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 <linux/jhash.h>
  45 #include "xdr4.h"
  46 #include "xdr4cb.h"
  47 #include "vfs.h"
  48 #include "current_stateid.h"
  49
  50 #include "netns.h"
  51 #include "pnfs.h"
  52
  53 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  54
  55 #define all_ones {{~0,~0},~0}
  56 static const stateid_t one_stateid = {
  57         .si_generation = ~0,
  58         .si_opaque = all_ones,
  59 };
  60 static const stateid_t zero_stateid = {
  61         /* all fields zero */
  62 };
  63 static const stateid_t currentstateid = {
  64         .si_generation = 1,
  65 };
  66 static const stateid_t close_stateid = {
  67         .si_generation = 0xffffffffU,
  68 };
  69
  70 static u64 current_sessionid = 1;
  71
  72 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  73 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  74 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  75 #define CLOSE_STATEID(stateid)  (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
  76
  77 /* forward declarations */
  78 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  79 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  80
  81 /* Locking: */
  82
  83 /*
  84  * Currently used for the del_recall_lru and file hash table.  In an
  85  * effort to decrease the scope of the client_mutex, this spinlock may
  86  * eventually cover more:
  87  */
  88 static DEFINE_SPINLOCK(state_lock);
  89
  90 enum nfsd4_st_mutex_lock_subclass {
  91         OPEN_STATEID_MUTEX = 0,
  92         LOCK_STATEID_MUTEX = 1,
  93 };
  94
  95 /*
  96  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
  97  * the refcount on the open stateid to drop.
  98  */
  99 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
 100
 101 static struct kmem_cache *client_slab;
 102 static struct kmem_cache *openowner_slab;
 103 static struct kmem_cache *lockowner_slab;
 104 static struct kmem_cache *file_slab;
 105 static struct kmem_cache *stateid_slab;
 106 static struct kmem_cache *deleg_slab;
 107 static struct kmem_cache *odstate_slab;
 108
 109 static void free_session(struct nfsd4_session *);
 110
 111 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 112 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
 113
 114 static bool is_session_dead(struct nfsd4_session *ses)
 115 {
 116         return ses->se_flags & NFS4_SESSION_DEAD;
 117 }
 118
 119 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 120 {
 121         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 122                 return nfserr_jukebox;
 123         ses->se_flags |= NFS4_SESSION_DEAD;
 124         return nfs_ok;
 125 }
 126
 127 static bool is_client_expired(struct nfs4_client *clp)
 128 {
 129         return clp->cl_time == 0;
 130 }
 131
 132 static __be32 get_client_locked(struct nfs4_client *clp)
 133 {
 134         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 135
 136         lockdep_assert_held(&nn->client_lock);
 137
 138         if (is_client_expired(clp))
 139                 return nfserr_expired;
 140         atomic_inc(&clp->cl_refcount);
 141         return nfs_ok;
 142 }
 143
 144 /* must be called under the client_lock */
 145 static inline void
 146 renew_client_locked(struct nfs4_client *clp)
 147 {
 148         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 149
 150         if (is_client_expired(clp)) {
 151                 WARN_ON(1);
 152                 printk("%s: client (clientid %08x/%08x) already expired\n",
 153                         __func__,
 154                         clp->cl_clientid.cl_boot,
 155                         clp->cl_clientid.cl_id);
 156                 return;
 157         }
 158
 159         dprintk("renewing client (clientid %08x/%08x)\n",
 160                         clp->cl_clientid.cl_boot,
 161                         clp->cl_clientid.cl_id);
 162         list_move_tail(&clp->cl_lru, &nn->client_lru);
 163         clp->cl_time = get_seconds();
 164 }
 165
 166 static void put_client_renew_locked(struct nfs4_client *clp)
 167 {
 168         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 169
 170         lockdep_assert_held(&nn->client_lock);
 171
 172         if (!atomic_dec_and_test(&clp->cl_refcount))
 173                 return;
 174         if (!is_client_expired(clp))
 175                 renew_client_locked(clp);
 176 }
 177
 178 static void put_client_renew(struct nfs4_client *clp)
 179 {
 180         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 181
 182         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
 183                 return;
 184         if (!is_client_expired(clp))
 185                 renew_client_locked(clp);
 186         spin_unlock(&nn->client_lock);
 187 }
 188
 189 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 190 {
 191         __be32 status;
 192
 193         if (is_session_dead(ses))
 194                 return nfserr_badsession;
 195         status = get_client_locked(ses->se_client);
 196         if (status)
 197                 return status;
 198         atomic_inc(&ses->se_ref);
 199         return nfs_ok;
 200 }
 201
 202 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 203 {
 204         struct nfs4_client *clp = ses->se_client;
 205         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 206
 207         lockdep_assert_held(&nn->client_lock);
 208
 209         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 210                 free_session(ses);
 211         put_client_renew_locked(clp);
 212 }
 213
 214 static void nfsd4_put_session(struct nfsd4_session *ses)
 215 {
 216         struct nfs4_client *clp = ses->se_client;
 217         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 218
 219         spin_lock(&nn->client_lock);
 220         nfsd4_put_session_locked(ses);
 221         spin_unlock(&nn->client_lock);
 222 }
 223
 224 static struct nfsd4_blocked_lock *
 225 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 226                         struct nfsd_net *nn)
 227 {
 228         struct nfsd4_blocked_lock *cur, *found = NULL;
 229
 230         spin_lock(&nn->blocked_locks_lock);
 231         list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
 232                 if (fh_match(fh, &cur->nbl_fh)) {
 233                         list_del_init(&cur->nbl_list);
 234                         list_del_init(&cur->nbl_lru);
 235                         found = cur;
 236                         break;
 237                 }
 238         }
 239         spin_unlock(&nn->blocked_locks_lock);
 240         if (found)
 241                 locks_delete_block(&found->nbl_lock);
 242         return found;
 243 }
 244
 245 static struct nfsd4_blocked_lock *
 246 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 247                         struct nfsd_net *nn)
 248 {
 249         struct nfsd4_blocked_lock *nbl;
 250
 251         nbl = find_blocked_lock(lo, fh, nn);
 252         if (!nbl) {
 253                 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
 254                 if (nbl) {
 255                         fh_copy_shallow(&nbl->nbl_fh, fh);
 256                         locks_init_lock(&nbl->nbl_lock);
 257                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
 258                                         &nfsd4_cb_notify_lock_ops,
 259                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
 260                 }
 261         }
 262         return nbl;
 263 }
 264
 265 static void
 266 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
 267 {
 268         locks_delete_block(&nbl->nbl_lock);
 269         locks_release_private(&nbl->nbl_lock);
 270         kfree(nbl);
 271 }
 272
 273 static void
 274 remove_blocked_locks(struct nfs4_lockowner *lo)
 275 {
 276         struct nfs4_client *clp = lo->lo_owner.so_client;
 277         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 278         struct nfsd4_blocked_lock *nbl;
 279         LIST_HEAD(reaplist);
 280
 281         /* Dequeue all blocked locks */
 282         spin_lock(&nn->blocked_locks_lock);
 283         while (!list_empty(&lo->lo_blocked)) {
 284                 nbl = list_first_entry(&lo->lo_blocked,
 285                                         struct nfsd4_blocked_lock,
 286                                         nbl_list);
 287                 list_del_init(&nbl->nbl_list);
 288                 list_move(&nbl->nbl_lru, &reaplist);
 289         }
 290         spin_unlock(&nn->blocked_locks_lock);
 291
 292         /* Now free them */
 293         while (!list_empty(&reaplist)) {
 294                 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
 295                                         nbl_lru);
 296                 list_del_init(&nbl->nbl_lru);
 297                 free_blocked_lock(nbl);
 298         }
 299 }
 300
 301 static void
 302 nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
 303 {
 304         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 305                                                 struct nfsd4_blocked_lock, nbl_cb);
 306         locks_delete_block(&nbl->nbl_lock);
 307 }
 308
 309 static int
 310 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
 311 {
 312         /*
 313          * Since this is just an optimization, we don't try very hard if it
 314          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
 315          * just quit trying on anything else.
 316          */
 317         switch (task->tk_status) {
 318         case -NFS4ERR_DELAY:
 319                 rpc_delay(task, 1 * HZ);
 320                 return 0;
 321         default:
 322                 return 1;
 323         }
 324 }
 325
 326 static void
 327 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
 328 {
 329         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 330                                                 struct nfsd4_blocked_lock, nbl_cb);
 331
 332         free_blocked_lock(nbl);
 333 }
 334
 335 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
 336         .prepare        = nfsd4_cb_notify_lock_prepare,
 337         .done           = nfsd4_cb_notify_lock_done,
 338         .release        = nfsd4_cb_notify_lock_release,
 339 };
 340
 341 static inline struct nfs4_stateowner *
 342 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 343 {
 344         atomic_inc(&sop->so_count);
 345         return sop;
 346 }
 347
 348 static int
 349 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 350 {
 351         return (sop->so_owner.len == owner->len) &&
 352                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 353 }
 354
 355 static struct nfs4_openowner *
 356 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 357                         struct nfs4_client *clp)
 358 {
 359         struct nfs4_stateowner *so;
 360
 361         lockdep_assert_held(&clp->cl_lock);
 362
 363         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 364                             so_strhash) {
 365                 if (!so->so_is_open_owner)
 366                         continue;
 367                 if (same_owner_str(so, &open->op_owner))
 368                         return openowner(nfs4_get_stateowner(so));
 369         }
 370         return NULL;
 371 }
 372
 373 static struct nfs4_openowner *
 374 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 375                         struct nfs4_client *clp)
 376 {
 377         struct nfs4_openowner *oo;
 378
 379         spin_lock(&clp->cl_lock);
 380         oo = find_openstateowner_str_locked(hashval, open, clp);
 381         spin_unlock(&clp->cl_lock);
 382         return oo;
 383 }
 384
 385 static inline u32
 386 opaque_hashval(const void *ptr, int nbytes)
 387 {
 388         unsigned char *cptr = (unsigned char *) ptr;
 389
 390         u32 x = 0;
 391         while (nbytes--) {
 392                 x *= 37;
 393                 x += *cptr++;
 394         }
 395         return x;
 396 }
 397
 398 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 399 {
 400         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 401
 402         kmem_cache_free(file_slab, fp);
 403 }
 404
 405 void
 406 put_nfs4_file(struct nfs4_file *fi)
 407 {
 408         might_lock(&state_lock);
 409
 410         if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
 411                 hlist_del_rcu(&fi->fi_hash);
 412                 spin_unlock(&state_lock);
 413                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 414                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 415                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 416         }
 417 }
 418
 419 static struct file *
 420 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 421 {
 422         if (f->fi_fds[oflag])
 423                 return get_file(f->fi_fds[oflag]);
 424         return NULL;
 425 }
 426
 427 static struct file *
 428 find_writeable_file_locked(struct nfs4_file *f)
 429 {
 430         struct file *ret;
 431
 432         lockdep_assert_held(&f->fi_lock);
 433
 434         ret = __nfs4_get_fd(f, O_WRONLY);
 435         if (!ret)
 436                 ret = __nfs4_get_fd(f, O_RDWR);
 437         return ret;
 438 }
 439
 440 static struct file *
 441 find_writeable_file(struct nfs4_file *f)
 442 {
 443         struct file *ret;
 444
 445         spin_lock(&f->fi_lock);
 446         ret = find_writeable_file_locked(f);
 447         spin_unlock(&f->fi_lock);
 448
 449         return ret;
 450 }
 451
 452 static struct file *find_readable_file_locked(struct nfs4_file *f)
 453 {
 454         struct file *ret;
 455
 456         lockdep_assert_held(&f->fi_lock);
 457
 458         ret = __nfs4_get_fd(f, O_RDONLY);
 459         if (!ret)
 460                 ret = __nfs4_get_fd(f, O_RDWR);
 461         return ret;
 462 }
 463
 464 static struct file *
 465 find_readable_file(struct nfs4_file *f)
 466 {
 467         struct file *ret;
 468
 469         spin_lock(&f->fi_lock);
 470         ret = find_readable_file_locked(f);
 471         spin_unlock(&f->fi_lock);
 472
 473         return ret;
 474 }
 475
 476 struct file *
 477 find_any_file(struct nfs4_file *f)
 478 {
 479         struct file *ret;
 480
 481         spin_lock(&f->fi_lock);
 482         ret = __nfs4_get_fd(f, O_RDWR);
 483         if (!ret) {
 484                 ret = __nfs4_get_fd(f, O_WRONLY);
 485                 if (!ret)
 486                         ret = __nfs4_get_fd(f, O_RDONLY);
 487         }
 488         spin_unlock(&f->fi_lock);
 489         return ret;
 490 }
 491
 492 static atomic_long_t num_delegations;
 493 unsigned long max_delegations;
 494
 495 /*
 496  * Open owner state (share locks)
 497  */
 498
 499 /* hash tables for lock and open owners */
 500 #define OWNER_HASH_BITS              8
 501 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 502 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 503
 504 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 505 {
 506         unsigned int ret;
 507
 508         ret = opaque_hashval(ownername->data, ownername->len);
 509         return ret & OWNER_HASH_MASK;
 510 }
 511
 512 /* hash table for nfs4_file */
 513 #define FILE_HASH_BITS                   8
 514 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 515
 516 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 517 {
 518         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 519 }
 520
 521 static unsigned int file_hashval(struct knfsd_fh *fh)
 522 {
 523         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 524 }
 525
 526 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 527
 528 static void
 529 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 530 {
 531         lockdep_assert_held(&fp->fi_lock);
 532
 533         if (access & NFS4_SHARE_ACCESS_WRITE)
 534                 atomic_inc(&fp->fi_access[O_WRONLY]);
 535         if (access & NFS4_SHARE_ACCESS_READ)
 536                 atomic_inc(&fp->fi_access[O_RDONLY]);
 537 }
 538
 539 static __be32
 540 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 541 {
 542         lockdep_assert_held(&fp->fi_lock);
 543
 544         /* Does this access mode make sense? */
 545         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 546                 return nfserr_inval;
 547
 548         /* Does it conflict with a deny mode already set? */
 549         if ((access & fp->fi_share_deny) != 0)
 550                 return nfserr_share_denied;
 551
 552         __nfs4_file_get_access(fp, access);
 553         return nfs_ok;
 554 }
 555
 556 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 557 {
 558         /* Common case is that there is no deny mode. */
 559         if (deny) {
 560                 /* Does this deny mode make sense? */
 561                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 562                         return nfserr_inval;
 563
 564                 if ((deny & NFS4_SHARE_DENY_READ) &&
 565                     atomic_read(&fp->fi_access[O_RDONLY]))
 566                         return nfserr_share_denied;
 567
 568                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 569                     atomic_read(&fp->fi_access[O_WRONLY]))
 570                         return nfserr_share_denied;
 571         }
 572         return nfs_ok;
 573 }
 574
 575 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 576 {
 577         might_lock(&fp->fi_lock);
 578
 579         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 580                 struct file *f1 = NULL;
 581                 struct file *f2 = NULL;
 582
 583                 swap(f1, fp->fi_fds[oflag]);
 584                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 585                         swap(f2, fp->fi_fds[O_RDWR]);
 586                 spin_unlock(&fp->fi_lock);
 587                 if (f1)
 588                         fput(f1);
 589                 if (f2)
 590                         fput(f2);
 591         }
 592 }
 593
 594 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 595 {
 596         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 597
 598         if (access & NFS4_SHARE_ACCESS_WRITE)
 599                 __nfs4_file_put_access(fp, O_WRONLY);
 600         if (access & NFS4_SHARE_ACCESS_READ)
 601                 __nfs4_file_put_access(fp, O_RDONLY);
 602 }
 603
 604 /*
 605  * Allocate a new open/delegation state counter. This is needed for
 606  * pNFS for proper return on close semantics.
 607  *
 608  * Note that we only allocate it for pNFS-enabled exports, otherwise
 609  * all pointers to struct nfs4_clnt_odstate are always NULL.
 610  */
 611 static struct nfs4_clnt_odstate *
 612 alloc_clnt_odstate(struct nfs4_client *clp)
 613 {
 614         struct nfs4_clnt_odstate *co;
 615
 616         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 617         if (co) {
 618                 co->co_client = clp;
 619                 refcount_set(&co->co_odcount, 1);
 620         }
 621         return co;
 622 }
 623
 624 static void
 625 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 626 {
 627         struct nfs4_file *fp = co->co_file;
 628
 629         lockdep_assert_held(&fp->fi_lock);
 630         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 631 }
 632
 633 static inline void
 634 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 635 {
 636         if (co)
 637                 refcount_inc(&co->co_odcount);
 638 }
 639
 640 static void
 641 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 642 {
 643         struct nfs4_file *fp;
 644
 645         if (!co)
 646                 return;
 647
 648         fp = co->co_file;
 649         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 650                 list_del(&co->co_perfile);
 651                 spin_unlock(&fp->fi_lock);
 652
 653                 nfsd4_return_all_file_layouts(co->co_client, fp);
 654                 kmem_cache_free(odstate_slab, co);
 655         }
 656 }
 657
 658 static struct nfs4_clnt_odstate *
 659 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 660 {
 661         struct nfs4_clnt_odstate *co;
 662         struct nfs4_client *cl;
 663
 664         if (!new)
 665                 return NULL;
 666
 667         cl = new->co_client;
 668
 669         spin_lock(&fp->fi_lock);
 670         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 671                 if (co->co_client == cl) {
 672                         get_clnt_odstate(co);
 673                         goto out;
 674                 }
 675         }
 676         co = new;
 677         co->co_file = fp;
 678         hash_clnt_odstate_locked(new);
 679 out:
 680         spin_unlock(&fp->fi_lock);
 681         return co;
 682 }
 683
 684 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 685                                   void (*sc_free)(struct nfs4_stid *))
 686 {
 687         struct nfs4_stid *stid;
 688         int new_id;
 689
 690         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 691         if (!stid)
 692                 return NULL;
 693
 694         idr_preload(GFP_KERNEL);
 695         spin_lock(&cl->cl_lock);
 696         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
 697         spin_unlock(&cl->cl_lock);
 698         idr_preload_end();
 699         if (new_id < 0)
 700                 goto out_free;
 701
 702         stid->sc_free = sc_free;
 703         stid->sc_client = cl;
 704         stid->sc_stateid.si_opaque.so_id = new_id;
 705         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 706         /* Will be incremented before return to client: */
 707         refcount_set(&stid->sc_count, 1);
 708         spin_lock_init(&stid->sc_lock);
 709
 710         /*
 711          * It shouldn't be a problem to reuse an opaque stateid value.
 712          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 713          * example, a stray write retransmission could be accepted by
 714          * the server when it should have been rejected.  Therefore,
 715          * adopt a trick from the sctp code to attempt to maximize the
 716          * amount of time until an id is reused, by ensuring they always
 717          * "increase" (mod INT_MAX):
 718          */
 719         return stid;
 720 out_free:
 721         kmem_cache_free(slab, stid);
 722         return NULL;
 723 }
 724
 725 /*
 726  * Create a unique stateid_t to represent each COPY.
 727  */
 728 int nfs4_init_cp_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
 729 {
 730         int new_id;
 731
 732         idr_preload(GFP_KERNEL);
 733         spin_lock(&nn->s2s_cp_lock);
 734         new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, copy, 0, 0, GFP_NOWAIT);
 735         spin_unlock(&nn->s2s_cp_lock);
 736         idr_preload_end();
 737         if (new_id < 0)
 738                 return 0;
 739         copy->cp_stateid.si_opaque.so_id = new_id;
 740         copy->cp_stateid.si_opaque.so_clid.cl_boot = nn->boot_time;
 741         copy->cp_stateid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
 742         return 1;
 743 }
 744
 745 void nfs4_free_cp_state(struct nfsd4_copy *copy)
 746 {
 747         struct nfsd_net *nn;
 748
 749         nn = net_generic(copy->cp_clp->net, nfsd_net_id);
 750         spin_lock(&nn->s2s_cp_lock);
 751         idr_remove(&nn->s2s_cp_stateids, copy->cp_stateid.si_opaque.so_id);
 752         spin_unlock(&nn->s2s_cp_lock);
 753 }
 754
 755 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 756 {
 757         struct nfs4_stid *stid;
 758
 759         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
 760         if (!stid)
 761                 return NULL;
 762
 763         return openlockstateid(stid);
 764 }
 765
 766 static void nfs4_free_deleg(struct nfs4_stid *stid)
 767 {
 768         kmem_cache_free(deleg_slab, stid);
 769         atomic_long_dec(&num_delegations);
 770 }
 771
 772 /*
 773  * When we recall a delegation, we should be careful not to hand it
 774  * out again straight away.
 775  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 776  * in which the filehandles of recalled delegations are "stored".
 777  * If a filehandle appear in either filter, a delegation is blocked.
 778  * When a delegation is recalled, the filehandle is stored in the "new"
 779  * filter.
 780  * Every 30 seconds we swap the filters and clear the "new" one,
 781  * unless both are empty of course.
 782  *
 783  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 784  * low 3 bytes as hash-table indices.
 785  *
 786  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 787  * is used to manage concurrent access.  Testing does not need the lock
 788  * except when swapping the two filters.
 789  */
 790 static DEFINE_SPINLOCK(blocked_delegations_lock);
 791 static struct bloom_pair {
 792         int     entries, old_entries;
 793         time_t  swap_time;
 794         int     new; /* index into 'set' */
 795         DECLARE_BITMAP(set[2], 256);
 796 } blocked_delegations;
 797
 798 static int delegation_blocked(struct knfsd_fh *fh)
 799 {
 800         u32 hash;
 801         struct bloom_pair *bd = &blocked_delegations;
 802
 803         if (bd->entries == 0)
 804                 return 0;
 805         if (seconds_since_boot() - bd->swap_time > 30) {
 806                 spin_lock(&blocked_delegations_lock);
 807                 if (seconds_since_boot() - bd->swap_time > 30) {
 808                         bd->entries -= bd->old_entries;
 809                         bd->old_entries = bd->entries;
 810                         memset(bd->set[bd->new], 0,
 811                                sizeof(bd->set[0]));
 812                         bd->new = 1-bd->new;
 813                         bd->swap_time = seconds_since_boot();
 814                 }
 815                 spin_unlock(&blocked_delegations_lock);
 816         }
 817         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 818         if (test_bit(hash&255, bd->set[0]) &&
 819             test_bit((hash>>8)&255, bd->set[0]) &&
 820             test_bit((hash>>16)&255, bd->set[0]))
 821                 return 1;
 822
 823         if (test_bit(hash&255, bd->set[1]) &&
 824             test_bit((hash>>8)&255, bd->set[1]) &&
 825             test_bit((hash>>16)&255, bd->set[1]))
 826                 return 1;
 827
 828         return 0;
 829 }
 830
 831 static void block_delegations(struct knfsd_fh *fh)
 832 {
 833         u32 hash;
 834         struct bloom_pair *bd = &blocked_delegations;
 835
 836         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 837
 838         spin_lock(&blocked_delegations_lock);
 839         __set_bit(hash&255, bd->set[bd->new]);
 840         __set_bit((hash>>8)&255, bd->set[bd->new]);
 841         __set_bit((hash>>16)&255, bd->set[bd->new]);
 842         if (bd->entries == 0)
 843                 bd->swap_time = seconds_since_boot();
 844         bd->entries += 1;
 845         spin_unlock(&blocked_delegations_lock);
 846 }
 847
 848 static struct nfs4_delegation *
 849 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
 850                  struct svc_fh *current_fh,
 851                  struct nfs4_clnt_odstate *odstate)
 852 {
 853         struct nfs4_delegation *dp;
 854         long n;
 855
 856         dprintk("NFSD alloc_init_deleg\n");
 857         n = atomic_long_inc_return(&num_delegations);
 858         if (n < 0 || n > max_delegations)
 859                 goto out_dec;
 860         if (delegation_blocked(&current_fh->fh_handle))
 861                 goto out_dec;
 862         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
 863         if (dp == NULL)
 864                 goto out_dec;
 865
 866         /*
 867          * delegation seqid's are never incremented.  The 4.1 special
 868          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 869          * 0 anyway just for consistency and use 1:
 870          */
 871         dp->dl_stid.sc_stateid.si_generation = 1;
 872         INIT_LIST_HEAD(&dp->dl_perfile);
 873         INIT_LIST_HEAD(&dp->dl_perclnt);
 874         INIT_LIST_HEAD(&dp->dl_recall_lru);
 875         dp->dl_clnt_odstate = odstate;
 876         get_clnt_odstate(odstate);
 877         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 878         dp->dl_retries = 1;
 879         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 880                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 881         get_nfs4_file(fp);
 882         dp->dl_stid.sc_file = fp;
 883         return dp;
 884 out_dec:
 885         atomic_long_dec(&num_delegations);
 886         return NULL;
 887 }
 888
 889 void
 890 nfs4_put_stid(struct nfs4_stid *s)
 891 {
 892         struct nfs4_file *fp = s->sc_file;
 893         struct nfs4_client *clp = s->sc_client;
 894
 895         might_lock(&clp->cl_lock);
 896
 897         if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 898                 wake_up_all(&close_wq);
 899                 return;
 900         }
 901         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 902         spin_unlock(&clp->cl_lock);
 903         s->sc_free(s);
 904         if (fp)
 905                 put_nfs4_file(fp);
 906 }
 907
 908 void
 909 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
 910 {
 911         stateid_t *src = &stid->sc_stateid;
 912
 913         spin_lock(&stid->sc_lock);
 914         if (unlikely(++src->si_generation == 0))
 915                 src->si_generation = 1;
 916         memcpy(dst, src, sizeof(*dst));
 917         spin_unlock(&stid->sc_lock);
 918 }
 919
 920 static void put_deleg_file(struct nfs4_file *fp)
 921 {
 922         struct file *filp = NULL;
 923
 924         spin_lock(&fp->fi_lock);
 925         if (--fp->fi_delegees == 0)
 926                 swap(filp, fp->fi_deleg_file);
 927         spin_unlock(&fp->fi_lock);
 928
 929         if (filp)
 930                 fput(filp);
 931 }
 932
 933 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
 934 {
 935         struct nfs4_file *fp = dp->dl_stid.sc_file;
 936         struct file *filp = fp->fi_deleg_file;
 937
 938         WARN_ON_ONCE(!fp->fi_delegees);
 939
 940         vfs_setlease(filp, F_UNLCK, NULL, (void **)&dp);
 941         put_deleg_file(fp);
 942 }
 943
 944 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
 945 {
 946         put_clnt_odstate(dp->dl_clnt_odstate);
 947         nfs4_unlock_deleg_lease(dp);
 948         nfs4_put_stid(&dp->dl_stid);
 949 }
 950
 951 void nfs4_unhash_stid(struct nfs4_stid *s)
 952 {
 953         s->sc_type = 0;
 954 }
 955
 956 /**
 957  * nfs4_delegation_exists - Discover if this delegation already exists
 958  * @clp:     a pointer to the nfs4_client we're granting a delegation to
 959  * @fp:      a pointer to the nfs4_file we're granting a delegation on
 960  *
 961  * Return:
 962  *      On success: true iff an existing delegation is found
 963  */
 964
 965 static bool
 966 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
 967 {
 968         struct nfs4_delegation *searchdp = NULL;
 969         struct nfs4_client *searchclp = NULL;
 970
 971         lockdep_assert_held(&state_lock);
 972         lockdep_assert_held(&fp->fi_lock);
 973
 974         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
 975                 searchclp = searchdp->dl_stid.sc_client;
 976                 if (clp == searchclp) {
 977                         return true;
 978                 }
 979         }
 980         return false;
 981 }
 982
 983 /**
 984  * hash_delegation_locked - Add a delegation to the appropriate lists
 985  * @dp:     a pointer to the nfs4_delegation we are adding.
 986  * @fp:     a pointer to the nfs4_file we're granting a delegation on
 987  *
 988  * Return:
 989  *      On success: NULL if the delegation was successfully hashed.
 990  *
 991  *      On error: -EAGAIN if one was previously granted to this
 992  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
 993  *
 994  */
 995
 996 static int
 997 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
 998 {
 999         struct nfs4_client *clp = dp->dl_stid.sc_client;
1000
1001         lockdep_assert_held(&state_lock);
1002         lockdep_assert_held(&fp->fi_lock);
1003
1004         if (nfs4_delegation_exists(clp, fp))
1005                 return -EAGAIN;
1006         refcount_inc(&dp->dl_stid.sc_count);
1007         dp->dl_stid.sc_type = NFS4_DELEG_STID;
1008         list_add(&dp->dl_perfile, &fp->fi_delegations);
1009         list_add(&dp->dl_perclnt, &clp->cl_delegations);
1010         return 0;
1011 }
1012
1013 static bool
1014 unhash_delegation_locked(struct nfs4_delegation *dp)
1015 {
1016         struct nfs4_file *fp = dp->dl_stid.sc_file;
1017
1018         lockdep_assert_held(&state_lock);
1019
1020         if (list_empty(&dp->dl_perfile))
1021                 return false;
1022
1023         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
1024         /* Ensure that deleg break won't try to requeue it */
1025         ++dp->dl_time;
1026         spin_lock(&fp->fi_lock);
1027         list_del_init(&dp->dl_perclnt);
1028         list_del_init(&dp->dl_recall_lru);
1029         list_del_init(&dp->dl_perfile);
1030         spin_unlock(&fp->fi_lock);
1031         return true;
1032 }
1033
1034 static void destroy_delegation(struct nfs4_delegation *dp)
1035 {
1036         bool unhashed;
1037
1038         spin_lock(&state_lock);
1039         unhashed = unhash_delegation_locked(dp);
1040         spin_unlock(&state_lock);
1041         if (unhashed)
1042                 destroy_unhashed_deleg(dp);
1043 }
1044
1045 static void revoke_delegation(struct nfs4_delegation *dp)
1046 {
1047         struct nfs4_client *clp = dp->dl_stid.sc_client;
1048
1049         WARN_ON(!list_empty(&dp->dl_recall_lru));
1050
1051         if (clp->cl_minorversion) {
1052                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1053                 refcount_inc(&dp->dl_stid.sc_count);
1054                 spin_lock(&clp->cl_lock);
1055                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1056                 spin_unlock(&clp->cl_lock);
1057         }
1058         destroy_unhashed_deleg(dp);
1059 }
1060
1061 /*
1062  * SETCLIENTID state
1063  */
1064
1065 static unsigned int clientid_hashval(u32 id)
1066 {
1067         return id & CLIENT_HASH_MASK;
1068 }
1069
1070 static unsigned int clientstr_hashval(const char *name)
1071 {
1072         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
1073 }
1074
1075 /*
1076  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1077  * st_{access,deny}_bmap field of the stateid, in order to track not
1078  * only what share bits are currently in force, but also what
1079  * combinations of share bits previous opens have used.  This allows us
1080  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1081  * return an error if the client attempt to downgrade to a combination
1082  * of share bits not explicable by closing some of its previous opens.
1083  *
1084  * XXX: This enforcement is actually incomplete, since we don't keep
1085  * track of access/deny bit combinations; so, e.g., we allow:
1086  *
1087  *      OPEN allow read, deny write
1088  *      OPEN allow both, deny none
1089  *      DOWNGRADE allow read, deny none
1090  *
1091  * which we should reject.
1092  */
1093 static unsigned int
1094 bmap_to_share_mode(unsigned long bmap) {
1095         int i;
1096         unsigned int access = 0;
1097
1098         for (i = 1; i < 4; i++) {
1099                 if (test_bit(i, &bmap))
1100                         access |= i;
1101         }
1102         return access;
1103 }
1104
1105 /* set share access for a given stateid */
1106 static inline void
1107 set_access(u32 access, struct nfs4_ol_stateid *stp)
1108 {
1109         unsigned char mask = 1 << access;
1110
1111         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1112         stp->st_access_bmap |= mask;
1113 }
1114
1115 /* clear share access for a given stateid */
1116 static inline void
1117 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1118 {
1119         unsigned char mask = 1 << access;
1120
1121         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1122         stp->st_access_bmap &= ~mask;
1123 }
1124
1125 /* test whether a given stateid has access */
1126 static inline bool
1127 test_access(u32 access, struct nfs4_ol_stateid *stp)
1128 {
1129         unsigned char mask = 1 << access;
1130
1131         return (bool)(stp->st_access_bmap & mask);
1132 }
1133
1134 /* set share deny for a given stateid */
1135 static inline void
1136 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1137 {
1138         unsigned char mask = 1 << deny;
1139
1140         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1141         stp->st_deny_bmap |= mask;
1142 }
1143
1144 /* clear share deny for a given stateid */
1145 static inline void
1146 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1147 {
1148         unsigned char mask = 1 << deny;
1149
1150         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1151         stp->st_deny_bmap &= ~mask;
1152 }
1153
1154 /* test whether a given stateid is denying specific access */
1155 static inline bool
1156 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1157 {
1158         unsigned char mask = 1 << deny;
1159
1160         return (bool)(stp->st_deny_bmap & mask);
1161 }
1162
1163 static int nfs4_access_to_omode(u32 access)
1164 {
1165         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1166         case NFS4_SHARE_ACCESS_READ:
1167                 return O_RDONLY;
1168         case NFS4_SHARE_ACCESS_WRITE:
1169                 return O_WRONLY;
1170         case NFS4_SHARE_ACCESS_BOTH:
1171                 return O_RDWR;
1172         }
1173         WARN_ON_ONCE(1);
1174         return O_RDONLY;
1175 }
1176
1177 /*
1178  * A stateid that had a deny mode associated with it is being released
1179  * or downgraded. Recalculate the deny mode on the file.
1180  */
1181 static void
1182 recalculate_deny_mode(struct nfs4_file *fp)
1183 {
1184         struct nfs4_ol_stateid *stp;
1185
1186         spin_lock(&fp->fi_lock);
1187         fp->fi_share_deny = 0;
1188         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1189                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1190         spin_unlock(&fp->fi_lock);
1191 }
1192
1193 static void
1194 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1195 {
1196         int i;
1197         bool change = false;
1198
1199         for (i = 1; i < 4; i++) {
1200                 if ((i & deny) != i) {
1201                         change = true;
1202                         clear_deny(i, stp);
1203                 }
1204         }
1205
1206         /* Recalculate per-file deny mode if there was a change */
1207         if (change)
1208                 recalculate_deny_mode(stp->st_stid.sc_file);
1209 }
1210
1211 /* release all access and file references for a given stateid */
1212 static void
1213 release_all_access(struct nfs4_ol_stateid *stp)
1214 {
1215         int i;
1216         struct nfs4_file *fp = stp->st_stid.sc_file;
1217
1218         if (fp && stp->st_deny_bmap != 0)
1219                 recalculate_deny_mode(fp);
1220
1221         for (i = 1; i < 4; i++) {
1222                 if (test_access(i, stp))
1223                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1224                 clear_access(i, stp);
1225         }
1226 }
1227
1228 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1229 {
1230         kfree(sop->so_owner.data);
1231         sop->so_ops->so_free(sop);
1232 }
1233
1234 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1235 {
1236         struct nfs4_client *clp = sop->so_client;
1237
1238         might_lock(&clp->cl_lock);
1239
1240         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1241                 return;
1242         sop->so_ops->so_unhash(sop);
1243         spin_unlock(&clp->cl_lock);
1244         nfs4_free_stateowner(sop);
1245 }
1246
1247 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1248 {
1249         struct nfs4_file *fp = stp->st_stid.sc_file;
1250
1251         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1252
1253         if (list_empty(&stp->st_perfile))
1254                 return false;
1255
1256         spin_lock(&fp->fi_lock);
1257         list_del_init(&stp->st_perfile);
1258         spin_unlock(&fp->fi_lock);
1259         list_del(&stp->st_perstateowner);
1260         return true;
1261 }
1262
1263 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1264 {
1265         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1266
1267         put_clnt_odstate(stp->st_clnt_odstate);
1268         release_all_access(stp);
1269         if (stp->st_stateowner)
1270                 nfs4_put_stateowner(stp->st_stateowner);
1271         kmem_cache_free(stateid_slab, stid);
1272 }
1273
1274 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1275 {
1276         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1277         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1278         struct file *file;
1279
1280         file = find_any_file(stp->st_stid.sc_file);
1281         if (file)
1282                 filp_close(file, (fl_owner_t)lo);
1283         nfs4_free_ol_stateid(stid);
1284 }
1285
1286 /*
1287  * Put the persistent reference to an already unhashed generic stateid, while
1288  * holding the cl_lock. If it's the last reference, then put it onto the
1289  * reaplist for later destruction.
1290  */
1291 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1292                                        struct list_head *reaplist)
1293 {
1294         struct nfs4_stid *s = &stp->st_stid;
1295         struct nfs4_client *clp = s->sc_client;
1296
1297         lockdep_assert_held(&clp->cl_lock);
1298
1299         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1300
1301         if (!refcount_dec_and_test(&s->sc_count)) {
1302                 wake_up_all(&close_wq);
1303                 return;
1304         }
1305
1306         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1307         list_add(&stp->st_locks, reaplist);
1308 }
1309
1310 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1311 {
1312         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1313
1314         list_del_init(&stp->st_locks);
1315         nfs4_unhash_stid(&stp->st_stid);
1316         return unhash_ol_stateid(stp);
1317 }
1318
1319 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1320 {
1321         struct nfs4_client *clp = stp->st_stid.sc_client;
1322         bool unhashed;
1323
1324         spin_lock(&clp->cl_lock);
1325         unhashed = unhash_lock_stateid(stp);
1326         spin_unlock(&clp->cl_lock);
1327         if (unhashed)
1328                 nfs4_put_stid(&stp->st_stid);
1329 }
1330
1331 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1332 {
1333         struct nfs4_client *clp = lo->lo_owner.so_client;
1334
1335         lockdep_assert_held(&clp->cl_lock);
1336
1337         list_del_init(&lo->lo_owner.so_strhash);
1338 }
1339
1340 /*
1341  * Free a list of generic stateids that were collected earlier after being
1342  * fully unhashed.
1343  */
1344 static void
1345 free_ol_stateid_reaplist(struct list_head *reaplist)
1346 {
1347         struct nfs4_ol_stateid *stp;
1348         struct nfs4_file *fp;
1349
1350         might_sleep();
1351
1352         while (!list_empty(reaplist)) {
1353                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1354                                        st_locks);
1355                 list_del(&stp->st_locks);
1356                 fp = stp->st_stid.sc_file;
1357                 stp->st_stid.sc_free(&stp->st_stid);
1358                 if (fp)
1359                         put_nfs4_file(fp);
1360         }
1361 }
1362
1363 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1364                                        struct list_head *reaplist)
1365 {
1366         struct nfs4_ol_stateid *stp;
1367
1368         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1369
1370         while (!list_empty(&open_stp->st_locks)) {
1371                 stp = list_entry(open_stp->st_locks.next,
1372                                 struct nfs4_ol_stateid, st_locks);
1373                 WARN_ON(!unhash_lock_stateid(stp));
1374                 put_ol_stateid_locked(stp, reaplist);
1375         }
1376 }
1377
1378 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1379                                 struct list_head *reaplist)
1380 {
1381         bool unhashed;
1382
1383         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1384
1385         unhashed = unhash_ol_stateid(stp);
1386         release_open_stateid_locks(stp, reaplist);
1387         return unhashed;
1388 }
1389
1390 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1391 {
1392         LIST_HEAD(reaplist);
1393
1394         spin_lock(&stp->st_stid.sc_client->cl_lock);
1395         if (unhash_open_stateid(stp, &reaplist))
1396                 put_ol_stateid_locked(stp, &reaplist);
1397         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1398         free_ol_stateid_reaplist(&reaplist);
1399 }
1400
1401 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1402 {
1403         struct nfs4_client *clp = oo->oo_owner.so_client;
1404
1405         lockdep_assert_held(&clp->cl_lock);
1406
1407         list_del_init(&oo->oo_owner.so_strhash);
1408         list_del_init(&oo->oo_perclient);
1409 }
1410
1411 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1412 {
1413         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1414                                           nfsd_net_id);
1415         struct nfs4_ol_stateid *s;
1416
1417         spin_lock(&nn->client_lock);
1418         s = oo->oo_last_closed_stid;
1419         if (s) {
1420                 list_del_init(&oo->oo_close_lru);
1421                 oo->oo_last_closed_stid = NULL;
1422         }
1423         spin_unlock(&nn->client_lock);
1424         if (s)
1425                 nfs4_put_stid(&s->st_stid);
1426 }
1427
1428 static void release_openowner(struct nfs4_openowner *oo)
1429 {
1430         struct nfs4_ol_stateid *stp;
1431         struct nfs4_client *clp = oo->oo_owner.so_client;
1432         struct list_head reaplist;
1433
1434         INIT_LIST_HEAD(&reaplist);
1435
1436         spin_lock(&clp->cl_lock);
1437         unhash_openowner_locked(oo);
1438         while (!list_empty(&oo->oo_owner.so_stateids)) {
1439                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1440                                 struct nfs4_ol_stateid, st_perstateowner);
1441                 if (unhash_open_stateid(stp, &reaplist))
1442                         put_ol_stateid_locked(stp, &reaplist);
1443         }
1444         spin_unlock(&clp->cl_lock);
1445         free_ol_stateid_reaplist(&reaplist);
1446         release_last_closed_stateid(oo);
1447         nfs4_put_stateowner(&oo->oo_owner);
1448 }
1449
1450 static inline int
1451 hash_sessionid(struct nfs4_sessionid *sessionid)
1452 {
1453         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1454
1455         return sid->sequence % SESSION_HASH_SIZE;
1456 }
1457
1458 #ifdef CONFIG_SUNRPC_DEBUG
1459 static inline void
1460 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1461 {
1462         u32 *ptr = (u32 *)(&sessionid->data[0]);
1463         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1464 }
1465 #else
1466 static inline void
1467 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1468 {
1469 }
1470 #endif
1471
1472 /*
1473  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1474  * won't be used for replay.
1475  */
1476 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1477 {
1478         struct nfs4_stateowner *so = cstate->replay_owner;
1479
1480         if (nfserr == nfserr_replay_me)
1481                 return;
1482
1483         if (!seqid_mutating_err(ntohl(nfserr))) {
1484                 nfsd4_cstate_clear_replay(cstate);
1485                 return;
1486         }
1487         if (!so)
1488                 return;
1489         if (so->so_is_open_owner)
1490                 release_last_closed_stateid(openowner(so));
1491         so->so_seqid++;
1492         return;
1493 }
1494
1495 static void
1496 gen_sessionid(struct nfsd4_session *ses)
1497 {
1498         struct nfs4_client *clp = ses->se_client;
1499         struct nfsd4_sessionid *sid;
1500
1501         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1502         sid->clientid = clp->cl_clientid;
1503         sid->sequence = current_sessionid++;
1504         sid->reserved = 0;
1505 }
1506
1507 /*
1508  * The protocol defines ca_maxresponssize_cached to include the size of
1509  * the rpc header, but all we need to cache is the data starting after
1510  * the end of the initial SEQUENCE operation--the rest we regenerate
1511  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1512  * value that is the number of bytes in our cache plus a few additional
1513  * bytes.  In order to stay on the safe side, and not promise more than
1514  * we can cache, those additional bytes must be the minimum possible: 24
1515  * bytes of rpc header (xid through accept state, with AUTH_NULL
1516  * verifier), 12 for the compound header (with zero-length tag), and 44
1517  * for the SEQUENCE op response:
1518  */
1519 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1520
1521 static void
1522 free_session_slots(struct nfsd4_session *ses)
1523 {
1524         int i;
1525
1526         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1527                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1528                 kfree(ses->se_slots[i]);
1529         }
1530 }
1531
1532 /*
1533  * We don't actually need to cache the rpc and session headers, so we
1534  * can allocate a little less for each slot:
1535  */
1536 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1537 {
1538         u32 size;
1539
1540         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1541                 size = 0;
1542         else
1543                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1544         return size + sizeof(struct nfsd4_slot);
1545 }
1546
1547 /*
1548  * XXX: If we run out of reserved DRC memory we could (up to a point)
1549  * re-negotiate active sessions and reduce their slot usage to make
1550  * room for new connections. For now we just fail the create session.
1551  */
1552 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1553 {
1554         u32 slotsize = slot_bytes(ca);
1555         u32 num = ca->maxreqs;
1556         unsigned long avail, total_avail;
1557
1558         spin_lock(&nfsd_drc_lock);
1559         total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1560         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1561         /*
1562          * Never use more than a third of the remaining memory,
1563          * unless it's the only way to give this client a slot:
1564          */
1565         avail = clamp_t(int, avail, slotsize, total_avail/3);
1566         num = min_t(int, num, avail / slotsize);
1567         nfsd_drc_mem_used += num * slotsize;
1568         spin_unlock(&nfsd_drc_lock);
1569
1570         return num;
1571 }
1572
1573 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1574 {
1575         int slotsize = slot_bytes(ca);
1576
1577         spin_lock(&nfsd_drc_lock);
1578         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1579         spin_unlock(&nfsd_drc_lock);
1580 }
1581
1582 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1583                                            struct nfsd4_channel_attrs *battrs)
1584 {
1585         int numslots = fattrs->maxreqs;
1586         int slotsize = slot_bytes(fattrs);
1587         struct nfsd4_session *new;
1588         int mem, i;
1589
1590         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1591                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1592         mem = numslots * sizeof(struct nfsd4_slot *);
1593
1594         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1595         if (!new)
1596                 return NULL;
1597         /* allocate each struct nfsd4_slot and data cache in one piece */
1598         for (i = 0; i < numslots; i++) {
1599                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1600                 if (!new->se_slots[i])
1601                         goto out_free;
1602         }
1603
1604         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1605         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1606
1607         return new;
1608 out_free:
1609         while (i--)
1610                 kfree(new->se_slots[i]);
1611         kfree(new);
1612         return NULL;
1613 }
1614
1615 static void free_conn(struct nfsd4_conn *c)
1616 {
1617         svc_xprt_put(c->cn_xprt);
1618         kfree(c);
1619 }
1620
1621 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1622 {
1623         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1624         struct nfs4_client *clp = c->cn_session->se_client;
1625
1626         spin_lock(&clp->cl_lock);
1627         if (!list_empty(&c->cn_persession)) {
1628                 list_del(&c->cn_persession);
1629                 free_conn(c);
1630         }
1631         nfsd4_probe_callback(clp);
1632         spin_unlock(&clp->cl_lock);
1633 }
1634
1635 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1636 {
1637         struct nfsd4_conn *conn;
1638
1639         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1640         if (!conn)
1641                 return NULL;
1642         svc_xprt_get(rqstp->rq_xprt);
1643         conn->cn_xprt = rqstp->rq_xprt;
1644         conn->cn_flags = flags;
1645         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1646         return conn;
1647 }
1648
1649 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1650 {
1651         conn->cn_session = ses;
1652         list_add(&conn->cn_persession, &ses->se_conns);
1653 }
1654
1655 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1656 {
1657         struct nfs4_client *clp = ses->se_client;
1658
1659         spin_lock(&clp->cl_lock);
1660         __nfsd4_hash_conn(conn, ses);
1661         spin_unlock(&clp->cl_lock);
1662 }
1663
1664 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1665 {
1666         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1667         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1668 }
1669
1670 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1671 {
1672         int ret;
1673
1674         nfsd4_hash_conn(conn, ses);
1675         ret = nfsd4_register_conn(conn);
1676         if (ret)
1677                 /* oops; xprt is already down: */
1678                 nfsd4_conn_lost(&conn->cn_xpt_user);
1679         /* We may have gained or lost a callback channel: */
1680         nfsd4_probe_callback_sync(ses->se_client);
1681 }
1682
1683 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1684 {
1685         u32 dir = NFS4_CDFC4_FORE;
1686
1687         if (cses->flags & SESSION4_BACK_CHAN)
1688                 dir |= NFS4_CDFC4_BACK;
1689         return alloc_conn(rqstp, dir);
1690 }
1691
1692 /* must be called under client_lock */
1693 static void nfsd4_del_conns(struct nfsd4_session *s)
1694 {
1695         struct nfs4_client *clp = s->se_client;
1696         struct nfsd4_conn *c;
1697
1698         spin_lock(&clp->cl_lock);
1699         while (!list_empty(&s->se_conns)) {
1700                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1701                 list_del_init(&c->cn_persession);
1702                 spin_unlock(&clp->cl_lock);
1703
1704                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1705                 free_conn(c);
1706
1707                 spin_lock(&clp->cl_lock);
1708         }
1709         spin_unlock(&clp->cl_lock);
1710 }
1711
1712 static void __free_session(struct nfsd4_session *ses)
1713 {
1714         free_session_slots(ses);
1715         kfree(ses);
1716 }
1717
1718 static void free_session(struct nfsd4_session *ses)
1719 {
1720         nfsd4_del_conns(ses);
1721         nfsd4_put_drc_mem(&ses->se_fchannel);
1722         __free_session(ses);
1723 }
1724
1725 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1726 {
1727         int idx;
1728         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1729
1730         new->se_client = clp;
1731         gen_sessionid(new);
1732
1733         INIT_LIST_HEAD(&new->se_conns);
1734
1735         new->se_cb_seq_nr = 1;
1736         new->se_flags = cses->flags;
1737         new->se_cb_prog = cses->callback_prog;
1738         new->se_cb_sec = cses->cb_sec;
1739         atomic_set(&new->se_ref, 0);
1740         idx = hash_sessionid(&new->se_sessionid);
1741         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1742         spin_lock(&clp->cl_lock);
1743         list_add(&new->se_perclnt, &clp->cl_sessions);
1744         spin_unlock(&clp->cl_lock);
1745
1746         {
1747                 struct sockaddr *sa = svc_addr(rqstp);
1748                 /*
1749                  * This is a little silly; with sessions there's no real
1750                  * use for the callback address.  Use the peer address
1751                  * as a reasonable default for now, but consider fixing
1752                  * the rpc client not to require an address in the
1753                  * future:
1754                  */
1755                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1756                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1757         }
1758 }
1759
1760 /* caller must hold client_lock */
1761 static struct nfsd4_session *
1762 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1763 {
1764         struct nfsd4_session *elem;
1765         int idx;
1766         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1767
1768         lockdep_assert_held(&nn->client_lock);
1769
1770         dump_sessionid(__func__, sessionid);
1771         idx = hash_sessionid(sessionid);
1772         /* Search in the appropriate list */
1773         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1774                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1775                             NFS4_MAX_SESSIONID_LEN)) {
1776                         return elem;
1777                 }
1778         }
1779
1780         dprintk("%s: session not found\n", __func__);
1781         return NULL;
1782 }
1783
1784 static struct nfsd4_session *
1785 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1786                 __be32 *ret)
1787 {
1788         struct nfsd4_session *session;
1789         __be32 status = nfserr_badsession;
1790
1791         session = __find_in_sessionid_hashtbl(sessionid, net);
1792         if (!session)
1793                 goto out;
1794         status = nfsd4_get_session_locked(session);
1795         if (status)
1796                 session = NULL;
1797 out:
1798         *ret = status;
1799         return session;
1800 }
1801
1802 /* caller must hold client_lock */
1803 static void
1804 unhash_session(struct nfsd4_session *ses)
1805 {
1806         struct nfs4_client *clp = ses->se_client;
1807         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1808
1809         lockdep_assert_held(&nn->client_lock);
1810
1811         list_del(&ses->se_hash);
1812         spin_lock(&ses->se_client->cl_lock);
1813         list_del(&ses->se_perclnt);
1814         spin_unlock(&ses->se_client->cl_lock);
1815 }
1816
1817 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1818 static int
1819 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1820 {
1821         /*
1822          * We're assuming the clid was not given out from a boot
1823          * precisely 2^32 (about 136 years) before this one.  That seems
1824          * a safe assumption:
1825          */
1826         if (clid->cl_boot == (u32)nn->boot_time)
1827                 return 0;
1828         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1829                 clid->cl_boot, clid->cl_id, nn->boot_time);
1830         return 1;
1831 }
1832
1833 /*
1834  * XXX Should we use a slab cache ?
1835  * This type of memory management is somewhat inefficient, but we use it
1836  * anyway since SETCLIENTID is not a common operation.
1837  */
1838 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1839 {
1840         struct nfs4_client *clp;
1841         int i;
1842
1843         clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
1844         if (clp == NULL)
1845                 return NULL;
1846         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1847         if (clp->cl_name.data == NULL)
1848                 goto err_no_name;
1849         clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
1850                                                  sizeof(struct list_head),
1851                                                  GFP_KERNEL);
1852         if (!clp->cl_ownerstr_hashtbl)
1853                 goto err_no_hashtbl;
1854         for (i = 0; i < OWNER_HASH_SIZE; i++)
1855                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1856         clp->cl_name.len = name.len;
1857         INIT_LIST_HEAD(&clp->cl_sessions);
1858         idr_init(&clp->cl_stateids);
1859         atomic_set(&clp->cl_refcount, 0);
1860         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1861         INIT_LIST_HEAD(&clp->cl_idhash);
1862         INIT_LIST_HEAD(&clp->cl_openowners);
1863         INIT_LIST_HEAD(&clp->cl_delegations);
1864         INIT_LIST_HEAD(&clp->cl_lru);
1865         INIT_LIST_HEAD(&clp->cl_revoked);
1866 #ifdef CONFIG_NFSD_PNFS
1867         INIT_LIST_HEAD(&clp->cl_lo_states);
1868 #endif
1869         INIT_LIST_HEAD(&clp->async_copies);
1870         spin_lock_init(&clp->async_lock);
1871         spin_lock_init(&clp->cl_lock);
1872         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1873         return clp;
1874 err_no_hashtbl:
1875         kfree(clp->cl_name.data);
1876 err_no_name:
1877         kmem_cache_free(client_slab, clp);
1878         return NULL;
1879 }
1880
1881 static void
1882 free_client(struct nfs4_client *clp)
1883 {
1884         while (!list_empty(&clp->cl_sessions)) {
1885                 struct nfsd4_session *ses;
1886                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1887                                 se_perclnt);
1888                 list_del(&ses->se_perclnt);
1889                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1890                 free_session(ses);
1891         }
1892         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1893         free_svc_cred(&clp->cl_cred);
1894         kfree(clp->cl_ownerstr_hashtbl);
1895         kfree(clp->cl_name.data);
1896         idr_destroy(&clp->cl_stateids);
1897         kmem_cache_free(client_slab, clp);
1898 }
1899
1900 /* must be called under the client_lock */
1901 static void
1902 unhash_client_locked(struct nfs4_client *clp)
1903 {
1904         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1905         struct nfsd4_session *ses;
1906
1907         lockdep_assert_held(&nn->client_lock);
1908
1909         /* Mark the client as expired! */
1910         clp->cl_time = 0;
1911         /* Make it invisible */
1912         if (!list_empty(&clp->cl_idhash)) {
1913                 list_del_init(&clp->cl_idhash);
1914                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1915                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1916                 else
1917                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1918         }
1919         list_del_init(&clp->cl_lru);
1920         spin_lock(&clp->cl_lock);
1921         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1922                 list_del_init(&ses->se_hash);
1923         spin_unlock(&clp->cl_lock);
1924 }
1925
1926 static void
1927 unhash_client(struct nfs4_client *clp)
1928 {
1929         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1930
1931         spin_lock(&nn->client_lock);
1932         unhash_client_locked(clp);
1933         spin_unlock(&nn->client_lock);
1934 }
1935
1936 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1937 {
1938         if (atomic_read(&clp->cl_refcount))
1939                 return nfserr_jukebox;
1940         unhash_client_locked(clp);
1941         return nfs_ok;
1942 }
1943
1944 static void
1945 __destroy_client(struct nfs4_client *clp)
1946 {
1947         int i;
1948         struct nfs4_openowner *oo;
1949         struct nfs4_delegation *dp;
1950         struct list_head reaplist;
1951
1952         INIT_LIST_HEAD(&reaplist);
1953         spin_lock(&state_lock);
1954         while (!list_empty(&clp->cl_delegations)) {
1955                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1956                 WARN_ON(!unhash_delegation_locked(dp));
1957                 list_add(&dp->dl_recall_lru, &reaplist);
1958         }
1959         spin_unlock(&state_lock);
1960         while (!list_empty(&reaplist)) {
1961                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1962                 list_del_init(&dp->dl_recall_lru);
1963                 destroy_unhashed_deleg(dp);
1964         }
1965         while (!list_empty(&clp->cl_revoked)) {
1966                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1967                 list_del_init(&dp->dl_recall_lru);
1968                 nfs4_put_stid(&dp->dl_stid);
1969         }
1970         while (!list_empty(&clp->cl_openowners)) {
1971                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1972                 nfs4_get_stateowner(&oo->oo_owner);
1973                 release_openowner(oo);
1974         }
1975         for (i = 0; i < OWNER_HASH_SIZE; i++) {
1976                 struct nfs4_stateowner *so, *tmp;
1977
1978                 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
1979                                          so_strhash) {
1980                         /* Should be no openowners at this point */
1981                         WARN_ON_ONCE(so->so_is_open_owner);
1982                         remove_blocked_locks(lockowner(so));
1983                 }
1984         }
1985         nfsd4_return_all_client_layouts(clp);
1986         nfsd4_shutdown_copy(clp);
1987         nfsd4_shutdown_callback(clp);
1988         if (clp->cl_cb_conn.cb_xprt)
1989                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1990         free_client(clp);
1991 }
1992
1993 static void
1994 destroy_client(struct nfs4_client *clp)
1995 {
1996         unhash_client(clp);
1997         __destroy_client(clp);
1998 }
1999
2000 static void expire_client(struct nfs4_client *clp)
2001 {
2002         unhash_client(clp);
2003         nfsd4_client_record_remove(clp);
2004         __destroy_client(clp);
2005 }
2006
2007 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2008 {
2009         memcpy(target->cl_verifier.data, source->data,
2010                         sizeof(target->cl_verifier.data));
2011 }
2012
2013 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2014 {
2015         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2016         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2017 }
2018
2019 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2020 {
2021         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2022         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2023                                                                 GFP_KERNEL);
2024         target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2025         if ((source->cr_principal && !target->cr_principal) ||
2026             (source->cr_raw_principal && !target->cr_raw_principal) ||
2027             (source->cr_targ_princ && !target->cr_targ_princ))
2028                 return -ENOMEM;
2029
2030         target->cr_flavor = source->cr_flavor;
2031         target->cr_uid = source->cr_uid;
2032         target->cr_gid = source->cr_gid;
2033         target->cr_group_info = source->cr_group_info;
2034         get_group_info(target->cr_group_info);
2035         target->cr_gss_mech = source->cr_gss_mech;
2036         if (source->cr_gss_mech)
2037                 gss_mech_get(source->cr_gss_mech);
2038         return 0;
2039 }
2040
2041 static int
2042 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2043 {
2044         if (o1->len < o2->len)
2045                 return -1;
2046         if (o1->len > o2->len)
2047                 return 1;
2048         return memcmp(o1->data, o2->data, o1->len);
2049 }
2050
2051 static int same_name(const char *n1, const char *n2)
2052 {
2053         return 0 == memcmp(n1, n2, HEXDIR_LEN);
2054 }
2055
2056 static int
2057 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2058 {
2059         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2060 }
2061
2062 static int
2063 same_clid(clientid_t *cl1, clientid_t *cl2)
2064 {
2065         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2066 }
2067
2068 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2069 {
2070         int i;
2071
2072         if (g1->ngroups != g2->ngroups)
2073                 return false;
2074         for (i=0; i<g1->ngroups; i++)
2075                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2076                         return false;
2077         return true;
2078 }
2079
2080 /*
2081  * RFC 3530 language requires clid_inuse be returned when the
2082  * "principal" associated with a requests differs from that previously
2083  * used.  We use uid, gid's, and gss principal string as our best
2084  * approximation.  We also don't want to allow non-gss use of a client
2085  * established using gss: in theory cr_principal should catch that
2086  * change, but in practice cr_principal can be null even in the gss case
2087  * since gssd doesn't always pass down a principal string.
2088  */
2089 static bool is_gss_cred(struct svc_cred *cr)
2090 {
2091         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2092         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2093 }
2094
2095
2096 static bool
2097 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2098 {
2099         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2100                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2101                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2102                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2103                 return false;
2104         /* XXX: check that cr_targ_princ fields match ? */
2105         if (cr1->cr_principal == cr2->cr_principal)
2106                 return true;
2107         if (!cr1->cr_principal || !cr2->cr_principal)
2108                 return false;
2109         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2110 }
2111
2112 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2113 {
2114         struct svc_cred *cr = &rqstp->rq_cred;
2115         u32 service;
2116
2117         if (!cr->cr_gss_mech)
2118                 return false;
2119         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2120         return service == RPC_GSS_SVC_INTEGRITY ||
2121                service == RPC_GSS_SVC_PRIVACY;
2122 }
2123
2124 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2125 {
2126         struct svc_cred *cr = &rqstp->rq_cred;
2127
2128         if (!cl->cl_mach_cred)
2129                 return true;
2130         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2131                 return false;
2132         if (!svc_rqst_integrity_protected(rqstp))
2133                 return false;
2134         if (cl->cl_cred.cr_raw_principal)
2135                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2136                                                 cr->cr_raw_principal);
2137         if (!cr->cr_principal)
2138                 return false;
2139         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2140 }
2141
2142 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2143 {
2144         __be32 verf[2];
2145
2146         /*
2147          * This is opaque to client, so no need to byte-swap. Use
2148          * __force to keep sparse happy
2149          */
2150         verf[0] = (__force __be32)get_seconds();
2151         verf[1] = (__force __be32)nn->clverifier_counter++;
2152         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2153 }
2154
2155 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2156 {
2157         clp->cl_clientid.cl_boot = nn->boot_time;
2158         clp->cl_clientid.cl_id = nn->clientid_counter++;
2159         gen_confirm(clp, nn);
2160 }
2161
2162 static struct nfs4_stid *
2163 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2164 {
2165         struct nfs4_stid *ret;
2166
2167         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2168         if (!ret || !ret->sc_type)
2169                 return NULL;
2170         return ret;
2171 }
2172
2173 static struct nfs4_stid *
2174 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2175 {
2176         struct nfs4_stid *s;
2177
2178         spin_lock(&cl->cl_lock);
2179         s = find_stateid_locked(cl, t);
2180         if (s != NULL) {
2181                 if (typemask & s->sc_type)
2182                         refcount_inc(&s->sc_count);
2183                 else
2184                         s = NULL;
2185         }
2186         spin_unlock(&cl->cl_lock);
2187         return s;
2188 }
2189
2190 static struct nfs4_client *create_client(struct xdr_netobj name,
2191                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2192 {
2193         struct nfs4_client *clp;
2194         struct sockaddr *sa = svc_addr(rqstp);
2195         int ret;
2196         struct net *net = SVC_NET(rqstp);
2197
2198         clp = alloc_client(name);
2199         if (clp == NULL)
2200                 return NULL;
2201
2202         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2203         if (ret) {
2204                 free_client(clp);
2205                 return NULL;
2206         }
2207         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2208         clp->cl_time = get_seconds();
2209         clear_bit(0, &clp->cl_cb_slot_busy);
2210         copy_verf(clp, verf);
2211         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2212         clp->cl_cb_session = NULL;
2213         clp->net = net;
2214         return clp;
2215 }
2216
2217 static void
2218 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2219 {
2220         struct rb_node **new = &(root->rb_node), *parent = NULL;
2221         struct nfs4_client *clp;
2222
2223         while (*new) {
2224                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2225                 parent = *new;
2226
2227                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2228                         new = &((*new)->rb_left);
2229                 else
2230                         new = &((*new)->rb_right);
2231         }
2232
2233         rb_link_node(&new_clp->cl_namenode, parent, new);
2234         rb_insert_color(&new_clp->cl_namenode, root);
2235 }
2236
2237 static struct nfs4_client *
2238 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2239 {
2240         int cmp;
2241         struct rb_node *node = root->rb_node;
2242         struct nfs4_client *clp;
2243
2244         while (node) {
2245                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2246                 cmp = compare_blob(&clp->cl_name, name);
2247                 if (cmp > 0)
2248                         node = node->rb_left;
2249                 else if (cmp < 0)
2250                         node = node->rb_right;
2251                 else
2252                         return clp;
2253         }
2254         return NULL;
2255 }
2256
2257 static void
2258 add_to_unconfirmed(struct nfs4_client *clp)
2259 {
2260         unsigned int idhashval;
2261         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2262
2263         lockdep_assert_held(&nn->client_lock);
2264
2265         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2266         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2267         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2268         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2269         renew_client_locked(clp);
2270 }
2271
2272 static void
2273 move_to_confirmed(struct nfs4_client *clp)
2274 {
2275         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2276         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2277
2278         lockdep_assert_held(&nn->client_lock);
2279
2280         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2281         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2282         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2283         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2284         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2285         renew_client_locked(clp);
2286 }
2287
2288 static struct nfs4_client *
2289 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2290 {
2291         struct nfs4_client *clp;
2292         unsigned int idhashval = clientid_hashval(clid->cl_id);
2293
2294         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2295                 if (same_clid(&clp->cl_clientid, clid)) {
2296                         if ((bool)clp->cl_minorversion != sessions)
2297                                 return NULL;
2298                         renew_client_locked(clp);
2299                         return clp;
2300                 }
2301         }
2302         return NULL;
2303 }
2304
2305 static struct nfs4_client *
2306 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2307 {
2308         struct list_head *tbl = nn->conf_id_hashtbl;
2309
2310         lockdep_assert_held(&nn->client_lock);
2311         return find_client_in_id_table(tbl, clid, sessions);
2312 }
2313
2314 static struct nfs4_client *
2315 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2316 {
2317         struct list_head *tbl = nn->unconf_id_hashtbl;
2318
2319         lockdep_assert_held(&nn->client_lock);
2320         return find_client_in_id_table(tbl, clid, sessions);
2321 }
2322
2323 static bool clp_used_exchangeid(struct nfs4_client *clp)
2324 {
2325         return clp->cl_exchange_flags != 0;
2326 }
2327
2328 static struct nfs4_client *
2329 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2330 {
2331         lockdep_assert_held(&nn->client_lock);
2332         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2333 }
2334
2335 static struct nfs4_client *
2336 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2337 {
2338         lockdep_assert_held(&nn->client_lock);
2339         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2340 }
2341
2342 static void
2343 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2344 {
2345         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2346         struct sockaddr *sa = svc_addr(rqstp);
2347         u32 scopeid = rpc_get_scope_id(sa);
2348         unsigned short expected_family;
2349
2350         /* Currently, we only support tcp and tcp6 for the callback channel */
2351         if (se->se_callback_netid_len == 3 &&
2352             !memcmp(se->se_callback_netid_val, "tcp", 3))
2353                 expected_family = AF_INET;
2354         else if (se->se_callback_netid_len == 4 &&
2355                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2356                 expected_family = AF_INET6;
2357         else
2358                 goto out_err;
2359
2360         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2361                                             se->se_callback_addr_len,
2362                                             (struct sockaddr *)&conn->cb_addr,
2363                                             sizeof(conn->cb_addr));
2364
2365         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2366                 goto out_err;
2367
2368         if (conn->cb_addr.ss_family == AF_INET6)
2369                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2370
2371         conn->cb_prog = se->se_callback_prog;
2372         conn->cb_ident = se->se_callback_ident;
2373         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2374         return;
2375 out_err:
2376         conn->cb_addr.ss_family = AF_UNSPEC;
2377         conn->cb_addrlen = 0;
2378         dprintk("NFSD: this client (clientid %08x/%08x) "
2379                 "will not receive delegations\n",
2380                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2381
2382         return;
2383 }
2384
2385 /*
2386  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2387  */
2388 static void
2389 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2390 {
2391         struct xdr_buf *buf = resp->xdr.buf;
2392         struct nfsd4_slot *slot = resp->cstate.slot;
2393         unsigned int base;
2394
2395         dprintk("--> %s slot %p\n", __func__, slot);
2396
2397         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2398         slot->sl_opcnt = resp->opcnt;
2399         slot->sl_status = resp->cstate.status;
2400         free_svc_cred(&slot->sl_cred);
2401         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2402
2403         if (!nfsd4_cache_this(resp)) {
2404                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2405                 return;
2406         }
2407         slot->sl_flags |= NFSD4_SLOT_CACHED;
2408
2409         base = resp->cstate.data_offset;
2410         slot->sl_datalen = buf->len - base;
2411         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2412                 WARN(1, "%s: sessions DRC could not cache compound\n",
2413                      __func__);
2414         return;
2415 }
2416
2417 /*
2418  * Encode the replay sequence operation from the slot values.
2419  * If cachethis is FALSE encode the uncached rep error on the next
2420  * operation which sets resp->p and increments resp->opcnt for
2421  * nfs4svc_encode_compoundres.
2422  *
2423  */
2424 static __be32
2425 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2426                           struct nfsd4_compoundres *resp)
2427 {
2428         struct nfsd4_op *op;
2429         struct nfsd4_slot *slot = resp->cstate.slot;
2430
2431         /* Encode the replayed sequence operation */
2432         op = &args->ops[resp->opcnt - 1];
2433         nfsd4_encode_operation(resp, op);
2434
2435         if (slot->sl_flags & NFSD4_SLOT_CACHED)
2436                 return op->status;
2437         if (args->opcnt == 1) {
2438                 /*
2439                  * The original operation wasn't a solo sequence--we
2440                  * always cache those--so this retry must not match the
2441                  * original:
2442                  */
2443                 op->status = nfserr_seq_false_retry;
2444         } else {
2445                 op = &args->ops[resp->opcnt++];
2446                 op->status = nfserr_retry_uncached_rep;
2447                 nfsd4_encode_operation(resp, op);
2448         }
2449         return op->status;
2450 }
2451
2452 /*
2453  * The sequence operation is not cached because we can use the slot and
2454  * session values.
2455  */
2456 static __be32
2457 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2458                          struct nfsd4_sequence *seq)
2459 {
2460         struct nfsd4_slot *slot = resp->cstate.slot;
2461         struct xdr_stream *xdr = &resp->xdr;
2462         __be32 *p;
2463         __be32 status;
2464
2465         dprintk("--> %s slot %p\n", __func__, slot);
2466
2467         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2468         if (status)
2469                 return status;
2470
2471         p = xdr_reserve_space(xdr, slot->sl_datalen);
2472         if (!p) {
2473                 WARN_ON_ONCE(1);
2474                 return nfserr_serverfault;
2475         }
2476         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2477         xdr_commit_encode(xdr);
2478
2479         resp->opcnt = slot->sl_opcnt;
2480         return slot->sl_status;
2481 }
2482
2483 /*
2484  * Set the exchange_id flags returned by the server.
2485  */
2486 static void
2487 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2488 {
2489 #ifdef CONFIG_NFSD_PNFS
2490         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2491 #else
2492         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2493 #endif
2494
2495         /* Referrals are supported, Migration is not. */
2496         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2497
2498         /* set the wire flags to return to client. */
2499         clid->flags = new->cl_exchange_flags;
2500 }
2501
2502 static bool client_has_openowners(struct nfs4_client *clp)
2503 {
2504         struct nfs4_openowner *oo;
2505
2506         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2507                 if (!list_empty(&oo->oo_owner.so_stateids))
2508                         return true;
2509         }
2510         return false;
2511 }
2512
2513 static bool client_has_state(struct nfs4_client *clp)
2514 {
2515         return client_has_openowners(clp)
2516 #ifdef CONFIG_NFSD_PNFS
2517                 || !list_empty(&clp->cl_lo_states)
2518 #endif
2519                 || !list_empty(&clp->cl_delegations)
2520                 || !list_empty(&clp->cl_sessions)
2521                 || !list_empty(&clp->async_copies);
2522 }
2523
2524 __be32
2525 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2526                 union nfsd4_op_u *u)
2527 {
2528         struct nfsd4_exchange_id *exid = &u->exchange_id;
2529         struct nfs4_client *conf, *new;
2530         struct nfs4_client *unconf = NULL;
2531         __be32 status;
2532         char                    addr_str[INET6_ADDRSTRLEN];
2533         nfs4_verifier           verf = exid->verifier;
2534         struct sockaddr         *sa = svc_addr(rqstp);
2535         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2536         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2537
2538         rpc_ntop(sa, addr_str, sizeof(addr_str));
2539         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2540                 "ip_addr=%s flags %x, spa_how %d\n",
2541                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2542                 addr_str, exid->flags, exid->spa_how);
2543
2544         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2545                 return nfserr_inval;
2546
2547         new = create_client(exid->clname, rqstp, &verf);
2548         if (new == NULL)
2549                 return nfserr_jukebox;
2550
2551         switch (exid->spa_how) {
2552         case SP4_MACH_CRED:
2553                 exid->spo_must_enforce[0] = 0;
2554                 exid->spo_must_enforce[1] = (
2555                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
2556                         1 << (OP_EXCHANGE_ID - 32) |
2557                         1 << (OP_CREATE_SESSION - 32) |
2558                         1 << (OP_DESTROY_SESSION - 32) |
2559                         1 << (OP_DESTROY_CLIENTID - 32));
2560
2561                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2562                                         1 << (OP_OPEN_DOWNGRADE) |
2563                                         1 << (OP_LOCKU) |
2564                                         1 << (OP_DELEGRETURN));
2565
2566                 exid->spo_must_allow[1] &= (
2567                                         1 << (OP_TEST_STATEID - 32) |
2568                                         1 << (OP_FREE_STATEID - 32));
2569                 if (!svc_rqst_integrity_protected(rqstp)) {
2570                         status = nfserr_inval;
2571                         goto out_nolock;
2572                 }
2573                 /*
2574                  * Sometimes userspace doesn't give us a principal.
2575                  * Which is a bug, really.  Anyway, we can't enforce
2576                  * MACH_CRED in that case, better to give up now:
2577                  */
2578                 if (!new->cl_cred.cr_principal &&
2579                                         !new->cl_cred.cr_raw_principal) {
2580                         status = nfserr_serverfault;
2581                         goto out_nolock;
2582                 }
2583                 new->cl_mach_cred = true;
2584         case SP4_NONE:
2585                 break;
2586         default:                                /* checked by xdr code */
2587                 WARN_ON_ONCE(1);
2588                 /* fall through */
2589         case SP4_SSV:
2590                 status = nfserr_encr_alg_unsupp;
2591                 goto out_nolock;
2592         }
2593
2594         /* Cases below refer to rfc 5661 section 18.35.4: */
2595         spin_lock(&nn->client_lock);
2596         conf = find_confirmed_client_by_name(&exid->clname, nn);
2597         if (conf) {
2598                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2599                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2600
2601                 if (update) {
2602                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2603                                 status = nfserr_inval;
2604                                 goto out;
2605                         }
2606                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
2607                                 status = nfserr_wrong_cred;
2608                                 goto out;
2609                         }
2610                         if (!creds_match) { /* case 9 */
2611                                 status = nfserr_perm;
2612                                 goto out;
2613                         }
2614                         if (!verfs_match) { /* case 8 */
2615                                 status = nfserr_not_same;
2616                                 goto out;
2617                         }
2618                         /* case 6 */
2619                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2620                         goto out_copy;
2621                 }
2622                 if (!creds_match) { /* case 3 */
2623                         if (client_has_state(conf)) {
2624                                 status = nfserr_clid_inuse;
2625                                 goto out;
2626                         }
2627                         goto out_new;
2628                 }
2629                 if (verfs_match) { /* case 2 */
2630                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2631                         goto out_copy;
2632                 }
2633                 /* case 5, client reboot */
2634                 conf = NULL;
2635                 goto out_new;
2636         }
2637
2638         if (update) { /* case 7 */
2639                 status = nfserr_noent;
2640                 goto out;
2641         }
2642
2643         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2644         if (unconf) /* case 4, possible retry or client restart */
2645                 unhash_client_locked(unconf);
2646
2647         /* case 1 (normal case) */
2648 out_new:
2649         if (conf) {
2650                 status = mark_client_expired_locked(conf);
2651                 if (status)
2652                         goto out;
2653         }
2654         new->cl_minorversion = cstate->minorversion;
2655         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2656         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2657
2658         gen_clid(new, nn);
2659         add_to_unconfirmed(new);
2660         swap(new, conf);
2661 out_copy:
2662         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2663         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2664
2665         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2666         nfsd4_set_ex_flags(conf, exid);
2667
2668         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2669                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2670         status = nfs_ok;
2671
2672 out:
2673         spin_unlock(&nn->client_lock);
2674 out_nolock:
2675         if (new)
2676                 expire_client(new);
2677         if (unconf)
2678                 expire_client(unconf);
2679         return status;
2680 }
2681
2682 static __be32
2683 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2684 {
2685         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2686                 slot_seqid);
2687
2688         /* The slot is in use, and no response has been sent. */
2689         if (slot_inuse) {
2690                 if (seqid == slot_seqid)
2691                         return nfserr_jukebox;
2692                 else
2693                         return nfserr_seq_misordered;
2694         }
2695         /* Note unsigned 32-bit arithmetic handles wraparound: */
2696         if (likely(seqid == slot_seqid + 1))
2697                 return nfs_ok;
2698         if (seqid == slot_seqid)
2699                 return nfserr_replay_cache;
2700         return nfserr_seq_misordered;
2701 }
2702
2703 /*
2704  * Cache the create session result into the create session single DRC
2705  * slot cache by saving the xdr structure. sl_seqid has been set.
2706  * Do this for solo or embedded create session operations.
2707  */
2708 static void
2709 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2710                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2711 {
2712         slot->sl_status = nfserr;
2713         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2714 }
2715
2716 static __be32
2717 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2718                             struct nfsd4_clid_slot *slot)
2719 {
2720         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2721         return slot->sl_status;
2722 }
2723
2724 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2725                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2726                         1 +     /* MIN tag is length with zero, only length */ \
2727                         3 +     /* version, opcount, opcode */ \
2728                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2729                                 /* seqid, slotID, slotID, cache */ \
2730                         4 ) * sizeof(__be32))
2731
2732 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2733                         2 +     /* verifier: AUTH_NULL, length 0 */\
2734                         1 +     /* status */ \
2735                         1 +     /* MIN tag is length with zero, only length */ \
2736                         3 +     /* opcount, opcode, opstatus*/ \
2737                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2738                                 /* seqid, slotID, slotID, slotID, status */ \
2739                         5 ) * sizeof(__be32))
2740
2741 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2742 {
2743         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2744
2745         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2746                 return nfserr_toosmall;
2747         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2748                 return nfserr_toosmall;
2749         ca->headerpadsz = 0;
2750         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2751         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2752         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2753         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2754                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2755         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2756         /*
2757          * Note decreasing slot size below client's request may make it
2758          * difficult for client to function correctly, whereas
2759          * decreasing the number of slots will (just?) affect
2760          * performance.  When short on memory we therefore prefer to
2761          * decrease number of slots instead of their size.  Clients that
2762          * request larger slots than they need will get poor results:
2763          */
2764         ca->maxreqs = nfsd4_get_drc_mem(ca);
2765         if (!ca->maxreqs)
2766                 return nfserr_jukebox;
2767
2768         return nfs_ok;
2769 }
2770
2771 /*
2772  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2773  * These are based on similar macros in linux/sunrpc/msg_prot.h .
2774  */
2775 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2776         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2777
2778 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2779         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2780
2781 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2782                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2783 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2784                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2785                                  sizeof(__be32))
2786
2787 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2788 {
2789         ca->headerpadsz = 0;
2790
2791         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2792                 return nfserr_toosmall;
2793         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2794                 return nfserr_toosmall;
2795         ca->maxresp_cached = 0;
2796         if (ca->maxops < 2)
2797                 return nfserr_toosmall;
2798
2799         return nfs_ok;
2800 }
2801
2802 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2803 {
2804         switch (cbs->flavor) {
2805         case RPC_AUTH_NULL:
2806         case RPC_AUTH_UNIX:
2807                 return nfs_ok;
2808         default:
2809                 /*
2810                  * GSS case: the spec doesn't allow us to return this
2811                  * error.  But it also doesn't allow us not to support
2812                  * GSS.
2813                  * I'd rather this fail hard than return some error the
2814                  * client might think it can already handle:
2815                  */
2816                 return nfserr_encr_alg_unsupp;
2817         }
2818 }
2819
2820 __be32
2821 nfsd4_create_session(struct svc_rqst *rqstp,
2822                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
2823 {
2824         struct nfsd4_create_session *cr_ses = &u->create_session;
2825         struct sockaddr *sa = svc_addr(rqstp);
2826         struct nfs4_client *conf, *unconf;
2827         struct nfs4_client *old = NULL;
2828         struct nfsd4_session *new;
2829         struct nfsd4_conn *conn;
2830         struct nfsd4_clid_slot *cs_slot = NULL;
2831         __be32 status = 0;
2832         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2833
2834         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2835                 return nfserr_inval;
2836         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2837         if (status)
2838                 return status;
2839         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2840         if (status)
2841                 return status;
2842         status = check_backchannel_attrs(&cr_ses->back_channel);
2843         if (status)
2844                 goto out_release_drc_mem;
2845         status = nfserr_jukebox;
2846         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2847         if (!new)
2848                 goto out_release_drc_mem;
2849         conn = alloc_conn_from_crses(rqstp, cr_ses);
2850         if (!conn)
2851                 goto out_free_session;
2852
2853         spin_lock(&nn->client_lock);
2854         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2855         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2856         WARN_ON_ONCE(conf && unconf);
2857
2858         if (conf) {
2859                 status = nfserr_wrong_cred;
2860                 if (!nfsd4_mach_creds_match(conf, rqstp))
2861                         goto out_free_conn;
2862                 cs_slot = &conf->cl_cs_slot;
2863                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2864                 if (status) {
2865                         if (status == nfserr_replay_cache)
2866                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2867                         goto out_free_conn;
2868                 }
2869         } else if (unconf) {
2870                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2871                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2872                         status = nfserr_clid_inuse;
2873                         goto out_free_conn;
2874                 }
2875                 status = nfserr_wrong_cred;
2876                 if (!nfsd4_mach_creds_match(unconf, rqstp))
2877                         goto out_free_conn;
2878                 cs_slot = &unconf->cl_cs_slot;
2879                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2880                 if (status) {
2881                         /* an unconfirmed replay returns misordered */
2882                         status = nfserr_seq_misordered;
2883                         goto out_free_conn;
2884                 }
2885                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2886                 if (old) {
2887                         status = mark_client_expired_locked(old);
2888                         if (status) {
2889                                 old = NULL;
2890                                 goto out_free_conn;
2891                         }
2892                 }
2893                 move_to_confirmed(unconf);
2894                 conf = unconf;
2895         } else {
2896                 status = nfserr_stale_clientid;
2897                 goto out_free_conn;
2898         }
2899         status = nfs_ok;
2900         /* Persistent sessions are not supported */
2901         cr_ses->flags &= ~SESSION4_PERSIST;
2902         /* Upshifting from TCP to RDMA is not supported */
2903         cr_ses->flags &= ~SESSION4_RDMA;
2904
2905         init_session(rqstp, new, conf, cr_ses);
2906         nfsd4_get_session_locked(new);
2907
2908         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2909                NFS4_MAX_SESSIONID_LEN);
2910         cs_slot->sl_seqid++;
2911         cr_ses->seqid = cs_slot->sl_seqid;
2912
2913         /* cache solo and embedded create sessions under the client_lock */
2914         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2915         spin_unlock(&nn->client_lock);
2916         /* init connection and backchannel */
2917         nfsd4_init_conn(rqstp, conn, new);
2918         nfsd4_put_session(new);
2919         if (old)
2920                 expire_client(old);
2921         return status;
2922 out_free_conn:
2923         spin_unlock(&nn->client_lock);
2924         free_conn(conn);
2925         if (old)
2926                 expire_client(old);
2927 out_free_session:
2928         __free_session(new);
2929 out_release_drc_mem:
2930         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2931         return status;
2932 }
2933
2934 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2935 {
2936         switch (*dir) {
2937         case NFS4_CDFC4_FORE:
2938         case NFS4_CDFC4_BACK:
2939                 return nfs_ok;
2940         case NFS4_CDFC4_FORE_OR_BOTH:
2941         case NFS4_CDFC4_BACK_OR_BOTH:
2942                 *dir = NFS4_CDFC4_BOTH;
2943                 return nfs_ok;
2944         };
2945         return nfserr_inval;
2946 }
2947
2948 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
2949                 struct nfsd4_compound_state *cstate,
2950                 union nfsd4_op_u *u)
2951 {
2952         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
2953         struct nfsd4_session *session = cstate->session;
2954         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2955         __be32 status;
2956
2957         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2958         if (status)
2959                 return status;
2960         spin_lock(&nn->client_lock);
2961         session->se_cb_prog = bc->bc_cb_program;
2962         session->se_cb_sec = bc->bc_cb_sec;
2963         spin_unlock(&nn->client_lock);
2964
2965         nfsd4_probe_callback(session->se_client);
2966
2967         return nfs_ok;
2968 }
2969
2970 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2971                      struct nfsd4_compound_state *cstate,
2972                      union nfsd4_op_u *u)
2973 {
2974         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
2975         __be32 status;
2976         struct nfsd4_conn *conn;
2977         struct nfsd4_session *session;
2978         struct net *net = SVC_NET(rqstp);
2979         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2980
2981         if (!nfsd4_last_compound_op(rqstp))
2982                 return nfserr_not_only_op;
2983         spin_lock(&nn->client_lock);
2984         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2985         spin_unlock(&nn->client_lock);
2986         if (!session)
2987                 goto out_no_session;
2988         status = nfserr_wrong_cred;
2989         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2990                 goto out;
2991         status = nfsd4_map_bcts_dir(&bcts->dir);
2992         if (status)
2993                 goto out;
2994         conn = alloc_conn(rqstp, bcts->dir);
2995         status = nfserr_jukebox;
2996         if (!conn)
2997                 goto out;
2998         nfsd4_init_conn(rqstp, conn, session);
2999         status = nfs_ok;
3000 out:
3001         nfsd4_put_session(session);
3002 out_no_session:
3003         return status;
3004 }
3005
3006 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3007 {
3008         if (!cstate->session)
3009                 return false;
3010         return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3011 }
3012
3013 __be32
3014 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3015                 union nfsd4_op_u *u)
3016 {
3017         struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3018         struct nfsd4_session *ses;
3019         __be32 status;
3020         int ref_held_by_me = 0;
3021         struct net *net = SVC_NET(r);
3022         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3023
3024         status = nfserr_not_only_op;
3025         if (nfsd4_compound_in_session(cstate, sessionid)) {
3026                 if (!nfsd4_last_compound_op(r))
3027                         goto out;
3028                 ref_held_by_me++;
3029         }
3030         dump_sessionid(__func__, sessionid);
3031         spin_lock(&nn->client_lock);
3032         ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3033         if (!ses)
3034                 goto out_client_lock;
3035         status = nfserr_wrong_cred;
3036         if (!nfsd4_mach_creds_match(ses->se_client, r))
3037                 goto out_put_session;
3038         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3039         if (status)
3040                 goto out_put_session;
3041         unhash_session(ses);
3042         spin_unlock(&nn->client_lock);
3043
3044         nfsd4_probe_callback_sync(ses->se_client);
3045
3046         spin_lock(&nn->client_lock);
3047         status = nfs_ok;
3048 out_put_session:
3049         nfsd4_put_session_locked(ses);
3050 out_client_lock:
3051         spin_unlock(&nn->client_lock);
3052 out:
3053         return status;
3054 }
3055
3056 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3057 {
3058         struct nfsd4_conn *c;
3059
3060         list_for_each_entry(c, &s->se_conns, cn_persession) {
3061                 if (c->cn_xprt == xpt) {
3062                         return c;
3063                 }
3064         }
3065         return NULL;
3066 }
3067
3068 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3069 {
3070         struct nfs4_client *clp = ses->se_client;
3071         struct nfsd4_conn *c;
3072         __be32 status = nfs_ok;
3073         int ret;
3074
3075         spin_lock(&clp->cl_lock);
3076         c = __nfsd4_find_conn(new->cn_xprt, ses);
3077         if (c)
3078                 goto out_free;
3079         status = nfserr_conn_not_bound_to_session;
3080         if (clp->cl_mach_cred)
3081                 goto out_free;
3082         __nfsd4_hash_conn(new, ses);
3083         spin_unlock(&clp->cl_lock);
3084         ret = nfsd4_register_conn(new);
3085         if (ret)
3086                 /* oops; xprt is already down: */
3087                 nfsd4_conn_lost(&new->cn_xpt_user);
3088         return nfs_ok;
3089 out_free:
3090         spin_unlock(&clp->cl_lock);
3091         free_conn(new);
3092         return status;
3093 }
3094
3095 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3096 {
3097         struct nfsd4_compoundargs *args = rqstp->rq_argp;
3098
3099         return args->opcnt > session->se_fchannel.maxops;
3100 }
3101
3102 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3103                                   struct nfsd4_session *session)
3104 {
3105         struct xdr_buf *xb = &rqstp->rq_arg;
3106
3107         return xb->len > session->se_fchannel.maxreq_sz;
3108 }
3109
3110 static bool replay_matches_cache(struct svc_rqst *rqstp,
3111                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3112 {
3113         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3114
3115         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3116             (bool)seq->cachethis)
3117                 return false;
3118         /*
3119          * If there's an error than the reply can have fewer ops than
3120          * the call.  But if we cached a reply with *more* ops than the
3121          * call you're sending us now, then this new call is clearly not
3122          * really a replay of the old one:
3123          */
3124         if (slot->sl_opcnt < argp->opcnt)
3125                 return false;
3126         /* This is the only check explicitly called by spec: */
3127         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3128                 return false;
3129         /*
3130          * There may be more comparisons we could actually do, but the
3131          * spec doesn't require us to catch every case where the calls
3132          * don't match (that would require caching the call as well as
3133          * the reply), so we don't bother.
3134          */
3135         return true;
3136 }
3137
3138 __be32
3139 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3140                 union nfsd4_op_u *u)
3141 {
3142         struct nfsd4_sequence *seq = &u->sequence;
3143         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3144         struct xdr_stream *xdr = &resp->xdr;
3145         struct nfsd4_session *session;
3146         struct nfs4_client *clp;
3147         struct nfsd4_slot *slot;
3148         struct nfsd4_conn *conn;
3149         __be32 status;
3150         int buflen;
3151         struct net *net = SVC_NET(rqstp);
3152         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3153
3154         if (resp->opcnt != 1)
3155                 return nfserr_sequence_pos;
3156
3157         /*
3158          * Will be either used or freed by nfsd4_sequence_check_conn
3159          * below.
3160          */
3161         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3162         if (!conn)
3163                 return nfserr_jukebox;
3164
3165         spin_lock(&nn->client_lock);
3166         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3167         if (!session)
3168                 goto out_no_session;
3169         clp = session->se_client;
3170
3171         status = nfserr_too_many_ops;
3172         if (nfsd4_session_too_many_ops(rqstp, session))
3173                 goto out_put_session;
3174
3175         status = nfserr_req_too_big;
3176         if (nfsd4_request_too_big(rqstp, session))
3177                 goto out_put_session;
3178
3179         status = nfserr_badslot;
3180         if (seq->slotid >= session->se_fchannel.maxreqs)
3181                 goto out_put_session;
3182
3183         slot = session->se_slots[seq->slotid];
3184         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3185
3186         /* We do not negotiate the number of slots yet, so set the
3187          * maxslots to the session maxreqs which is used to encode
3188          * sr_highest_slotid and the sr_target_slot id to maxslots */
3189         seq->maxslots = session->se_fchannel.maxreqs;
3190
3191         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3192                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3193         if (status == nfserr_replay_cache) {
3194                 status = nfserr_seq_misordered;
3195                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3196                         goto out_put_session;
3197                 status = nfserr_seq_false_retry;
3198                 if (!replay_matches_cache(rqstp, seq, slot))
3199                         goto out_put_session;
3200                 cstate->slot = slot;
3201                 cstate->session = session;
3202                 cstate->clp = clp;
3203                 /* Return the cached reply status and set cstate->status
3204                  * for nfsd4_proc_compound processing */
3205                 status = nfsd4_replay_cache_entry(resp, seq);
3206                 cstate->status = nfserr_replay_cache;
3207                 goto out;
3208         }
3209         if (status)
3210                 goto out_put_session;
3211
3212         status = nfsd4_sequence_check_conn(conn, session);
3213         conn = NULL;
3214         if (status)
3215                 goto out_put_session;
3216
3217         buflen = (seq->cachethis) ?
3218                         session->se_fchannel.maxresp_cached :
3219                         session->se_fchannel.maxresp_sz;
3220         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3221                                     nfserr_rep_too_big;
3222         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3223                 goto out_put_session;
3224         svc_reserve(rqstp, buflen);
3225
3226         status = nfs_ok;
3227         /* Success! bump slot seqid */
3228         slot->sl_seqid = seq->seqid;
3229         slot->sl_flags |= NFSD4_SLOT_INUSE;
3230         if (seq->cachethis)
3231                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3232         else
3233                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3234
3235         cstate->slot = slot;
3236         cstate->session = session;
3237         cstate->clp = clp;
3238
3239 out:
3240         switch (clp->cl_cb_state) {
3241         case NFSD4_CB_DOWN:
3242                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3243                 break;
3244         case NFSD4_CB_FAULT:
3245                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3246                 break;
3247         default:
3248                 seq->status_flags = 0;
3249         }
3250         if (!list_empty(&clp->cl_revoked))
3251                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3252 out_no_session:
3253         if (conn)
3254                 free_conn(conn);
3255         spin_unlock(&nn->client_lock);
3256         return status;
3257 out_put_session:
3258         nfsd4_put_session_locked(session);
3259         goto out_no_session;
3260 }
3261
3262 void
3263 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3264 {
3265         struct nfsd4_compound_state *cs = &resp->cstate;
3266
3267         if (nfsd4_has_session(cs)) {
3268                 if (cs->status != nfserr_replay_cache) {
3269                         nfsd4_store_cache_entry(resp);
3270                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3271                 }
3272                 /* Drop session reference that was taken in nfsd4_sequence() */
3273                 nfsd4_put_session(cs->session);
3274         } else if (cs->clp)
3275                 put_client_renew(cs->clp);
3276 }
3277
3278 __be32
3279 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3280                 struct nfsd4_compound_state *cstate,
3281                 union nfsd4_op_u *u)
3282 {
3283         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3284         struct nfs4_client *conf, *unconf;
3285         struct nfs4_client *clp = NULL;
3286         __be32 status = 0;
3287         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3288
3289         spin_lock(&nn->client_lock);
3290         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3291         conf = find_confirmed_client(&dc->clientid, true, nn);
3292         WARN_ON_ONCE(conf && unconf);
3293
3294         if (conf) {
3295                 if (client_has_state(conf)) {
3296                         status = nfserr_clientid_busy;
3297                         goto out;
3298                 }
3299                 status = mark_client_expired_locked(conf);
3300                 if (status)
3301                         goto out;
3302                 clp = conf;
3303         } else if (unconf)
3304                 clp = unconf;
3305         else {
3306                 status = nfserr_stale_clientid;
3307                 goto out;
3308         }
3309         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3310                 clp = NULL;
3311                 status = nfserr_wrong_cred;
3312                 goto out;
3313         }
3314         unhash_client_locked(clp);
3315 out:
3316         spin_unlock(&nn->client_lock);
3317         if (clp)
3318                 expire_client(clp);
3319         return status;
3320 }
3321
3322 __be32
3323 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3324                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3325 {
3326         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3327         __be32 status = 0;
3328
3329         if (rc->rca_one_fs) {
3330                 if (!cstate->current_fh.fh_dentry)
3331                         return nfserr_nofilehandle;
3332                 /*
3333                  * We don't take advantage of the rca_one_fs case.
3334                  * That's OK, it's optional, we can safely ignore it.
3335                  */
3336                 return nfs_ok;
3337         }
3338
3339         status = nfserr_complete_already;
3340         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3341                              &cstate->session->se_client->cl_flags))
3342                 goto out;
3343
3344         status = nfserr_stale_clientid;
3345         if (is_client_expired(cstate->session->se_client))
3346                 /*
3347                  * The following error isn't really legal.
3348                  * But we only get here if the client just explicitly
3349                  * destroyed the client.  Surely it no longer cares what
3350                  * error it gets back on an operation for the dead
3351                  * client.
3352                  */
3353                 goto out;
3354
3355         status = nfs_ok;
3356         nfsd4_client_record_create(cstate->session->se_client);
3357 out:
3358         return status;
3359 }
3360
3361 __be32
3362 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3363                   union nfsd4_op_u *u)
3364 {
3365         struct nfsd4_setclientid *setclid = &u->setclientid;
3366         struct xdr_netobj       clname = setclid->se_name;
3367         nfs4_verifier           clverifier = setclid->se_verf;
3368         struct nfs4_client      *conf, *new;
3369         struct nfs4_client      *unconf = NULL;
3370         __be32                  status;
3371         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3372
3373         new = create_client(clname, rqstp, &clverifier);
3374         if (new == NULL)
3375                 return nfserr_jukebox;
3376         /* Cases below refer to rfc 3530 section 14.2.33: */
3377         spin_lock(&nn->client_lock);
3378         conf = find_confirmed_client_by_name(&clname, nn);
3379         if (conf && client_has_state(conf)) {
3380                 /* case 0: */
3381                 status = nfserr_clid_inuse;
3382                 if (clp_used_exchangeid(conf))
3383                         goto out;
3384                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3385                         char addr_str[INET6_ADDRSTRLEN];
3386                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3387                                  sizeof(addr_str));
3388                         dprintk("NFSD: setclientid: string in use by client "
3389                                 "at %s\n", addr_str);
3390                         goto out;
3391                 }
3392         }
3393         unconf = find_unconfirmed_client_by_name(&clname, nn);
3394         if (unconf)
3395                 unhash_client_locked(unconf);
3396         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3397                 /* case 1: probable callback update */
3398                 copy_clid(new, conf);
3399                 gen_confirm(new, nn);
3400         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3401                 gen_clid(new, nn);
3402         new->cl_minorversion = 0;
3403         gen_callback(new, setclid, rqstp);
3404         add_to_unconfirmed(new);
3405         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3406         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3407         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3408         new = NULL;
3409         status = nfs_ok;
3410 out:
3411         spin_unlock(&nn->client_lock);
3412         if (new)
3413                 free_client(new);
3414         if (unconf)
3415                 expire_client(unconf);
3416         return status;
3417 }
3418
3419
3420 __be32
3421 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3422                         struct nfsd4_compound_state *cstate,
3423                         union nfsd4_op_u *u)
3424 {
3425         struct nfsd4_setclientid_confirm *setclientid_confirm =
3426                         &u->setclientid_confirm;
3427         struct nfs4_client *conf, *unconf;
3428         struct nfs4_client *old = NULL;
3429         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3430         clientid_t * clid = &setclientid_confirm->sc_clientid;
3431         __be32 status;
3432         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3433
3434         if (STALE_CLIENTID(clid, nn))
3435                 return nfserr_stale_clientid;
3436
3437         spin_lock(&nn->client_lock);
3438         conf = find_confirmed_client(clid, false, nn);
3439         unconf = find_unconfirmed_client(clid, false, nn);
3440         /*
3441          * We try hard to give out unique clientid's, so if we get an
3442          * attempt to confirm the same clientid with a different cred,
3443          * the client may be buggy; this should never happen.
3444          *
3445          * Nevertheless, RFC 7530 recommends INUSE for this case:
3446          */
3447         status = nfserr_clid_inuse;
3448         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3449                 goto out;
3450         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3451                 goto out;
3452         /* cases below refer to rfc 3530 section 14.2.34: */
3453         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3454                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3455                         /* case 2: probable retransmit */
3456                         status = nfs_ok;
3457                 } else /* case 4: client hasn't noticed we rebooted yet? */
3458                         status = nfserr_stale_clientid;
3459                 goto out;
3460         }
3461         status = nfs_ok;
3462         if (conf) { /* case 1: callback update */
3463                 old = unconf;
3464                 unhash_client_locked(old);
3465                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3466         } else { /* case 3: normal case; new or rebooted client */
3467                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3468                 if (old) {
3469                         status = nfserr_clid_inuse;
3470                         if (client_has_state(old)
3471                                         && !same_creds(&unconf->cl_cred,
3472                                                         &old->cl_cred))
3473                                 goto out;
3474                         status = mark_client_expired_locked(old);
3475                         if (status) {
3476                                 old = NULL;
3477                                 goto out;
3478                         }
3479                 }
3480                 move_to_confirmed(unconf);
3481                 conf = unconf;
3482         }
3483         get_client_locked(conf);
3484         spin_unlock(&nn->client_lock);
3485         nfsd4_probe_callback(conf);
3486         spin_lock(&nn->client_lock);
3487         put_client_renew_locked(conf);
3488 out:
3489         spin_unlock(&nn->client_lock);
3490         if (old)
3491                 expire_client(old);
3492         return status;
3493 }
3494
3495 static struct nfs4_file *nfsd4_alloc_file(void)
3496 {
3497         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3498 }
3499
3500 /* OPEN Share state helper functions */
3501 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3502                                 struct nfs4_file *fp)
3503 {
3504         lockdep_assert_held(&state_lock);
3505
3506         refcount_set(&fp->fi_ref, 1);
3507         spin_lock_init(&fp->fi_lock);
3508         INIT_LIST_HEAD(&fp->fi_stateids);
3509         INIT_LIST_HEAD(&fp->fi_delegations);
3510         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3511         fh_copy_shallow(&fp->fi_fhandle, fh);
3512         fp->fi_deleg_file = NULL;
3513         fp->fi_had_conflict = false;
3514         fp->fi_share_deny = 0;
3515         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3516         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3517 #ifdef CONFIG_NFSD_PNFS
3518         INIT_LIST_HEAD(&fp->fi_lo_states);
3519         atomic_set(&fp->fi_lo_recalls, 0);
3520 #endif
3521         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3522 }
3523
3524 void
3525 nfsd4_free_slabs(void)
3526 {
3527         kmem_cache_destroy(client_slab);
3528         kmem_cache_destroy(openowner_slab);
3529         kmem_cache_destroy(lockowner_slab);
3530         kmem_cache_destroy(file_slab);
3531         kmem_cache_destroy(stateid_slab);
3532         kmem_cache_destroy(deleg_slab);
3533         kmem_cache_destroy(odstate_slab);
3534 }
3535
3536 int
3537 nfsd4_init_slabs(void)
3538 {
3539         client_slab = kmem_cache_create("nfsd4_clients",
3540                         sizeof(struct nfs4_client), 0, 0, NULL);
3541         if (client_slab == NULL)
3542                 goto out;
3543         openowner_slab = kmem_cache_create("nfsd4_openowners",
3544                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3545         if (openowner_slab == NULL)
3546                 goto out_free_client_slab;
3547         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3548                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3549         if (lockowner_slab == NULL)
3550                 goto out_free_openowner_slab;
3551         file_slab = kmem_cache_create("nfsd4_files",
3552                         sizeof(struct nfs4_file), 0, 0, NULL);
3553         if (file_slab == NULL)
3554                 goto out_free_lockowner_slab;
3555         stateid_slab = kmem_cache_create("nfsd4_stateids",
3556                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3557         if (stateid_slab == NULL)
3558                 goto out_free_file_slab;
3559         deleg_slab = kmem_cache_create("nfsd4_delegations",
3560                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3561         if (deleg_slab == NULL)
3562                 goto out_free_stateid_slab;
3563         odstate_slab = kmem_cache_create("nfsd4_odstate",
3564                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3565         if (odstate_slab == NULL)
3566                 goto out_free_deleg_slab;
3567         return 0;
3568
3569 out_free_deleg_slab:
3570         kmem_cache_destroy(deleg_slab);
3571 out_free_stateid_slab:
3572         kmem_cache_destroy(stateid_slab);
3573 out_free_file_slab:
3574         kmem_cache_destroy(file_slab);
3575 out_free_lockowner_slab:
3576         kmem_cache_destroy(lockowner_slab);
3577 out_free_openowner_slab:
3578         kmem_cache_destroy(openowner_slab);
3579 out_free_client_slab:
3580         kmem_cache_destroy(client_slab);
3581 out:
3582         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3583         return -ENOMEM;
3584 }
3585
3586 static void init_nfs4_replay(struct nfs4_replay *rp)
3587 {
3588         rp->rp_status = nfserr_serverfault;
3589         rp->rp_buflen = 0;
3590         rp->rp_buf = rp->rp_ibuf;
3591         mutex_init(&rp->rp_mutex);
3592 }
3593
3594 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3595                 struct nfs4_stateowner *so)
3596 {
3597         if (!nfsd4_has_session(cstate)) {
3598                 mutex_lock(&so->so_replay.rp_mutex);
3599                 cstate->replay_owner = nfs4_get_stateowner(so);
3600         }
3601 }
3602
3603 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3604 {
3605         struct nfs4_stateowner *so = cstate->replay_owner;
3606
3607         if (so != NULL) {
3608                 cstate->replay_owner = NULL;
3609                 mutex_unlock(&so->so_replay.rp_mutex);
3610                 nfs4_put_stateowner(so);
3611         }
3612 }
3613
3614 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3615 {
3616         struct nfs4_stateowner *sop;
3617
3618         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3619         if (!sop)
3620                 return NULL;
3621
3622         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3623         if (!sop->so_owner.data) {
3624                 kmem_cache_free(slab, sop);
3625                 return NULL;
3626         }
3627         sop->so_owner.len = owner->len;
3628
3629         INIT_LIST_HEAD(&sop->so_stateids);
3630         sop->so_client = clp;
3631         init_nfs4_replay(&sop->so_replay);
3632         atomic_set(&sop->so_count, 1);
3633         return sop;
3634 }
3635
3636 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3637 {
3638         lockdep_assert_held(&clp->cl_lock);
3639
3640         list_add(&oo->oo_owner.so_strhash,
3641                  &clp->cl_ownerstr_hashtbl[strhashval]);
3642         list_add(&oo->oo_perclient, &clp->cl_openowners);
3643 }
3644
3645 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3646 {
3647         unhash_openowner_locked(openowner(so));
3648 }
3649
3650 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3651 {
3652         struct nfs4_openowner *oo = openowner(so);
3653
3654         kmem_cache_free(openowner_slab, oo);
3655 }
3656
3657 static const struct nfs4_stateowner_operations openowner_ops = {
3658         .so_unhash =    nfs4_unhash_openowner,
3659         .so_free =      nfs4_free_openowner,
3660 };
3661
3662 static struct nfs4_ol_stateid *
3663 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3664 {
3665         struct nfs4_ol_stateid *local, *ret = NULL;
3666         struct nfs4_openowner *oo = open->op_openowner;
3667
3668         lockdep_assert_held(&fp->fi_lock);
3669
3670         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3671                 /* ignore lock owners */
3672                 if (local->st_stateowner->so_is_open_owner == 0)
3673                         continue;
3674                 if (local->st_stateowner != &oo->oo_owner)
3675                         continue;
3676                 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
3677                         ret = local;
3678                         refcount_inc(&ret->st_stid.sc_count);
3679                         break;
3680                 }
3681         }
3682         return ret;
3683 }
3684
3685 static __be32
3686 nfsd4_verify_open_stid(struct nfs4_stid *s)
3687 {
3688         __be32 ret = nfs_ok;
3689
3690         switch (s->sc_type) {
3691         default:
3692                 break;
3693         case 0:
3694         case NFS4_CLOSED_STID:
3695         case NFS4_CLOSED_DELEG_STID:
3696                 ret = nfserr_bad_stateid;
3697                 break;
3698         case NFS4_REVOKED_DELEG_STID:
3699                 ret = nfserr_deleg_revoked;
3700         }
3701         return ret;
3702 }
3703
3704 /* Lock the stateid st_mutex, and deal with races with CLOSE */
3705 static __be32
3706 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
3707 {
3708         __be32 ret;
3709
3710         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
3711         ret = nfsd4_verify_open_stid(&stp->st_stid);
3712         if (ret != nfs_ok)
3713                 mutex_unlock(&stp->st_mutex);
3714         return ret;
3715 }
3716
3717 static struct nfs4_ol_stateid *
3718 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3719 {
3720         struct nfs4_ol_stateid *stp;
3721         for (;;) {
3722                 spin_lock(&fp->fi_lock);
3723                 stp = nfsd4_find_existing_open(fp, open);
3724                 spin_unlock(&fp->fi_lock);
3725                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
3726                         break;
3727                 nfs4_put_stid(&stp->st_stid);
3728         }
3729         return stp;
3730 }
3731
3732 static struct nfs4_openowner *
3733 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3734                            struct nfsd4_compound_state *cstate)
3735 {
3736         struct nfs4_client *clp = cstate->clp;
3737         struct nfs4_openowner *oo, *ret;
3738
3739         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3740         if (!oo)
3741                 return NULL;
3742         oo->oo_owner.so_ops = &openowner_ops;
3743         oo->oo_owner.so_is_open_owner = 1;
3744         oo->oo_owner.so_seqid = open->op_seqid;
3745         oo->oo_flags = 0;
3746         if (nfsd4_has_session(cstate))
3747                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3748         oo->oo_time = 0;
3749         oo->oo_last_closed_stid = NULL;
3750         INIT_LIST_HEAD(&oo->oo_close_lru);
3751         spin_lock(&clp->cl_lock);
3752         ret = find_openstateowner_str_locked(strhashval, open, clp);
3753         if (ret == NULL) {
3754                 hash_openowner(oo, clp, strhashval);
3755                 ret = oo;
3756         } else
3757                 nfs4_free_stateowner(&oo->oo_owner);
3758
3759         spin_unlock(&clp->cl_lock);
3760         return ret;
3761 }
3762
3763 static struct nfs4_ol_stateid *
3764 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3765 {
3766
3767         struct nfs4_openowner *oo = open->op_openowner;
3768         struct nfs4_ol_stateid *retstp = NULL;
3769         struct nfs4_ol_stateid *stp;
3770
3771         stp = open->op_stp;
3772         /* We are moving these outside of the spinlocks to avoid the warnings */
3773         mutex_init(&stp->st_mutex);
3774         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
3775
3776 retry:
3777         spin_lock(&oo->oo_owner.so_client->cl_lock);
3778         spin_lock(&fp->fi_lock);
3779
3780         retstp = nfsd4_find_existing_open(fp, open);
3781         if (retstp)
3782                 goto out_unlock;
3783
3784         open->op_stp = NULL;
3785         refcount_inc(&stp->st_stid.sc_count);
3786         stp->st_stid.sc_type = NFS4_OPEN_STID;
3787         INIT_LIST_HEAD(&stp->st_locks);
3788         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3789         get_nfs4_file(fp);
3790         stp->st_stid.sc_file = fp;
3791         stp->st_access_bmap = 0;
3792         stp->st_deny_bmap = 0;
3793         stp->st_openstp = NULL;
3794         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3795         list_add(&stp->st_perfile, &fp->fi_stateids);
3796
3797 out_unlock:
3798         spin_unlock(&fp->fi_lock);
3799         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3800         if (retstp) {
3801                 /* Handle races with CLOSE */
3802                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
3803                         nfs4_put_stid(&retstp->st_stid);
3804                         goto retry;
3805                 }
3806                 /* To keep mutex tracking happy */
3807                 mutex_unlock(&stp->st_mutex);
3808                 stp = retstp;
3809         }
3810         return stp;
3811 }
3812
3813 /*
3814  * In the 4.0 case we need to keep the owners around a little while to handle
3815  * CLOSE replay. We still do need to release any file access that is held by
3816  * them before returning however.
3817  */
3818 static void
3819 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3820 {
3821         struct nfs4_ol_stateid *last;
3822         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3823         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3824                                                 nfsd_net_id);
3825
3826         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3827
3828         /*
3829          * We know that we hold one reference via nfsd4_close, and another
3830          * "persistent" reference for the client. If the refcount is higher
3831          * than 2, then there are still calls in progress that are using this
3832          * stateid. We can't put the sc_file reference until they are finished.
3833          * Wait for the refcount to drop to 2. Since it has been unhashed,
3834          * there should be no danger of the refcount going back up again at
3835          * this point.
3836          */
3837         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
3838
3839         release_all_access(s);
3840         if (s->st_stid.sc_file) {
3841                 put_nfs4_file(s->st_stid.sc_file);
3842                 s->st_stid.sc_file = NULL;
3843         }
3844
3845         spin_lock(&nn->client_lock);
3846         last = oo->oo_last_closed_stid;
3847         oo->oo_last_closed_stid = s;
3848         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3849         oo->oo_time = get_seconds();
3850         spin_unlock(&nn->client_lock);
3851         if (last)
3852                 nfs4_put_stid(&last->st_stid);
3853 }
3854
3855 /* search file_hashtbl[] for file */
3856 static struct nfs4_file *
3857 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3858 {
3859         struct nfs4_file *fp;
3860
3861         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3862                 if (fh_match(&fp->fi_fhandle, fh)) {
3863                         if (refcount_inc_not_zero(&fp->fi_ref))
3864                                 return fp;
3865                 }
3866         }
3867         return NULL;
3868 }
3869
3870 struct nfs4_file *
3871 find_file(struct knfsd_fh *fh)
3872 {
3873         struct nfs4_file *fp;
3874         unsigned int hashval = file_hashval(fh);
3875
3876         rcu_read_lock();
3877         fp = find_file_locked(fh, hashval);
3878         rcu_read_unlock();
3879         return fp;
3880 }
3881
3882 static struct nfs4_file *
3883 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3884 {
3885         struct nfs4_file *fp;
3886         unsigned int hashval = file_hashval(fh);
3887
3888         rcu_read_lock();
3889         fp = find_file_locked(fh, hashval);
3890         rcu_read_unlock();
3891         if (fp)
3892                 return fp;
3893
3894         spin_lock(&state_lock);
3895         fp = find_file_locked(fh, hashval);
3896         if (likely(fp == NULL)) {
3897                 nfsd4_init_file(fh, hashval, new);
3898                 fp = new;
3899         }
3900         spin_unlock(&state_lock);
3901
3902         return fp;
3903 }
3904
3905 /*
3906  * Called to check deny when READ with all zero stateid or
3907  * WRITE with all zero or all one stateid
3908  */
3909 static __be32
3910 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3911 {
3912         struct nfs4_file *fp;
3913         __be32 ret = nfs_ok;
3914
3915         fp = find_file(&current_fh->fh_handle);
3916         if (!fp)
3917                 return ret;
3918         /* Check for conflicting share reservations */
3919         spin_lock(&fp->fi_lock);
3920         if (fp->fi_share_deny & deny_type)
3921                 ret = nfserr_locked;
3922         spin_unlock(&fp->fi_lock);
3923         put_nfs4_file(fp);
3924         return ret;
3925 }
3926
3927 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3928 {
3929         struct nfs4_delegation *dp = cb_to_delegation(cb);
3930         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3931                                           nfsd_net_id);
3932
3933         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3934
3935         /*
3936          * We can't do this in nfsd_break_deleg_cb because it is
3937          * already holding inode->i_lock.
3938          *
3939          * If the dl_time != 0, then we know that it has already been
3940          * queued for a lease break. Don't queue it again.
3941          */
3942         spin_lock(&state_lock);
3943         if (dp->dl_time == 0) {
3944                 dp->dl_time = get_seconds();
3945                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3946         }
3947         spin_unlock(&state_lock);
3948 }
3949
3950 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3951                 struct rpc_task *task)
3952 {
3953         struct nfs4_delegation *dp = cb_to_delegation(cb);
3954
3955         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3956                 return 1;
3957
3958         switch (task->tk_status) {
3959         case 0:
3960                 return 1;
3961         case -EBADHANDLE:
3962         case -NFS4ERR_BAD_STATEID:
3963                 /*
3964                  * Race: client probably got cb_recall before open reply
3965                  * granting delegation.
3966                  */
3967                 if (dp->dl_retries--) {
3968                         rpc_delay(task, 2 * HZ);
3969                         return 0;
3970                 }
3971                 /*FALLTHRU*/
3972         default:
3973                 return -1;
3974         }
3975 }
3976
3977 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3978 {
3979         struct nfs4_delegation *dp = cb_to_delegation(cb);
3980
3981         nfs4_put_stid(&dp->dl_stid);
3982 }
3983
3984 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3985         .prepare        = nfsd4_cb_recall_prepare,
3986         .done           = nfsd4_cb_recall_done,
3987         .release        = nfsd4_cb_recall_release,
3988 };
3989
3990 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3991 {
3992         /*
3993          * We're assuming the state code never drops its reference
3994          * without first removing the lease.  Since we're in this lease
3995          * callback (and since the lease code is serialized by the
3996          * i_lock) we know the server hasn't removed the lease yet, and
3997          * we know it's safe to take a reference.
3998          */
3999         refcount_inc(&dp->dl_stid.sc_count);
4000         nfsd4_run_cb(&dp->dl_recall);
4001 }
4002
4003 /* Called from break_lease() with i_lock held. */
4004 static bool
4005 nfsd_break_deleg_cb(struct file_lock *fl)
4006 {
4007         bool ret = false;
4008         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4009         struct nfs4_file *fp = dp->dl_stid.sc_file;
4010
4011         /*
4012          * We don't want the locks code to timeout the lease for us;
4013          * we'll remove it ourself if a delegation isn't returned
4014          * in time:
4015          */
4016         fl->fl_break_time = 0;
4017
4018         spin_lock(&fp->fi_lock);
4019         fp->fi_had_conflict = true;
4020         nfsd_break_one_deleg(dp);
4021         spin_unlock(&fp->fi_lock);
4022         return ret;
4023 }
4024
4025 static int
4026 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4027                      struct list_head *dispose)
4028 {
4029         if (arg & F_UNLCK)
4030                 return lease_modify(onlist, arg, dispose);
4031         else
4032                 return -EAGAIN;
4033 }
4034
4035 static const struct lock_manager_operations nfsd_lease_mng_ops = {
4036         .lm_break = nfsd_break_deleg_cb,
4037         .lm_change = nfsd_change_deleg_cb,
4038 };
4039
4040 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4041 {
4042         if (nfsd4_has_session(cstate))
4043                 return nfs_ok;
4044         if (seqid == so->so_seqid - 1)
4045                 return nfserr_replay_me;
4046         if (seqid == so->so_seqid)
4047                 return nfs_ok;
4048         return nfserr_bad_seqid;
4049 }
4050
4051 static __be32 lookup_clientid(clientid_t *clid,
4052                 struct nfsd4_compound_state *cstate,
4053                 struct nfsd_net *nn)
4054 {
4055         struct nfs4_client *found;
4056
4057         if (cstate->clp) {
4058                 found = cstate->clp;
4059                 if (!same_clid(&found->cl_clientid, clid))
4060                         return nfserr_stale_clientid;
4061                 return nfs_ok;
4062         }
4063
4064         if (STALE_CLIENTID(clid, nn))
4065                 return nfserr_stale_clientid;
4066
4067         /*
4068          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4069          * cached already then we know this is for is for v4.0 and "sessions"
4070          * will be false.
4071          */
4072         WARN_ON_ONCE(cstate->session);
4073         spin_lock(&nn->client_lock);
4074         found = find_confirmed_client(clid, false, nn);
4075         if (!found) {
4076                 spin_unlock(&nn->client_lock);
4077                 return nfserr_expired;
4078         }
4079         atomic_inc(&found->cl_refcount);
4080         spin_unlock(&nn->client_lock);
4081
4082         /* Cache the nfs4_client in cstate! */
4083         cstate->clp = found;
4084         return nfs_ok;
4085 }
4086
4087 __be32
4088 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4089                     struct nfsd4_open *open, struct nfsd_net *nn)
4090 {
4091         clientid_t *clientid = &open->op_clientid;
4092         struct nfs4_client *clp = NULL;
4093         unsigned int strhashval;
4094         struct nfs4_openowner *oo = NULL;
4095         __be32 status;
4096
4097         if (STALE_CLIENTID(&open->op_clientid, nn))
4098                 return nfserr_stale_clientid;
4099         /*
4100          * In case we need it later, after we've already created the
4101          * file and don't want to risk a further failure:
4102          */
4103         open->op_file = nfsd4_alloc_file();
4104         if (open->op_file == NULL)
4105                 return nfserr_jukebox;
4106
4107         status = lookup_clientid(clientid, cstate, nn);
4108         if (status)
4109                 return status;
4110         clp = cstate->clp;
4111
4112         strhashval = ownerstr_hashval(&open->op_owner);
4113         oo = find_openstateowner_str(strhashval, open, clp);
4114         open->op_openowner = oo;
4115         if (!oo) {
4116                 goto new_owner;
4117         }
4118         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4119                 /* Replace unconfirmed owners without checking for replay. */
4120                 release_openowner(oo);
4121                 open->op_openowner = NULL;
4122                 goto new_owner;
4123         }
4124         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4125         if (status)
4126                 return status;
4127         goto alloc_stateid;
4128 new_owner:
4129         oo = alloc_init_open_stateowner(strhashval, open, cstate);
4130         if (oo == NULL)
4131                 return nfserr_jukebox;
4132         open->op_openowner = oo;
4133 alloc_stateid:
4134         open->op_stp = nfs4_alloc_open_stateid(clp);
4135         if (!open->op_stp)
4136                 return nfserr_jukebox;
4137
4138         if (nfsd4_has_session(cstate) &&
4139             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4140                 open->op_odstate = alloc_clnt_odstate(clp);
4141                 if (!open->op_odstate)
4142                         return nfserr_jukebox;
4143         }
4144
4145         return nfs_ok;
4146 }
4147
4148 static inline __be32
4149 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4150 {
4151         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4152                 return nfserr_openmode;
4153         else
4154                 return nfs_ok;
4155 }
4156
4157 static int share_access_to_flags(u32 share_access)
4158 {
4159         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4160 }
4161
4162 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4163 {
4164         struct nfs4_stid *ret;
4165
4166         ret = find_stateid_by_type(cl, s,
4167                                 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4168         if (!ret)
4169                 return NULL;
4170         return delegstateid(ret);
4171 }
4172
4173 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4174 {
4175         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4176                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4177 }
4178
4179 static __be32
4180 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4181                 struct nfs4_delegation **dp)
4182 {
4183         int flags;
4184         __be32 status = nfserr_bad_stateid;
4185         struct nfs4_delegation *deleg;
4186
4187         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4188         if (deleg == NULL)
4189                 goto out;
4190         if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4191                 nfs4_put_stid(&deleg->dl_stid);
4192                 if (cl->cl_minorversion)
4193                         status = nfserr_deleg_revoked;
4194                 goto out;
4195         }
4196         flags = share_access_to_flags(open->op_share_access);
4197         status = nfs4_check_delegmode(deleg, flags);
4198         if (status) {
4199                 nfs4_put_stid(&deleg->dl_stid);
4200                 goto out;
4201         }
4202         *dp = deleg;
4203 out:
4204         if (!nfsd4_is_deleg_cur(open))
4205                 return nfs_ok;
4206         if (status)
4207                 return status;
4208         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4209         return nfs_ok;
4210 }
4211
4212 static inline int nfs4_access_to_access(u32 nfs4_access)
4213 {
4214         int flags = 0;
4215
4216         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4217                 flags |= NFSD_MAY_READ;
4218         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4219                 flags |= NFSD_MAY_WRITE;
4220         return flags;
4221 }
4222
4223 static inline __be32
4224 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4225                 struct nfsd4_open *open)
4226 {
4227         struct iattr iattr = {
4228                 .ia_valid = ATTR_SIZE,
4229                 .ia_size = 0,
4230         };
4231         if (!open->op_truncate)
4232                 return 0;
4233         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4234                 return nfserr_inval;
4235         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4236 }
4237
4238 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4239                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4240                 struct nfsd4_open *open)
4241 {
4242         struct file *filp = NULL;
4243         __be32 status;
4244         int oflag = nfs4_access_to_omode(open->op_share_access);
4245         int access = nfs4_access_to_access(open->op_share_access);
4246         unsigned char old_access_bmap, old_deny_bmap;
4247
4248         spin_lock(&fp->fi_lock);
4249
4250         /*
4251          * Are we trying to set a deny mode that would conflict with
4252          * current access?
4253          */
4254         status = nfs4_file_check_deny(fp, open->op_share_deny);
4255         if (status != nfs_ok) {
4256                 spin_unlock(&fp->fi_lock);
4257                 goto out;
4258         }
4259
4260         /* set access to the file */
4261         status = nfs4_file_get_access(fp, open->op_share_access);
4262         if (status != nfs_ok) {
4263                 spin_unlock(&fp->fi_lock);
4264                 goto out;
4265         }
4266
4267         /* Set access bits in stateid */
4268         old_access_bmap = stp->st_access_bmap;
4269         set_access(open->op_share_access, stp);
4270
4271         /* Set new deny mask */
4272         old_deny_bmap = stp->st_deny_bmap;
4273         set_deny(open->op_share_deny, stp);
4274         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4275
4276         if (!fp->fi_fds[oflag]) {
4277                 spin_unlock(&fp->fi_lock);
4278                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4279                 if (status)
4280                         goto out_put_access;
4281                 spin_lock(&fp->fi_lock);
4282                 if (!fp->fi_fds[oflag]) {
4283                         fp->fi_fds[oflag] = filp;
4284                         filp = NULL;
4285                 }
4286         }
4287         spin_unlock(&fp->fi_lock);
4288         if (filp)
4289                 fput(filp);
4290
4291         status = nfsd4_truncate(rqstp, cur_fh, open);
4292         if (status)
4293                 goto out_put_access;
4294 out:
4295         return status;
4296 out_put_access:
4297         stp->st_access_bmap = old_access_bmap;
4298         nfs4_file_put_access(fp, open->op_share_access);
4299         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4300         goto out;
4301 }
4302
4303 static __be32
4304 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)
4305 {
4306         __be32 status;
4307         unsigned char old_deny_bmap = stp->st_deny_bmap;
4308
4309         if (!test_access(open->op_share_access, stp))
4310                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4311
4312         /* test and set deny mode */
4313         spin_lock(&fp->fi_lock);
4314         status = nfs4_file_check_deny(fp, open->op_share_deny);
4315         if (status == nfs_ok) {
4316                 set_deny(open->op_share_deny, stp);
4317                 fp->fi_share_deny |=
4318                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4319         }
4320         spin_unlock(&fp->fi_lock);
4321
4322         if (status != nfs_ok)
4323                 return status;
4324
4325         status = nfsd4_truncate(rqstp, cur_fh, open);
4326         if (status != nfs_ok)
4327                 reset_union_bmap_deny(old_deny_bmap, stp);
4328         return status;
4329 }
4330
4331 /* Should we give out recallable state?: */
4332 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4333 {
4334         if (clp->cl_cb_state == NFSD4_CB_UP)
4335                 return true;
4336         /*
4337          * In the sessions case, since we don't have to establish a
4338          * separate connection for callbacks, we assume it's OK
4339          * until we hear otherwise:
4340          */
4341         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4342 }
4343
4344 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
4345                                                 int flag)
4346 {
4347         struct file_lock *fl;
4348
4349         fl = locks_alloc_lock();
4350         if (!fl)
4351                 return NULL;
4352         fl->fl_lmops = &nfsd_lease_mng_ops;
4353         fl->fl_flags = FL_DELEG;
4354         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4355         fl->fl_end = OFFSET_MAX;
4356         fl->fl_owner = (fl_owner_t)dp;
4357         fl->fl_pid = current->tgid;
4358         fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file;
4359         return fl;
4360 }
4361
4362 static struct nfs4_delegation *
4363 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4364                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4365 {
4366         int status = 0;
4367         struct nfs4_delegation *dp;
4368         struct file *filp;
4369         struct file_lock *fl;
4370
4371         /*
4372          * The fi_had_conflict and nfs_get_existing_delegation checks
4373          * here are just optimizations; we'll need to recheck them at
4374          * the end:
4375          */
4376         if (fp->fi_had_conflict)
4377                 return ERR_PTR(-EAGAIN);
4378
4379         filp = find_readable_file(fp);
4380         if (!filp) {
4381                 /* We should always have a readable file here */
4382                 WARN_ON_ONCE(1);
4383                 return ERR_PTR(-EBADF);
4384         }
4385         spin_lock(&state_lock);
4386         spin_lock(&fp->fi_lock);
4387         if (nfs4_delegation_exists(clp, fp))
4388                 status = -EAGAIN;
4389         else if (!fp->fi_deleg_file) {
4390                 fp->fi_deleg_file = filp;
4391                 /* increment early to prevent fi_deleg_file from being
4392                  * cleared */
4393                 fp->fi_delegees = 1;
4394                 filp = NULL;
4395         } else
4396                 fp->fi_delegees++;
4397         spin_unlock(&fp->fi_lock);
4398         spin_unlock(&state_lock);
4399         if (filp)
4400                 fput(filp);
4401         if (status)
4402                 return ERR_PTR(status);
4403
4404         status = -ENOMEM;
4405         dp = alloc_init_deleg(clp, fp, fh, odstate);
4406         if (!dp)
4407                 goto out_delegees;
4408
4409         fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
4410         if (!fl)
4411                 goto out_clnt_odstate;
4412
4413         status = vfs_setlease(fp->fi_deleg_file, fl->fl_type, &fl, NULL);
4414         if (fl)
4415                 locks_free_lock(fl);
4416         if (status)
4417                 goto out_clnt_odstate;
4418
4419         spin_lock(&state_lock);
4420         spin_lock(&fp->fi_lock);
4421         if (fp->fi_had_conflict)
4422                 status = -EAGAIN;
4423         else
4424                 status = hash_delegation_locked(dp, fp);
4425         spin_unlock(&fp->fi_lock);
4426         spin_unlock(&state_lock);
4427
4428         if (status)
4429                 goto out_unlock;
4430
4431         return dp;
4432 out_unlock:
4433         vfs_setlease(fp->fi_deleg_file, F_UNLCK, NULL, (void **)&dp);
4434 out_clnt_odstate:
4435         put_clnt_odstate(dp->dl_clnt_odstate);
4436         nfs4_put_stid(&dp->dl_stid);
4437 out_delegees:
4438         put_deleg_file(fp);
4439         return ERR_PTR(status);
4440 }
4441
4442 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4443 {
4444         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4445         if (status == -EAGAIN)
4446                 open->op_why_no_deleg = WND4_CONTENTION;
4447         else {
4448                 open->op_why_no_deleg = WND4_RESOURCE;
4449                 switch (open->op_deleg_want) {
4450                 case NFS4_SHARE_WANT_READ_DELEG:
4451                 case NFS4_SHARE_WANT_WRITE_DELEG:
4452                 case NFS4_SHARE_WANT_ANY_DELEG:
4453                         break;
4454                 case NFS4_SHARE_WANT_CANCEL:
4455                         open->op_why_no_deleg = WND4_CANCELLED;
4456                         break;
4457                 case NFS4_SHARE_WANT_NO_DELEG:
4458                         WARN_ON_ONCE(1);
4459                 }
4460         }
4461 }
4462
4463 /*
4464  * Attempt to hand out a delegation.
4465  *
4466  * Note we don't support write delegations, and won't until the vfs has
4467  * proper support for them.
4468  */
4469 static void
4470 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4471                         struct nfs4_ol_stateid *stp)
4472 {
4473         struct nfs4_delegation *dp;
4474         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4475         struct nfs4_client *clp = stp->st_stid.sc_client;
4476         int cb_up;
4477         int status = 0;
4478
4479         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4480         open->op_recall = 0;
4481         switch (open->op_claim_type) {
4482                 case NFS4_OPEN_CLAIM_PREVIOUS:
4483                         if (!cb_up)
4484                                 open->op_recall = 1;
4485                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4486                                 goto out_no_deleg;
4487                         break;
4488                 case NFS4_OPEN_CLAIM_NULL:
4489                 case NFS4_OPEN_CLAIM_FH:
4490                         /*
4491                          * Let's not give out any delegations till everyone's
4492                          * had the chance to reclaim theirs, *and* until
4493                          * NLM locks have all been reclaimed:
4494                          */
4495                         if (locks_in_grace(clp->net))
4496                                 goto out_no_deleg;
4497                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4498                                 goto out_no_deleg;
4499                         /*
4500                          * Also, if the file was opened for write or
4501                          * create, there's a good chance the client's
4502                          * about to write to it, resulting in an
4503                          * immediate recall (since we don't support
4504                          * write delegations):
4505                          */
4506                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4507                                 goto out_no_deleg;
4508                         if (open->op_create == NFS4_OPEN_CREATE)
4509                                 goto out_no_deleg;
4510                         break;
4511                 default:
4512                         goto out_no_deleg;
4513         }
4514         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4515         if (IS_ERR(dp))
4516                 goto out_no_deleg;
4517
4518         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4519
4520         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4521                 STATEID_VAL(&dp->dl_stid.sc_stateid));
4522         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4523         nfs4_put_stid(&dp->dl_stid);
4524         return;
4525 out_no_deleg:
4526         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4527         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4528             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4529                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4530                 open->op_recall = 1;
4531         }
4532
4533         /* 4.1 client asking for a delegation? */
4534         if (open->op_deleg_want)
4535                 nfsd4_open_deleg_none_ext(open, status);
4536         return;
4537 }
4538
4539 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4540                                         struct nfs4_delegation *dp)
4541 {
4542         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4543             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4544                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4545                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4546         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4547                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4548                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4549                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4550         }
4551         /* Otherwise the client must be confused wanting a delegation
4552          * it already has, therefore we don't return
4553          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4554          */
4555 }
4556
4557 __be32
4558 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4559 {
4560         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4561         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4562         struct nfs4_file *fp = NULL;
4563         struct nfs4_ol_stateid *stp = NULL;
4564         struct nfs4_delegation *dp = NULL;
4565         __be32 status;
4566         bool new_stp = false;
4567
4568         /*
4569          * Lookup file; if found, lookup stateid and check open request,
4570          * and check for delegations in the process of being recalled.
4571          * If not found, create the nfs4_file struct
4572          */
4573         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4574         if (fp != open->op_file) {
4575                 status = nfs4_check_deleg(cl, open, &dp);
4576                 if (status)
4577                         goto out;
4578                 stp = nfsd4_find_and_lock_existing_open(fp, open);
4579         } else {
4580                 open->op_file = NULL;
4581                 status = nfserr_bad_stateid;
4582                 if (nfsd4_is_deleg_cur(open))
4583                         goto out;
4584         }
4585
4586         if (!stp) {
4587                 stp = init_open_stateid(fp, open);
4588                 if (!open->op_stp)
4589                         new_stp = true;
4590         }
4591
4592         /*
4593          * OPEN the file, or upgrade an existing OPEN.
4594          * If truncate fails, the OPEN fails.
4595          *
4596          * stp is already locked.
4597          */
4598         if (!new_stp) {
4599                 /* Stateid was found, this is an OPEN upgrade */
4600                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4601                 if (status) {
4602                         mutex_unlock(&stp->st_mutex);
4603                         goto out;
4604                 }
4605         } else {
4606                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4607                 if (status) {
4608                         stp->st_stid.sc_type = NFS4_CLOSED_STID;
4609                         release_open_stateid(stp);
4610                         mutex_unlock(&stp->st_mutex);
4611                         goto out;
4612                 }
4613
4614                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4615                                                         open->op_odstate);
4616                 if (stp->st_clnt_odstate == open->op_odstate)
4617                         open->op_odstate = NULL;
4618         }
4619
4620         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4621         mutex_unlock(&stp->st_mutex);
4622
4623         if (nfsd4_has_session(&resp->cstate)) {
4624                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4625                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4626                         open->op_why_no_deleg = WND4_NOT_WANTED;
4627                         goto nodeleg;
4628                 }
4629         }
4630
4631         /*
4632         * Attempt to hand out a delegation. No error return, because the
4633         * OPEN succeeds even if we fail.
4634         */
4635         nfs4_open_delegation(current_fh, open, stp);
4636 nodeleg:
4637         status = nfs_ok;
4638
4639         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4640                 STATEID_VAL(&stp->st_stid.sc_stateid));
4641 out:
4642         /* 4.1 client trying to upgrade/downgrade delegation? */
4643         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4644             open->op_deleg_want)
4645                 nfsd4_deleg_xgrade_none_ext(open, dp);
4646
4647         if (fp)
4648                 put_nfs4_file(fp);
4649         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4650                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4651         /*
4652         * To finish the open response, we just need to set the rflags.
4653         */
4654         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4655         if (nfsd4_has_session(&resp->cstate))
4656                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
4657         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
4658                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4659
4660         if (dp)
4661                 nfs4_put_stid(&dp->dl_stid);
4662         if (stp)
4663                 nfs4_put_stid(&stp->st_stid);
4664
4665         return status;
4666 }
4667
4668 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4669                               struct nfsd4_open *open)
4670 {
4671         if (open->op_openowner) {
4672                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4673
4674                 nfsd4_cstate_assign_replay(cstate, so);
4675                 nfs4_put_stateowner(so);
4676         }
4677         if (open->op_file)
4678                 kmem_cache_free(file_slab, open->op_file);
4679         if (open->op_stp)
4680                 nfs4_put_stid(&open->op_stp->st_stid);
4681         if (open->op_odstate)
4682                 kmem_cache_free(odstate_slab, open->op_odstate);
4683 }
4684
4685 __be32
4686 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4687             union nfsd4_op_u *u)
4688 {
4689         clientid_t *clid = &u->renew;
4690         struct nfs4_client *clp;
4691         __be32 status;
4692         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4693
4694         dprintk("process_renew(%08x/%08x): starting\n",
4695                         clid->cl_boot, clid->cl_id);
4696         status = lookup_clientid(clid, cstate, nn);
4697         if (status)
4698                 goto out;
4699         clp = cstate->clp;
4700         status = nfserr_cb_path_down;
4701         if (!list_empty(&clp->cl_delegations)
4702                         && clp->cl_cb_state != NFSD4_CB_UP)
4703                 goto out;
4704         status = nfs_ok;
4705 out:
4706         return status;
4707 }
4708
4709 void
4710 nfsd4_end_grace(struct nfsd_net *nn)
4711 {
4712         /* do nothing if grace period already ended */
4713         if (nn->grace_ended)
4714                 return;
4715
4716         dprintk("NFSD: end of grace period\n");
4717         nn->grace_ended = true;
4718         /*
4719          * If the server goes down again right now, an NFSv4
4720          * client will still be allowed to reclaim after it comes back up,
4721          * even if it hasn't yet had a chance to reclaim state this time.
4722          *
4723          */
4724         nfsd4_record_grace_done(nn);
4725         /*
4726          * At this point, NFSv4 clients can still reclaim.  But if the
4727          * server crashes, any that have not yet reclaimed will be out
4728          * of luck on the next boot.
4729          *
4730          * (NFSv4.1+ clients are considered to have reclaimed once they
4731          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
4732          * have reclaimed after their first OPEN.)
4733          */
4734         locks_end_grace(&nn->nfsd4_manager);
4735         /*
4736          * At this point, and once lockd and/or any other containers
4737          * exit their grace period, further reclaims will fail and
4738          * regular locking can resume.
4739          */
4740 }
4741
4742 /*
4743  * If we've waited a lease period but there are still clients trying to
4744  * reclaim, wait a little longer to give them a chance to finish.
4745  */
4746 static bool clients_still_reclaiming(struct nfsd_net *nn)
4747 {
4748         unsigned long now = get_seconds();
4749         unsigned long double_grace_period_end = nn->boot_time +
4750                                                 2 * nn->nfsd4_lease;
4751
4752         if (!nn->somebody_reclaimed)
4753                 return false;
4754         nn->somebody_reclaimed = false;
4755         /*
4756          * If we've given them *two* lease times to reclaim, and they're
4757          * still not done, give up:
4758          */
4759         if (time_after(now, double_grace_period_end))
4760                 return false;
4761         return true;
4762 }
4763
4764 static time_t
4765 nfs4_laundromat(struct nfsd_net *nn)
4766 {
4767         struct nfs4_client *clp;
4768         struct nfs4_openowner *oo;
4769         struct nfs4_delegation *dp;
4770         struct nfs4_ol_stateid *stp;
4771         struct nfsd4_blocked_lock *nbl;
4772         struct list_head *pos, *next, reaplist;
4773         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4774         time_t t, new_timeo = nn->nfsd4_lease;
4775
4776         dprintk("NFSD: laundromat service - starting\n");
4777
4778         if (clients_still_reclaiming(nn)) {
4779                 new_timeo = 0;
4780                 goto out;
4781         }
4782         nfsd4_end_grace(nn);
4783         INIT_LIST_HEAD(&reaplist);
4784         spin_lock(&nn->client_lock);
4785         list_for_each_safe(pos, next, &nn->client_lru) {
4786                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4787                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4788                         t = clp->cl_time - cutoff;
4789                         new_timeo = min(new_timeo, t);
4790                         break;
4791                 }
4792                 if (mark_client_expired_locked(clp)) {
4793                         dprintk("NFSD: client in use (clientid %08x)\n",
4794                                 clp->cl_clientid.cl_id);
4795                         continue;
4796                 }
4797                 list_add(&clp->cl_lru, &reaplist);
4798         }
4799         spin_unlock(&nn->client_lock);
4800         list_for_each_safe(pos, next, &reaplist) {
4801                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4802                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4803                         clp->cl_clientid.cl_id);
4804                 list_del_init(&clp->cl_lru);
4805                 expire_client(clp);
4806         }
4807         spin_lock(&state_lock);
4808         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4809                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4810                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4811                         t = dp->dl_time - cutoff;
4812                         new_timeo = min(new_timeo, t);
4813                         break;
4814                 }
4815                 WARN_ON(!unhash_delegation_locked(dp));
4816                 list_add(&dp->dl_recall_lru, &reaplist);
4817         }
4818         spin_unlock(&state_lock);
4819         while (!list_empty(&reaplist)) {
4820                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4821                                         dl_recall_lru);
4822                 list_del_init(&dp->dl_recall_lru);
4823                 revoke_delegation(dp);
4824         }
4825
4826         spin_lock(&nn->client_lock);
4827         while (!list_empty(&nn->close_lru)) {
4828                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4829                                         oo_close_lru);
4830                 if (time_after((unsigned long)oo->oo_time,
4831                                (unsigned long)cutoff)) {
4832                         t = oo->oo_time - cutoff;
4833                         new_timeo = min(new_timeo, t);
4834                         break;
4835                 }
4836                 list_del_init(&oo->oo_close_lru);
4837                 stp = oo->oo_last_closed_stid;
4838                 oo->oo_last_closed_stid = NULL;
4839                 spin_unlock(&nn->client_lock);
4840                 nfs4_put_stid(&stp->st_stid);
4841                 spin_lock(&nn->client_lock);
4842         }
4843         spin_unlock(&nn->client_lock);
4844
4845         /*
4846          * It's possible for a client to try and acquire an already held lock
4847          * that is being held for a long time, and then lose interest in it.
4848          * So, we clean out any un-revisited request after a lease period
4849          * under the assumption that the client is no longer interested.
4850          *
4851          * RFC5661, sec. 9.6 states that the client must not rely on getting
4852          * notifications and must continue to poll for locks, even when the
4853          * server supports them. Thus this shouldn't lead to clients blocking
4854          * indefinitely once the lock does become free.
4855          */
4856         BUG_ON(!list_empty(&reaplist));
4857         spin_lock(&nn->blocked_locks_lock);
4858         while (!list_empty(&nn->blocked_locks_lru)) {
4859                 nbl = list_first_entry(&nn->blocked_locks_lru,
4860                                         struct nfsd4_blocked_lock, nbl_lru);
4861                 if (time_after((unsigned long)nbl->nbl_time,
4862                                (unsigned long)cutoff)) {
4863                         t = nbl->nbl_time - cutoff;
4864                         new_timeo = min(new_timeo, t);
4865                         break;
4866                 }
4867                 list_move(&nbl->nbl_lru, &reaplist);
4868                 list_del_init(&nbl->nbl_list);
4869         }
4870         spin_unlock(&nn->blocked_locks_lock);
4871
4872         while (!list_empty(&reaplist)) {
4873                 nbl = list_first_entry(&reaplist,
4874                                         struct nfsd4_blocked_lock, nbl_lru);
4875                 list_del_init(&nbl->nbl_lru);
4876                 free_blocked_lock(nbl);
4877         }
4878 out:
4879         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4880         return new_timeo;
4881 }
4882
4883 static struct workqueue_struct *laundry_wq;
4884 static void laundromat_main(struct work_struct *);
4885
4886 static void
4887 laundromat_main(struct work_struct *laundry)
4888 {
4889         time_t t;
4890         struct delayed_work *dwork = to_delayed_work(laundry);
4891         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4892                                            laundromat_work);
4893
4894         t = nfs4_laundromat(nn);
4895         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4896         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4897 }
4898
4899 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4900 {
4901         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4902                 return nfserr_bad_stateid;
4903         return nfs_ok;
4904 }
4905
4906 static inline int
4907 access_permit_read(struct nfs4_ol_stateid *stp)
4908 {
4909         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4910                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4911                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4912 }
4913
4914 static inline int
4915 access_permit_write(struct nfs4_ol_stateid *stp)
4916 {
4917         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4918                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4919 }
4920
4921 static
4922 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4923 {
4924         __be32 status = nfserr_openmode;
4925
4926         /* For lock stateid's, we test the parent open, not the lock: */
4927         if (stp->st_openstp)
4928                 stp = stp->st_openstp;
4929         if ((flags & WR_STATE) && !access_permit_write(stp))
4930                 goto out;
4931         if ((flags & RD_STATE) && !access_permit_read(stp))
4932                 goto out;
4933         status = nfs_ok;
4934 out:
4935         return status;
4936 }
4937
4938 static inline __be32
4939 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4940 {
4941         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4942                 return nfs_ok;
4943         else if (opens_in_grace(net)) {
4944                 /* Answer in remaining cases depends on existence of
4945                  * conflicting state; so we must wait out the grace period. */
4946                 return nfserr_grace;
4947         } else if (flags & WR_STATE)
4948                 return nfs4_share_conflict(current_fh,
4949                                 NFS4_SHARE_DENY_WRITE);
4950         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4951                 return nfs4_share_conflict(current_fh,
4952                                 NFS4_SHARE_DENY_READ);
4953 }
4954
4955 /*
4956  * Allow READ/WRITE during grace period on recovered state only for files
4957  * that are not able to provide mandatory locking.
4958  */
4959 static inline int
4960 grace_disallows_io(struct net *net, struct inode *inode)
4961 {
4962         return opens_in_grace(net) && mandatory_lock(inode);
4963 }
4964
4965 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4966 {
4967         /*
4968          * When sessions are used the stateid generation number is ignored
4969          * when it is zero.
4970          */
4971         if (has_session && in->si_generation == 0)
4972                 return nfs_ok;
4973
4974         if (in->si_generation == ref->si_generation)
4975                 return nfs_ok;
4976
4977         /* If the client sends us a stateid from the future, it's buggy: */
4978         if (nfsd4_stateid_generation_after(in, ref))
4979                 return nfserr_bad_stateid;
4980         /*
4981          * However, we could see a stateid from the past, even from a
4982          * non-buggy client.  For example, if the client sends a lock
4983          * while some IO is outstanding, the lock may bump si_generation
4984          * while the IO is still in flight.  The client could avoid that
4985          * situation by waiting for responses on all the IO requests,
4986          * but better performance may result in retrying IO that
4987          * receives an old_stateid error if requests are rarely
4988          * reordered in flight:
4989          */
4990         return nfserr_old_stateid;
4991 }
4992
4993 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
4994 {
4995         __be32 ret;
4996
4997         spin_lock(&s->sc_lock);
4998         ret = nfsd4_verify_open_stid(s);
4999         if (ret == nfs_ok)
5000                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
5001         spin_unlock(&s->sc_lock);
5002         return ret;
5003 }
5004
5005 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
5006 {
5007         if (ols->st_stateowner->so_is_open_owner &&
5008             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
5009                 return nfserr_bad_stateid;
5010         return nfs_ok;
5011 }
5012
5013 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
5014 {
5015         struct nfs4_stid *s;
5016         __be32 status = nfserr_bad_stateid;
5017
5018         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5019                 CLOSE_STATEID(stateid))
5020                 return status;
5021         /* Client debugging aid. */
5022         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
5023                 char addr_str[INET6_ADDRSTRLEN];
5024                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
5025                                  sizeof(addr_str));
5026                 pr_warn_ratelimited("NFSD: client %s testing state ID "
5027                                         "with incorrect client ID\n", addr_str);
5028                 return status;
5029         }
5030         spin_lock(&cl->cl_lock);
5031         s = find_stateid_locked(cl, stateid);
5032         if (!s)
5033                 goto out_unlock;
5034         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
5035         if (status)
5036                 goto out_unlock;
5037         switch (s->sc_type) {
5038         case NFS4_DELEG_STID:
5039                 status = nfs_ok;
5040                 break;
5041         case NFS4_REVOKED_DELEG_STID:
5042                 status = nfserr_deleg_revoked;
5043                 break;
5044         case NFS4_OPEN_STID:
5045         case NFS4_LOCK_STID:
5046                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5047                 break;
5048         default:
5049                 printk("unknown stateid type %x\n", s->sc_type);
5050                 /* Fallthrough */
5051         case NFS4_CLOSED_STID:
5052         case NFS4_CLOSED_DELEG_STID:
5053                 status = nfserr_bad_stateid;
5054         }
5055 out_unlock:
5056         spin_unlock(&cl->cl_lock);
5057         return status;
5058 }
5059
5060 __be32
5061 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5062                      stateid_t *stateid, unsigned char typemask,
5063                      struct nfs4_stid **s, struct nfsd_net *nn)
5064 {
5065         __be32 status;
5066         bool return_revoked = false;
5067
5068         /*
5069          *  only return revoked delegations if explicitly asked.
5070          *  otherwise we report revoked or bad_stateid status.
5071          */
5072         if (typemask & NFS4_REVOKED_DELEG_STID)
5073                 return_revoked = true;
5074         else if (typemask & NFS4_DELEG_STID)
5075                 typemask |= NFS4_REVOKED_DELEG_STID;
5076
5077         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5078                 CLOSE_STATEID(stateid))
5079                 return nfserr_bad_stateid;
5080         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
5081         if (status == nfserr_stale_clientid) {
5082                 if (cstate->session)
5083                         return nfserr_bad_stateid;
5084                 return nfserr_stale_stateid;
5085         }
5086         if (status)
5087                 return status;
5088         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5089         if (!*s)
5090                 return nfserr_bad_stateid;
5091         if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5092                 nfs4_put_stid(*s);
5093                 if (cstate->minorversion)
5094                         return nfserr_deleg_revoked;
5095                 return nfserr_bad_stateid;
5096         }
5097         return nfs_ok;
5098 }
5099
5100 static struct file *
5101 nfs4_find_file(struct nfs4_stid *s, int flags)
5102 {
5103         if (!s)
5104                 return NULL;
5105
5106         switch (s->sc_type) {
5107         case NFS4_DELEG_STID:
5108                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5109                         return NULL;
5110                 return get_file(s->sc_file->fi_deleg_file);
5111         case NFS4_OPEN_STID:
5112         case NFS4_LOCK_STID:
5113                 if (flags & RD_STATE)
5114                         return find_readable_file(s->sc_file);
5115                 else
5116                         return find_writeable_file(s->sc_file);
5117                 break;
5118         }
5119
5120         return NULL;
5121 }
5122
5123 static __be32
5124 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
5125 {
5126         __be32 status;
5127
5128         status = nfsd4_check_openowner_confirmed(ols);
5129         if (status)
5130                 return status;
5131         return nfs4_check_openmode(ols, flags);
5132 }
5133
5134 static __be32
5135 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5136                 struct file **filpp, bool *tmp_file, int flags)
5137 {
5138         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5139         struct file *file;
5140         __be32 status;
5141
5142         file = nfs4_find_file(s, flags);
5143         if (file) {
5144                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5145                                 acc | NFSD_MAY_OWNER_OVERRIDE);
5146                 if (status) {
5147                         fput(file);
5148                         return status;
5149                 }
5150
5151                 *filpp = file;
5152         } else {
5153                 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
5154                 if (status)
5155                         return status;
5156
5157                 if (tmp_file)
5158                         *tmp_file = true;
5159         }
5160
5161         return 0;
5162 }
5163
5164 /*
5165  * Checks for stateid operations
5166  */
5167 __be32
5168 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5169                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5170                 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
5171 {
5172         struct inode *ino = d_inode(fhp->fh_dentry);
5173         struct net *net = SVC_NET(rqstp);
5174         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5175         struct nfs4_stid *s = NULL;
5176         __be32 status;
5177
5178         if (filpp)
5179                 *filpp = NULL;
5180         if (tmp_file)
5181                 *tmp_file = false;
5182
5183         if (grace_disallows_io(net, ino))
5184                 return nfserr_grace;
5185
5186         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5187                 status = check_special_stateids(net, fhp, stateid, flags);
5188                 goto done;
5189         }
5190
5191         status = nfsd4_lookup_stateid(cstate, stateid,
5192                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5193                                 &s, nn);
5194         if (status)
5195                 return status;
5196         status = nfsd4_stid_check_stateid_generation(stateid, s,
5197                         nfsd4_has_session(cstate));
5198         if (status)
5199                 goto out;
5200
5201         switch (s->sc_type) {
5202         case NFS4_DELEG_STID:
5203                 status = nfs4_check_delegmode(delegstateid(s), flags);
5204                 break;
5205         case NFS4_OPEN_STID:
5206         case NFS4_LOCK_STID:
5207                 status = nfs4_check_olstateid(openlockstateid(s), flags);
5208                 break;
5209         default:
5210                 status = nfserr_bad_stateid;
5211                 break;
5212         }
5213         if (status)
5214                 goto out;
5215         status = nfs4_check_fh(fhp, s);
5216
5217 done:
5218         if (!status && filpp)
5219                 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
5220 out:
5221         if (s)
5222                 nfs4_put_stid(s);
5223         return status;
5224 }
5225
5226 /*
5227  * Test if the stateid is valid
5228  */
5229 __be32
5230 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5231                    union nfsd4_op_u *u)
5232 {
5233         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5234         struct nfsd4_test_stateid_id *stateid;
5235         struct nfs4_client *cl = cstate->session->se_client;
5236
5237         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5238                 stateid->ts_id_status =
5239                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5240
5241         return nfs_ok;
5242 }
5243
5244 static __be32
5245 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5246 {
5247         struct nfs4_ol_stateid *stp = openlockstateid(s);
5248         __be32 ret;
5249
5250         ret = nfsd4_lock_ol_stateid(stp);
5251         if (ret)
5252                 goto out_put_stid;
5253
5254         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5255         if (ret)
5256                 goto out;
5257
5258         ret = nfserr_locks_held;
5259         if (check_for_locks(stp->st_stid.sc_file,
5260                             lockowner(stp->st_stateowner)))
5261                 goto out;
5262
5263         release_lock_stateid(stp);
5264         ret = nfs_ok;
5265
5266 out:
5267         mutex_unlock(&stp->st_mutex);
5268 out_put_stid:
5269         nfs4_put_stid(s);
5270         return ret;
5271 }
5272
5273 __be32
5274 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5275                    union nfsd4_op_u *u)
5276 {
5277         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5278         stateid_t *stateid = &free_stateid->fr_stateid;
5279         struct nfs4_stid *s;
5280         struct nfs4_delegation *dp;
5281         struct nfs4_client *cl = cstate->session->se_client;
5282         __be32 ret = nfserr_bad_stateid;
5283
5284         spin_lock(&cl->cl_lock);
5285         s = find_stateid_locked(cl, stateid);
5286         if (!s)
5287                 goto out_unlock;
5288         spin_lock(&s->sc_lock);
5289         switch (s->sc_type) {
5290         case NFS4_DELEG_STID:
5291                 ret = nfserr_locks_held;
5292                 break;
5293         case NFS4_OPEN_STID:
5294                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5295                 if (ret)
5296                         break;
5297                 ret = nfserr_locks_held;
5298                 break;
5299         case NFS4_LOCK_STID:
5300                 spin_unlock(&s->sc_lock);
5301                 refcount_inc(&s->sc_count);
5302                 spin_unlock(&cl->cl_lock);
5303                 ret = nfsd4_free_lock_stateid(stateid, s);
5304                 goto out;
5305         case NFS4_REVOKED_DELEG_STID:
5306                 spin_unlock(&s->sc_lock);
5307                 dp = delegstateid(s);
5308                 list_del_init(&dp->dl_recall_lru);
5309                 spin_unlock(&cl->cl_lock);
5310                 nfs4_put_stid(s);
5311                 ret = nfs_ok;
5312                 goto out;
5313         /* Default falls through and returns nfserr_bad_stateid */
5314         }
5315         spin_unlock(&s->sc_lock);
5316 out_unlock:
5317         spin_unlock(&cl->cl_lock);
5318 out:
5319         return ret;
5320 }
5321
5322 static inline int
5323 setlkflg (int type)
5324 {
5325         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5326                 RD_STATE : WR_STATE;
5327 }
5328
5329 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5330 {
5331         struct svc_fh *current_fh = &cstate->current_fh;
5332         struct nfs4_stateowner *sop = stp->st_stateowner;
5333         __be32 status;
5334
5335         status = nfsd4_check_seqid(cstate, sop, seqid);
5336         if (status)
5337                 return status;
5338         status = nfsd4_lock_ol_stateid(stp);
5339         if (status != nfs_ok)
5340                 return status;
5341         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5342         if (status == nfs_ok)
5343                 status = nfs4_check_fh(current_fh, &stp->st_stid);
5344         if (status != nfs_ok)
5345                 mutex_unlock(&stp->st_mutex);
5346         return status;
5347 }
5348
5349 /*
5350  * Checks for sequence id mutating operations.
5351  */
5352 static __be32
5353 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5354                          stateid_t *stateid, char typemask,
5355                          struct nfs4_ol_stateid **stpp,
5356                          struct nfsd_net *nn)
5357 {
5358         __be32 status;
5359         struct nfs4_stid *s;
5360         struct nfs4_ol_stateid *stp = NULL;
5361
5362         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5363                 seqid, STATEID_VAL(stateid));
5364
5365         *stpp = NULL;
5366         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5367         if (status)
5368                 return status;
5369         stp = openlockstateid(s);
5370         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5371
5372         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5373         if (!status)
5374                 *stpp = stp;
5375         else
5376                 nfs4_put_stid(&stp->st_stid);
5377         return status;
5378 }
5379
5380 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5381                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5382 {
5383         __be32 status;
5384         struct nfs4_openowner *oo;
5385         struct nfs4_ol_stateid *stp;
5386
5387         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5388                                                 NFS4_OPEN_STID, &stp, nn);
5389         if (status)
5390                 return status;
5391         oo = openowner(stp->st_stateowner);
5392         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5393                 mutex_unlock(&stp->st_mutex);
5394                 nfs4_put_stid(&stp->st_stid);
5395                 return nfserr_bad_stateid;
5396         }
5397         *stpp = stp;
5398         return nfs_ok;
5399 }
5400
5401 __be32
5402 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5403                    union nfsd4_op_u *u)
5404 {
5405         struct nfsd4_open_confirm *oc = &u->open_confirm;
5406         __be32 status;
5407         struct nfs4_openowner *oo;
5408         struct nfs4_ol_stateid *stp;
5409         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5410
5411         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5412                         cstate->current_fh.fh_dentry);
5413
5414         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5415         if (status)
5416                 return status;
5417
5418         status = nfs4_preprocess_seqid_op(cstate,
5419                                         oc->oc_seqid, &oc->oc_req_stateid,
5420                                         NFS4_OPEN_STID, &stp, nn);
5421         if (status)
5422                 goto out;
5423         oo = openowner(stp->st_stateowner);
5424         status = nfserr_bad_stateid;
5425         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5426                 mutex_unlock(&stp->st_mutex);
5427                 goto put_stateid;
5428         }
5429         oo->oo_flags |= NFS4_OO_CONFIRMED;
5430         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5431         mutex_unlock(&stp->st_mutex);
5432         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5433                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5434
5435         nfsd4_client_record_create(oo->oo_owner.so_client);
5436         status = nfs_ok;
5437 put_stateid:
5438         nfs4_put_stid(&stp->st_stid);
5439 out:
5440         nfsd4_bump_seqid(cstate, status);
5441         return status;
5442 }
5443
5444 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5445 {
5446         if (!test_access(access, stp))
5447                 return;
5448         nfs4_file_put_access(stp->st_stid.sc_file, access);
5449         clear_access(access, stp);
5450 }
5451
5452 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5453 {
5454         switch (to_access) {
5455         case NFS4_SHARE_ACCESS_READ:
5456                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5457                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5458                 break;
5459         case NFS4_SHARE_ACCESS_WRITE:
5460                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5461                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5462                 break;
5463         case NFS4_SHARE_ACCESS_BOTH:
5464                 break;
5465         default:
5466                 WARN_ON_ONCE(1);
5467         }
5468 }
5469
5470 __be32
5471 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5472                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
5473 {
5474         struct nfsd4_open_downgrade *od = &u->open_downgrade;
5475         __be32 status;
5476         struct nfs4_ol_stateid *stp;
5477         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5478
5479         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5480                         cstate->current_fh.fh_dentry);
5481
5482         /* We don't yet support WANT bits: */
5483         if (od->od_deleg_want)
5484                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5485                         od->od_deleg_want);
5486
5487         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5488                                         &od->od_stateid, &stp, nn);
5489         if (status)
5490                 goto out;
5491         status = nfserr_inval;
5492         if (!test_access(od->od_share_access, stp)) {
5493                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5494                         stp->st_access_bmap, od->od_share_access);
5495                 goto put_stateid;
5496         }
5497         if (!test_deny(od->od_share_deny, stp)) {
5498                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5499                         stp->st_deny_bmap, od->od_share_deny);
5500                 goto put_stateid;
5501         }
5502         nfs4_stateid_downgrade(stp, od->od_share_access);
5503         reset_union_bmap_deny(od->od_share_deny, stp);
5504         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5505         status = nfs_ok;
5506 put_stateid:
5507         mutex_unlock(&stp->st_mutex);
5508         nfs4_put_stid(&stp->st_stid);
5509 out:
5510         nfsd4_bump_seqid(cstate, status);
5511         return status;
5512 }
5513
5514 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5515 {
5516         struct nfs4_client *clp = s->st_stid.sc_client;
5517         bool unhashed;
5518         LIST_HEAD(reaplist);
5519
5520         spin_lock(&clp->cl_lock);
5521         unhashed = unhash_open_stateid(s, &reaplist);
5522
5523         if (clp->cl_minorversion) {
5524                 if (unhashed)
5525                         put_ol_stateid_locked(s, &reaplist);
5526                 spin_unlock(&clp->cl_lock);
5527                 free_ol_stateid_reaplist(&reaplist);
5528         } else {
5529                 spin_unlock(&clp->cl_lock);
5530                 free_ol_stateid_reaplist(&reaplist);
5531                 if (unhashed)
5532                         move_to_close_lru(s, clp->net);
5533         }
5534 }
5535
5536 /*
5537  * nfs4_unlock_state() called after encode
5538  */
5539 __be32
5540 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5541                 union nfsd4_op_u *u)
5542 {
5543         struct nfsd4_close *close = &u->close;
5544         __be32 status;
5545         struct nfs4_ol_stateid *stp;
5546         struct net *net = SVC_NET(rqstp);
5547         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5548
5549         dprintk("NFSD: nfsd4_close on file %pd\n",
5550                         cstate->current_fh.fh_dentry);
5551
5552         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5553                                         &close->cl_stateid,
5554                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
5555                                         &stp, nn);
5556         nfsd4_bump_seqid(cstate, status);
5557         if (status)
5558                 goto out;
5559
5560         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5561
5562         /*
5563          * Technically we don't _really_ have to increment or copy it, since
5564          * it should just be gone after this operation and we clobber the
5565          * copied value below, but we continue to do so here just to ensure
5566          * that racing ops see that there was a state change.
5567          */
5568         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5569
5570         nfsd4_close_open_stateid(stp);
5571         mutex_unlock(&stp->st_mutex);
5572
5573         /* v4.1+ suggests that we send a special stateid in here, since the
5574          * clients should just ignore this anyway. Since this is not useful
5575          * for v4.0 clients either, we set it to the special close_stateid
5576          * universally.
5577          *
5578          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
5579          */
5580         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
5581
5582         /* put reference from nfs4_preprocess_seqid_op */
5583         nfs4_put_stid(&stp->st_stid);
5584 out:
5585         return status;
5586 }
5587
5588 __be32
5589 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5590                   union nfsd4_op_u *u)
5591 {
5592         struct nfsd4_delegreturn *dr = &u->delegreturn;
5593         struct nfs4_delegation *dp;
5594         stateid_t *stateid = &dr->dr_stateid;
5595         struct nfs4_stid *s;
5596         __be32 status;
5597         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5598
5599         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5600                 return status;
5601
5602         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5603         if (status)
5604                 goto out;
5605         dp = delegstateid(s);
5606         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
5607         if (status)
5608                 goto put_stateid;
5609
5610         destroy_delegation(dp);
5611 put_stateid:
5612         nfs4_put_stid(&dp->dl_stid);
5613 out:
5614         return status;
5615 }
5616
5617 static inline u64
5618 end_offset(u64 start, u64 len)
5619 {
5620         u64 end;
5621
5622         end = start + len;
5623         return end >= start ? end: NFS4_MAX_UINT64;
5624 }
5625
5626 /* last octet in a range */
5627 static inline u64
5628 last_byte_offset(u64 start, u64 len)
5629 {
5630         u64 end;
5631
5632         WARN_ON_ONCE(!len);
5633         end = start + len;
5634         return end > start ? end - 1: NFS4_MAX_UINT64;
5635 }
5636
5637 /*
5638  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5639  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5640  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
5641  * locking, this prevents us from being completely protocol-compliant.  The
5642  * real solution to this problem is to start using unsigned file offsets in
5643  * the VFS, but this is a very deep change!
5644  */
5645 static inline void
5646 nfs4_transform_lock_offset(struct file_lock *lock)
5647 {
5648         if (lock->fl_start < 0)
5649                 lock->fl_start = OFFSET_MAX;
5650         if (lock->fl_end < 0)
5651                 lock->fl_end = OFFSET_MAX;
5652 }
5653
5654 static fl_owner_t
5655 nfsd4_fl_get_owner(fl_owner_t owner)
5656 {
5657         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5658
5659         nfs4_get_stateowner(&lo->lo_owner);
5660         return owner;
5661 }
5662
5663 static void
5664 nfsd4_fl_put_owner(fl_owner_t owner)
5665 {
5666         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5667
5668         if (lo)
5669                 nfs4_put_stateowner(&lo->lo_owner);
5670 }
5671
5672 static void
5673 nfsd4_lm_notify(struct file_lock *fl)
5674 {
5675         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
5676         struct net                      *net = lo->lo_owner.so_client->net;
5677         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
5678         struct nfsd4_blocked_lock       *nbl = container_of(fl,
5679                                                 struct nfsd4_blocked_lock, nbl_lock);
5680         bool queue = false;
5681
5682         /* An empty list means that something else is going to be using it */
5683         spin_lock(&nn->blocked_locks_lock);
5684         if (!list_empty(&nbl->nbl_list)) {
5685                 list_del_init(&nbl->nbl_list);
5686                 list_del_init(&nbl->nbl_lru);
5687                 queue = true;
5688         }
5689         spin_unlock(&nn->blocked_locks_lock);
5690
5691         if (queue)
5692                 nfsd4_run_cb(&nbl->nbl_cb);
5693 }
5694
5695 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
5696         .lm_notify = nfsd4_lm_notify,
5697         .lm_get_owner = nfsd4_fl_get_owner,
5698         .lm_put_owner = nfsd4_fl_put_owner,
5699 };
5700
5701 static inline void
5702 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5703 {
5704         struct nfs4_lockowner *lo;
5705
5706         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5707                 lo = (struct nfs4_lockowner *) fl->fl_owner;
5708                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5709                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
5710                 if (!deny->ld_owner.data)
5711                         /* We just don't care that much */
5712                         goto nevermind;
5713                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5714                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5715         } else {
5716 nevermind:
5717                 deny->ld_owner.len = 0;
5718                 deny->ld_owner.data = NULL;
5719                 deny->ld_clientid.cl_boot = 0;
5720                 deny->ld_clientid.cl_id = 0;
5721         }
5722         deny->ld_start = fl->fl_start;
5723         deny->ld_length = NFS4_MAX_UINT64;
5724         if (fl->fl_end != NFS4_MAX_UINT64)
5725                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5726         deny->ld_type = NFS4_READ_LT;
5727         if (fl->fl_type != F_RDLCK)
5728                 deny->ld_type = NFS4_WRITE_LT;
5729 }
5730
5731 static struct nfs4_lockowner *
5732 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5733 {
5734         unsigned int strhashval = ownerstr_hashval(owner);
5735         struct nfs4_stateowner *so;
5736
5737         lockdep_assert_held(&clp->cl_lock);
5738
5739         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5740                             so_strhash) {
5741                 if (so->so_is_open_owner)
5742                         continue;
5743                 if (same_owner_str(so, owner))
5744                         return lockowner(nfs4_get_stateowner(so));
5745         }
5746         return NULL;
5747 }
5748
5749 static struct nfs4_lockowner *
5750 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5751 {
5752         struct nfs4_lockowner *lo;
5753
5754         spin_lock(&clp->cl_lock);
5755         lo = find_lockowner_str_locked(clp, owner);
5756         spin_unlock(&clp->cl_lock);
5757         return lo;
5758 }
5759
5760 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5761 {
5762         unhash_lockowner_locked(lockowner(sop));
5763 }
5764
5765 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5766 {
5767         struct nfs4_lockowner *lo = lockowner(sop);
5768
5769         kmem_cache_free(lockowner_slab, lo);
5770 }
5771
5772 static const struct nfs4_stateowner_operations lockowner_ops = {
5773         .so_unhash =    nfs4_unhash_lockowner,
5774         .so_free =      nfs4_free_lockowner,
5775 };
5776
5777 /*
5778  * Alloc a lock owner structure.
5779  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5780  * occurred.
5781  *
5782  * strhashval = ownerstr_hashval
5783  */
5784 static struct nfs4_lockowner *
5785 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5786                            struct nfs4_ol_stateid *open_stp,
5787                            struct nfsd4_lock *lock)
5788 {
5789         struct nfs4_lockowner *lo, *ret;
5790
5791         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5792         if (!lo)
5793                 return NULL;
5794         INIT_LIST_HEAD(&lo->lo_blocked);
5795         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5796         lo->lo_owner.so_is_open_owner = 0;
5797         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5798         lo->lo_owner.so_ops = &lockowner_ops;
5799         spin_lock(&clp->cl_lock);
5800         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5801         if (ret == NULL) {
5802                 list_add(&lo->lo_owner.so_strhash,
5803                          &clp->cl_ownerstr_hashtbl[strhashval]);
5804                 ret = lo;
5805         } else
5806                 nfs4_free_stateowner(&lo->lo_owner);
5807
5808         spin_unlock(&clp->cl_lock);
5809         return ret;
5810 }
5811
5812 static struct nfs4_ol_stateid *
5813 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5814 {
5815         struct nfs4_ol_stateid *lst;
5816         struct nfs4_client *clp = lo->lo_owner.so_client;
5817
5818         lockdep_assert_held(&clp->cl_lock);
5819
5820         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5821                 if (lst->st_stid.sc_type != NFS4_LOCK_STID)
5822                         continue;
5823                 if (lst->st_stid.sc_file == fp) {
5824                         refcount_inc(&lst->st_stid.sc_count);
5825                         return lst;
5826                 }
5827         }
5828         return NULL;
5829 }
5830
5831 static struct nfs4_ol_stateid *
5832 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5833                   struct nfs4_file *fp, struct inode *inode,
5834                   struct nfs4_ol_stateid *open_stp)
5835 {
5836         struct nfs4_client *clp = lo->lo_owner.so_client;
5837         struct nfs4_ol_stateid *retstp;
5838
5839         mutex_init(&stp->st_mutex);
5840         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
5841 retry:
5842         spin_lock(&clp->cl_lock);
5843         spin_lock(&fp->fi_lock);
5844         retstp = find_lock_stateid(lo, fp);
5845         if (retstp)
5846                 goto out_unlock;
5847
5848         refcount_inc(&stp->st_stid.sc_count);
5849         stp->st_stid.sc_type = NFS4_LOCK_STID;
5850         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5851         get_nfs4_file(fp);
5852         stp->st_stid.sc_file = fp;
5853         stp->st_access_bmap = 0;
5854         stp->st_deny_bmap = open_stp->st_deny_bmap;
5855         stp->st_openstp = open_stp;
5856         list_add(&stp->st_locks, &open_stp->st_locks);
5857         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5858         list_add(&stp->st_perfile, &fp->fi_stateids);
5859 out_unlock:
5860         spin_unlock(&fp->fi_lock);
5861         spin_unlock(&clp->cl_lock);
5862         if (retstp) {
5863                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
5864                         nfs4_put_stid(&retstp->st_stid);
5865                         goto retry;
5866                 }
5867                 /* To keep mutex tracking happy */
5868                 mutex_unlock(&stp->st_mutex);
5869                 stp = retstp;
5870         }
5871         return stp;
5872 }
5873
5874 static struct nfs4_ol_stateid *
5875 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5876                             struct inode *inode, struct nfs4_ol_stateid *ost,
5877                             bool *new)
5878 {
5879         struct nfs4_stid *ns = NULL;
5880         struct nfs4_ol_stateid *lst;
5881         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5882         struct nfs4_client *clp = oo->oo_owner.so_client;
5883
5884         *new = false;
5885         spin_lock(&clp->cl_lock);
5886         lst = find_lock_stateid(lo, fi);
5887         spin_unlock(&clp->cl_lock);
5888         if (lst != NULL) {
5889                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
5890                         goto out;
5891                 nfs4_put_stid(&lst->st_stid);
5892         }
5893         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
5894         if (ns == NULL)
5895                 return NULL;
5896
5897         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
5898         if (lst == openlockstateid(ns))
5899                 *new = true;
5900         else
5901                 nfs4_put_stid(ns);
5902 out:
5903         return lst;
5904 }
5905
5906 static int
5907 check_lock_length(u64 offset, u64 length)
5908 {
5909         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5910                 (length > ~offset)));
5911 }
5912
5913 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5914 {
5915         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5916
5917         lockdep_assert_held(&fp->fi_lock);
5918
5919         if (test_access(access, lock_stp))
5920                 return;
5921         __nfs4_file_get_access(fp, access);
5922         set_access(access, lock_stp);
5923 }
5924
5925 static __be32
5926 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5927                             struct nfs4_ol_stateid *ost,
5928                             struct nfsd4_lock *lock,
5929                             struct nfs4_ol_stateid **plst, bool *new)
5930 {
5931         __be32 status;
5932         struct nfs4_file *fi = ost->st_stid.sc_file;
5933         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5934         struct nfs4_client *cl = oo->oo_owner.so_client;
5935         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5936         struct nfs4_lockowner *lo;
5937         struct nfs4_ol_stateid *lst;
5938         unsigned int strhashval;
5939
5940         lo = find_lockowner_str(cl, &lock->lk_new_owner);
5941         if (!lo) {
5942                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5943                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5944                 if (lo == NULL)
5945                         return nfserr_jukebox;
5946         } else {
5947                 /* with an existing lockowner, seqids must be the same */
5948                 status = nfserr_bad_seqid;
5949                 if (!cstate->minorversion &&
5950                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5951                         goto out;
5952         }
5953
5954         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5955         if (lst == NULL) {
5956                 status = nfserr_jukebox;
5957                 goto out;
5958         }
5959
5960         status = nfs_ok;
5961         *plst = lst;
5962 out:
5963         nfs4_put_stateowner(&lo->lo_owner);
5964         return status;
5965 }
5966
5967 /*
5968  *  LOCK operation
5969  */
5970 __be32
5971 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5972            union nfsd4_op_u *u)
5973 {
5974         struct nfsd4_lock *lock = &u->lock;
5975         struct nfs4_openowner *open_sop = NULL;
5976         struct nfs4_lockowner *lock_sop = NULL;
5977         struct nfs4_ol_stateid *lock_stp = NULL;
5978         struct nfs4_ol_stateid *open_stp = NULL;
5979         struct nfs4_file *fp;
5980         struct file *filp = NULL;
5981         struct nfsd4_blocked_lock *nbl = NULL;
5982         struct file_lock *file_lock = NULL;
5983         struct file_lock *conflock = NULL;
5984         __be32 status = 0;
5985         int lkflg;
5986         int err;
5987         bool new = false;
5988         unsigned char fl_type;
5989         unsigned int fl_flags = FL_POSIX;
5990         struct net *net = SVC_NET(rqstp);
5991         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5992
5993         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5994                 (long long) lock->lk_offset,
5995                 (long long) lock->lk_length);
5996
5997         if (check_lock_length(lock->lk_offset, lock->lk_length))
5998                  return nfserr_inval;
5999
6000         if ((status = fh_verify(rqstp, &cstate->current_fh,
6001                                 S_IFREG, NFSD_MAY_LOCK))) {
6002                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
6003                 return status;
6004         }
6005
6006         if (lock->lk_is_new) {
6007                 if (nfsd4_has_session(cstate))
6008                         /* See rfc 5661 18.10.3: given clientid is ignored: */
6009                         memcpy(&lock->lk_new_clientid,
6010                                 &cstate->session->se_client->cl_clientid,
6011                                 sizeof(clientid_t));
6012
6013                 status = nfserr_stale_clientid;
6014                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
6015                         goto out;
6016
6017                 /* validate and update open stateid and open seqid */
6018                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
6019                                         lock->lk_new_open_seqid,
6020                                         &lock->lk_new_open_stateid,
6021                                         &open_stp, nn);
6022                 if (status)
6023                         goto out;
6024                 mutex_unlock(&open_stp->st_mutex);
6025                 open_sop = openowner(open_stp->st_stateowner);
6026                 status = nfserr_bad_stateid;
6027                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
6028                                                 &lock->lk_new_clientid))
6029                         goto out;
6030                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
6031                                                         &lock_stp, &new);
6032         } else {
6033                 status = nfs4_preprocess_seqid_op(cstate,
6034                                        lock->lk_old_lock_seqid,
6035                                        &lock->lk_old_lock_stateid,
6036                                        NFS4_LOCK_STID, &lock_stp, nn);
6037         }
6038         if (status)
6039                 goto out;
6040         lock_sop = lockowner(lock_stp->st_stateowner);
6041
6042         lkflg = setlkflg(lock->lk_type);
6043         status = nfs4_check_openmode(lock_stp, lkflg);
6044         if (status)
6045                 goto out;
6046
6047         status = nfserr_grace;
6048         if (locks_in_grace(net) && !lock->lk_reclaim)
6049                 goto out;
6050         status = nfserr_no_grace;
6051         if (!locks_in_grace(net) && lock->lk_reclaim)
6052                 goto out;
6053
6054         fp = lock_stp->st_stid.sc_file;
6055         switch (lock->lk_type) {
6056                 case NFS4_READW_LT:
6057                         if (nfsd4_has_session(cstate))
6058                                 fl_flags |= FL_SLEEP;
6059                         /* Fallthrough */
6060                 case NFS4_READ_LT:
6061                         spin_lock(&fp->fi_lock);
6062                         filp = find_readable_file_locked(fp);
6063                         if (filp)
6064                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
6065                         spin_unlock(&fp->fi_lock);
6066                         fl_type = F_RDLCK;
6067                         break;
6068                 case NFS4_WRITEW_LT:
6069                         if (nfsd4_has_session(cstate))
6070                                 fl_flags |= FL_SLEEP;
6071                         /* Fallthrough */
6072                 case NFS4_WRITE_LT:
6073                         spin_lock(&fp->fi_lock);
6074                         filp = find_writeable_file_locked(fp);
6075                         if (filp)
6076                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6077                         spin_unlock(&fp->fi_lock);
6078                         fl_type = F_WRLCK;
6079                         break;
6080                 default:
6081                         status = nfserr_inval;
6082                 goto out;
6083         }
6084
6085         if (!filp) {
6086                 status = nfserr_openmode;
6087                 goto out;
6088         }
6089
6090         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6091         if (!nbl) {
6092                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6093                 status = nfserr_jukebox;
6094                 goto out;
6095         }
6096
6097         file_lock = &nbl->nbl_lock;
6098         file_lock->fl_type = fl_type;
6099         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6100         file_lock->fl_pid = current->tgid;
6101         file_lock->fl_file = filp;
6102         file_lock->fl_flags = fl_flags;
6103         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6104         file_lock->fl_start = lock->lk_offset;
6105         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6106         nfs4_transform_lock_offset(file_lock);
6107
6108         conflock = locks_alloc_lock();
6109         if (!conflock) {
6110                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6111                 status = nfserr_jukebox;
6112                 goto out;
6113         }
6114
6115         if (fl_flags & FL_SLEEP) {
6116                 nbl->nbl_time = jiffies;
6117                 spin_lock(&nn->blocked_locks_lock);
6118                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6119                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6120                 spin_unlock(&nn->blocked_locks_lock);
6121         }
6122
6123         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
6124         switch (err) {
6125         case 0: /* success! */
6126                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6127                 status = 0;
6128                 if (lock->lk_reclaim)
6129                         nn->somebody_reclaimed = true;
6130                 break;
6131         case FILE_LOCK_DEFERRED:
6132                 nbl = NULL;
6133                 /* Fallthrough */
6134         case -EAGAIN:           /* conflock holds conflicting lock */
6135                 status = nfserr_denied;
6136                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6137                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6138                 break;
6139         case -EDEADLK:
6140                 status = nfserr_deadlock;
6141                 break;
6142         default:
6143                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6144                 status = nfserrno(err);
6145                 break;
6146         }
6147 out:
6148         if (nbl) {
6149                 /* dequeue it if we queued it before */
6150                 if (fl_flags & FL_SLEEP) {
6151                         spin_lock(&nn->blocked_locks_lock);
6152                         list_del_init(&nbl->nbl_list);
6153                         list_del_init(&nbl->nbl_lru);
6154                         spin_unlock(&nn->blocked_locks_lock);
6155                 }
6156                 free_blocked_lock(nbl);
6157         }
6158         if (filp)
6159                 fput(filp);
6160         if (lock_stp) {
6161                 /* Bump seqid manually if the 4.0 replay owner is openowner */
6162                 if (cstate->replay_owner &&
6163                     cstate->replay_owner != &lock_sop->lo_owner &&
6164                     seqid_mutating_err(ntohl(status)))
6165                         lock_sop->lo_owner.so_seqid++;
6166
6167                 /*
6168                  * If this is a new, never-before-used stateid, and we are
6169                  * returning an error, then just go ahead and release it.
6170                  */
6171                 if (status && new)
6172                         release_lock_stateid(lock_stp);
6173
6174                 mutex_unlock(&lock_stp->st_mutex);
6175
6176                 nfs4_put_stid(&lock_stp->st_stid);
6177         }
6178         if (open_stp)
6179                 nfs4_put_stid(&open_stp->st_stid);
6180         nfsd4_bump_seqid(cstate, status);
6181         if (conflock)
6182                 locks_free_lock(conflock);
6183         return status;
6184 }
6185
6186 /*
6187  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6188  * so we do a temporary open here just to get an open file to pass to
6189  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
6190  * inode operation.)
6191  */
6192 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6193 {
6194         struct file *file;
6195         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
6196         if (!err) {
6197                 err = nfserrno(vfs_test_lock(file, lock));
6198                 fput(file);
6199         }
6200         return err;
6201 }
6202
6203 /*
6204  * LOCKT operation
6205  */
6206 __be32
6207 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6208             union nfsd4_op_u *u)
6209 {
6210         struct nfsd4_lockt *lockt = &u->lockt;
6211         struct file_lock *file_lock = NULL;
6212         struct nfs4_lockowner *lo = NULL;
6213         __be32 status;
6214         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6215
6216         if (locks_in_grace(SVC_NET(rqstp)))
6217                 return nfserr_grace;
6218
6219         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6220                  return nfserr_inval;
6221
6222         if (!nfsd4_has_session(cstate)) {
6223                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
6224                 if (status)
6225                         goto out;
6226         }
6227
6228         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6229                 goto out;
6230
6231         file_lock = locks_alloc_lock();
6232         if (!file_lock) {
6233                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6234                 status = nfserr_jukebox;
6235                 goto out;
6236         }
6237
6238         switch (lockt->lt_type) {
6239                 case NFS4_READ_LT:
6240                 case NFS4_READW_LT:
6241                         file_lock->fl_type = F_RDLCK;
6242                         break;
6243                 case NFS4_WRITE_LT:
6244                 case NFS4_WRITEW_LT:
6245                         file_lock->fl_type = F_WRLCK;
6246                         break;
6247                 default:
6248                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6249                         status = nfserr_inval;
6250                         goto out;
6251         }
6252
6253         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6254         if (lo)
6255                 file_lock->fl_owner = (fl_owner_t)lo;
6256         file_lock->fl_pid = current->tgid;
6257         file_lock->fl_flags = FL_POSIX;
6258
6259         file_lock->fl_start = lockt->lt_offset;
6260         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6261
6262         nfs4_transform_lock_offset(file_lock);
6263
6264         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6265         if (status)
6266                 goto out;
6267
6268         if (file_lock->fl_type != F_UNLCK) {
6269                 status = nfserr_denied;
6270                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6271         }
6272 out:
6273         if (lo)
6274                 nfs4_put_stateowner(&lo->lo_owner);
6275         if (file_lock)
6276                 locks_free_lock(file_lock);
6277         return status;
6278 }
6279
6280 __be32
6281 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6282             union nfsd4_op_u *u)
6283 {
6284         struct nfsd4_locku *locku = &u->locku;
6285         struct nfs4_ol_stateid *stp;
6286         struct file *filp = NULL;
6287         struct file_lock *file_lock = NULL;
6288         __be32 status;
6289         int err;
6290         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6291
6292         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6293                 (long long) locku->lu_offset,
6294                 (long long) locku->lu_length);
6295
6296         if (check_lock_length(locku->lu_offset, locku->lu_length))
6297                  return nfserr_inval;
6298
6299         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6300                                         &locku->lu_stateid, NFS4_LOCK_STID,
6301                                         &stp, nn);
6302         if (status)
6303                 goto out;
6304         filp = find_any_file(stp->st_stid.sc_file);
6305         if (!filp) {
6306                 status = nfserr_lock_range;
6307                 goto put_stateid;
6308         }
6309         file_lock = locks_alloc_lock();
6310         if (!file_lock) {
6311                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6312                 status = nfserr_jukebox;
6313                 goto fput;
6314         }
6315
6316         file_lock->fl_type = F_UNLCK;
6317         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6318         file_lock->fl_pid = current->tgid;
6319         file_lock->fl_file = filp;
6320         file_lock->fl_flags = FL_POSIX;
6321         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6322         file_lock->fl_start = locku->lu_offset;
6323
6324         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6325                                                 locku->lu_length);
6326         nfs4_transform_lock_offset(file_lock);
6327
6328         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6329         if (err) {
6330                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6331                 goto out_nfserr;
6332         }
6333         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6334 fput:
6335         fput(filp);
6336 put_stateid:
6337         mutex_unlock(&stp->st_mutex);
6338         nfs4_put_stid(&stp->st_stid);
6339 out:
6340         nfsd4_bump_seqid(cstate, status);
6341         if (file_lock)
6342                 locks_free_lock(file_lock);
6343         return status;
6344
6345 out_nfserr:
6346         status = nfserrno(err);
6347         goto fput;
6348 }
6349
6350 /*
6351  * returns
6352  *      true:  locks held by lockowner
6353  *      false: no locks held by lockowner
6354  */
6355 static bool
6356 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6357 {
6358         struct file_lock *fl;
6359         int status = false;
6360         struct file *filp = find_any_file(fp);
6361         struct inode *inode;
6362         struct file_lock_context *flctx;
6363
6364         if (!filp) {
6365                 /* Any valid lock stateid should have some sort of access */
6366                 WARN_ON_ONCE(1);
6367                 return status;
6368         }
6369
6370         inode = locks_inode(filp);
6371         flctx = inode->i_flctx;
6372
6373         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6374                 spin_lock(&flctx->flc_lock);
6375                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6376                         if (fl->fl_owner == (fl_owner_t)lowner) {
6377                                 status = true;
6378                                 break;
6379                         }
6380                 }
6381                 spin_unlock(&flctx->flc_lock);
6382         }
6383         fput(filp);
6384         return status;
6385 }
6386
6387 __be32
6388 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6389                         struct nfsd4_compound_state *cstate,
6390                         union nfsd4_op_u *u)
6391 {
6392         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6393         clientid_t *clid = &rlockowner->rl_clientid;
6394         struct nfs4_stateowner *sop;
6395         struct nfs4_lockowner *lo = NULL;
6396         struct nfs4_ol_stateid *stp;
6397         struct xdr_netobj *owner = &rlockowner->rl_owner;
6398         unsigned int hashval = ownerstr_hashval(owner);
6399         __be32 status;
6400         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6401         struct nfs4_client *clp;
6402         LIST_HEAD (reaplist);
6403
6404         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6405                 clid->cl_boot, clid->cl_id);
6406
6407         status = lookup_clientid(clid, cstate, nn);
6408         if (status)
6409                 return status;
6410
6411         clp = cstate->clp;
6412         /* Find the matching lock stateowner */
6413         spin_lock(&clp->cl_lock);
6414         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6415                             so_strhash) {
6416
6417                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6418                         continue;
6419
6420                 /* see if there are still any locks associated with it */
6421                 lo = lockowner(sop);
6422                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6423                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
6424                                 status = nfserr_locks_held;
6425                                 spin_unlock(&clp->cl_lock);
6426                                 return status;
6427                         }
6428                 }
6429
6430                 nfs4_get_stateowner(sop);
6431                 break;
6432         }
6433         if (!lo) {
6434                 spin_unlock(&clp->cl_lock);
6435                 return status;
6436         }
6437
6438         unhash_lockowner_locked(lo);
6439         while (!list_empty(&lo->lo_owner.so_stateids)) {
6440                 stp = list_first_entry(&lo->lo_owner.so_stateids,
6441                                        struct nfs4_ol_stateid,
6442                                        st_perstateowner);
6443                 WARN_ON(!unhash_lock_stateid(stp));
6444                 put_ol_stateid_locked(stp, &reaplist);
6445         }
6446         spin_unlock(&clp->cl_lock);
6447         free_ol_stateid_reaplist(&reaplist);
6448         remove_blocked_locks(lo);
6449         nfs4_put_stateowner(&lo->lo_owner);
6450
6451         return status;
6452 }
6453
6454 static inline struct nfs4_client_reclaim *
6455 alloc_reclaim(void)
6456 {
6457         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6458 }
6459
6460 bool
6461 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
6462 {
6463         struct nfs4_client_reclaim *crp;
6464
6465         crp = nfsd4_find_reclaim_client(name, nn);
6466         return (crp && crp->cr_clp);
6467 }
6468
6469 /*
6470  * failure => all reset bets are off, nfserr_no_grace...
6471  */
6472 struct nfs4_client_reclaim *
6473 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6474 {
6475         unsigned int strhashval;
6476         struct nfs4_client_reclaim *crp;
6477
6478         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6479         crp = alloc_reclaim();
6480         if (crp) {
6481                 strhashval = clientstr_hashval(name);
6482                 INIT_LIST_HEAD(&crp->cr_strhash);
6483                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6484                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6485                 crp->cr_clp = NULL;
6486                 nn->reclaim_str_hashtbl_size++;
6487         }
6488         return crp;
6489 }
6490
6491 void
6492 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6493 {
6494         list_del(&crp->cr_strhash);
6495         kfree(crp);
6496         nn->reclaim_str_hashtbl_size--;
6497 }
6498
6499 void
6500 nfs4_release_reclaim(struct nfsd_net *nn)
6501 {
6502         struct nfs4_client_reclaim *crp = NULL;
6503         int i;
6504
6505         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6506                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6507                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6508                                         struct nfs4_client_reclaim, cr_strhash);
6509                         nfs4_remove_reclaim_record(crp, nn);
6510                 }
6511         }
6512         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6513 }
6514
6515 /*
6516  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6517 struct nfs4_client_reclaim *
6518 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6519 {
6520         unsigned int strhashval;
6521         struct nfs4_client_reclaim *crp = NULL;
6522
6523         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6524
6525         strhashval = clientstr_hashval(recdir);
6526         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6527                 if (same_name(crp->cr_recdir, recdir)) {
6528                         return crp;
6529                 }
6530         }
6531         return NULL;
6532 }
6533
6534 /*
6535 * Called from OPEN. Look for clientid in reclaim list.
6536 */
6537 __be32
6538 nfs4_check_open_reclaim(clientid_t *clid,
6539                 struct nfsd4_compound_state *cstate,
6540                 struct nfsd_net *nn)
6541 {
6542         __be32 status;
6543
6544         /* find clientid in conf_id_hashtbl */
6545         status = lookup_clientid(clid, cstate, nn);
6546         if (status)
6547                 return nfserr_reclaim_bad;
6548
6549         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6550                 return nfserr_no_grace;
6551
6552         if (nfsd4_client_record_check(cstate->clp))
6553                 return nfserr_reclaim_bad;
6554
6555         return nfs_ok;
6556 }
6557
6558 #ifdef CONFIG_NFSD_FAULT_INJECTION
6559 static inline void
6560 put_client(struct nfs4_client *clp)
6561 {
6562         atomic_dec(&clp->cl_refcount);
6563 }
6564
6565 static struct nfs4_client *
6566 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6567 {
6568         struct nfs4_client *clp;
6569         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6570                                           nfsd_net_id);
6571
6572         if (!nfsd_netns_ready(nn))
6573                 return NULL;
6574
6575         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6576                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6577                         return clp;
6578         }
6579         return NULL;
6580 }
6581
6582 u64
6583 nfsd_inject_print_clients(void)
6584 {
6585         struct nfs4_client *clp;
6586         u64 count = 0;
6587         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6588                                           nfsd_net_id);
6589         char buf[INET6_ADDRSTRLEN];
6590
6591         if (!nfsd_netns_ready(nn))
6592                 return 0;
6593
6594         spin_lock(&nn->client_lock);
6595         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6596                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6597                 pr_info("NFS Client: %s\n", buf);
6598                 ++count;
6599         }
6600         spin_unlock(&nn->client_lock);
6601
6602         return count;
6603 }
6604
6605 u64
6606 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6607 {
6608         u64 count = 0;
6609         struct nfs4_client *clp;
6610         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6611                                           nfsd_net_id);
6612
6613         if (!nfsd_netns_ready(nn))
6614                 return count;
6615
6616         spin_lock(&nn->client_lock);
6617         clp = nfsd_find_client(addr, addr_size);
6618         if (clp) {
6619                 if (mark_client_expired_locked(clp) == nfs_ok)
6620                         ++count;
6621                 else
6622                         clp = NULL;
6623         }
6624         spin_unlock(&nn->client_lock);
6625
6626         if (clp)
6627                 expire_client(clp);
6628
6629         return count;
6630 }
6631
6632 u64
6633 nfsd_inject_forget_clients(u64 max)
6634 {
6635         u64 count = 0;
6636         struct nfs4_client *clp, *next;
6637         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6638                                                 nfsd_net_id);
6639         LIST_HEAD(reaplist);
6640
6641         if (!nfsd_netns_ready(nn))
6642                 return count;
6643
6644         spin_lock(&nn->client_lock);
6645         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6646                 if (mark_client_expired_locked(clp) == nfs_ok) {
6647                         list_add(&clp->cl_lru, &reaplist);
6648                         if (max != 0 && ++count >= max)
6649                                 break;
6650                 }
6651         }
6652         spin_unlock(&nn->client_lock);
6653
6654         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6655                 expire_client(clp);
6656
6657         return count;
6658 }
6659
6660 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6661                              const char *type)
6662 {
6663         char buf[INET6_ADDRSTRLEN];
6664         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6665         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6666 }
6667
6668 static void
6669 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6670                              struct list_head *collect)
6671 {
6672         struct nfs4_client *clp = lst->st_stid.sc_client;
6673         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6674                                           nfsd_net_id);
6675
6676         if (!collect)
6677                 return;
6678
6679         lockdep_assert_held(&nn->client_lock);
6680         atomic_inc(&clp->cl_refcount);
6681         list_add(&lst->st_locks, collect);
6682 }
6683
6684 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6685                                     struct list_head *collect,
6686                                     bool (*func)(struct nfs4_ol_stateid *))
6687 {
6688         struct nfs4_openowner *oop;
6689         struct nfs4_ol_stateid *stp, *st_next;
6690         struct nfs4_ol_stateid *lst, *lst_next;
6691         u64 count = 0;
6692
6693         spin_lock(&clp->cl_lock);
6694         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6695                 list_for_each_entry_safe(stp, st_next,
6696                                 &oop->oo_owner.so_stateids, st_perstateowner) {
6697                         list_for_each_entry_safe(lst, lst_next,
6698                                         &stp->st_locks, st_locks) {
6699                                 if (func) {
6700                                         if (func(lst))
6701                                                 nfsd_inject_add_lock_to_list(lst,
6702                                                                         collect);
6703                                 }
6704                                 ++count;
6705                                 /*
6706                                  * Despite the fact that these functions deal
6707                                  * with 64-bit integers for "count", we must
6708                                  * ensure that it doesn't blow up the
6709                                  * clp->cl_refcount. Throw a warning if we
6710                                  * start to approach INT_MAX here.
6711                                  */
6712                                 WARN_ON_ONCE(count == (INT_MAX / 2));
6713                                 if (count == max)
6714                                         goto out;
6715                         }
6716                 }
6717         }
6718 out:
6719         spin_unlock(&clp->cl_lock);
6720
6721         return count;
6722 }
6723
6724 static u64
6725 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6726                           u64 max)
6727 {
6728         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6729 }
6730
6731 static u64
6732 nfsd_print_client_locks(struct nfs4_client *clp)
6733 {
6734         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6735         nfsd_print_count(clp, count, "locked files");
6736         return count;
6737 }
6738
6739 u64
6740 nfsd_inject_print_locks(void)
6741 {
6742         struct nfs4_client *clp;
6743         u64 count = 0;
6744         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6745                                                 nfsd_net_id);
6746
6747         if (!nfsd_netns_ready(nn))
6748                 return 0;
6749
6750         spin_lock(&nn->client_lock);
6751         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6752                 count += nfsd_print_client_locks(clp);
6753         spin_unlock(&nn->client_lock);
6754
6755         return count;
6756 }
6757
6758 static void
6759 nfsd_reap_locks(struct list_head *reaplist)
6760 {
6761         struct nfs4_client *clp;
6762         struct nfs4_ol_stateid *stp, *next;
6763
6764         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6765                 list_del_init(&stp->st_locks);
6766                 clp = stp->st_stid.sc_client;
6767                 nfs4_put_stid(&stp->st_stid);
6768                 put_client(clp);
6769         }
6770 }
6771
6772 u64
6773 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6774 {
6775         unsigned int count = 0;
6776         struct nfs4_client *clp;
6777         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6778                                                 nfsd_net_id);
6779         LIST_HEAD(reaplist);
6780
6781         if (!nfsd_netns_ready(nn))
6782                 return count;
6783
6784         spin_lock(&nn->client_lock);
6785         clp = nfsd_find_client(addr, addr_size);
6786         if (clp)
6787                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6788         spin_unlock(&nn->client_lock);
6789         nfsd_reap_locks(&reaplist);
6790         return count;
6791 }
6792
6793 u64
6794 nfsd_inject_forget_locks(u64 max)
6795 {
6796         u64 count = 0;
6797         struct nfs4_client *clp;
6798         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6799                                                 nfsd_net_id);
6800         LIST_HEAD(reaplist);
6801
6802         if (!nfsd_netns_ready(nn))
6803                 return count;
6804
6805         spin_lock(&nn->client_lock);
6806         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6807                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6808                 if (max != 0 && count >= max)
6809                         break;
6810         }
6811         spin_unlock(&nn->client_lock);
6812         nfsd_reap_locks(&reaplist);
6813         return count;
6814 }
6815
6816 static u64
6817 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6818                               struct list_head *collect,
6819                               void (*func)(struct nfs4_openowner *))
6820 {
6821         struct nfs4_openowner *oop, *next;
6822         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6823                                                 nfsd_net_id);
6824         u64 count = 0;
6825
6826         lockdep_assert_held(&nn->client_lock);
6827
6828         spin_lock(&clp->cl_lock);
6829         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6830                 if (func) {
6831                         func(oop);
6832                         if (collect) {
6833                                 atomic_inc(&clp->cl_refcount);
6834                                 list_add(&oop->oo_perclient, collect);
6835                         }
6836                 }
6837                 ++count;
6838                 /*
6839                  * Despite the fact that these functions deal with
6840                  * 64-bit integers for "count", we must ensure that
6841                  * it doesn't blow up the clp->cl_refcount. Throw a
6842                  * warning if we start to approach INT_MAX here.
6843                  */
6844                 WARN_ON_ONCE(count == (INT_MAX / 2));
6845                 if (count == max)
6846                         break;
6847         }
6848         spin_unlock(&clp->cl_lock);
6849
6850         return count;
6851 }
6852
6853 static u64
6854 nfsd_print_client_openowners(struct nfs4_client *clp)
6855 {
6856         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6857
6858         nfsd_print_count(clp, count, "openowners");
6859         return count;
6860 }
6861
6862 static u64
6863 nfsd_collect_client_openowners(struct nfs4_client *clp,
6864                                struct list_head *collect, u64 max)
6865 {
6866         return nfsd_foreach_client_openowner(clp, max, collect,
6867                                                 unhash_openowner_locked);
6868 }
6869
6870 u64
6871 nfsd_inject_print_openowners(void)
6872 {
6873         struct nfs4_client *clp;
6874         u64 count = 0;
6875         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6876                                                 nfsd_net_id);
6877
6878         if (!nfsd_netns_ready(nn))
6879                 return 0;
6880
6881         spin_lock(&nn->client_lock);
6882         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6883                 count += nfsd_print_client_openowners(clp);
6884         spin_unlock(&nn->client_lock);
6885
6886         return count;
6887 }
6888
6889 static void
6890 nfsd_reap_openowners(struct list_head *reaplist)
6891 {
6892         struct nfs4_client *clp;
6893         struct nfs4_openowner *oop, *next;
6894
6895         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6896                 list_del_init(&oop->oo_perclient);
6897                 clp = oop->oo_owner.so_client;
6898                 release_openowner(oop);
6899                 put_client(clp);
6900         }
6901 }
6902
6903 u64
6904 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6905                                      size_t addr_size)
6906 {
6907         unsigned int count = 0;
6908         struct nfs4_client *clp;
6909         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6910                                                 nfsd_net_id);
6911         LIST_HEAD(reaplist);
6912
6913         if (!nfsd_netns_ready(nn))
6914                 return count;
6915
6916         spin_lock(&nn->client_lock);
6917         clp = nfsd_find_client(addr, addr_size);
6918         if (clp)
6919                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6920         spin_unlock(&nn->client_lock);
6921         nfsd_reap_openowners(&reaplist);
6922         return count;
6923 }
6924
6925 u64
6926 nfsd_inject_forget_openowners(u64 max)
6927 {
6928         u64 count = 0;
6929         struct nfs4_client *clp;
6930         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6931                                                 nfsd_net_id);
6932         LIST_HEAD(reaplist);
6933
6934         if (!nfsd_netns_ready(nn))
6935                 return count;
6936
6937         spin_lock(&nn->client_lock);
6938         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6939                 count += nfsd_collect_client_openowners(clp, &reaplist,
6940                                                         max - count);
6941                 if (max != 0 && count >= max)
6942                         break;
6943         }
6944         spin_unlock(&nn->client_lock);
6945         nfsd_reap_openowners(&reaplist);
6946         return count;
6947 }
6948
6949 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6950                                      struct list_head *victims)
6951 {
6952         struct nfs4_delegation *dp, *next;
6953         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6954                                                 nfsd_net_id);
6955         u64 count = 0;
6956
6957         lockdep_assert_held(&nn->client_lock);
6958
6959         spin_lock(&state_lock);
6960         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6961                 if (victims) {
6962                         /*
6963                          * It's not safe to mess with delegations that have a
6964                          * non-zero dl_time. They might have already been broken
6965                          * and could be processed by the laundromat outside of
6966                          * the state_lock. Just leave them be.
6967                          */
6968                         if (dp->dl_time != 0)
6969                                 continue;
6970
6971                         atomic_inc(&clp->cl_refcount);
6972                         WARN_ON(!unhash_delegation_locked(dp));
6973                         list_add(&dp->dl_recall_lru, victims);
6974                 }
6975                 ++count;
6976                 /*
6977                  * Despite the fact that these functions deal with
6978                  * 64-bit integers for "count", we must ensure that
6979                  * it doesn't blow up the clp->cl_refcount. Throw a
6980                  * warning if we start to approach INT_MAX here.
6981                  */
6982                 WARN_ON_ONCE(count == (INT_MAX / 2));
6983                 if (count == max)
6984                         break;
6985         }
6986         spin_unlock(&state_lock);
6987         return count;
6988 }
6989
6990 static u64
6991 nfsd_print_client_delegations(struct nfs4_client *clp)
6992 {
6993         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6994
6995         nfsd_print_count(clp, count, "delegations");
6996         return count;
6997 }
6998
6999 u64
7000 nfsd_inject_print_delegations(void)
7001 {
7002         struct nfs4_client *clp;
7003         u64 count = 0;
7004         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7005                                                 nfsd_net_id);
7006
7007         if (!nfsd_netns_ready(nn))
7008                 return 0;
7009
7010         spin_lock(&nn->client_lock);
7011         list_for_each_entry(clp, &nn->client_lru, cl_lru)
7012                 count += nfsd_print_client_delegations(clp);
7013         spin_unlock(&nn->client_lock);
7014
7015         return count;
7016 }
7017
7018 static void
7019 nfsd_forget_delegations(struct list_head *reaplist)
7020 {
7021         struct nfs4_client *clp;
7022         struct nfs4_delegation *dp, *next;
7023
7024         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7025                 list_del_init(&dp->dl_recall_lru);
7026                 clp = dp->dl_stid.sc_client;
7027                 revoke_delegation(dp);
7028                 put_client(clp);
7029         }
7030 }
7031
7032 u64
7033 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
7034                                       size_t addr_size)
7035 {
7036         u64 count = 0;
7037         struct nfs4_client *clp;
7038         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7039                                                 nfsd_net_id);
7040         LIST_HEAD(reaplist);
7041
7042         if (!nfsd_netns_ready(nn))
7043                 return count;
7044
7045         spin_lock(&nn->client_lock);
7046         clp = nfsd_find_client(addr, addr_size);
7047         if (clp)
7048                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7049         spin_unlock(&nn->client_lock);
7050
7051         nfsd_forget_delegations(&reaplist);
7052         return count;
7053 }
7054
7055 u64
7056 nfsd_inject_forget_delegations(u64 max)
7057 {
7058         u64 count = 0;
7059         struct nfs4_client *clp;
7060         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7061                                                 nfsd_net_id);
7062         LIST_HEAD(reaplist);
7063
7064         if (!nfsd_netns_ready(nn))
7065                 return count;
7066
7067         spin_lock(&nn->client_lock);
7068         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7069                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7070                 if (max != 0 && count >= max)
7071                         break;
7072         }
7073         spin_unlock(&nn->client_lock);
7074         nfsd_forget_delegations(&reaplist);
7075         return count;
7076 }
7077
7078 static void
7079 nfsd_recall_delegations(struct list_head *reaplist)
7080 {
7081         struct nfs4_client *clp;
7082         struct nfs4_delegation *dp, *next;
7083
7084         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7085                 list_del_init(&dp->dl_recall_lru);
7086                 clp = dp->dl_stid.sc_client;
7087                 /*
7088                  * We skipped all entries that had a zero dl_time before,
7089                  * so we can now reset the dl_time back to 0. If a delegation
7090                  * break comes in now, then it won't make any difference since
7091                  * we're recalling it either way.
7092                  */
7093                 spin_lock(&state_lock);
7094                 dp->dl_time = 0;
7095                 spin_unlock(&state_lock);
7096                 nfsd_break_one_deleg(dp);
7097                 put_client(clp);
7098         }
7099 }
7100
7101 u64
7102 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7103                                       size_t addr_size)
7104 {
7105         u64 count = 0;
7106         struct nfs4_client *clp;
7107         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7108                                                 nfsd_net_id);
7109         LIST_HEAD(reaplist);
7110
7111         if (!nfsd_netns_ready(nn))
7112                 return count;
7113
7114         spin_lock(&nn->client_lock);
7115         clp = nfsd_find_client(addr, addr_size);
7116         if (clp)
7117                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7118         spin_unlock(&nn->client_lock);
7119
7120         nfsd_recall_delegations(&reaplist);
7121         return count;
7122 }
7123
7124 u64
7125 nfsd_inject_recall_delegations(u64 max)
7126 {
7127         u64 count = 0;
7128         struct nfs4_client *clp, *next;
7129         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7130                                                 nfsd_net_id);
7131         LIST_HEAD(reaplist);
7132
7133         if (!nfsd_netns_ready(nn))
7134                 return count;
7135
7136         spin_lock(&nn->client_lock);
7137         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7138                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7139                 if (max != 0 && ++count >= max)
7140                         break;
7141         }
7142         spin_unlock(&nn->client_lock);
7143         nfsd_recall_delegations(&reaplist);
7144         return count;
7145 }
7146 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7147
7148 /*
7149  * Since the lifetime of a delegation isn't limited to that of an open, a
7150  * client may quite reasonably hang on to a delegation as long as it has
7151  * the inode cached.  This becomes an obvious problem the first time a
7152  * client's inode cache approaches the size of the server's total memory.
7153  *
7154  * For now we avoid this problem by imposing a hard limit on the number
7155  * of delegations, which varies according to the server's memory size.
7156  */
7157 static void
7158 set_max_delegations(void)
7159 {
7160         /*
7161          * Allow at most 4 delegations per megabyte of RAM.  Quick
7162          * estimates suggest that in the worst case (where every delegation
7163          * is for a different inode), a delegation could take about 1.5K,
7164          * giving a worst case usage of about 6% of memory.
7165          */
7166         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7167 }
7168
7169 static int nfs4_state_create_net(struct net *net)
7170 {
7171         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7172         int i;
7173
7174         nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7175                                             sizeof(struct list_head),
7176                                             GFP_KERNEL);
7177         if (!nn->conf_id_hashtbl)
7178                 goto err;
7179         nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7180                                               sizeof(struct list_head),
7181                                               GFP_KERNEL);
7182         if (!nn->unconf_id_hashtbl)
7183                 goto err_unconf_id;
7184         nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7185                                               sizeof(struct list_head),
7186                                               GFP_KERNEL);
7187         if (!nn->sessionid_hashtbl)
7188                 goto err_sessionid;
7189
7190         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7191                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7192                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7193         }
7194         for (i = 0; i < SESSION_HASH_SIZE; i++)
7195                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7196         nn->conf_name_tree = RB_ROOT;
7197         nn->unconf_name_tree = RB_ROOT;
7198         nn->boot_time = get_seconds();
7199         nn->grace_ended = false;
7200         nn->nfsd4_manager.block_opens = true;
7201         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7202         INIT_LIST_HEAD(&nn->client_lru);
7203         INIT_LIST_HEAD(&nn->close_lru);
7204         INIT_LIST_HEAD(&nn->del_recall_lru);
7205         spin_lock_init(&nn->client_lock);
7206         spin_lock_init(&nn->s2s_cp_lock);
7207         idr_init(&nn->s2s_cp_stateids);
7208
7209         spin_lock_init(&nn->blocked_locks_lock);
7210         INIT_LIST_HEAD(&nn->blocked_locks_lru);
7211
7212         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7213         get_net(net);
7214
7215         return 0;
7216
7217 err_sessionid:
7218         kfree(nn->unconf_id_hashtbl);
7219 err_unconf_id:
7220         kfree(nn->conf_id_hashtbl);
7221 err:
7222         return -ENOMEM;
7223 }
7224
7225 static void
7226 nfs4_state_destroy_net(struct net *net)
7227 {
7228         int i;
7229         struct nfs4_client *clp = NULL;
7230         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7231
7232         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7233                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7234                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7235                         destroy_client(clp);
7236                 }
7237         }
7238
7239         WARN_ON(!list_empty(&nn->blocked_locks_lru));
7240
7241         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7242                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7243                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7244                         destroy_client(clp);
7245                 }
7246         }
7247
7248         kfree(nn->sessionid_hashtbl);
7249         kfree(nn->unconf_id_hashtbl);
7250         kfree(nn->conf_id_hashtbl);
7251         put_net(net);
7252 }
7253
7254 int
7255 nfs4_state_start_net(struct net *net)
7256 {
7257         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7258         int ret;
7259
7260         ret = nfs4_state_create_net(net);
7261         if (ret)
7262                 return ret;
7263         locks_start_grace(net, &nn->nfsd4_manager);
7264         nfsd4_client_tracking_init(net);
7265         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %x)\n",
7266                nn->nfsd4_grace, net->ns.inum);
7267         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7268         return 0;
7269 }
7270
7271 /* initialization to perform when the nfsd service is started: */
7272
7273 int
7274 nfs4_state_start(void)
7275 {
7276         int ret;
7277
7278         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7279         if (laundry_wq == NULL) {
7280                 ret = -ENOMEM;
7281                 goto out;
7282         }
7283         ret = nfsd4_create_callback_queue();
7284         if (ret)
7285                 goto out_free_laundry;
7286
7287         set_max_delegations();
7288         return 0;
7289
7290 out_free_laundry:
7291         destroy_workqueue(laundry_wq);
7292 out:
7293         return ret;
7294 }
7295
7296 void
7297 nfs4_state_shutdown_net(struct net *net)
7298 {
7299         struct nfs4_delegation *dp = NULL;
7300         struct list_head *pos, *next, reaplist;
7301         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7302
7303         cancel_delayed_work_sync(&nn->laundromat_work);
7304         locks_end_grace(&nn->nfsd4_manager);
7305
7306         INIT_LIST_HEAD(&reaplist);
7307         spin_lock(&state_lock);
7308         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7309                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7310                 WARN_ON(!unhash_delegation_locked(dp));
7311                 list_add(&dp->dl_recall_lru, &reaplist);
7312         }
7313         spin_unlock(&state_lock);
7314         list_for_each_safe(pos, next, &reaplist) {
7315                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7316                 list_del_init(&dp->dl_recall_lru);
7317                 destroy_unhashed_deleg(dp);
7318         }
7319
7320         nfsd4_client_tracking_exit(net);
7321         nfs4_state_destroy_net(net);
7322 }
7323
7324 void
7325 nfs4_state_shutdown(void)
7326 {
7327         destroy_workqueue(laundry_wq);
7328         nfsd4_destroy_callback_queue();
7329 }
7330
7331 static void
7332 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7333 {
7334         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7335                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7336 }
7337
7338 static void
7339 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7340 {
7341         if (cstate->minorversion) {
7342                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7343                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7344         }
7345 }
7346
7347 void
7348 clear_current_stateid(struct nfsd4_compound_state *cstate)
7349 {
7350         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7351 }
7352
7353 /*
7354  * functions to set current state id
7355  */
7356 void
7357 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7358                 union nfsd4_op_u *u)
7359 {
7360         put_stateid(cstate, &u->open_downgrade.od_stateid);
7361 }
7362
7363 void
7364 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7365                 union nfsd4_op_u *u)
7366 {
7367         put_stateid(cstate, &u->open.op_stateid);
7368 }
7369
7370 void
7371 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7372                 union nfsd4_op_u *u)
7373 {
7374         put_stateid(cstate, &u->close.cl_stateid);
7375 }
7376
7377 void
7378 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7379                 union nfsd4_op_u *u)
7380 {
7381         put_stateid(cstate, &u->lock.lk_resp_stateid);
7382 }
7383
7384 /*
7385  * functions to consume current state id
7386  */
7387
7388 void
7389 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7390                 union nfsd4_op_u *u)
7391 {
7392         get_stateid(cstate, &u->open_downgrade.od_stateid);
7393 }
7394
7395 void
7396 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7397                 union nfsd4_op_u *u)
7398 {
7399         get_stateid(cstate, &u->delegreturn.dr_stateid);
7400 }
7401
7402 void
7403 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7404                 union nfsd4_op_u *u)
7405 {
7406         get_stateid(cstate, &u->free_stateid.fr_stateid);
7407 }
7408
7409 void
7410 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7411                 union nfsd4_op_u *u)
7412 {
7413         get_stateid(cstate, &u->setattr.sa_stateid);
7414 }
7415
7416 void
7417 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7418                 union nfsd4_op_u *u)
7419 {
7420         get_stateid(cstate, &u->close.cl_stateid);
7421 }
7422
7423 void
7424 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7425                 union nfsd4_op_u *u)
7426 {
7427         get_stateid(cstate, &u->locku.lu_stateid);
7428 }
7429
7430 void
7431 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7432                 union nfsd4_op_u *u)
7433 {
7434         get_stateid(cstate, &u->read.rd_stateid);
7435 }
7436
7437 void
7438 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7439                 union nfsd4_op_u *u)
7440 {
7441         get_stateid(cstate, &u->write.wr_stateid);
7442 }