);
3552         if (!s)
3553                 goto out;
3554         switch (s->sc_type) {
3555         case NFS4_DELEG_STID:
3556                 ret = nfserr_locks_held;
3557                 goto out;
3558         case NFS4_OPEN_STID:
3559         case NFS4_LOCK_STID:
3560                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
3561                 if (ret)
3562                         goto out;
3563                 if (s->sc_type == NFS4_LOCK_STID)
3564                         ret = nfsd4_free_lock_stateid(openlockstateid(s));
3565                 else
3566                         ret = nfserr_locks_held;
3567                 break;
3568         default:
3569                 ret = nfserr_bad_stateid;
3570         }
3571 out:
3572         nfs4_unlock_state();
3573         return ret;
3574 }
3575
3576 static inline int
3577 setlkflg (int type)
3578 {
3579         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3580                 RD_STATE : WR_STATE;
3581 }
3582
3583 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
3584 {
3585         struct svc_fh *current_fh = &cstate->current_fh;
3586         struct nfs4_stateowner *sop = stp->st_stateowner;
3587         __be32 status;
3588
3589         status = nfsd4_check_seqid(cstate, sop, seqid);
3590         if (status)
3591                 return status;
3592         if (stp->st_stid.sc_type == NFS4_CLOSED_STID)
3593                 /*
3594                  * "Closed" stateid's exist *only* to return
3595                  * nfserr_replay_me from the previous step.
3596                  */
3597                 return nfserr_bad_stateid;
3598         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
3599         if (status)
3600                 return status;
3601         return nfs4_check_fh(current_fh, stp);
3602 }
3603
3604 /*
3605  * Checks for sequence id mutating operations.
3606  */
3607 static __be32
3608 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3609                          stateid_t *stateid, char typemask,
3610                          struct nfs4_ol_stateid **stpp)
3611 {
3612         __be32 status;
3613         struct nfs4_stid *s;
3614
3615         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3616                 seqid, STATEID_VAL(stateid));
3617
3618         *stpp = NULL;
3619         status = nfsd4_lookup_stateid(stateid, typemask, &s);
3620         if (status)
3621                 return status;
3622         *stpp = openlockstateid(s);
3623         cstate->replay_owner = (*stpp)->st_stateowner;
3624
3625         return nfs4_seqid_op_checks(cstate, stateid, seqid, *stpp);
3626 }
3627
3628 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid, stateid_t *stateid, struct nfs4_ol_stateid **stpp)
3629 {
3630         __be32 status;
3631         struct nfs4_openowner *oo;
3632
3633         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
3634                                                 NFS4_OPEN_STID, stpp);
3635         if (status)
3636                 return status;
3637         oo = openowner((*stpp)->st_stateowner);
3638         if (!(oo->oo_flags & NFS4_OO_CONFIRMED))
3639                 return nfserr_bad_stateid;
3640         return nfs_ok;
3641 }
3642
3643 __be32
3644 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3645                    struct nfsd4_open_confirm *oc)
3646 {
3647         __be32 status;
3648         struct nfs4_openowner *oo;
3649         struct nfs4_ol_stateid *stp;
3650
3651         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3652                         (int)cstate->current_fh.fh_dentry->d_name.len,
3653                         cstate->current_fh.fh_dentry->d_name.name);
3654
3655         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3656         if (status)
3657                 return status;
3658
3659         nfs4_lock_state();
3660
3661         status = nfs4_preprocess_seqid_op(cstate,
3662                                         oc->oc_seqid, &oc->oc_req_stateid,
3663                                         NFS4_OPEN_STID, &stp);
3664         if (status)
3665                 goto out;
3666         oo = openowner(stp->st_stateowner);
3667         status = nfserr_bad_stateid;
3668         if (oo->oo_flags & NFS4_OO_CONFIRMED)
3669                 goto out;
3670         oo->oo_flags |= NFS4_OO_CONFIRMED;
3671         update_stateid(&stp->st_stid.sc_stateid);
3672         memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3673         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3674                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
3675
3676         nfsd4_client_record_create(oo->oo_owner.so_client);
3677         status = nfs_ok;
3678 out:
3679         if (!cstate->replay_owner)
3680                 nfs4_unlock_state();
3681         return status;
3682 }
3683
3684 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
3685 {
3686         if (!test_access(access, stp))
3687                 return;
3688         nfs4_file_put_access(stp->st_file, nfs4_access_to_omode(access));
3689         clear_access(access, stp);
3690 }
3691
3692 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
3693 {
3694         switch (to_access) {
3695         case NFS4_SHARE_ACCESS_READ:
3696                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
3697                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3698                 break;
3699         case NFS4_SHARE_ACCESS_WRITE:
3700                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
3701                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3702                 break;
3703         case NFS4_SHARE_ACCESS_BOTH:
3704                 break;
3705         default:
3706                 BUG();
3707         }
3708 }
3709
3710 static void
3711 reset_union_bmap_deny(unsigned long deny, struct nfs4_ol_stateid *stp)
3712 {
3713         int i;
3714         for (i = 0; i < 4; i++) {
3715                 if ((i & deny) != i)
3716                         clear_deny(i, stp);
3717         }
3718 }
3719
3720 __be32
3721 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3722                      struct nfsd4_compound_state *cstate,
3723                      struct nfsd4_open_downgrade *od)
3724 {
3725         __be32 status;
3726         struct nfs4_ol_stateid *stp;
3727
3728         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3729                         (int)cstate->current_fh.fh_dentry->d_name.len,
3730                         cstate->current_fh.fh_dentry->d_name.name);
3731
3732         /* We don't yet support WANT bits: */
3733         if (od->od_deleg_want)
3734                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
3735                         od->od_deleg_want);
3736
3737         nfs4_lock_state();
3738         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
3739                                         &od->od_stateid, &stp);
3740         if (status)
3741                 goto out;
3742         status = nfserr_inval;
3743         if (!test_access(od->od_share_access, stp)) {
3744                 dprintk("NFSD: access not a subset current bitmap: 0x%lx, input access=%08x\n",
3745                         stp->st_access_bmap, od->od_share_access);
3746                 goto out;
3747         }
3748         if (!test_deny(od->od_share_deny, stp)) {
3749                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3750                         stp->st_deny_bmap, od->od_share_deny);
3751                 goto out;
3752         }
3753         nfs4_stateid_downgrade(stp, od->od_share_access);
3754
3755         reset_union_bmap_deny(od->od_share_deny, stp);
3756
3757         update_stateid(&stp->st_stid.sc_stateid);
3758         memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3759         status = nfs_ok;
3760 out:
3761         if (!cstate->replay_owner)
3762                 nfs4_unlock_state();
3763         return status;
3764 }
3765
3766 void nfsd4_purge_closed_stateid(struct nfs4_stateowner *so)
3767 {
3768         struct nfs4_openowner *oo;
3769         struct nfs4_ol_stateid *s;
3770
3771         if (!so->so_is_open_owner)
3772                 return;
3773         oo = openowner(so);
3774         s = oo->oo_last_closed_stid;
3775         if (!s)
3776                 return;
3777         if (!(oo->oo_flags & NFS4_OO_PURGE_CLOSE)) {
3778                 /* Release the last_closed_stid on the next seqid bump: */
3779                 oo->oo_flags |= NFS4_OO_PURGE_CLOSE;
3780                 return;
3781         }
3782         oo->oo_flags &= ~NFS4_OO_PURGE_CLOSE;
3783         release_last_closed_stateid(oo);
3784 }
3785
3786 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
3787 {
3788         unhash_open_stateid(s);
3789         s->st_stid.sc_type = NFS4_CLOSED_STID;
3790 }
3791
3792 /*
3793  * nfs4_unlock_state() called after encode
3794  */
3795 __be32
3796 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3797             struct nfsd4_close *close)
3798 {
3799         __be32 status;
3800         struct nfs4_openowner *oo;
3801         struct nfs4_ol_stateid *stp;
3802
3803         dprintk("NFSD: nfsd4_close on file %.*s\n",
3804                         (int)cstate->current_fh.fh_dentry->d_name.len,
3805                         cstate->current_fh.fh_dentry->d_name.name);
3806
3807         nfs4_lock_state();
3808         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
3809                                         &close->cl_stateid,
3810                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
3811                                         &stp);
3812         if (status)
3813                 goto out;
3814         oo = openowner(stp->st_stateowner);
3815         status = nfs_ok;
3816         update_stateid(&stp->st_stid.sc_stateid);
3817         memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3818
3819         nfsd4_close_open_stateid(stp);
3820         oo->oo_last_closed_stid = stp;
3821
3822         /* place unused nfs4_stateowners on so_close_lru list to be
3823          * released by the laundromat service after the lease period
3824          * to enable us to handle CLOSE replay
3825          */
3826         if (list_empty(&oo->oo_owner.so_stateids))
3827                 move_to_close_lru(oo);
3828 out:
3829         if (!cstate->replay_owner)
3830                 nfs4_unlock_state();
3831         return status;
3832 }
3833
3834 __be32
3835 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3836                   struct nfsd4_delegreturn *dr)
3837 {
3838         struct nfs4_delegation *dp;
3839         stateid_t *stateid = &dr->dr_stateid;
3840         struct nfs4_stid *s;
3841         struct inode *inode;
3842         __be32 status;
3843
3844         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3845                 return status;
3846         inode = cstate->current_fh.fh_dentry->d_inode;
3847
3848         nfs4_lock_state();
3849         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID, &s);
3850         if (status)
3851                 goto out;
3852         dp = delegstateid(s);
3853         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
3854         if (status)
3855                 goto out;
3856
3857         unhash_delegation(dp);
3858 out:
3859         nfs4_unlock_state();
3860
3861         return status;
3862 }
3863
3864
3865 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3866
3867 #define LOCKOWNER_INO_HASH_BITS 8
3868 #define LOCKOWNER_INO_HASH_SIZE (1 << LOCKOWNER_INO_HASH_BITS)
3869 #define LOCKOWNER_INO_HASH_MASK (LOCKOWNER_INO_HASH_SIZE - 1)
3870
3871 static inline u64
3872 end_offset(u64 start, u64 len)
3873 {
3874         u64 end;
3875
3876         end = start + len;
3877         return end >= start ? end: NFS4_MAX_UINT64;
3878 }
3879
3880 /* last octet in a range */
3881 static inline u64
3882 last_byte_offset(u64 start, u64 len)
3883 {
3884         u64 end;
3885
3886         BUG_ON(!len);
3887         end = start + len;
3888         return end > start ? end - 1: NFS4_MAX_UINT64;
3889 }
3890
3891 static unsigned int lockowner_ino_hashval(struct inode *inode, u32 cl_id, struct xdr_netobj *ownername)
3892 {
3893         return (file_hashval(inode) + cl_id
3894                         + opaque_hashval(ownername->data, ownername->len))
3895                 & LOCKOWNER_INO_HASH_MASK;
3896 }
3897
3898 static struct list_head lockowner_ino_hashtbl[LOCKOWNER_INO_HASH_SIZE];
3899
3900 /*
3901  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3902  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3903  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3904  * locking, this prevents us from being completely protocol-compliant.  The
3905  * real solution to this problem is to start using unsigned file offsets in
3906  * the VFS, but this is a very deep change!
3907  */
3908 static inline void
3909 nfs4_transform_lock_offset(struct file_lock *lock)
3910 {
3911         if (lock->fl_start < 0)
3912                 lock->fl_start = OFFSET_MAX;
3913         if (lock->fl_end < 0)
3914                 lock->fl_end = OFFSET_MAX;
3915 }
3916
3917 /* Hack!: For now, we're defining this just so we can use a pointer to it
3918  * as a unique cookie to identify our (NFSv4's) posix locks. */
3919 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
3920 };
3921
3922 static inline void
3923 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3924 {
3925         struct nfs4_lockowner *lo;
3926
3927         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3928                 lo = (struct nfs4_lockowner *) fl->fl_owner;
3929                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
3930                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
3931                 if (!deny->ld_owner.data)
3932                         /* We just don't care that much */
3933                         goto nevermind;
3934                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
3935                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
3936         } else {
3937 nevermind:
3938                 deny->ld_owner.len = 0;
3939                 deny->ld_owner.data = NULL;
3940                 deny->ld_clientid.cl_boot = 0;
3941                 deny->ld_clientid.cl_id = 0;
3942         }
3943         deny->ld_start = fl->fl_start;
3944         deny->ld_length = NFS4_MAX_UINT64;
3945         if (fl->fl_end != NFS4_MAX_UINT64)
3946                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3947         deny->ld_type = NFS4_READ_LT;
3948         if (fl->fl_type != F_RDLCK)
3949                 deny->ld_type = NFS4_WRITE_LT;
3950 }
3951
3952 static bool same_lockowner_ino(struct nfs4_lockowner *lo, struct inode *inode, clientid_t *clid, struct xdr_netobj *owner)
3953 {
3954         struct nfs4_ol_stateid *lst;
3955
3956         if (!same_owner_str(&lo->lo_owner, owner, clid))
3957                 return false;
3958         lst = list_first_entry(&lo->lo_owner.so_stateids,
3959                                struct nfs4_ol_stateid, st_perstateowner);
3960         return lst->st_file->fi_inode == inode;
3961 }
3962
3963 static struct nfs4_lockowner *
3964 find_lockowner_str(struct inode *inode, clientid_t *clid,
3965                 struct xdr_netobj *owner)
3966 {
3967         unsigned int hashval = lockowner_ino_hashval(inode, clid->cl_id, owner);
3968         struct nfs4_lockowner *lo;
3969
3970         list_for_each_entry(lo, &lockowner_ino_hashtbl[hashval], lo_owner_ino_hash) {
3971                 if (same_lockowner_ino(lo, inode, clid, owner))
3972                         return lo;
3973         }
3974         return NULL;
3975 }
3976
3977 static void hash_lockowner(struct nfs4_lockowner *lo, unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp)
3978 {
3979         struct inode *inode = open_stp->st_file->fi_inode;
3980         unsigned int inohash = lockowner_ino_hashval(inode,
3981                         clp->cl_clientid.cl_id, &lo->lo_owner.so_owner);
3982
3983         list_add(&lo->lo_owner.so_strhash, &ownerstr_hashtbl[strhashval]);
3984         list_add(&lo->lo_owner_ino_hash, &lockowner_ino_hashtbl[inohash]);
3985         list_add(&lo->lo_perstateid, &open_stp->st_lockowners);
3986 }
3987
3988 /*
3989  * Alloc a lock owner structure.
3990  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3991  * occurred.
3992  *
3993  * strhashval = ownerstr_hashval
3994  */
3995
3996 static struct nfs4_lockowner *
3997 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp, struct nfsd4_lock *lock) {
3998         struct nfs4_lockowner *lo;
3999
4000         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
4001         if (!lo)
4002                 return NULL;
4003         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
4004         lo->lo_owner.so_is_open_owner = 0;
4005         /* It is the openowner seqid that will be incremented in encode in the
4006          * case of new lockowners; so increment the lock seqid manually: */
4007         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid + 1;
4008         hash_lockowner(lo, strhashval, clp, open_stp);
4009         return lo;
4010 }
4011
4012 static struct nfs4_ol_stateid *
4013 alloc_init_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp, struct nfs4_ol_stateid *open_stp)
4014 {
4015         struct nfs4_ol_stateid *stp;
4016         struct nfs4_client *clp = lo->lo_owner.so_client;
4017
4018         stp = nfs4_alloc_stateid(clp);
4019         if (stp == NULL)
4020                 return NULL;
4021         init_stid(&stp->st_stid, clp, NFS4_LOCK_STID);
4022         list_add(&stp->st_perfile, &fp->fi_stateids);
4023         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
4024         stp->st_stateowner = &lo->lo_owner;
4025         get_nfs4_file(fp);
4026         stp->st_file = fp;
4027         stp->st_access_bmap = 0;
4028         stp->st_deny_bmap = open_stp->st_deny_bmap;
4029         stp->st_openstp = open_stp;
4030         return stp;
4031 }
4032
4033 static int
4034 check_lock_length(u64 offset, u64 length)
4035 {
4036         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
4037              LOFF_OVERFLOW(offset, length)));
4038 }
4039
4040 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
4041 {
4042         struct nfs4_file *fp = lock_stp->st_file;
4043         int oflag = nfs4_access_to_omode(access);
4044
4045         if (test_access(access, lock_stp))
4046                 return;
4047         nfs4_file_get_access(fp, oflag);
4048         set_access(access, lock_stp);
4049 }
4050
4051 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)
4052 {
4053         struct nfs4_file *fi = ost->st_file;
4054         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
4055         struct nfs4_client *cl = oo->oo_owner.so_client;
4056         struct nfs4_lockowner *lo;
4057         unsigned int strhashval;
4058
4059         lo = find_lockowner_str(fi->fi_inode, &cl->cl_clientid, &lock->v.new.owner);
4060         if (lo) {
4061                 if (!cstate->minorversion)
4062                         return nfserr_bad_seqid;
4063                 /* XXX: a lockowner always has exactly one stateid: */
4064                 *lst = list_first_entry(&lo->lo_owner.so_stateids,
4065                                 struct nfs4_ol_stateid, st_perstateowner);
4066                 return nfs_ok;
4067         }
4068         strhashval = ownerstr_hashval(cl->cl_clientid.cl_id,
4069                         &lock->v.new.owner);
4070         lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
4071         if (lo == NULL)
4072                 return nfserr_jukebox;
4073         *lst = alloc_init_lock_stateid(lo, fi, ost);
4074         if (*lst == NULL) {
4075                 release_lockowner(lo);
4076                 return nfserr_jukebox;
4077         }
4078         *new = true;
4079         return nfs_ok;
4080 }
4081
4082 /*
4083  *  LOCK operation
4084  */
4085 __be32
4086 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4087            struct nfsd4_lock *lock)
4088 {
4089         struct nfs4_openowner *open_sop = NULL;
4090         struct nfs4_lockowner *lock_sop = NULL;
4091         struct nfs4_ol_stateid *lock_stp;
4092         struct file *filp = NULL;
4093         struct file_lock file_lock;
4094         struct file_lock conflock;
4095         __be32 status = 0;
4096         bool new_state = false;
4097         int lkflg;
4098         int err;
4099
4100         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
4101                 (long long) lock->lk_offset,
4102                 (long long) lock->lk_length);
4103
4104         if (check_lock_length(lock->lk_offset, lock->lk_length))
4105                  return nfserr_inval;
4106
4107         if ((status = fh_verify(rqstp, &cstate->current_fh,
4108                                 S_IFREG, NFSD_MAY_LOCK))) {
4109                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
4110                 return status;
4111         }
4112
4113         nfs4_lock_state();
4114
4115         if (lock->lk_is_new) {
4116                 /*
4117                  * Client indicates that this is a new lockowner.
4118                  * Use open owner and open stateid to create lock owner and
4119                  * lock stateid.
4120                  */
4121                 struct nfs4_ol_stateid *open_stp = NULL;
4122
4123                 if (nfsd4_has_session(cstate))
4124                         /* See rfc 5661 18.10.3: given clientid is ignored: */
4125                         memcpy(&lock->v.new.clientid,
4126                                 &cstate->session->se_client->cl_clientid,
4127                                 sizeof(clientid_t));
4128
4129                 status = nfserr_stale_clientid;
4130                 if (STALE_CLIENTID(&lock->lk_new_clientid))
4131                         goto out;
4132
4133                 /* validate and update open stateid and open seqid */
4134                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
4135                                         lock->lk_new_open_seqid,
4136                                         &lock->lk_new_open_stateid,
4137                                         &open_stp);
4138                 if (status)
4139                         goto out;
4140                 open_sop = openowner(open_stp->st_stateowner);
4141                 status = nfserr_bad_stateid;
4142                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
4143                                                 &lock->v.new.clientid))
4144                         goto out;
4145                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
4146                                                         &lock_stp, &new_state);
4147                 if (status)
4148                         goto out;
4149         } else {
4150                 /* lock (lock owner + lock stateid) already exists */
4151                 status = nfs4_preprocess_seqid_op(cstate,
4152                                        lock->lk_old_lock_seqid,
4153                                        &lock->lk_old_lock_stateid,
4154                                        NFS4_LOCK_STID, &lock_stp);
4155                 if (status)
4156                         goto out;
4157         }
4158         lock_sop = lockowner(lock_stp->st_stateowner);
4159
4160         lkflg = setlkflg(lock->lk_type);
4161         status = nfs4_check_openmode(lock_stp, lkflg);
4162         if (status)
4163                 goto out;
4164
4165         status = nfserr_grace;
4166         if (locks_in_grace() && !lock->lk_reclaim)
4167                 goto out;
4168         status = nfserr_no_grace;
4169         if (!locks_in_grace() && lock->lk_reclaim)
4170                 goto out;
4171
4172         locks_init_lock(&file_lock);
4173         switch (lock->lk_type) {
4174                 case NFS4_READ_LT:
4175                 case NFS4_READW_LT:
4176                         filp = find_readable_file(lock_stp->st_file);
4177                         if (filp)
4178                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
4179                         file_lock.fl_type = F_RDLCK;
4180                         break;
4181                 case NFS4_WRITE_LT:
4182                 case NFS4_WRITEW_LT:
4183                         filp = find_writeable_file(lock_stp->st_file);
4184                         if (filp)
4185                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
4186                         file_lock.fl_type = F_WRLCK;
4187                         break;
4188                 default:
4189                         status = nfserr_inval;
4190                 goto out;
4191         }
4192         if (!filp) {
4193                 status = nfserr_openmode;
4194                 goto out;
4195         }
4196         file_lock.fl_owner = (fl_owner_t)lock_sop;
4197         file_lock.fl_pid = current->tgid;
4198         file_lock.fl_file = filp;
4199         file_lock.fl_flags = FL_POSIX;
4200         file_lock.fl_lmops = &nfsd_posix_mng_ops;
4201
4202         file_lock.fl_start = lock->lk_offset;
4203         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
4204         nfs4_transform_lock_offset(&file_lock);
4205
4206         /*
4207         * Try to lock the file in the VFS.
4208         * Note: locks.c uses the BKL to protect the inode's lock list.
4209         */
4210
4211         err = vfs_lock_file(filp, F_SETLK, &file_lock, &conflock);
4212         switch (-err) {
4213         case 0: /* success! */
4214                 update_stateid(&lock_stp->st_stid.sc_stateid);
4215                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
4216                                 sizeof(stateid_t));
4217                 status = 0;
4218                 break;
4219         case (EAGAIN):          /* conflock holds conflicting lock */
4220                 status = nfserr_denied;
4221                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
4222                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
4223                 break;
4224         case (EDEADLK):
4225                 status = nfserr_deadlock;
4226                 break;
4227         default:
4228                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4229                 status = nfserrno(err);
4230                 break;
4231         }
4232 out:
4233         if (status && new_state)
4234                 release_lockowner(lock_sop);
4235         if (!cstate->replay_owner)
4236                 nfs4_unlock_state();
4237         return status;
4238 }
4239
4240 /*
4241  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4242  * so we do a temporary open here just to get an open file to pass to
4243  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
4244  * inode operation.)
4245  */
4246 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4247 {
4248         struct file *file;
4249         int err;
4250
4251         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4252         if (err)
4253                 return err;
4254         err = vfs_test_lock(file, lock);
4255         nfsd_close(file);
4256         return err;
4257 }
4258
4259 /*
4260  * LOCKT operation
4261  */
4262 __be32
4263 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4264             struct nfsd4_lockt *lockt)
4265 {
4266         struct inode *inode;
4267         struct file_lock file_lock;
4268         struct nfs4_lockowner *lo;
4269         int error;
4270         __be32 status;
4271
4272         if (locks_in_grace())
4273                 return nfserr_grace;
4274
4275         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4276                  return nfserr_inval;
4277
4278         nfs4_lock_state();
4279
4280         status = nfserr_stale_clientid;
4281         if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
4282                 goto out;
4283
4284         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4285                 goto out;
4286
4287         inode = cstate->current_fh.fh_dentry->d_inode;
4288         locks_init_lock(&file_lock);
4289         switch (lockt->lt_type) {
4290                 case NFS4_READ_LT:
4291                 case NFS4_READW_LT:
4292                         file_lock.fl_type = F_RDLCK;
4293                 break;
4294                 case NFS4_WRITE_LT:
4295                 case NFS4_WRITEW_LT:
4296                         file_lock.fl_type = F_WRLCK;
4297                 break;
4298                 default:
4299                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4300                         status = nfserr_inval;
4301                 goto out;
4302         }
4303
4304         lo = find_lockowner_str(inode, &lockt->lt_clientid, &lockt->lt_owner);
4305         if (lo)
4306                 file_lock.fl_owner = (fl_owner_t)lo;
4307         file_lock.fl_pid = current->tgid;
4308         file_lock.fl_flags = FL_POSIX;
4309
4310         file_lock.fl_start = lockt->lt_offset;
4311         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4312
4313         nfs4_transform_lock_offset(&file_lock);
4314
4315         status = nfs_ok;
4316         error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
4317         if (error) {
4318                 status = nfserrno(error);
4319                 goto out;
4320         }
4321         if (file_lock.fl_type != F_UNLCK) {
4322                 status = nfserr_denied;
4323                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
4324         }
4325 out:
4326         nfs4_unlock_state();
4327         return status;
4328 }
4329
4330 __be32
4331 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4332             struct nfsd4_locku *locku)
4333 {
4334         struct nfs4_ol_stateid *stp;
4335         struct file *filp = NULL;
4336         struct file_lock file_lock;
4337         __be32 status;
4338         int err;
4339
4340         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4341                 (long long) locku->lu_offset,
4342                 (long long) locku->lu_length);
4343
4344         if (check_lock_length(locku->lu_offset, locku->lu_length))
4345                  return nfserr_inval;
4346
4347         nfs4_lock_state();
4348
4349         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
4350                                         &locku->lu_stateid, NFS4_LOCK_STID, &stp);
4351         if (status)
4352                 goto out;
4353         filp = find_any_file(stp->st_file);
4354         if (!filp) {
4355                 status = nfserr_lock_range;
4356                 goto out;
4357         }
4358         BUG_ON(!filp);
4359         locks_init_lock(&file_lock);
4360         file_lock.fl_type = F_UNLCK;
4361         file_lock.fl_owner = (fl_owner_t)lockowner(stp->st_stateowner);
4362         file_lock.fl_pid = current->tgid;
4363         file_lock.fl_file = filp;
4364         file_lock.fl_flags = FL_POSIX;
4365         file_lock.fl_lmops = &nfsd_posix_mng_ops;
4366         file_lock.fl_start = locku->lu_offset;
4367
4368         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
4369         nfs4_transform_lock_offset(&file_lock);
4370
4371         /*
4372         *  Try to unlock the file in the VFS.
4373         */
4374         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
4375         if (err) {
4376                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4377                 goto out_nfserr;
4378         }
4379         /*
4380         * OK, unlock succeeded; the only thing left to do is update the stateid.
4381         */
4382         update_stateid(&stp->st_stid.sc_stateid);
4383         memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4384
4385 out:
4386         if (!cstate->replay_owner)
4387                 nfs4_unlock_state();
4388         return status;
4389
4390 out_nfserr:
4391         status = nfserrno(err);
4392         goto out;
4393 }
4394
4395 /*
4396  * returns
4397  *      1: locks held by lockowner
4398  *      0: no locks held by lockowner
4399  */
4400 static int
4401 check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner)
4402 {
4403         struct file_lock **flpp;
4404         struct inode *inode = filp->fi_inode;
4405         int status = 0;
4406
4407         lock_flocks();
4408         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4409                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4410                         status = 1;
4411                         goto out;
4412                 }
4413         }
4414 out:
4415         unlock_flocks();
4416         return status;
4417 }
4418
4419 __be32
4420 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4421                         struct nfsd4_compound_state *cstate,
4422                         struct nfsd4_release_lockowner *rlockowner)
4423 {
4424         clientid_t *clid = &rlockowner->rl_clientid;
4425         struct nfs4_stateowner *sop;
4426         struct nfs4_lockowner *lo;
4427         struct nfs4_ol_stateid *stp;
4428         struct xdr_netobj *owner = &rlockowner->rl_owner;
4429         struct list_head matches;
4430         unsigned int hashval = ownerstr_hashval(clid->cl_id, owner);
4431         __be32 status;
4432
4433         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4434                 clid->cl_boot, clid->cl_id);
4435
4436         /* XXX check for lease expiration */
4437
4438         status = nfserr_stale_clientid;
4439         if (STALE_CLIENTID(clid))
4440                 return status;
4441
4442         nfs4_lock_state();
4443
4444         status = nfserr_locks_held;
4445         INIT_LIST_HEAD(&matches);
4446
4447         list_for_each_entry(sop, &ownerstr_hashtbl[hashval], so_strhash) {
4448                 if (sop->so_is_open_owner)
4449                         continue;
4450                 if (!same_owner_str(sop, owner, clid))
4451                         continue;
4452                 list_for_each_entry(stp, &sop->so_stateids,
4453                                 st_perstateowner) {
4454                         lo = lockowner(sop);
4455                         if (check_for_locks(stp->st_file, lo))
4456                                 goto out;
4457                         list_add(&lo->lo_list, &matches);
4458                 }
4459         }
4460         /* Clients probably won't expect us to return with some (but not all)
4461          * of the lockowner state released; so don't release any until all
4462          * have been checked. */
4463         status = nfs_ok;
4464         while (!list_empty(&matches)) {
4465                 lo = list_entry(matches.next, struct nfs4_lockowner,
4466                                                                 lo_list);
4467                 /* unhash_stateowner deletes so_perclient only
4468                  * for openowners. */
4469                 list_del(&lo->lo_list);
4470                 release_lockowner(lo);
4471         }
4472 out:
4473         nfs4_unlock_state();
4474         return status;
4475 }
4476
4477 static inline struct nfs4_client_reclaim *
4478 alloc_reclaim(void)
4479 {
4480         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4481 }
4482
4483 int
4484 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
4485 {
4486         unsigned int strhashval = clientstr_hashval(name);
4487         struct nfs4_client *clp;
4488
4489         clp = find_confirmed_client_by_str(name, strhashval);
4490         if (!clp)
4491                 return 0;
4492         return test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
4493 }
4494
4495 /*
4496  * failure => all reset bets are off, nfserr_no_grace...
4497  */
4498 int
4499 nfs4_client_to_reclaim(const char *name)
4500 {
4501         unsigned int strhashval;
4502         struct nfs4_client_reclaim *crp = NULL;
4503
4504         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4505         crp = alloc_reclaim();
4506         if (!crp)
4507                 return 0;
4508         strhashval = clientstr_hashval(name);
4509         INIT_LIST_HEAD(&crp->cr_strhash);
4510         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4511         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4512         reclaim_str_hashtbl_size++;
4513         return 1;
4514 }
4515
4516 void
4517 nfs4_release_reclaim(void)
4518 {
4519         struct nfs4_client_reclaim *crp = NULL;
4520         int i;
4521
4522         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4523                 while (!list_empty(&reclaim_str_hashtbl[i])) {
4524                         crp = list_entry(reclaim_str_hashtbl[i].next,
4525                                         struct nfs4_client_reclaim, cr_strhash);
4526                         list_del(&crp->cr_strhash);
4527                         kfree(crp);
4528                         reclaim_str_hashtbl_size--;
4529                 }
4530         }
4531         BUG_ON(reclaim_str_hashtbl_size);
4532 }
4533
4534 /*
4535  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4536 struct nfs4_client_reclaim *
4537 nfsd4_find_reclaim_client(struct nfs4_client *clp)
4538 {
4539         unsigned int strhashval;
4540         struct nfs4_client_reclaim *crp = NULL;
4541
4542         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4543                             clp->cl_name.len, clp->cl_name.data,
4544                             clp->cl_recdir);
4545
4546         /* find clp->cl_name in reclaim_str_hashtbl */
4547         strhashval = clientstr_hashval(clp->cl_recdir);
4548         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4549                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4550                         return crp;
4551                 }
4552         }
4553         return NULL;
4554 }
4555
4556 /*
4557 * Called from OPEN. Look for clientid in reclaim list.
4558 */
4559 __be32
4560 nfs4_check_open_reclaim(clientid_t *clid)
4561 {
4562         struct nfs4_client *clp;
4563
4564         /* find clientid in conf_id_hashtbl */
4565         clp = find_confirmed_client(clid);
4566         if (clp == NULL)
4567                 return nfserr_reclaim_bad;
4568
4569         return nfsd4_client_record_check(clp) ? nfserr_reclaim_bad : nfs_ok;
4570 }
4571
4572 #ifdef CONFIG_NFSD_FAULT_INJECTION
4573
4574 void nfsd_forget_clients(u64 num)
4575 {
4576         struct nfs4_client *clp, *next;
4577         int count = 0;
4578
4579         nfs4_lock_state();
4580         list_for_each_entry_safe(clp, next, &client_lru, cl_lru) {
4581                 nfsd4_client_record_remove(clp);
4582                 expire_client(clp);
4583                 if (++count == num)
4584                         break;
4585         }
4586         nfs4_unlock_state();
4587
4588         printk(KERN_INFO "NFSD: Forgot %d clients", count);
4589 }
4590
4591 static void release_lockowner_sop(struct nfs4_stateowner *sop)
4592 {
4593         release_lockowner(lockowner(sop));
4594 }
4595
4596 static void release_openowner_sop(struct nfs4_stateowner *sop)
4597 {
4598         release_openowner(openowner(sop));
4599 }
4600
4601 static int nfsd_release_n_owners(u64 num, bool is_open_owner,
4602                                 void (*release_sop)(struct nfs4_stateowner *))
4603 {
4604         int i, count = 0;
4605         struct nfs4_stateowner *sop, *next;
4606
4607         for (i = 0; i < OWNER_HASH_SIZE; i++) {
4608                 list_for_each_entry_safe(sop, next, &ownerstr_hashtbl[i], so_strhash) {
4609                         if (sop->so_is_open_owner != is_open_owner)
4610                                 continue;
4611                         release_sop(sop);
4612                         if (++count == num)
4613                                 return count;
4614                 }
4615         }
4616         return count;
4617 }
4618
4619 void nfsd_forget_locks(u64 num)
4620 {
4621         int count;
4622
4623         nfs4_lock_state();
4624         count = nfsd_release_n_owners(num, false, release_lockowner_sop);
4625         nfs4_unlock_state();
4626
4627         printk(KERN_INFO "NFSD: Forgot %d locks", count);
4628 }
4629
4630 void nfsd_forget_openowners(u64 num)
4631 {
4632         int count;
4633
4634         nfs4_lock_state();
4635         count = nfsd_release_n_owners(num, true, release_openowner_sop);
4636         nfs4_unlock_state();
4637
4638         printk(KERN_INFO "NFSD: Forgot %d open owners", count);
4639 }
4640
4641 int nfsd_process_n_delegations(u64 num, void (*deleg_func)(struct nfs4_delegation *))
4642 {
4643         int i, count = 0;
4644         struct nfs4_file *fp, *fnext;
4645         struct nfs4_delegation *dp, *dnext;
4646
4647         for (i = 0; i < FILE_HASH_SIZE; i++) {
4648                 list_for_each_entry_safe(fp, fnext, &file_hashtbl[i], fi_hash) {
4649                         list_for_each_entry_safe(dp, dnext, &fp->fi_delegations, dl_perfile) {
4650                                 deleg_func(dp);
4651                                 if (++count == num)
4652                                         return count;
4653                         }
4654                 }
4655         }
4656
4657         return count;
4658 }
4659
4660 void nfsd_forget_delegations(u64 num)
4661 {
4662         unsigned int count;
4663
4664         nfs4_lock_state();
4665         count = nfsd_process_n_delegations(num, unhash_delegation);
4666         nfs4_unlock_state();
4667
4668         printk(KERN_INFO "NFSD: Forgot %d delegations", count);
4669 }
4670
4671 void nfsd_recall_delegations(u64 num)
4672 {
4673         unsigned int count;
4674
4675         nfs4_lock_state();
4676         spin_lock(&recall_lock);
4677         count = nfsd_process_n_delegations(num, nfsd_break_one_deleg);
4678         spin_unlock(&recall_lock);
4679         nfs4_unlock_state();
4680
4681         printk(KERN_INFO "NFSD: Recalled %d delegations", count);
4682 }
4683
4684 #endif /* CONFIG_NFSD_FAULT_INJECTION */
4685
4686 /* initialization to perform at module load time: */
4687
4688 void
4689 nfs4_state_init(void)
4690 {
4691         int i;
4692
4693         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4694                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4695                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4696                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4697                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4698                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4699         }
4700         for (i = 0; i < SESSION_HASH_SIZE; i++)
4701                 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4702         for (i = 0; i < FILE_HASH_SIZE; i++) {
4703                 INIT_LIST_HEAD(&file_hashtbl[i]);
4704         }
4705         for (i = 0; i < OWNER_HASH_SIZE; i++) {
4706                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4707         }
4708         for (i = 0; i < LOCKOWNER_INO_HASH_SIZE; i++)
4709                 INIT_LIST_HEAD(&lockowner_ino_hashtbl[i]);
4710         INIT_LIST_HEAD(&close_lru);
4711         INIT_LIST_HEAD(&client_lru);
4712         INIT_LIST_HEAD(&del_recall_lru);
4713         reclaim_str_hashtbl_size = 0;
4714 }
4715
4716 /*
4717  * Since the lifetime of a delegation isn't limited to that of an open, a
4718  * client may quite reasonably hang on to a delegation as long as it has
4719  * the inode cached.  This becomes an obvious problem the first time a
4720  * client's inode cache approaches the size of the server's total memory.
4721  *
4722  * For now we avoid this problem by imposing a hard limit on the number
4723  * of delegations, which varies according to the server's memory size.
4724  */
4725 static void
4726 set_max_delegations(void)
4727 {
4728         /*
4729          * Allow at most 4 delegations per megabyte of RAM.  Quick
4730          * estimates suggest that in the worst case (where every delegation
4731          * is for a different inode), a delegation could take about 1.5K,
4732          * giving a worst case usage of about 6% of memory.
4733          */
4734         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4735 }
4736
4737 /* initialization to perform when the nfsd service is started: */
4738
4739 int
4740 nfs4_state_start(void)
4741 {
4742         int ret;
4743
4744         /*
4745          * FIXME: For now, we hang most of the pernet global stuff off of
4746          * init_net until nfsd is fully containerized. Eventually, we'll
4747          * need to pass a net pointer into this function, take a reference
4748          * to that instead and then do most of the rest of this on a per-net
4749          * basis.
4750          */
4751         get_net(&init_net);
4752         nfsd4_client_tracking_init(&init_net);
4753         boot_time = get_seconds();
4754         locks_start_grace(&nfsd4_manager);
4755         grace_ended = false;
4756         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4757                nfsd4_grace);
4758         ret = set_callback_cred();
4759         if (ret) {
4760                 ret = -ENOMEM;
4761                 goto out_recovery;
4762         }
4763         laundry_wq = create_singlethread_workqueue("nfsd4");
4764         if (laundry_wq == NULL) {
4765                 ret = -ENOMEM;
4766                 goto out_recovery;
4767         }
4768         ret = nfsd4_create_callback_queue();
4769         if (ret)
4770                 goto out_free_laundry;
4771         queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4772         set_max_delegations();
4773         return 0;
4774 out_free_laundry:
4775         destroy_workqueue(laundry_wq);
4776 out_recovery:
4777         nfsd4_client_tracking_exit(&init_net);
4778         put_net(&init_net);
4779         return ret;
4780 }
4781
4782 static void
4783 __nfs4_state_shutdown(void)
4784 {
4785         int i;
4786         struct nfs4_client *clp = NULL;
4787         struct nfs4_delegation *dp = NULL;
4788         struct list_head *pos, *next, reaplist;
4789
4790         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4791                 while (!list_empty(&conf_id_hashtbl[i])) {
4792                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4793                         expire_client(clp);
4794                 }
4795                 while (!list_empty(&unconf_str_hashtbl[i])) {
4796                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4797                         expire_client(clp);
4798                 }
4799         }
4800         INIT_LIST_HEAD(&reaplist);
4801         spin_lock(&recall_lock);
4802         list_for_each_safe(pos, next, &del_recall_lru) {
4803                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4804                 list_move(&dp->dl_recall_lru, &reaplist);
4805         }
4806         spin_unlock(&recall_lock);
4807         list_for_each_safe(pos, next, &reaplist) {
4808                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4809                 unhash_delegation(dp);
4810         }
4811
4812         nfsd4_client_tracking_exit(&init_net);
4813         put_net(&init_net);
4814 }
4815
4816 void
4817 nfs4_state_shutdown(void)
4818 {
4819         cancel_delayed_work_sync(&laundromat_work);
4820         destroy_workqueue(laundry_wq);
4821         locks_end_grace(&nfsd4_manager);
4822         nfs4_lock_state();
4823         __nfs4_state_shutdown();
4824         nfs4_unlock_state();
4825         nfsd4_destroy_callback_queue();
4826 }
4827
4828 static void
4829 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
4830 {
4831         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
4832                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
4833 }
4834
4835 static void
4836 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
4837 {
4838         if (cstate->minorversion) {
4839                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
4840                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
4841         }
4842 }
4843
4844 void
4845 clear_current_stateid(struct nfsd4_compound_state *cstate)
4846 {
4847         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
4848 }
4849
4850 /*
4851  * functions to set current state id
4852  */
4853 void
4854 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
4855 {
4856         put_stateid(cstate, &odp->od_stateid);
4857 }
4858
4859 void
4860 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
4861 {
4862         put_stateid(cstate, &open->op_stateid);
4863 }
4864
4865 void
4866 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
4867 {
4868         put_stateid(cstate, &close->cl_stateid);
4869 }
4870
4871 void
4872 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
4873 {
4874         put_stateid(cstate, &lock->lk_resp_stateid);
4875 }
4876
4877 /*
4878  * functions to consume current state id
4879  */
4880
4881 void
4882 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
4883 {
4884         get_stateid(cstate, &odp->od_stateid);
4885 }
4886
4887 void
4888 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
4889 {
4890         get_stateid(cstate, &drp->dr_stateid);
4891 }
4892
4893 void
4894 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
4895 {
4896         get_stateid(cstate, &fsp->fr_stateid);
4897 }
4898
4899 void
4900 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
4901 {
4902         get_stateid(cstate, &setattr->sa_stateid);
4903 }
4904
4905 void
4906 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
4907 {
4908         get_stateid(cstate, &close->cl_stateid);
4909 }
4910
4911 void
4912 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
4913 {
4914         get_stateid(cstate, &locku->lu_stateid);
4915 }
4916
4917 void
4918 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
4919 {
4920         get_stateid(cstate, &read->rd_stateid);
4921 }
4922
4923 void
4924 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
4925 {
4926         get_stateid(cstate, &write->wr_stateid);
4927 }