fs/nfsd/nfs4state.c

   1 /*
   2 *  Copyright (c) 2001 The Regents of the University of Michigan.
   3 *  All rights reserved.
   4 *
   5 *  Kendrick Smith <kmsmith@umich.edu>
   6 *  Andy Adamson <kandros@umich.edu>
   7 *
   8 *  Redistribution and use in source and binary forms, with or without
   9 *  modification, are permitted provided that the following conditions
  10 *  are met:
  11 *
  12 *  1. Redistributions of source code must retain the above copyright
  13 *     notice, this list of conditions and the following disclaimer.
  14 *  2. Redistributions in binary form must reproduce the above copyright
  15 *     notice, this list of conditions and the following disclaimer in the
  16 *     documentation and/or other materials provided with the distribution.
  17 *  3. Neither the name of the University nor the names of its
  18 *     contributors may be used to endorse or promote products derived
  19 *     from this software without specific prior written permission.
  20 *
  21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32 *
  33 */
  34
  35 #include <linux/file.h>
  36 #include <linux/fs.h>
  37 #include <linux/slab.h>
  38 #include <linux/namei.h>
  39 #include <linux/swap.h>
  40 #include <linux/pagemap.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/sunrpc/svcauth_gss.h>
  43 #include <linux/sunrpc/clnt.h>
  44 #include "xdr4.h"
  45 #include "vfs.h"
  46 #include "current_stateid.h"
  47 #include "fault_inject.h"
  48
  49 #include "netns.h"
  50
  51 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  52
  53 /* Globals */
  54 time_t nfsd4_lease = 90;     /* default lease time */
  55 time_t nfsd4_grace = 90;
  56 static time_t boot_time;
  57
  58 #define all_ones {{~0,~0},~0}
  59 static const stateid_t one_stateid = {
  60         .si_generation = ~0,
  61         .si_opaque = all_ones,
  62 };
  63 static const stateid_t zero_stateid = {
  64         /* all fields zero */
  65 };
  66 static const stateid_t currentstateid = {
  67         .si_generation = 1,
  68 };
  69
  70 static u64 current_sessionid = 1;
  71
  72 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  73 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  74 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  75
  76 /* forward declarations */
  77 static int check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner);
  78
  79 /* Locking: */
  80
  81 /* Currently used for almost all code touching nfsv4 state: */
  82 static DEFINE_MUTEX(client_mutex);
  83
  84 /*
  85  * Currently used for the del_recall_lru and file hash table.  In an
  86  * effort to decrease the scope of the client_mutex, this spinlock may
  87  * eventually cover more:
  88  */
  89 static DEFINE_SPINLOCK(recall_lock);
  90
  91 static struct kmem_cache *openowner_slab = NULL;
  92 static struct kmem_cache *lockowner_slab = NULL;
  93 static struct kmem_cache *file_slab = NULL;
  94 static struct kmem_cache *stateid_slab = NULL;
  95 static struct kmem_cache *deleg_slab = NULL;
  96
  97 void
  98 nfs4_lock_state(void)
  99 {
 100         mutex_lock(&client_mutex);
 101 }
 102
 103 static void free_session(struct kref *);
 104
 105 /* Must be called under the client_lock */
 106 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 107 {
 108         kref_put(&ses->se_ref, free_session);
 109 }
 110
 111 static void nfsd4_get_session(struct nfsd4_session *ses)
 112 {
 113         kref_get(&ses->se_ref);
 114 }
 115
 116 void
 117 nfs4_unlock_state(void)
 118 {
 119         mutex_unlock(&client_mutex);
 120 }
 121
 122 static inline u32
 123 opaque_hashval(const void *ptr, int nbytes)
 124 {
 125         unsigned char *cptr = (unsigned char *) ptr;
 126
 127         u32 x = 0;
 128         while (nbytes--) {
 129                 x *= 37;
 130                 x += *cptr++;
 131         }
 132         return x;
 133 }
 134
 135 static struct list_head del_recall_lru;
 136
 137 static void nfsd4_free_file(struct nfs4_file *f)
 138 {
 139         kmem_cache_free(file_slab, f);
 140 }
 141
 142 static inline void
 143 put_nfs4_file(struct nfs4_file *fi)
 144 {
 145         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
 146                 list_del(&fi->fi_hash);
 147                 spin_unlock(&recall_lock);
 148                 iput(fi->fi_inode);
 149                 nfsd4_free_file(fi);
 150         }
 151 }
 152
 153 static inline void
 154 get_nfs4_file(struct nfs4_file *fi)
 155 {
 156         atomic_inc(&fi->fi_ref);
 157 }
 158
 159 static int num_delegations;
 160 unsigned int max_delegations;
 161
 162 /*
 163  * Open owner state (share locks)
 164  */
 165
 166 /* hash tables for lock and open owners */
 167 #define OWNER_HASH_BITS              8
 168 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 169 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 170
 171 static unsigned int ownerstr_hashval(u32 clientid, struct xdr_netobj *ownername)
 172 {
 173         unsigned int ret;
 174
 175         ret = opaque_hashval(ownername->data, ownername->len);
 176         ret += clientid;
 177         return ret & OWNER_HASH_MASK;
 178 }
 179
 180 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
 181
 182 /* hash table for nfs4_file */
 183 #define FILE_HASH_BITS                   8
 184 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 185
 186 static unsigned int file_hashval(struct inode *ino)
 187 {
 188         /* XXX: why are we hashing on inode pointer, anyway? */
 189         return hash_ptr(ino, FILE_HASH_BITS);
 190 }
 191
 192 static struct list_head file_hashtbl[FILE_HASH_SIZE];
 193
 194 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
 195 {
 196         BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
 197         atomic_inc(&fp->fi_access[oflag]);
 198 }
 199
 200 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
 201 {
 202         if (oflag == O_RDWR) {
 203                 __nfs4_file_get_access(fp, O_RDONLY);
 204                 __nfs4_file_get_access(fp, O_WRONLY);
 205         } else
 206                 __nfs4_file_get_access(fp, oflag);
 207 }
 208
 209 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
 210 {
 211         if (fp->fi_fds[oflag]) {
 212                 fput(fp->fi_fds[oflag]);
 213                 fp->fi_fds[oflag] = NULL;
 214         }
 215 }
 216
 217 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 218 {
 219         if (atomic_dec_and_test(&fp->fi_access[oflag])) {
 220                 nfs4_file_put_fd(fp, oflag);
 221                 /*
 222                  * It's also safe to get rid of the RDWR open *if*
 223                  * we no longer have need of the other kind of access
 224                  * or if we already have the other kind of open:
 225                  */
 226                 if (fp->fi_fds[1-oflag]
 227                         || atomic_read(&fp->fi_access[1 - oflag]) == 0)
 228                         nfs4_file_put_fd(fp, O_RDWR);
 229         }
 230 }
 231
 232 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 233 {
 234         if (oflag == O_RDWR) {
 235                 __nfs4_file_put_access(fp, O_RDONLY);
 236                 __nfs4_file_put_access(fp, O_WRONLY);
 237         } else
 238                 __nfs4_file_put_access(fp, oflag);
 239 }
 240
 241 static inline int get_new_stid(struct nfs4_stid *stid)
 242 {
 243         static int min_stateid = 0;
 244         struct idr *stateids = &stid->sc_client->cl_stateids;
 245         int new_stid;
 246         int error;
 247
 248         error = idr_get_new_above(stateids, stid, min_stateid, &new_stid);
 249         /*
 250          * Note: the necessary preallocation was done in
 251          * nfs4_alloc_stateid().  The idr code caps the number of
 252          * preallocations that can exist at a time, but the state lock
 253          * prevents anyone from using ours before we get here:
 254          */
 255         BUG_ON(error);
 256         /*
 257          * It shouldn't be a problem to reuse an opaque stateid value.
 258          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 259          * example, a stray write retransmission could be accepted by
 260          * the server when it should have been rejected.  Therefore,
 261          * adopt a trick from the sctp code to attempt to maximize the
 262          * amount of time until an id is reused, by ensuring they always
 263          * "increase" (mod INT_MAX):
 264          */
 265
 266         min_stateid = new_stid+1;
 267         if (min_stateid == INT_MAX)
 268                 min_stateid = 0;
 269         return new_stid;
 270 }
 271
 272 static void init_stid(struct nfs4_stid *stid, struct nfs4_client *cl, unsigned char type)
 273 {
 274         stateid_t *s = &stid->sc_stateid;
 275         int new_id;
 276
 277         stid->sc_type = type;
 278         stid->sc_client = cl;
 279         s->si_opaque.so_clid = cl->cl_clientid;
 280         new_id = get_new_stid(stid);
 281         s->si_opaque.so_id = (u32)new_id;
 282         /* Will be incremented before return to client: */
 283         s->si_generation = 0;
 284 }
 285
 286 static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab)
 287 {
 288         struct idr *stateids = &cl->cl_stateids;
 289
 290         if (!idr_pre_get(stateids, GFP_KERNEL))
 291                 return NULL;
 292         /*
 293          * Note: if we fail here (or any time between now and the time
 294          * we actually get the new idr), we won't need to undo the idr
 295          * preallocation, since the idr code caps the number of
 296          * preallocated entries.
 297          */
 298         return kmem_cache_alloc(slab, GFP_KERNEL);
 299 }
 300
 301 static struct nfs4_ol_stateid * nfs4_alloc_stateid(struct nfs4_client *clp)
 302 {
 303         return openlockstateid(nfs4_alloc_stid(clp, stateid_slab));
 304 }
 305
 306 static struct nfs4_delegation *
 307 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_ol_stateid *stp, struct svc_fh *current_fh, u32 type)
 308 {
 309         struct nfs4_delegation *dp;
 310         struct nfs4_file *fp = stp->st_file;
 311
 312         dprintk("NFSD alloc_init_deleg\n");
 313         /*
 314          * Major work on the lease subsystem (for example, to support
 315          * calbacks on stat) will be required before we can support
 316          * write delegations properly.
 317          */
 318         if (type != NFS4_OPEN_DELEGATE_READ)
 319                 return NULL;
 320         if (fp->fi_had_conflict)
 321                 return NULL;
 322         if (num_delegations > max_delegations)
 323                 return NULL;
 324         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
 325         if (dp == NULL)
 326                 return dp;
 327         init_stid(&dp->dl_stid, clp, NFS4_DELEG_STID);
 328         /*
 329          * delegation seqid's are never incremented.  The 4.1 special
 330          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 331          * 0 anyway just for consistency and use 1:
 332          */
 333         dp->dl_stid.sc_stateid.si_generation = 1;
 334         num_delegations++;
 335         INIT_LIST_HEAD(&dp->dl_perfile);
 336         INIT_LIST_HEAD(&dp->dl_perclnt);
 337         INIT_LIST_HEAD(&dp->dl_recall_lru);
 338         get_nfs4_file(fp);
 339         dp->dl_file = fp;
 340         dp->dl_type = type;
 341         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
 342         dp->dl_time = 0;
 343         atomic_set(&dp->dl_count, 1);
 344         INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
 345         return dp;
 346 }
 347
 348 void
 349 nfs4_put_delegation(struct nfs4_delegation *dp)
 350 {
 351         if (atomic_dec_and_test(&dp->dl_count)) {
 352                 dprintk("NFSD: freeing dp %p\n",dp);
 353                 put_nfs4_file(dp->dl_file);
 354                 kmem_cache_free(deleg_slab, dp);
 355                 num_delegations--;
 356         }
 357 }
 358
 359 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 360 {
 361         if (atomic_dec_and_test(&fp->fi_delegees)) {
 362                 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
 363                 fp->fi_lease = NULL;
 364                 fput(fp->fi_deleg_file);
 365                 fp->fi_deleg_file = NULL;
 366         }
 367 }
 368
 369 static void unhash_stid(struct nfs4_stid *s)
 370 {
 371         struct idr *stateids = &s->sc_client->cl_stateids;
 372
 373         idr_remove(stateids, s->sc_stateid.si_opaque.so_id);
 374 }
 375
 376 /* Called under the state lock. */
 377 static void
 378 unhash_delegation(struct nfs4_delegation *dp)
 379 {
 380         unhash_stid(&dp->dl_stid);
 381         list_del_init(&dp->dl_perclnt);
 382         spin_lock(&recall_lock);
 383         list_del_init(&dp->dl_perfile);
 384         list_del_init(&dp->dl_recall_lru);
 385         spin_unlock(&recall_lock);
 386         nfs4_put_deleg_lease(dp->dl_file);
 387         nfs4_put_delegation(dp);
 388 }
 389
 390 /*
 391  * SETCLIENTID state
 392  */
 393
 394 /* client_lock protects the client lru list and session hash table */
 395 static DEFINE_SPINLOCK(client_lock);
 396
 397 /* Hash tables for nfs4_clientid state */
 398 #define CLIENT_HASH_BITS                 4
 399 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
 400 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
 401
 402 static unsigned int clientid_hashval(u32 id)
 403 {
 404         return id & CLIENT_HASH_MASK;
 405 }
 406
 407 static unsigned int clientstr_hashval(const char *name)
 408 {
 409         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
 410 }
 411
 412 /*
 413  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
 414  * used in reboot/reset lease grace period processing
 415  *
 416  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
 417  * setclientid_confirmed info.
 418  *
 419  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
 420  * setclientid info.
 421  *
 422  * client_lru holds client queue ordered by nfs4_client.cl_time
 423  * for lease renewal.
 424  *
 425  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
 426  * for last close replay.
 427  */
 428 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
 429 static int reclaim_str_hashtbl_size = 0;
 430 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
 431 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
 432 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
 433 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
 434 static struct list_head client_lru;
 435 static struct list_head close_lru;
 436
 437 /*
 438  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 439  * st_{access,deny}_bmap field of the stateid, in order to track not
 440  * only what share bits are currently in force, but also what
 441  * combinations of share bits previous opens have used.  This allows us
 442  * to enforce the recommendation of rfc 3530 14.2.19 that the server
 443  * return an error if the client attempt to downgrade to a combination
 444  * of share bits not explicable by closing some of its previous opens.
 445  *
 446  * XXX: This enforcement is actually incomplete, since we don't keep
 447  * track of access/deny bit combinations; so, e.g., we allow:
 448  *
 449  *      OPEN allow read, deny write
 450  *      OPEN allow both, deny none
 451  *      DOWNGRADE allow read, deny none
 452  *
 453  * which we should reject.
 454  */
 455 static unsigned int
 456 bmap_to_share_mode(unsigned long bmap) {
 457         int i;
 458         unsigned int access = 0;
 459
 460         for (i = 1; i < 4; i++) {
 461                 if (test_bit(i, &bmap))
 462                         access |= i;
 463         }
 464         return access;
 465 }
 466
 467 static bool
 468 test_share(struct nfs4_ol_stateid *stp, struct nfsd4_open *open) {
 469         unsigned int access, deny;
 470
 471         access = bmap_to_share_mode(stp->st_access_bmap);
 472         deny = bmap_to_share_mode(stp->st_deny_bmap);
 473         if ((access & open->op_share_deny) || (deny & open->op_share_access))
 474                 return false;
 475         return true;
 476 }
 477
 478 /* set share access for a given stateid */
 479 static inline void
 480 set_access(u32 access, struct nfs4_ol_stateid *stp)
 481 {
 482         __set_bit(access, &stp->st_access_bmap);
 483 }
 484
 485 /* clear share access for a given stateid */
 486 static inline void
 487 clear_access(u32 access, struct nfs4_ol_stateid *stp)
 488 {
 489         __clear_bit(access, &stp->st_access_bmap);
 490 }
 491
 492 /* test whether a given stateid has access */
 493 static inline bool
 494 test_access(u32 access, struct nfs4_ol_stateid *stp)
 495 {
 496         return test_bit(access, &stp->st_access_bmap);
 497 }
 498
 499 /* set share deny for a given stateid */
 500 static inline void
 501 set_deny(u32 access, struct nfs4_ol_stateid *stp)
 502 {
 503         __set_bit(access, &stp->st_deny_bmap);
 504 }
 505
 506 /* clear share deny for a given stateid */
 507 static inline void
 508 clear_deny(u32 access, struct nfs4_ol_stateid *stp)
 509 {
 510         __clear_bit(access, &stp->st_deny_bmap);
 511 }
 512
 513 /* test whether a given stateid is denying specific access */
 514 static inline bool
 515 test_deny(u32 access, struct nfs4_ol_stateid *stp)
 516 {
 517         return test_bit(access, &stp->st_deny_bmap);
 518 }
 519
 520 static int nfs4_access_to_omode(u32 access)
 521 {
 522         switch (access & NFS4_SHARE_ACCESS_BOTH) {
 523         case NFS4_SHARE_ACCESS_READ:
 524                 return O_RDONLY;
 525         case NFS4_SHARE_ACCESS_WRITE:
 526                 return O_WRONLY;
 527         case NFS4_SHARE_ACCESS_BOTH:
 528                 return O_RDWR;
 529         }
 530         BUG();
 531 }
 532
 533 /* release all access and file references for a given stateid */
 534 static void
 535 release_all_access(struct nfs4_ol_stateid *stp)
 536 {
 537         int i;
 538
 539         for (i = 1; i < 4; i++) {
 540                 if (test_access(i, stp))
 541                         nfs4_file_put_access(stp->st_file,
 542                                              nfs4_access_to_omode(i));
 543                 clear_access(i, stp);
 544         }
 545 }
 546
 547 static void unhash_generic_stateid(struct nfs4_ol_stateid *stp)
 548 {
 549         list_del(&stp->st_perfile);
 550         list_del(&stp->st_perstateowner);
 551 }
 552
 553 static void close_generic_stateid(struct nfs4_ol_stateid *stp)
 554 {
 555         release_all_access(stp);
 556         put_nfs4_file(stp->st_file);
 557         stp->st_file = NULL;
 558 }
 559
 560 static void free_generic_stateid(struct nfs4_ol_stateid *stp)
 561 {
 562         kmem_cache_free(stateid_slab, stp);
 563 }
 564
 565 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
 566 {
 567         struct file *file;
 568
 569         unhash_generic_stateid(stp);
 570         unhash_stid(&stp->st_stid);
 571         file = find_any_file(stp->st_file);
 572         if (file)
 573                 locks_remove_posix(file, (fl_owner_t)lockowner(stp->st_stateowner));
 574         close_generic_stateid(stp);
 575         free_generic_stateid(stp);
 576 }
 577
 578 static void unhash_lockowner(struct nfs4_lockowner *lo)
 579 {
 580         struct nfs4_ol_stateid *stp;
 581
 582         list_del(&lo->lo_owner.so_strhash);
 583         list_del(&lo->lo_perstateid);
 584         list_del(&lo->lo_owner_ino_hash);
 585         while (!list_empty(&lo->lo_owner.so_stateids)) {
 586                 stp = list_first_entry(&lo->lo_owner.so_stateids,
 587                                 struct nfs4_ol_stateid, st_perstateowner);
 588                 release_lock_stateid(stp);
 589         }
 590 }
 591
 592 static void release_lockowner(struct nfs4_lockowner *lo)
 593 {
 594         unhash_lockowner(lo);
 595         nfs4_free_lockowner(lo);
 596 }
 597
 598 static void
 599 release_stateid_lockowners(struct nfs4_ol_stateid *open_stp)
 600 {
 601         struct nfs4_lockowner *lo;
 602
 603         while (!list_empty(&open_stp->st_lockowners)) {
 604                 lo = list_entry(open_stp->st_lockowners.next,
 605                                 struct nfs4_lockowner, lo_perstateid);
 606                 release_lockowner(lo);
 607         }
 608 }
 609
 610 static void unhash_open_stateid(struct nfs4_ol_stateid *stp)
 611 {
 612         unhash_generic_stateid(stp);
 613         release_stateid_lockowners(stp);
 614         close_generic_stateid(stp);
 615 }
 616
 617 static void release_open_stateid(struct nfs4_ol_stateid *stp)
 618 {
 619         unhash_open_stateid(stp);
 620         unhash_stid(&stp->st_stid);
 621         free_generic_stateid(stp);
 622 }
 623
 624 static void unhash_openowner(struct nfs4_openowner *oo)
 625 {
 626         struct nfs4_ol_stateid *stp;
 627
 628         list_del(&oo->oo_owner.so_strhash);
 629         list_del(&oo->oo_perclient);
 630         while (!list_empty(&oo->oo_owner.so_stateids)) {
 631                 stp = list_first_entry(&oo->oo_owner.so_stateids,
 632                                 struct nfs4_ol_stateid, st_perstateowner);
 633                 release_open_stateid(stp);
 634         }
 635 }
 636
 637 static void release_last_closed_stateid(struct nfs4_openowner *oo)
 638 {
 639         struct nfs4_ol_stateid *s = oo->oo_last_closed_stid;
 640
 641         if (s) {
 642                 unhash_stid(&s->st_stid);
 643                 free_generic_stateid(s);
 644                 oo->oo_last_closed_stid = NULL;
 645         }
 646 }
 647
 648 static void release_openowner(struct nfs4_openowner *oo)
 649 {
 650         unhash_openowner(oo);
 651         list_del(&oo->oo_close_lru);
 652         release_last_closed_stateid(oo);
 653         nfs4_free_openowner(oo);
 654 }
 655
 656 #define SESSION_HASH_SIZE       512
 657 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
 658
 659 static inline int
 660 hash_sessionid(struct nfs4_sessionid *sessionid)
 661 {
 662         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
 663
 664         return sid->sequence % SESSION_HASH_SIZE;
 665 }
 666
 667 #ifdef NFSD_DEBUG
 668 static inline void
 669 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 670 {
 671         u32 *ptr = (u32 *)(&sessionid->data[0]);
 672         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
 673 }
 674 #else
 675 static inline void
 676 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 677 {
 678 }
 679 #endif
 680
 681
 682 static void
 683 gen_sessionid(struct nfsd4_session *ses)
 684 {
 685         struct nfs4_client *clp = ses->se_client;
 686         struct nfsd4_sessionid *sid;
 687
 688         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
 689         sid->clientid = clp->cl_clientid;
 690         sid->sequence = current_sessionid++;
 691         sid->reserved = 0;
 692 }
 693
 694 /*
 695  * The protocol defines ca_maxresponssize_cached to include the size of
 696  * the rpc header, but all we need to cache is the data starting after
 697  * the end of the initial SEQUENCE operation--the rest we regenerate
 698  * each time.  Therefore we can advertise a ca_maxresponssize_cached
 699  * value that is the number of bytes in our cache plus a few additional
 700  * bytes.  In order to stay on the safe side, and not promise more than
 701  * we can cache, those additional bytes must be the minimum possible: 24
 702  * bytes of rpc header (xid through accept state, with AUTH_NULL
 703  * verifier), 12 for the compound header (with zero-length tag), and 44
 704  * for the SEQUENCE op response:
 705  */
 706 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
 707
 708 static void
 709 free_session_slots(struct nfsd4_session *ses)
 710 {
 711         int i;
 712
 713         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
 714                 kfree(ses->se_slots[i]);
 715 }
 716
 717 /*
 718  * We don't actually need to cache the rpc and session headers, so we
 719  * can allocate a little less for each slot:
 720  */
 721 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
 722 {
 723         return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
 724 }
 725
 726 static int nfsd4_sanitize_slot_size(u32 size)
 727 {
 728         size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
 729         size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
 730
 731         return size;
 732 }
 733
 734 /*
 735  * XXX: If we run out of reserved DRC memory we could (up to a point)
 736  * re-negotiate active sessions and reduce their slot usage to make
 737  * room for new connections. For now we just fail the create session.
 738  */
 739 static int nfsd4_get_drc_mem(int slotsize, u32 num)
 740 {
 741         int avail;
 742
 743         num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
 744
 745         spin_lock(&nfsd_drc_lock);
 746         avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
 747                         nfsd_drc_max_mem - nfsd_drc_mem_used);
 748         num = min_t(int, num, avail / slotsize);
 749         nfsd_drc_mem_used += num * slotsize;
 750         spin_unlock(&nfsd_drc_lock);
 751
 752         return num;
 753 }
 754
 755 static void nfsd4_put_drc_mem(int slotsize, int num)
 756 {
 757         spin_lock(&nfsd_drc_lock);
 758         nfsd_drc_mem_used -= slotsize * num;
 759         spin_unlock(&nfsd_drc_lock);
 760 }
 761
 762 static struct nfsd4_session *alloc_session(int slotsize, int numslots)
 763 {
 764         struct nfsd4_session *new;
 765         int mem, i;
 766
 767         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
 768                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
 769         mem = numslots * sizeof(struct nfsd4_slot *);
 770
 771         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
 772         if (!new)
 773                 return NULL;
 774         /* allocate each struct nfsd4_slot and data cache in one piece */
 775         for (i = 0; i < numslots; i++) {
 776                 mem = sizeof(struct nfsd4_slot) + slotsize;
 777                 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
 778                 if (!new->se_slots[i])
 779                         goto out_free;
 780         }
 781         return new;
 782 out_free:
 783         while (i--)
 784                 kfree(new->se_slots[i]);
 785         kfree(new);
 786         return NULL;
 787 }
 788
 789 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
 790 {
 791         u32 maxrpc = nfsd_serv->sv_max_mesg;
 792
 793         new->maxreqs = numslots;
 794         new->maxresp_cached = min_t(u32, req->maxresp_cached,
 795                                         slotsize + NFSD_MIN_HDR_SEQ_SZ);
 796         new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
 797         new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
 798         new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
 799 }
 800
 801 static void free_conn(struct nfsd4_conn *c)
 802 {
 803         svc_xprt_put(c->cn_xprt);
 804         kfree(c);
 805 }
 806
 807 static void nfsd4_conn_lost(struct svc_xpt_user *u)
 808 {
 809         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
 810         struct nfs4_client *clp = c->cn_session->se_client;
 811
 812         spin_lock(&clp->cl_lock);
 813         if (!list_empty(&c->cn_persession)) {
 814                 list_del(&c->cn_persession);
 815                 free_conn(c);
 816         }
 817         spin_unlock(&clp->cl_lock);
 818         nfsd4_probe_callback(clp);
 819 }
 820
 821 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
 822 {
 823         struct nfsd4_conn *conn;
 824
 825         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
 826         if (!conn)
 827                 return NULL;
 828         svc_xprt_get(rqstp->rq_xprt);
 829         conn->cn_xprt = rqstp->rq_xprt;
 830         conn->cn_flags = flags;
 831         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
 832         return conn;
 833 }
 834
 835 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 836 {
 837         conn->cn_session = ses;
 838         list_add(&conn->cn_persession, &ses->se_conns);
 839 }
 840
 841 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 842 {
 843         struct nfs4_client *clp = ses->se_client;
 844
 845         spin_lock(&clp->cl_lock);
 846         __nfsd4_hash_conn(conn, ses);
 847         spin_unlock(&clp->cl_lock);
 848 }
 849
 850 static int nfsd4_register_conn(struct nfsd4_conn *conn)
 851 {
 852         conn->cn_xpt_user.callback = nfsd4_conn_lost;
 853         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
 854 }
 855
 856 static __be32 nfsd4_new_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses, u32 dir)
 857 {
 858         struct nfsd4_conn *conn;
 859         int ret;
 860
 861         conn = alloc_conn(rqstp, dir);
 862         if (!conn)
 863                 return nfserr_jukebox;
 864         nfsd4_hash_conn(conn, ses);
 865         ret = nfsd4_register_conn(conn);
 866         if (ret)
 867                 /* oops; xprt is already down: */
 868                 nfsd4_conn_lost(&conn->cn_xpt_user);
 869         if (ses->se_client->cl_cb_state == NFSD4_CB_DOWN &&
 870                 dir & NFS4_CDFC4_BACK) {
 871                 /* callback channel may be back up */
 872                 nfsd4_probe_callback(ses->se_client);
 873         }
 874         return nfs_ok;
 875 }
 876
 877 static __be32 nfsd4_new_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_session *ses)
 878 {
 879         u32 dir = NFS4_CDFC4_FORE;
 880
 881         if (ses->se_flags & SESSION4_BACK_CHAN)
 882                 dir |= NFS4_CDFC4_BACK;
 883
 884         return nfsd4_new_conn(rqstp, ses, dir);
 885 }
 886
 887 /* must be called under client_lock */
 888 static void nfsd4_del_conns(struct nfsd4_session *s)
 889 {
 890         struct nfs4_client *clp = s->se_client;
 891         struct nfsd4_conn *c;
 892
 893         spin_lock(&clp->cl_lock);
 894         while (!list_empty(&s->se_conns)) {
 895                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
 896                 list_del_init(&c->cn_persession);
 897                 spin_unlock(&clp->cl_lock);
 898
 899                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
 900                 free_conn(c);
 901
 902                 spin_lock(&clp->cl_lock);
 903         }
 904         spin_unlock(&clp->cl_lock);
 905 }
 906
 907 static void free_session(struct kref *kref)
 908 {
 909         struct nfsd4_session *ses;
 910         int mem;
 911
 912         lockdep_assert_held(&client_lock);
 913         ses = container_of(kref, struct nfsd4_session, se_ref);
 914         nfsd4_del_conns(ses);
 915         spin_lock(&nfsd_drc_lock);
 916         mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
 917         nfsd_drc_mem_used -= mem;
 918         spin_unlock(&nfsd_drc_lock);
 919         free_session_slots(ses);
 920         kfree(ses);
 921 }
 922
 923 void nfsd4_put_session(struct nfsd4_session *ses)
 924 {
 925         spin_lock(&client_lock);
 926         nfsd4_put_session_locked(ses);
 927         spin_unlock(&client_lock);
 928 }
 929
 930 static struct nfsd4_session *alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp, struct nfsd4_create_session *cses)
 931 {
 932         struct nfsd4_session *new;
 933         struct nfsd4_channel_attrs *fchan = &cses->fore_channel;
 934         int numslots, slotsize;
 935         __be32 status;
 936         int idx;
 937
 938         /*
 939          * Note decreasing slot size below client's request may
 940          * make it difficult for client to function correctly, whereas
 941          * decreasing the number of slots will (just?) affect
 942          * performance.  When short on memory we therefore prefer to
 943          * decrease number of slots instead of their size.
 944          */
 945         slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
 946         numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
 947         if (numslots < 1)
 948                 return NULL;
 949
 950         new = alloc_session(slotsize, numslots);
 951         if (!new) {
 952                 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
 953                 return NULL;
 954         }
 955         init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
 956
 957         new->se_client = clp;
 958         gen_sessionid(new);
 959
 960         INIT_LIST_HEAD(&new->se_conns);
 961
 962         new->se_cb_seq_nr = 1;
 963         new->se_flags = cses->flags;
 964         new->se_cb_prog = cses->callback_prog;
 965         kref_init(&new->se_ref);
 966         idx = hash_sessionid(&new->se_sessionid);
 967         spin_lock(&client_lock);
 968         list_add(&new->se_hash, &sessionid_hashtbl[idx]);
 969         spin_lock(&clp->cl_lock);
 970         list_add(&new->se_perclnt, &clp->cl_sessions);
 971         spin_unlock(&clp->cl_lock);
 972         spin_unlock(&client_lock);
 973
 974         status = nfsd4_new_conn_from_crses(rqstp, new);
 975         /* whoops: benny points out, status is ignored! (err, or bogus) */
 976         if (status) {
 977                 spin_lock(&client_lock);
 978                 free_session(&new->se_ref);
 979                 spin_unlock(&client_lock);
 980                 return NULL;
 981         }
 982         if (cses->flags & SESSION4_BACK_CHAN) {
 983                 struct sockaddr *sa = svc_addr(rqstp);
 984                 /*
 985                  * This is a little silly; with sessions there's no real
 986                  * use for the callback address.  Use the peer address
 987                  * as a reasonable default for now, but consider fixing
 988                  * the rpc client not to require an address in the
 989                  * future:
 990                  */
 991                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
 992                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
 993         }
 994         nfsd4_probe_callback(clp);
 995         return new;
 996 }
 997
 998 /* caller must hold client_lock */
 999 static struct nfsd4_session *
1000 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
1001 {
1002         struct nfsd4_session *elem;
1003         int idx;
1004
1005         dump_sessionid(__func__, sessionid);
1006         idx = hash_sessionid(sessionid);
1007         /* Search in the appropriate list */
1008         list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
1009                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1010                             NFS4_MAX_SESSIONID_LEN)) {
1011                         return elem;
1012                 }
1013         }
1014
1015         dprintk("%s: session not found\n", __func__);
1016         return NULL;
1017 }
1018
1019 /* caller must hold client_lock */
1020 static void
1021 unhash_session(struct nfsd4_session *ses)
1022 {
1023         list_del(&ses->se_hash);
1024         spin_lock(&ses->se_client->cl_lock);
1025         list_del(&ses->se_perclnt);
1026         spin_unlock(&ses->se_client->cl_lock);
1027 }
1028
1029 /* must be called under the client_lock */
1030 static inline void
1031 renew_client_locked(struct nfs4_client *clp)
1032 {
1033         if (is_client_expired(clp)) {
1034                 WARN_ON(1);
1035                 printk("%s: client (clientid %08x/%08x) already expired\n",
1036                         __func__,
1037                         clp->cl_clientid.cl_boot,
1038                         clp->cl_clientid.cl_id);
1039                 return;
1040         }
1041
1042         dprintk("renewing client (clientid %08x/%08x)\n",
1043                         clp->cl_clientid.cl_boot,
1044                         clp->cl_clientid.cl_id);
1045         list_move_tail(&clp->cl_lru, &client_lru);
1046         clp->cl_time = get_seconds();
1047 }
1048
1049 static inline void
1050 renew_client(struct nfs4_client *clp)
1051 {
1052         spin_lock(&client_lock);
1053         renew_client_locked(clp);
1054         spin_unlock(&client_lock);
1055 }
1056
1057 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1058 static int
1059 STALE_CLIENTID(clientid_t *clid)
1060 {
1061         if (clid->cl_boot == boot_time)
1062                 return 0;
1063         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1064                 clid->cl_boot, clid->cl_id, boot_time);
1065         return 1;
1066 }
1067
1068 /*
1069  * XXX Should we use a slab cache ?
1070  * This type of memory management is somewhat inefficient, but we use it
1071  * anyway since SETCLIENTID is not a common operation.
1072  */
1073 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1074 {
1075         struct nfs4_client *clp;
1076
1077         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1078         if (clp == NULL)
1079                 return NULL;
1080         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1081         if (clp->cl_name.data == NULL) {
1082                 kfree(clp);
1083                 return NULL;
1084         }
1085         clp->cl_name.len = name.len;
1086         return clp;
1087 }
1088
1089 static inline void
1090 free_client(struct nfs4_client *clp)
1091 {
1092         lockdep_assert_held(&client_lock);
1093         while (!list_empty(&clp->cl_sessions)) {
1094                 struct nfsd4_session *ses;
1095                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1096                                 se_perclnt);
1097                 list_del(&ses->se_perclnt);
1098                 nfsd4_put_session_locked(ses);
1099         }
1100         free_svc_cred(&clp->cl_cred);
1101         kfree(clp->cl_name.data);
1102         kfree(clp);
1103 }
1104
1105 void
1106 release_session_client(struct nfsd4_session *session)
1107 {
1108         struct nfs4_client *clp = session->se_client;
1109
1110         if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
1111                 return;
1112         if (is_client_expired(clp)) {
1113                 free_client(clp);
1114                 session->se_client = NULL;
1115         } else
1116                 renew_client_locked(clp);
1117         spin_unlock(&client_lock);
1118 }
1119
1120 /* must be called under the client_lock */
1121 static inline void
1122 unhash_client_locked(struct nfs4_client *clp)
1123 {
1124         struct nfsd4_session *ses;
1125
1126         mark_client_expired(clp);
1127         list_del(&clp->cl_lru);
1128         spin_lock(&clp->cl_lock);
1129         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1130                 list_del_init(&ses->se_hash);
1131         spin_unlock(&clp->cl_lock);
1132 }
1133
1134 static void
1135 expire_client(struct nfs4_client *clp)
1136 {
1137         struct nfs4_openowner *oo;
1138         struct nfs4_delegation *dp;
1139         struct list_head reaplist;
1140
1141         INIT_LIST_HEAD(&reaplist);
1142         spin_lock(&recall_lock);
1143         while (!list_empty(&clp->cl_delegations)) {
1144                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1145                 list_del_init(&dp->dl_perclnt);
1146                 list_move(&dp->dl_recall_lru, &reaplist);
1147         }
1148         spin_unlock(&recall_lock);
1149         while (!list_empty(&reaplist)) {
1150                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1151                 unhash_delegation(dp);
1152         }
1153         while (!list_empty(&clp->cl_openowners)) {
1154                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1155                 release_openowner(oo);
1156         }
1157         nfsd4_shutdown_callback(clp);
1158         if (clp->cl_cb_conn.cb_xprt)
1159                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1160         list_del(&clp->cl_idhash);
1161         list_del(&clp->cl_strhash);
1162         spin_lock(&client_lock);
1163         unhash_client_locked(clp);
1164         if (atomic_read(&clp->cl_refcount) == 0)
1165                 free_client(clp);
1166         spin_unlock(&client_lock);
1167 }
1168
1169 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1170 {
1171         memcpy(target->cl_verifier.data, source->data,
1172                         sizeof(target->cl_verifier.data));
1173 }
1174
1175 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1176 {
1177         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1178         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1179 }
1180
1181 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1182 {
1183         if (source->cr_principal) {
1184                 target->cr_principal =
1185                                 kstrdup(source->cr_principal, GFP_KERNEL);
1186                 if (target->cr_principal == NULL)
1187                         return -ENOMEM;
1188         } else
1189                 target->cr_principal = NULL;
1190         target->cr_flavor = source->cr_flavor;
1191         target->cr_uid = source->cr_uid;
1192         target->cr_gid = source->cr_gid;
1193         target->cr_group_info = source->cr_group_info;
1194         get_group_info(target->cr_group_info);
1195         return 0;
1196 }
1197
1198 static int same_name(const char *n1, const char *n2)
1199 {
1200         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1201 }
1202
1203 static int
1204 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1205 {
1206         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1207 }
1208
1209 static int
1210 same_clid(clientid_t *cl1, clientid_t *cl2)
1211 {
1212         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1213 }
1214
1215 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1216 {
1217         int i;
1218
1219         if (g1->ngroups != g2->ngroups)
1220                 return false;
1221         for (i=0; i<g1->ngroups; i++)
1222                 if (GROUP_AT(g1, i) != GROUP_AT(g2, i))
1223                         return false;
1224         return true;
1225 }
1226
1227 static bool
1228 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1229 {
1230         if ((cr1->cr_flavor != cr2->cr_flavor)
1231                 || (cr1->cr_uid != cr2->cr_uid)
1232                 || (cr1->cr_gid != cr2->cr_gid)
1233                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1234                 return false;
1235         if (cr1->cr_principal == cr2->cr_principal)
1236                 return true;
1237         if (!cr1->cr_principal || !cr2->cr_principal)
1238                 return false;
1239         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1240 }
1241
1242 static void gen_clid(struct nfs4_client *clp)
1243 {
1244         static u32 current_clientid = 1;
1245
1246         clp->cl_clientid.cl_boot = boot_time;
1247         clp->cl_clientid.cl_id = current_clientid++;
1248 }
1249
1250 static void gen_confirm(struct nfs4_client *clp)
1251 {
1252         __be32 verf[2];
1253         static u32 i;
1254
1255         verf[0] = (__be32)get_seconds();
1256         verf[1] = (__be32)i++;
1257         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1258 }
1259
1260 static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1261 {
1262         return idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1263 }
1264
1265 static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1266 {
1267         struct nfs4_stid *s;
1268
1269         s = find_stateid(cl, t);
1270         if (!s)
1271                 return NULL;
1272         if (typemask & s->sc_type)
1273                 return s;
1274         return NULL;
1275 }
1276
1277 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1278                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1279 {
1280         struct nfs4_client *clp;
1281         struct sockaddr *sa = svc_addr(rqstp);
1282         int ret;
1283
1284         clp = alloc_client(name);
1285         if (clp == NULL)
1286                 return NULL;
1287
1288         INIT_LIST_HEAD(&clp->cl_sessions);
1289         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1290         if (ret) {
1291                 spin_lock(&client_lock);
1292                 free_client(clp);
1293                 spin_unlock(&client_lock);
1294                 return NULL;
1295         }
1296         idr_init(&clp->cl_stateids);
1297         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1298         atomic_set(&clp->cl_refcount, 0);
1299         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1300         INIT_LIST_HEAD(&clp->cl_idhash);
1301         INIT_LIST_HEAD(&clp->cl_strhash);
1302         INIT_LIST_HEAD(&clp->cl_openowners);
1303         INIT_LIST_HEAD(&clp->cl_delegations);
1304         INIT_LIST_HEAD(&clp->cl_lru);
1305         INIT_LIST_HEAD(&clp->cl_callbacks);
1306         spin_lock_init(&clp->cl_lock);
1307         INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1308         clp->cl_time = get_seconds();
1309         clear_bit(0, &clp->cl_cb_slot_busy);
1310         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1311         copy_verf(clp, verf);
1312         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1313         gen_confirm(clp);
1314         clp->cl_cb_session = NULL;
1315         return clp;
1316 }
1317
1318 static void
1319 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1320 {
1321         unsigned int idhashval;
1322
1323         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1324         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1325         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1326         renew_client(clp);
1327 }
1328
1329 static void
1330 move_to_confirmed(struct nfs4_client *clp)
1331 {
1332         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1333         unsigned int strhashval;
1334
1335         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1336         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1337         strhashval = clientstr_hashval(clp->cl_recdir);
1338         list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1339         renew_client(clp);
1340 }
1341
1342 static struct nfs4_client *
1343 find_confirmed_client(clientid_t *clid)
1344 {
1345         struct nfs4_client *clp;
1346         unsigned int idhashval = clientid_hashval(clid->cl_id);
1347
1348         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1349                 if (same_clid(&clp->cl_clientid, clid)) {
1350                         renew_client(clp);
1351                         return clp;
1352                 }
1353         }
1354         return NULL;
1355 }
1356
1357 static struct nfs4_client *
1358 find_unconfirmed_client(clientid_t *clid)
1359 {
1360         struct nfs4_client *clp;
1361         unsigned int idhashval = clientid_hashval(clid->cl_id);
1362
1363         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1364                 if (same_clid(&clp->cl_clientid, clid))
1365                         return clp;
1366         }
1367         return NULL;
1368 }
1369
1370 static bool clp_used_exchangeid(struct nfs4_client *clp)
1371 {
1372         return clp->cl_exchange_flags != 0;
1373 }
1374
1375 static struct nfs4_client *
1376 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
1377 {
1378         struct nfs4_client *clp;
1379
1380         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1381                 if (same_name(clp->cl_recdir, dname))
1382                         return clp;
1383         }
1384         return NULL;
1385 }
1386
1387 static struct nfs4_client *
1388 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
1389 {
1390         struct nfs4_client *clp;
1391
1392         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1393                 if (same_name(clp->cl_recdir, dname))
1394                         return clp;
1395         }
1396         return NULL;
1397 }
1398
1399 static void
1400 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1401 {
1402         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1403         struct sockaddr *sa = svc_addr(rqstp);
1404         u32 scopeid = rpc_get_scope_id(sa);
1405         unsigned short expected_family;
1406
1407         /* Currently, we only support tcp and tcp6 for the callback channel */
1408         if (se->se_callback_netid_len == 3 &&
1409             !memcmp(se->se_callback_netid_val, "tcp", 3))
1410                 expected_family = AF_INET;
1411         else if (se->se_callback_netid_len == 4 &&
1412                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1413                 expected_family = AF_INET6;
1414         else
1415                 goto out_err;
1416
1417         conn->cb_addrlen = rpc_uaddr2sockaddr(&init_net, se->se_callback_addr_val,
1418                                             se->se_callback_addr_len,
1419                                             (struct sockaddr *)&conn->cb_addr,
1420                                             sizeof(conn->cb_addr));
1421
1422         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1423                 goto out_err;
1424
1425         if (conn->cb_addr.ss_family == AF_INET6)
1426                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1427
1428         conn->cb_prog = se->se_callback_prog;
1429         conn->cb_ident = se->se_callback_ident;
1430         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1431         return;
1432 out_err:
1433         conn->cb_addr.ss_family = AF_UNSPEC;
1434         conn->cb_addrlen = 0;
1435         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1436                 "will not receive delegations\n",
1437                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1438
1439         return;
1440 }
1441
1442 /*
1443  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1444  */
1445 void
1446 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1447 {
1448         struct nfsd4_slot *slot = resp->cstate.slot;
1449         unsigned int base;
1450
1451         dprintk("--> %s slot %p\n", __func__, slot);
1452
1453         slot->sl_opcnt = resp->opcnt;
1454         slot->sl_status = resp->cstate.status;
1455
1456         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
1457         if (nfsd4_not_cached(resp)) {
1458                 slot->sl_datalen = 0;
1459                 return;
1460         }
1461         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1462         base = (char *)resp->cstate.datap -
1463                                         (char *)resp->xbuf->head[0].iov_base;
1464         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1465                                     slot->sl_datalen))
1466                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1467         return;
1468 }
1469
1470 /*
1471  * Encode the replay sequence operation from the slot values.
1472  * If cachethis is FALSE encode the uncached rep error on the next
1473  * operation which sets resp->p and increments resp->opcnt for
1474  * nfs4svc_encode_compoundres.
1475  *
1476  */
1477 static __be32
1478 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1479                           struct nfsd4_compoundres *resp)
1480 {
1481         struct nfsd4_op *op;
1482         struct nfsd4_slot *slot = resp->cstate.slot;
1483
1484         /* Encode the replayed sequence operation */
1485         op = &args->ops[resp->opcnt - 1];
1486         nfsd4_encode_operation(resp, op);
1487
1488         /* Return nfserr_retry_uncached_rep in next operation. */
1489         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
1490                 op = &args->ops[resp->opcnt++];
1491                 op->status = nfserr_retry_uncached_rep;
1492                 nfsd4_encode_operation(resp, op);
1493         }
1494         return op->status;
1495 }
1496
1497 /*
1498  * The sequence operation is not cached because we can use the slot and
1499  * session values.
1500  */
1501 __be32
1502 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1503                          struct nfsd4_sequence *seq)
1504 {
1505         struct nfsd4_slot *slot = resp->cstate.slot;
1506         __be32 status;
1507
1508         dprintk("--> %s slot %p\n", __func__, slot);
1509
1510         /* Either returns 0 or nfserr_retry_uncached */
1511         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1512         if (status == nfserr_retry_uncached_rep)
1513                 return status;
1514
1515         /* The sequence operation has been encoded, cstate->datap set. */
1516         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1517
1518         resp->opcnt = slot->sl_opcnt;
1519         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1520         status = slot->sl_status;
1521
1522         return status;
1523 }
1524
1525 /*
1526  * Set the exchange_id flags returned by the server.
1527  */
1528 static void
1529 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1530 {
1531         /* pNFS is not supported */
1532         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1533
1534         /* Referrals are supported, Migration is not. */
1535         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1536
1537         /* set the wire flags to return to client. */
1538         clid->flags = new->cl_exchange_flags;
1539 }
1540
1541 static bool client_has_state(struct nfs4_client *clp)
1542 {
1543         /*
1544          * Note clp->cl_openowners check isn't quite right: there's no
1545          * need to count owners without stateid's.
1546          *
1547          * Also note we should probably be using this in 4.0 case too.
1548          */
1549         return !list_empty(&clp->cl_openowners)
1550                 || !list_empty(&clp->cl_delegations)
1551                 || !list_empty(&clp->cl_sessions);
1552 }
1553
1554 __be32
1555 nfsd4_exchange_id(struct svc_rqst *rqstp,
1556                   struct nfsd4_compound_state *cstate,
1557                   struct nfsd4_exchange_id *exid)
1558 {
1559         struct nfs4_client *unconf, *conf, *new;
1560         __be32 status;
1561         unsigned int            strhashval;
1562         char                    dname[HEXDIR_LEN];
1563         char                    addr_str[INET6_ADDRSTRLEN];
1564         nfs4_verifier           verf = exid->verifier;
1565         struct sockaddr         *sa = svc_addr(rqstp);
1566         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
1567
1568         rpc_ntop(sa, addr_str, sizeof(addr_str));
1569         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1570                 "ip_addr=%s flags %x, spa_how %d\n",
1571                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1572                 addr_str, exid->flags, exid->spa_how);
1573
1574         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1575                 return nfserr_inval;
1576
1577         /* Currently only support SP4_NONE */
1578         switch (exid->spa_how) {
1579         case SP4_NONE:
1580                 break;
1581         case SP4_SSV:
1582                 return nfserr_serverfault;
1583         default:
1584                 BUG();                          /* checked by xdr code */
1585         case SP4_MACH_CRED:
1586                 return nfserr_serverfault;      /* no excuse :-/ */
1587         }
1588
1589         status = nfs4_make_rec_clidname(dname, &exid->clname);
1590
1591         if (status)
1592                 return status;
1593
1594         strhashval = clientstr_hashval(dname);
1595
1596         /* Cases below refer to rfc 5661 section 18.35.4: */
1597         nfs4_lock_state();
1598         conf = find_confirmed_client_by_str(dname, strhashval);
1599         if (conf) {
1600                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
1601                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
1602
1603                 if (update) {
1604                         if (!clp_used_exchangeid(conf)) { /* buggy client */
1605                                 status = nfserr_inval;
1606                                 goto out;
1607                         }
1608                         if (!creds_match) { /* case 9 */
1609                                 status = nfserr_perm;
1610                                 goto out;
1611                         }
1612                         if (!verfs_match) { /* case 8 */
1613                                 status = nfserr_not_same;
1614                                 goto out;
1615                         }
1616                         /* case 6 */
1617                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1618                         new = conf;
1619                         goto out_copy;
1620                 }
1621                 if (!creds_match) { /* case 3 */
1622                         if (client_has_state(conf)) {
1623                                 status = nfserr_clid_inuse;
1624                                 goto out;
1625                         }
1626                         expire_client(conf);
1627                         goto out_new;
1628                 }
1629                 if (verfs_match) { /* case 2 */
1630                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
1631                         new = conf;
1632                         goto out_copy;
1633                 }
1634                 /* case 5, client reboot */
1635                 goto out_new;
1636         }
1637
1638         if (update) { /* case 7 */
1639                 status = nfserr_noent;
1640                 goto out;
1641         }
1642
1643         unconf  = find_unconfirmed_client_by_str(dname, strhashval);
1644         if (unconf) /* case 4, possible retry or client restart */
1645                 expire_client(unconf);
1646
1647         /* case 1 (normal case) */
1648 out_new:
1649         new = create_client(exid->clname, dname, rqstp, &verf);
1650         if (new == NULL) {
1651                 status = nfserr_jukebox;
1652                 goto out;
1653         }
1654
1655         gen_clid(new);
1656         add_to_unconfirmed(new, strhashval);
1657 out_copy:
1658         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1659         exid->clientid.cl_id = new->cl_clientid.cl_id;
1660
1661         exid->seqid = new->cl_cs_slot.sl_seqid + 1;
1662         nfsd4_set_ex_flags(new, exid);
1663
1664         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1665                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1666         status = nfs_ok;
1667
1668 out:
1669         nfs4_unlock_state();
1670         return status;
1671 }
1672
1673 static __be32
1674 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1675 {
1676         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1677                 slot_seqid);
1678
1679         /* The slot is in use, and no response has been sent. */
1680         if (slot_inuse) {
1681                 if (seqid == slot_seqid)
1682                         return nfserr_jukebox;
1683                 else
1684                         return nfserr_seq_misordered;
1685         }
1686         /* Note unsigned 32-bit arithmetic handles wraparound: */
1687         if (likely(seqid == slot_seqid + 1))
1688                 return nfs_ok;
1689         if (seqid == slot_seqid)
1690                 return nfserr_replay_cache;
1691         return nfserr_seq_misordered;
1692 }
1693
1694 /*
1695  * Cache the create session result into the create session single DRC
1696  * slot cache by saving the xdr structure. sl_seqid has been set.
1697  * Do this for solo or embedded create session operations.
1698  */
1699 static void
1700 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1701                            struct nfsd4_clid_slot *slot, __be32 nfserr)
1702 {
1703         slot->sl_status = nfserr;
1704         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1705 }
1706
1707 static __be32
1708 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1709                             struct nfsd4_clid_slot *slot)
1710 {
1711         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1712         return slot->sl_status;
1713 }
1714
1715 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1716                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1717                         1 +     /* MIN tag is length with zero, only length */ \
1718                         3 +     /* version, opcount, opcode */ \
1719                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1720                                 /* seqid, slotID, slotID, cache */ \
1721                         4 ) * sizeof(__be32))
1722
1723 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1724                         2 +     /* verifier: AUTH_NULL, length 0 */\
1725                         1 +     /* status */ \
1726                         1 +     /* MIN tag is length with zero, only length */ \
1727                         3 +     /* opcount, opcode, opstatus*/ \
1728                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1729                                 /* seqid, slotID, slotID, slotID, status */ \
1730                         5 ) * sizeof(__be32))
1731
1732 static bool check_forechannel_attrs(struct nfsd4_channel_attrs fchannel)
1733 {
1734         return fchannel.maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ
1735                 || fchannel.maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ;
1736 }
1737
1738 __be32
1739 nfsd4_create_session(struct svc_rqst *rqstp,
1740                      struct nfsd4_compound_state *cstate,
1741                      struct nfsd4_create_session *cr_ses)
1742 {
1743         struct sockaddr *sa = svc_addr(rqstp);
1744         struct nfs4_client *conf, *unconf;
1745         struct nfsd4_session *new;
1746         struct nfsd4_clid_slot *cs_slot = NULL;
1747         bool confirm_me = false;
1748         __be32 status = 0;
1749
1750         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1751                 return nfserr_inval;
1752
1753         nfs4_lock_state();
1754         unconf = find_unconfirmed_client(&cr_ses->clientid);
1755         conf = find_confirmed_client(&cr_ses->clientid);
1756
1757         if (conf) {
1758                 cs_slot = &conf->cl_cs_slot;
1759                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1760                 if (status == nfserr_replay_cache) {
1761                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1762                         goto out;
1763                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1764                         status = nfserr_seq_misordered;
1765                         goto out;
1766                 }
1767         } else if (unconf) {
1768                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1769                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1770                         status = nfserr_clid_inuse;
1771                         goto out;
1772                 }
1773                 cs_slot = &unconf->cl_cs_slot;
1774                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1775                 if (status) {
1776                         /* an unconfirmed replay returns misordered */
1777                         status = nfserr_seq_misordered;
1778                         goto out;
1779                 }
1780                 confirm_me = true;
1781                 conf = unconf;
1782         } else {
1783                 status = nfserr_stale_clientid;
1784                 goto out;
1785         }
1786
1787         /*
1788          * XXX: we should probably set this at creation time, and check
1789          * for consistent minorversion use throughout:
1790          */
1791         conf->cl_minorversion = 1;
1792         /*
1793          * We do not support RDMA or persistent sessions
1794          */
1795         cr_ses->flags &= ~SESSION4_PERSIST;
1796         cr_ses->flags &= ~SESSION4_RDMA;
1797
1798         status = nfserr_toosmall;
1799         if (check_forechannel_attrs(cr_ses->fore_channel))
1800                 goto out;
1801
1802         status = nfserr_jukebox;
1803         new = alloc_init_session(rqstp, conf, cr_ses);
1804         if (!new)
1805                 goto out;
1806         status = nfs_ok;
1807         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1808                NFS4_MAX_SESSIONID_LEN);
1809         memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1810                 sizeof(struct nfsd4_channel_attrs));
1811         cs_slot->sl_seqid++;
1812         cr_ses->seqid = cs_slot->sl_seqid;
1813
1814         /* cache solo and embedded create sessions under the state lock */
1815         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1816         if (confirm_me) {
1817                 unsigned int hash = clientstr_hashval(unconf->cl_recdir);
1818                 struct nfs4_client *old =
1819                         find_confirmed_client_by_str(conf->cl_recdir, hash);
1820                 if (old)
1821                         expire_client(old);
1822                 move_to_confirmed(conf);
1823         }
1824 out:
1825         nfs4_unlock_state();
1826         dprintk("%s returns %d\n", __func__, ntohl(status));
1827         return status;
1828 }
1829
1830 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1831 {
1832         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1833         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1834
1835         return argp->opcnt == resp->opcnt;
1836 }
1837
1838 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1839 {
1840         switch (*dir) {
1841         case NFS4_CDFC4_FORE:
1842         case NFS4_CDFC4_BACK:
1843                 return nfs_ok;
1844         case NFS4_CDFC4_FORE_OR_BOTH:
1845         case NFS4_CDFC4_BACK_OR_BOTH:
1846                 *dir = NFS4_CDFC4_BOTH;
1847                 return nfs_ok;
1848         };
1849         return nfserr_inval;
1850 }
1851
1852 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1853                      struct nfsd4_compound_state *cstate,
1854                      struct nfsd4_bind_conn_to_session *bcts)
1855 {
1856         __be32 status;
1857
1858         if (!nfsd4_last_compound_op(rqstp))
1859                 return nfserr_not_only_op;
1860         spin_lock(&client_lock);
1861         cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid);
1862         /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1863          * client_lock iself: */
1864         if (cstate->session) {
1865                 nfsd4_get_session(cstate->session);
1866                 atomic_inc(&cstate->session->se_client->cl_refcount);
1867         }
1868         spin_unlock(&client_lock);
1869         if (!cstate->session)
1870                 return nfserr_badsession;
1871
1872         status = nfsd4_map_bcts_dir(&bcts->dir);
1873         if (!status)
1874                 nfsd4_new_conn(rqstp, cstate->session, bcts->dir);
1875         return status;
1876 }
1877
1878 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1879 {
1880         if (!session)
1881                 return 0;
1882         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1883 }
1884
1885 __be32
1886 nfsd4_destroy_session(struct svc_rqst *r,
1887                       struct nfsd4_compound_state *cstate,
1888                       struct nfsd4_destroy_session *sessionid)
1889 {
1890         struct nfsd4_session *ses;
1891         __be32 status = nfserr_badsession;
1892
1893         /* Notes:
1894          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1895          * - Should we return nfserr_back_chan_busy if waiting for
1896          *   callbacks on to-be-destroyed session?
1897          * - Do we need to clear any callback info from previous session?
1898          */
1899
1900         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1901                 if (!nfsd4_last_compound_op(r))
1902                         return nfserr_not_only_op;
1903         }
1904         dump_sessionid(__func__, &sessionid->sessionid);
1905         spin_lock(&client_lock);
1906         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1907         if (!ses) {
1908                 spin_unlock(&client_lock);
1909                 goto out;
1910         }
1911
1912         unhash_session(ses);
1913         spin_unlock(&client_lock);
1914
1915         nfs4_lock_state();
1916         nfsd4_probe_callback_sync(ses->se_client);
1917         nfs4_unlock_state();
1918
1919         spin_lock(&client_lock);
1920         nfsd4_del_conns(ses);
1921         nfsd4_put_session_locked(ses);
1922         spin_unlock(&client_lock);
1923         status = nfs_ok;
1924 out:
1925         dprintk("%s returns %d\n", __func__, ntohl(status));
1926         return status;
1927 }
1928
1929 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1930 {
1931         struct nfsd4_conn *c;
1932
1933         list_for_each_entry(c, &s->se_conns, cn_persession) {
1934                 if (c->cn_xprt == xpt) {
1935                         return c;
1936                 }
1937         }
1938         return NULL;
1939 }
1940
1941 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1942 {
1943         struct nfs4_client *clp = ses->se_client;
1944         struct nfsd4_conn *c;
1945         int ret;
1946
1947         spin_lock(&clp->cl_lock);
1948         c = __nfsd4_find_conn(new->cn_xprt, ses);
1949         if (c) {
1950                 spin_unlock(&clp->cl_lock);
1951                 free_conn(new);
1952                 return;
1953         }
1954         __nfsd4_hash_conn(new, ses);
1955         spin_unlock(&clp->cl_lock);
1956         ret = nfsd4_register_conn(new);
1957         if (ret)
1958                 /* oops; xprt is already down: */
1959                 nfsd4_conn_lost(&new->cn_xpt_user);
1960         return;
1961 }
1962
1963 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
1964 {
1965         struct nfsd4_compoundargs *args = rqstp->rq_argp;
1966
1967         return args->opcnt > session->se_fchannel.maxops;
1968 }
1969
1970 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
1971                                   struct nfsd4_session *session)
1972 {
1973         struct xdr_buf *xb = &rqstp->rq_arg;
1974
1975         return xb->len > session->se_fchannel.maxreq_sz;
1976 }
1977
1978 __be32
1979 nfsd4_sequence(struct svc_rqst *rqstp,
1980                struct nfsd4_compound_state *cstate,
1981                struct nfsd4_sequence *seq)
1982 {
1983         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1984         struct nfsd4_session *session;
1985         struct nfsd4_slot *slot;
1986         struct nfsd4_conn *conn;
1987         __be32 status;
1988
1989         if (resp->opcnt != 1)
1990                 return nfserr_sequence_pos;
1991
1992         /*
1993          * Will be either used or freed by nfsd4_sequence_check_conn
1994          * below.
1995          */
1996         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
1997         if (!conn)
1998                 return nfserr_jukebox;
1999
2000         spin_lock(&client_lock);
2001         status = nfserr_badsession;
2002         session = find_in_sessionid_hashtbl(&seq->sessionid);
2003         if (!session)
2004                 goto out;
2005
2006         status = nfserr_too_many_ops;
2007         if (nfsd4_session_too_many_ops(rqstp, session))
2008                 goto out;
2009
2010         status = nfserr_req_too_big;
2011         if (nfsd4_request_too_big(rqstp, session))
2012                 goto out;
2013
2014         status = nfserr_badslot;
2015         if (seq->slotid >= session->se_fchannel.maxreqs)
2016                 goto out;
2017
2018         slot = session->se_slots[seq->slotid];
2019         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2020
2021         /* We do not negotiate the number of slots yet, so set the
2022          * maxslots to the session maxreqs which is used to encode
2023          * sr_highest_slotid and the sr_target_slot id to maxslots */
2024         seq->maxslots = session->se_fchannel.maxreqs;
2025
2026         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2027                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2028         if (status == nfserr_replay_cache) {
2029                 status = nfserr_seq_misordered;
2030                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2031                         goto out;
2032                 cstate->slot = slot;
2033                 cstate->session = session;
2034                 /* Return the cached reply status and set cstate->status
2035                  * for nfsd4_proc_compound processing */
2036                 status = nfsd4_replay_cache_entry(resp, seq);
2037                 cstate->status = nfserr_replay_cache;
2038                 goto out;
2039         }
2040         if (status)
2041                 goto out;
2042
2043         nfsd4_sequence_check_conn(conn, session);
2044         conn = NULL;
2045
2046         /* Success! bump slot seqid */
2047         slot->sl_seqid = seq->seqid;
2048         slot->sl_flags |= NFSD4_SLOT_INUSE;
2049         if (seq->cachethis)
2050                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2051         else
2052                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2053
2054         cstate->slot = slot;
2055         cstate->session = session;
2056
2057 out:
2058         /* Hold a session reference until done processing the compound. */
2059         if (cstate->session) {
2060                 struct nfs4_client *clp = session->se_client;
2061
2062                 nfsd4_get_session(cstate->session);
2063                 atomic_inc(&clp->cl_refcount);
2064                 switch (clp->cl_cb_state) {
2065                 case NFSD4_CB_DOWN:
2066                         seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2067                         break;
2068                 case NFSD4_CB_FAULT:
2069                         seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2070                         break;
2071                 default:
2072                         seq->status_flags = 0;
2073                 }
2074         }
2075         kfree(conn);
2076         spin_unlock(&client_lock);
2077         dprintk("%s: return %d\n", __func__, ntohl(status));
2078         return status;
2079 }
2080
2081 __be32
2082 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2083 {
2084         struct nfs4_client *conf, *unconf, *clp;
2085         __be32 status = 0;
2086
2087         nfs4_lock_state();
2088         unconf = find_unconfirmed_client(&dc->clientid);
2089         conf = find_confirmed_client(&dc->clientid);
2090
2091         if (conf) {
2092                 clp = conf;
2093
2094                 if (!is_client_expired(conf) && client_has_state(conf)) {
2095                         status = nfserr_clientid_busy;
2096                         goto out;
2097                 }
2098
2099                 /* rfc5661 18.50.3 */
2100                 if (cstate->session && conf == cstate->session->se_client) {
2101                         status = nfserr_clientid_busy;
2102                         goto out;
2103                 }
2104         } else if (unconf)
2105                 clp = unconf;
2106         else {
2107                 status = nfserr_stale_clientid;
2108                 goto out;
2109         }
2110
2111         expire_client(clp);
2112 out:
2113         nfs4_unlock_state();
2114         dprintk("%s return %d\n", __func__, ntohl(status));
2115         return status;
2116 }
2117
2118 __be32
2119 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2120 {
2121         __be32 status = 0;
2122
2123         if (rc->rca_one_fs) {
2124                 if (!cstate->current_fh.fh_dentry)
2125                         return nfserr_nofilehandle;
2126                 /*
2127                  * We don't take advantage of the rca_one_fs case.
2128                  * That's OK, it's optional, we can safely ignore it.
2129                  */
2130                  return nfs_ok;
2131         }
2132
2133         nfs4_lock_state();
2134         status = nfserr_complete_already;
2135         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2136                              &cstate->session->se_client->cl_flags))
2137                 goto out;
2138
2139         status = nfserr_stale_clientid;
2140         if (is_client_expired(cstate->session->se_client))
2141                 /*
2142                  * The following error isn't really legal.
2143                  * But we only get here if the client just explicitly
2144                  * destroyed the client.  Surely it no longer cares what
2145                  * error it gets back on an operation for the dead
2146                  * client.
2147                  */
2148                 goto out;
2149
2150         status = nfs_ok;
2151         nfsd4_client_record_create(cstate->session->se_client);
2152 out:
2153         nfs4_unlock_state();
2154         return status;
2155 }
2156
2157 __be32
2158 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2159                   struct nfsd4_setclientid *setclid)
2160 {
2161         struct xdr_netobj       clname = setclid->se_name;
2162         nfs4_verifier           clverifier = setclid->se_verf;
2163         unsigned int            strhashval;
2164         struct nfs4_client      *conf, *unconf, *new;
2165         __be32                  status;
2166         char                    dname[HEXDIR_LEN];
2167
2168         status = nfs4_make_rec_clidname(dname, &clname);
2169         if (status)
2170                 return status;
2171
2172         strhashval = clientstr_hashval(dname);
2173
2174         /* Cases below refer to rfc 3530 section 14.2.33: */
2175         nfs4_lock_state();
2176         conf = find_confirmed_client_by_str(dname, strhashval);
2177         if (conf) {
2178                 /* case 0: */
2179                 status = nfserr_clid_inuse;
2180                 if (clp_used_exchangeid(conf))
2181                         goto out;
2182                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2183                         char addr_str[INET6_ADDRSTRLEN];
2184                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2185                                  sizeof(addr_str));
2186                         dprintk("NFSD: setclientid: string in use by client "
2187                                 "at %s\n", addr_str);
2188                         goto out;
2189                 }
2190         }
2191         unconf = find_unconfirmed_client_by_str(dname, strhashval);
2192         if (unconf)
2193                 expire_client(unconf);
2194         status = nfserr_jukebox;
2195         new = create_client(clname, dname, rqstp, &clverifier);
2196         if (new == NULL)
2197                 goto out;
2198         if (conf && same_verf(&conf->cl_verifier, &clverifier))
2199                 /* case 1: probable callback update */
2200                 copy_clid(new, conf);
2201         else /* case 4 (new client) or cases 2, 3 (client reboot): */
2202                 gen_clid(new);
2203         /*
2204          * XXX: we should probably set this at creation time, and check
2205          * for consistent minorversion use throughout:
2206          */
2207         new->cl_minorversion = 0;
2208         gen_callback(new, setclid, rqstp);
2209         add_to_unconfirmed(new, strhashval);
2210         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2211         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2212         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2213         status = nfs_ok;
2214 out:
2215         nfs4_unlock_state();
2216         return status;
2217 }
2218
2219
2220 __be32
2221 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2222                          struct nfsd4_compound_state *cstate,
2223                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2224 {
2225         struct nfs4_client *conf, *unconf;
2226         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2227         clientid_t * clid = &setclientid_confirm->sc_clientid;
2228         __be32 status;
2229
2230         if (STALE_CLIENTID(clid))
2231                 return nfserr_stale_clientid;
2232         nfs4_lock_state();
2233
2234         conf = find_confirmed_client(clid);
2235         unconf = find_unconfirmed_client(clid);
2236         /*
2237          * We try hard to give out unique clientid's, so if we get an
2238          * attempt to confirm the same clientid with a different cred,
2239          * there's a bug somewhere.  Let's charitably assume it's our
2240          * bug.
2241          */
2242         status = nfserr_serverfault;
2243         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
2244                 goto out;
2245         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
2246                 goto out;
2247         /* cases below refer to rfc 3530 section 14.2.34: */
2248         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
2249                 if (conf && !unconf) /* case 2: probable retransmit */
2250                         status = nfs_ok;
2251                 else /* case 4: client hasn't noticed we rebooted yet? */
2252                         status = nfserr_stale_clientid;
2253                 goto out;
2254         }
2255         status = nfs_ok;
2256         if (conf) { /* case 1: callback update */
2257                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2258                 nfsd4_probe_callback(conf);
2259                 expire_client(unconf);
2260         } else { /* case 3: normal case; new or rebooted client */
2261                 unsigned int hash = clientstr_hashval(unconf->cl_recdir);
2262
2263                 conf = find_confirmed_client_by_str(unconf->cl_recdir, hash);
2264                 if (conf) {
2265                         nfsd4_client_record_remove(conf);
2266                         expire_client(conf);
2267                 }
2268                 move_to_confirmed(unconf);
2269                 nfsd4_probe_callback(unconf);
2270         }
2271 out:
2272         nfs4_unlock_state();
2273         return status;
2274 }
2275
2276 static struct nfs4_file *nfsd4_alloc_file(void)
2277 {
2278         return kmem_cache_alloc(file_slab, GFP_KERNEL);
2279 }
2280
2281 /* OPEN Share state helper functions */
2282 static void nfsd4_init_file(struct nfs4_file *fp, struct inode *ino)
2283 {
2284         unsigned int hashval = file_hashval(ino);
2285
2286         atomic_set(&fp->fi_ref, 1);
2287         INIT_LIST_HEAD(&fp->fi_hash);
2288         INIT_LIST_HEAD(&fp->fi_stateids);
2289         INIT_LIST_HEAD(&fp->fi_delegations);
2290         fp->fi_inode = igrab(ino);
2291         fp->fi_had_conflict = false;
2292         fp->fi_lease = NULL;
2293         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2294         memset(fp->fi_access, 0, sizeof(fp->fi_access));
2295         spin_lock(&recall_lock);
2296         list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2297         spin_unlock(&recall_lock);
2298 }
2299
2300 static void
2301 nfsd4_free_slab(struct kmem_cache **slab)
2302 {
2303         if (*slab == NULL)
2304                 return;
2305         kmem_cache_destroy(*slab);
2306         *slab = NULL;
2307 }
2308
2309 void
2310 nfsd4_free_slabs(void)
2311 {
2312         nfsd4_free_slab(&openowner_slab);
2313         nfsd4_free_slab(&lockowner_slab);
2314         nfsd4_free_slab(&file_slab);
2315         nfsd4_free_slab(&stateid_slab);
2316         nfsd4_free_slab(&deleg_slab);
2317 }
2318
2319 int
2320 nfsd4_init_slabs(void)
2321 {
2322         openowner_slab = kmem_cache_create("nfsd4_openowners",
2323                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2324         if (openowner_slab == NULL)
2325                 goto out_nomem;
2326         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2327                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2328         if (lockowner_slab == NULL)
2329                 goto out_nomem;
2330         file_slab = kmem_cache_create("nfsd4_files",
2331                         sizeof(struct nfs4_file), 0, 0, NULL);
2332         if (file_slab == NULL)
2333                 goto out_nomem;
2334         stateid_slab = kmem_cache_create("nfsd4_stateids",
2335                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2336         if (stateid_slab == NULL)
2337                 goto out_nomem;
2338         deleg_slab = kmem_cache_create("nfsd4_delegations",
2339                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2340         if (deleg_slab == NULL)
2341                 goto out_nomem;
2342         return 0;
2343 out_nomem:
2344         nfsd4_free_slabs();
2345         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2346         return -ENOMEM;
2347 }
2348
2349 void nfs4_free_openowner(struct nfs4_openowner *oo)
2350 {
2351         kfree(oo->oo_owner.so_owner.data);
2352         kmem_cache_free(openowner_slab, oo);
2353 }
2354
2355 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2356 {
2357         kfree(lo->lo_owner.so_owner.data);
2358         kmem_cache_free(lockowner_slab, lo);
2359 }
2360
2361 static void init_nfs4_replay(struct nfs4_replay *rp)
2362 {
2363         rp->rp_status = nfserr_serverfault;
2364         rp->rp_buflen = 0;
2365         rp->rp_buf = rp->rp_ibuf;
2366 }
2367
2368 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2369 {
2370         struct nfs4_stateowner *sop;
2371
2372         sop = kmem_cache_alloc(slab, GFP_KERNEL);
2373         if (!sop)
2374                 return NULL;
2375
2376         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2377         if (!sop->so_owner.data) {
2378                 kmem_cache_free(slab, sop);
2379                 return NULL;
2380         }
2381         sop->so_owner.len = owner->len;
2382
2383         INIT_LIST_HEAD(&sop->so_stateids);
2384         sop->so_client = clp;
2385         init_nfs4_replay(&sop->so_replay);
2386         return sop;
2387 }
2388
2389 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2390 {
2391         list_add(&oo->oo_owner.so_strhash, &ownerstr_hashtbl[strhashval]);
2392         list_add(&oo->oo_perclient, &clp->cl_openowners);
2393 }
2394
2395 static struct nfs4_openowner *
2396 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2397         struct nfs4_openowner *oo;
2398
2399         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2400         if (!oo)
2401                 return NULL;
2402         oo->oo_owner.so_is_open_owner = 1;
2403         oo->oo_owner.so_seqid = open->op_seqid;
2404         oo->oo_flags = NFS4_OO_NEW;
2405         oo->oo_time = 0;
2406         oo->oo_last_closed_stid = NULL;
2407         INIT_LIST_HEAD(&oo->oo_close_lru);
2408         hash_openowner(oo, clp, strhashval);
2409         return oo;
2410 }
2411
2412 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2413         struct nfs4_openowner *oo = open->op_openowner;
2414         struct nfs4_client *clp = oo->oo_owner.so_client;
2415
2416         init_stid(&stp->st_stid, clp, NFS4_OPEN_STID);
2417         INIT_LIST_HEAD(&stp->st_lockowners);
2418         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2419         list_add(&stp->st_perfile, &fp->fi_stateids);
2420         stp->st_stateowner = &oo->oo_owner;
2421         get_nfs4_file(fp);
2422         stp->st_file = fp;
2423         stp->st_access_bmap = 0;
2424         stp->st_deny_bmap = 0;
2425         set_access(open->op_share_access, stp);
2426         set_deny(open->op_share_deny, stp);
2427         stp->st_openstp = NULL;
2428 }
2429
2430 static void
2431 move_to_close_lru(struct nfs4_openowner *oo)
2432 {
2433         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2434
2435         list_move_tail(&oo->oo_close_lru, &close_lru);
2436         oo->oo_time = get_seconds();
2437 }
2438
2439 static int
2440 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2441                                                         clientid_t *clid)
2442 {
2443         return (sop->so_owner.len == owner->len) &&
2444                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2445                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2446 }
2447
2448 static struct nfs4_openowner *
2449 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2450 {
2451         struct nfs4_stateowner *so;
2452         struct nfs4_openowner *oo;
2453
2454         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2455                 if (!so->so_is_open_owner)
2456                         continue;
2457                 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2458                         oo = openowner(so);
2459                         renew_client(oo->oo_owner.so_client);
2460                         return oo;
2461                 }
2462         }
2463         return NULL;
2464 }
2465
2466 /* search file_hashtbl[] for file */
2467 static struct nfs4_file *
2468 find_file(struct inode *ino)
2469 {
2470         unsigned int hashval = file_hashval(ino);
2471         struct nfs4_file *fp;
2472
2473         spin_lock(&recall_lock);
2474         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2475                 if (fp->fi_inode == ino) {
2476                         get_nfs4_file(fp);
2477                         spin_unlock(&recall_lock);
2478                         return fp;
2479                 }
2480         }
2481         spin_unlock(&recall_lock);
2482         return NULL;
2483 }
2484
2485 /*
2486  * Called to check deny when READ with all zero stateid or
2487  * WRITE with all zero or all one stateid
2488  */
2489 static __be32
2490 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2491 {
2492         struct inode *ino = current_fh->fh_dentry->d_inode;
2493         struct nfs4_file *fp;
2494         struct nfs4_ol_stateid *stp;
2495         __be32 ret;
2496
2497         dprintk("NFSD: nfs4_share_conflict\n");
2498
2499         fp = find_file(ino);
2500         if (!fp)
2501                 return nfs_ok;
2502         ret = nfserr_locked;
2503         /* Search for conflicting share reservations */
2504         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2505                 if (test_deny(deny_type, stp) ||
2506                     test_deny(NFS4_SHARE_DENY_BOTH, stp))
2507                         goto out;
2508         }
2509         ret = nfs_ok;
2510 out:
2511         put_nfs4_file(fp);
2512         return ret;
2513 }
2514
2515 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2516 {
2517         /* We're assuming the state code never drops its reference
2518          * without first removing the lease.  Since we're in this lease
2519          * callback (and since the lease code is serialized by the kernel
2520          * lock) we know the server hasn't removed the lease yet, we know
2521          * it's safe to take a reference: */
2522         atomic_inc(&dp->dl_count);
2523
2524         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2525
2526         /* only place dl_time is set. protected by lock_flocks*/
2527         dp->dl_time = get_seconds();
2528
2529         nfsd4_cb_recall(dp);
2530 }
2531
2532 /* Called from break_lease() with lock_flocks() held. */
2533 static void nfsd_break_deleg_cb(struct file_lock *fl)
2534 {
2535         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2536         struct nfs4_delegation *dp;
2537
2538         BUG_ON(!fp);
2539         /* We assume break_lease is only called once per lease: */
2540         BUG_ON(fp->fi_had_conflict);
2541         /*
2542          * We don't want the locks code to timeout the lease for us;
2543          * we'll remove it ourself if a delegation isn't returned
2544          * in time:
2545          */
2546         fl->fl_break_time = 0;
2547
2548         spin_lock(&recall_lock);
2549         fp->fi_had_conflict = true;
2550         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2551                 nfsd_break_one_deleg(dp);
2552         spin_unlock(&recall_lock);
2553 }
2554
2555 static
2556 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2557 {
2558         if (arg & F_UNLCK)
2559                 return lease_modify(onlist, arg);
2560         else
2561                 return -EAGAIN;
2562 }
2563
2564 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2565         .lm_break = nfsd_break_deleg_cb,
2566         .lm_change = nfsd_change_deleg_cb,
2567 };
2568
2569 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2570 {
2571         if (nfsd4_has_session(cstate))
2572                 return nfs_ok;
2573         if (seqid == so->so_seqid - 1)
2574                 return nfserr_replay_me;
2575         if (seqid == so->so_seqid)
2576                 return nfs_ok;
2577         return nfserr_bad_seqid;
2578 }
2579
2580 __be32
2581 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2582                     struct nfsd4_open *open)
2583 {
2584         clientid_t *clientid = &open->op_clientid;
2585         struct nfs4_client *clp = NULL;
2586         unsigned int strhashval;
2587         struct nfs4_openowner *oo = NULL;
2588         __be32 status;
2589
2590         if (STALE_CLIENTID(&open->op_clientid))
2591                 return nfserr_stale_clientid;
2592         /*
2593          * In case we need it later, after we've already created the
2594          * file and don't want to risk a further failure:
2595          */
2596         open->op_file = nfsd4_alloc_file();
2597         if (open->op_file == NULL)
2598                 return nfserr_jukebox;
2599
2600         strhashval = ownerstr_hashval(clientid->cl_id, &open->op_owner);
2601         oo = find_openstateowner_str(strhashval, open);
2602         open->op_openowner = oo;
2603         if (!oo) {
2604                 clp = find_confirmed_client(clientid);
2605                 if (clp == NULL)
2606                         return nfserr_expired;
2607                 goto new_owner;
2608         }
2609         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2610                 /* Replace unconfirmed owners without checking for replay. */
2611                 clp = oo->oo_owner.so_client;
2612                 release_openowner(oo);
2613                 open->op_openowner = NULL;
2614                 goto new_owner;
2615         }
2616         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2617         if (status)
2618                 return status;
2619         clp = oo->oo_owner.so_client;
2620         goto alloc_stateid;
2621 new_owner:
2622         oo = alloc_init_open_stateowner(strhashval, clp, open);
2623         if (oo == NULL)
2624                 return nfserr_jukebox;
2625         open->op_openowner = oo;
2626 alloc_stateid:
2627         open->op_stp = nfs4_alloc_stateid(clp);
2628         if (!open->op_stp)
2629                 return nfserr_jukebox;
2630         return nfs_ok;
2631 }
2632
2633 static inline __be32
2634 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2635 {
2636         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2637                 return nfserr_openmode;
2638         else
2639                 return nfs_ok;
2640 }
2641
2642 static int share_access_to_flags(u32 share_access)
2643 {
2644         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2645 }
2646
2647 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2648 {
2649         struct nfs4_stid *ret;
2650
2651         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2652         if (!ret)
2653                 return NULL;
2654         return delegstateid(ret);
2655 }
2656
2657 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
2658 {
2659         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
2660                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
2661 }
2662
2663 static __be32
2664 nfs4_check_deleg(struct nfs4_client *cl, struct nfs4_file *fp, struct nfsd4_open *open,
2665                 struct nfs4_delegation **dp)
2666 {
2667         int flags;
2668         __be32 status = nfserr_bad_stateid;
2669
2670         *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2671         if (*dp == NULL)
2672                 goto out;
2673         flags = share_access_to_flags(open->op_share_access);
2674         status = nfs4_check_delegmode(*dp, flags);
2675         if (status)
2676                 *dp = NULL;
2677 out:
2678         if (!nfsd4_is_deleg_cur(open))
2679                 return nfs_ok;
2680         if (status)
2681                 return status;
2682         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2683         return nfs_ok;
2684 }
2685
2686 static __be32
2687 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2688 {
2689         struct nfs4_ol_stateid *local;
2690         struct nfs4_openowner *oo = open->op_openowner;
2691
2692         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2693                 /* ignore lock owners */
2694                 if (local->st_stateowner->so_is_open_owner == 0)
2695                         continue;
2696                 /* remember if we have seen this open owner */
2697                 if (local->st_stateowner == &oo->oo_owner)
2698                         *stpp = local;
2699                 /* check for conflicting share reservations */
2700                 if (!test_share(local, open))
2701                         return nfserr_share_denied;
2702         }
2703         return nfs_ok;
2704 }
2705
2706 static void nfs4_free_stateid(struct nfs4_ol_stateid *s)
2707 {
2708         kmem_cache_free(stateid_slab, s);
2709 }
2710
2711 static inline int nfs4_access_to_access(u32 nfs4_access)
2712 {
2713         int flags = 0;
2714
2715         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2716                 flags |= NFSD_MAY_READ;
2717         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2718                 flags |= NFSD_MAY_WRITE;
2719         return flags;
2720 }
2721
2722 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2723                 struct svc_fh *cur_fh, struct nfsd4_open *open)
2724 {
2725         __be32 status;
2726         int oflag = nfs4_access_to_omode(open->op_share_access);
2727         int access = nfs4_access_to_access(open->op_share_access);
2728
2729         if (!fp->fi_fds[oflag]) {
2730                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2731                         &fp->fi_fds[oflag]);
2732                 if (status)
2733                         return status;
2734         }
2735         nfs4_file_get_access(fp, oflag);
2736
2737         return nfs_ok;
2738 }
2739
2740 static inline __be32
2741 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2742                 struct nfsd4_open *open)
2743 {
2744         struct iattr iattr = {
2745                 .ia_valid = ATTR_SIZE,
2746                 .ia_size = 0,
2747         };
2748         if (!open->op_truncate)
2749                 return 0;
2750         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2751                 return nfserr_inval;
2752         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2753 }
2754
2755 static __be32
2756 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)
2757 {
2758         u32 op_share_access = open->op_share_access;
2759         bool new_access;
2760         __be32 status;
2761
2762         new_access = !test_access(op_share_access, stp);
2763         if (new_access) {
2764                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2765                 if (status)
2766                         return status;
2767         }
2768         status = nfsd4_truncate(rqstp, cur_fh, open);
2769         if (status) {
2770                 if (new_access) {
2771                         int oflag = nfs4_access_to_omode(op_share_access);
2772                         nfs4_file_put_access(fp, oflag);
2773                 }
2774                 return status;
2775         }
2776         /* remember the open */
2777         set_access(op_share_access, stp);
2778         set_deny(open->op_share_deny, stp);
2779
2780         return nfs_ok;
2781 }
2782
2783
2784 static void
2785 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
2786 {
2787         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2788 }
2789
2790 /* Should we give out recallable state?: */
2791 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2792 {
2793         if (clp->cl_cb_state == NFSD4_CB_UP)
2794                 return true;
2795         /*
2796          * In the sessions case, since we don't have to establish a
2797          * separate connection for callbacks, we assume it's OK
2798          * until we hear otherwise:
2799          */
2800         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2801 }
2802
2803 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2804 {
2805         struct file_lock *fl;
2806
2807         fl = locks_alloc_lock();
2808         if (!fl)
2809                 return NULL;
2810         locks_init_lock(fl);
2811         fl->fl_lmops = &nfsd_lease_mng_ops;
2812         fl->fl_flags = FL_LEASE;
2813         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2814         fl->fl_end = OFFSET_MAX;
2815         fl->fl_owner = (fl_owner_t)(dp->dl_file);
2816         fl->fl_pid = current->tgid;
2817         return fl;
2818 }
2819
2820 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2821 {
2822         struct nfs4_file *fp = dp->dl_file;
2823         struct file_lock *fl;
2824         int status;
2825
2826         fl = nfs4_alloc_init_lease(dp, flag);
2827         if (!fl)
2828                 return -ENOMEM;
2829         fl->fl_file = find_readable_file(fp);
2830         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2831         status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2832         if (status) {
2833                 list_del_init(&dp->dl_perclnt);
2834                 locks_free_lock(fl);
2835                 return -ENOMEM;
2836         }
2837         fp->fi_lease = fl;
2838         fp->fi_deleg_file = fl->fl_file;
2839         get_file(fp->fi_deleg_file);
2840         atomic_set(&fp->fi_delegees, 1);
2841         list_add(&dp->dl_perfile, &fp->fi_delegations);
2842         return 0;
2843 }
2844
2845 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2846 {
2847         struct nfs4_file *fp = dp->dl_file;
2848
2849         if (!fp->fi_lease)
2850                 return nfs4_setlease(dp, flag);
2851         spin_lock(&recall_lock);
2852         if (fp->fi_had_conflict) {
2853                 spin_unlock(&recall_lock);
2854                 return -EAGAIN;
2855         }
2856         atomic_inc(&fp->fi_delegees);
2857         list_add(&dp->dl_perfile, &fp->fi_delegations);
2858         spin_unlock(&recall_lock);
2859         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2860         return 0;
2861 }
2862
2863 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
2864 {
2865         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2866         if (status == -EAGAIN)
2867                 open->op_why_no_deleg = WND4_CONTENTION;
2868         else {
2869                 open->op_why_no_deleg = WND4_RESOURCE;
2870                 switch (open->op_deleg_want) {
2871                 case NFS4_SHARE_WANT_READ_DELEG:
2872                 case NFS4_SHARE_WANT_WRITE_DELEG:
2873                 case NFS4_SHARE_WANT_ANY_DELEG:
2874                         break;
2875                 case NFS4_SHARE_WANT_CANCEL:
2876                         open->op_why_no_deleg = WND4_CANCELLED;
2877                         break;
2878                 case NFS4_SHARE_WANT_NO_DELEG:
2879                         BUG();  /* not supposed to get here */
2880                 }
2881         }
2882 }
2883
2884 /*
2885  * Attempt to hand out a delegation.
2886  */
2887 static void
2888 nfs4_open_delegation(struct net *net, struct svc_fh *fh,
2889                      struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
2890 {
2891         struct nfs4_delegation *dp;
2892         struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
2893         int cb_up;
2894         int status = 0, flag = 0;
2895
2896         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
2897         flag = NFS4_OPEN_DELEGATE_NONE;
2898         open->op_recall = 0;
2899         switch (open->op_claim_type) {
2900                 case NFS4_OPEN_CLAIM_PREVIOUS:
2901                         if (!cb_up)
2902                                 open->op_recall = 1;
2903                         flag = open->op_delegate_type;
2904                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2905                                 goto out;
2906                         break;
2907                 case NFS4_OPEN_CLAIM_NULL:
2908                         /* Let's not give out any delegations till everyone's
2909                          * had the chance to reclaim theirs.... */
2910                         if (locks_in_grace(net))
2911                                 goto out;
2912                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
2913                                 goto out;
2914                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2915                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2916                         else
2917                                 flag = NFS4_OPEN_DELEGATE_READ;
2918                         break;
2919                 default:
2920                         goto out;
2921         }
2922
2923         dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh, flag);
2924         if (dp == NULL)
2925                 goto out_no_deleg;
2926         status = nfs4_set_delegation(dp, flag);
2927         if (status)
2928                 goto out_free;
2929
2930         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
2931
2932         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2933                 STATEID_VAL(&dp->dl_stid.sc_stateid));
2934 out:
2935         open->op_delegate_type = flag;
2936         if (flag == NFS4_OPEN_DELEGATE_NONE) {
2937                 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
2938                     open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2939                         dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2940
2941                 /* 4.1 client asking for a delegation? */
2942                 if (open->op_deleg_want)
2943                         nfsd4_open_deleg_none_ext(open, status);
2944         }
2945         return;
2946 out_free:
2947         nfs4_put_delegation(dp);
2948 out_no_deleg:
2949         flag = NFS4_OPEN_DELEGATE_NONE;
2950         goto out;
2951 }
2952
2953 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
2954                                         struct nfs4_delegation *dp)
2955 {
2956         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
2957             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
2958                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2959                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
2960         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
2961                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
2962                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2963                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
2964         }
2965         /* Otherwise the client must be confused wanting a delegation
2966          * it already has, therefore we don't return
2967          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
2968          */
2969 }
2970
2971 /*
2972  * called with nfs4_lock_state() held.
2973  */
2974 __be32
2975 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2976 {
2977         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2978         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
2979         struct nfs4_file *fp = NULL;
2980         struct inode *ino = current_fh->fh_dentry->d_inode;
2981         struct nfs4_ol_stateid *stp = NULL;
2982         struct nfs4_delegation *dp = NULL;
2983         __be32 status;
2984
2985         /*
2986          * Lookup file; if found, lookup stateid and check open request,
2987          * and check for delegations in the process of being recalled.
2988          * If not found, create the nfs4_file struct
2989          */
2990         fp = find_file(ino);
2991         if (fp) {
2992                 if ((status = nfs4_check_open(fp, open, &stp)))
2993                         goto out;
2994                 status = nfs4_check_deleg(cl, fp, open, &dp);
2995                 if (status)
2996                         goto out;
2997         } else {
2998                 status = nfserr_bad_stateid;
2999                 if (nfsd4_is_deleg_cur(open))
3000                         goto out;
3001                 status = nfserr_jukebox;
3002                 fp = open->op_file;
3003                 open->op_file = NULL;
3004                 nfsd4_init_file(fp, ino);
3005         }
3006
3007         /*
3008          * OPEN the file, or upgrade an existing OPEN.
3009          * If truncate fails, the OPEN fails.
3010          */
3011         if (stp) {
3012                 /* Stateid was found, this is an OPEN upgrade */
3013                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
3014                 if (status)
3015                         goto out;
3016         } else {
3017                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, open);
3018                 if (status)
3019                         goto out;
3020                 status = nfsd4_truncate(rqstp, current_fh, open);
3021                 if (status)
3022                         goto out;
3023                 stp = open->op_stp;
3024                 open->op_stp = NULL;
3025                 init_open_stateid(stp, fp, open);
3026         }
3027         update_stateid(&stp->st_stid.sc_stateid);
3028         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3029
3030         if (nfsd4_has_session(&resp->cstate)) {
3031                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3032
3033                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
3034                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3035                         open->op_why_no_deleg = WND4_NOT_WANTED;
3036                         goto nodeleg;
3037                 }
3038         }
3039
3040         /*
3041         * Attempt to hand out a delegation. No error return, because the
3042         * OPEN succeeds even if we fail.
3043         */
3044         nfs4_open_delegation(SVC_NET(rqstp), current_fh, open, stp);
3045 nodeleg:
3046         status = nfs_ok;
3047
3048         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
3049                 STATEID_VAL(&stp->st_stid.sc_stateid));
3050 out:
3051         /* 4.1 client trying to upgrade/downgrade delegation? */
3052         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
3053             open->op_deleg_want)
3054                 nfsd4_deleg_xgrade_none_ext(open, dp);
3055
3056         if (fp)
3057                 put_nfs4_file(fp);
3058         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
3059                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
3060         /*
3061         * To finish the open response, we just need to set the rflags.
3062         */
3063         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
3064         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
3065             !nfsd4_has_session(&resp->cstate))
3066                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
3067
3068         return status;
3069 }
3070
3071 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
3072 {
3073         if (open->op_openowner) {
3074                 struct nfs4_openowner *oo = open->op_openowner;
3075
3076                 if (!list_empty(&oo->oo_owner.so_stateids))
3077                         list_del_init(&oo->oo_close_lru);
3078                 if (oo->oo_flags & NFS4_OO_NEW) {
3079                         if (status) {
3080                                 release_openowner(oo);
3081                                 open->op_openowner = NULL;
3082                         } else
3083                                 oo->oo_flags &= ~NFS4_OO_NEW;
3084                 }
3085         }
3086         if (open->op_file)
3087                 nfsd4_free_file(open->op_file);
3088         if (open->op_stp)
3089                 nfs4_free_stateid(open->op_stp);
3090 }
3091
3092 __be32
3093 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3094             clientid_t *clid)
3095 {
3096         struct nfs4_client *clp;
3097         __be32 status;
3098
3099         nfs4_lock_state();
3100         dprintk("process_renew(%08x/%08x): starting\n",
3101                         clid->cl_boot, clid->cl_id);
3102         status = nfserr_stale_clientid;
3103         if (STALE_CLIENTID(clid))
3104                 goto out;
3105         clp = find_confirmed_client(clid);
3106         status = nfserr_expired;
3107         if (clp == NULL) {
3108                 /* We assume the client took too long to RENEW. */
3109                 dprintk("nfsd4_renew: clientid not found!\n");
3110                 goto out;
3111         }
3112         status = nfserr_cb_path_down;
3113         if (!list_empty(&clp->cl_delegations)
3114                         && clp->cl_cb_state != NFSD4_CB_UP)
3115                 goto out;
3116         status = nfs_ok;
3117 out:
3118         nfs4_unlock_state();
3119         return status;
3120 }
3121
3122 static bool grace_ended;
3123
3124 static void
3125 nfsd4_end_grace(struct net *net)
3126 {
3127         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3128
3129         /* do nothing if grace period already ended */
3130         if (grace_ended)
3131                 return;
3132
3133         dprintk("NFSD: end of grace period\n");
3134         grace_ended = true;
3135         nfsd4_record_grace_done(net, boot_time);
3136         locks_end_grace(&nn->nfsd4_manager);
3137         /*
3138          * Now that every NFSv4 client has had the chance to recover and
3139          * to see the (possibly new, possibly shorter) lease time, we
3140          * can safely set the next grace time to the current lease time:
3141          */
3142         nfsd4_grace = nfsd4_lease;
3143 }
3144
3145 static time_t
3146 nfs4_laundromat(void)
3147 {
3148         struct nfs4_client *clp;
3149         struct nfs4_openowner *oo;
3150         struct nfs4_delegation *dp;
3151         struct list_head *pos, *next, reaplist;
3152         time_t cutoff = get_seconds() - nfsd4_lease;
3153         time_t t, clientid_val = nfsd4_lease;
3154         time_t u, test_val = nfsd4_lease;
3155
3156         nfs4_lock_state();
3157
3158         dprintk("NFSD: laundromat service - starting\n");
3159         nfsd4_end_grace(&init_net);
3160         INIT_LIST_HEAD(&reaplist);
3161         spin_lock(&client_lock);
3162         list_for_each_safe(pos, next, &client_lru) {
3163                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3164                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3165                         t = clp->cl_time - cutoff;
3166                         if (clientid_val > t)
3167                                 clientid_val = t;
3168                         break;
3169                 }
3170                 if (atomic_read(&clp->cl_refcount)) {
3171                         dprintk("NFSD: client in use (clientid %08x)\n",
3172                                 clp->cl_clientid.cl_id);
3173                         continue;
3174                 }
3175                 unhash_client_locked(clp);
3176                 list_add(&clp->cl_lru, &reaplist);
3177         }
3178         spin_unlock(&client_lock);
3179         list_for_each_safe(pos, next, &reaplist) {
3180                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3181                 dprintk("NFSD: purging unused client (clientid %08x)\n",
3182                         clp->cl_clientid.cl_id);
3183                 nfsd4_client_record_remove(clp);
3184                 expire_client(clp);
3185         }
3186         spin_lock(&recall_lock);
3187         list_for_each_safe(pos, next, &del_recall_lru) {
3188                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3189                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3190                         u = dp->dl_time - cutoff;
3191                         if (test_val > u)
3192                                 test_val = u;
3193                         break;
3194                 }
3195                 list_move(&dp->dl_recall_lru, &reaplist);
3196         }
3197         spin_unlock(&recall_lock);
3198         list_for_each_safe(pos, next, &reaplist) {
3199                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3200                 unhash_delegation(dp);
3201         }
3202         test_val = nfsd4_lease;
3203         list_for_each_safe(pos, next, &close_lru) {
3204                 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3205                 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3206                         u = oo->oo_time - cutoff;
3207                         if (test_val > u)
3208                                 test_val = u;
3209                         break;
3210                 }
3211                 release_openowner(oo);
3212         }
3213         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3214                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3215         nfs4_unlock_state();
3216         return clientid_val;
3217 }
3218
3219 static struct workqueue_struct *laundry_wq;
3220 static void laundromat_main(struct work_struct *);
3221 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
3222
3223 static void
3224 laundromat_main(struct work_struct *not_used)
3225 {
3226         time_t t;
3227
3228         t = nfs4_laundromat();
3229         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3230         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
3231 }
3232
3233 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3234 {
3235         if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3236                 return nfserr_bad_stateid;
3237         return nfs_ok;
3238 }
3239
3240 static int
3241 STALE_STATEID(stateid_t *stateid)
3242 {
3243         if (stateid->si_opaque.so_clid.cl_boot == boot_time)
3244                 return 0;
3245         dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
3246                 STATEID_VAL(stateid));
3247         return 1;
3248 }
3249
3250 static inline int
3251 access_permit_read(struct nfs4_ol_stateid *stp)
3252 {
3253         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
3254                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
3255                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
3256 }
3257
3258 static inline int
3259 access_permit_write(struct nfs4_ol_stateid *stp)
3260 {
3261         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
3262                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
3263 }
3264
3265 static
3266 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3267 {
3268         __be32 status = nfserr_openmode;
3269
3270         /* For lock stateid's, we test the parent open, not the lock: */
3271         if (stp->st_openstp)
3272                 stp = stp->st_openstp;
3273         if ((flags & WR_STATE) && !access_permit_write(stp))
3274                 goto out;
3275         if ((flags & RD_STATE) && !access_permit_read(stp))
3276                 goto out;
3277         status = nfs_ok;
3278 out:
3279         return status;
3280 }
3281
3282 static inline __be32
3283 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
3284 {
3285         if (ONE_STATEID(stateid) && (flags & RD_STATE))
3286                 return nfs_ok;
3287         else if (locks_in_grace(net)) {
3288                 /* Answer in remaining cases depends on existence of
3289                  * conflicting state; so we must wait out the grace period. */
3290                 return nfserr_grace;
3291         } else if (flags & WR_STATE)
3292                 return nfs4_share_conflict(current_fh,
3293                                 NFS4_SHARE_DENY_WRITE);
3294         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3295                 return nfs4_share_conflict(current_fh,
3296                                 NFS4_SHARE_DENY_READ);
3297 }
3298
3299 /*
3300  * Allow READ/WRITE during grace period on recovered state only for files
3301  * that are not able to provide mandatory locking.
3302  */
3303 static inline int
3304 grace_disallows_io(struct net *net, struct inode *inode)
3305 {
3306         return locks_in_grace(net) && mandatory_lock(inode);
3307 }
3308
3309 /* Returns true iff a is later than b: */
3310 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3311 {
3312         return (s32)a->si_generation - (s32)b->si_generation > 0;
3313 }
3314
3315 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3316 {
3317         /*
3318          * When sessions are used the stateid generation number is ignored
3319          * when it is zero.
3320          */
3321         if (has_session && in->si_generation == 0)
3322                 return nfs_ok;
3323
3324         if (in->si_generation == ref->si_generation)
3325                 return nfs_ok;
3326
3327         /* If the client sends us a stateid from the future, it's buggy: */
3328         if (stateid_generation_after(in, ref))
3329                 return nfserr_bad_stateid;
3330         /*
3331          * However, we could see a stateid from the past, even from a
3332          * non-buggy client.  For example, if the client sends a lock
3333          * while some IO is outstanding, the lock may bump si_generation
3334          * while the IO is still in flight.  The client could avoid that
3335          * situation by waiting for responses on all the IO requests,
3336          * but better performance may result in retrying IO that
3337          * receives an old_stateid error if requests are rarely
3338          * reordered in flight:
3339          */
3340         return nfserr_old_stateid;
3341 }
3342
3343 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
3344 {
3345         struct nfs4_stid *s;
3346         struct nfs4_ol_stateid *ols;
3347         __be32 status;
3348
3349         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3350                 return nfserr_bad_stateid;
3351         /* Client debugging aid. */
3352         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
3353                 char addr_str[INET6_ADDRSTRLEN];
3354                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
3355                                  sizeof(addr_str));
3356                 pr_warn_ratelimited("NFSD: client %s testing state ID "
3357                                         "with incorrect client ID\n", addr_str);
3358                 return nfserr_bad_stateid;
3359         }
3360         s = find_stateid(cl, stateid);
3361         if (!s)
3362                 return nfserr_bad_stateid;
3363         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3364         if (status)
3365                 return status;
3366         if (!(s->sc_type & (NFS4_OPEN_STID | NFS4_LOCK_STID)))
3367                 return nfs_ok;
3368         ols = openlockstateid(s);
3369         if (ols->st_stateowner->so_is_open_owner
3370             && !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3371                 return nfserr_bad_stateid;
3372         return nfs_ok;
3373 }
3374
3375 static __be32 nfsd4_lookup_stateid(stateid_t *stateid, unsigned char typemask, struct nfs4_stid **s)
3376 {
3377         struct nfs4_client *cl;
3378
3379         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3380                 return nfserr_bad_stateid;
3381         if (STALE_STATEID(stateid))
3382                 return nfserr_stale_stateid;
3383         cl = find_confirmed_client(&stateid->si_opaque.so_clid);
3384         if (!cl)
3385                 return nfserr_expired;
3386         *s = find_stateid_by_type(cl, stateid, typemask);
3387         if (!*s)
3388                 return nfserr_bad_stateid;
3389         return nfs_ok;
3390
3391 }
3392
3393 /*
3394 * Checks for stateid operations
3395 */
3396 __be32
3397 nfs4_preprocess_stateid_op(struct net *net, struct nfsd4_compound_state *cstate,
3398                            stateid_t *stateid, int flags, struct file **filpp)
3399 {
3400         struct nfs4_stid *s;
3401         struct nfs4_ol_stateid *stp = NULL;
3402         struct nfs4_delegation *dp = NULL;
3403         struct svc_fh *current_fh = &cstate->current_fh;
3404         struct inode *ino = current_fh->fh_dentry->d_inode;
3405         __be32 status;
3406
3407         if (filpp)
3408                 *filpp = NULL;
3409
3410         if (grace_disallows_io(net, ino))
3411                 return nfserr_grace;
3412
3413         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3414                 return check_special_stateids(net, current_fh, stateid, flags);
3415
3416         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID, &s);
3417         if (status)
3418                 return status;
3419         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
3420         if (status)
3421                 goto out;
3422         switch (s->sc_type) {
3423         case NFS4_DELEG_STID:
3424                 dp = delegstateid(s);
3425                 status = nfs4_check_delegmode(dp, flags);
3426                 if (status)
3427                         goto out;
3428                 if (filpp) {
3429                         *filpp = dp->dl_file->fi_deleg_file;
3430                         BUG_ON(!*filpp);
3431                 }
3432                 break;
3433         case NFS4_OPEN_STID:
3434         case NFS4_LOCK_STID:
3435                 stp = openlockstateid(s);
3436                 status = nfs4_check_fh(current_fh, stp);
3437                 if (status)
3438                         goto out;
3439                 if (stp->st_stateowner->so_is_open_owner
3440                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3441                         goto out;
3442                 status = nfs4_check_openmode(stp, flags);
3443                 if (status)
3444                         goto out;
3445                 if (filpp) {
3446                         if (flags & RD_STATE)
3447                                 *filpp = find_readable_file(stp->st_file);
3448                         else
3449                                 *filpp = find_writeable_file(stp->st_file);
3450                 }
3451                 break;
3452         default:
3453                 return nfserr_bad_stateid;
3454         }
3455         status = nfs_ok;
3456 out:
3457         return status;
3458 }
3459
3460 static __be32
3461 nfsd4_free_lock_stateid(struct nfs4_ol_stateid *stp)
3462 {
3463         if (check_for_locks(stp->st_file, lockowner(stp->st_stateowner)))
3464                 return nfserr_locks_held;
3465         release_lock_stateid(stp);
3466         return nfs_ok;
3467 }
3468
3469 /*
3470  * Test if the stateid is valid
3471  */
3472 __be32
3473 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3474                    struct nfsd4_test_stateid *test_stateid)
3475 {
3476         struct nfsd4_test_stateid_id *stateid;
3477         struct nfs4_client *cl = cstate->session->se_client;
3478
3479         nfs4_lock_state();
3480         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
3481                 stateid->ts_id_status =
3482                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
3483         nfs4_unlock_state();
3484
3485         return nfs_ok;
3486 }
3487
3488 __be32
3489 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3490                    struct nfsd4_free_stateid *free_stateid)
3491 {
3492         stateid_t *stateid = &free_stateid->fr_stateid;
3493         struct nfs4_stid *s;
3494         struct nfs4_client *cl = cstate->session->se_client;
3495         __be32 ret = nfserr_bad_stateid;
3496
3497         nfs4_lock_state();
3498         s = find_stateid(cl, stateid);
3499         if (!s)
3500                 goto out;
3501         switch (s->sc_type) {
3502         case NFS4_DELEG_STID:
3503                 ret = nfserr_locks_held;
3504                 goto out;
3505         case NFS4_OPEN_STID:
3506         case NFS4_LOCK_STID:
3507                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
3508                 if (ret)
3509                         goto out;
3510                 if (s->sc_type == NFS4_LOCK_STID)
3511                         ret = nfsd4_free_lock_stateid(openlockstateid(s));
3512                 else
3513                         ret = nfserr_locks_held;
3514                 break;
3515         default:
3516                 ret = nfserr_bad_stateid;
3517         }
3518 out:
3519         nfs4_unlock_state();
3520         return ret;
3521 }
3522
3523 static inline int
3524 setlkflg (int type)
3525 {
3526         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3527                 RD_STATE : WR_STATE;
3528 }
3529
3530 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
3531 {
3532         struct svc_fh *current_fh = &cstate->current_fh;
3533         struct nfs4_stateowner *sop = stp->st_stateowner;
3534         __be32 status;
3535
3536         status = nfsd4_check_seqid(cstate, sop, seqid);
3537         if (status)
3538                 return status;
3539         if (stp->st_stid.sc_type == NFS4_CLOSED_STID)
3540                 /*
3541                  * "Closed" stateid's exist *only* to return
3542                  * nfserr_replay_me from the previous step.
3543                  */
3544                 return nfserr_bad_stateid;
3545         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
3546         if (status)
3547                 return status;
3548         return nfs4_check_fh(current_fh, stp);
3549 }
3550
3551 /*
3552  * Checks for sequence id mutating operations.
3553  */
3554 static __be32
3555 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3556                          stateid_t *stateid, char typemask,
3557                          struct nfs4_ol_stateid **stpp)
3558 {
3559         __be32 status;
3560         struct nfs4_stid *s;
3561
3562         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3563                 seqid, STATEID_VAL(stateid));
3564
3565         *stpp = NULL;
3566         status = nfsd4_lookup_stateid(stateid, typemask, &s);
3567         if (status)
3568                 return status;
3569         *stpp = openlockstateid(s);
3570         cstate->replay_owner = (*stpp)->st_stateowner;
3571
3572         return nfs4_seqid_op_checks(cstate, stateid, seqid, *stpp);
3573 }
3574
3575 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid, stateid_t *stateid, struct nfs4_ol_stateid **stpp)
3576 {
3577         __be32 status;
3578         struct nfs4_openowner *oo;
3579
3580         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
3581                                                 NFS4_OPEN_STID, stpp);
3582         if (status)
3583                 return status;
3584         oo = openowner((*stpp)->st_stateowner);
3585         if (!(oo->oo_flags & NFS4_OO_CONFIRMED))
3586                 return nfserr_bad_stateid;
3587         return nfs_ok;
3588 }
3589
3590 __be32
3591 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3592                    struct nfsd4_open_confirm *oc)
3593 {
3594         __be32 status;
3595         struct nfs4_openowner *oo;
3596         struct nfs4_ol_stateid *stp;
3597
3598         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3599                         (int)cstate->current_fh.fh_dentry->d_name.len,
3600                         cstate->current_fh.fh_dentry->d_name.name);
3601
3602         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3603         if (status)
3604                 return status;
3605
3606         nfs4_lock_state();
3607
3608         status = nfs4_preprocess_seqid_op(cstate,
3609                                         oc->oc_seqid, &oc->oc_req_stateid,
3610                                         NFS4_OPEN_STID, &stp);
3611         if (status)
3612                 goto out;
3613         oo = openowner(stp->st_stateowner);
3614         status = nfserr_bad_stateid;
3615         if (oo->oo_flags & NFS4_OO_CONFIRMED)
3616                 goto out;
3617         oo->oo_flags |= NFS4_OO_CONFIRMED;
3618         update_stateid(&stp->st_stid.sc_stateid);
3619         memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3620         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3621                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
3622
3623         nfsd4_client_record_create(oo->oo_owner.so_client);
3624         status = nfs_ok;
3625 out:
3626         if (!cstate->replay_owner)
3627                 nfs4_unlock_state();
3628         return status;
3629 }
3630
3631 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
3632 {
3633         if (!test_access(access, stp))
3634                 return;
3635         nfs4_file_put_access(stp->st_file, nfs4_access_to_omode(access));
3636         clear_access(access, stp);
3637 }
3638
3639 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
3640 {
3641         switch (to_access) {
3642         case NFS4_SHARE_ACCESS_READ:
3643                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
3644                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3645                 break;
3646         case NFS4_SHARE_ACCESS_WRITE:
3647                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
3648                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3649                 break;
3650         case NFS4_SHARE_ACCESS_BOTH:
3651                 break;
3652         default:
3653                 BUG();
3654         }
3655 }
3656
3657 static void
3658 reset_union_bmap_deny(unsigned long deny, struct nfs4_ol_stateid *stp)
3659 {
3660         int i;
3661         for (i = 0; i < 4; i++) {
3662                 if ((i & deny) != i)
3663                         clear_deny(i, stp);
3664         }
3665 }
3666
3667 __be32
3668 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3669                      struct nfsd4_compound_state *cstate,
3670                      struct nfsd4_open_downgrade *od)
3671 {
3672         __be32 status;
3673         struct nfs4_ol_stateid *stp;
3674
3675         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3676                         (int)cstate->current_fh.fh_dentry->d_name.len,
3677                         cstate->current_fh.fh_dentry->d_name.name);
3678
3679         /* We don't yet support WANT bits: */
3680         if (od->od_deleg_want)
3681                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
3682                         od->od_deleg_want);
3683
3684         nfs4_lock_state();
3685         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
3686                                         &od->od_stateid, &stp);
3687         if (status)
3688                 goto out;
3689         status = nfserr_inval;
3690         if (!test_access(od->od_share_access, stp)) {
3691                 dprintk("NFSD: access not a subset current bitmap: 0x%lx, input access=%08x\n",
3692                         stp->st_access_bmap, od->od_share_access);
3693                 goto out;
3694         }
3695         if (!test_deny(od->od_share_deny, stp)) {
3696                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3697                         stp->st_deny_bmap, od->od_share_deny);
3698                 goto out;
3699         }
3700         nfs4_stateid_downgrade(stp, od->od_share_access);
3701
3702         reset_union_bmap_deny(od->od_share_deny, stp);
3703
3704         update_stateid(&stp->st_stid.sc_stateid);
3705         memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3706         status = nfs_ok;
3707 out:
3708         if (!cstate->replay_owner)
3709                 nfs4_unlock_state();
3710         return status;
3711 }
3712
3713 void nfsd4_purge_closed_stateid(struct nfs4_stateowner *so)
3714 {
3715         struct nfs4_openowner *oo;
3716         struct nfs4_ol_stateid *s;
3717
3718         if (!so->so_is_open_owner)
3719                 return;
3720         oo = openowner(so);
3721         s = oo->oo_last_closed_stid;
3722         if (!s)
3723                 return;
3724         if (!(oo->oo_flags & NFS4_OO_PURGE_CLOSE)) {
3725                 /* Release the last_closed_stid on the next seqid bump: */
3726                 oo->oo_flags |= NFS4_OO_PURGE_CLOSE;
3727                 return;
3728         }
3729         oo->oo_flags &= ~NFS4_OO_PURGE_CLOSE;
3730         release_last_closed_stateid(oo);
3731 }
3732
3733 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
3734 {
3735         unhash_open_stateid(s);
3736         s->st_stid.sc_type = NFS4_CLOSED_STID;
3737 }
3738
3739 /*
3740  * nfs4_unlock_state() called after encode
3741  */
3742 __be32
3743 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3744             struct nfsd4_close *close)
3745 {
3746         __be32 status;
3747         struct nfs4_openowner *oo;
3748         struct nfs4_ol_stateid *stp;
3749
3750         dprintk("NFSD: nfsd4_close on file %.*s\n",
3751                         (int)cstate->current_fh.fh_dentry->d_name.len,
3752                         cstate->current_fh.fh_dentry->d_name.name);
3753
3754         nfs4_lock_state();
3755         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
3756                                         &close->cl_stateid,
3757                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
3758                                         &stp);
3759         if (status)
3760                 goto out;
3761         oo = openowner(stp->st_stateowner);
3762         status = nfs_ok;
3763         update_stateid(&stp->st_stid.sc_stateid);
3764         memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3765
3766         nfsd4_close_open_stateid(stp);
3767         oo->oo_last_closed_stid = stp;
3768
3769         if (list_empty(&oo->oo_owner.so_stateids)) {
3770                 if (cstate->minorversion) {
3771                         release_openowner(oo);
3772                         cstate->replay_owner = NULL;
3773                 } else {
3774                         /*
3775                          * In the 4.0 case we need to keep the owners around a
3776                          * little while to handle CLOSE replay.
3777                          */
3778                         if (list_empty(&oo->oo_owner.so_stateids))
3779                                 move_to_close_lru(oo);
3780                 }
3781         }
3782 out:
3783         if (!cstate->replay_owner)
3784                 nfs4_unlock_state();
3785         return status;
3786 }
3787
3788 __be32
3789 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3790                   struct nfsd4_delegreturn *dr)
3791 {
3792         struct nfs4_delegation *dp;
3793         stateid_t *stateid = &dr->dr_stateid;
3794         struct nfs4_stid *s;
3795         struct inode *inode;
3796         __be32 status;
3797
3798         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3799                 return status;
3800         inode = cstate->current_fh.fh_dentry->d_inode;
3801
3802         nfs4_lock_state();
3803         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID, &s);
3804         if (status)
3805                 goto out;
3806         dp = delegstateid(s);
3807         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
3808         if (status)
3809                 goto out;
3810
3811         unhash_delegation(dp);
3812 out:
3813         nfs4_unlock_state();
3814
3815         return status;
3816 }
3817
3818
3819 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3820
3821 #define LOCKOWNER_INO_HASH_BITS 8
3822 #define LOCKOWNER_INO_HASH_SIZE (1 << LOCKOWNER_INO_HASH_BITS)
3823 #define LOCKOWNER_INO_HASH_MASK (LOCKOWNER_INO_HASH_SIZE - 1)
3824
3825 static inline u64
3826 end_offset(u64 start, u64 len)
3827 {
3828         u64 end;
3829
3830         end = start + len;
3831         return end >= start ? end: NFS4_MAX_UINT64;
3832 }
3833
3834 /* last octet in a range */
3835 static inline u64
3836 last_byte_offset(u64 start, u64 len)
3837 {
3838         u64 end;
3839
3840         BUG_ON(!len);
3841         end = start + len;
3842         return end > start ? end - 1: NFS4_MAX_UINT64;
3843 }
3844
3845 static unsigned int lockowner_ino_hashval(struct inode *inode, u32 cl_id, struct xdr_netobj *ownername)
3846 {
3847         return (file_hashval(inode) + cl_id
3848                         + opaque_hashval(ownername->data, ownername->len))
3849                 & LOCKOWNER_INO_HASH_MASK;
3850 }
3851
3852 static struct list_head lockowner_ino_hashtbl[LOCKOWNER_INO_HASH_SIZE];
3853
3854 /*
3855  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3856  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3857  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3858  * locking, this prevents us from being completely protocol-compliant.  The
3859  * real solution to this problem is to start using unsigned file offsets in
3860  * the VFS, but this is a very deep change!
3861  */
3862 static inline void
3863 nfs4_transform_lock_offset(struct file_lock *lock)
3864 {
3865         if (lock->fl_start < 0)
3866                 lock->fl_start = OFFSET_MAX;
3867         if (lock->fl_end < 0)
3868                 lock->fl_end = OFFSET_MAX;
3869 }
3870
3871 /* Hack!: For now, we're defining this just so we can use a pointer to it
3872  * as a unique cookie to identify our (NFSv4's) posix locks. */
3873 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
3874 };
3875
3876 static inline void
3877 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3878 {
3879         struct nfs4_lockowner *lo;
3880
3881         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3882                 lo = (struct nfs4_lockowner *) fl->fl_owner;
3883                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
3884                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
3885                 if (!deny->ld_owner.data)
3886                         /* We just don't care that much */
3887                         goto nevermind;
3888                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
3889                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
3890         } else {
3891 nevermind:
3892                 deny->ld_owner.len = 0;
3893                 deny->ld_owner.data = NULL;
3894                 deny->ld_clientid.cl_boot = 0;
3895                 deny->ld_clientid.cl_id = 0;
3896         }
3897         deny->ld_start = fl->fl_start;
3898         deny->ld_length = NFS4_MAX_UINT64;
3899         if (fl->fl_end != NFS4_MAX_UINT64)
3900                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3901         deny->ld_type = NFS4_READ_LT;
3902         if (fl->fl_type != F_RDLCK)
3903                 deny->ld_type = NFS4_WRITE_LT;
3904 }
3905
3906 static bool same_lockowner_ino(struct nfs4_lockowner *lo, struct inode *inode, clientid_t *clid, struct xdr_netobj *owner)
3907 {
3908         struct nfs4_ol_stateid *lst;
3909
3910         if (!same_owner_str(&lo->lo_owner, owner, clid))
3911                 return false;
3912         lst = list_first_entry(&lo->lo_owner.so_stateids,
3913                                struct nfs4_ol_stateid, st_perstateowner);
3914         return lst->st_file->fi_inode == inode;
3915 }
3916
3917 static struct nfs4_lockowner *
3918 find_lockowner_str(struct inode *inode, clientid_t *clid,
3919                 struct xdr_netobj *owner)
3920 {
3921         unsigned int hashval = lockowner_ino_hashval(inode, clid->cl_id, owner);
3922         struct nfs4_lockowner *lo;
3923
3924         list_for_each_entry(lo, &lockowner_ino_hashtbl[hashval], lo_owner_ino_hash) {
3925                 if (same_lockowner_ino(lo, inode, clid, owner))
3926                         return lo;
3927         }
3928         return NULL;
3929 }
3930
3931 static void hash_lockowner(struct nfs4_lockowner *lo, unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp)
3932 {
3933         struct inode *inode = open_stp->st_file->fi_inode;
3934         unsigned int inohash = lockowner_ino_hashval(inode,
3935                         clp->cl_clientid.cl_id, &lo->lo_owner.so_owner);
3936
3937         list_add(&lo->lo_owner.so_strhash, &ownerstr_hashtbl[strhashval]);
3938         list_add(&lo->lo_owner_ino_hash, &lockowner_ino_hashtbl[inohash]);
3939         list_add(&lo->lo_perstateid, &open_stp->st_lockowners);
3940 }
3941
3942 /*
3943  * Alloc a lock owner structure.
3944  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3945  * occurred.
3946  *
3947  * strhashval = ownerstr_hashval
3948  */
3949
3950 static struct nfs4_lockowner *
3951 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp, struct nfsd4_lock *lock) {
3952         struct nfs4_lockowner *lo;
3953
3954         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
3955         if (!lo)
3956                 return NULL;
3957         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
3958         lo->lo_owner.so_is_open_owner = 0;
3959         /* It is the openowner seqid that will be incremented in encode in the
3960          * case of new lockowners; so increment the lock seqid manually: */
3961         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid + 1;
3962         hash_lockowner(lo, strhashval, clp, open_stp);
3963         return lo;
3964 }
3965
3966 static struct nfs4_ol_stateid *
3967 alloc_init_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp, struct nfs4_ol_stateid *open_stp)
3968 {
3969         struct nfs4_ol_stateid *stp;
3970         struct nfs4_client *clp = lo->lo_owner.so_client;
3971
3972         stp = nfs4_alloc_stateid(clp);
3973         if (stp == NULL)
3974                 return NULL;
3975         init_stid(&stp->st_stid, clp, NFS4_LOCK_STID);
3976         list_add(&stp->st_perfile, &fp->fi_stateids);
3977         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
3978         stp->st_stateowner = &lo->lo_owner;
3979         get_nfs4_file(fp);
3980         stp->st_file = fp;
3981         stp->st_access_bmap = 0;
3982         stp->st_deny_bmap = open_stp->st_deny_bmap;
3983         stp->st_openstp = open_stp;
3984         return stp;
3985 }
3986
3987 static int
3988 check_lock_length(u64 offset, u64 length)
3989 {
3990         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
3991              LOFF_OVERFLOW(offset, length)));
3992 }
3993
3994 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
3995 {
3996         struct nfs4_file *fp = lock_stp->st_file;
3997         int oflag = nfs4_access_to_omode(access);
3998
3999         if (test_access(access, lock_stp))
4000                 return;
4001         nfs4_file_get_access(fp, oflag);
4002         set_access(access, lock_stp);
4003 }
4004
4005 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)
4006 {
4007         struct nfs4_file *fi = ost->st_file;
4008         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
4009         struct nfs4_client *cl = oo->oo_owner.so_client;
4010         struct nfs4_lockowner *lo;
4011         unsigned int strhashval;
4012
4013         lo = find_lockowner_str(fi->fi_inode, &cl->cl_clientid, &lock->v.new.owner);
4014         if (lo) {
4015                 if (!cstate->minorversion)
4016                         return nfserr_bad_seqid;
4017                 /* XXX: a lockowner always has exactly one stateid: */
4018                 *lst = list_first_entry(&lo->lo_owner.so_stateids,
4019                                 struct nfs4_ol_stateid, st_perstateowner);
4020                 return nfs_ok;
4021         }
4022         strhashval = ownerstr_hashval(cl->cl_clientid.cl_id,
4023                         &lock->v.new.owner);
4024         lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
4025         if (lo == NULL)
4026                 return nfserr_jukebox;
4027         *lst = alloc_init_lock_stateid(lo, fi, ost);
4028         if (*lst == NULL) {
4029                 release_lockowner(lo);
4030                 return nfserr_jukebox;
4031         }
4032         *new = true;
4033         return nfs_ok;
4034 }
4035
4036 /*
4037  *  LOCK operation
4038  */
4039 __be32
4040 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4041            struct nfsd4_lock *lock)
4042 {
4043         struct nfs4_openowner *open_sop = NULL;
4044         struct nfs4_lockowner *lock_sop = NULL;
4045         struct nfs4_ol_stateid *lock_stp;
4046         struct file *filp = NULL;
4047         struct file_lock file_lock;
4048         struct file_lock conflock;
4049         __be32 status = 0;
4050         bool new_state = false;
4051         int lkflg;
4052         int err;
4053
4054         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
4055                 (long long) lock->lk_offset,
4056                 (long long) lock->lk_length);
4057
4058         if (check_lock_length(lock->lk_offset, lock->lk_length))
4059                  return nfserr_inval;
4060
4061         if ((status = fh_verify(rqstp, &cstate->current_fh,
4062                                 S_IFREG, NFSD_MAY_LOCK))) {
4063                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
4064                 return status;
4065         }
4066
4067         nfs4_lock_state();
4068
4069         if (lock->lk_is_new) {
4070                 struct nfs4_ol_stateid *open_stp = NULL;
4071
4072                 if (nfsd4_has_session(cstate))
4073                         /* See rfc 5661 18.10.3: given clientid is ignored: */
4074                         memcpy(&lock->v.new.clientid,
4075                                 &cstate->session->se_client->cl_clientid,
4076                                 sizeof(clientid_t));
4077
4078                 status = nfserr_stale_clientid;
4079                 if (STALE_CLIENTID(&lock->lk_new_clientid))
4080                         goto out;
4081
4082                 /* validate and update open stateid and open seqid */
4083                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
4084                                         lock->lk_new_open_seqid,
4085                                         &lock->lk_new_open_stateid,
4086                                         &open_stp);
4087                 if (status)
4088                         goto out;
4089                 open_sop = openowner(open_stp->st_stateowner);
4090                 status = nfserr_bad_stateid;
4091                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
4092                                                 &lock->v.new.clientid))
4093                         goto out;
4094                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
4095                                                         &lock_stp, &new_state);
4096         } else
4097                 status = nfs4_preprocess_seqid_op(cstate,
4098                                        lock->lk_old_lock_seqid,
4099                                        &lock->lk_old_lock_stateid,
4100                                        NFS4_LOCK_STID, &lock_stp);
4101         if (status)
4102                 goto out;
4103         lock_sop = lockowner(lock_stp->st_stateowner);
4104
4105         lkflg = setlkflg(lock->lk_type);
4106         status = nfs4_check_openmode(lock_stp, lkflg);
4107         if (status)
4108                 goto out;
4109
4110         status = nfserr_grace;
4111         if (locks_in_grace(SVC_NET(rqstp)) && !lock->lk_reclaim)
4112                 goto out;
4113         status = nfserr_no_grace;
4114         if (!locks_in_grace(SVC_NET(rqstp)) && lock->lk_reclaim)
4115                 goto out;
4116
4117         locks_init_lock(&file_lock);
4118         switch (lock->lk_type) {
4119                 case NFS4_READ_LT:
4120                 case NFS4_READW_LT:
4121                         filp = find_readable_file(lock_stp->st_file);
4122                         if (filp)
4123                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
4124                         file_lock.fl_type = F_RDLCK;
4125                         break;
4126                 case NFS4_WRITE_LT:
4127                 case NFS4_WRITEW_LT:
4128                         filp = find_writeable_file(lock_stp->st_file);
4129                         if (filp)
4130                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
4131                         file_lock.fl_type = F_WRLCK;
4132                         break;
4133                 default:
4134                         status = nfserr_inval;
4135                 goto out;
4136         }
4137         if (!filp) {
4138                 status = nfserr_openmode;
4139                 goto out;
4140         }
4141         file_lock.fl_owner = (fl_owner_t)lock_sop;
4142         file_lock.fl_pid = current->tgid;
4143         file_lock.fl_file = filp;
4144         file_lock.fl_flags = FL_POSIX;
4145         file_lock.fl_lmops = &nfsd_posix_mng_ops;
4146
4147         file_lock.fl_start = lock->lk_offset;
4148         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
4149         nfs4_transform_lock_offset(&file_lock);
4150
4151         /*
4152         * Try to lock the file in the VFS.
4153         * Note: locks.c uses the BKL to protect the inode's lock list.
4154         */
4155
4156         err = vfs_lock_file(filp, F_SETLK, &file_lock, &conflock);
4157         switch (-err) {
4158         case 0: /* success! */
4159                 update_stateid(&lock_stp->st_stid.sc_stateid);
4160                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
4161                                 sizeof(stateid_t));
4162                 status = 0;
4163                 break;
4164         case (EAGAIN):          /* conflock holds conflicting lock */
4165                 status = nfserr_denied;
4166                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
4167                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
4168                 break;
4169         case (EDEADLK):
4170                 status = nfserr_deadlock;
4171                 break;
4172         default:
4173                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4174                 status = nfserrno(err);
4175                 break;
4176         }
4177 out:
4178         if (status && new_state)
4179                 release_lockowner(lock_sop);
4180         if (!cstate->replay_owner)
4181                 nfs4_unlock_state();
4182         return status;
4183 }
4184
4185 /*
4186  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4187  * so we do a temporary open here just to get an open file to pass to
4188  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
4189  * inode operation.)
4190  */
4191 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4192 {
4193         struct file *file;
4194         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4195         if (!err) {
4196                 err = nfserrno(vfs_test_lock(file, lock));
4197                 nfsd_close(file);
4198         }
4199         return err;
4200 }
4201
4202 /*
4203  * LOCKT operation
4204  */
4205 __be32
4206 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4207             struct nfsd4_lockt *lockt)
4208 {
4209         struct inode *inode;
4210         struct file_lock file_lock;
4211         struct nfs4_lockowner *lo;
4212         __be32 status;
4213
4214         if (locks_in_grace(SVC_NET(rqstp)))
4215                 return nfserr_grace;
4216
4217         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4218                  return nfserr_inval;
4219
4220         nfs4_lock_state();
4221
4222         status = nfserr_stale_clientid;
4223         if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
4224                 goto out;
4225
4226         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4227                 goto out;
4228
4229         inode = cstate->current_fh.fh_dentry->d_inode;
4230         locks_init_lock(&file_lock);
4231         switch (lockt->lt_type) {
4232                 case NFS4_READ_LT:
4233                 case NFS4_READW_LT:
4234                         file_lock.fl_type = F_RDLCK;
4235                 break;
4236                 case NFS4_WRITE_LT:
4237                 case NFS4_WRITEW_LT:
4238                         file_lock.fl_type = F_WRLCK;
4239                 break;
4240                 default:
4241                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4242                         status = nfserr_inval;
4243                 goto out;
4244         }
4245
4246         lo = find_lockowner_str(inode, &lockt->lt_clientid, &lockt->lt_owner);
4247         if (lo)
4248                 file_lock.fl_owner = (fl_owner_t)lo;
4249         file_lock.fl_pid = current->tgid;
4250         file_lock.fl_flags = FL_POSIX;
4251
4252         file_lock.fl_start = lockt->lt_offset;
4253         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4254
4255         nfs4_transform_lock_offset(&file_lock);
4256
4257         status = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
4258         if (status)
4259                 goto out;
4260
4261         if (file_lock.fl_type != F_UNLCK) {
4262                 status = nfserr_denied;
4263                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
4264         }
4265 out:
4266         nfs4_unlock_state();
4267         return status;
4268 }
4269
4270 __be32
4271 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4272             struct nfsd4_locku *locku)
4273 {
4274         struct nfs4_ol_stateid *stp;
4275         struct file *filp = NULL;
4276         struct file_lock file_lock;
4277         __be32 status;
4278         int err;
4279
4280         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4281                 (long long) locku->lu_offset,
4282                 (long long) locku->lu_length);
4283
4284         if (check_lock_length(locku->lu_offset, locku->lu_length))
4285                  return nfserr_inval;
4286
4287         nfs4_lock_state();
4288
4289         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
4290                                         &locku->lu_stateid, NFS4_LOCK_STID, &stp);
4291         if (status)
4292                 goto out;
4293         filp = find_any_file(stp->st_file);
4294         if (!filp) {
4295                 status = nfserr_lock_range;
4296                 goto out;
4297         }
4298         BUG_ON(!filp);
4299         locks_init_lock(&file_lock);
4300         file_lock.fl_type = F_UNLCK;
4301         file_lock.fl_owner = (fl_owner_t)lockowner(stp->st_stateowner);
4302         file_lock.fl_pid = current->tgid;
4303         file_lock.fl_file = filp;
4304         file_lock.fl_flags = FL_POSIX;
4305         file_lock.fl_lmops = &nfsd_posix_mng_ops;
4306         file_lock.fl_start = locku->lu_offset;
4307
4308         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
4309         nfs4_transform_lock_offset(&file_lock);
4310
4311         /*
4312         *  Try to unlock the file in the VFS.
4313         */
4314         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
4315         if (err) {
4316                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4317                 goto out_nfserr;
4318         }
4319         /*
4320         * OK, unlock succeeded; the only thing left to do is update the stateid.
4321         */
4322         update_stateid(&stp->st_stid.sc_stateid);
4323         memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4324
4325 out:
4326         if (!cstate->replay_owner)
4327                 nfs4_unlock_state();
4328         return status;
4329
4330 out_nfserr:
4331         status = nfserrno(err);
4332         goto out;
4333 }
4334
4335 /*
4336  * returns
4337  *      1: locks held by lockowner
4338  *      0: no locks held by lockowner
4339  */
4340 static int
4341 check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner)
4342 {
4343         struct file_lock **flpp;
4344         struct inode *inode = filp->fi_inode;
4345         int status = 0;
4346
4347         lock_flocks();
4348         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4349                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4350                         status = 1;
4351                         goto out;
4352                 }
4353         }
4354 out:
4355         unlock_flocks();
4356         return status;
4357 }
4358
4359 __be32
4360 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4361                         struct nfsd4_compound_state *cstate,
4362                         struct nfsd4_release_lockowner *rlockowner)
4363 {
4364         clientid_t *clid = &rlockowner->rl_clientid;
4365         struct nfs4_stateowner *sop;
4366         struct nfs4_lockowner *lo;
4367         struct nfs4_ol_stateid *stp;
4368         struct xdr_netobj *owner = &rlockowner->rl_owner;
4369         struct list_head matches;
4370         unsigned int hashval = ownerstr_hashval(clid->cl_id, owner);
4371         __be32 status;
4372
4373         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4374                 clid->cl_boot, clid->cl_id);
4375
4376         /* XXX check for lease expiration */
4377
4378         status = nfserr_stale_clientid;
4379         if (STALE_CLIENTID(clid))
4380                 return status;
4381
4382         nfs4_lock_state();
4383
4384         status = nfserr_locks_held;
4385         INIT_LIST_HEAD(&matches);
4386
4387         list_for_each_entry(sop, &ownerstr_hashtbl[hashval], so_strhash) {
4388                 if (sop->so_is_open_owner)
4389                         continue;
4390                 if (!same_owner_str(sop, owner, clid))
4391                         continue;
4392                 list_for_each_entry(stp, &sop->so_stateids,
4393                                 st_perstateowner) {
4394                         lo = lockowner(sop);
4395                         if (check_for_locks(stp->st_file, lo))
4396                                 goto out;
4397                         list_add(&lo->lo_list, &matches);
4398                 }
4399         }
4400         /* Clients probably won't expect us to return with some (but not all)
4401          * of the lockowner state released; so don't release any until all
4402          * have been checked. */
4403         status = nfs_ok;
4404         while (!list_empty(&matches)) {
4405                 lo = list_entry(matches.next, struct nfs4_lockowner,
4406                                                                 lo_list);
4407                 /* unhash_stateowner deletes so_perclient only
4408                  * for openowners. */
4409                 list_del(&lo->lo_list);
4410                 release_lockowner(lo);
4411         }
4412 out:
4413         nfs4_unlock_state();
4414         return status;
4415 }
4416
4417 static inline struct nfs4_client_reclaim *
4418 alloc_reclaim(void)
4419 {
4420         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4421 }
4422
4423 int
4424 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
4425 {
4426         unsigned int strhashval = clientstr_hashval(name);
4427         struct nfs4_client *clp;
4428
4429         clp = find_confirmed_client_by_str(name, strhashval);
4430         if (!clp)
4431                 return 0;
4432         return test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
4433 }
4434
4435 /*
4436  * failure => all reset bets are off, nfserr_no_grace...
4437  */
4438 int
4439 nfs4_client_to_reclaim(const char *name)
4440 {
4441         unsigned int strhashval;
4442         struct nfs4_client_reclaim *crp = NULL;
4443
4444         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4445         crp = alloc_reclaim();
4446         if (!crp)
4447                 return 0;
4448         strhashval = clientstr_hashval(name);
4449         INIT_LIST_HEAD(&crp->cr_strhash);
4450         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4451         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4452         reclaim_str_hashtbl_size++;
4453         return 1;
4454 }
4455
4456 void
4457 nfs4_release_reclaim(void)
4458 {
4459         struct nfs4_client_reclaim *crp = NULL;
4460         int i;
4461
4462         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4463                 while (!list_empty(&reclaim_str_hashtbl[i])) {
4464                         crp = list_entry(reclaim_str_hashtbl[i].next,
4465                                         struct nfs4_client_reclaim, cr_strhash);
4466                         list_del(&crp->cr_strhash);
4467                         kfree(crp);
4468                         reclaim_str_hashtbl_size--;
4469                 }
4470         }
4471         BUG_ON(reclaim_str_hashtbl_size);
4472 }
4473
4474 /*
4475  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4476 struct nfs4_client_reclaim *
4477 nfsd4_find_reclaim_client(struct nfs4_client *clp)
4478 {
4479         unsigned int strhashval;
4480         struct nfs4_client_reclaim *crp = NULL;
4481
4482         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4483                             clp->cl_name.len, clp->cl_name.data,
4484                             clp->cl_recdir);
4485
4486         /* find clp->cl_name in reclaim_str_hashtbl */
4487         strhashval = clientstr_hashval(clp->cl_recdir);
4488         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4489                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4490                         return crp;
4491                 }
4492         }
4493         return NULL;
4494 }
4495
4496 /*
4497 * Called from OPEN. Look for clientid in reclaim list.
4498 */
4499 __be32
4500 nfs4_check_open_reclaim(clientid_t *clid)
4501 {
4502         struct nfs4_client *clp;
4503
4504         /* find clientid in conf_id_hashtbl */
4505         clp = find_confirmed_client(clid);
4506         if (clp == NULL)
4507                 return nfserr_reclaim_bad;
4508
4509         return nfsd4_client_record_check(clp) ? nfserr_reclaim_bad : nfs_ok;
4510 }
4511
4512 #ifdef CONFIG_NFSD_FAULT_INJECTION
4513
4514 void nfsd_forget_clients(u64 num)
4515 {
4516         struct nfs4_client *clp, *next;
4517         int count = 0;
4518
4519         nfs4_lock_state();
4520         list_for_each_entry_safe(clp, next, &client_lru, cl_lru) {
4521                 nfsd4_client_record_remove(clp);
4522                 expire_client(clp);
4523                 if (++count == num)
4524                         break;
4525         }
4526         nfs4_unlock_state();
4527
4528         printk(KERN_INFO "NFSD: Forgot %d clients", count);
4529 }
4530
4531 static void release_lockowner_sop(struct nfs4_stateowner *sop)
4532 {
4533         release_lockowner(lockowner(sop));
4534 }
4535
4536 static void release_openowner_sop(struct nfs4_stateowner *sop)
4537 {
4538         release_openowner(openowner(sop));
4539 }
4540
4541 static int nfsd_release_n_owners(u64 num, bool is_open_owner,
4542                                 void (*release_sop)(struct nfs4_stateowner *))
4543 {
4544         int i, count = 0;
4545         struct nfs4_stateowner *sop, *next;
4546
4547         for (i = 0; i < OWNER_HASH_SIZE; i++) {
4548                 list_for_each_entry_safe(sop, next, &ownerstr_hashtbl[i], so_strhash) {
4549                         if (sop->so_is_open_owner != is_open_owner)
4550                                 continue;
4551                         release_sop(sop);
4552                         if (++count == num)
4553                                 return count;
4554                 }
4555         }
4556         return count;
4557 }
4558
4559 void nfsd_forget_locks(u64 num)
4560 {
4561         int count;
4562
4563         nfs4_lock_state();
4564         count = nfsd_release_n_owners(num, false, release_lockowner_sop);
4565         nfs4_unlock_state();
4566
4567         printk(KERN_INFO "NFSD: Forgot %d locks", count);
4568 }
4569
4570 void nfsd_forget_openowners(u64 num)
4571 {
4572         int count;
4573
4574         nfs4_lock_state();
4575         count = nfsd_release_n_owners(num, true, release_openowner_sop);
4576         nfs4_unlock_state();
4577
4578         printk(KERN_INFO "NFSD: Forgot %d open owners", count);
4579 }
4580
4581 int nfsd_process_n_delegations(u64 num, struct list_head *list)
4582 {
4583         int i, count = 0;
4584         struct nfs4_file *fp, *fnext;
4585         struct nfs4_delegation *dp, *dnext;
4586
4587         for (i = 0; i < FILE_HASH_SIZE; i++) {
4588                 list_for_each_entry_safe(fp, fnext, &file_hashtbl[i], fi_hash) {
4589                         list_for_each_entry_safe(dp, dnext, &fp->fi_delegations, dl_perfile) {
4590                                 list_move(&dp->dl_recall_lru, list);
4591                                 if (++count == num)
4592                                         return count;
4593                         }
4594                 }
4595         }
4596
4597         return count;
4598 }
4599
4600 void nfsd_forget_delegations(u64 num)
4601 {
4602         unsigned int count;
4603         LIST_HEAD(victims);
4604         struct nfs4_delegation *dp, *dnext;
4605
4606         spin_lock(&recall_lock);
4607         count = nfsd_process_n_delegations(num, &victims);
4608         spin_unlock(&recall_lock);
4609
4610         nfs4_lock_state();
4611         list_for_each_entry_safe(dp, dnext, &victims, dl_recall_lru)
4612                 unhash_delegation(dp);
4613         nfs4_unlock_state();
4614
4615         printk(KERN_INFO "NFSD: Forgot %d delegations", count);
4616 }
4617
4618 void nfsd_recall_delegations(u64 num)
4619 {
4620         unsigned int count;
4621         LIST_HEAD(victims);
4622         struct nfs4_delegation *dp, *dnext;
4623
4624         spin_lock(&recall_lock);
4625         count = nfsd_process_n_delegations(num, &victims);
4626         list_for_each_entry_safe(dp, dnext, &victims, dl_recall_lru) {
4627                 list_del(&dp->dl_recall_lru);
4628                 nfsd_break_one_deleg(dp);
4629         }
4630         spin_unlock(&recall_lock);
4631
4632         printk(KERN_INFO "NFSD: Recalled %d delegations", count);
4633 }
4634
4635 #endif /* CONFIG_NFSD_FAULT_INJECTION */
4636
4637 /* initialization to perform at module load time: */
4638
4639 void
4640 nfs4_state_init(void)
4641 {
4642         int i;
4643
4644         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4645                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4646                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4647                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4648                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4649                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4650         }
4651         for (i = 0; i < SESSION_HASH_SIZE; i++)
4652                 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4653         for (i = 0; i < FILE_HASH_SIZE; i++) {
4654                 INIT_LIST_HEAD(&file_hashtbl[i]);
4655         }
4656         for (i = 0; i < OWNER_HASH_SIZE; i++) {
4657                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4658         }
4659         for (i = 0; i < LOCKOWNER_INO_HASH_SIZE; i++)
4660                 INIT_LIST_HEAD(&lockowner_ino_hashtbl[i]);
4661         INIT_LIST_HEAD(&close_lru);
4662         INIT_LIST_HEAD(&client_lru);
4663         INIT_LIST_HEAD(&del_recall_lru);
4664         reclaim_str_hashtbl_size = 0;
4665 }
4666
4667 /*
4668  * Since the lifetime of a delegation isn't limited to that of an open, a
4669  * client may quite reasonably hang on to a delegation as long as it has
4670  * the inode cached.  This becomes an obvious problem the first time a
4671  * client's inode cache approaches the size of the server's total memory.
4672  *
4673  * For now we avoid this problem by imposing a hard limit on the number
4674  * of delegations, which varies according to the server's memory size.
4675  */
4676 static void
4677 set_max_delegations(void)
4678 {
4679         /*
4680          * Allow at most 4 delegations per megabyte of RAM.  Quick
4681          * estimates suggest that in the worst case (where every delegation
4682          * is for a different inode), a delegation could take about 1.5K,
4683          * giving a worst case usage of about 6% of memory.
4684          */
4685         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4686 }
4687
4688 /* initialization to perform when the nfsd service is started: */
4689
4690 int
4691 nfs4_state_start(void)
4692 {
4693         struct net *net = &init_net;
4694         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4695         int ret;
4696
4697         /*
4698          * FIXME: For now, we hang most of the pernet global stuff off of
4699          * init_net until nfsd is fully containerized. Eventually, we'll
4700          * need to pass a net pointer into this function, take a reference
4701          * to that instead and then do most of the rest of this on a per-net
4702          * basis.
4703          */
4704         get_net(net);
4705         nfsd4_client_tracking_init(net);
4706         boot_time = get_seconds();
4707         locks_start_grace(net, &nn->nfsd4_manager);
4708         grace_ended = false;
4709         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4710                nfsd4_grace);
4711         ret = set_callback_cred();
4712         if (ret) {
4713                 ret = -ENOMEM;
4714                 goto out_recovery;
4715         }
4716         laundry_wq = create_singlethread_workqueue("nfsd4");
4717         if (laundry_wq == NULL) {
4718                 ret = -ENOMEM;
4719                 goto out_recovery;
4720         }
4721         ret = nfsd4_create_callback_queue();
4722         if (ret)
4723                 goto out_free_laundry;
4724         queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4725         set_max_delegations();
4726         return 0;
4727 out_free_laundry:
4728         destroy_workqueue(laundry_wq);
4729 out_recovery:
4730         nfsd4_client_tracking_exit(net);
4731         put_net(net);
4732         return ret;
4733 }
4734
4735 static void
4736 __nfs4_state_shutdown(void)
4737 {
4738         int i;
4739         struct nfs4_client *clp = NULL;
4740         struct nfs4_delegation *dp = NULL;
4741         struct list_head *pos, *next, reaplist;
4742
4743         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4744                 while (!list_empty(&conf_id_hashtbl[i])) {
4745                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4746                         expire_client(clp);
4747                 }
4748                 while (!list_empty(&unconf_str_hashtbl[i])) {
4749                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4750                         expire_client(clp);
4751                 }
4752         }
4753         INIT_LIST_HEAD(&reaplist);
4754         spin_lock(&recall_lock);
4755         list_for_each_safe(pos, next, &del_recall_lru) {
4756                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4757                 list_move(&dp->dl_recall_lru, &reaplist);
4758         }
4759         spin_unlock(&recall_lock);
4760         list_for_each_safe(pos, next, &reaplist) {
4761                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4762                 unhash_delegation(dp);
4763         }
4764
4765         nfsd4_client_tracking_exit(&init_net);
4766         put_net(&init_net);
4767 }
4768
4769 void
4770 nfs4_state_shutdown(void)
4771 {
4772         struct net *net = &init_net;
4773         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4774
4775         cancel_delayed_work_sync(&laundromat_work);
4776         destroy_workqueue(laundry_wq);
4777         locks_end_grace(&nn->nfsd4_manager);
4778         nfs4_lock_state();
4779         __nfs4_state_shutdown();
4780         nfs4_unlock_state();
4781         nfsd4_destroy_callback_queue();
4782 }
4783
4784 static void
4785 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
4786 {
4787         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
4788                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
4789 }
4790
4791 static void
4792 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
4793 {
4794         if (cstate->minorversion) {
4795                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
4796                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
4797         }
4798 }
4799
4800 void
4801 clear_current_stateid(struct nfsd4_compound_state *cstate)
4802 {
4803         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
4804 }
4805
4806 /*
4807  * functions to set current state id
4808  */
4809 void
4810 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
4811 {
4812         put_stateid(cstate, &odp->od_stateid);
4813 }
4814
4815 void
4816 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
4817 {
4818         put_stateid(cstate, &open->op_stateid);
4819 }
4820
4821 void
4822 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
4823 {
4824         put_stateid(cstate, &close->cl_stateid);
4825 }
4826
4827 void
4828 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
4829 {
4830         put_stateid(cstate, &lock->lk_resp_stateid);
4831 }
4832
4833 /*
4834  * functions to consume current state id
4835  */
4836
4837 void
4838 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
4839 {
4840         get_stateid(cstate, &odp->od_stateid);
4841 }
4842
4843 void
4844 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
4845 {
4846         get_stateid(cstate, &drp->dr_stateid);
4847 }
4848
4849 void
4850 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
4851 {
4852         get_stateid(cstate, &fsp->fr_stateid);
4853 }
4854
4855 void
4856 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
4857 {
4858         get_stateid(cstate, &setattr->sa_stateid);
4859 }
4860
4861 void
4862 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
4863 {
4864         get_stateid(cstate, &close->cl_stateid);
4865 }
4866
4867 void
4868 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
4869 {
4870         get_stateid(cstate, &locku->lu_stateid);
4871 }
4872
4873 void
4874 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
4875 {
4876         get_stateid(cstate, &read->rd_stateid);
4877 }
4878
4879 void
4880 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
4881 {
4882         get_stateid(cstate, &write->wr_stateid);
4883 }