fs/nfsd/nfs4state.c

   1 /*
   2 *  Copyright (c) 2001 The Regents of the University of Michigan.
   3 *  All rights reserved.
   4 *
   5 *  Kendrick Smith <kmsmith@umich.edu>
   6 *  Andy Adamson <kandros@umich.edu>
   7 *
   8 *  Redistribution and use in source and binary forms, with or without
   9 *  modification, are permitted provided that the following conditions
  10 *  are met:
  11 *
  12 *  1. Redistributions of source code must retain the above copyright
  13 *     notice, this list of conditions and the following disclaimer.
  14 *  2. Redistributions in binary form must reproduce the above copyright
  15 *     notice, this list of conditions and the following disclaimer in the
  16 *     documentation and/or other materials provided with the distribution.
  17 *  3. Neither the name of the University nor the names of its
  18 *     contributors may be used to endorse or promote products derived
  19 *     from this software without specific prior written permission.
  20 *
  21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32 *
  33 */
  34
  35 #include <linux/file.h>
  36 #include <linux/fs.h>
  37 #include <linux/slab.h>
  38 #include <linux/namei.h>
  39 #include <linux/swap.h>
  40 #include <linux/pagemap.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/sunrpc/svcauth_gss.h>
  43 #include <linux/sunrpc/addr.h>
  44 #include "xdr4.h"
  45 #include "vfs.h"
  46 #include "current_stateid.h"
  47
  48 #include "netns.h"
  49
  50 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  51
  52 #define all_ones {{~0,~0},~0}
  53 static const stateid_t one_stateid = {
  54         .si_generation = ~0,
  55         .si_opaque = all_ones,
  56 };
  57 static const stateid_t zero_stateid = {
  58         /* all fields zero */
  59 };
  60 static const stateid_t currentstateid = {
  61         .si_generation = 1,
  62 };
  63
  64 static u64 current_sessionid = 1;
  65
  66 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  67 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  68 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  69
  70 /* forward declarations */
  71 static int check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner);
  72
  73 /* Locking: */
  74
  75 /* Currently used for almost all code touching nfsv4 state: */
  76 static DEFINE_MUTEX(client_mutex);
  77
  78 /*
  79  * Currently used for the del_recall_lru and file hash table.  In an
  80  * effort to decrease the scope of the client_mutex, this spinlock may
  81  * eventually cover more:
  82  */
  83 static DEFINE_SPINLOCK(recall_lock);
  84
  85 static struct kmem_cache *openowner_slab = NULL;
  86 static struct kmem_cache *lockowner_slab = NULL;
  87 static struct kmem_cache *file_slab = NULL;
  88 static struct kmem_cache *stateid_slab = NULL;
  89 static struct kmem_cache *deleg_slab = NULL;
  90
  91 void
  92 nfs4_lock_state(void)
  93 {
  94         mutex_lock(&client_mutex);
  95 }
  96
  97 static void free_session(struct kref *);
  98
  99 /* Must be called under the client_lock */
 100 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 101 {
 102         kref_put(&ses->se_ref, free_session);
 103 }
 104
 105 static void nfsd4_get_session(struct nfsd4_session *ses)
 106 {
 107         kref_get(&ses->se_ref);
 108 }
 109
 110 void
 111 nfs4_unlock_state(void)
 112 {
 113         mutex_unlock(&client_mutex);
 114 }
 115
 116 static inline u32
 117 opaque_hashval(const void *ptr, int nbytes)
 118 {
 119         unsigned char *cptr = (unsigned char *) ptr;
 120
 121         u32 x = 0;
 122         while (nbytes--) {
 123                 x *= 37;
 124                 x += *cptr++;
 125         }
 126         return x;
 127 }
 128
 129 static struct list_head del_recall_lru;
 130
 131 static void nfsd4_free_file(struct nfs4_file *f)
 132 {
 133         kmem_cache_free(file_slab, f);
 134 }
 135
 136 static inline void
 137 put_nfs4_file(struct nfs4_file *fi)
 138 {
 139         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
 140                 list_del(&fi->fi_hash);
 141                 spin_unlock(&recall_lock);
 142                 iput(fi->fi_inode);
 143                 nfsd4_free_file(fi);
 144         }
 145 }
 146
 147 static inline void
 148 get_nfs4_file(struct nfs4_file *fi)
 149 {
 150         atomic_inc(&fi->fi_ref);
 151 }
 152
 153 static int num_delegations;
 154 unsigned long max_delegations;
 155
 156 /*
 157  * Open owner state (share locks)
 158  */
 159
 160 /* hash tables for lock and open owners */
 161 #define OWNER_HASH_BITS              8
 162 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 163 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 164
 165 static unsigned int ownerstr_hashval(u32 clientid, struct xdr_netobj *ownername)
 166 {
 167         unsigned int ret;
 168
 169         ret = opaque_hashval(ownername->data, ownername->len);
 170         ret += clientid;
 171         return ret & OWNER_HASH_MASK;
 172 }
 173
 174 /* hash table for nfs4_file */
 175 #define FILE_HASH_BITS                   8
 176 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 177
 178 static unsigned int file_hashval(struct inode *ino)
 179 {
 180         /* XXX: why are we hashing on inode pointer, anyway? */
 181         return hash_ptr(ino, FILE_HASH_BITS);
 182 }
 183
 184 static struct list_head file_hashtbl[FILE_HASH_SIZE];
 185
 186 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
 187 {
 188         WARN_ON_ONCE(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
 189         atomic_inc(&fp->fi_access[oflag]);
 190 }
 191
 192 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
 193 {
 194         if (oflag == O_RDWR) {
 195                 __nfs4_file_get_access(fp, O_RDONLY);
 196                 __nfs4_file_get_access(fp, O_WRONLY);
 197         } else
 198                 __nfs4_file_get_access(fp, oflag);
 199 }
 200
 201 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
 202 {
 203         if (fp->fi_fds[oflag]) {
 204                 fput(fp->fi_fds[oflag]);
 205                 fp->fi_fds[oflag] = NULL;
 206         }
 207 }
 208
 209 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 210 {
 211         if (atomic_dec_and_test(&fp->fi_access[oflag])) {
 212                 nfs4_file_put_fd(fp, oflag);
 213                 /*
 214                  * It's also safe to get rid of the RDWR open *if*
 215                  * we no longer have need of the other kind of access
 216                  * or if we already have the other kind of open:
 217                  */
 218                 if (fp->fi_fds[1-oflag]
 219                         || atomic_read(&fp->fi_access[1 - oflag]) == 0)
 220                         nfs4_file_put_fd(fp, O_RDWR);
 221         }
 222 }
 223
 224 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 225 {
 226         if (oflag == O_RDWR) {
 227                 __nfs4_file_put_access(fp, O_RDONLY);
 228                 __nfs4_file_put_access(fp, O_WRONLY);
 229         } else
 230                 __nfs4_file_put_access(fp, oflag);
 231 }
 232
 233 static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct
 234 kmem_cache *slab)
 235 {
 236         struct idr *stateids = &cl->cl_stateids;
 237         static int min_stateid = 0;
 238         struct nfs4_stid *stid;
 239         int new_id;
 240
 241         stid = kmem_cache_alloc(slab, GFP_KERNEL);
 242         if (!stid)
 243                 return NULL;
 244
 245         new_id = idr_alloc(stateids, stid, min_stateid, 0, GFP_KERNEL);
 246         if (new_id < 0)
 247                 goto out_free;
 248         stid->sc_client = cl;
 249         stid->sc_type = 0;
 250         stid->sc_stateid.si_opaque.so_id = new_id;
 251         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 252         /* Will be incremented before return to client: */
 253         stid->sc_stateid.si_generation = 0;
 254
 255         /*
 256          * It shouldn't be a problem to reuse an opaque stateid value.
 257          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 258          * example, a stray write retransmission could be accepted by
 259          * the server when it should have been rejected.  Therefore,
 260          * adopt a trick from the sctp code to attempt to maximize the
 261          * amount of time until an id is reused, by ensuring they always
 262          * "increase" (mod INT_MAX):
 263          */
 264
 265         min_stateid = new_id+1;
 266         if (min_stateid == INT_MAX)
 267                 min_stateid = 0;
 268         return stid;
 269 out_free:
 270         kfree(stid);
 271         return NULL;
 272 }
 273
 274 static struct nfs4_ol_stateid * nfs4_alloc_stateid(struct nfs4_client *clp)
 275 {
 276         return openlockstateid(nfs4_alloc_stid(clp, stateid_slab));
 277 }
 278
 279 static struct nfs4_delegation *
 280 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_ol_stateid *stp, struct svc_fh *current_fh, u32 type)
 281 {
 282         struct nfs4_delegation *dp;
 283         struct nfs4_file *fp = stp->st_file;
 284
 285         dprintk("NFSD alloc_init_deleg\n");
 286         /*
 287          * Major work on the lease subsystem (for example, to support
 288          * calbacks on stat) will be required before we can support
 289          * write delegations properly.
 290          */
 291         if (type != NFS4_OPEN_DELEGATE_READ)
 292                 return NULL;
 293         if (fp->fi_had_conflict)
 294                 return NULL;
 295         if (num_delegations > max_delegations)
 296                 return NULL;
 297         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
 298         if (dp == NULL)
 299                 return dp;
 300         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 301         /*
 302          * delegation seqid's are never incremented.  The 4.1 special
 303          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 304          * 0 anyway just for consistency and use 1:
 305          */
 306         dp->dl_stid.sc_stateid.si_generation = 1;
 307         num_delegations++;
 308         INIT_LIST_HEAD(&dp->dl_perfile);
 309         INIT_LIST_HEAD(&dp->dl_perclnt);
 310         INIT_LIST_HEAD(&dp->dl_recall_lru);
 311         get_nfs4_file(fp);
 312         dp->dl_file = fp;
 313         dp->dl_type = type;
 314         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
 315         dp->dl_time = 0;
 316         atomic_set(&dp->dl_count, 1);
 317         nfsd4_init_callback(&dp->dl_recall);
 318         return dp;
 319 }
 320
 321 static void free_stid(struct nfs4_stid *s, struct kmem_cache *slab)
 322 {
 323         struct idr *stateids = &s->sc_client->cl_stateids;
 324
 325         idr_remove(stateids, s->sc_stateid.si_opaque.so_id);
 326         kmem_cache_free(slab, s);
 327 }
 328
 329 void
 330 nfs4_put_delegation(struct nfs4_delegation *dp)
 331 {
 332         if (atomic_dec_and_test(&dp->dl_count)) {
 333                 dprintk("NFSD: freeing dp %p\n",dp);
 334                 put_nfs4_file(dp->dl_file);
 335                 free_stid(&dp->dl_stid, deleg_slab);
 336                 num_delegations--;
 337         }
 338 }
 339
 340 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 341 {
 342         if (atomic_dec_and_test(&fp->fi_delegees)) {
 343                 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
 344                 fp->fi_lease = NULL;
 345                 fput(fp->fi_deleg_file);
 346                 fp->fi_deleg_file = NULL;
 347         }
 348 }
 349
 350 static void unhash_stid(struct nfs4_stid *s)
 351 {
 352         s->sc_type = 0;
 353 }
 354
 355 /* Called under the state lock. */
 356 static void
 357 unhash_delegation(struct nfs4_delegation *dp)
 358 {
 359         unhash_stid(&dp->dl_stid);
 360         list_del_init(&dp->dl_perclnt);
 361         spin_lock(&recall_lock);
 362         list_del_init(&dp->dl_perfile);
 363         list_del_init(&dp->dl_recall_lru);
 364         spin_unlock(&recall_lock);
 365         nfs4_put_deleg_lease(dp->dl_file);
 366         nfs4_put_delegation(dp);
 367 }
 368
 369 /*
 370  * SETCLIENTID state
 371  */
 372
 373 static unsigned int clientid_hashval(u32 id)
 374 {
 375         return id & CLIENT_HASH_MASK;
 376 }
 377
 378 static unsigned int clientstr_hashval(const char *name)
 379 {
 380         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
 381 }
 382
 383 /*
 384  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 385  * st_{access,deny}_bmap field of the stateid, in order to track not
 386  * only what share bits are currently in force, but also what
 387  * combinations of share bits previous opens have used.  This allows us
 388  * to enforce the recommendation of rfc 3530 14.2.19 that the server
 389  * return an error if the client attempt to downgrade to a combination
 390  * of share bits not explicable by closing some of its previous opens.
 391  *
 392  * XXX: This enforcement is actually incomplete, since we don't keep
 393  * track of access/deny bit combinations; so, e.g., we allow:
 394  *
 395  *      OPEN allow read, deny write
 396  *      OPEN allow both, deny none
 397  *      DOWNGRADE allow read, deny none
 398  *
 399  * which we should reject.
 400  */
 401 static unsigned int
 402 bmap_to_share_mode(unsigned long bmap) {
 403         int i;
 404         unsigned int access = 0;
 405
 406         for (i = 1; i < 4; i++) {
 407                 if (test_bit(i, &bmap))
 408                         access |= i;
 409         }
 410         return access;
 411 }
 412
 413 static bool
 414 test_share(struct nfs4_ol_stateid *stp, struct nfsd4_open *open) {
 415         unsigned int access, deny;
 416
 417         access = bmap_to_share_mode(stp->st_access_bmap);
 418         deny = bmap_to_share_mode(stp->st_deny_bmap);
 419         if ((access & open->op_share_deny) || (deny & open->op_share_access))
 420                 return false;
 421         return true;
 422 }
 423
 424 /* set share access for a given stateid */
 425 static inline void
 426 set_access(u32 access, struct nfs4_ol_stateid *stp)
 427 {
 428         __set_bit(access, &stp->st_access_bmap);
 429 }
 430
 431 /* clear share access for a given stateid */
 432 static inline void
 433 clear_access(u32 access, struct nfs4_ol_stateid *stp)
 434 {
 435         __clear_bit(access, &stp->st_access_bmap);
 436 }
 437
 438 /* test whether a given stateid has access */
 439 static inline bool
 440 test_access(u32 access, struct nfs4_ol_stateid *stp)
 441 {
 442         return test_bit(access, &stp->st_access_bmap);
 443 }
 444
 445 /* set share deny for a given stateid */
 446 static inline void
 447 set_deny(u32 access, struct nfs4_ol_stateid *stp)
 448 {
 449         __set_bit(access, &stp->st_deny_bmap);
 450 }
 451
 452 /* clear share deny for a given stateid */
 453 static inline void
 454 clear_deny(u32 access, struct nfs4_ol_stateid *stp)
 455 {
 456         __clear_bit(access, &stp->st_deny_bmap);
 457 }
 458
 459 /* test whether a given stateid is denying specific access */
 460 static inline bool
 461 test_deny(u32 access, struct nfs4_ol_stateid *stp)
 462 {
 463         return test_bit(access, &stp->st_deny_bmap);
 464 }
 465
 466 static int nfs4_access_to_omode(u32 access)
 467 {
 468         switch (access & NFS4_SHARE_ACCESS_BOTH) {
 469         case NFS4_SHARE_ACCESS_READ:
 470                 return O_RDONLY;
 471         case NFS4_SHARE_ACCESS_WRITE:
 472                 return O_WRONLY;
 473         case NFS4_SHARE_ACCESS_BOTH:
 474                 return O_RDWR;
 475         }
 476         WARN_ON_ONCE(1);
 477         return O_RDONLY;
 478 }
 479
 480 /* release all access and file references for a given stateid */
 481 static void
 482 release_all_access(struct nfs4_ol_stateid *stp)
 483 {
 484         int i;
 485
 486         for (i = 1; i < 4; i++) {
 487                 if (test_access(i, stp))
 488                         nfs4_file_put_access(stp->st_file,
 489                                              nfs4_access_to_omode(i));
 490                 clear_access(i, stp);
 491         }
 492 }
 493
 494 static void unhash_generic_stateid(struct nfs4_ol_stateid *stp)
 495 {
 496         list_del(&stp->st_perfile);
 497         list_del(&stp->st_perstateowner);
 498 }
 499
 500 static void close_generic_stateid(struct nfs4_ol_stateid *stp)
 501 {
 502         release_all_access(stp);
 503         put_nfs4_file(stp->st_file);
 504         stp->st_file = NULL;
 505 }
 506
 507 static void free_generic_stateid(struct nfs4_ol_stateid *stp)
 508 {
 509         free_stid(&stp->st_stid, stateid_slab);
 510 }
 511
 512 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
 513 {
 514         struct file *file;
 515
 516         unhash_generic_stateid(stp);
 517         unhash_stid(&stp->st_stid);
 518         file = find_any_file(stp->st_file);
 519         if (file)
 520                 locks_remove_posix(file, (fl_owner_t)lockowner(stp->st_stateowner));
 521         close_generic_stateid(stp);
 522         free_generic_stateid(stp);
 523 }
 524
 525 static void unhash_lockowner(struct nfs4_lockowner *lo)
 526 {
 527         struct nfs4_ol_stateid *stp;
 528
 529         list_del(&lo->lo_owner.so_strhash);
 530         list_del(&lo->lo_perstateid);
 531         list_del(&lo->lo_owner_ino_hash);
 532         while (!list_empty(&lo->lo_owner.so_stateids)) {
 533                 stp = list_first_entry(&lo->lo_owner.so_stateids,
 534                                 struct nfs4_ol_stateid, st_perstateowner);
 535                 release_lock_stateid(stp);
 536         }
 537 }
 538
 539 static void release_lockowner(struct nfs4_lockowner *lo)
 540 {
 541         unhash_lockowner(lo);
 542         nfs4_free_lockowner(lo);
 543 }
 544
 545 static void
 546 release_stateid_lockowners(struct nfs4_ol_stateid *open_stp)
 547 {
 548         struct nfs4_lockowner *lo;
 549
 550         while (!list_empty(&open_stp->st_lockowners)) {
 551                 lo = list_entry(open_stp->st_lockowners.next,
 552                                 struct nfs4_lockowner, lo_perstateid);
 553                 release_lockowner(lo);
 554         }
 555 }
 556
 557 static void unhash_open_stateid(struct nfs4_ol_stateid *stp)
 558 {
 559         unhash_generic_stateid(stp);
 560         release_stateid_lockowners(stp);
 561         close_generic_stateid(stp);
 562 }
 563
 564 static void release_open_stateid(struct nfs4_ol_stateid *stp)
 565 {
 566         unhash_open_stateid(stp);
 567         unhash_stid(&stp->st_stid);
 568         free_generic_stateid(stp);
 569 }
 570
 571 static void unhash_openowner(struct nfs4_openowner *oo)
 572 {
 573         struct nfs4_ol_stateid *stp;
 574
 575         list_del(&oo->oo_owner.so_strhash);
 576         list_del(&oo->oo_perclient);
 577         while (!list_empty(&oo->oo_owner.so_stateids)) {
 578                 stp = list_first_entry(&oo->oo_owner.so_stateids,
 579                                 struct nfs4_ol_stateid, st_perstateowner);
 580                 release_open_stateid(stp);
 581         }
 582 }
 583
 584 static void release_last_closed_stateid(struct nfs4_openowner *oo)
 585 {
 586         struct nfs4_ol_stateid *s = oo->oo_last_closed_stid;
 587
 588         if (s) {
 589                 unhash_stid(&s->st_stid);
 590                 free_generic_stateid(s);
 591                 oo->oo_last_closed_stid = NULL;
 592         }
 593 }
 594
 595 static void release_openowner(struct nfs4_openowner *oo)
 596 {
 597         unhash_openowner(oo);
 598         list_del(&oo->oo_close_lru);
 599         release_last_closed_stateid(oo);
 600         nfs4_free_openowner(oo);
 601 }
 602
 603 static inline int
 604 hash_sessionid(struct nfs4_sessionid *sessionid)
 605 {
 606         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
 607
 608         return sid->sequence % SESSION_HASH_SIZE;
 609 }
 610
 611 #ifdef NFSD_DEBUG
 612 static inline void
 613 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 614 {
 615         u32 *ptr = (u32 *)(&sessionid->data[0]);
 616         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
 617 }
 618 #else
 619 static inline void
 620 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 621 {
 622 }
 623 #endif
 624
 625
 626 static void
 627 gen_sessionid(struct nfsd4_session *ses)
 628 {
 629         struct nfs4_client *clp = ses->se_client;
 630         struct nfsd4_sessionid *sid;
 631
 632         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
 633         sid->clientid = clp->cl_clientid;
 634         sid->sequence = current_sessionid++;
 635         sid->reserved = 0;
 636 }
 637
 638 /*
 639  * The protocol defines ca_maxresponssize_cached to include the size of
 640  * the rpc header, but all we need to cache is the data starting after
 641  * the end of the initial SEQUENCE operation--the rest we regenerate
 642  * each time.  Therefore we can advertise a ca_maxresponssize_cached
 643  * value that is the number of bytes in our cache plus a few additional
 644  * bytes.  In order to stay on the safe side, and not promise more than
 645  * we can cache, those additional bytes must be the minimum possible: 24
 646  * bytes of rpc header (xid through accept state, with AUTH_NULL
 647  * verifier), 12 for the compound header (with zero-length tag), and 44
 648  * for the SEQUENCE op response:
 649  */
 650 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
 651
 652 static void
 653 free_session_slots(struct nfsd4_session *ses)
 654 {
 655         int i;
 656
 657         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
 658                 kfree(ses->se_slots[i]);
 659 }
 660
 661 /*
 662  * We don't actually need to cache the rpc and session headers, so we
 663  * can allocate a little less for each slot:
 664  */
 665 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
 666 {
 667         return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
 668 }
 669
 670 static int nfsd4_sanitize_slot_size(u32 size)
 671 {
 672         size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
 673         size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
 674
 675         return size;
 676 }
 677
 678 /*
 679  * XXX: If we run out of reserved DRC memory we could (up to a point)
 680  * re-negotiate active sessions and reduce their slot usage to make
 681  * room for new connections. For now we just fail the create session.
 682  */
 683 static int nfsd4_get_drc_mem(int slotsize, u32 num)
 684 {
 685         int avail;
 686
 687         num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
 688
 689         spin_lock(&nfsd_drc_lock);
 690         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
 691                     nfsd_drc_max_mem - nfsd_drc_mem_used);
 692         num = min_t(int, num, avail / slotsize);
 693         nfsd_drc_mem_used += num * slotsize;
 694         spin_unlock(&nfsd_drc_lock);
 695
 696         return num;
 697 }
 698
 699 static void nfsd4_put_drc_mem(int slotsize, int num)
 700 {
 701         spin_lock(&nfsd_drc_lock);
 702         nfsd_drc_mem_used -= slotsize * num;
 703         spin_unlock(&nfsd_drc_lock);
 704 }
 705
 706 static struct nfsd4_session *__alloc_session(int slotsize, int numslots)
 707 {
 708         struct nfsd4_session *new;
 709         int mem, i;
 710
 711         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
 712                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
 713         mem = numslots * sizeof(struct nfsd4_slot *);
 714
 715         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
 716         if (!new)
 717                 return NULL;
 718         /* allocate each struct nfsd4_slot and data cache in one piece */
 719         for (i = 0; i < numslots; i++) {
 720                 mem = sizeof(struct nfsd4_slot) + slotsize;
 721                 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
 722                 if (!new->se_slots[i])
 723                         goto out_free;
 724         }
 725         return new;
 726 out_free:
 727         while (i--)
 728                 kfree(new->se_slots[i]);
 729         kfree(new);
 730         return NULL;
 731 }
 732
 733 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new,
 734                                    struct nfsd4_channel_attrs *req,
 735                                    int numslots, int slotsize,
 736                                    struct nfsd_net *nn)
 737 {
 738         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
 739
 740         new->maxreqs = numslots;
 741         new->maxresp_cached = min_t(u32, req->maxresp_cached,
 742                                         slotsize + NFSD_MIN_HDR_SEQ_SZ);
 743         new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
 744         new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
 745         new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
 746 }
 747
 748 static void free_conn(struct nfsd4_conn *c)
 749 {
 750         svc_xprt_put(c->cn_xprt);
 751         kfree(c);
 752 }
 753
 754 static void nfsd4_conn_lost(struct svc_xpt_user *u)
 755 {
 756         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
 757         struct nfs4_client *clp = c->cn_session->se_client;
 758
 759         spin_lock(&clp->cl_lock);
 760         if (!list_empty(&c->cn_persession)) {
 761                 list_del(&c->cn_persession);
 762                 free_conn(c);
 763         }
 764         nfsd4_probe_callback(clp);
 765         spin_unlock(&clp->cl_lock);
 766 }
 767
 768 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
 769 {
 770         struct nfsd4_conn *conn;
 771
 772         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
 773         if (!conn)
 774                 return NULL;
 775         svc_xprt_get(rqstp->rq_xprt);
 776         conn->cn_xprt = rqstp->rq_xprt;
 777         conn->cn_flags = flags;
 778         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
 779         return conn;
 780 }
 781
 782 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 783 {
 784         conn->cn_session = ses;
 785         list_add(&conn->cn_persession, &ses->se_conns);
 786 }
 787
 788 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 789 {
 790         struct nfs4_client *clp = ses->se_client;
 791
 792         spin_lock(&clp->cl_lock);
 793         __nfsd4_hash_conn(conn, ses);
 794         spin_unlock(&clp->cl_lock);
 795 }
 796
 797 static int nfsd4_register_conn(struct nfsd4_conn *conn)
 798 {
 799         conn->cn_xpt_user.callback = nfsd4_conn_lost;
 800         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
 801 }
 802
 803 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
 804 {
 805         int ret;
 806
 807         nfsd4_hash_conn(conn, ses);
 808         ret = nfsd4_register_conn(conn);
 809         if (ret)
 810                 /* oops; xprt is already down: */
 811                 nfsd4_conn_lost(&conn->cn_xpt_user);
 812         if (conn->cn_flags & NFS4_CDFC4_BACK) {
 813                 /* callback channel may be back up */
 814                 nfsd4_probe_callback(ses->se_client);
 815         }
 816 }
 817
 818 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
 819 {
 820         u32 dir = NFS4_CDFC4_FORE;
 821
 822         if (cses->flags & SESSION4_BACK_CHAN)
 823                 dir |= NFS4_CDFC4_BACK;
 824         return alloc_conn(rqstp, dir);
 825 }
 826
 827 /* must be called under client_lock */
 828 static void nfsd4_del_conns(struct nfsd4_session *s)
 829 {
 830         struct nfs4_client *clp = s->se_client;
 831         struct nfsd4_conn *c;
 832
 833         spin_lock(&clp->cl_lock);
 834         while (!list_empty(&s->se_conns)) {
 835                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
 836                 list_del_init(&c->cn_persession);
 837                 spin_unlock(&clp->cl_lock);
 838
 839                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
 840                 free_conn(c);
 841
 842                 spin_lock(&clp->cl_lock);
 843         }
 844         spin_unlock(&clp->cl_lock);
 845 }
 846
 847 static void __free_session(struct nfsd4_session *ses)
 848 {
 849         nfsd4_put_drc_mem(slot_bytes(&ses->se_fchannel), ses->se_fchannel.maxreqs);
 850         free_session_slots(ses);
 851         kfree(ses);
 852 }
 853
 854 static void free_session(struct kref *kref)
 855 {
 856         struct nfsd4_session *ses;
 857         struct nfsd_net *nn;
 858
 859         ses = container_of(kref, struct nfsd4_session, se_ref);
 860         nn = net_generic(ses->se_client->net, nfsd_net_id);
 861
 862         lockdep_assert_held(&nn->client_lock);
 863         nfsd4_del_conns(ses);
 864         __free_session(ses);
 865 }
 866
 867 static void nfsd4_put_session(struct nfsd4_session *ses)
 868 {
 869         struct nfsd_net *nn = net_generic(ses->se_client->net, nfsd_net_id);
 870
 871         spin_lock(&nn->client_lock);
 872         nfsd4_put_session_locked(ses);
 873         spin_unlock(&nn->client_lock);
 874 }
 875
 876 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fchan,
 877                                            struct nfsd_net *nn)
 878 {
 879         struct nfsd4_session *new;
 880         int numslots, slotsize;
 881         /*
 882          * Note decreasing slot size below client's request may
 883          * make it difficult for client to function correctly, whereas
 884          * decreasing the number of slots will (just?) affect
 885          * performance.  When short on memory we therefore prefer to
 886          * decrease number of slots instead of their size.
 887          */
 888         slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
 889         numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
 890         if (numslots < 1)
 891                 return NULL;
 892
 893         new = __alloc_session(slotsize, numslots);
 894         if (!new) {
 895                 nfsd4_put_drc_mem(slotsize, numslots);
 896                 return NULL;
 897         }
 898         init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize, nn);
 899         return new;
 900 }
 901
 902 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
 903 {
 904         int idx;
 905         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
 906
 907         new->se_client = clp;
 908         gen_sessionid(new);
 909
 910         INIT_LIST_HEAD(&new->se_conns);
 911
 912         new->se_cb_seq_nr = 1;
 913         new->se_flags = cses->flags;
 914         new->se_cb_prog = cses->callback_prog;
 915         new->se_cb_sec = cses->cb_sec;
 916         kref_init(&new->se_ref);
 917         idx = hash_sessionid(&new->se_sessionid);
 918         spin_lock(&nn->client_lock);
 919         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
 920         spin_lock(&clp->cl_lock);
 921         list_add(&new->se_perclnt, &clp->cl_sessions);
 922         spin_unlock(&clp->cl_lock);
 923         spin_unlock(&nn->client_lock);
 924
 925         if (cses->flags & SESSION4_BACK_CHAN) {
 926                 struct sockaddr *sa = svc_addr(rqstp);
 927                 /*
 928                  * This is a little silly; with sessions there's no real
 929                  * use for the callback address.  Use the peer address
 930                  * as a reasonable default for now, but consider fixing
 931                  * the rpc client not to require an address in the
 932                  * future:
 933                  */
 934                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
 935                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
 936         }
 937 }
 938
 939 /* caller must hold client_lock */
 940 static struct nfsd4_session *
 941 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
 942 {
 943         struct nfsd4_session *elem;
 944         int idx;
 945         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 946
 947         dump_sessionid(__func__, sessionid);
 948         idx = hash_sessionid(sessionid);
 949         /* Search in the appropriate list */
 950         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
 951                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
 952                             NFS4_MAX_SESSIONID_LEN)) {
 953                         return elem;
 954                 }
 955         }
 956
 957         dprintk("%s: session not found\n", __func__);
 958         return NULL;
 959 }
 960
 961 /* caller must hold client_lock */
 962 static void
 963 unhash_session(struct nfsd4_session *ses)
 964 {
 965         list_del(&ses->se_hash);
 966         spin_lock(&ses->se_client->cl_lock);
 967         list_del(&ses->se_perclnt);
 968         spin_unlock(&ses->se_client->cl_lock);
 969 }
 970
 971 /* must be called under the client_lock */
 972 static inline void
 973 renew_client_locked(struct nfs4_client *clp)
 974 {
 975         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 976
 977         if (is_client_expired(clp)) {
 978                 WARN_ON(1);
 979                 printk("%s: client (clientid %08x/%08x) already expired\n",
 980                         __func__,
 981                         clp->cl_clientid.cl_boot,
 982                         clp->cl_clientid.cl_id);
 983                 return;
 984         }
 985
 986         dprintk("renewing client (clientid %08x/%08x)\n",
 987                         clp->cl_clientid.cl_boot,
 988                         clp->cl_clientid.cl_id);
 989         list_move_tail(&clp->cl_lru, &nn->client_lru);
 990         clp->cl_time = get_seconds();
 991 }
 992
 993 static inline void
 994 renew_client(struct nfs4_client *clp)
 995 {
 996         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 997
 998         spin_lock(&nn->client_lock);
 999         renew_client_locked(clp);
1000         spin_unlock(&nn->client_lock);
1001 }
1002
1003 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1004 static int
1005 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1006 {
1007         if (clid->cl_boot == nn->boot_time)
1008                 return 0;
1009         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1010                 clid->cl_boot, clid->cl_id, nn->boot_time);
1011         return 1;
1012 }
1013
1014 /*
1015  * XXX Should we use a slab cache ?
1016  * This type of memory management is somewhat inefficient, but we use it
1017  * anyway since SETCLIENTID is not a common operation.
1018  */
1019 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1020 {
1021         struct nfs4_client *clp;
1022
1023         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1024         if (clp == NULL)
1025                 return NULL;
1026         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1027         if (clp->cl_name.data == NULL) {
1028                 kfree(clp);
1029                 return NULL;
1030         }
1031         clp->cl_name.len = name.len;
1032         return clp;
1033 }
1034
1035 static inline void
1036 free_client(struct nfs4_client *clp)
1037 {
1038         struct nfsd_net __maybe_unused *nn = net_generic(clp->net, nfsd_net_id);
1039
1040         lockdep_assert_held(&nn->client_lock);
1041         while (!list_empty(&clp->cl_sessions)) {
1042                 struct nfsd4_session *ses;
1043                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1044                                 se_perclnt);
1045                 list_del(&ses->se_perclnt);
1046                 nfsd4_put_session_locked(ses);
1047         }
1048         free_svc_cred(&clp->cl_cred);
1049         kfree(clp->cl_name.data);
1050         idr_destroy(&clp->cl_stateids);
1051         kfree(clp);
1052 }
1053
1054 void
1055 release_session_client(struct nfsd4_session *session)
1056 {
1057         struct nfs4_client *clp = session->se_client;
1058         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1059
1060         nfsd4_put_session(session);
1061         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
1062                 return;
1063         /*
1064          * At this point we also know all sessions have refcnt 1,
1065          * so free_client will delete them all if necessary:
1066          */
1067         if (is_client_expired(clp))
1068                 free_client(clp);
1069         else
1070                 renew_client_locked(clp);
1071         spin_unlock(&nn->client_lock);
1072 }
1073
1074 /* must be called under the client_lock */
1075 static inline void
1076 unhash_client_locked(struct nfs4_client *clp)
1077 {
1078         struct nfsd4_session *ses;
1079
1080         mark_client_expired(clp);
1081         list_del(&clp->cl_lru);
1082         spin_lock(&clp->cl_lock);
1083         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1084                 list_del_init(&ses->se_hash);
1085         spin_unlock(&clp->cl_lock);
1086 }
1087
1088 static void
1089 destroy_client(struct nfs4_client *clp)
1090 {
1091         struct nfs4_openowner *oo;
1092         struct nfs4_delegation *dp;
1093         struct list_head reaplist;
1094         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1095
1096         INIT_LIST_HEAD(&reaplist);
1097         spin_lock(&recall_lock);
1098         while (!list_empty(&clp->cl_delegations)) {
1099                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1100                 list_del_init(&dp->dl_perclnt);
1101                 list_move(&dp->dl_recall_lru, &reaplist);
1102         }
1103         spin_unlock(&recall_lock);
1104         while (!list_empty(&reaplist)) {
1105                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1106                 unhash_delegation(dp);
1107         }
1108         while (!list_empty(&clp->cl_openowners)) {
1109                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1110                 release_openowner(oo);
1111         }
1112         nfsd4_shutdown_callback(clp);
1113         if (clp->cl_cb_conn.cb_xprt)
1114                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1115         list_del(&clp->cl_idhash);
1116         if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1117                 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1118         else
1119                 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1120         spin_lock(&nn->client_lock);
1121         unhash_client_locked(clp);
1122         if (atomic_read(&clp->cl_refcount) == 0)
1123                 free_client(clp);
1124         spin_unlock(&nn->client_lock);
1125 }
1126
1127 static void expire_client(struct nfs4_client *clp)
1128 {
1129         nfsd4_client_record_remove(clp);
1130         destroy_client(clp);
1131 }
1132
1133 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1134 {
1135         memcpy(target->cl_verifier.data, source->data,
1136                         sizeof(target->cl_verifier.data));
1137 }
1138
1139 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1140 {
1141         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1142         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1143 }
1144
1145 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1146 {
1147         if (source->cr_principal) {
1148                 target->cr_principal =
1149                                 kstrdup(source->cr_principal, GFP_KERNEL);
1150                 if (target->cr_principal == NULL)
1151                         return -ENOMEM;
1152         } else
1153                 target->cr_principal = NULL;
1154         target->cr_flavor = source->cr_flavor;
1155         target->cr_uid = source->cr_uid;
1156         target->cr_gid = source->cr_gid;
1157         target->cr_group_info = source->cr_group_info;
1158         get_group_info(target->cr_group_info);
1159         return 0;
1160 }
1161
1162 static long long
1163 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1164 {
1165         long long res;
1166
1167         res = o1->len - o2->len;
1168         if (res)
1169                 return res;
1170         return (long long)memcmp(o1->data, o2->data, o1->len);
1171 }
1172
1173 static int same_name(const char *n1, const char *n2)
1174 {
1175         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1176 }
1177
1178 static int
1179 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1180 {
1181         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1182 }
1183
1184 static int
1185 same_clid(clientid_t *cl1, clientid_t *cl2)
1186 {
1187         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1188 }
1189
1190 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1191 {
1192         int i;
1193
1194         if (g1->ngroups != g2->ngroups)
1195                 return false;
1196         for (i=0; i<g1->ngroups; i++)
1197                 if (!gid_eq(GROUP_AT(g1, i), GROUP_AT(g2, i)))
1198                         return false;
1199         return true;
1200 }
1201
1202 /*
1203  * RFC 3530 language requires clid_inuse be returned when the
1204  * "principal" associated with a requests differs from that previously
1205  * used.  We use uid, gid's, and gss principal string as our best
1206  * approximation.  We also don't want to allow non-gss use of a client
1207  * established using gss: in theory cr_principal should catch that
1208  * change, but in practice cr_principal can be null even in the gss case
1209  * since gssd doesn't always pass down a principal string.
1210  */
1211 static bool is_gss_cred(struct svc_cred *cr)
1212 {
1213         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1214         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1215 }
1216
1217
1218 static bool
1219 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1220 {
1221         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1222                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1223                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1224                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1225                 return false;
1226         if (cr1->cr_principal == cr2->cr_principal)
1227                 return true;
1228         if (!cr1->cr_principal || !cr2->cr_principal)
1229                 return false;
1230         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1231 }
1232
1233 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
1234 {
1235         static u32 current_clientid = 1;
1236
1237         clp->cl_clientid.cl_boot = nn->boot_time;
1238         clp->cl_clientid.cl_id = current_clientid++;
1239 }
1240
1241 static void gen_confirm(struct nfs4_client *clp)
1242 {
1243         __be32 verf[2];
1244         static u32 i;
1245
1246         verf[0] = (__be32)get_seconds();
1247         verf[1] = (__be32)i++;
1248         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1249 }
1250
1251 static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1252 {
1253         struct nfs4_stid *ret;
1254
1255         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1256         if (!ret || !ret->sc_type)
1257                 return NULL;
1258         return ret;
1259 }
1260
1261 static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1262 {
1263         struct nfs4_stid *s;
1264
1265         s = find_stateid(cl, t);
1266         if (!s)
1267                 return NULL;
1268         if (typemask & s->sc_type)
1269                 return s;
1270         return NULL;
1271 }
1272
1273 static struct nfs4_client *create_client(struct xdr_netobj name,
1274                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1275 {
1276         struct nfs4_client *clp;
1277         struct sockaddr *sa = svc_addr(rqstp);
1278         int ret;
1279         struct net *net = SVC_NET(rqstp);
1280         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1281
1282         clp = alloc_client(name);
1283         if (clp == NULL)
1284                 return NULL;
1285
1286         INIT_LIST_HEAD(&clp->cl_sessions);
1287         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1288         if (ret) {
1289                 spin_lock(&nn->client_lock);
1290                 free_client(clp);
1291                 spin_unlock(&nn->client_lock);
1292                 return NULL;
1293         }
1294         idr_init(&clp->cl_stateids);
1295         atomic_set(&clp->cl_refcount, 0);
1296         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1297         INIT_LIST_HEAD(&clp->cl_idhash);
1298         INIT_LIST_HEAD(&clp->cl_openowners);
1299         INIT_LIST_HEAD(&clp->cl_delegations);
1300         INIT_LIST_HEAD(&clp->cl_lru);
1301         INIT_LIST_HEAD(&clp->cl_callbacks);
1302         spin_lock_init(&clp->cl_lock);
1303         nfsd4_init_callback(&clp->cl_cb_null);
1304         clp->cl_time = get_seconds();
1305         clear_bit(0, &clp->cl_cb_slot_busy);
1306         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1307         copy_verf(clp, verf);
1308         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1309         gen_confirm(clp);
1310         clp->cl_cb_session = NULL;
1311         clp->net = net;
1312         return clp;
1313 }
1314
1315 static void
1316 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
1317 {
1318         struct rb_node **new = &(root->rb_node), *parent = NULL;
1319         struct nfs4_client *clp;
1320
1321         while (*new) {
1322                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
1323                 parent = *new;
1324
1325                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
1326                         new = &((*new)->rb_left);
1327                 else
1328                         new = &((*new)->rb_right);
1329         }
1330
1331         rb_link_node(&new_clp->cl_namenode, parent, new);
1332         rb_insert_color(&new_clp->cl_namenode, root);
1333 }
1334
1335 static struct nfs4_client *
1336 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
1337 {
1338         long long cmp;
1339         struct rb_node *node = root->rb_node;
1340         struct nfs4_client *clp;
1341
1342         while (node) {
1343                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
1344                 cmp = compare_blob(&clp->cl_name, name);
1345                 if (cmp > 0)
1346                         node = node->rb_left;
1347                 else if (cmp < 0)
1348                         node = node->rb_right;
1349                 else
1350                         return clp;
1351         }
1352         return NULL;
1353 }
1354
1355 static void
1356 add_to_unconfirmed(struct nfs4_client *clp)
1357 {
1358         unsigned int idhashval;
1359         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1360
1361         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1362         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
1363         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1364         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
1365         renew_client(clp);
1366 }
1367
1368 static void
1369 move_to_confirmed(struct nfs4_client *clp)
1370 {
1371         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1372         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1373
1374         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1375         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
1376         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1377         add_clp_to_name_tree(clp, &nn->conf_name_tree);
1378         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1379         renew_client(clp);
1380 }
1381
1382 static struct nfs4_client *
1383 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1384 {
1385         struct nfs4_client *clp;
1386         unsigned int idhashval = clientid_hashval(clid->cl_id);
1387
1388         list_for_each_entry(clp, &nn->conf_id_hashtbl[idhashval], cl_idhash) {
1389                 if (same_clid(&clp->cl_clientid, clid)) {
1390                         if ((bool)clp->cl_minorversion != sessions)
1391                                 return NULL;
1392                         renew_client(clp);
1393                         return clp;
1394                 }
1395         }
1396         return NULL;
1397 }
1398
1399 static struct nfs4_client *
1400 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1401 {
1402         struct nfs4_client *clp;
1403         unsigned int idhashval = clientid_hashval(clid->cl_id);
1404
1405         list_for_each_entry(clp, &nn->unconf_id_hashtbl[idhashval], cl_idhash) {
1406                 if (same_clid(&clp->cl_clientid, clid)) {
1407                         if ((bool)clp->cl_minorversion != sessions)
1408                                 return NULL;
1409                         return clp;
1410                 }
1411         }
1412         return NULL;
1413 }
1414
1415 static bool clp_used_exchangeid(struct nfs4_client *clp)
1416 {
1417         return clp->cl_exchange_flags != 0;
1418 }
1419
1420 static struct nfs4_client *
1421 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1422 {
1423         return find_clp_in_name_tree(name, &nn->conf_name_tree);
1424 }
1425
1426 static struct nfs4_client *
1427 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1428 {
1429         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
1430 }
1431
1432 static void
1433 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1434 {
1435         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1436         struct sockaddr *sa = svc_addr(rqstp);
1437         u32 scopeid = rpc_get_scope_id(sa);
1438         unsigned short expected_family;
1439
1440         /* Currently, we only support tcp and tcp6 for the callback channel */
1441         if (se->se_callback_netid_len == 3 &&
1442             !memcmp(se->se_callback_netid_val, "tcp", 3))
1443                 expected_family = AF_INET;
1444         else if (se->se_callback_netid_len == 4 &&
1445                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1446                 expected_family = AF_INET6;
1447         else
1448                 goto out_err;
1449
1450         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
1451                                             se->se_callback_addr_len,
1452                                             (struct sockaddr *)&conn->cb_addr,
1453                                             sizeof(conn->cb_addr));
1454
1455         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1456                 goto out_err;
1457
1458         if (conn->cb_addr.ss_family == AF_INET6)
1459                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1460
1461         conn->cb_prog = se->se_callback_prog;
1462         conn->cb_ident = se->se_callback_ident;
1463         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1464         return;
1465 out_err:
1466         conn->cb_addr.ss_family = AF_UNSPEC;
1467         conn->cb_addrlen = 0;
1468         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1469                 "will not receive delegations\n",
1470                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1471
1472         return;
1473 }
1474
1475 /*
1476  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1477  */
1478 void
1479 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1480 {
1481         struct nfsd4_slot *slot = resp->cstate.slot;
1482         unsigned int base;
1483
1484         dprintk("--> %s slot %p\n", __func__, slot);
1485
1486         slot->sl_opcnt = resp->opcnt;
1487         slot->sl_status = resp->cstate.status;
1488
1489         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
1490         if (nfsd4_not_cached(resp)) {
1491                 slot->sl_datalen = 0;
1492                 return;
1493         }
1494         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1495         base = (char *)resp->cstate.datap -
1496                                         (char *)resp->xbuf->head[0].iov_base;
1497         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1498                                     slot->sl_datalen))
1499                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1500         return;
1501 }
1502
1503 /*
1504  * Encode the replay sequence operation from the slot values.
1505  * If cachethis is FALSE encode the uncached rep error on the next
1506  * operation which sets resp->p and increments resp->opcnt for
1507  * nfs4svc_encode_compoundres.
1508  *
1509  */
1510 static __be32
1511 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1512                           struct nfsd4_compoundres *resp)
1513 {
1514         struct nfsd4_op *op;
1515         struct nfsd4_slot *slot = resp->cstate.slot;
1516
1517         /* Encode the replayed sequence operation */
1518         op = &args->ops[resp->opcnt - 1];
1519         nfsd4_encode_operation(resp, op);
1520
1521         /* Return nfserr_retry_uncached_rep in next operation. */
1522         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
1523                 op = &args->ops[resp->opcnt++];
1524                 op->status = nfserr_retry_uncached_rep;
1525                 nfsd4_encode_operation(resp, op);
1526         }
1527         return op->status;
1528 }
1529
1530 /*
1531  * The sequence operation is not cached because we can use the slot and
1532  * session values.
1533  */
1534 __be32
1535 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1536                          struct nfsd4_sequence *seq)
1537 {
1538         struct nfsd4_slot *slot = resp->cstate.slot;
1539         __be32 status;
1540
1541         dprintk("--> %s slot %p\n", __func__, slot);
1542
1543         /* Either returns 0 or nfserr_retry_uncached */
1544         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1545         if (status == nfserr_retry_uncached_rep)
1546                 return status;
1547
1548         /* The sequence operation has been encoded, cstate->datap set. */
1549         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1550
1551         resp->opcnt = slot->sl_opcnt;
1552         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1553         status = slot->sl_status;
1554
1555         return status;
1556 }
1557
1558 /*
1559  * Set the exchange_id flags returned by the server.
1560  */
1561 static void
1562 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1563 {
1564         /* pNFS is not supported */
1565         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1566
1567         /* Referrals are supported, Migration is not. */
1568         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1569
1570         /* set the wire flags to return to client. */
1571         clid->flags = new->cl_exchange_flags;
1572 }
1573
1574 static bool client_has_state(struct nfs4_client *clp)
1575 {
1576         /*
1577          * Note clp->cl_openowners check isn't quite right: there's no
1578          * need to count owners without stateid's.
1579          *
1580          * Also note we should probably be using this in 4.0 case too.
1581          */
1582         return !list_empty(&clp->cl_openowners)
1583                 || !list_empty(&clp->cl_delegations)
1584                 || !list_empty(&clp->cl_sessions);
1585 }
1586
1587 __be32
1588 nfsd4_exchange_id(struct svc_rqst *rqstp,
1589                   struct nfsd4_compound_state *cstate,
1590                   struct nfsd4_exchange_id *exid)
1591 {
1592         struct nfs4_client *unconf, *conf, *new;
1593         __be32 status;
1594         char                    addr_str[INET6_ADDRSTRLEN];
1595         nfs4_verifier           verf = exid->verifier;
1596         struct sockaddr         *sa = svc_addr(rqstp);
1597         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
1598         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1599
1600         rpc_ntop(sa, addr_str, sizeof(addr_str));
1601         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1602                 "ip_addr=%s flags %x, spa_how %d\n",
1603                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1604                 addr_str, exid->flags, exid->spa_how);
1605
1606         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1607                 return nfserr_inval;
1608
1609         /* Currently only support SP4_NONE */
1610         switch (exid->spa_how) {
1611         case SP4_NONE:
1612                 break;
1613         default:                                /* checked by xdr code */
1614                 WARN_ON_ONCE(1);
1615         case SP4_SSV:
1616         case SP4_MACH_CRED:
1617                 return nfserr_serverfault;      /* no excuse :-/ */
1618         }
1619
1620         /* Cases below refer to rfc 5661 section 18.35.4: */
1621         nfs4_lock_state();
1622         conf = find_confirmed_client_by_name(&exid->clname, nn);
1623         if (conf) {
1624                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
1625                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
1626
1627                 if (update) {
1628                         if (!clp_used_exchangeid(conf)) { /* buggy client */
1629                                 status = nfserr_inval;
1630                                 goto out;
1631                         }
1632                         if (!creds_match) { /* case 9 */
1633                                 status = nfserr_perm;
1634                                 goto out;
1635                         }
1636                         if (!verfs_match) { /* case 8 */
1637                                 status = nfserr_not_same;
1638                                 goto out;
1639                         }
1640                         /* case 6 */
1641                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1642                         new = conf;
1643                         goto out_copy;
1644                 }
1645                 if (!creds_match) { /* case 3 */
1646                         if (client_has_state(conf)) {
1647                                 status = nfserr_clid_inuse;
1648                                 goto out;
1649                         }
1650                         expire_client(conf);
1651                         goto out_new;
1652                 }
1653                 if (verfs_match) { /* case 2 */
1654                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
1655                         new = conf;
1656                         goto out_copy;
1657                 }
1658                 /* case 5, client reboot */
1659                 goto out_new;
1660         }
1661
1662         if (update) { /* case 7 */
1663                 status = nfserr_noent;
1664                 goto out;
1665         }
1666
1667         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
1668         if (unconf) /* case 4, possible retry or client restart */
1669                 expire_client(unconf);
1670
1671         /* case 1 (normal case) */
1672 out_new:
1673         new = create_client(exid->clname, rqstp, &verf);
1674         if (new == NULL) {
1675                 status = nfserr_jukebox;
1676                 goto out;
1677         }
1678         new->cl_minorversion = 1;
1679
1680         gen_clid(new, nn);
1681         add_to_unconfirmed(new);
1682 out_copy:
1683         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1684         exid->clientid.cl_id = new->cl_clientid.cl_id;
1685
1686         exid->seqid = new->cl_cs_slot.sl_seqid + 1;
1687         nfsd4_set_ex_flags(new, exid);
1688
1689         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1690                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1691         status = nfs_ok;
1692
1693 out:
1694         nfs4_unlock_state();
1695         return status;
1696 }
1697
1698 static __be32
1699 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1700 {
1701         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1702                 slot_seqid);
1703
1704         /* The slot is in use, and no response has been sent. */
1705         if (slot_inuse) {
1706                 if (seqid == slot_seqid)
1707                         return nfserr_jukebox;
1708                 else
1709                         return nfserr_seq_misordered;
1710         }
1711         /* Note unsigned 32-bit arithmetic handles wraparound: */
1712         if (likely(seqid == slot_seqid + 1))
1713                 return nfs_ok;
1714         if (seqid == slot_seqid)
1715                 return nfserr_replay_cache;
1716         return nfserr_seq_misordered;
1717 }
1718
1719 /*
1720  * Cache the create session result into the create session single DRC
1721  * slot cache by saving the xdr structure. sl_seqid has been set.
1722  * Do this for solo or embedded create session operations.
1723  */
1724 static void
1725 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1726                            struct nfsd4_clid_slot *slot, __be32 nfserr)
1727 {
1728         slot->sl_status = nfserr;
1729         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1730 }
1731
1732 static __be32
1733 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1734                             struct nfsd4_clid_slot *slot)
1735 {
1736         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1737         return slot->sl_status;
1738 }
1739
1740 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1741                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1742                         1 +     /* MIN tag is length with zero, only length */ \
1743                         3 +     /* version, opcount, opcode */ \
1744                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1745                                 /* seqid, slotID, slotID, cache */ \
1746                         4 ) * sizeof(__be32))
1747
1748 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1749                         2 +     /* verifier: AUTH_NULL, length 0 */\
1750                         1 +     /* status */ \
1751                         1 +     /* MIN tag is length with zero, only length */ \
1752                         3 +     /* opcount, opcode, opstatus*/ \
1753                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1754                                 /* seqid, slotID, slotID, slotID, status */ \
1755                         5 ) * sizeof(__be32))
1756
1757 static bool check_forechannel_attrs(struct nfsd4_channel_attrs fchannel)
1758 {
1759         return fchannel.maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ
1760                 || fchannel.maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ;
1761 }
1762
1763 __be32
1764 nfsd4_create_session(struct svc_rqst *rqstp,
1765                      struct nfsd4_compound_state *cstate,
1766                      struct nfsd4_create_session *cr_ses)
1767 {
1768         struct sockaddr *sa = svc_addr(rqstp);
1769         struct nfs4_client *conf, *unconf;
1770         struct nfsd4_session *new;
1771         struct nfsd4_conn *conn;
1772         struct nfsd4_clid_slot *cs_slot = NULL;
1773         __be32 status = 0;
1774         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1775
1776         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1777                 return nfserr_inval;
1778         if (check_forechannel_attrs(cr_ses->fore_channel))
1779                 return nfserr_toosmall;
1780         new = alloc_session(&cr_ses->fore_channel, nn);
1781         if (!new)
1782                 return nfserr_jukebox;
1783         status = nfserr_jukebox;
1784         conn = alloc_conn_from_crses(rqstp, cr_ses);
1785         if (!conn)
1786                 goto out_free_session;
1787
1788         nfs4_lock_state();
1789         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
1790         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
1791         WARN_ON_ONCE(conf && unconf);
1792
1793         if (conf) {
1794                 cs_slot = &conf->cl_cs_slot;
1795                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1796                 if (status == nfserr_replay_cache) {
1797                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1798                         goto out_free_conn;
1799                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1800                         status = nfserr_seq_misordered;
1801                         goto out_free_conn;
1802                 }
1803         } else if (unconf) {
1804                 struct nfs4_client *old;
1805                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1806                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1807                         status = nfserr_clid_inuse;
1808                         goto out_free_conn;
1809                 }
1810                 cs_slot = &unconf->cl_cs_slot;
1811                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1812                 if (status) {
1813                         /* an unconfirmed replay returns misordered */
1814                         status = nfserr_seq_misordered;
1815                         goto out_free_conn;
1816                 }
1817                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
1818                 if (old)
1819                         expire_client(old);
1820                 move_to_confirmed(unconf);
1821                 conf = unconf;
1822         } else {
1823                 status = nfserr_stale_clientid;
1824                 goto out_free_conn;
1825         }
1826         status = nfs_ok;
1827         /*
1828          * We do not support RDMA or persistent sessions
1829          */
1830         cr_ses->flags &= ~SESSION4_PERSIST;
1831         cr_ses->flags &= ~SESSION4_RDMA;
1832
1833         init_session(rqstp, new, conf, cr_ses);
1834         nfsd4_init_conn(rqstp, conn, new);
1835
1836         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1837                NFS4_MAX_SESSIONID_LEN);
1838         memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1839                 sizeof(struct nfsd4_channel_attrs));
1840         cs_slot->sl_seqid++;
1841         cr_ses->seqid = cs_slot->sl_seqid;
1842
1843         /* cache solo and embedded create sessions under the state lock */
1844         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1845         nfs4_unlock_state();
1846         return status;
1847 out_free_conn:
1848         nfs4_unlock_state();
1849         free_conn(conn);
1850 out_free_session:
1851         __free_session(new);
1852         return status;
1853 }
1854
1855 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1856 {
1857         switch (*dir) {
1858         case NFS4_CDFC4_FORE:
1859         case NFS4_CDFC4_BACK:
1860                 return nfs_ok;
1861         case NFS4_CDFC4_FORE_OR_BOTH:
1862         case NFS4_CDFC4_BACK_OR_BOTH:
1863                 *dir = NFS4_CDFC4_BOTH;
1864                 return nfs_ok;
1865         };
1866         return nfserr_inval;
1867 }
1868
1869 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
1870 {
1871         struct nfsd4_session *session = cstate->session;
1872         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1873
1874         spin_lock(&nn->client_lock);
1875         session->se_cb_prog = bc->bc_cb_program;
1876         session->se_cb_sec = bc->bc_cb_sec;
1877         spin_unlock(&nn->client_lock);
1878
1879         nfsd4_probe_callback(session->se_client);
1880
1881         return nfs_ok;
1882 }
1883
1884 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1885                      struct nfsd4_compound_state *cstate,
1886                      struct nfsd4_bind_conn_to_session *bcts)
1887 {
1888         __be32 status;
1889         struct nfsd4_conn *conn;
1890         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1891
1892         if (!nfsd4_last_compound_op(rqstp))
1893                 return nfserr_not_only_op;
1894         spin_lock(&nn->client_lock);
1895         cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid, SVC_NET(rqstp));
1896         /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1897          * client_lock iself: */
1898         if (cstate->session) {
1899                 nfsd4_get_session(cstate->session);
1900                 atomic_inc(&cstate->session->se_client->cl_refcount);
1901         }
1902         spin_unlock(&nn->client_lock);
1903         if (!cstate->session)
1904                 return nfserr_badsession;
1905
1906         status = nfsd4_map_bcts_dir(&bcts->dir);
1907         if (status)
1908                 return status;
1909         conn = alloc_conn(rqstp, bcts->dir);
1910         if (!conn)
1911                 return nfserr_jukebox;
1912         nfsd4_init_conn(rqstp, conn, cstate->session);
1913         return nfs_ok;
1914 }
1915
1916 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1917 {
1918         if (!session)
1919                 return 0;
1920         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1921 }
1922
1923 __be32
1924 nfsd4_destroy_session(struct svc_rqst *r,
1925                       struct nfsd4_compound_state *cstate,
1926                       struct nfsd4_destroy_session *sessionid)
1927 {
1928         struct nfsd4_session *ses;
1929         __be32 status = nfserr_badsession;
1930         struct nfsd_net *nn = net_generic(SVC_NET(r), nfsd_net_id);
1931
1932         /* Notes:
1933          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1934          * - Should we return nfserr_back_chan_busy if waiting for
1935          *   callbacks on to-be-destroyed session?
1936          * - Do we need to clear any callback info from previous session?
1937          */
1938
1939         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1940                 if (!nfsd4_last_compound_op(r))
1941                         return nfserr_not_only_op;
1942         }
1943         dump_sessionid(__func__, &sessionid->sessionid);
1944         spin_lock(&nn->client_lock);
1945         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, SVC_NET(r));
1946         if (!ses) {
1947                 spin_unlock(&nn->client_lock);
1948                 goto out;
1949         }
1950
1951         unhash_session(ses);
1952         spin_unlock(&nn->client_lock);
1953
1954         nfs4_lock_state();
1955         nfsd4_probe_callback_sync(ses->se_client);
1956         nfs4_unlock_state();
1957
1958         spin_lock(&nn->client_lock);
1959         nfsd4_del_conns(ses);
1960         nfsd4_put_session_locked(ses);
1961         spin_unlock(&nn->client_lock);
1962         status = nfs_ok;
1963 out:
1964         return status;
1965 }
1966
1967 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1968 {
1969         struct nfsd4_conn *c;
1970
1971         list_for_each_entry(c, &s->se_conns, cn_persession) {
1972                 if (c->cn_xprt == xpt) {
1973                         return c;
1974                 }
1975         }
1976         return NULL;
1977 }
1978
1979 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1980 {
1981         struct nfs4_client *clp = ses->se_client;
1982         struct nfsd4_conn *c;
1983         int ret;
1984
1985         spin_lock(&clp->cl_lock);
1986         c = __nfsd4_find_conn(new->cn_xprt, ses);
1987         if (c) {
1988                 spin_unlock(&clp->cl_lock);
1989                 free_conn(new);
1990                 return;
1991         }
1992         __nfsd4_hash_conn(new, ses);
1993         spin_unlock(&clp->cl_lock);
1994         ret = nfsd4_register_conn(new);
1995         if (ret)
1996                 /* oops; xprt is already down: */
1997                 nfsd4_conn_lost(&new->cn_xpt_user);
1998         return;
1999 }
2000
2001 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2002 {
2003         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2004
2005         return args->opcnt > session->se_fchannel.maxops;
2006 }
2007
2008 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2009                                   struct nfsd4_session *session)
2010 {
2011         struct xdr_buf *xb = &rqstp->rq_arg;
2012
2013         return xb->len > session->se_fchannel.maxreq_sz;
2014 }
2015
2016 __be32
2017 nfsd4_sequence(struct svc_rqst *rqstp,
2018                struct nfsd4_compound_state *cstate,
2019                struct nfsd4_sequence *seq)
2020 {
2021         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2022         struct nfsd4_session *session;
2023         struct nfsd4_slot *slot;
2024         struct nfsd4_conn *conn;
2025         __be32 status;
2026         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2027
2028         if (resp->opcnt != 1)
2029                 return nfserr_sequence_pos;
2030
2031         /*
2032          * Will be either used or freed by nfsd4_sequence_check_conn
2033          * below.
2034          */
2035         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2036         if (!conn)
2037                 return nfserr_jukebox;
2038
2039         spin_lock(&nn->client_lock);
2040         status = nfserr_badsession;
2041         session = find_in_sessionid_hashtbl(&seq->sessionid, SVC_NET(rqstp));
2042         if (!session)
2043                 goto out;
2044
2045         status = nfserr_too_many_ops;
2046         if (nfsd4_session_too_many_ops(rqstp, session))
2047                 goto out;
2048
2049         status = nfserr_req_too_big;
2050         if (nfsd4_request_too_big(rqstp, session))
2051                 goto out;
2052
2053         status = nfserr_badslot;
2054         if (seq->slotid >= session->se_fchannel.maxreqs)
2055                 goto out;
2056
2057         slot = session->se_slots[seq->slotid];
2058         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2059
2060         /* We do not negotiate the number of slots yet, so set the
2061          * maxslots to the session maxreqs which is used to encode
2062          * sr_highest_slotid and the sr_target_slot id to maxslots */
2063         seq->maxslots = session->se_fchannel.maxreqs;
2064
2065         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2066                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2067         if (status == nfserr_replay_cache) {
2068                 status = nfserr_seq_misordered;
2069                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2070                         goto out;
2071                 cstate->slot = slot;
2072                 cstate->session = session;
2073                 /* Return the cached reply status and set cstate->status
2074                  * for nfsd4_proc_compound processing */
2075                 status = nfsd4_replay_cache_entry(resp, seq);
2076                 cstate->status = nfserr_replay_cache;
2077                 goto out;
2078         }
2079         if (status)
2080                 goto out;
2081
2082         nfsd4_sequence_check_conn(conn, session);
2083         conn = NULL;
2084
2085         /* Success! bump slot seqid */
2086         slot->sl_seqid = seq->seqid;
2087         slot->sl_flags |= NFSD4_SLOT_INUSE;
2088         if (seq->cachethis)
2089                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2090         else
2091                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2092
2093         cstate->slot = slot;
2094         cstate->session = session;
2095
2096 out:
2097         /* Hold a session reference until done processing the compound. */
2098         if (cstate->session) {
2099                 struct nfs4_client *clp = session->se_client;
2100
2101                 nfsd4_get_session(cstate->session);
2102                 atomic_inc(&clp->cl_refcount);
2103                 switch (clp->cl_cb_state) {
2104                 case NFSD4_CB_DOWN:
2105                         seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2106                         break;
2107                 case NFSD4_CB_FAULT:
2108                         seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2109                         break;
2110                 default:
2111                         seq->status_flags = 0;
2112                 }
2113         }
2114         kfree(conn);
2115         spin_unlock(&nn->client_lock);
2116         return status;
2117 }
2118
2119 __be32
2120 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2121 {
2122         struct nfs4_client *conf, *unconf, *clp;
2123         __be32 status = 0;
2124         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2125
2126         nfs4_lock_state();
2127         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
2128         conf = find_confirmed_client(&dc->clientid, true, nn);
2129         WARN_ON_ONCE(conf && unconf);
2130
2131         if (conf) {
2132                 clp = conf;
2133
2134                 if (!is_client_expired(conf) && client_has_state(conf)) {
2135                         status = nfserr_clientid_busy;
2136                         goto out;
2137                 }
2138
2139                 /* rfc5661 18.50.3 */
2140                 if (cstate->session && conf == cstate->session->se_client) {
2141                         status = nfserr_clientid_busy;
2142                         goto out;
2143                 }
2144         } else if (unconf)
2145                 clp = unconf;
2146         else {
2147                 status = nfserr_stale_clientid;
2148                 goto out;
2149         }
2150
2151         expire_client(clp);
2152 out:
2153         nfs4_unlock_state();
2154         return status;
2155 }
2156
2157 __be32
2158 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2159 {
2160         __be32 status = 0;
2161
2162         if (rc->rca_one_fs) {
2163                 if (!cstate->current_fh.fh_dentry)
2164                         return nfserr_nofilehandle;
2165                 /*
2166                  * We don't take advantage of the rca_one_fs case.
2167                  * That's OK, it's optional, we can safely ignore it.
2168                  */
2169                  return nfs_ok;
2170         }
2171
2172         nfs4_lock_state();
2173         status = nfserr_complete_already;
2174         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2175                              &cstate->session->se_client->cl_flags))
2176                 goto out;
2177
2178         status = nfserr_stale_clientid;
2179         if (is_client_expired(cstate->session->se_client))
2180                 /*
2181                  * The following error isn't really legal.
2182                  * But we only get here if the client just explicitly
2183                  * destroyed the client.  Surely it no longer cares what
2184                  * error it gets back on an operation for the dead
2185                  * client.
2186                  */
2187                 goto out;
2188
2189         status = nfs_ok;
2190         nfsd4_client_record_create(cstate->session->se_client);
2191 out:
2192         nfs4_unlock_state();
2193         return status;
2194 }
2195
2196 __be32
2197 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2198                   struct nfsd4_setclientid *setclid)
2199 {
2200         struct xdr_netobj       clname = setclid->se_name;
2201         nfs4_verifier           clverifier = setclid->se_verf;
2202         struct nfs4_client      *conf, *unconf, *new;
2203         __be32                  status;
2204         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2205
2206         /* Cases below refer to rfc 3530 section 14.2.33: */
2207         nfs4_lock_state();
2208         conf = find_confirmed_client_by_name(&clname, nn);
2209         if (conf) {
2210                 /* case 0: */
2211                 status = nfserr_clid_inuse;
2212                 if (clp_used_exchangeid(conf))
2213                         goto out;
2214                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2215                         char addr_str[INET6_ADDRSTRLEN];
2216                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2217                                  sizeof(addr_str));
2218                         dprintk("NFSD: setclientid: string in use by client "
2219                                 "at %s\n", addr_str);
2220                         goto out;
2221                 }
2222         }
2223         unconf = find_unconfirmed_client_by_name(&clname, nn);
2224         if (unconf)
2225                 expire_client(unconf);
2226         status = nfserr_jukebox;
2227         new = create_client(clname, rqstp, &clverifier);
2228         if (new == NULL)
2229                 goto out;
2230         if (conf && same_verf(&conf->cl_verifier, &clverifier))
2231                 /* case 1: probable callback update */
2232                 copy_clid(new, conf);
2233         else /* case 4 (new client) or cases 2, 3 (client reboot): */
2234                 gen_clid(new, nn);
2235         new->cl_minorversion = 0;
2236         gen_callback(new, setclid, rqstp);
2237         add_to_unconfirmed(new);
2238         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2239         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2240         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2241         status = nfs_ok;
2242 out:
2243         nfs4_unlock_state();
2244         return status;
2245 }
2246
2247
2248 __be32
2249 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2250                          struct nfsd4_compound_state *cstate,
2251                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2252 {
2253         struct nfs4_client *conf, *unconf;
2254         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2255         clientid_t * clid = &setclientid_confirm->sc_clientid;
2256         __be32 status;
2257         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2258
2259         if (STALE_CLIENTID(clid, nn))
2260                 return nfserr_stale_clientid;
2261         nfs4_lock_state();
2262
2263         conf = find_confirmed_client(clid, false, nn);
2264         unconf = find_unconfirmed_client(clid, false, nn);
2265         /*
2266          * We try hard to give out unique clientid's, so if we get an
2267          * attempt to confirm the same clientid with a different cred,
2268          * there's a bug somewhere.  Let's charitably assume it's our
2269          * bug.
2270          */
2271         status = nfserr_serverfault;
2272         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
2273                 goto out;
2274         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
2275                 goto out;
2276         /* cases below refer to rfc 3530 section 14.2.34: */
2277         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
2278                 if (conf && !unconf) /* case 2: probable retransmit */
2279                         status = nfs_ok;
2280                 else /* case 4: client hasn't noticed we rebooted yet? */
2281                         status = nfserr_stale_clientid;
2282                 goto out;
2283         }
2284         status = nfs_ok;
2285         if (conf) { /* case 1: callback update */
2286                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2287                 nfsd4_probe_callback(conf);
2288                 expire_client(unconf);
2289         } else { /* case 3: normal case; new or rebooted client */
2290                 conf = find_confirmed_client_by_name(&unconf->cl_name, nn);
2291                 if (conf)
2292                         expire_client(conf);
2293                 move_to_confirmed(unconf);
2294                 nfsd4_probe_callback(unconf);
2295         }
2296 out:
2297         nfs4_unlock_state();
2298         return status;
2299 }
2300
2301 static struct nfs4_file *nfsd4_alloc_file(void)
2302 {
2303         return kmem_cache_alloc(file_slab, GFP_KERNEL);
2304 }
2305
2306 /* OPEN Share state helper functions */
2307 static void nfsd4_init_file(struct nfs4_file *fp, struct inode *ino)
2308 {
2309         unsigned int hashval = file_hashval(ino);
2310
2311         atomic_set(&fp->fi_ref, 1);
2312         INIT_LIST_HEAD(&fp->fi_hash);
2313         INIT_LIST_HEAD(&fp->fi_stateids);
2314         INIT_LIST_HEAD(&fp->fi_delegations);
2315         fp->fi_inode = igrab(ino);
2316         fp->fi_had_conflict = false;
2317         fp->fi_lease = NULL;
2318         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2319         memset(fp->fi_access, 0, sizeof(fp->fi_access));
2320         spin_lock(&recall_lock);
2321         list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2322         spin_unlock(&recall_lock);
2323 }
2324
2325 static void
2326 nfsd4_free_slab(struct kmem_cache **slab)
2327 {
2328         if (*slab == NULL)
2329                 return;
2330         kmem_cache_destroy(*slab);
2331         *slab = NULL;
2332 }
2333
2334 void
2335 nfsd4_free_slabs(void)
2336 {
2337         nfsd4_free_slab(&openowner_slab);
2338         nfsd4_free_slab(&lockowner_slab);
2339         nfsd4_free_slab(&file_slab);
2340         nfsd4_free_slab(&stateid_slab);
2341         nfsd4_free_slab(&deleg_slab);
2342 }
2343
2344 int
2345 nfsd4_init_slabs(void)
2346 {
2347         openowner_slab = kmem_cache_create("nfsd4_openowners",
2348                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2349         if (openowner_slab == NULL)
2350                 goto out_nomem;
2351         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2352                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
2353         if (lockowner_slab == NULL)
2354                 goto out_nomem;
2355         file_slab = kmem_cache_create("nfsd4_files",
2356                         sizeof(struct nfs4_file), 0, 0, NULL);
2357         if (file_slab == NULL)
2358                 goto out_nomem;
2359         stateid_slab = kmem_cache_create("nfsd4_stateids",
2360                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2361         if (stateid_slab == NULL)
2362                 goto out_nomem;
2363         deleg_slab = kmem_cache_create("nfsd4_delegations",
2364                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2365         if (deleg_slab == NULL)
2366                 goto out_nomem;
2367         return 0;
2368 out_nomem:
2369         nfsd4_free_slabs();
2370         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2371         return -ENOMEM;
2372 }
2373
2374 void nfs4_free_openowner(struct nfs4_openowner *oo)
2375 {
2376         kfree(oo->oo_owner.so_owner.data);
2377         kmem_cache_free(openowner_slab, oo);
2378 }
2379
2380 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2381 {
2382         kfree(lo->lo_owner.so_owner.data);
2383         kmem_cache_free(lockowner_slab, lo);
2384 }
2385
2386 static void init_nfs4_replay(struct nfs4_replay *rp)
2387 {
2388         rp->rp_status = nfserr_serverfault;
2389         rp->rp_buflen = 0;
2390         rp->rp_buf = rp->rp_ibuf;
2391 }
2392
2393 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2394 {
2395         struct nfs4_stateowner *sop;
2396
2397         sop = kmem_cache_alloc(slab, GFP_KERNEL);
2398         if (!sop)
2399                 return NULL;
2400
2401         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2402         if (!sop->so_owner.data) {
2403                 kmem_cache_free(slab, sop);
2404                 return NULL;
2405         }
2406         sop->so_owner.len = owner->len;
2407
2408         INIT_LIST_HEAD(&sop->so_stateids);
2409         sop->so_client = clp;
2410         init_nfs4_replay(&sop->so_replay);
2411         return sop;
2412 }
2413
2414 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2415 {
2416         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2417
2418         list_add(&oo->oo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
2419         list_add(&oo->oo_perclient, &clp->cl_openowners);
2420 }
2421
2422 static struct nfs4_openowner *
2423 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2424         struct nfs4_openowner *oo;
2425
2426         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2427         if (!oo)
2428                 return NULL;
2429         oo->oo_owner.so_is_open_owner = 1;
2430         oo->oo_owner.so_seqid = open->op_seqid;
2431         oo->oo_flags = NFS4_OO_NEW;
2432         oo->oo_time = 0;
2433         oo->oo_last_closed_stid = NULL;
2434         INIT_LIST_HEAD(&oo->oo_close_lru);
2435         hash_openowner(oo, clp, strhashval);
2436         return oo;
2437 }
2438
2439 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2440         struct nfs4_openowner *oo = open->op_openowner;
2441
2442         stp->st_stid.sc_type = NFS4_OPEN_STID;
2443         INIT_LIST_HEAD(&stp->st_lockowners);
2444         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2445         list_add(&stp->st_perfile, &fp->fi_stateids);
2446         stp->st_stateowner = &oo->oo_owner;
2447         get_nfs4_file(fp);
2448         stp->st_file = fp;
2449         stp->st_access_bmap = 0;
2450         stp->st_deny_bmap = 0;
2451         set_access(open->op_share_access, stp);
2452         set_deny(open->op_share_deny, stp);
2453         stp->st_openstp = NULL;
2454 }
2455
2456 static void
2457 move_to_close_lru(struct nfs4_openowner *oo, struct net *net)
2458 {
2459         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2460
2461         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2462
2463         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
2464         oo->oo_time = get_seconds();
2465 }
2466
2467 static int
2468 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2469                                                         clientid_t *clid)
2470 {
2471         return (sop->so_owner.len == owner->len) &&
2472                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2473                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2474 }
2475
2476 static struct nfs4_openowner *
2477 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
2478                         bool sessions, struct nfsd_net *nn)
2479 {
2480         struct nfs4_stateowner *so;
2481         struct nfs4_openowner *oo;
2482         struct nfs4_client *clp;
2483
2484         list_for_each_entry(so, &nn->ownerstr_hashtbl[hashval], so_strhash) {
2485                 if (!so->so_is_open_owner)
2486                         continue;
2487                 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2488                         oo = openowner(so);
2489                         clp = oo->oo_owner.so_client;
2490                         if ((bool)clp->cl_minorversion != sessions)
2491                                 return NULL;
2492                         renew_client(oo->oo_owner.so_client);
2493                         return oo;
2494                 }
2495         }
2496         return NULL;
2497 }
2498
2499 /* search file_hashtbl[] for file */
2500 static struct nfs4_file *
2501 find_file(struct inode *ino)
2502 {
2503         unsigned int hashval = file_hashval(ino);
2504         struct nfs4_file *fp;
2505
2506         spin_lock(&recall_lock);
2507         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2508                 if (fp->fi_inode == ino) {
2509                         get_nfs4_file(fp);
2510                         spin_unlock(&recall_lock);
2511                         return fp;
2512                 }
2513         }
2514         spin_unlock(&recall_lock);
2515         return NULL;
2516 }
2517
2518 /*
2519  * Called to check deny when READ with all zero stateid or
2520  * WRITE with all zero or all one stateid
2521  */
2522 static __be32
2523 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2524 {
2525         struct inode *ino = current_fh->fh_dentry->d_inode;
2526         struct nfs4_file *fp;
2527         struct nfs4_ol_stateid *stp;
2528         __be32 ret;
2529
2530         fp = find_file(ino);
2531         if (!fp)
2532                 return nfs_ok;
2533         ret = nfserr_locked;
2534         /* Search for conflicting share reservations */
2535         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2536                 if (test_deny(deny_type, stp) ||
2537                     test_deny(NFS4_SHARE_DENY_BOTH, stp))
2538                         goto out;
2539         }
2540         ret = nfs_ok;
2541 out:
2542         put_nfs4_file(fp);
2543         return ret;
2544 }
2545
2546 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2547 {
2548         /* We're assuming the state code never drops its reference
2549          * without first removing the lease.  Since we're in this lease
2550          * callback (and since the lease code is serialized by the kernel
2551          * lock) we know the server hasn't removed the lease yet, we know
2552          * it's safe to take a reference: */
2553         atomic_inc(&dp->dl_count);
2554
2555         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2556
2557         /* only place dl_time is set. protected by lock_flocks*/
2558         dp->dl_time = get_seconds();
2559
2560         nfsd4_cb_recall(dp);
2561 }
2562
2563 /* Called from break_lease() with lock_flocks() held. */
2564 static void nfsd_break_deleg_cb(struct file_lock *fl)
2565 {
2566         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2567         struct nfs4_delegation *dp;
2568
2569         if (!fp) {
2570                 WARN(1, "(%p)->fl_owner NULL\n", fl);
2571                 return;
2572         }
2573         if (fp->fi_had_conflict) {
2574                 WARN(1, "duplicate break on %p\n", fp);
2575                 return;
2576         }
2577         /*
2578          * We don't want the locks code to timeout the lease for us;
2579          * we'll remove it ourself if a delegation isn't returned
2580          * in time:
2581          */
2582         fl->fl_break_time = 0;
2583
2584         spin_lock(&recall_lock);
2585         fp->fi_had_conflict = true;
2586         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2587                 nfsd_break_one_deleg(dp);
2588         spin_unlock(&recall_lock);
2589 }
2590
2591 static
2592 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2593 {
2594         if (arg & F_UNLCK)
2595                 return lease_modify(onlist, arg);
2596         else
2597                 return -EAGAIN;
2598 }
2599
2600 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2601         .lm_break = nfsd_break_deleg_cb,
2602         .lm_change = nfsd_change_deleg_cb,
2603 };
2604
2605 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2606 {
2607         if (nfsd4_has_session(cstate))
2608                 return nfs_ok;
2609         if (seqid == so->so_seqid - 1)
2610                 return nfserr_replay_me;
2611         if (seqid == so->so_seqid)
2612                 return nfs_ok;
2613         return nfserr_bad_seqid;
2614 }
2615
2616 __be32
2617 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2618                     struct nfsd4_open *open, struct nfsd_net *nn)
2619 {
2620         clientid_t *clientid = &open->op_clientid;
2621         struct nfs4_client *clp = NULL;
2622         unsigned int strhashval;
2623         struct nfs4_openowner *oo = NULL;
2624         __be32 status;
2625
2626         if (STALE_CLIENTID(&open->op_clientid, nn))
2627                 return nfserr_stale_clientid;
2628         /*
2629          * In case we need it later, after we've already created the
2630          * file and don't want to risk a further failure:
2631          */
2632         open->op_file = nfsd4_alloc_file();
2633         if (open->op_file == NULL)
2634                 return nfserr_jukebox;
2635
2636         strhashval = ownerstr_hashval(clientid->cl_id, &open->op_owner);
2637         oo = find_openstateowner_str(strhashval, open, cstate->minorversion, nn);
2638         open->op_openowner = oo;
2639         if (!oo) {
2640                 clp = find_confirmed_client(clientid, cstate->minorversion,
2641                                             nn);
2642                 if (clp == NULL)
2643                         return nfserr_expired;
2644                 goto new_owner;
2645         }
2646         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2647                 /* Replace unconfirmed owners without checking for replay. */
2648                 clp = oo->oo_owner.so_client;
2649                 release_openowner(oo);
2650                 open->op_openowner = NULL;
2651                 goto new_owner;
2652         }
2653         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2654         if (status)
2655                 return status;
2656         clp = oo->oo_owner.so_client;
2657         goto alloc_stateid;
2658 new_owner:
2659         oo = alloc_init_open_stateowner(strhashval, clp, open);
2660         if (oo == NULL)
2661                 return nfserr_jukebox;
2662         open->op_openowner = oo;
2663 alloc_stateid:
2664         open->op_stp = nfs4_alloc_stateid(clp);
2665         if (!open->op_stp)
2666                 return nfserr_jukebox;
2667         return nfs_ok;
2668 }
2669
2670 static inline __be32
2671 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2672 {
2673         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2674                 return nfserr_openmode;
2675         else
2676                 return nfs_ok;
2677 }
2678
2679 static int share_access_to_flags(u32 share_access)
2680 {
2681         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2682 }
2683
2684 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2685 {
2686         struct nfs4_stid *ret;
2687
2688         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2689         if (!ret)
2690                 return NULL;
2691         return delegstateid(ret);
2692 }
2693
2694 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
2695 {
2696         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
2697                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
2698 }
2699
2700 static __be32
2701 nfs4_check_deleg(struct nfs4_client *cl, struct nfs4_file *fp, struct nfsd4_open *open,
2702                 struct nfs4_delegation **dp)
2703 {
2704         int flags;
2705         __be32 status = nfserr_bad_stateid;
2706
2707         *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2708         if (*dp == NULL)
2709                 goto out;
2710         flags = share_access_to_flags(open->op_share_access);
2711         status = nfs4_check_delegmode(*dp, flags);
2712         if (status)
2713                 *dp = NULL;
2714 out:
2715         if (!nfsd4_is_deleg_cur(open))
2716                 return nfs_ok;
2717         if (status)
2718                 return status;
2719         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2720         return nfs_ok;
2721 }
2722
2723 static __be32
2724 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2725 {
2726         struct nfs4_ol_stateid *local;
2727         struct nfs4_openowner *oo = open->op_openowner;
2728
2729         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2730                 /* ignore lock owners */
2731                 if (local->st_stateowner->so_is_open_owner == 0)
2732                         continue;
2733                 /* remember if we have seen this open owner */
2734                 if (local->st_stateowner == &oo->oo_owner)
2735                         *stpp = local;
2736                 /* check for conflicting share reservations */
2737                 if (!test_share(local, open))
2738                         return nfserr_share_denied;
2739         }
2740         return nfs_ok;
2741 }
2742
2743 static inline int nfs4_access_to_access(u32 nfs4_access)
2744 {
2745         int flags = 0;
2746
2747         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2748                 flags |= NFSD_MAY_READ;
2749         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2750                 flags |= NFSD_MAY_WRITE;
2751         return flags;
2752 }
2753
2754 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2755                 struct svc_fh *cur_fh, struct nfsd4_open *open)
2756 {
2757         __be32 status;
2758         int oflag = nfs4_access_to_omode(open->op_share_access);
2759         int access = nfs4_access_to_access(open->op_share_access);
2760
2761         if (!fp->fi_fds[oflag]) {
2762                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2763                         &fp->fi_fds[oflag]);
2764                 if (status)
2765                         return status;
2766         }
2767         nfs4_file_get_access(fp, oflag);
2768
2769         return nfs_ok;
2770 }
2771
2772 static inline __be32
2773 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2774                 struct nfsd4_open *open)
2775 {
2776         struct iattr iattr = {
2777                 .ia_valid = ATTR_SIZE,
2778                 .ia_size = 0,
2779         };
2780         if (!open->op_truncate)
2781                 return 0;
2782         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2783                 return nfserr_inval;
2784         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2785 }
2786
2787 static __be32
2788 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)
2789 {
2790         u32 op_share_access = open->op_share_access;
2791         bool new_access;
2792         __be32 status;
2793
2794         new_access = !test_access(op_share_access, stp);
2795         if (new_access) {
2796                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2797                 if (status)
2798                         return status;
2799         }
2800         status = nfsd4_truncate(rqstp, cur_fh, open);
2801         if (status) {
2802                 if (new_access) {
2803                         int oflag = nfs4_access_to_omode(op_share_access);
2804                         nfs4_file_put_access(fp, oflag);
2805                 }
2806                 return status;
2807         }
2808         /* remember the open */
2809         set_access(op_share_access, stp);
2810         set_deny(open->op_share_deny, stp);
2811
2812         return nfs_ok;
2813 }
2814
2815
2816 static void
2817 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
2818 {
2819         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2820 }
2821
2822 /* Should we give out recallable state?: */
2823 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2824 {
2825         if (clp->cl_cb_state == NFSD4_CB_UP)
2826                 return true;
2827         /*
2828          * In the sessions case, since we don't have to establish a
2829          * separate connection for callbacks, we assume it's OK
2830          * until we hear otherwise:
2831          */
2832         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2833 }
2834
2835 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2836 {
2837         struct file_lock *fl;
2838
2839         fl = locks_alloc_lock();
2840         if (!fl)
2841                 return NULL;
2842         locks_init_lock(fl);
2843         fl->fl_lmops = &nfsd_lease_mng_ops;
2844         fl->fl_flags = FL_LEASE;
2845         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2846         fl->fl_end = OFFSET_MAX;
2847         fl->fl_owner = (fl_owner_t)(dp->dl_file);
2848         fl->fl_pid = current->tgid;
2849         return fl;
2850 }
2851
2852 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2853 {
2854         struct nfs4_file *fp = dp->dl_file;
2855         struct file_lock *fl;
2856         int status;
2857
2858         fl = nfs4_alloc_init_lease(dp, flag);
2859         if (!fl)
2860                 return -ENOMEM;
2861         fl->fl_file = find_readable_file(fp);
2862         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2863         status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2864         if (status) {
2865                 list_del_init(&dp->dl_perclnt);
2866                 locks_free_lock(fl);
2867                 return -ENOMEM;
2868         }
2869         fp->fi_lease = fl;
2870         fp->fi_deleg_file = get_file(fl->fl_file);
2871         atomic_set(&fp->fi_delegees, 1);
2872         list_add(&dp->dl_perfile, &fp->fi_delegations);
2873         return 0;
2874 }
2875
2876 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2877 {
2878         struct nfs4_file *fp = dp->dl_file;
2879
2880         if (!fp->fi_lease)
2881                 return nfs4_setlease(dp, flag);
2882         spin_lock(&recall_lock);
2883         if (fp->fi_had_conflict) {
2884                 spin_unlock(&recall_lock);
2885                 return -EAGAIN;
2886         }
2887         atomic_inc(&fp->fi_delegees);
2888         list_add(&dp->dl_perfile, &fp->fi_delegations);
2889         spin_unlock(&recall_lock);
2890         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2891         return 0;
2892 }
2893
2894 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
2895 {
2896         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2897         if (status == -EAGAIN)
2898                 open->op_why_no_deleg = WND4_CONTENTION;
2899         else {
2900                 open->op_why_no_deleg = WND4_RESOURCE;
2901                 switch (open->op_deleg_want) {
2902                 case NFS4_SHARE_WANT_READ_DELEG:
2903                 case NFS4_SHARE_WANT_WRITE_DELEG:
2904                 case NFS4_SHARE_WANT_ANY_DELEG:
2905                         break;
2906                 case NFS4_SHARE_WANT_CANCEL:
2907                         open->op_why_no_deleg = WND4_CANCELLED;
2908                         break;
2909                 case NFS4_SHARE_WANT_NO_DELEG:
2910                         WARN_ON_ONCE(1);
2911                 }
2912         }
2913 }
2914
2915 /*
2916  * Attempt to hand out a delegation.
2917  */
2918 static void
2919 nfs4_open_delegation(struct net *net, struct svc_fh *fh,
2920                      struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
2921 {
2922         struct nfs4_delegation *dp;
2923         struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
2924         int cb_up;
2925         int status = 0, flag = 0;
2926
2927         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
2928         flag = NFS4_OPEN_DELEGATE_NONE;
2929         open->op_recall = 0;
2930         switch (open->op_claim_type) {
2931                 case NFS4_OPEN_CLAIM_PREVIOUS:
2932                         if (!cb_up)
2933                                 open->op_recall = 1;
2934                         flag = open->op_delegate_type;
2935                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2936                                 goto out;
2937                         break;
2938                 case NFS4_OPEN_CLAIM_NULL:
2939                         /* Let's not give out any delegations till everyone's
2940                          * had the chance to reclaim theirs.... */
2941                         if (locks_in_grace(net))
2942                                 goto out;
2943                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
2944                                 goto out;
2945                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2946                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2947                         else
2948                                 flag = NFS4_OPEN_DELEGATE_READ;
2949                         break;
2950                 default:
2951                         goto out;
2952         }
2953
2954         dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh, flag);
2955         if (dp == NULL)
2956                 goto out_no_deleg;
2957         status = nfs4_set_delegation(dp, flag);
2958         if (status)
2959                 goto out_free;
2960
2961         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
2962
2963         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2964                 STATEID_VAL(&dp->dl_stid.sc_stateid));
2965 out:
2966         open->op_delegate_type = flag;
2967         if (flag == NFS4_OPEN_DELEGATE_NONE) {
2968                 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
2969                     open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2970                         dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2971
2972                 /* 4.1 client asking for a delegation? */
2973                 if (open->op_deleg_want)
2974                         nfsd4_open_deleg_none_ext(open, status);
2975         }
2976         return;
2977 out_free:
2978         unhash_stid(&dp->dl_stid);
2979         nfs4_put_delegation(dp);
2980 out_no_deleg:
2981         flag = NFS4_OPEN_DELEGATE_NONE;
2982         goto out;
2983 }
2984
2985 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
2986                                         struct nfs4_delegation *dp)
2987 {
2988         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
2989             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
2990                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2991                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
2992         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
2993                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
2994                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2995                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
2996         }
2997         /* Otherwise the client must be confused wanting a delegation
2998          * it already has, therefore we don't return
2999          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
3000          */
3001 }
3002
3003 /*
3004  * called with nfs4_lock_state() held.
3005  */
3006 __be32
3007 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
3008 {
3009         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3010         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
3011         struct nfs4_file *fp = NULL;
3012         struct inode *ino = current_fh->fh_dentry->d_inode;
3013         struct nfs4_ol_stateid *stp = NULL;
3014         struct nfs4_delegation *dp = NULL;
3015         __be32 status;
3016
3017         /*
3018          * Lookup file; if found, lookup stateid and check open request,
3019          * and check for delegations in the process of being recalled.
3020          * If not found, create the nfs4_file struct
3021          */
3022         fp = find_file(ino);
3023         if (fp) {
3024                 if ((status = nfs4_check_open(fp, open, &stp)))
3025                         goto out;
3026                 status = nfs4_check_deleg(cl, fp, open, &dp);
3027                 if (status)
3028                         goto out;
3029         } else {
3030                 status = nfserr_bad_stateid;
3031                 if (nfsd4_is_deleg_cur(open))
3032                         goto out;
3033                 status = nfserr_jukebox;
3034                 fp = open->op_file;
3035                 open->op_file = NULL;
3036                 nfsd4_init_file(fp, ino);
3037         }
3038
3039         /*
3040          * OPEN the file, or upgrade an existing OPEN.
3041          * If truncate fails, the OPEN fails.
3042          */
3043         if (stp) {
3044                 /* Stateid was found, this is an OPEN upgrade */
3045                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
3046                 if (status)
3047                         goto out;
3048         } else {
3049                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, open);
3050                 if (status)
3051                         goto out;
3052                 status = nfsd4_truncate(rqstp, current_fh, open);
3053                 if (status)
3054                         goto out;
3055                 stp = open->op_stp;
3056                 open->op_stp = NULL;
3057                 init_open_stateid(stp, fp, open);
3058         }
3059         update_stateid(&stp->st_stid.sc_stateid);
3060         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3061
3062         if (nfsd4_has_session(&resp->cstate)) {
3063                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3064
3065                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
3066                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3067                         open->op_why_no_deleg = WND4_NOT_WANTED;
3068                         goto nodeleg;
3069                 }
3070         }
3071
3072         /*
3073         * Attempt to hand out a delegation. No error return, because the
3074         * OPEN succeeds even if we fail.
3075         */
3076         nfs4_open_delegation(SVC_NET(rqstp), current_fh, open, stp);
3077 nodeleg:
3078         status = nfs_ok;
3079
3080         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
3081                 STATEID_VAL(&stp->st_stid.sc_stateid));
3082 out:
3083         /* 4.1 client trying to upgrade/downgrade delegation? */
3084         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
3085             open->op_deleg_want)
3086                 nfsd4_deleg_xgrade_none_ext(open, dp);
3087
3088         if (fp)
3089                 put_nfs4_file(fp);
3090         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
3091                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
3092         /*
3093         * To finish the open response, we just need to set the rflags.
3094         */
3095         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
3096         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
3097             !nfsd4_has_session(&resp->cstate))
3098                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
3099
3100         return status;
3101 }
3102
3103 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
3104 {
3105         if (open->op_openowner) {
3106                 struct nfs4_openowner *oo = open->op_openowner;
3107
3108                 if (!list_empty(&oo->oo_owner.so_stateids))
3109                         list_del_init(&oo->oo_close_lru);
3110                 if (oo->oo_flags & NFS4_OO_NEW) {
3111                         if (status) {
3112                                 release_openowner(oo);
3113                                 open->op_openowner = NULL;
3114                         } else
3115                                 oo->oo_flags &= ~NFS4_OO_NEW;
3116                 }
3117         }
3118         if (open->op_file)
3119                 nfsd4_free_file(open->op_file);
3120         if (open->op_stp)
3121                 free_generic_stateid(open->op_stp);
3122 }
3123
3124 static __be32 lookup_clientid(clientid_t *clid, bool session, struct nfsd_net *nn, struct nfs4_client **clp)
3125 {
3126         struct nfs4_client *found;
3127
3128         if (STALE_CLIENTID(clid, nn))
3129                 return nfserr_stale_clientid;
3130         found = find_confirmed_client(clid, session, nn);
3131         if (clp)
3132                 *clp = found;
3133         return found ? nfs_ok : nfserr_expired;
3134 }
3135
3136 __be32
3137 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3138             clientid_t *clid)
3139 {
3140         struct nfs4_client *clp;
3141         __be32 status;
3142         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3143
3144         nfs4_lock_state();
3145         dprintk("process_renew(%08x/%08x): starting\n",
3146                         clid->cl_boot, clid->cl_id);
3147         status = lookup_clientid(clid, cstate->minorversion, nn, &clp);
3148         if (status)
3149                 goto out;
3150         status = nfserr_cb_path_down;
3151         if (!list_empty(&clp->cl_delegations)
3152                         && clp->cl_cb_state != NFSD4_CB_UP)
3153                 goto out;
3154         status = nfs_ok;
3155 out:
3156         nfs4_unlock_state();
3157         return status;
3158 }
3159
3160 static void
3161 nfsd4_end_grace(struct nfsd_net *nn)
3162 {
3163         /* do nothing if grace period already ended */
3164         if (nn->grace_ended)
3165                 return;
3166
3167         dprintk("NFSD: end of grace period\n");
3168         nn->grace_ended = true;
3169         nfsd4_record_grace_done(nn, nn->boot_time);
3170         locks_end_grace(&nn->nfsd4_manager);
3171         /*
3172          * Now that every NFSv4 client has had the chance to recover and
3173          * to see the (possibly new, possibly shorter) lease time, we
3174          * can safely set the next grace time to the current lease time:
3175          */
3176         nn->nfsd4_grace = nn->nfsd4_lease;
3177 }
3178
3179 static time_t
3180 nfs4_laundromat(struct nfsd_net *nn)
3181 {
3182         struct nfs4_client *clp;
3183         struct nfs4_openowner *oo;
3184         struct nfs4_delegation *dp;
3185         struct list_head *pos, *next, reaplist;
3186         time_t cutoff = get_seconds() - nn->nfsd4_lease;
3187         time_t t, clientid_val = nn->nfsd4_lease;
3188         time_t u, test_val = nn->nfsd4_lease;
3189
3190         nfs4_lock_state();
3191
3192         dprintk("NFSD: laundromat service - starting\n");
3193         nfsd4_end_grace(nn);
3194         INIT_LIST_HEAD(&reaplist);
3195         spin_lock(&nn->client_lock);
3196         list_for_each_safe(pos, next, &nn->client_lru) {
3197                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3198                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3199                         t = clp->cl_time - cutoff;
3200                         if (clientid_val > t)
3201                                 clientid_val = t;
3202                         break;
3203                 }
3204                 if (atomic_read(&clp->cl_refcount)) {
3205                         dprintk("NFSD: client in use (clientid %08x)\n",
3206                                 clp->cl_clientid.cl_id);
3207                         continue;
3208                 }
3209                 unhash_client_locked(clp);
3210                 list_add(&clp->cl_lru, &reaplist);
3211         }
3212         spin_unlock(&nn->client_lock);
3213         list_for_each_safe(pos, next, &reaplist) {
3214                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3215                 dprintk("NFSD: purging unused client (clientid %08x)\n",
3216                         clp->cl_clientid.cl_id);
3217                 expire_client(clp);
3218         }
3219         spin_lock(&recall_lock);
3220         list_for_each_safe(pos, next, &del_recall_lru) {
3221                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3222                 if (net_generic(dp->dl_stid.sc_client->net, nfsd_net_id) != nn)
3223                         continue;
3224                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3225                         u = dp->dl_time - cutoff;
3226                         if (test_val > u)
3227                                 test_val = u;
3228                         break;
3229                 }
3230                 list_move(&dp->dl_recall_lru, &reaplist);
3231         }
3232         spin_unlock(&recall_lock);
3233         list_for_each_safe(pos, next, &reaplist) {
3234                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3235                 unhash_delegation(dp);
3236         }
3237         test_val = nn->nfsd4_lease;
3238         list_for_each_safe(pos, next, &nn->close_lru) {
3239                 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3240                 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3241                         u = oo->oo_time - cutoff;
3242                         if (test_val > u)
3243                                 test_val = u;
3244                         break;
3245                 }
3246                 release_openowner(oo);
3247         }
3248         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3249                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3250         nfs4_unlock_state();
3251         return clientid_val;
3252 }
3253
3254 static struct workqueue_struct *laundry_wq;
3255 static void laundromat_main(struct work_struct *);
3256
3257 static void
3258 laundromat_main(struct work_struct *laundry)
3259 {
3260         time_t t;
3261         struct delayed_work *dwork = container_of(laundry, struct delayed_work,
3262                                                   work);
3263         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
3264                                            laundromat_work);
3265
3266         t = nfs4_laundromat(nn);
3267         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3268         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
3269 }
3270
3271 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3272 {
3273         if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3274                 return nfserr_bad_stateid;
3275         return nfs_ok;
3276 }
3277
3278 static inline int
3279 access_permit_read(struct nfs4_ol_stateid *stp)
3280 {
3281         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
3282                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
3283                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
3284 }
3285
3286 static inline int
3287 access_permit_write(struct nfs4_ol_stateid *stp)
3288 {
3289         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
3290                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
3291 }
3292
3293 static
3294 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3295 {
3296         __be32 status = nfserr_openmode;
3297
3298         /* For lock stateid's, we test the parent open, not the lock: */
3299         if (stp->st_openstp)
3300                 stp = stp->st_openstp;
3301         if ((flags & WR_STATE) && !access_permit_write(stp))
3302                 goto out;
3303         if ((flags & RD_STATE) && !access_permit_read(stp))
3304                 goto out;
3305         status = nfs_ok;
3306 out:
3307         return status;
3308 }
3309
3310 static inline __be32
3311 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
3312 {
3313         if (ONE_STATEID(stateid) && (flags & RD_STATE))
3314                 return nfs_ok;
3315         else if (locks_in_grace(net)) {
3316                 /* Answer in remaining cases depends on existence of
3317                  * conflicting state; so we must wait out the grace period. */
3318                 return nfserr_grace;
3319         } else if (flags & WR_STATE)
3320                 return nfs4_share_conflict(current_fh,
3321                                 NFS4_SHARE_DENY_WRITE);
3322         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3323                 return nfs4_share_conflict(current_fh,
3324                                 NFS4_SHARE_DENY_READ);
3325 }
3326
3327 /*
3328  * Allow READ/WRITE during grace period on recovered state only for files
3329  * that are not able to provide mandatory locking.
3330  */
3331 static inline int
3332 grace_disallows_io(struct net *net, struct inode *inode)
3333 {
3334         return locks_in_grace(net) && mandatory_lock(inode);
3335 }
3336
3337 /* Returns true iff a is later than b: */
3338 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3339 {
3340         return (s32)a->si_generation - (s32)b->si_generation > 0;
3341 }
3342
3343 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3344 {
3345         /*
3346          * When sessions are used the stateid generation number is ignored
3347          * when it is zero.
3348          */
3349         if (has_session && in->si_generation == 0)
3350                 return nfs_ok;
3351
3352         if (in->si_generation == ref->si_generation)
3353                 return nfs_ok;
3354
3355         /* If the client sends us a stateid from the future, it's buggy: */
3356         if (stateid_generation_after(in, ref))
3357                 return nfserr_bad_stateid;
3358         /*
3359          * However, we could see a stateid from the past, even from a
3360          * non-buggy client.  For example, if the client sends a lock
3361          * while some IO is outstanding, the lock may bump si_generation
3362          * while the IO is still in flight.  The client could avoid that
3363          * situation by waiting for responses on all the IO requests,
3364          * but better performance may result in retrying IO that
3365          * receives an old_stateid error if requests are rarely
3366          * reordered in flight:
3367          */
3368         return nfserr_old_stateid;
3369 }
3370
3371 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
3372 {
3373         struct nfs4_stid *s;
3374         struct nfs4_ol_stateid *ols;
3375         __be32 status;
3376
3377         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3378                 return nfserr_bad_stateid;
3379         /* Client debugging aid. */
3380         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
3381                 char addr_str[INET6_ADDRSTRLEN];
3382                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
3383                                  sizeof(addr_str));
3384                 pr_warn_ratelimited("NFSD: client %s testing state ID "
3385                                         "with incorrect client ID\n", addr_str);
3386                 return nfserr_bad_stateid;
3387         }
3388         s = find_stateid(cl, stateid);
3389         if (!s)
3390                 return nfserr_bad_stateid;
3391         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3392         if (status)
3393                 return status;
3394         if (!(s->sc_type & (NFS4_OPEN_STID | NFS4_LOCK_STID)))
3395                 return nfs_ok;
3396         ols = openlockstateid(s);
3397         if (ols->st_stateowner->so_is_open_owner
3398             && !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3399                 return nfserr_bad_stateid;
3400         return nfs_ok;
3401 }
3402
3403 static __be32 nfsd4_lookup_stateid(stateid_t *stateid, unsigned char typemask,
3404                                    struct nfs4_stid **s, bool sessions,
3405                                    struct nfsd_net *nn)
3406 {
3407         struct nfs4_client *cl;
3408         __be32 status;
3409
3410         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3411                 return nfserr_bad_stateid;
3412         status = lookup_clientid(&stateid->si_opaque.so_clid, sessions,
3413                                                         nn, &cl);
3414         if (status == nfserr_stale_clientid)
3415                 return nfserr_stale_stateid;
3416         if (status)
3417                 return status;
3418         *s = find_stateid_by_type(cl, stateid, typemask);
3419         if (!*s)
3420                 return nfserr_bad_stateid;
3421         return nfs_ok;
3422 }
3423
3424 /*
3425 * Checks for stateid operations
3426 */
3427 __be32
3428 nfs4_preprocess_stateid_op(struct net *net, struct nfsd4_compound_state *cstate,
3429                            stateid_t *stateid, int flags, struct file **filpp)
3430 {
3431         struct nfs4_stid *s;
3432         struct nfs4_ol_stateid *stp = NULL;
3433         struct nfs4_delegation *dp = NULL;
3434         struct svc_fh *current_fh = &cstate->current_fh;
3435         struct inode *ino = current_fh->fh_dentry->d_inode;
3436         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3437         __be32 status;
3438
3439         if (filpp)
3440                 *filpp = NULL;
3441
3442         if (grace_disallows_io(net, ino))
3443                 return nfserr_grace;
3444
3445         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3446                 return check_special_stateids(net, current_fh, stateid, flags);
3447
3448         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
3449                                       &s, cstate->minorversion, nn);
3450         if (status)
3451                 return status;
3452         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
3453         if (status)
3454                 goto out;
3455         switch (s->sc_type) {
3456         case NFS4_DELEG_STID:
3457                 dp = delegstateid(s);
3458                 status = nfs4_check_delegmode(dp, flags);
3459                 if (status)
3460                         goto out;
3461                 if (filpp) {
3462                         *filpp = dp->dl_file->fi_deleg_file;
3463                         if (!*filpp) {
3464                                 WARN_ON_ONCE(1);
3465                                 status = nfserr_serverfault;
3466                                 goto out;
3467                         }
3468                 }
3469                 break;
3470         case NFS4_OPEN_STID:
3471         case NFS4_LOCK_STID:
3472                 stp = openlockstateid(s);
3473                 status = nfs4_check_fh(current_fh, stp);
3474                 if (status)
3475                         goto out;
3476                 if (stp->st_stateowner->so_is_open_owner
3477                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3478                         goto out;
3479                 status = nfs4_check_openmode(stp, flags);
3480                 if (status)
3481                         goto out;
3482                 if (filpp) {
3483                         if (flags & RD_STATE)
3484                                 *filpp = find_readable_file(stp->st_file);
3485                         else
3486                                 *filpp = find_writeable_file(stp->st_file);
3487                 }
3488                 break;
3489         default:
3490                 return nfserr_bad_stateid;
3491         }
3492         status = nfs_ok;
3493 out:
3494         return status;
3495 }
3496
3497 static __be32
3498 nfsd4_free_lock_stateid(struct nfs4_ol_stateid *stp)
3499 {
3500         if (check_for_locks(stp->st_file, lockowner(stp->st_stateowner)))
3501                 return nfserr_locks_held;
3502         release_lock_stateid(stp);
3503         return nfs_ok;
3504 }
3505
3506 /*
3507  * Test if the stateid is valid
3508  */
3509 __be32
3510 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3511                    struct nfsd4_test_stateid *test_stateid)
3512 {
3513         struct nfsd4_test_stateid_id *stateid;
3514         struct nfs4_client *cl = cstate->session->se_client;
3515
3516         nfs4_lock_state();
3517         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
3518                 stateid->ts_id_status =
3519                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
3520         nfs4_unlock_state();
3521
3522         return nfs_ok;
3523 }
3524
3525 __be32
3526 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3527                    struct nfsd4_free_stateid *free_stateid)
3528 {
3529         stateid_t *stateid = &free_stateid->fr_stateid;
3530         struct nfs4_stid *s;
3531         struct nfs4_client *cl = cstate->session->se_client;
3532         __be32 ret = nfserr_bad_stateid;
3533
3534         nfs4_lock_state();
3535         s = find_stateid(cl, stateid);
3536         if (!s)
3537                 goto out;
3538         switch (s->sc_type) {
3539         case NFS4_DELEG_STID:
3540                 ret = nfserr_locks_held;
3541                 goto out;
3542         case NFS4_OPEN_STID:
3543         case NFS4_LOCK_STID:
3544                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
3545                 if (ret)
3546                         goto out;
3547                 if (s->sc_type == NFS4_LOCK_STID)
3548                         ret = nfsd4_free_lock_stateid(openlockstateid(s));
3549                 else
3550                         ret = nfserr_locks_held;
3551                 break;
3552         default:
3553                 ret = nfserr_bad_stateid;
3554         }
3555 out:
3556         nfs4_unlock_state();
3557         return ret;
3558 }
3559
3560 static inline int
3561 setlkflg (int type)
3562 {
3563         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3564                 RD_STATE : WR_STATE;
3565 }
3566
3567 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
3568 {
3569         struct svc_fh *current_fh = &cstate->current_fh;
3570         struct nfs4_stateowner *sop = stp->st_stateowner;
3571         __be32 status;
3572
3573         status = nfsd4_check_seqid(cstate, sop, seqid);
3574         if (status)
3575                 return status;
3576         if (stp->st_stid.sc_type == NFS4_CLOSED_STID)
3577                 /*
3578                  * "Closed" stateid's exist *only* to return
3579                  * nfserr_replay_me from the previous step.
3580                  */
3581                 return nfserr_bad_stateid;
3582         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
3583         if (status)
3584                 return status;
3585         return nfs4_check_fh(current_fh, stp);
3586 }
3587
3588 /*
3589  * Checks for sequence id mutating operations.
3590  */
3591 static __be32
3592 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3593                          stateid_t *stateid, char typemask,
3594                          struct nfs4_ol_stateid **stpp,
3595                          struct nfsd_net *nn)
3596 {
3597         __be32 status;
3598         struct nfs4_stid *s;
3599
3600         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3601                 seqid, STATEID_VAL(stateid));
3602
3603         *stpp = NULL;
3604         status = nfsd4_lookup_stateid(stateid, typemask, &s,
3605                                       cstate->minorversion, nn);
3606         if (status)
3607                 return status;
3608         *stpp = openlockstateid(s);
3609         cstate->replay_owner = (*stpp)->st_stateowner;
3610
3611         return nfs4_seqid_op_checks(cstate, stateid, seqid, *stpp);
3612 }
3613
3614 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3615                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
3616 {
3617         __be32 status;
3618         struct nfs4_openowner *oo;
3619
3620         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
3621                                                 NFS4_OPEN_STID, stpp, nn);
3622         if (status)
3623                 return status;
3624         oo = openowner((*stpp)->st_stateowner);
3625         if (!(oo->oo_flags & NFS4_OO_CONFIRMED))
3626                 return nfserr_bad_stateid;
3627         return nfs_ok;
3628 }
3629
3630 __be32
3631 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3632                    struct nfsd4_open_confirm *oc)
3633 {
3634         __be32 status;
3635         struct nfs4_openowner *oo;
3636         struct nfs4_ol_stateid *stp;
3637         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3638
3639         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3640                         (int)cstate->current_fh.fh_dentry->d_name.len,
3641                         cstate->current_fh.fh_dentry->d_name.name);
3642
3643         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3644         if (status)
3645                 return status;
3646
3647         nfs4_lock_state();
3648
3649         status = nfs4_preprocess_seqid_op(cstate,
3650                                         oc->oc_seqid, &oc->oc_req_stateid,
3651                                         NFS4_OPEN_STID, &stp, nn);
3652         if (status)
3653                 goto out;
3654         oo = openowner(stp->st_stateowner);
3655         status = nfserr_bad_stateid;
3656         if (oo->oo_flags & NFS4_OO_CONFIRMED)
3657                 goto out;
3658         oo->oo_flags |= NFS4_OO_CONFIRMED;
3659         update_stateid(&stp->st_stid.sc_stateid);
3660         memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3661         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3662                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
3663
3664         nfsd4_client_record_create(oo->oo_owner.so_client);
3665         status = nfs_ok;
3666 out:
3667         if (!cstate->replay_owner)
3668                 nfs4_unlock_state();
3669         return status;
3670 }
3671
3672 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
3673 {
3674         if (!test_access(access, stp))
3675                 return;
3676         nfs4_file_put_access(stp->st_file, nfs4_access_to_omode(access));
3677         clear_access(access, stp);
3678 }
3679
3680 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
3681 {
3682         switch (to_access) {
3683         case NFS4_SHARE_ACCESS_READ:
3684                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
3685                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3686                 break;
3687         case NFS4_SHARE_ACCESS_WRITE:
3688                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
3689                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3690                 break;
3691         case NFS4_SHARE_ACCESS_BOTH:
3692                 break;
3693         default:
3694                 WARN_ON_ONCE(1);
3695         }
3696 }
3697
3698 static void
3699 reset_union_bmap_deny(unsigned long deny, struct nfs4_ol_stateid *stp)
3700 {
3701         int i;
3702         for (i = 0; i < 4; i++) {
3703                 if ((i & deny) != i)
3704                         clear_deny(i, stp);
3705         }
3706 }
3707
3708 __be32
3709 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3710                      struct nfsd4_compound_state *cstate,
3711                      struct nfsd4_open_downgrade *od)
3712 {
3713         __be32 status;
3714         struct nfs4_ol_stateid *stp;
3715         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3716
3717         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3718                         (int)cstate->current_fh.fh_dentry->d_name.len,
3719                         cstate->current_fh.fh_dentry->d_name.name);
3720
3721         /* We don't yet support WANT bits: */
3722         if (od->od_deleg_want)
3723                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
3724                         od->od_deleg_want);
3725
3726         nfs4_lock_state();
3727         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
3728                                         &od->od_stateid, &stp, nn);
3729         if (status)
3730                 goto out;
3731         status = nfserr_inval;
3732         if (!test_access(od->od_share_access, stp)) {
3733                 dprintk("NFSD: access not a subset current bitmap: 0x%lx, input access=%08x\n",
3734                         stp->st_access_bmap, od->od_share_access);
3735                 goto out;
3736         }
3737         if (!test_deny(od->od_share_deny, stp)) {
3738                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3739                         stp->st_deny_bmap, od->od_share_deny);
3740                 goto out;
3741         }
3742         nfs4_stateid_downgrade(stp, od->od_share_access);
3743
3744         reset_union_bmap_deny(od->od_share_deny, stp);
3745
3746         update_stateid(&stp->st_stid.sc_stateid);
3747         memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3748         status = nfs_ok;
3749 out:
3750         if (!cstate->replay_owner)
3751                 nfs4_unlock_state();
3752         return status;
3753 }
3754
3755 void nfsd4_purge_closed_stateid(struct nfs4_stateowner *so)
3756 {
3757         struct nfs4_openowner *oo;
3758         struct nfs4_ol_stateid *s;
3759
3760         if (!so->so_is_open_owner)
3761                 return;
3762         oo = openowner(so);
3763         s = oo->oo_last_closed_stid;
3764         if (!s)
3765                 return;
3766         if (!(oo->oo_flags & NFS4_OO_PURGE_CLOSE)) {
3767                 /* Release the last_closed_stid on the next seqid bump: */
3768                 oo->oo_flags |= NFS4_OO_PURGE_CLOSE;
3769                 return;
3770         }
3771         oo->oo_flags &= ~NFS4_OO_PURGE_CLOSE;
3772         release_last_closed_stateid(oo);
3773 }
3774
3775 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
3776 {
3777         unhash_open_stateid(s);
3778         s->st_stid.sc_type = NFS4_CLOSED_STID;
3779 }
3780
3781 /*
3782  * nfs4_unlock_state() called after encode
3783  */
3784 __be32
3785 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3786             struct nfsd4_close *close)
3787 {
3788         __be32 status;
3789         struct nfs4_openowner *oo;
3790         struct nfs4_ol_stateid *stp;
3791         struct net *net = SVC_NET(rqstp);
3792         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3793
3794         dprintk("NFSD: nfsd4_close on file %.*s\n",
3795                         (int)cstate->current_fh.fh_dentry->d_name.len,
3796                         cstate->current_fh.fh_dentry->d_name.name);
3797
3798         nfs4_lock_state();
3799         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
3800                                         &close->cl_stateid,
3801                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
3802                                         &stp, nn);
3803         if (status)
3804                 goto out;
3805         oo = openowner(stp->st_stateowner);
3806         status = nfs_ok;
3807         update_stateid(&stp->st_stid.sc_stateid);
3808         memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3809
3810         nfsd4_close_open_stateid(stp);
3811         release_last_closed_stateid(oo);
3812         oo->oo_flags &= ~NFS4_OO_PURGE_CLOSE;
3813         oo->oo_last_closed_stid = stp;
3814
3815         if (list_empty(&oo->oo_owner.so_stateids)) {
3816                 if (cstate->minorversion) {
3817                         release_openowner(oo);
3818                         cstate->replay_owner = NULL;
3819                 } else {
3820                         /*
3821                          * In the 4.0 case we need to keep the owners around a
3822                          * little while to handle CLOSE replay.
3823                          */
3824                         if (list_empty(&oo->oo_owner.so_stateids))
3825                                 move_to_close_lru(oo, SVC_NET(rqstp));
3826                 }
3827         }
3828 out:
3829         if (!cstate->replay_owner)
3830                 nfs4_unlock_state();
3831         return status;
3832 }
3833
3834 __be32
3835 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3836                   struct nfsd4_delegreturn *dr)
3837 {
3838         struct nfs4_delegation *dp;
3839         stateid_t *stateid = &dr->dr_stateid;
3840         struct nfs4_stid *s;
3841         __be32 status;
3842         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3843
3844         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3845                 return status;
3846
3847         nfs4_lock_state();
3848         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID, &s,
3849                                       cstate->minorversion, nn);
3850         if (status)
3851                 goto out;
3852         dp = delegstateid(s);
3853         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
3854         if (status)
3855                 goto out;
3856
3857         unhash_delegation(dp);
3858 out:
3859         nfs4_unlock_state();
3860
3861         return status;
3862 }
3863
3864
3865 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3866
3867 #define LOCKOWNER_INO_HASH_MASK (LOCKOWNER_INO_HASH_SIZE - 1)
3868
3869 static inline u64
3870 end_offset(u64 start, u64 len)
3871 {
3872         u64 end;
3873
3874         end = start + len;
3875         return end >= start ? end: NFS4_MAX_UINT64;
3876 }
3877
3878 /* last octet in a range */
3879 static inline u64
3880 last_byte_offset(u64 start, u64 len)
3881 {
3882         u64 end;
3883
3884         WARN_ON_ONCE(!len);
3885         end = start + len;
3886         return end > start ? end - 1: NFS4_MAX_UINT64;
3887 }
3888
3889 static unsigned int lockowner_ino_hashval(struct inode *inode, u32 cl_id, struct xdr_netobj *ownername)
3890 {
3891         return (file_hashval(inode) + cl_id
3892                         + opaque_hashval(ownername->data, ownername->len))
3893                 & LOCKOWNER_INO_HASH_MASK;
3894 }
3895
3896 /*
3897  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3898  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3899  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3900  * locking, this prevents us from being completely protocol-compliant.  The
3901  * real solution to this problem is to start using unsigned file offsets in
3902  * the VFS, but this is a very deep change!
3903  */
3904 static inline void
3905 nfs4_transform_lock_offset(struct file_lock *lock)
3906 {
3907         if (lock->fl_start < 0)
3908                 lock->fl_start = OFFSET_MAX;
3909         if (lock->fl_end < 0)
3910                 lock->fl_end = OFFSET_MAX;
3911 }
3912
3913 /* Hack!: For now, we're defining this just so we can use a pointer to it
3914  * as a unique cookie to identify our (NFSv4's) posix locks. */
3915 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
3916 };
3917
3918 static inline void
3919 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3920 {
3921         struct nfs4_lockowner *lo;
3922
3923         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3924                 lo = (struct nfs4_lockowner *) fl->fl_owner;
3925                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
3926                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
3927                 if (!deny->ld_owner.data)
3928                         /* We just don't care that much */
3929                         goto nevermind;
3930                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
3931                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
3932         } else {
3933 nevermind:
3934                 deny->ld_owner.len = 0;
3935                 deny->ld_owner.data = NULL;
3936                 deny->ld_clientid.cl_boot = 0;
3937                 deny->ld_clientid.cl_id = 0;
3938         }
3939         deny->ld_start = fl->fl_start;
3940         deny->ld_length = NFS4_MAX_UINT64;
3941         if (fl->fl_end != NFS4_MAX_UINT64)
3942                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3943         deny->ld_type = NFS4_READ_LT;
3944         if (fl->fl_type != F_RDLCK)
3945                 deny->ld_type = NFS4_WRITE_LT;
3946 }
3947
3948 static bool same_lockowner_ino(struct nfs4_lockowner *lo, struct inode *inode, clientid_t *clid, struct xdr_netobj *owner)
3949 {
3950         struct nfs4_ol_stateid *lst;
3951
3952         if (!same_owner_str(&lo->lo_owner, owner, clid))
3953                 return false;
3954         lst = list_first_entry(&lo->lo_owner.so_stateids,
3955                                struct nfs4_ol_stateid, st_perstateowner);
3956         return lst->st_file->fi_inode == inode;
3957 }
3958
3959 static struct nfs4_lockowner *
3960 find_lockowner_str(struct inode *inode, clientid_t *clid,
3961                    struct xdr_netobj *owner, struct nfsd_net *nn)
3962 {
3963         unsigned int hashval = lockowner_ino_hashval(inode, clid->cl_id, owner);
3964         struct nfs4_lockowner *lo;
3965
3966         list_for_each_entry(lo, &nn->lockowner_ino_hashtbl[hashval], lo_owner_ino_hash) {
3967                 if (same_lockowner_ino(lo, inode, clid, owner))
3968                         return lo;
3969         }
3970         return NULL;
3971 }
3972
3973 static void hash_lockowner(struct nfs4_lockowner *lo, unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp)
3974 {
3975         struct inode *inode = open_stp->st_file->fi_inode;
3976         unsigned int inohash = lockowner_ino_hashval(inode,
3977                         clp->cl_clientid.cl_id, &lo->lo_owner.so_owner);
3978         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3979
3980         list_add(&lo->lo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
3981         list_add(&lo->lo_owner_ino_hash, &nn->lockowner_ino_hashtbl[inohash]);
3982         list_add(&lo->lo_perstateid, &open_stp->st_lockowners);
3983 }
3984
3985 /*
3986  * Alloc a lock owner structure.
3987  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3988  * occurred.
3989  *
3990  * strhashval = ownerstr_hashval
3991  */
3992
3993 static struct nfs4_lockowner *
3994 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp, struct nfsd4_lock *lock) {
3995         struct nfs4_lockowner *lo;
3996
3997         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
3998         if (!lo)
3999                 return NULL;
4000         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
4001         lo->lo_owner.so_is_open_owner = 0;
4002         /* It is the openowner seqid that will be incremented in encode in the
4003          * case of new lockowners; so increment the lock seqid manually: */
4004         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid + 1;
4005         hash_lockowner(lo, strhashval, clp, open_stp);
4006         return lo;
4007 }
4008
4009 static struct nfs4_ol_stateid *
4010 alloc_init_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp, struct nfs4_ol_stateid *open_stp)
4011 {
4012         struct nfs4_ol_stateid *stp;
4013         struct nfs4_client *clp = lo->lo_owner.so_client;
4014
4015         stp = nfs4_alloc_stateid(clp);
4016         if (stp == NULL)
4017                 return NULL;
4018         stp->st_stid.sc_type = NFS4_LOCK_STID;
4019         list_add(&stp->st_perfile, &fp->fi_stateids);
4020         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
4021         stp->st_stateowner = &lo->lo_owner;
4022         get_nfs4_file(fp);
4023         stp->st_file = fp;
4024         stp->st_access_bmap = 0;
4025         stp->st_deny_bmap = open_stp->st_deny_bmap;
4026         stp->st_openstp = open_stp;
4027         return stp;
4028 }
4029
4030 static int
4031 check_lock_length(u64 offset, u64 length)
4032 {
4033         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
4034              LOFF_OVERFLOW(offset, length)));
4035 }
4036
4037 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
4038 {
4039         struct nfs4_file *fp = lock_stp->st_file;
4040         int oflag = nfs4_access_to_omode(access);
4041
4042         if (test_access(access, lock_stp))
4043                 return;
4044         nfs4_file_get_access(fp, oflag);
4045         set_access(access, lock_stp);
4046 }
4047
4048 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)
4049 {
4050         struct nfs4_file *fi = ost->st_file;
4051         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
4052         struct nfs4_client *cl = oo->oo_owner.so_client;
4053         struct nfs4_lockowner *lo;
4054         unsigned int strhashval;
4055         struct nfsd_net *nn = net_generic(cl->net, nfsd_net_id);
4056
4057         lo = find_lockowner_str(fi->fi_inode, &cl->cl_clientid,
4058                                 &lock->v.new.owner, nn);
4059         if (lo) {
4060                 if (!cstate->minorversion)
4061                         return nfserr_bad_seqid;
4062                 /* XXX: a lockowner always has exactly one stateid: */
4063                 *lst = list_first_entry(&lo->lo_owner.so_stateids,
4064                                 struct nfs4_ol_stateid, st_perstateowner);
4065                 return nfs_ok;
4066         }
4067         strhashval = ownerstr_hashval(cl->cl_clientid.cl_id,
4068                         &lock->v.new.owner);
4069         lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
4070         if (lo == NULL)
4071                 return nfserr_jukebox;
4072         *lst = alloc_init_lock_stateid(lo, fi, ost);
4073         if (*lst == NULL) {
4074                 release_lockowner(lo);
4075                 return nfserr_jukebox;
4076         }
4077         *new = true;
4078         return nfs_ok;
4079 }
4080
4081 /*
4082  *  LOCK operation
4083  */
4084 __be32
4085 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4086            struct nfsd4_lock *lock)
4087 {
4088         struct nfs4_openowner *open_sop = NULL;
4089         struct nfs4_lockowner *lock_sop = NULL;
4090         struct nfs4_ol_stateid *lock_stp;
4091         struct file *filp = NULL;
4092         struct file_lock *file_lock = NULL;
4093         struct file_lock *conflock = NULL;
4094         __be32 status = 0;
4095         bool new_state = false;
4096         int lkflg;
4097         int err;
4098         struct net *net = SVC_NET(rqstp);
4099         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4100
4101         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
4102                 (long long) lock->lk_offset,
4103                 (long long) lock->lk_length);
4104
4105         if (check_lock_length(lock->lk_offset, lock->lk_length))
4106                  return nfserr_inval;
4107
4108         if ((status = fh_verify(rqstp, &cstate->current_fh,
4109                                 S_IFREG, NFSD_MAY_LOCK))) {
4110                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
4111                 return status;
4112         }
4113
4114         nfs4_lock_state();
4115
4116         if (lock->lk_is_new) {
4117                 struct nfs4_ol_stateid *open_stp = NULL;
4118
4119                 if (nfsd4_has_session(cstate))
4120                         /* See rfc 5661 18.10.3: given clientid is ignored: */
4121                         memcpy(&lock->v.new.clientid,
4122                                 &cstate->session->se_client->cl_clientid,
4123                                 sizeof(clientid_t));
4124
4125                 status = nfserr_stale_clientid;
4126                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
4127                         goto out;
4128
4129                 /* validate and update open stateid and open seqid */
4130                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
4131                                         lock->lk_new_open_seqid,
4132                                         &lock->lk_new_open_stateid,
4133                                         &open_stp, nn);
4134                 if (status)
4135                         goto out;
4136                 open_sop = openowner(open_stp->st_stateowner);
4137                 status = nfserr_bad_stateid;
4138                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
4139                                                 &lock->v.new.clientid))
4140                         goto out;
4141                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
4142                                                         &lock_stp, &new_state);
4143         } else
4144                 status = nfs4_preprocess_seqid_op(cstate,
4145                                        lock->lk_old_lock_seqid,
4146                                        &lock->lk_old_lock_stateid,
4147                                        NFS4_LOCK_STID, &lock_stp, nn);
4148         if (status)
4149                 goto out;
4150         lock_sop = lockowner(lock_stp->st_stateowner);
4151
4152         lkflg = setlkflg(lock->lk_type);
4153         status = nfs4_check_openmode(lock_stp, lkflg);
4154         if (status)
4155                 goto out;
4156
4157         status = nfserr_grace;
4158         if (locks_in_grace(net) && !lock->lk_reclaim)
4159                 goto out;
4160         status = nfserr_no_grace;
4161         if (!locks_in_grace(net) && lock->lk_reclaim)
4162                 goto out;
4163
4164         file_lock = locks_alloc_lock();
4165         if (!file_lock) {
4166                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4167                 status = nfserr_jukebox;
4168                 goto out;
4169         }
4170
4171         locks_init_lock(file_lock);
4172         switch (lock->lk_type) {
4173                 case NFS4_READ_LT:
4174                 case NFS4_READW_LT:
4175                         filp = find_readable_file(lock_stp->st_file);
4176                         if (filp)
4177                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
4178                         file_lock->fl_type = F_RDLCK;
4179                         break;
4180                 case NFS4_WRITE_LT:
4181                 case NFS4_WRITEW_LT:
4182                         filp = find_writeable_file(lock_stp->st_file);
4183                         if (filp)
4184                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
4185                         file_lock->fl_type = F_WRLCK;
4186                         break;
4187                 default:
4188                         status = nfserr_inval;
4189                 goto out;
4190         }
4191         if (!filp) {
4192                 status = nfserr_openmode;
4193                 goto out;
4194         }
4195         file_lock->fl_owner = (fl_owner_t)lock_sop;
4196         file_lock->fl_pid = current->tgid;
4197         file_lock->fl_file = filp;
4198         file_lock->fl_flags = FL_POSIX;
4199         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4200         file_lock->fl_start = lock->lk_offset;
4201         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
4202         nfs4_transform_lock_offset(file_lock);
4203
4204         conflock = locks_alloc_lock();
4205         if (!conflock) {
4206                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4207                 status = nfserr_jukebox;
4208                 goto out;
4209         }
4210
4211         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
4212         switch (-err) {
4213         case 0: /* success! */
4214                 update_stateid(&lock_stp->st_stid.sc_stateid);
4215                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
4216                                 sizeof(stateid_t));
4217                 status = 0;
4218                 break;
4219         case (EAGAIN):          /* conflock holds conflicting lock */
4220                 status = nfserr_denied;
4221                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
4222                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
4223                 break;
4224         case (EDEADLK):
4225                 status = nfserr_deadlock;
4226                 break;
4227         default:
4228                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4229                 status = nfserrno(err);
4230                 break;
4231         }
4232 out:
4233         if (status && new_state)
4234                 release_lockowner(lock_sop);
4235         if (!cstate->replay_owner)
4236                 nfs4_unlock_state();
4237         if (file_lock)
4238                 locks_free_lock(file_lock);
4239         if (conflock)
4240                 locks_free_lock(conflock);
4241         return status;
4242 }
4243
4244 /*
4245  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4246  * so we do a temporary open here just to get an open file to pass to
4247  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
4248  * inode operation.)
4249  */
4250 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4251 {
4252         struct file *file;
4253         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4254         if (!err) {
4255                 err = nfserrno(vfs_test_lock(file, lock));
4256                 nfsd_close(file);
4257         }
4258         return err;
4259 }
4260
4261 /*
4262  * LOCKT operation
4263  */
4264 __be32
4265 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4266             struct nfsd4_lockt *lockt)
4267 {
4268         struct inode *inode;
4269         struct file_lock *file_lock = NULL;
4270         struct nfs4_lockowner *lo;
4271         __be32 status;
4272         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4273
4274         if (locks_in_grace(SVC_NET(rqstp)))
4275                 return nfserr_grace;
4276
4277         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4278                  return nfserr_inval;
4279
4280         nfs4_lock_state();
4281
4282         if (!nfsd4_has_session(cstate)) {
4283                 status = lookup_clientid(&lockt->lt_clientid, false, nn, NULL);
4284                 if (status)
4285                         goto out;
4286         }
4287
4288         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4289                 goto out;
4290
4291         inode = cstate->current_fh.fh_dentry->d_inode;
4292         file_lock = locks_alloc_lock();
4293         if (!file_lock) {
4294                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4295                 status = nfserr_jukebox;
4296                 goto out;
4297         }
4298         locks_init_lock(file_lock);
4299         switch (lockt->lt_type) {
4300                 case NFS4_READ_LT:
4301                 case NFS4_READW_LT:
4302                         file_lock->fl_type = F_RDLCK;
4303                 break;
4304                 case NFS4_WRITE_LT:
4305                 case NFS4_WRITEW_LT:
4306                         file_lock->fl_type = F_WRLCK;
4307                 break;
4308                 default:
4309                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4310                         status = nfserr_inval;
4311                 goto out;
4312         }
4313
4314         lo = find_lockowner_str(inode, &lockt->lt_clientid, &lockt->lt_owner, nn);
4315         if (lo)
4316                 file_lock->fl_owner = (fl_owner_t)lo;
4317         file_lock->fl_pid = current->tgid;
4318         file_lock->fl_flags = FL_POSIX;
4319
4320         file_lock->fl_start = lockt->lt_offset;
4321         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4322
4323         nfs4_transform_lock_offset(file_lock);
4324
4325         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
4326         if (status)
4327                 goto out;
4328
4329         if (file_lock->fl_type != F_UNLCK) {
4330                 status = nfserr_denied;
4331                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
4332         }
4333 out:
4334         nfs4_unlock_state();
4335         if (file_lock)
4336                 locks_free_lock(file_lock);
4337         return status;
4338 }
4339
4340 __be32
4341 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4342             struct nfsd4_locku *locku)
4343 {
4344         struct nfs4_ol_stateid *stp;
4345         struct file *filp = NULL;
4346         struct file_lock *file_lock = NULL;
4347         __be32 status;
4348         int err;
4349         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4350
4351         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4352                 (long long) locku->lu_offset,
4353                 (long long) locku->lu_length);
4354
4355         if (check_lock_length(locku->lu_offset, locku->lu_length))
4356                  return nfserr_inval;
4357
4358         nfs4_lock_state();
4359
4360         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
4361                                         &locku->lu_stateid, NFS4_LOCK_STID,
4362                                         &stp, nn);
4363         if (status)
4364                 goto out;
4365         filp = find_any_file(stp->st_file);
4366         if (!filp) {
4367                 status = nfserr_lock_range;
4368                 goto out;
4369         }
4370         file_lock = locks_alloc_lock();
4371         if (!file_lock) {
4372                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4373                 status = nfserr_jukebox;
4374                 goto out;
4375         }
4376         locks_init_lock(file_lock);
4377         file_lock->fl_type = F_UNLCK;
4378         file_lock->fl_owner = (fl_owner_t)lockowner(stp->st_stateowner);
4379         file_lock->fl_pid = current->tgid;
4380         file_lock->fl_file = filp;
4381         file_lock->fl_flags = FL_POSIX;
4382         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4383         file_lock->fl_start = locku->lu_offset;
4384
4385         file_lock->fl_end = last_byte_offset(locku->lu_offset,
4386                                                 locku->lu_length);
4387         nfs4_transform_lock_offset(file_lock);
4388
4389         /*
4390         *  Try to unlock the file in the VFS.
4391         */
4392         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
4393         if (err) {
4394                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4395                 goto out_nfserr;
4396         }
4397         /*
4398         * OK, unlock succeeded; the only thing left to do is update the stateid.
4399         */
4400         update_stateid(&stp->st_stid.sc_stateid);
4401         memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4402
4403 out:
4404         if (!cstate->replay_owner)
4405                 nfs4_unlock_state();
4406         if (file_lock)
4407                 locks_free_lock(file_lock);
4408         return status;
4409
4410 out_nfserr:
4411         status = nfserrno(err);
4412         goto out;
4413 }
4414
4415 /*
4416  * returns
4417  *      1: locks held by lockowner
4418  *      0: no locks held by lockowner
4419  */
4420 static int
4421 check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner)
4422 {
4423         struct file_lock **flpp;
4424         struct inode *inode = filp->fi_inode;
4425         int status = 0;
4426
4427         lock_flocks();
4428         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4429                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4430                         status = 1;
4431                         goto out;
4432                 }
4433         }
4434 out:
4435         unlock_flocks();
4436         return status;
4437 }
4438
4439 __be32
4440 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4441                         struct nfsd4_compound_state *cstate,
4442                         struct nfsd4_release_lockowner *rlockowner)
4443 {
4444         clientid_t *clid = &rlockowner->rl_clientid;
4445         struct nfs4_stateowner *sop;
4446         struct nfs4_lockowner *lo;
4447         struct nfs4_ol_stateid *stp;
4448         struct xdr_netobj *owner = &rlockowner->rl_owner;
4449         struct list_head matches;
4450         unsigned int hashval = ownerstr_hashval(clid->cl_id, owner);
4451         __be32 status;
4452         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4453
4454         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4455                 clid->cl_boot, clid->cl_id);
4456
4457         nfs4_lock_state();
4458
4459         status = lookup_clientid(clid, cstate->minorversion, nn, NULL);
4460         if (status)
4461                 goto out;
4462
4463         status = nfserr_locks_held;
4464         INIT_LIST_HEAD(&matches);
4465
4466         list_for_each_entry(sop, &nn->ownerstr_hashtbl[hashval], so_strhash) {
4467                 if (sop->so_is_open_owner)
4468                         continue;
4469                 if (!same_owner_str(sop, owner, clid))
4470                         continue;
4471                 list_for_each_entry(stp, &sop->so_stateids,
4472                                 st_perstateowner) {
4473                         lo = lockowner(sop);
4474                         if (check_for_locks(stp->st_file, lo))
4475                                 goto out;
4476                         list_add(&lo->lo_list, &matches);
4477                 }
4478         }
4479         /* Clients probably won't expect us to return with some (but not all)
4480          * of the lockowner state released; so don't release any until all
4481          * have been checked. */
4482         status = nfs_ok;
4483         while (!list_empty(&matches)) {
4484                 lo = list_entry(matches.next, struct nfs4_lockowner,
4485                                                                 lo_list);
4486                 /* unhash_stateowner deletes so_perclient only
4487                  * for openowners. */
4488                 list_del(&lo->lo_list);
4489                 release_lockowner(lo);
4490         }
4491 out:
4492         nfs4_unlock_state();
4493         return status;
4494 }
4495
4496 static inline struct nfs4_client_reclaim *
4497 alloc_reclaim(void)
4498 {
4499         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4500 }
4501
4502 bool
4503 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
4504 {
4505         struct nfs4_client_reclaim *crp;
4506
4507         crp = nfsd4_find_reclaim_client(name, nn);
4508         return (crp && crp->cr_clp);
4509 }
4510
4511 /*
4512  * failure => all reset bets are off, nfserr_no_grace...
4513  */
4514 struct nfs4_client_reclaim *
4515 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
4516 {
4517         unsigned int strhashval;
4518         struct nfs4_client_reclaim *crp;
4519
4520         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4521         crp = alloc_reclaim();
4522         if (crp) {
4523                 strhashval = clientstr_hashval(name);
4524                 INIT_LIST_HEAD(&crp->cr_strhash);
4525                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
4526                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4527                 crp->cr_clp = NULL;
4528                 nn->reclaim_str_hashtbl_size++;
4529         }
4530         return crp;
4531 }
4532
4533 void
4534 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
4535 {
4536         list_del(&crp->cr_strhash);
4537         kfree(crp);
4538         nn->reclaim_str_hashtbl_size--;
4539 }
4540
4541 void
4542 nfs4_release_reclaim(struct nfsd_net *nn)
4543 {
4544         struct nfs4_client_reclaim *crp = NULL;
4545         int i;
4546
4547         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4548                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
4549                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
4550                                         struct nfs4_client_reclaim, cr_strhash);
4551                         nfs4_remove_reclaim_record(crp, nn);
4552                 }
4553         }
4554         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
4555 }
4556
4557 /*
4558  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4559 struct nfs4_client_reclaim *
4560 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
4561 {
4562         unsigned int strhashval;
4563         struct nfs4_client_reclaim *crp = NULL;
4564
4565         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
4566
4567         strhashval = clientstr_hashval(recdir);
4568         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
4569                 if (same_name(crp->cr_recdir, recdir)) {
4570                         return crp;
4571                 }
4572         }
4573         return NULL;
4574 }
4575
4576 /*
4577 * Called from OPEN. Look for clientid in reclaim list.
4578 */
4579 __be32
4580 nfs4_check_open_reclaim(clientid_t *clid, bool sessions, struct nfsd_net *nn)
4581 {
4582         struct nfs4_client *clp;
4583
4584         /* find clientid in conf_id_hashtbl */
4585         clp = find_confirmed_client(clid, sessions, nn);
4586         if (clp == NULL)
4587                 return nfserr_reclaim_bad;
4588
4589         return nfsd4_client_record_check(clp) ? nfserr_reclaim_bad : nfs_ok;
4590 }
4591
4592 #ifdef CONFIG_NFSD_FAULT_INJECTION
4593
4594 u64 nfsd_forget_client(struct nfs4_client *clp, u64 max)
4595 {
4596         expire_client(clp);
4597         return 1;
4598 }
4599
4600 u64 nfsd_print_client(struct nfs4_client *clp, u64 num)
4601 {
4602         char buf[INET6_ADDRSTRLEN];
4603         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
4604         printk(KERN_INFO "NFS Client: %s\n", buf);
4605         return 1;
4606 }
4607
4608 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
4609                              const char *type)
4610 {
4611         char buf[INET6_ADDRSTRLEN];
4612         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
4613         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
4614 }
4615
4616 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max, void (*func)(struct nfs4_lockowner *))
4617 {
4618         struct nfs4_openowner *oop;
4619         struct nfs4_lockowner *lop, *lo_next;
4620         struct nfs4_ol_stateid *stp, *st_next;
4621         u64 count = 0;
4622
4623         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
4624                 list_for_each_entry_safe(stp, st_next, &oop->oo_owner.so_stateids, st_perstateowner) {
4625                         list_for_each_entry_safe(lop, lo_next, &stp->st_lockowners, lo_perstateid) {
4626                                 if (func)
4627                                         func(lop);
4628                                 if (++count == max)
4629                                         return count;
4630                         }
4631                 }
4632         }
4633
4634         return count;
4635 }
4636
4637 u64 nfsd_forget_client_locks(struct nfs4_client *clp, u64 max)
4638 {
4639         return nfsd_foreach_client_lock(clp, max, release_lockowner);
4640 }
4641
4642 u64 nfsd_print_client_locks(struct nfs4_client *clp, u64 max)
4643 {
4644         u64 count = nfsd_foreach_client_lock(clp, max, NULL);
4645         nfsd_print_count(clp, count, "locked files");
4646         return count;
4647 }
4648
4649 static u64 nfsd_foreach_client_open(struct nfs4_client *clp, u64 max, void (*func)(struct nfs4_openowner *))
4650 {
4651         struct nfs4_openowner *oop, *next;
4652         u64 count = 0;
4653
4654         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
4655                 if (func)
4656                         func(oop);
4657                 if (++count == max)
4658                         break;
4659         }
4660
4661         return count;
4662 }
4663
4664 u64 nfsd_forget_client_openowners(struct nfs4_client *clp, u64 max)
4665 {
4666         return nfsd_foreach_client_open(clp, max, release_openowner);
4667 }
4668
4669 u64 nfsd_print_client_openowners(struct nfs4_client *clp, u64 max)
4670 {
4671         u64 count = nfsd_foreach_client_open(clp, max, NULL);
4672         nfsd_print_count(clp, count, "open files");
4673         return count;
4674 }
4675
4676 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
4677                                      struct list_head *victims)
4678 {
4679         struct nfs4_delegation *dp, *next;
4680         u64 count = 0;
4681
4682         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
4683                 if (victims)
4684                         list_move(&dp->dl_recall_lru, victims);
4685                 if (++count == max)
4686                         break;
4687         }
4688         return count;
4689 }
4690
4691 u64 nfsd_forget_client_delegations(struct nfs4_client *clp, u64 max)
4692 {
4693         struct nfs4_delegation *dp, *next;
4694         LIST_HEAD(victims);
4695         u64 count;
4696
4697         spin_lock(&recall_lock);
4698         count = nfsd_find_all_delegations(clp, max, &victims);
4699         spin_unlock(&recall_lock);
4700
4701         list_for_each_entry_safe(dp, next, &victims, dl_recall_lru)
4702                 unhash_delegation(dp);
4703
4704         return count;
4705 }
4706
4707 u64 nfsd_recall_client_delegations(struct nfs4_client *clp, u64 max)
4708 {
4709         struct nfs4_delegation *dp, *next;
4710         LIST_HEAD(victims);
4711         u64 count;
4712
4713         spin_lock(&recall_lock);
4714         count = nfsd_find_all_delegations(clp, max, &victims);
4715         list_for_each_entry_safe(dp, next, &victims, dl_recall_lru)
4716                 nfsd_break_one_deleg(dp);
4717         spin_unlock(&recall_lock);
4718
4719         return count;
4720 }
4721
4722 u64 nfsd_print_client_delegations(struct nfs4_client *clp, u64 max)
4723 {
4724         u64 count = 0;
4725
4726         spin_lock(&recall_lock);
4727         count = nfsd_find_all_delegations(clp, max, NULL);
4728         spin_unlock(&recall_lock);
4729
4730         nfsd_print_count(clp, count, "delegations");
4731         return count;
4732 }
4733
4734 u64 nfsd_for_n_state(u64 max, u64 (*func)(struct nfs4_client *, u64))
4735 {
4736         struct nfs4_client *clp, *next;
4737         u64 count = 0;
4738         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
4739
4740         if (!nfsd_netns_ready(nn))
4741                 return 0;
4742
4743         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
4744                 count += func(clp, max - count);
4745                 if ((max != 0) && (count >= max))
4746                         break;
4747         }
4748
4749         return count;
4750 }
4751
4752 struct nfs4_client *nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
4753 {
4754         struct nfs4_client *clp;
4755         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
4756
4757         if (!nfsd_netns_ready(nn))
4758                 return NULL;
4759
4760         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
4761                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
4762                         return clp;
4763         }
4764         return NULL;
4765 }
4766
4767 #endif /* CONFIG_NFSD_FAULT_INJECTION */
4768
4769 /* initialization to perform at module load time: */
4770
4771 void
4772 nfs4_state_init(void)
4773 {
4774         int i;
4775
4776         for (i = 0; i < FILE_HASH_SIZE; i++) {
4777                 INIT_LIST_HEAD(&file_hashtbl[i]);
4778         }
4779         INIT_LIST_HEAD(&del_recall_lru);
4780 }
4781
4782 /*
4783  * Since the lifetime of a delegation isn't limited to that of an open, a
4784  * client may quite reasonably hang on to a delegation as long as it has
4785  * the inode cached.  This becomes an obvious problem the first time a
4786  * client's inode cache approaches the size of the server's total memory.
4787  *
4788  * For now we avoid this problem by imposing a hard limit on the number
4789  * of delegations, which varies according to the server's memory size.
4790  */
4791 static void
4792 set_max_delegations(void)
4793 {
4794         /*
4795          * Allow at most 4 delegations per megabyte of RAM.  Quick
4796          * estimates suggest that in the worst case (where every delegation
4797          * is for a different inode), a delegation could take about 1.5K,
4798          * giving a worst case usage of about 6% of memory.
4799          */
4800         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4801 }
4802
4803 static int nfs4_state_create_net(struct net *net)
4804 {
4805         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4806         int i;
4807
4808         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
4809                         CLIENT_HASH_SIZE, GFP_KERNEL);
4810         if (!nn->conf_id_hashtbl)
4811                 goto err;
4812         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
4813                         CLIENT_HASH_SIZE, GFP_KERNEL);
4814         if (!nn->unconf_id_hashtbl)
4815                 goto err_unconf_id;
4816         nn->ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
4817                         OWNER_HASH_SIZE, GFP_KERNEL);
4818         if (!nn->ownerstr_hashtbl)
4819                 goto err_ownerstr;
4820         nn->lockowner_ino_hashtbl = kmalloc(sizeof(struct list_head) *
4821                         LOCKOWNER_INO_HASH_SIZE, GFP_KERNEL);
4822         if (!nn->lockowner_ino_hashtbl)
4823                 goto err_lockowner_ino;
4824         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
4825                         SESSION_HASH_SIZE, GFP_KERNEL);
4826         if (!nn->sessionid_hashtbl)
4827                 goto err_sessionid;
4828
4829         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4830                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
4831                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
4832         }
4833         for (i = 0; i < OWNER_HASH_SIZE; i++)
4834                 INIT_LIST_HEAD(&nn->ownerstr_hashtbl[i]);
4835         for (i = 0; i < LOCKOWNER_INO_HASH_SIZE; i++)
4836                 INIT_LIST_HEAD(&nn->lockowner_ino_hashtbl[i]);
4837         for (i = 0; i < SESSION_HASH_SIZE; i++)
4838                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
4839         nn->conf_name_tree = RB_ROOT;
4840         nn->unconf_name_tree = RB_ROOT;
4841         INIT_LIST_HEAD(&nn->client_lru);
4842         INIT_LIST_HEAD(&nn->close_lru);
4843         spin_lock_init(&nn->client_lock);
4844
4845         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
4846         get_net(net);
4847
4848         return 0;
4849
4850 err_sessionid:
4851         kfree(nn->lockowner_ino_hashtbl);
4852 err_lockowner_ino:
4853         kfree(nn->ownerstr_hashtbl);
4854 err_ownerstr:
4855         kfree(nn->unconf_id_hashtbl);
4856 err_unconf_id:
4857         kfree(nn->conf_id_hashtbl);
4858 err:
4859         return -ENOMEM;
4860 }
4861
4862 static void
4863 nfs4_state_destroy_net(struct net *net)
4864 {
4865         int i;
4866         struct nfs4_client *clp = NULL;
4867         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4868         struct rb_node *node, *tmp;
4869
4870         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4871                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
4872                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4873                         destroy_client(clp);
4874                 }
4875         }
4876
4877         node = rb_first(&nn->unconf_name_tree);
4878         while (node != NULL) {
4879                 tmp = node;
4880                 node = rb_next(tmp);
4881                 clp = rb_entry(tmp, struct nfs4_client, cl_namenode);
4882                 rb_erase(tmp, &nn->unconf_name_tree);
4883                 destroy_client(clp);
4884         }
4885
4886         kfree(nn->sessionid_hashtbl);
4887         kfree(nn->lockowner_ino_hashtbl);
4888         kfree(nn->ownerstr_hashtbl);
4889         kfree(nn->unconf_id_hashtbl);
4890         kfree(nn->conf_id_hashtbl);
4891         put_net(net);
4892 }
4893
4894 int
4895 nfs4_state_start_net(struct net *net)
4896 {
4897         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4898         int ret;
4899
4900         ret = nfs4_state_create_net(net);
4901         if (ret)
4902                 return ret;
4903         nfsd4_client_tracking_init(net);
4904         nn->boot_time = get_seconds();
4905         locks_start_grace(net, &nn->nfsd4_manager);
4906         nn->grace_ended = false;
4907         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
4908                nn->nfsd4_grace, net);
4909         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
4910         return 0;
4911 }
4912
4913 /* initialization to perform when the nfsd service is started: */
4914
4915 int
4916 nfs4_state_start(void)
4917 {
4918         int ret;
4919
4920         ret = set_callback_cred();
4921         if (ret)
4922                 return -ENOMEM;
4923         laundry_wq = create_singlethread_workqueue("nfsd4");
4924         if (laundry_wq == NULL) {
4925                 ret = -ENOMEM;
4926                 goto out_recovery;
4927         }
4928         ret = nfsd4_create_callback_queue();
4929         if (ret)
4930                 goto out_free_laundry;
4931
4932         set_max_delegations();
4933
4934         return 0;
4935
4936 out_free_laundry:
4937         destroy_workqueue(laundry_wq);
4938 out_recovery:
4939         return ret;
4940 }
4941
4942 /* should be called with the state lock held */
4943 void
4944 nfs4_state_shutdown_net(struct net *net)
4945 {
4946         struct nfs4_delegation *dp = NULL;
4947         struct list_head *pos, *next, reaplist;
4948         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4949
4950         cancel_delayed_work_sync(&nn->laundromat_work);
4951         locks_end_grace(&nn->nfsd4_manager);
4952
4953         INIT_LIST_HEAD(&reaplist);
4954         spin_lock(&recall_lock);
4955         list_for_each_safe(pos, next, &del_recall_lru) {
4956                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4957                 if (dp->dl_stid.sc_client->net != net)
4958                         continue;
4959                 list_move(&dp->dl_recall_lru, &reaplist);
4960         }
4961         spin_unlock(&recall_lock);
4962         list_for_each_safe(pos, next, &reaplist) {
4963                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4964                 unhash_delegation(dp);
4965         }
4966
4967         nfsd4_client_tracking_exit(net);
4968         nfs4_state_destroy_net(net);
4969 }
4970
4971 void
4972 nfs4_state_shutdown(void)
4973 {
4974         destroy_workqueue(laundry_wq);
4975         nfsd4_destroy_callback_queue();
4976 }
4977
4978 static void
4979 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
4980 {
4981         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
4982                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
4983 }
4984
4985 static void
4986 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
4987 {
4988         if (cstate->minorversion) {
4989                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
4990                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
4991         }
4992 }
4993
4994 void
4995 clear_current_stateid(struct nfsd4_compound_state *cstate)
4996 {
4997         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
4998 }
4999
5000 /*
5001  * functions to set current state id
5002  */
5003 void
5004 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5005 {
5006         put_stateid(cstate, &odp->od_stateid);
5007 }
5008
5009 void
5010 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
5011 {
5012         put_stateid(cstate, &open->op_stateid);
5013 }
5014
5015 void
5016 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5017 {
5018         put_stateid(cstate, &close->cl_stateid);
5019 }
5020
5021 void
5022 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
5023 {
5024         put_stateid(cstate, &lock->lk_resp_stateid);
5025 }
5026
5027 /*
5028  * functions to consume current state id
5029  */
5030
5031 void
5032 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5033 {
5034         get_stateid(cstate, &odp->od_stateid);
5035 }
5036
5037 void
5038 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
5039 {
5040         get_stateid(cstate, &drp->dr_stateid);
5041 }
5042
5043 void
5044 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
5045 {
5046         get_stateid(cstate, &fsp->fr_stateid);
5047 }
5048
5049 void
5050 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
5051 {
5052         get_stateid(cstate, &setattr->sa_stateid);
5053 }
5054
5055 void
5056 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5057 {
5058         get_stateid(cstate, &close->cl_stateid);
5059 }
5060
5061 void
5062 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
5063 {
5064         get_stateid(cstate, &locku->lu_stateid);
5065 }
5066
5067 void
5068 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
5069 {
5070         get_stateid(cstate, &read->rd_stateid);
5071 }
5072
5073 void
5074 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
5075 {
5076         get_stateid(cstate, &write->wr_stateid);
5077 }