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         case SP4_SSV:
2589                 status = nfserr_encr_alg_unsupp;
2590                 goto out_nolock;
2591         }
2592
2593         /* Cases below refer to rfc 5661 section 18.35.4: */
2594         spin_lock(&nn->client_lock);
2595         conf = find_confirmed_client_by_name(&exid->clname, nn);
2596         if (conf) {
2597                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2598                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2599
2600                 if (update) {
2601                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2602                                 status = nfserr_inval;
2603                                 goto out;
2604                         }
2605                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
2606                                 status = nfserr_wrong_cred;
2607                                 goto out;
2608                         }
2609                         if (!creds_match) { /* case 9 */
2610                                 status = nfserr_perm;
2611                                 goto out;
2612                         }
2613                         if (!verfs_match) { /* case 8 */
2614                                 status = nfserr_not_same;
2615                                 goto out;
2616                         }
2617                         /* case 6 */
2618                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2619                         goto out_copy;
2620                 }
2621                 if (!creds_match) { /* case 3 */
2622                         if (client_has_state(conf)) {
2623                                 status = nfserr_clid_inuse;
2624                                 goto out;
2625                         }
2626                         goto out_new;
2627                 }
2628                 if (verfs_match) { /* case 2 */
2629                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2630                         goto out_copy;
2631                 }
2632                 /* case 5, client reboot */
2633                 conf = NULL;
2634                 goto out_new;
2635         }
2636
2637         if (update) { /* case 7 */
2638                 status = nfserr_noent;
2639                 goto out;
2640         }
2641
2642         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2643         if (unconf) /* case 4, possible retry or client restart */
2644                 unhash_client_locked(unconf);
2645
2646         /* case 1 (normal case) */
2647 out_new:
2648         if (conf) {
2649                 status = mark_client_expired_locked(conf);
2650                 if (status)
2651                         goto out;
2652         }
2653         new->cl_minorversion = cstate->minorversion;
2654         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2655         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2656
2657         gen_clid(new, nn);
2658         add_to_unconfirmed(new);
2659         swap(new, conf);
2660 out_copy:
2661         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2662         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2663
2664         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2665         nfsd4_set_ex_flags(conf, exid);
2666
2667         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2668                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2669         status = nfs_ok;
2670
2671 out:
2672         spin_unlock(&nn->client_lock);
2673 out_nolock:
2674         if (new)
2675                 expire_client(new);
2676         if (unconf)
2677                 expire_client(unconf);
2678         return status;
2679 }
2680
2681 static __be32
2682 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2683 {
2684         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2685                 slot_seqid);
2686
2687         /* The slot is in use, and no response has been sent. */
2688         if (slot_inuse) {
2689                 if (seqid == slot_seqid)
2690                         return nfserr_jukebox;
2691                 else
2692                         return nfserr_seq_misordered;
2693         }
2694         /* Note unsigned 32-bit arithmetic handles wraparound: */
2695         if (likely(seqid == slot_seqid + 1))
2696                 return nfs_ok;
2697         if (seqid == slot_seqid)
2698                 return nfserr_replay_cache;
2699         return nfserr_seq_misordered;
2700 }
2701
2702 /*
2703  * Cache the create session result into the create session single DRC
2704  * slot cache by saving the xdr structure. sl_seqid has been set.
2705  * Do this for solo or embedded create session operations.
2706  */
2707 static void
2708 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2709                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2710 {
2711         slot->sl_status = nfserr;
2712         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2713 }
2714
2715 static __be32
2716 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2717                             struct nfsd4_clid_slot *slot)
2718 {
2719         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2720         return slot->sl_status;
2721 }
2722
2723 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2724                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2725                         1 +     /* MIN tag is length with zero, only length */ \
2726                         3 +     /* version, opcount, opcode */ \
2727                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2728                                 /* seqid, slotID, slotID, cache */ \
2729                         4 ) * sizeof(__be32))
2730
2731 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2732                         2 +     /* verifier: AUTH_NULL, length 0 */\
2733                         1 +     /* status */ \
2734                         1 +     /* MIN tag is length with zero, only length */ \
2735                         3 +     /* opcount, opcode, opstatus*/ \
2736                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2737                                 /* seqid, slotID, slotID, slotID, status */ \
2738                         5 ) * sizeof(__be32))
2739
2740 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2741 {
2742         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2743
2744         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2745                 return nfserr_toosmall;
2746         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2747                 return nfserr_toosmall;
2748         ca->headerpadsz = 0;
2749         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2750         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2751         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2752         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2753                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2754         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2755         /*
2756          * Note decreasing slot size below client's request may make it
2757          * difficult for client to function correctly, whereas
2758          * decreasing the number of slots will (just?) affect
2759          * performance.  When short on memory we therefore prefer to
2760          * decrease number of slots instead of their size.  Clients that
2761          * request larger slots than they need will get poor results:
2762          */
2763         ca->maxreqs = nfsd4_get_drc_mem(ca);
2764         if (!ca->maxreqs)
2765                 return nfserr_jukebox;
2766
2767         return nfs_ok;
2768 }
2769
2770 /*
2771  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2772  * These are based on similar macros in linux/sunrpc/msg_prot.h .
2773  */
2774 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2775         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2776
2777 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2778         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2779
2780 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2781                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2782 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2783                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2784                                  sizeof(__be32))
2785
2786 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2787 {
2788         ca->headerpadsz = 0;
2789
2790         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2791                 return nfserr_toosmall;
2792         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2793                 return nfserr_toosmall;
2794         ca->maxresp_cached = 0;
2795         if (ca->maxops < 2)
2796                 return nfserr_toosmall;
2797
2798         return nfs_ok;
2799 }
2800
2801 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2802 {
2803         switch (cbs->flavor) {
2804         case RPC_AUTH_NULL:
2805         case RPC_AUTH_UNIX:
2806                 return nfs_ok;
2807         default:
2808                 /*
2809                  * GSS case: the spec doesn't allow us to return this
2810                  * error.  But it also doesn't allow us not to support
2811                  * GSS.
2812                  * I'd rather this fail hard than return some error the
2813                  * client might think it can already handle:
2814                  */
2815                 return nfserr_encr_alg_unsupp;
2816         }
2817 }
2818
2819 __be32
2820 nfsd4_create_session(struct svc_rqst *rqstp,
2821                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
2822 {
2823         struct nfsd4_create_session *cr_ses = &u->create_session;
2824         struct sockaddr *sa = svc_addr(rqstp);
2825         struct nfs4_client *conf, *unconf;
2826         struct nfs4_client *old = NULL;
2827         struct nfsd4_session *new;
2828         struct nfsd4_conn *conn;
2829         struct nfsd4_clid_slot *cs_slot = NULL;
2830         __be32 status = 0;
2831         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2832
2833         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2834                 return nfserr_inval;
2835         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2836         if (status)
2837                 return status;
2838         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2839         if (status)
2840                 return status;
2841         status = check_backchannel_attrs(&cr_ses->back_channel);
2842         if (status)
2843                 goto out_release_drc_mem;
2844         status = nfserr_jukebox;
2845         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2846         if (!new)
2847                 goto out_release_drc_mem;
2848         conn = alloc_conn_from_crses(rqstp, cr_ses);
2849         if (!conn)
2850                 goto out_free_session;
2851
2852         spin_lock(&nn->client_lock);
2853         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2854         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2855         WARN_ON_ONCE(conf && unconf);
2856
2857         if (conf) {
2858                 status = nfserr_wrong_cred;
2859                 if (!nfsd4_mach_creds_match(conf, rqstp))
2860                         goto out_free_conn;
2861                 cs_slot = &conf->cl_cs_slot;
2862                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2863                 if (status) {
2864                         if (status == nfserr_replay_cache)
2865                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2866                         goto out_free_conn;
2867                 }
2868         } else if (unconf) {
2869                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2870                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2871                         status = nfserr_clid_inuse;
2872                         goto out_free_conn;
2873                 }
2874                 status = nfserr_wrong_cred;
2875                 if (!nfsd4_mach_creds_match(unconf, rqstp))
2876                         goto out_free_conn;
2877                 cs_slot = &unconf->cl_cs_slot;
2878                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2879                 if (status) {
2880                         /* an unconfirmed replay returns misordered */
2881                         status = nfserr_seq_misordered;
2882                         goto out_free_conn;
2883                 }
2884                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2885                 if (old) {
2886                         status = mark_client_expired_locked(old);
2887                         if (status) {
2888                                 old = NULL;
2889                                 goto out_free_conn;
2890                         }
2891                 }
2892                 move_to_confirmed(unconf);
2893                 conf = unconf;
2894         } else {
2895                 status = nfserr_stale_clientid;
2896                 goto out_free_conn;
2897         }
2898         status = nfs_ok;
2899         /* Persistent sessions are not supported */
2900         cr_ses->flags &= ~SESSION4_PERSIST;
2901         /* Upshifting from TCP to RDMA is not supported */
2902         cr_ses->flags &= ~SESSION4_RDMA;
2903
2904         init_session(rqstp, new, conf, cr_ses);
2905         nfsd4_get_session_locked(new);
2906
2907         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2908                NFS4_MAX_SESSIONID_LEN);
2909         cs_slot->sl_seqid++;
2910         cr_ses->seqid = cs_slot->sl_seqid;
2911
2912         /* cache solo and embedded create sessions under the client_lock */
2913         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2914         spin_unlock(&nn->client_lock);
2915         /* init connection and backchannel */
2916         nfsd4_init_conn(rqstp, conn, new);
2917         nfsd4_put_session(new);
2918         if (old)
2919                 expire_client(old);
2920         return status;
2921 out_free_conn:
2922         spin_unlock(&nn->client_lock);
2923         free_conn(conn);
2924         if (old)
2925                 expire_client(old);
2926 out_free_session:
2927         __free_session(new);
2928 out_release_drc_mem:
2929         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2930         return status;
2931 }
2932
2933 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2934 {
2935         switch (*dir) {
2936         case NFS4_CDFC4_FORE:
2937         case NFS4_CDFC4_BACK:
2938                 return nfs_ok;
2939         case NFS4_CDFC4_FORE_OR_BOTH:
2940         case NFS4_CDFC4_BACK_OR_BOTH:
2941                 *dir = NFS4_CDFC4_BOTH;
2942                 return nfs_ok;
2943         };
2944         return nfserr_inval;
2945 }
2946
2947 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
2948                 struct nfsd4_compound_state *cstate,
2949                 union nfsd4_op_u *u)
2950 {
2951         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
2952         struct nfsd4_session *session = cstate->session;
2953         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2954         __be32 status;
2955
2956         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2957         if (status)
2958                 return status;
2959         spin_lock(&nn->client_lock);
2960         session->se_cb_prog = bc->bc_cb_program;
2961         session->se_cb_sec = bc->bc_cb_sec;
2962         spin_unlock(&nn->client_lock);
2963
2964         nfsd4_probe_callback(session->se_client);
2965
2966         return nfs_ok;
2967 }
2968
2969 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2970                      struct nfsd4_compound_state *cstate,
2971                      union nfsd4_op_u *u)
2972 {
2973         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
2974         __be32 status;
2975         struct nfsd4_conn *conn;
2976         struct nfsd4_session *session;
2977         struct net *net = SVC_NET(rqstp);
2978         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2979
2980         if (!nfsd4_last_compound_op(rqstp))
2981                 return nfserr_not_only_op;
2982         spin_lock(&nn->client_lock);
2983         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2984         spin_unlock(&nn->client_lock);
2985         if (!session)
2986                 goto out_no_session;
2987         status = nfserr_wrong_cred;
2988         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2989                 goto out;
2990         status = nfsd4_map_bcts_dir(&bcts->dir);
2991         if (status)
2992                 goto out;
2993         conn = alloc_conn(rqstp, bcts->dir);
2994         status = nfserr_jukebox;
2995         if (!conn)
2996                 goto out;
2997         nfsd4_init_conn(rqstp, conn, session);
2998         status = nfs_ok;
2999 out:
3000         nfsd4_put_session(session);
3001 out_no_session:
3002         return status;
3003 }
3004
3005 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3006 {
3007         if (!cstate->session)
3008                 return false;
3009         return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3010 }
3011
3012 __be32
3013 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3014                 union nfsd4_op_u *u)
3015 {
3016         struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3017         struct nfsd4_session *ses;
3018         __be32 status;
3019         int ref_held_by_me = 0;
3020         struct net *net = SVC_NET(r);
3021         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3022
3023         status = nfserr_not_only_op;
3024         if (nfsd4_compound_in_session(cstate, sessionid)) {
3025                 if (!nfsd4_last_compound_op(r))
3026                         goto out;
3027                 ref_held_by_me++;
3028         }
3029         dump_sessionid(__func__, sessionid);
3030         spin_lock(&nn->client_lock);
3031         ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3032         if (!ses)
3033                 goto out_client_lock;
3034         status = nfserr_wrong_cred;
3035         if (!nfsd4_mach_creds_match(ses->se_client, r))
3036                 goto out_put_session;
3037         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3038         if (status)
3039                 goto out_put_session;
3040         unhash_session(ses);
3041         spin_unlock(&nn->client_lock);
3042
3043         nfsd4_probe_callback_sync(ses->se_client);
3044
3045         spin_lock(&nn->client_lock);
3046         status = nfs_ok;
3047 out_put_session:
3048         nfsd4_put_session_locked(ses);
3049 out_client_lock:
3050         spin_unlock(&nn->client_lock);
3051 out:
3052         return status;
3053 }
3054
3055 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3056 {
3057         struct nfsd4_conn *c;
3058
3059         list_for_each_entry(c, &s->se_conns, cn_persession) {
3060                 if (c->cn_xprt == xpt) {
3061                         return c;
3062                 }
3063         }
3064         return NULL;
3065 }
3066
3067 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3068 {
3069         struct nfs4_client *clp = ses->se_client;
3070         struct nfsd4_conn *c;
3071         __be32 status = nfs_ok;
3072         int ret;
3073
3074         spin_lock(&clp->cl_lock);
3075         c = __nfsd4_find_conn(new->cn_xprt, ses);
3076         if (c)
3077                 goto out_free;
3078         status = nfserr_conn_not_bound_to_session;
3079         if (clp->cl_mach_cred)
3080                 goto out_free;
3081         __nfsd4_hash_conn(new, ses);
3082         spin_unlock(&clp->cl_lock);
3083         ret = nfsd4_register_conn(new);
3084         if (ret)
3085                 /* oops; xprt is already down: */
3086                 nfsd4_conn_lost(&new->cn_xpt_user);
3087         return nfs_ok;
3088 out_free:
3089         spin_unlock(&clp->cl_lock);
3090         free_conn(new);
3091         return status;
3092 }
3093
3094 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3095 {
3096         struct nfsd4_compoundargs *args = rqstp->rq_argp;
3097
3098         return args->opcnt > session->se_fchannel.maxops;
3099 }
3100
3101 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3102                                   struct nfsd4_session *session)
3103 {
3104         struct xdr_buf *xb = &rqstp->rq_arg;
3105
3106         return xb->len > session->se_fchannel.maxreq_sz;
3107 }
3108
3109 static bool replay_matches_cache(struct svc_rqst *rqstp,
3110                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3111 {
3112         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3113
3114         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3115             (bool)seq->cachethis)
3116                 return false;
3117         /*
3118          * If there's an error than the reply can have fewer ops than
3119          * the call.  But if we cached a reply with *more* ops than the
3120          * call you're sending us now, then this new call is clearly not
3121          * really a replay of the old one:
3122          */
3123         if (slot->sl_opcnt < argp->opcnt)
3124                 return false;
3125         /* This is the only check explicitly called by spec: */
3126         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3127                 return false;
3128         /*
3129          * There may be more comparisons we could actually do, but the
3130          * spec doesn't require us to catch every case where the calls
3131          * don't match (that would require caching the call as well as
3132          * the reply), so we don't bother.
3133          */
3134         return true;
3135 }
3136
3137 __be32
3138 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3139                 union nfsd4_op_u *u)
3140 {
3141         struct nfsd4_sequence *seq = &u->sequence;
3142         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3143         struct xdr_stream *xdr = &resp->xdr;
3144         struct nfsd4_session *session;
3145         struct nfs4_client *clp;
3146         struct nfsd4_slot *slot;
3147         struct nfsd4_conn *conn;
3148         __be32 status;
3149         int buflen;
3150         struct net *net = SVC_NET(rqstp);
3151         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3152
3153         if (resp->opcnt != 1)
3154                 return nfserr_sequence_pos;
3155
3156         /*
3157          * Will be either used or freed by nfsd4_sequence_check_conn
3158          * below.
3159          */
3160         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3161         if (!conn)
3162                 return nfserr_jukebox;
3163
3164         spin_lock(&nn->client_lock);
3165         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3166         if (!session)
3167                 goto out_no_session;
3168         clp = session->se_client;
3169
3170         status = nfserr_too_many_ops;
3171         if (nfsd4_session_too_many_ops(rqstp, session))
3172                 goto out_put_session;
3173
3174         status = nfserr_req_too_big;
3175         if (nfsd4_request_too_big(rqstp, session))
3176                 goto out_put_session;
3177
3178         status = nfserr_badslot;
3179         if (seq->slotid >= session->se_fchannel.maxreqs)
3180                 goto out_put_session;
3181
3182         slot = session->se_slots[seq->slotid];
3183         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3184
3185         /* We do not negotiate the number of slots yet, so set the
3186          * maxslots to the session maxreqs which is used to encode
3187          * sr_highest_slotid and the sr_target_slot id to maxslots */
3188         seq->maxslots = session->se_fchannel.maxreqs;
3189
3190         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3191                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3192         if (status == nfserr_replay_cache) {
3193                 status = nfserr_seq_misordered;
3194                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3195                         goto out_put_session;
3196                 status = nfserr_seq_false_retry;
3197                 if (!replay_matches_cache(rqstp, seq, slot))
3198                         goto out_put_session;
3199                 cstate->slot = slot;
3200                 cstate->session = session;
3201                 cstate->clp = clp;
3202                 /* Return the cached reply status and set cstate->status
3203                  * for nfsd4_proc_compound processing */
3204                 status = nfsd4_replay_cache_entry(resp, seq);
3205                 cstate->status = nfserr_replay_cache;
3206                 goto out;
3207         }
3208         if (status)
3209                 goto out_put_session;
3210
3211         status = nfsd4_sequence_check_conn(conn, session);
3212         conn = NULL;
3213         if (status)
3214                 goto out_put_session;
3215
3216         buflen = (seq->cachethis) ?
3217                         session->se_fchannel.maxresp_cached :
3218                         session->se_fchannel.maxresp_sz;
3219         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3220                                     nfserr_rep_too_big;
3221         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3222                 goto out_put_session;
3223         svc_reserve(rqstp, buflen);
3224
3225         status = nfs_ok;
3226         /* Success! bump slot seqid */
3227         slot->sl_seqid = seq->seqid;
3228         slot->sl_flags |= NFSD4_SLOT_INUSE;
3229         if (seq->cachethis)
3230                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3231         else
3232                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3233
3234         cstate->slot = slot;
3235         cstate->session = session;
3236         cstate->clp = clp;
3237
3238 out:
3239         switch (clp->cl_cb_state) {
3240         case NFSD4_CB_DOWN:
3241                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3242                 break;
3243         case NFSD4_CB_FAULT:
3244                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3245                 break;
3246         default:
3247                 seq->status_flags = 0;
3248         }
3249         if (!list_empty(&clp->cl_revoked))
3250                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3251 out_no_session:
3252         if (conn)
3253                 free_conn(conn);
3254         spin_unlock(&nn->client_lock);
3255         return status;
3256 out_put_session:
3257         nfsd4_put_session_locked(session);
3258         goto out_no_session;
3259 }
3260
3261 void
3262 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3263 {
3264         struct nfsd4_compound_state *cs = &resp->cstate;
3265
3266         if (nfsd4_has_session(cs)) {
3267                 if (cs->status != nfserr_replay_cache) {
3268                         nfsd4_store_cache_entry(resp);
3269                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3270                 }
3271                 /* Drop session reference that was taken in nfsd4_sequence() */
3272                 nfsd4_put_session(cs->session);
3273         } else if (cs->clp)
3274                 put_client_renew(cs->clp);
3275 }
3276
3277 __be32
3278 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3279                 struct nfsd4_compound_state *cstate,
3280                 union nfsd4_op_u *u)
3281 {
3282         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3283         struct nfs4_client *conf, *unconf;
3284         struct nfs4_client *clp = NULL;
3285         __be32 status = 0;
3286         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3287
3288         spin_lock(&nn->client_lock);
3289         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3290         conf = find_confirmed_client(&dc->clientid, true, nn);
3291         WARN_ON_ONCE(conf && unconf);
3292
3293         if (conf) {
3294                 if (client_has_state(conf)) {
3295                         status = nfserr_clientid_busy;
3296                         goto out;
3297                 }
3298                 status = mark_client_expired_locked(conf);
3299                 if (status)
3300                         goto out;
3301                 clp = conf;
3302         } else if (unconf)
3303                 clp = unconf;
3304         else {
3305                 status = nfserr_stale_clientid;
3306                 goto out;
3307         }
3308         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3309                 clp = NULL;
3310                 status = nfserr_wrong_cred;
3311                 goto out;
3312         }
3313         unhash_client_locked(clp);
3314 out:
3315         spin_unlock(&nn->client_lock);
3316         if (clp)
3317                 expire_client(clp);
3318         return status;
3319 }
3320
3321 __be32
3322 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3323                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3324 {
3325         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3326         __be32 status = 0;
3327
3328         if (rc->rca_one_fs) {
3329                 if (!cstate->current_fh.fh_dentry)
3330                         return nfserr_nofilehandle;
3331                 /*
3332                  * We don't take advantage of the rca_one_fs case.
3333                  * That's OK, it's optional, we can safely ignore it.
3334                  */
3335                 return nfs_ok;
3336         }
3337
3338         status = nfserr_complete_already;
3339         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3340                              &cstate->session->se_client->cl_flags))
3341                 goto out;
3342
3343         status = nfserr_stale_clientid;
3344         if (is_client_expired(cstate->session->se_client))
3345                 /*
3346                  * The following error isn't really legal.
3347                  * But we only get here if the client just explicitly
3348                  * destroyed the client.  Surely it no longer cares what
3349                  * error it gets back on an operation for the dead
3350                  * client.
3351                  */
3352                 goto out;
3353
3354         status = nfs_ok;
3355         nfsd4_client_record_create(cstate->session->se_client);
3356 out:
3357         return status;
3358 }
3359
3360 __be32
3361 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3362                   union nfsd4_op_u *u)
3363 {
3364         struct nfsd4_setclientid *setclid = &u->setclientid;
3365         struct xdr_netobj       clname = setclid->se_name;
3366         nfs4_verifier           clverifier = setclid->se_verf;
3367         struct nfs4_client      *conf, *new;
3368         struct nfs4_client      *unconf = NULL;
3369         __be32                  status;
3370         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3371
3372         new = create_client(clname, rqstp, &clverifier);
3373         if (new == NULL)
3374                 return nfserr_jukebox;
3375         /* Cases below refer to rfc 3530 section 14.2.33: */
3376         spin_lock(&nn->client_lock);
3377         conf = find_confirmed_client_by_name(&clname, nn);
3378         if (conf && client_has_state(conf)) {
3379                 /* case 0: */
3380                 status = nfserr_clid_inuse;
3381                 if (clp_used_exchangeid(conf))
3382                         goto out;
3383                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3384                         char addr_str[INET6_ADDRSTRLEN];
3385                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3386                                  sizeof(addr_str));
3387                         dprintk("NFSD: setclientid: string in use by client "
3388                                 "at %s\n", addr_str);
3389                         goto out;
3390                 }
3391         }
3392         unconf = find_unconfirmed_client_by_name(&clname, nn);
3393         if (unconf)
3394                 unhash_client_locked(unconf);
3395         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3396                 /* case 1: probable callback update */
3397                 copy_clid(new, conf);
3398                 gen_confirm(new, nn);
3399         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3400                 gen_clid(new, nn);
3401         new->cl_minorversion = 0;
3402         gen_callback(new, setclid, rqstp);
3403         add_to_unconfirmed(new);
3404         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3405         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3406         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3407         new = NULL;
3408         status = nfs_ok;
3409 out:
3410         spin_unlock(&nn->client_lock);
3411         if (new)
3412                 free_client(new);
3413         if (unconf)
3414                 expire_client(unconf);
3415         return status;
3416 }
3417
3418
3419 __be32
3420 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3421                         struct nfsd4_compound_state *cstate,
3422                         union nfsd4_op_u *u)
3423 {
3424         struct nfsd4_setclientid_confirm *setclientid_confirm =
3425                         &u->setclientid_confirm;
3426         struct nfs4_client *conf, *unconf;
3427         struct nfs4_client *old = NULL;
3428         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3429         clientid_t * clid = &setclientid_confirm->sc_clientid;
3430         __be32 status;
3431         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3432
3433         if (STALE_CLIENTID(clid, nn))
3434                 return nfserr_stale_clientid;
3435
3436         spin_lock(&nn->client_lock);
3437         conf = find_confirmed_client(clid, false, nn);
3438         unconf = find_unconfirmed_client(clid, false, nn);
3439         /*
3440          * We try hard to give out unique clientid's, so if we get an
3441          * attempt to confirm the same clientid with a different cred,
3442          * the client may be buggy; this should never happen.
3443          *
3444          * Nevertheless, RFC 7530 recommends INUSE for this case:
3445          */
3446         status = nfserr_clid_inuse;
3447         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3448                 goto out;
3449         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3450                 goto out;
3451         /* cases below refer to rfc 3530 section 14.2.34: */
3452         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3453                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3454                         /* case 2: probable retransmit */
3455                         status = nfs_ok;
3456                 } else /* case 4: client hasn't noticed we rebooted yet? */
3457                         status = nfserr_stale_clientid;
3458                 goto out;
3459         }
3460         status = nfs_ok;
3461         if (conf) { /* case 1: callback update */
3462                 old = unconf;
3463                 unhash_client_locked(old);
3464                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3465         } else { /* case 3: normal case; new or rebooted client */
3466                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3467                 if (old) {
3468                         status = nfserr_clid_inuse;
3469                         if (client_has_state(old)
3470                                         && !same_creds(&unconf->cl_cred,
3471                                                         &old->cl_cred))
3472                                 goto out;
3473                         status = mark_client_expired_locked(old);
3474                         if (status) {
3475                                 old = NULL;
3476                                 goto out;
3477                         }
3478                 }
3479                 move_to_confirmed(unconf);
3480                 conf = unconf;
3481         }
3482         get_client_locked(conf);
3483         spin_unlock(&nn->client_lock);
3484         nfsd4_probe_callback(conf);
3485         spin_lock(&nn->client_lock);
3486         put_client_renew_locked(conf);
3487 out:
3488         spin_unlock(&nn->client_lock);
3489         if (old)
3490                 expire_client(old);
3491         return status;
3492 }
3493
3494 static struct nfs4_file *nfsd4_alloc_file(void)
3495 {
3496         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3497 }
3498
3499 /* OPEN Share state helper functions */
3500 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3501                                 struct nfs4_file *fp)
3502 {
3503         lockdep_assert_held(&state_lock);
3504
3505         refcount_set(&fp->fi_ref, 1);
3506         spin_lock_init(&fp->fi_lock);
3507         INIT_LIST_HEAD(&fp->fi_stateids);
3508         INIT_LIST_HEAD(&fp->fi_delegations);
3509         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3510         fh_copy_shallow(&fp->fi_fhandle, fh);
3511         fp->fi_deleg_file = NULL;
3512         fp->fi_had_conflict = false;
3513         fp->fi_share_deny = 0;
3514         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3515         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3516 #ifdef CONFIG_NFSD_PNFS
3517         INIT_LIST_HEAD(&fp->fi_lo_states);
3518         atomic_set(&fp->fi_lo_recalls, 0);
3519 #endif
3520         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3521 }
3522
3523 void
3524 nfsd4_free_slabs(void)
3525 {
3526         kmem_cache_destroy(client_slab);
3527         kmem_cache_destroy(openowner_slab);
3528         kmem_cache_destroy(lockowner_slab);
3529         kmem_cache_destroy(file_slab);
3530         kmem_cache_destroy(stateid_slab);
3531         kmem_cache_destroy(deleg_slab);
3532         kmem_cache_destroy(odstate_slab);
3533 }
3534
3535 int
3536 nfsd4_init_slabs(void)
3537 {
3538         client_slab = kmem_cache_create("nfsd4_clients",
3539                         sizeof(struct nfs4_client), 0, 0, NULL);
3540         if (client_slab == NULL)
3541                 goto out;
3542         openowner_slab = kmem_cache_create("nfsd4_openowners",
3543                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3544         if (openowner_slab == NULL)
3545                 goto out_free_client_slab;
3546         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3547                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3548         if (lockowner_slab == NULL)
3549                 goto out_free_openowner_slab;
3550         file_slab = kmem_cache_create("nfsd4_files",
3551                         sizeof(struct nfs4_file), 0, 0, NULL);
3552         if (file_slab == NULL)
3553                 goto out_free_lockowner_slab;
3554         stateid_slab = kmem_cache_create("nfsd4_stateids",
3555                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3556         if (stateid_slab == NULL)
3557                 goto out_free_file_slab;
3558         deleg_slab = kmem_cache_create("nfsd4_delegations",
3559                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3560         if (deleg_slab == NULL)
3561                 goto out_free_stateid_slab;
3562         odstate_slab = kmem_cache_create("nfsd4_odstate",
3563                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3564         if (odstate_slab == NULL)
3565                 goto out_free_deleg_slab;
3566         return 0;
3567
3568 out_free_deleg_slab:
3569         kmem_cache_destroy(deleg_slab);
3570 out_free_stateid_slab:
3571         kmem_cache_destroy(stateid_slab);
3572 out_free_file_slab:
3573         kmem_cache_destroy(file_slab);
3574 out_free_lockowner_slab:
3575         kmem_cache_destroy(lockowner_slab);
3576 out_free_openowner_slab:
3577         kmem_cache_destroy(openowner_slab);
3578 out_free_client_slab:
3579         kmem_cache_destroy(client_slab);
3580 out:
3581         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3582         return -ENOMEM;
3583 }
3584
3585 static void init_nfs4_replay(struct nfs4_replay *rp)
3586 {
3587         rp->rp_status = nfserr_serverfault;
3588         rp->rp_buflen = 0;
3589         rp->rp_buf = rp->rp_ibuf;
3590         mutex_init(&rp->rp_mutex);
3591 }
3592
3593 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3594                 struct nfs4_stateowner *so)
3595 {
3596         if (!nfsd4_has_session(cstate)) {
3597                 mutex_lock(&so->so_replay.rp_mutex);
3598                 cstate->replay_owner = nfs4_get_stateowner(so);
3599         }
3600 }
3601
3602 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3603 {
3604         struct nfs4_stateowner *so = cstate->replay_owner;
3605
3606         if (so != NULL) {
3607                 cstate->replay_owner = NULL;
3608                 mutex_unlock(&so->so_replay.rp_mutex);
3609                 nfs4_put_stateowner(so);
3610         }
3611 }
3612
3613 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3614 {
3615         struct nfs4_stateowner *sop;
3616
3617         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3618         if (!sop)
3619                 return NULL;
3620
3621         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3622         if (!sop->so_owner.data) {
3623                 kmem_cache_free(slab, sop);
3624                 return NULL;
3625         }
3626         sop->so_owner.len = owner->len;
3627
3628         INIT_LIST_HEAD(&sop->so_stateids);
3629         sop->so_client = clp;
3630         init_nfs4_replay(&sop->so_replay);
3631         atomic_set(&sop->so_count, 1);
3632         return sop;
3633 }
3634
3635 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3636 {
3637         lockdep_assert_held(&clp->cl_lock);
3638
3639         list_add(&oo->oo_owner.so_strhash,
3640                  &clp->cl_ownerstr_hashtbl[strhashval]);
3641         list_add(&oo->oo_perclient, &clp->cl_openowners);
3642 }
3643
3644 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3645 {
3646         unhash_openowner_locked(openowner(so));
3647 }
3648
3649 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3650 {
3651         struct nfs4_openowner *oo = openowner(so);
3652
3653         kmem_cache_free(openowner_slab, oo);
3654 }
3655
3656 static const struct nfs4_stateowner_operations openowner_ops = {
3657         .so_unhash =    nfs4_unhash_openowner,
3658         .so_free =      nfs4_free_openowner,
3659 };
3660
3661 static struct nfs4_ol_stateid *
3662 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3663 {
3664         struct nfs4_ol_stateid *local, *ret = NULL;
3665         struct nfs4_openowner *oo = open->op_openowner;
3666
3667         lockdep_assert_held(&fp->fi_lock);
3668
3669         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3670                 /* ignore lock owners */
3671                 if (local->st_stateowner->so_is_open_owner == 0)
3672                         continue;
3673                 if (local->st_stateowner != &oo->oo_owner)
3674                         continue;
3675                 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
3676                         ret = local;
3677                         refcount_inc(&ret->st_stid.sc_count);
3678                         break;
3679                 }
3680         }
3681         return ret;
3682 }
3683
3684 static __be32
3685 nfsd4_verify_open_stid(struct nfs4_stid *s)
3686 {
3687         __be32 ret = nfs_ok;
3688
3689         switch (s->sc_type) {
3690         default:
3691                 break;
3692         case 0:
3693         case NFS4_CLOSED_STID:
3694         case NFS4_CLOSED_DELEG_STID:
3695                 ret = nfserr_bad_stateid;
3696                 break;
3697         case NFS4_REVOKED_DELEG_STID:
3698                 ret = nfserr_deleg_revoked;
3699         }
3700         return ret;
3701 }
3702
3703 /* Lock the stateid st_mutex, and deal with races with CLOSE */
3704 static __be32
3705 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
3706 {
3707         __be32 ret;
3708
3709         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
3710         ret = nfsd4_verify_open_stid(&stp->st_stid);
3711         if (ret != nfs_ok)
3712                 mutex_unlock(&stp->st_mutex);
3713         return ret;
3714 }
3715
3716 static struct nfs4_ol_stateid *
3717 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3718 {
3719         struct nfs4_ol_stateid *stp;
3720         for (;;) {
3721                 spin_lock(&fp->fi_lock);
3722                 stp = nfsd4_find_existing_open(fp, open);
3723                 spin_unlock(&fp->fi_lock);
3724                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
3725                         break;
3726                 nfs4_put_stid(&stp->st_stid);
3727         }
3728         return stp;
3729 }
3730
3731 static struct nfs4_openowner *
3732 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3733                            struct nfsd4_compound_state *cstate)
3734 {
3735         struct nfs4_client *clp = cstate->clp;
3736         struct nfs4_openowner *oo, *ret;
3737
3738         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3739         if (!oo)
3740                 return NULL;
3741         oo->oo_owner.so_ops = &openowner_ops;
3742         oo->oo_owner.so_is_open_owner = 1;
3743         oo->oo_owner.so_seqid = open->op_seqid;
3744         oo->oo_flags = 0;
3745         if (nfsd4_has_session(cstate))
3746                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3747         oo->oo_time = 0;
3748         oo->oo_last_closed_stid = NULL;
3749         INIT_LIST_HEAD(&oo->oo_close_lru);
3750         spin_lock(&clp->cl_lock);
3751         ret = find_openstateowner_str_locked(strhashval, open, clp);
3752         if (ret == NULL) {
3753                 hash_openowner(oo, clp, strhashval);
3754                 ret = oo;
3755         } else
3756                 nfs4_free_stateowner(&oo->oo_owner);
3757
3758         spin_unlock(&clp->cl_lock);
3759         return ret;
3760 }
3761
3762 static struct nfs4_ol_stateid *
3763 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3764 {
3765
3766         struct nfs4_openowner *oo = open->op_openowner;
3767         struct nfs4_ol_stateid *retstp = NULL;
3768         struct nfs4_ol_stateid *stp;
3769
3770         stp = open->op_stp;
3771         /* We are moving these outside of the spinlocks to avoid the warnings */
3772         mutex_init(&stp->st_mutex);
3773         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
3774
3775 retry:
3776         spin_lock(&oo->oo_owner.so_client->cl_lock);
3777         spin_lock(&fp->fi_lock);
3778
3779         retstp = nfsd4_find_existing_open(fp, open);
3780         if (retstp)
3781                 goto out_unlock;
3782
3783         open->op_stp = NULL;
3784         refcount_inc(&stp->st_stid.sc_count);
3785         stp->st_stid.sc_type = NFS4_OPEN_STID;
3786         INIT_LIST_HEAD(&stp->st_locks);
3787         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3788         get_nfs4_file(fp);
3789         stp->st_stid.sc_file = fp;
3790         stp->st_access_bmap = 0;
3791         stp->st_deny_bmap = 0;
3792         stp->st_openstp = NULL;
3793         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3794         list_add(&stp->st_perfile, &fp->fi_stateids);
3795
3796 out_unlock:
3797         spin_unlock(&fp->fi_lock);
3798         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3799         if (retstp) {
3800                 /* Handle races with CLOSE */
3801                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
3802                         nfs4_put_stid(&retstp->st_stid);
3803                         goto retry;
3804                 }
3805                 /* To keep mutex tracking happy */
3806                 mutex_unlock(&stp->st_mutex);
3807                 stp = retstp;
3808         }
3809         return stp;
3810 }
3811
3812 /*
3813  * In the 4.0 case we need to keep the owners around a little while to handle
3814  * CLOSE replay. We still do need to release any file access that is held by
3815  * them before returning however.
3816  */
3817 static void
3818 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3819 {
3820         struct nfs4_ol_stateid *last;
3821         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3822         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3823                                                 nfsd_net_id);
3824
3825         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3826
3827         /*
3828          * We know that we hold one reference via nfsd4_close, and another
3829          * "persistent" reference for the client. If the refcount is higher
3830          * than 2, then there are still calls in progress that are using this
3831          * stateid. We can't put the sc_file reference until they are finished.
3832          * Wait for the refcount to drop to 2. Since it has been unhashed,
3833          * there should be no danger of the refcount going back up again at
3834          * this point.
3835          */
3836         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
3837
3838         release_all_access(s);
3839         if (s->st_stid.sc_file) {
3840                 put_nfs4_file(s->st_stid.sc_file);
3841                 s->st_stid.sc_file = NULL;
3842         }
3843
3844         spin_lock(&nn->client_lock);
3845         last = oo->oo_last_closed_stid;
3846         oo->oo_last_closed_stid = s;
3847         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3848         oo->oo_time = get_seconds();
3849         spin_unlock(&nn->client_lock);
3850         if (last)
3851                 nfs4_put_stid(&last->st_stid);
3852 }
3853
3854 /* search file_hashtbl[] for file */
3855 static struct nfs4_file *
3856 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3857 {
3858         struct nfs4_file *fp;
3859
3860         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3861                 if (fh_match(&fp->fi_fhandle, fh)) {
3862                         if (refcount_inc_not_zero(&fp->fi_ref))
3863                                 return fp;
3864                 }
3865         }
3866         return NULL;
3867 }
3868
3869 struct nfs4_file *
3870 find_file(struct knfsd_fh *fh)
3871 {
3872         struct nfs4_file *fp;
3873         unsigned int hashval = file_hashval(fh);
3874
3875         rcu_read_lock();
3876         fp = find_file_locked(fh, hashval);
3877         rcu_read_unlock();
3878         return fp;
3879 }
3880
3881 static struct nfs4_file *
3882 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3883 {
3884         struct nfs4_file *fp;
3885         unsigned int hashval = file_hashval(fh);
3886
3887         rcu_read_lock();
3888         fp = find_file_locked(fh, hashval);
3889         rcu_read_unlock();
3890         if (fp)
3891                 return fp;
3892
3893         spin_lock(&state_lock);
3894         fp = find_file_locked(fh, hashval);
3895         if (likely(fp == NULL)) {
3896                 nfsd4_init_file(fh, hashval, new);
3897                 fp = new;
3898         }
3899         spin_unlock(&state_lock);
3900
3901         return fp;
3902 }
3903
3904 /*
3905  * Called to check deny when READ with all zero stateid or
3906  * WRITE with all zero or all one stateid
3907  */
3908 static __be32
3909 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3910 {
3911         struct nfs4_file *fp;
3912         __be32 ret = nfs_ok;
3913
3914         fp = find_file(&current_fh->fh_handle);
3915         if (!fp)
3916                 return ret;
3917         /* Check for conflicting share reservations */
3918         spin_lock(&fp->fi_lock);
3919         if (fp->fi_share_deny & deny_type)
3920                 ret = nfserr_locked;
3921         spin_unlock(&fp->fi_lock);
3922         put_nfs4_file(fp);
3923         return ret;
3924 }
3925
3926 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3927 {
3928         struct nfs4_delegation *dp = cb_to_delegation(cb);
3929         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3930                                           nfsd_net_id);
3931
3932         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3933
3934         /*
3935          * We can't do this in nfsd_break_deleg_cb because it is
3936          * already holding inode->i_lock.
3937          *
3938          * If the dl_time != 0, then we know that it has already been
3939          * queued for a lease break. Don't queue it again.
3940          */
3941         spin_lock(&state_lock);
3942         if (dp->dl_time == 0) {
3943                 dp->dl_time = get_seconds();
3944                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3945         }
3946         spin_unlock(&state_lock);
3947 }
3948
3949 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3950                 struct rpc_task *task)
3951 {
3952         struct nfs4_delegation *dp = cb_to_delegation(cb);
3953
3954         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3955                 return 1;
3956
3957         switch (task->tk_status) {
3958         case 0:
3959                 return 1;
3960         case -EBADHANDLE:
3961         case -NFS4ERR_BAD_STATEID:
3962                 /*
3963                  * Race: client probably got cb_recall before open reply
3964                  * granting delegation.
3965                  */
3966                 if (dp->dl_retries--) {
3967                         rpc_delay(task, 2 * HZ);
3968                         return 0;
3969                 }
3970                 /*FALLTHRU*/
3971         default:
3972                 return -1;
3973         }
3974 }
3975
3976 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3977 {
3978         struct nfs4_delegation *dp = cb_to_delegation(cb);
3979
3980         nfs4_put_stid(&dp->dl_stid);
3981 }
3982
3983 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3984         .prepare        = nfsd4_cb_recall_prepare,
3985         .done           = nfsd4_cb_recall_done,
3986         .release        = nfsd4_cb_recall_release,
3987 };
3988
3989 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3990 {
3991         /*
3992          * We're assuming the state code never drops its reference
3993          * without first removing the lease.  Since we're in this lease
3994          * callback (and since the lease code is serialized by the
3995          * i_lock) we know the server hasn't removed the lease yet, and
3996          * we know it's safe to take a reference.
3997          */
3998         refcount_inc(&dp->dl_stid.sc_count);
3999         nfsd4_run_cb(&dp->dl_recall);
4000 }
4001
4002 /* Called from break_lease() with i_lock held. */
4003 static bool
4004 nfsd_break_deleg_cb(struct file_lock *fl)
4005 {
4006         bool ret = false;
4007         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4008         struct nfs4_file *fp = dp->dl_stid.sc_file;
4009
4010         /*
4011          * We don't want the locks code to timeout the lease for us;
4012          * we'll remove it ourself if a delegation isn't returned
4013          * in time:
4014          */
4015         fl->fl_break_time = 0;
4016
4017         spin_lock(&fp->fi_lock);
4018         fp->fi_had_conflict = true;
4019         nfsd_break_one_deleg(dp);
4020         spin_unlock(&fp->fi_lock);
4021         return ret;
4022 }
4023
4024 static int
4025 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4026                      struct list_head *dispose)
4027 {
4028         if (arg & F_UNLCK)
4029                 return lease_modify(onlist, arg, dispose);
4030         else
4031                 return -EAGAIN;
4032 }
4033
4034 static const struct lock_manager_operations nfsd_lease_mng_ops = {
4035         .lm_break = nfsd_break_deleg_cb,
4036         .lm_change = nfsd_change_deleg_cb,
4037 };
4038
4039 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4040 {
4041         if (nfsd4_has_session(cstate))
4042                 return nfs_ok;
4043         if (seqid == so->so_seqid - 1)
4044                 return nfserr_replay_me;
4045         if (seqid == so->so_seqid)
4046                 return nfs_ok;
4047         return nfserr_bad_seqid;
4048 }
4049
4050 static __be32 lookup_clientid(clientid_t *clid,
4051                 struct nfsd4_compound_state *cstate,
4052                 struct nfsd_net *nn)
4053 {
4054         struct nfs4_client *found;
4055
4056         if (cstate->clp) {
4057                 found = cstate->clp;
4058                 if (!same_clid(&found->cl_clientid, clid))
4059                         return nfserr_stale_clientid;
4060                 return nfs_ok;
4061         }
4062
4063         if (STALE_CLIENTID(clid, nn))
4064                 return nfserr_stale_clientid;
4065
4066         /*
4067          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4068          * cached already then we know this is for is for v4.0 and "sessions"
4069          * will be false.
4070          */
4071         WARN_ON_ONCE(cstate->session);
4072         spin_lock(&nn->client_lock);
4073         found = find_confirmed_client(clid, false, nn);
4074         if (!found) {
4075                 spin_unlock(&nn->client_lock);
4076                 return nfserr_expired;
4077         }
4078         atomic_inc(&found->cl_refcount);
4079         spin_unlock(&nn->client_lock);
4080
4081         /* Cache the nfs4_client in cstate! */
4082         cstate->clp = found;
4083         return nfs_ok;
4084 }
4085
4086 __be32
4087 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4088                     struct nfsd4_open *open, struct nfsd_net *nn)
4089 {
4090         clientid_t *clientid = &open->op_clientid;
4091         struct nfs4_client *clp = NULL;
4092         unsigned int strhashval;
4093         struct nfs4_openowner *oo = NULL;
4094         __be32 status;
4095
4096         if (STALE_CLIENTID(&open->op_clientid, nn))
4097                 return nfserr_stale_clientid;
4098         /*
4099          * In case we need it later, after we've already created the
4100          * file and don't want to risk a further failure:
4101          */
4102         open->op_file = nfsd4_alloc_file();
4103         if (open->op_file == NULL)
4104                 return nfserr_jukebox;
4105
4106         status = lookup_clientid(clientid, cstate, nn);
4107         if (status)
4108                 return status;
4109         clp = cstate->clp;
4110
4111         strhashval = ownerstr_hashval(&open->op_owner);
4112         oo = find_openstateowner_str(strhashval, open, clp);
4113         open->op_openowner = oo;
4114         if (!oo) {
4115                 goto new_owner;
4116         }
4117         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4118                 /* Replace unconfirmed owners without checking for replay. */
4119                 release_openowner(oo);
4120                 open->op_openowner = NULL;
4121                 goto new_owner;
4122         }
4123         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4124         if (status)
4125                 return status;
4126         goto alloc_stateid;
4127 new_owner:
4128         oo = alloc_init_open_stateowner(strhashval, open, cstate);
4129         if (oo == NULL)
4130                 return nfserr_jukebox;
4131         open->op_openowner = oo;
4132 alloc_stateid:
4133         open->op_stp = nfs4_alloc_open_stateid(clp);
4134         if (!open->op_stp)
4135                 return nfserr_jukebox;
4136
4137         if (nfsd4_has_session(cstate) &&
4138             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4139                 open->op_odstate = alloc_clnt_odstate(clp);
4140                 if (!open->op_odstate)
4141                         return nfserr_jukebox;
4142         }
4143
4144         return nfs_ok;
4145 }
4146
4147 static inline __be32
4148 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4149 {
4150         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4151                 return nfserr_openmode;
4152         else
4153                 return nfs_ok;
4154 }
4155
4156 static int share_access_to_flags(u32 share_access)
4157 {
4158         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4159 }
4160
4161 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4162 {
4163         struct nfs4_stid *ret;
4164
4165         ret = find_stateid_by_type(cl, s,
4166                                 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4167         if (!ret)
4168                 return NULL;
4169         return delegstateid(ret);
4170 }
4171
4172 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4173 {
4174         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4175                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4176 }
4177
4178 static __be32
4179 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4180                 struct nfs4_delegation **dp)
4181 {
4182         int flags;
4183         __be32 status = nfserr_bad_stateid;
4184         struct nfs4_delegation *deleg;
4185
4186         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4187         if (deleg == NULL)
4188                 goto out;
4189         if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4190                 nfs4_put_stid(&deleg->dl_stid);
4191                 if (cl->cl_minorversion)
4192                         status = nfserr_deleg_revoked;
4193                 goto out;
4194         }
4195         flags = share_access_to_flags(open->op_share_access);
4196         status = nfs4_check_delegmode(deleg, flags);
4197         if (status) {
4198                 nfs4_put_stid(&deleg->dl_stid);
4199                 goto out;
4200         }
4201         *dp = deleg;
4202 out:
4203         if (!nfsd4_is_deleg_cur(open))
4204                 return nfs_ok;
4205         if (status)
4206                 return status;
4207         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4208         return nfs_ok;
4209 }
4210
4211 static inline int nfs4_access_to_access(u32 nfs4_access)
4212 {
4213         int flags = 0;
4214
4215         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4216                 flags |= NFSD_MAY_READ;
4217         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4218                 flags |= NFSD_MAY_WRITE;
4219         return flags;
4220 }
4221
4222 static inline __be32
4223 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4224                 struct nfsd4_open *open)
4225 {
4226         struct iattr iattr = {
4227                 .ia_valid = ATTR_SIZE,
4228                 .ia_size = 0,
4229         };
4230         if (!open->op_truncate)
4231                 return 0;
4232         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4233                 return nfserr_inval;
4234         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4235 }
4236
4237 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4238                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4239                 struct nfsd4_open *open)
4240 {
4241         struct file *filp = NULL;
4242         __be32 status;
4243         int oflag = nfs4_access_to_omode(open->op_share_access);
4244         int access = nfs4_access_to_access(open->op_share_access);
4245         unsigned char old_access_bmap, old_deny_bmap;
4246
4247         spin_lock(&fp->fi_lock);
4248
4249         /*
4250          * Are we trying to set a deny mode that would conflict with
4251          * current access?
4252          */
4253         status = nfs4_file_check_deny(fp, open->op_share_deny);
4254         if (status != nfs_ok) {
4255                 spin_unlock(&fp->fi_lock);
4256                 goto out;
4257         }
4258
4259         /* set access to the file */
4260         status = nfs4_file_get_access(fp, open->op_share_access);
4261         if (status != nfs_ok) {
4262                 spin_unlock(&fp->fi_lock);
4263                 goto out;
4264         }
4265
4266         /* Set access bits in stateid */
4267         old_access_bmap = stp->st_access_bmap;
4268         set_access(open->op_share_access, stp);
4269
4270         /* Set new deny mask */
4271         old_deny_bmap = stp->st_deny_bmap;
4272         set_deny(open->op_share_deny, stp);
4273         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4274
4275         if (!fp->fi_fds[oflag]) {
4276                 spin_unlock(&fp->fi_lock);
4277                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4278                 if (status)
4279                         goto out_put_access;
4280                 spin_lock(&fp->fi_lock);
4281                 if (!fp->fi_fds[oflag]) {
4282                         fp->fi_fds[oflag] = filp;
4283                         filp = NULL;
4284                 }
4285         }
4286         spin_unlock(&fp->fi_lock);
4287         if (filp)
4288                 fput(filp);
4289
4290         status = nfsd4_truncate(rqstp, cur_fh, open);
4291         if (status)
4292                 goto out_put_access;
4293 out:
4294         return status;
4295 out_put_access:
4296         stp->st_access_bmap = old_access_bmap;
4297         nfs4_file_put_access(fp, open->op_share_access);
4298         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4299         goto out;
4300 }
4301
4302 static __be32
4303 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)
4304 {
4305         __be32 status;
4306         unsigned char old_deny_bmap = stp->st_deny_bmap;
4307
4308         if (!test_access(open->op_share_access, stp))
4309                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4310
4311         /* test and set deny mode */
4312         spin_lock(&fp->fi_lock);
4313         status = nfs4_file_check_deny(fp, open->op_share_deny);
4314         if (status == nfs_ok) {
4315                 set_deny(open->op_share_deny, stp);
4316                 fp->fi_share_deny |=
4317                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4318         }
4319         spin_unlock(&fp->fi_lock);
4320
4321         if (status != nfs_ok)
4322                 return status;
4323
4324         status = nfsd4_truncate(rqstp, cur_fh, open);
4325         if (status != nfs_ok)
4326                 reset_union_bmap_deny(old_deny_bmap, stp);
4327         return status;
4328 }
4329
4330 /* Should we give out recallable state?: */
4331 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4332 {
4333         if (clp->cl_cb_state == NFSD4_CB_UP)
4334                 return true;
4335         /*
4336          * In the sessions case, since we don't have to establish a
4337          * separate connection for callbacks, we assume it's OK
4338          * until we hear otherwise:
4339          */
4340         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4341 }
4342
4343 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
4344                                                 int flag)
4345 {
4346         struct file_lock *fl;
4347
4348         fl = locks_alloc_lock();
4349         if (!fl)
4350                 return NULL;
4351         fl->fl_lmops = &nfsd_lease_mng_ops;
4352         fl->fl_flags = FL_DELEG;
4353         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4354         fl->fl_end = OFFSET_MAX;
4355         fl->fl_owner = (fl_owner_t)dp;
4356         fl->fl_pid = current->tgid;
4357         fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file;
4358         return fl;
4359 }
4360
4361 static struct nfs4_delegation *
4362 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4363                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4364 {
4365         int status = 0;
4366         struct nfs4_delegation *dp;
4367         struct file *filp;
4368         struct file_lock *fl;
4369
4370         /*
4371          * The fi_had_conflict and nfs_get_existing_delegation checks
4372          * here are just optimizations; we'll need to recheck them at
4373          * the end:
4374          */
4375         if (fp->fi_had_conflict)
4376                 return ERR_PTR(-EAGAIN);
4377
4378         filp = find_readable_file(fp);
4379         if (!filp) {
4380                 /* We should always have a readable file here */
4381                 WARN_ON_ONCE(1);
4382                 return ERR_PTR(-EBADF);
4383         }
4384         spin_lock(&state_lock);
4385         spin_lock(&fp->fi_lock);
4386         if (nfs4_delegation_exists(clp, fp))
4387                 status = -EAGAIN;
4388         else if (!fp->fi_deleg_file) {
4389                 fp->fi_deleg_file = filp;
4390                 /* increment early to prevent fi_deleg_file from being
4391                  * cleared */
4392                 fp->fi_delegees = 1;
4393                 filp = NULL;
4394         } else
4395                 fp->fi_delegees++;
4396         spin_unlock(&fp->fi_lock);
4397         spin_unlock(&state_lock);
4398         if (filp)
4399                 fput(filp);
4400         if (status)
4401                 return ERR_PTR(status);
4402
4403         status = -ENOMEM;
4404         dp = alloc_init_deleg(clp, fp, fh, odstate);
4405         if (!dp)
4406                 goto out_delegees;
4407
4408         fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
4409         if (!fl)
4410                 goto out_clnt_odstate;
4411
4412         status = vfs_setlease(fp->fi_deleg_file, fl->fl_type, &fl, NULL);
4413         if (fl)
4414                 locks_free_lock(fl);
4415         if (status)
4416                 goto out_clnt_odstate;
4417
4418         spin_lock(&state_lock);
4419         spin_lock(&fp->fi_lock);
4420         if (fp->fi_had_conflict)
4421                 status = -EAGAIN;
4422         else
4423                 status = hash_delegation_locked(dp, fp);
4424         spin_unlock(&fp->fi_lock);
4425         spin_unlock(&state_lock);
4426
4427         if (status)
4428                 goto out_unlock;
4429
4430         return dp;
4431 out_unlock:
4432         vfs_setlease(fp->fi_deleg_file, F_UNLCK, NULL, (void **)&dp);
4433 out_clnt_odstate:
4434         put_clnt_odstate(dp->dl_clnt_odstate);
4435         nfs4_put_stid(&dp->dl_stid);
4436 out_delegees:
4437         put_deleg_file(fp);
4438         return ERR_PTR(status);
4439 }
4440
4441 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4442 {
4443         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4444         if (status == -EAGAIN)
4445                 open->op_why_no_deleg = WND4_CONTENTION;
4446         else {
4447                 open->op_why_no_deleg = WND4_RESOURCE;
4448                 switch (open->op_deleg_want) {
4449                 case NFS4_SHARE_WANT_READ_DELEG:
4450                 case NFS4_SHARE_WANT_WRITE_DELEG:
4451                 case NFS4_SHARE_WANT_ANY_DELEG:
4452                         break;
4453                 case NFS4_SHARE_WANT_CANCEL:
4454                         open->op_why_no_deleg = WND4_CANCELLED;
4455                         break;
4456                 case NFS4_SHARE_WANT_NO_DELEG:
4457                         WARN_ON_ONCE(1);
4458                 }
4459         }
4460 }
4461
4462 /*
4463  * Attempt to hand out a delegation.
4464  *
4465  * Note we don't support write delegations, and won't until the vfs has
4466  * proper support for them.
4467  */
4468 static void
4469 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4470                         struct nfs4_ol_stateid *stp)
4471 {
4472         struct nfs4_delegation *dp;
4473         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4474         struct nfs4_client *clp = stp->st_stid.sc_client;
4475         int cb_up;
4476         int status = 0;
4477
4478         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4479         open->op_recall = 0;
4480         switch (open->op_claim_type) {
4481                 case NFS4_OPEN_CLAIM_PREVIOUS:
4482                         if (!cb_up)
4483                                 open->op_recall = 1;
4484                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4485                                 goto out_no_deleg;
4486                         break;
4487                 case NFS4_OPEN_CLAIM_NULL:
4488                 case NFS4_OPEN_CLAIM_FH:
4489                         /*
4490                          * Let's not give out any delegations till everyone's
4491                          * had the chance to reclaim theirs, *and* until
4492                          * NLM locks have all been reclaimed:
4493                          */
4494                         if (locks_in_grace(clp->net))
4495                                 goto out_no_deleg;
4496                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4497                                 goto out_no_deleg;
4498                         /*
4499                          * Also, if the file was opened for write or
4500                          * create, there's a good chance the client's
4501                          * about to write to it, resulting in an
4502                          * immediate recall (since we don't support
4503                          * write delegations):
4504                          */
4505                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4506                                 goto out_no_deleg;
4507                         if (open->op_create == NFS4_OPEN_CREATE)
4508                                 goto out_no_deleg;
4509                         break;
4510                 default:
4511                         goto out_no_deleg;
4512         }
4513         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4514         if (IS_ERR(dp))
4515                 goto out_no_deleg;
4516
4517         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4518
4519         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4520                 STATEID_VAL(&dp->dl_stid.sc_stateid));
4521         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4522         nfs4_put_stid(&dp->dl_stid);
4523         return;
4524 out_no_deleg:
4525         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4526         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4527             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4528                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4529                 open->op_recall = 1;
4530         }
4531
4532         /* 4.1 client asking for a delegation? */
4533         if (open->op_deleg_want)
4534                 nfsd4_open_deleg_none_ext(open, status);
4535         return;
4536 }
4537
4538 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4539                                         struct nfs4_delegation *dp)
4540 {
4541         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4542             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4543                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4544                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4545         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4546                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4547                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4548                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4549         }
4550         /* Otherwise the client must be confused wanting a delegation
4551          * it already has, therefore we don't return
4552          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4553          */
4554 }
4555
4556 __be32
4557 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4558 {
4559         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4560         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4561         struct nfs4_file *fp = NULL;
4562         struct nfs4_ol_stateid *stp = NULL;
4563         struct nfs4_delegation *dp = NULL;
4564         __be32 status;
4565         bool new_stp = false;
4566
4567         /*
4568          * Lookup file; if found, lookup stateid and check open request,
4569          * and check for delegations in the process of being recalled.
4570          * If not found, create the nfs4_file struct
4571          */
4572         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4573         if (fp != open->op_file) {
4574                 status = nfs4_check_deleg(cl, open, &dp);
4575                 if (status)
4576                         goto out;
4577                 stp = nfsd4_find_and_lock_existing_open(fp, open);
4578         } else {
4579                 open->op_file = NULL;
4580                 status = nfserr_bad_stateid;
4581                 if (nfsd4_is_deleg_cur(open))
4582                         goto out;
4583         }
4584
4585         if (!stp) {
4586                 stp = init_open_stateid(fp, open);
4587                 if (!open->op_stp)
4588                         new_stp = true;
4589         }
4590
4591         /*
4592          * OPEN the file, or upgrade an existing OPEN.
4593          * If truncate fails, the OPEN fails.
4594          *
4595          * stp is already locked.
4596          */
4597         if (!new_stp) {
4598                 /* Stateid was found, this is an OPEN upgrade */
4599                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4600                 if (status) {
4601                         mutex_unlock(&stp->st_mutex);
4602                         goto out;
4603                 }
4604         } else {
4605                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4606                 if (status) {
4607                         stp->st_stid.sc_type = NFS4_CLOSED_STID;
4608                         release_open_stateid(stp);
4609                         mutex_unlock(&stp->st_mutex);
4610                         goto out;
4611                 }
4612
4613                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4614                                                         open->op_odstate);
4615                 if (stp->st_clnt_odstate == open->op_odstate)
4616                         open->op_odstate = NULL;
4617         }
4618
4619         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4620         mutex_unlock(&stp->st_mutex);
4621
4622         if (nfsd4_has_session(&resp->cstate)) {
4623                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4624                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4625                         open->op_why_no_deleg = WND4_NOT_WANTED;
4626                         goto nodeleg;
4627                 }
4628         }
4629
4630         /*
4631         * Attempt to hand out a delegation. No error return, because the
4632         * OPEN succeeds even if we fail.
4633         */
4634         nfs4_open_delegation(current_fh, open, stp);
4635 nodeleg:
4636         status = nfs_ok;
4637
4638         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4639                 STATEID_VAL(&stp->st_stid.sc_stateid));
4640 out:
4641         /* 4.1 client trying to upgrade/downgrade delegation? */
4642         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4643             open->op_deleg_want)
4644                 nfsd4_deleg_xgrade_none_ext(open, dp);
4645
4646         if (fp)
4647                 put_nfs4_file(fp);
4648         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4649                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4650         /*
4651         * To finish the open response, we just need to set the rflags.
4652         */
4653         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4654         if (nfsd4_has_session(&resp->cstate))
4655                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
4656         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
4657                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4658
4659         if (dp)
4660                 nfs4_put_stid(&dp->dl_stid);
4661         if (stp)
4662                 nfs4_put_stid(&stp->st_stid);
4663
4664         return status;
4665 }
4666
4667 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4668                               struct nfsd4_open *open)
4669 {
4670         if (open->op_openowner) {
4671                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4672
4673                 nfsd4_cstate_assign_replay(cstate, so);
4674                 nfs4_put_stateowner(so);
4675         }
4676         if (open->op_file)
4677                 kmem_cache_free(file_slab, open->op_file);
4678         if (open->op_stp)
4679                 nfs4_put_stid(&open->op_stp->st_stid);
4680         if (open->op_odstate)
4681                 kmem_cache_free(odstate_slab, open->op_odstate);
4682 }
4683
4684 __be32
4685 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4686             union nfsd4_op_u *u)
4687 {
4688         clientid_t *clid = &u->renew;
4689         struct nfs4_client *clp;
4690         __be32 status;
4691         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4692
4693         dprintk("process_renew(%08x/%08x): starting\n",
4694                         clid->cl_boot, clid->cl_id);
4695         status = lookup_clientid(clid, cstate, nn);
4696         if (status)
4697                 goto out;
4698         clp = cstate->clp;
4699         status = nfserr_cb_path_down;
4700         if (!list_empty(&clp->cl_delegations)
4701                         && clp->cl_cb_state != NFSD4_CB_UP)
4702                 goto out;
4703         status = nfs_ok;
4704 out:
4705         return status;
4706 }
4707
4708 void
4709 nfsd4_end_grace(struct nfsd_net *nn)
4710 {
4711         /* do nothing if grace period already ended */
4712         if (nn->grace_ended)
4713                 return;
4714
4715         dprintk("NFSD: end of grace period\n");
4716         nn->grace_ended = true;
4717         /*
4718          * If the server goes down again right now, an NFSv4
4719          * client will still be allowed to reclaim after it comes back up,
4720          * even if it hasn't yet had a chance to reclaim state this time.
4721          *
4722          */
4723         nfsd4_record_grace_done(nn);
4724         /*
4725          * At this point, NFSv4 clients can still reclaim.  But if the
4726          * server crashes, any that have not yet reclaimed will be out
4727          * of luck on the next boot.
4728          *
4729          * (NFSv4.1+ clients are considered to have reclaimed once they
4730          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
4731          * have reclaimed after their first OPEN.)
4732          */
4733         locks_end_grace(&nn->nfsd4_manager);
4734         /*
4735          * At this point, and once lockd and/or any other containers
4736          * exit their grace period, further reclaims will fail and
4737          * regular locking can resume.
4738          */
4739 }
4740
4741 /*
4742  * If we've waited a lease period but there are still clients trying to
4743  * reclaim, wait a little longer to give them a chance to finish.
4744  */
4745 static bool clients_still_reclaiming(struct nfsd_net *nn)
4746 {
4747         unsigned long now = get_seconds();
4748         unsigned long double_grace_period_end = nn->boot_time +
4749                                                 2 * nn->nfsd4_lease;
4750
4751         if (!nn->somebody_reclaimed)
4752                 return false;
4753         nn->somebody_reclaimed = false;
4754         /*
4755          * If we've given them *two* lease times to reclaim, and they're
4756          * still not done, give up:
4757          */
4758         if (time_after(now, double_grace_period_end))
4759                 return false;
4760         return true;
4761 }
4762
4763 static time_t
4764 nfs4_laundromat(struct nfsd_net *nn)
4765 {
4766         struct nfs4_client *clp;
4767         struct nfs4_openowner *oo;
4768         struct nfs4_delegation *dp;
4769         struct nfs4_ol_stateid *stp;
4770         struct nfsd4_blocked_lock *nbl;
4771         struct list_head *pos, *next, reaplist;
4772         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4773         time_t t, new_timeo = nn->nfsd4_lease;
4774
4775         dprintk("NFSD: laundromat service - starting\n");
4776
4777         if (clients_still_reclaiming(nn)) {
4778                 new_timeo = 0;
4779                 goto out;
4780         }
4781         nfsd4_end_grace(nn);
4782         INIT_LIST_HEAD(&reaplist);
4783         spin_lock(&nn->client_lock);
4784         list_for_each_safe(pos, next, &nn->client_lru) {
4785                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4786                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4787                         t = clp->cl_time - cutoff;
4788                         new_timeo = min(new_timeo, t);
4789                         break;
4790                 }
4791                 if (mark_client_expired_locked(clp)) {
4792                         dprintk("NFSD: client in use (clientid %08x)\n",
4793                                 clp->cl_clientid.cl_id);
4794                         continue;
4795                 }
4796                 list_add(&clp->cl_lru, &reaplist);
4797         }
4798         spin_unlock(&nn->client_lock);
4799         list_for_each_safe(pos, next, &reaplist) {
4800                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4801                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4802                         clp->cl_clientid.cl_id);
4803                 list_del_init(&clp->cl_lru);
4804                 expire_client(clp);
4805         }
4806         spin_lock(&state_lock);
4807         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4808                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4809                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4810                         t = dp->dl_time - cutoff;
4811                         new_timeo = min(new_timeo, t);
4812                         break;
4813                 }
4814                 WARN_ON(!unhash_delegation_locked(dp));
4815                 list_add(&dp->dl_recall_lru, &reaplist);
4816         }
4817         spin_unlock(&state_lock);
4818         while (!list_empty(&reaplist)) {
4819                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4820                                         dl_recall_lru);
4821                 list_del_init(&dp->dl_recall_lru);
4822                 revoke_delegation(dp);
4823         }
4824
4825         spin_lock(&nn->client_lock);
4826         while (!list_empty(&nn->close_lru)) {
4827                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4828                                         oo_close_lru);
4829                 if (time_after((unsigned long)oo->oo_time,
4830                                (unsigned long)cutoff)) {
4831                         t = oo->oo_time - cutoff;
4832                         new_timeo = min(new_timeo, t);
4833                         break;
4834                 }
4835                 list_del_init(&oo->oo_close_lru);
4836                 stp = oo->oo_last_closed_stid;
4837                 oo->oo_last_closed_stid = NULL;
4838                 spin_unlock(&nn->client_lock);
4839                 nfs4_put_stid(&stp->st_stid);
4840                 spin_lock(&nn->client_lock);
4841         }
4842         spin_unlock(&nn->client_lock);
4843
4844         /*
4845          * It's possible for a client to try and acquire an already held lock
4846          * that is being held for a long time, and then lose interest in it.
4847          * So, we clean out any un-revisited request after a lease period
4848          * under the assumption that the client is no longer interested.
4849          *
4850          * RFC5661, sec. 9.6 states that the client must not rely on getting
4851          * notifications and must continue to poll for locks, even when the
4852          * server supports them. Thus this shouldn't lead to clients blocking
4853          * indefinitely once the lock does become free.
4854          */
4855         BUG_ON(!list_empty(&reaplist));
4856         spin_lock(&nn->blocked_locks_lock);
4857         while (!list_empty(&nn->blocked_locks_lru)) {
4858                 nbl = list_first_entry(&nn->blocked_locks_lru,
4859                                         struct nfsd4_blocked_lock, nbl_lru);
4860                 if (time_after((unsigned long)nbl->nbl_time,
4861                                (unsigned long)cutoff)) {
4862                         t = nbl->nbl_time - cutoff;
4863                         new_timeo = min(new_timeo, t);
4864                         break;
4865                 }
4866                 list_move(&nbl->nbl_lru, &reaplist);
4867                 list_del_init(&nbl->nbl_list);
4868         }
4869         spin_unlock(&nn->blocked_locks_lock);
4870
4871         while (!list_empty(&reaplist)) {
4872                 nbl = list_first_entry(&reaplist,
4873                                         struct nfsd4_blocked_lock, nbl_lru);
4874                 list_del_init(&nbl->nbl_lru);
4875                 free_blocked_lock(nbl);
4876         }
4877 out:
4878         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4879         return new_timeo;
4880 }
4881
4882 static struct workqueue_struct *laundry_wq;
4883 static void laundromat_main(struct work_struct *);
4884
4885 static void
4886 laundromat_main(struct work_struct *laundry)
4887 {
4888         time_t t;
4889         struct delayed_work *dwork = to_delayed_work(laundry);
4890         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4891                                            laundromat_work);
4892
4893         t = nfs4_laundromat(nn);
4894         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4895         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4896 }
4897
4898 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4899 {
4900         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4901                 return nfserr_bad_stateid;
4902         return nfs_ok;
4903 }
4904
4905 static inline int
4906 access_permit_read(struct nfs4_ol_stateid *stp)
4907 {
4908         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4909                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4910                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4911 }
4912
4913 static inline int
4914 access_permit_write(struct nfs4_ol_stateid *stp)
4915 {
4916         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4917                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4918 }
4919
4920 static
4921 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4922 {
4923         __be32 status = nfserr_openmode;
4924
4925         /* For lock stateid's, we test the parent open, not the lock: */
4926         if (stp->st_openstp)
4927                 stp = stp->st_openstp;
4928         if ((flags & WR_STATE) && !access_permit_write(stp))
4929                 goto out;
4930         if ((flags & RD_STATE) && !access_permit_read(stp))
4931                 goto out;
4932         status = nfs_ok;
4933 out:
4934         return status;
4935 }
4936
4937 static inline __be32
4938 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4939 {
4940         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4941                 return nfs_ok;
4942         else if (opens_in_grace(net)) {
4943                 /* Answer in remaining cases depends on existence of
4944                  * conflicting state; so we must wait out the grace period. */
4945                 return nfserr_grace;
4946         } else if (flags & WR_STATE)
4947                 return nfs4_share_conflict(current_fh,
4948                                 NFS4_SHARE_DENY_WRITE);
4949         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4950                 return nfs4_share_conflict(current_fh,
4951                                 NFS4_SHARE_DENY_READ);
4952 }
4953
4954 /*
4955  * Allow READ/WRITE during grace period on recovered state only for files
4956  * that are not able to provide mandatory locking.
4957  */
4958 static inline int
4959 grace_disallows_io(struct net *net, struct inode *inode)
4960 {
4961         return opens_in_grace(net) && mandatory_lock(inode);
4962 }
4963
4964 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4965 {
4966         /*
4967          * When sessions are used the stateid generation number is ignored
4968          * when it is zero.
4969          */
4970         if (has_session && in->si_generation == 0)
4971                 return nfs_ok;
4972
4973         if (in->si_generation == ref->si_generation)
4974                 return nfs_ok;
4975
4976         /* If the client sends us a stateid from the future, it's buggy: */
4977         if (nfsd4_stateid_generation_after(in, ref))
4978                 return nfserr_bad_stateid;
4979         /*
4980          * However, we could see a stateid from the past, even from a
4981          * non-buggy client.  For example, if the client sends a lock
4982          * while some IO is outstanding, the lock may bump si_generation
4983          * while the IO is still in flight.  The client could avoid that
4984          * situation by waiting for responses on all the IO requests,
4985          * but better performance may result in retrying IO that
4986          * receives an old_stateid error if requests are rarely
4987          * reordered in flight:
4988          */
4989         return nfserr_old_stateid;
4990 }
4991
4992 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
4993 {
4994         __be32 ret;
4995
4996         spin_lock(&s->sc_lock);
4997         ret = nfsd4_verify_open_stid(s);
4998         if (ret == nfs_ok)
4999                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
5000         spin_unlock(&s->sc_lock);
5001         return ret;
5002 }
5003
5004 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
5005 {
5006         if (ols->st_stateowner->so_is_open_owner &&
5007             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
5008                 return nfserr_bad_stateid;
5009         return nfs_ok;
5010 }
5011
5012 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
5013 {
5014         struct nfs4_stid *s;
5015         __be32 status = nfserr_bad_stateid;
5016
5017         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5018                 CLOSE_STATEID(stateid))
5019                 return status;
5020         /* Client debugging aid. */
5021         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
5022                 char addr_str[INET6_ADDRSTRLEN];
5023                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
5024                                  sizeof(addr_str));
5025                 pr_warn_ratelimited("NFSD: client %s testing state ID "
5026                                         "with incorrect client ID\n", addr_str);
5027                 return status;
5028         }
5029         spin_lock(&cl->cl_lock);
5030         s = find_stateid_locked(cl, stateid);
5031         if (!s)
5032                 goto out_unlock;
5033         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
5034         if (status)
5035                 goto out_unlock;
5036         switch (s->sc_type) {
5037         case NFS4_DELEG_STID:
5038                 status = nfs_ok;
5039                 break;
5040         case NFS4_REVOKED_DELEG_STID:
5041                 status = nfserr_deleg_revoked;
5042                 break;
5043         case NFS4_OPEN_STID:
5044         case NFS4_LOCK_STID:
5045                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5046                 break;
5047         default:
5048                 printk("unknown stateid type %x\n", s->sc_type);
5049                 /* Fallthrough */
5050         case NFS4_CLOSED_STID:
5051         case NFS4_CLOSED_DELEG_STID:
5052                 status = nfserr_bad_stateid;
5053         }
5054 out_unlock:
5055         spin_unlock(&cl->cl_lock);
5056         return status;
5057 }
5058
5059 __be32
5060 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5061                      stateid_t *stateid, unsigned char typemask,
5062                      struct nfs4_stid **s, struct nfsd_net *nn)
5063 {
5064         __be32 status;
5065         bool return_revoked = false;
5066
5067         /*
5068          *  only return revoked delegations if explicitly asked.
5069          *  otherwise we report revoked or bad_stateid status.
5070          */
5071         if (typemask & NFS4_REVOKED_DELEG_STID)
5072                 return_revoked = true;
5073         else if (typemask & NFS4_DELEG_STID)
5074                 typemask |= NFS4_REVOKED_DELEG_STID;
5075
5076         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5077                 CLOSE_STATEID(stateid))
5078                 return nfserr_bad_stateid;
5079         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
5080         if (status == nfserr_stale_clientid) {
5081                 if (cstate->session)
5082                         return nfserr_bad_stateid;
5083                 return nfserr_stale_stateid;
5084         }
5085         if (status)
5086                 return status;
5087         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5088         if (!*s)
5089                 return nfserr_bad_stateid;
5090         if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5091                 nfs4_put_stid(*s);
5092                 if (cstate->minorversion)
5093                         return nfserr_deleg_revoked;
5094                 return nfserr_bad_stateid;
5095         }
5096         return nfs_ok;
5097 }
5098
5099 static struct file *
5100 nfs4_find_file(struct nfs4_stid *s, int flags)
5101 {
5102         if (!s)
5103                 return NULL;
5104
5105         switch (s->sc_type) {
5106         case NFS4_DELEG_STID:
5107                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5108                         return NULL;
5109                 return get_file(s->sc_file->fi_deleg_file);
5110         case NFS4_OPEN_STID:
5111         case NFS4_LOCK_STID:
5112                 if (flags & RD_STATE)
5113                         return find_readable_file(s->sc_file);
5114                 else
5115                         return find_writeable_file(s->sc_file);
5116                 break;
5117         }
5118
5119         return NULL;
5120 }
5121
5122 static __be32
5123 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
5124 {
5125         __be32 status;
5126
5127         status = nfsd4_check_openowner_confirmed(ols);
5128         if (status)
5129                 return status;
5130         return nfs4_check_openmode(ols, flags);
5131 }
5132
5133 static __be32
5134 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5135                 struct file **filpp, bool *tmp_file, int flags)
5136 {
5137         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5138         struct file *file;
5139         __be32 status;
5140
5141         file = nfs4_find_file(s, flags);
5142         if (file) {
5143                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5144                                 acc | NFSD_MAY_OWNER_OVERRIDE);
5145                 if (status) {
5146                         fput(file);
5147                         return status;
5148                 }
5149
5150                 *filpp = file;
5151         } else {
5152                 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
5153                 if (status)
5154                         return status;
5155
5156                 if (tmp_file)
5157                         *tmp_file = true;
5158         }
5159
5160         return 0;
5161 }
5162
5163 /*
5164  * Checks for stateid operations
5165  */
5166 __be32
5167 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5168                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5169                 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
5170 {
5171         struct inode *ino = d_inode(fhp->fh_dentry);
5172         struct net *net = SVC_NET(rqstp);
5173         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5174         struct nfs4_stid *s = NULL;
5175         __be32 status;
5176
5177         if (filpp)
5178                 *filpp = NULL;
5179         if (tmp_file)
5180                 *tmp_file = false;
5181
5182         if (grace_disallows_io(net, ino))
5183                 return nfserr_grace;
5184
5185         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5186                 status = check_special_stateids(net, fhp, stateid, flags);
5187                 goto done;
5188         }
5189
5190         status = nfsd4_lookup_stateid(cstate, stateid,
5191                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5192                                 &s, nn);
5193         if (status)
5194                 return status;
5195         status = nfsd4_stid_check_stateid_generation(stateid, s,
5196                         nfsd4_has_session(cstate));
5197         if (status)
5198                 goto out;
5199
5200         switch (s->sc_type) {
5201         case NFS4_DELEG_STID:
5202                 status = nfs4_check_delegmode(delegstateid(s), flags);
5203                 break;
5204         case NFS4_OPEN_STID:
5205         case NFS4_LOCK_STID:
5206                 status = nfs4_check_olstateid(openlockstateid(s), flags);
5207                 break;
5208         default:
5209                 status = nfserr_bad_stateid;
5210                 break;
5211         }
5212         if (status)
5213                 goto out;
5214         status = nfs4_check_fh(fhp, s);
5215
5216 done:
5217         if (!status && filpp)
5218                 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
5219 out:
5220         if (s)
5221                 nfs4_put_stid(s);
5222         return status;
5223 }
5224
5225 /*
5226  * Test if the stateid is valid
5227  */
5228 __be32
5229 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5230                    union nfsd4_op_u *u)
5231 {
5232         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5233         struct nfsd4_test_stateid_id *stateid;
5234         struct nfs4_client *cl = cstate->session->se_client;
5235
5236         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5237                 stateid->ts_id_status =
5238                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5239
5240         return nfs_ok;
5241 }
5242
5243 static __be32
5244 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5245 {
5246         struct nfs4_ol_stateid *stp = openlockstateid(s);
5247         __be32 ret;
5248
5249         ret = nfsd4_lock_ol_stateid(stp);
5250         if (ret)
5251                 goto out_put_stid;
5252
5253         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5254         if (ret)
5255                 goto out;
5256
5257         ret = nfserr_locks_held;
5258         if (check_for_locks(stp->st_stid.sc_file,
5259                             lockowner(stp->st_stateowner)))
5260                 goto out;
5261
5262         release_lock_stateid(stp);
5263         ret = nfs_ok;
5264
5265 out:
5266         mutex_unlock(&stp->st_mutex);
5267 out_put_stid:
5268         nfs4_put_stid(s);
5269         return ret;
5270 }
5271
5272 __be32
5273 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5274                    union nfsd4_op_u *u)
5275 {
5276         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5277         stateid_t *stateid = &free_stateid->fr_stateid;
5278         struct nfs4_stid *s;
5279         struct nfs4_delegation *dp;
5280         struct nfs4_client *cl = cstate->session->se_client;
5281         __be32 ret = nfserr_bad_stateid;
5282
5283         spin_lock(&cl->cl_lock);
5284         s = find_stateid_locked(cl, stateid);
5285         if (!s)
5286                 goto out_unlock;
5287         spin_lock(&s->sc_lock);
5288         switch (s->sc_type) {
5289         case NFS4_DELEG_STID:
5290                 ret = nfserr_locks_held;
5291                 break;
5292         case NFS4_OPEN_STID:
5293                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5294                 if (ret)
5295                         break;
5296                 ret = nfserr_locks_held;
5297                 break;
5298         case NFS4_LOCK_STID:
5299                 spin_unlock(&s->sc_lock);
5300                 refcount_inc(&s->sc_count);
5301                 spin_unlock(&cl->cl_lock);
5302                 ret = nfsd4_free_lock_stateid(stateid, s);
5303                 goto out;
5304         case NFS4_REVOKED_DELEG_STID:
5305                 spin_unlock(&s->sc_lock);
5306                 dp = delegstateid(s);
5307                 list_del_init(&dp->dl_recall_lru);
5308                 spin_unlock(&cl->cl_lock);
5309                 nfs4_put_stid(s);
5310                 ret = nfs_ok;
5311                 goto out;
5312         /* Default falls through and returns nfserr_bad_stateid */
5313         }
5314         spin_unlock(&s->sc_lock);
5315 out_unlock:
5316         spin_unlock(&cl->cl_lock);
5317 out:
5318         return ret;
5319 }
5320
5321 static inline int
5322 setlkflg (int type)
5323 {
5324         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5325                 RD_STATE : WR_STATE;
5326 }
5327
5328 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5329 {
5330         struct svc_fh *current_fh = &cstate->current_fh;
5331         struct nfs4_stateowner *sop = stp->st_stateowner;
5332         __be32 status;
5333
5334         status = nfsd4_check_seqid(cstate, sop, seqid);
5335         if (status)
5336                 return status;
5337         status = nfsd4_lock_ol_stateid(stp);
5338         if (status != nfs_ok)
5339                 return status;
5340         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5341         if (status == nfs_ok)
5342                 status = nfs4_check_fh(current_fh, &stp->st_stid);
5343         if (status != nfs_ok)
5344                 mutex_unlock(&stp->st_mutex);
5345         return status;
5346 }
5347
5348 /*
5349  * Checks for sequence id mutating operations.
5350  */
5351 static __be32
5352 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5353                          stateid_t *stateid, char typemask,
5354                          struct nfs4_ol_stateid **stpp,
5355                          struct nfsd_net *nn)
5356 {
5357         __be32 status;
5358         struct nfs4_stid *s;
5359         struct nfs4_ol_stateid *stp = NULL;
5360
5361         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5362                 seqid, STATEID_VAL(stateid));
5363
5364         *stpp = NULL;
5365         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5366         if (status)
5367                 return status;
5368         stp = openlockstateid(s);
5369         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5370
5371         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5372         if (!status)
5373                 *stpp = stp;
5374         else
5375                 nfs4_put_stid(&stp->st_stid);
5376         return status;
5377 }
5378
5379 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5380                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5381 {
5382         __be32 status;
5383         struct nfs4_openowner *oo;
5384         struct nfs4_ol_stateid *stp;
5385
5386         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5387                                                 NFS4_OPEN_STID, &stp, nn);
5388         if (status)
5389                 return status;
5390         oo = openowner(stp->st_stateowner);
5391         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5392                 mutex_unlock(&stp->st_mutex);
5393                 nfs4_put_stid(&stp->st_stid);
5394                 return nfserr_bad_stateid;
5395         }
5396         *stpp = stp;
5397         return nfs_ok;
5398 }
5399
5400 __be32
5401 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5402                    union nfsd4_op_u *u)
5403 {
5404         struct nfsd4_open_confirm *oc = &u->open_confirm;
5405         __be32 status;
5406         struct nfs4_openowner *oo;
5407         struct nfs4_ol_stateid *stp;
5408         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5409
5410         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5411                         cstate->current_fh.fh_dentry);
5412
5413         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5414         if (status)
5415                 return status;
5416
5417         status = nfs4_preprocess_seqid_op(cstate,
5418                                         oc->oc_seqid, &oc->oc_req_stateid,
5419                                         NFS4_OPEN_STID, &stp, nn);
5420         if (status)
5421                 goto out;
5422         oo = openowner(stp->st_stateowner);
5423         status = nfserr_bad_stateid;
5424         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5425                 mutex_unlock(&stp->st_mutex);
5426                 goto put_stateid;
5427         }
5428         oo->oo_flags |= NFS4_OO_CONFIRMED;
5429         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5430         mutex_unlock(&stp->st_mutex);
5431         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5432                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5433
5434         nfsd4_client_record_create(oo->oo_owner.so_client);
5435         status = nfs_ok;
5436 put_stateid:
5437         nfs4_put_stid(&stp->st_stid);
5438 out:
5439         nfsd4_bump_seqid(cstate, status);
5440         return status;
5441 }
5442
5443 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5444 {
5445         if (!test_access(access, stp))
5446                 return;
5447         nfs4_file_put_access(stp->st_stid.sc_file, access);
5448         clear_access(access, stp);
5449 }
5450
5451 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5452 {
5453         switch (to_access) {
5454         case NFS4_SHARE_ACCESS_READ:
5455                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5456                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5457                 break;
5458         case NFS4_SHARE_ACCESS_WRITE:
5459                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5460                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5461                 break;
5462         case NFS4_SHARE_ACCESS_BOTH:
5463                 break;
5464         default:
5465                 WARN_ON_ONCE(1);
5466         }
5467 }
5468
5469 __be32
5470 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5471                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
5472 {
5473         struct nfsd4_open_downgrade *od = &u->open_downgrade;
5474         __be32 status;
5475         struct nfs4_ol_stateid *stp;
5476         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5477
5478         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5479                         cstate->current_fh.fh_dentry);
5480
5481         /* We don't yet support WANT bits: */
5482         if (od->od_deleg_want)
5483                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5484                         od->od_deleg_want);
5485
5486         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5487                                         &od->od_stateid, &stp, nn);
5488         if (status)
5489                 goto out;
5490         status = nfserr_inval;
5491         if (!test_access(od->od_share_access, stp)) {
5492                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5493                         stp->st_access_bmap, od->od_share_access);
5494                 goto put_stateid;
5495         }
5496         if (!test_deny(od->od_share_deny, stp)) {
5497                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5498                         stp->st_deny_bmap, od->od_share_deny);
5499                 goto put_stateid;
5500         }
5501         nfs4_stateid_downgrade(stp, od->od_share_access);
5502         reset_union_bmap_deny(od->od_share_deny, stp);
5503         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5504         status = nfs_ok;
5505 put_stateid:
5506         mutex_unlock(&stp->st_mutex);
5507         nfs4_put_stid(&stp->st_stid);
5508 out:
5509         nfsd4_bump_seqid(cstate, status);
5510         return status;
5511 }
5512
5513 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5514 {
5515         struct nfs4_client *clp = s->st_stid.sc_client;
5516         bool unhashed;
5517         LIST_HEAD(reaplist);
5518
5519         spin_lock(&clp->cl_lock);
5520         unhashed = unhash_open_stateid(s, &reaplist);
5521
5522         if (clp->cl_minorversion) {
5523                 if (unhashed)
5524                         put_ol_stateid_locked(s, &reaplist);
5525                 spin_unlock(&clp->cl_lock);
5526                 free_ol_stateid_reaplist(&reaplist);
5527         } else {
5528                 spin_unlock(&clp->cl_lock);
5529                 free_ol_stateid_reaplist(&reaplist);
5530                 if (unhashed)
5531                         move_to_close_lru(s, clp->net);
5532         }
5533 }
5534
5535 /*
5536  * nfs4_unlock_state() called after encode
5537  */
5538 __be32
5539 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5540                 union nfsd4_op_u *u)
5541 {
5542         struct nfsd4_close *close = &u->close;
5543         __be32 status;
5544         struct nfs4_ol_stateid *stp;
5545         struct net *net = SVC_NET(rqstp);
5546         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5547
5548         dprintk("NFSD: nfsd4_close on file %pd\n",
5549                         cstate->current_fh.fh_dentry);
5550
5551         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5552                                         &close->cl_stateid,
5553                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
5554                                         &stp, nn);
5555         nfsd4_bump_seqid(cstate, status);
5556         if (status)
5557                 goto out;
5558
5559         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5560
5561         /*
5562          * Technically we don't _really_ have to increment or copy it, since
5563          * it should just be gone after this operation and we clobber the
5564          * copied value below, but we continue to do so here just to ensure
5565          * that racing ops see that there was a state change.
5566          */
5567         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5568
5569         nfsd4_close_open_stateid(stp);
5570         mutex_unlock(&stp->st_mutex);
5571
5572         /* v4.1+ suggests that we send a special stateid in here, since the
5573          * clients should just ignore this anyway. Since this is not useful
5574          * for v4.0 clients either, we set it to the special close_stateid
5575          * universally.
5576          *
5577          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
5578          */
5579         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
5580
5581         /* put reference from nfs4_preprocess_seqid_op */
5582         nfs4_put_stid(&stp->st_stid);
5583 out:
5584         return status;
5585 }
5586
5587 __be32
5588 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5589                   union nfsd4_op_u *u)
5590 {
5591         struct nfsd4_delegreturn *dr = &u->delegreturn;
5592         struct nfs4_delegation *dp;
5593         stateid_t *stateid = &dr->dr_stateid;
5594         struct nfs4_stid *s;
5595         __be32 status;
5596         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5597
5598         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5599                 return status;
5600
5601         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5602         if (status)
5603                 goto out;
5604         dp = delegstateid(s);
5605         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
5606         if (status)
5607                 goto put_stateid;
5608
5609         destroy_delegation(dp);
5610 put_stateid:
5611         nfs4_put_stid(&dp->dl_stid);
5612 out:
5613         return status;
5614 }
5615
5616 static inline u64
5617 end_offset(u64 start, u64 len)
5618 {
5619         u64 end;
5620
5621         end = start + len;
5622         return end >= start ? end: NFS4_MAX_UINT64;
5623 }
5624
5625 /* last octet in a range */
5626 static inline u64
5627 last_byte_offset(u64 start, u64 len)
5628 {
5629         u64 end;
5630
5631         WARN_ON_ONCE(!len);
5632         end = start + len;
5633         return end > start ? end - 1: NFS4_MAX_UINT64;
5634 }
5635
5636 /*
5637  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5638  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5639  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
5640  * locking, this prevents us from being completely protocol-compliant.  The
5641  * real solution to this problem is to start using unsigned file offsets in
5642  * the VFS, but this is a very deep change!
5643  */
5644 static inline void
5645 nfs4_transform_lock_offset(struct file_lock *lock)
5646 {
5647         if (lock->fl_start < 0)
5648                 lock->fl_start = OFFSET_MAX;
5649         if (lock->fl_end < 0)
5650                 lock->fl_end = OFFSET_MAX;
5651 }
5652
5653 static fl_owner_t
5654 nfsd4_fl_get_owner(fl_owner_t owner)
5655 {
5656         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5657
5658         nfs4_get_stateowner(&lo->lo_owner);
5659         return owner;
5660 }
5661
5662 static void
5663 nfsd4_fl_put_owner(fl_owner_t owner)
5664 {
5665         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5666
5667         if (lo)
5668                 nfs4_put_stateowner(&lo->lo_owner);
5669 }
5670
5671 static void
5672 nfsd4_lm_notify(struct file_lock *fl)
5673 {
5674         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
5675         struct net                      *net = lo->lo_owner.so_client->net;
5676         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
5677         struct nfsd4_blocked_lock       *nbl = container_of(fl,
5678                                                 struct nfsd4_blocked_lock, nbl_lock);
5679         bool queue = false;
5680
5681         /* An empty list means that something else is going to be using it */
5682         spin_lock(&nn->blocked_locks_lock);
5683         if (!list_empty(&nbl->nbl_list)) {
5684                 list_del_init(&nbl->nbl_list);
5685                 list_del_init(&nbl->nbl_lru);
5686                 queue = true;
5687         }
5688         spin_unlock(&nn->blocked_locks_lock);
5689
5690         if (queue)
5691                 nfsd4_run_cb(&nbl->nbl_cb);
5692 }
5693
5694 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
5695         .lm_notify = nfsd4_lm_notify,
5696         .lm_get_owner = nfsd4_fl_get_owner,
5697         .lm_put_owner = nfsd4_fl_put_owner,
5698 };
5699
5700 static inline void
5701 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5702 {
5703         struct nfs4_lockowner *lo;
5704
5705         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5706                 lo = (struct nfs4_lockowner *) fl->fl_owner;
5707                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5708                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
5709                 if (!deny->ld_owner.data)
5710                         /* We just don't care that much */
5711                         goto nevermind;
5712                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5713                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5714         } else {
5715 nevermind:
5716                 deny->ld_owner.len = 0;
5717                 deny->ld_owner.data = NULL;
5718                 deny->ld_clientid.cl_boot = 0;
5719                 deny->ld_clientid.cl_id = 0;
5720         }
5721         deny->ld_start = fl->fl_start;
5722         deny->ld_length = NFS4_MAX_UINT64;
5723         if (fl->fl_end != NFS4_MAX_UINT64)
5724                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5725         deny->ld_type = NFS4_READ_LT;
5726         if (fl->fl_type != F_RDLCK)
5727                 deny->ld_type = NFS4_WRITE_LT;
5728 }
5729
5730 static struct nfs4_lockowner *
5731 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5732 {
5733         unsigned int strhashval = ownerstr_hashval(owner);
5734         struct nfs4_stateowner *so;
5735
5736         lockdep_assert_held(&clp->cl_lock);
5737
5738         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5739                             so_strhash) {
5740                 if (so->so_is_open_owner)
5741                         continue;
5742                 if (same_owner_str(so, owner))
5743                         return lockowner(nfs4_get_stateowner(so));
5744         }
5745         return NULL;
5746 }
5747
5748 static struct nfs4_lockowner *
5749 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5750 {
5751         struct nfs4_lockowner *lo;
5752
5753         spin_lock(&clp->cl_lock);
5754         lo = find_lockowner_str_locked(clp, owner);
5755         spin_unlock(&clp->cl_lock);
5756         return lo;
5757 }
5758
5759 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5760 {
5761         unhash_lockowner_locked(lockowner(sop));
5762 }
5763
5764 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5765 {
5766         struct nfs4_lockowner *lo = lockowner(sop);
5767
5768         kmem_cache_free(lockowner_slab, lo);
5769 }
5770
5771 static const struct nfs4_stateowner_operations lockowner_ops = {
5772         .so_unhash =    nfs4_unhash_lockowner,
5773         .so_free =      nfs4_free_lockowner,
5774 };
5775
5776 /*
5777  * Alloc a lock owner structure.
5778  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5779  * occurred.
5780  *
5781  * strhashval = ownerstr_hashval
5782  */
5783 static struct nfs4_lockowner *
5784 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5785                            struct nfs4_ol_stateid *open_stp,
5786                            struct nfsd4_lock *lock)
5787 {
5788         struct nfs4_lockowner *lo, *ret;
5789
5790         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5791         if (!lo)
5792                 return NULL;
5793         INIT_LIST_HEAD(&lo->lo_blocked);
5794         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5795         lo->lo_owner.so_is_open_owner = 0;
5796         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5797         lo->lo_owner.so_ops = &lockowner_ops;
5798         spin_lock(&clp->cl_lock);
5799         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5800         if (ret == NULL) {
5801                 list_add(&lo->lo_owner.so_strhash,
5802                          &clp->cl_ownerstr_hashtbl[strhashval]);
5803                 ret = lo;
5804         } else
5805                 nfs4_free_stateowner(&lo->lo_owner);
5806
5807         spin_unlock(&clp->cl_lock);
5808         return ret;
5809 }
5810
5811 static struct nfs4_ol_stateid *
5812 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5813 {
5814         struct nfs4_ol_stateid *lst;
5815         struct nfs4_client *clp = lo->lo_owner.so_client;
5816
5817         lockdep_assert_held(&clp->cl_lock);
5818
5819         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5820                 if (lst->st_stid.sc_type != NFS4_LOCK_STID)
5821                         continue;
5822                 if (lst->st_stid.sc_file == fp) {
5823                         refcount_inc(&lst->st_stid.sc_count);
5824                         return lst;
5825                 }
5826         }
5827         return NULL;
5828 }
5829
5830 static struct nfs4_ol_stateid *
5831 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5832                   struct nfs4_file *fp, struct inode *inode,
5833                   struct nfs4_ol_stateid *open_stp)
5834 {
5835         struct nfs4_client *clp = lo->lo_owner.so_client;
5836         struct nfs4_ol_stateid *retstp;
5837
5838         mutex_init(&stp->st_mutex);
5839         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
5840 retry:
5841         spin_lock(&clp->cl_lock);
5842         spin_lock(&fp->fi_lock);
5843         retstp = find_lock_stateid(lo, fp);
5844         if (retstp)
5845                 goto out_unlock;
5846
5847         refcount_inc(&stp->st_stid.sc_count);
5848         stp->st_stid.sc_type = NFS4_LOCK_STID;
5849         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5850         get_nfs4_file(fp);
5851         stp->st_stid.sc_file = fp;
5852         stp->st_access_bmap = 0;
5853         stp->st_deny_bmap = open_stp->st_deny_bmap;
5854         stp->st_openstp = open_stp;
5855         list_add(&stp->st_locks, &open_stp->st_locks);
5856         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5857         list_add(&stp->st_perfile, &fp->fi_stateids);
5858 out_unlock:
5859         spin_unlock(&fp->fi_lock);
5860         spin_unlock(&clp->cl_lock);
5861         if (retstp) {
5862                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
5863                         nfs4_put_stid(&retstp->st_stid);
5864                         goto retry;
5865                 }
5866                 /* To keep mutex tracking happy */
5867                 mutex_unlock(&stp->st_mutex);
5868                 stp = retstp;
5869         }
5870         return stp;
5871 }
5872
5873 static struct nfs4_ol_stateid *
5874 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5875                             struct inode *inode, struct nfs4_ol_stateid *ost,
5876                             bool *new)
5877 {
5878         struct nfs4_stid *ns = NULL;
5879         struct nfs4_ol_stateid *lst;
5880         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5881         struct nfs4_client *clp = oo->oo_owner.so_client;
5882
5883         *new = false;
5884         spin_lock(&clp->cl_lock);
5885         lst = find_lock_stateid(lo, fi);
5886         spin_unlock(&clp->cl_lock);
5887         if (lst != NULL) {
5888                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
5889                         goto out;
5890                 nfs4_put_stid(&lst->st_stid);
5891         }
5892         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
5893         if (ns == NULL)
5894                 return NULL;
5895
5896         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
5897         if (lst == openlockstateid(ns))
5898                 *new = true;
5899         else
5900                 nfs4_put_stid(ns);
5901 out:
5902         return lst;
5903 }
5904
5905 static int
5906 check_lock_length(u64 offset, u64 length)
5907 {
5908         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5909                 (length > ~offset)));
5910 }
5911
5912 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5913 {
5914         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5915
5916         lockdep_assert_held(&fp->fi_lock);
5917
5918         if (test_access(access, lock_stp))
5919                 return;
5920         __nfs4_file_get_access(fp, access);
5921         set_access(access, lock_stp);
5922 }
5923
5924 static __be32
5925 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5926                             struct nfs4_ol_stateid *ost,
5927                             struct nfsd4_lock *lock,
5928                             struct nfs4_ol_stateid **plst, bool *new)
5929 {
5930         __be32 status;
5931         struct nfs4_file *fi = ost->st_stid.sc_file;
5932         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5933         struct nfs4_client *cl = oo->oo_owner.so_client;
5934         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5935         struct nfs4_lockowner *lo;
5936         struct nfs4_ol_stateid *lst;
5937         unsigned int strhashval;
5938
5939         lo = find_lockowner_str(cl, &lock->lk_new_owner);
5940         if (!lo) {
5941                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5942                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5943                 if (lo == NULL)
5944                         return nfserr_jukebox;
5945         } else {
5946                 /* with an existing lockowner, seqids must be the same */
5947                 status = nfserr_bad_seqid;
5948                 if (!cstate->minorversion &&
5949                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5950                         goto out;
5951         }
5952
5953         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5954         if (lst == NULL) {
5955                 status = nfserr_jukebox;
5956                 goto out;
5957         }
5958
5959         status = nfs_ok;
5960         *plst = lst;
5961 out:
5962         nfs4_put_stateowner(&lo->lo_owner);
5963         return status;
5964 }
5965
5966 /*
5967  *  LOCK operation
5968  */
5969 __be32
5970 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5971            union nfsd4_op_u *u)
5972 {
5973         struct nfsd4_lock *lock = &u->lock;
5974         struct nfs4_openowner *open_sop = NULL;
5975         struct nfs4_lockowner *lock_sop = NULL;
5976         struct nfs4_ol_stateid *lock_stp = NULL;
5977         struct nfs4_ol_stateid *open_stp = NULL;
5978         struct nfs4_file *fp;
5979         struct file *filp = NULL;
5980         struct nfsd4_blocked_lock *nbl = NULL;
5981         struct file_lock *file_lock = NULL;
5982         struct file_lock *conflock = NULL;
5983         __be32 status = 0;
5984         int lkflg;
5985         int err;
5986         bool new = false;
5987         unsigned char fl_type;
5988         unsigned int fl_flags = FL_POSIX;
5989         struct net *net = SVC_NET(rqstp);
5990         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5991
5992         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5993                 (long long) lock->lk_offset,
5994                 (long long) lock->lk_length);
5995
5996         if (check_lock_length(lock->lk_offset, lock->lk_length))
5997                  return nfserr_inval;
5998
5999         if ((status = fh_verify(rqstp, &cstate->current_fh,
6000                                 S_IFREG, NFSD_MAY_LOCK))) {
6001                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
6002                 return status;
6003         }
6004
6005         if (lock->lk_is_new) {
6006                 if (nfsd4_has_session(cstate))
6007                         /* See rfc 5661 18.10.3: given clientid is ignored: */
6008                         memcpy(&lock->lk_new_clientid,
6009                                 &cstate->session->se_client->cl_clientid,
6010                                 sizeof(clientid_t));
6011
6012                 status = nfserr_stale_clientid;
6013                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
6014                         goto out;
6015
6016                 /* validate and update open stateid and open seqid */
6017                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
6018                                         lock->lk_new_open_seqid,
6019                                         &lock->lk_new_open_stateid,
6020                                         &open_stp, nn);
6021                 if (status)
6022                         goto out;
6023                 mutex_unlock(&open_stp->st_mutex);
6024                 open_sop = openowner(open_stp->st_stateowner);
6025                 status = nfserr_bad_stateid;
6026                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
6027                                                 &lock->lk_new_clientid))
6028                         goto out;
6029                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
6030                                                         &lock_stp, &new);
6031         } else {
6032                 status = nfs4_preprocess_seqid_op(cstate,
6033                                        lock->lk_old_lock_seqid,
6034                                        &lock->lk_old_lock_stateid,
6035                                        NFS4_LOCK_STID, &lock_stp, nn);
6036         }
6037         if (status)
6038                 goto out;
6039         lock_sop = lockowner(lock_stp->st_stateowner);
6040
6041         lkflg = setlkflg(lock->lk_type);
6042         status = nfs4_check_openmode(lock_stp, lkflg);
6043         if (status)
6044                 goto out;
6045
6046         status = nfserr_grace;
6047         if (locks_in_grace(net) && !lock->lk_reclaim)
6048                 goto out;
6049         status = nfserr_no_grace;
6050         if (!locks_in_grace(net) && lock->lk_reclaim)
6051                 goto out;
6052
6053         fp = lock_stp->st_stid.sc_file;
6054         switch (lock->lk_type) {
6055                 case NFS4_READW_LT:
6056                         if (nfsd4_has_session(cstate))
6057                                 fl_flags |= FL_SLEEP;
6058                         /* Fallthrough */
6059                 case NFS4_READ_LT:
6060                         spin_lock(&fp->fi_lock);
6061                         filp = find_readable_file_locked(fp);
6062                         if (filp)
6063                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
6064                         spin_unlock(&fp->fi_lock);
6065                         fl_type = F_RDLCK;
6066                         break;
6067                 case NFS4_WRITEW_LT:
6068                         if (nfsd4_has_session(cstate))
6069                                 fl_flags |= FL_SLEEP;
6070                         /* Fallthrough */
6071                 case NFS4_WRITE_LT:
6072                         spin_lock(&fp->fi_lock);
6073                         filp = find_writeable_file_locked(fp);
6074                         if (filp)
6075                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6076                         spin_unlock(&fp->fi_lock);
6077                         fl_type = F_WRLCK;
6078                         break;
6079                 default:
6080                         status = nfserr_inval;
6081                 goto out;
6082         }
6083
6084         if (!filp) {
6085                 status = nfserr_openmode;
6086                 goto out;
6087         }
6088
6089         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6090         if (!nbl) {
6091                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6092                 status = nfserr_jukebox;
6093                 goto out;
6094         }
6095
6096         file_lock = &nbl->nbl_lock;
6097         file_lock->fl_type = fl_type;
6098         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6099         file_lock->fl_pid = current->tgid;
6100         file_lock->fl_file = filp;
6101         file_lock->fl_flags = fl_flags;
6102         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6103         file_lock->fl_start = lock->lk_offset;
6104         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6105         nfs4_transform_lock_offset(file_lock);
6106
6107         conflock = locks_alloc_lock();
6108         if (!conflock) {
6109                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6110                 status = nfserr_jukebox;
6111                 goto out;
6112         }
6113
6114         if (fl_flags & FL_SLEEP) {
6115                 nbl->nbl_time = jiffies;
6116                 spin_lock(&nn->blocked_locks_lock);
6117                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6118                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6119                 spin_unlock(&nn->blocked_locks_lock);
6120         }
6121
6122         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
6123         switch (err) {
6124         case 0: /* success! */
6125                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6126                 status = 0;
6127                 if (lock->lk_reclaim)
6128                         nn->somebody_reclaimed = true;
6129                 break;
6130         case FILE_LOCK_DEFERRED:
6131                 nbl = NULL;
6132                 /* Fallthrough */
6133         case -EAGAIN:           /* conflock holds conflicting lock */
6134                 status = nfserr_denied;
6135                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6136                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6137                 break;
6138         case -EDEADLK:
6139                 status = nfserr_deadlock;
6140                 break;
6141         default:
6142                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6143                 status = nfserrno(err);
6144                 break;
6145         }
6146 out:
6147         if (nbl) {
6148                 /* dequeue it if we queued it before */
6149                 if (fl_flags & FL_SLEEP) {
6150                         spin_lock(&nn->blocked_locks_lock);
6151                         list_del_init(&nbl->nbl_list);
6152                         list_del_init(&nbl->nbl_lru);
6153                         spin_unlock(&nn->blocked_locks_lock);
6154                 }
6155                 free_blocked_lock(nbl);
6156         }
6157         if (filp)
6158                 fput(filp);
6159         if (lock_stp) {
6160                 /* Bump seqid manually if the 4.0 replay owner is openowner */
6161                 if (cstate->replay_owner &&
6162                     cstate->replay_owner != &lock_sop->lo_owner &&
6163                     seqid_mutating_err(ntohl(status)))
6164                         lock_sop->lo_owner.so_seqid++;
6165
6166                 /*
6167                  * If this is a new, never-before-used stateid, and we are
6168                  * returning an error, then just go ahead and release it.
6169                  */
6170                 if (status && new)
6171                         release_lock_stateid(lock_stp);
6172
6173                 mutex_unlock(&lock_stp->st_mutex);
6174
6175                 nfs4_put_stid(&lock_stp->st_stid);
6176         }
6177         if (open_stp)
6178                 nfs4_put_stid(&open_stp->st_stid);
6179         nfsd4_bump_seqid(cstate, status);
6180         if (conflock)
6181                 locks_free_lock(conflock);
6182         return status;
6183 }
6184
6185 /*
6186  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6187  * so we do a temporary open here just to get an open file to pass to
6188  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
6189  * inode operation.)
6190  */
6191 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6192 {
6193         struct file *file;
6194         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
6195         if (!err) {
6196                 err = nfserrno(vfs_test_lock(file, lock));
6197                 fput(file);
6198         }
6199         return err;
6200 }
6201
6202 /*
6203  * LOCKT operation
6204  */
6205 __be32
6206 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6207             union nfsd4_op_u *u)
6208 {
6209         struct nfsd4_lockt *lockt = &u->lockt;
6210         struct file_lock *file_lock = NULL;
6211         struct nfs4_lockowner *lo = NULL;
6212         __be32 status;
6213         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6214
6215         if (locks_in_grace(SVC_NET(rqstp)))
6216                 return nfserr_grace;
6217
6218         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6219                  return nfserr_inval;
6220
6221         if (!nfsd4_has_session(cstate)) {
6222                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
6223                 if (status)
6224                         goto out;
6225         }
6226
6227         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6228                 goto out;
6229
6230         file_lock = locks_alloc_lock();
6231         if (!file_lock) {
6232                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6233                 status = nfserr_jukebox;
6234                 goto out;
6235         }
6236
6237         switch (lockt->lt_type) {
6238                 case NFS4_READ_LT:
6239                 case NFS4_READW_LT:
6240                         file_lock->fl_type = F_RDLCK;
6241                         break;
6242                 case NFS4_WRITE_LT:
6243                 case NFS4_WRITEW_LT:
6244                         file_lock->fl_type = F_WRLCK;
6245                         break;
6246                 default:
6247                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6248                         status = nfserr_inval;
6249                         goto out;
6250         }
6251
6252         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6253         if (lo)
6254                 file_lock->fl_owner = (fl_owner_t)lo;
6255         file_lock->fl_pid = current->tgid;
6256         file_lock->fl_flags = FL_POSIX;
6257
6258         file_lock->fl_start = lockt->lt_offset;
6259         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6260
6261         nfs4_transform_lock_offset(file_lock);
6262
6263         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6264         if (status)
6265                 goto out;
6266
6267         if (file_lock->fl_type != F_UNLCK) {
6268                 status = nfserr_denied;
6269                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6270         }
6271 out:
6272         if (lo)
6273                 nfs4_put_stateowner(&lo->lo_owner);
6274         if (file_lock)
6275                 locks_free_lock(file_lock);
6276         return status;
6277 }
6278
6279 __be32
6280 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6281             union nfsd4_op_u *u)
6282 {
6283         struct nfsd4_locku *locku = &u->locku;
6284         struct nfs4_ol_stateid *stp;
6285         struct file *filp = NULL;
6286         struct file_lock *file_lock = NULL;
6287         __be32 status;
6288         int err;
6289         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6290
6291         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6292                 (long long) locku->lu_offset,
6293                 (long long) locku->lu_length);
6294
6295         if (check_lock_length(locku->lu_offset, locku->lu_length))
6296                  return nfserr_inval;
6297
6298         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6299                                         &locku->lu_stateid, NFS4_LOCK_STID,
6300                                         &stp, nn);
6301         if (status)
6302                 goto out;
6303         filp = find_any_file(stp->st_stid.sc_file);
6304         if (!filp) {
6305                 status = nfserr_lock_range;
6306                 goto put_stateid;
6307         }
6308         file_lock = locks_alloc_lock();
6309         if (!file_lock) {
6310                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6311                 status = nfserr_jukebox;
6312                 goto fput;
6313         }
6314
6315         file_lock->fl_type = F_UNLCK;
6316         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6317         file_lock->fl_pid = current->tgid;
6318         file_lock->fl_file = filp;
6319         file_lock->fl_flags = FL_POSIX;
6320         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6321         file_lock->fl_start = locku->lu_offset;
6322
6323         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6324                                                 locku->lu_length);
6325         nfs4_transform_lock_offset(file_lock);
6326
6327         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6328         if (err) {
6329                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6330                 goto out_nfserr;
6331         }
6332         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6333 fput:
6334         fput(filp);
6335 put_stateid:
6336         mutex_unlock(&stp->st_mutex);
6337         nfs4_put_stid(&stp->st_stid);
6338 out:
6339         nfsd4_bump_seqid(cstate, status);
6340         if (file_lock)
6341                 locks_free_lock(file_lock);
6342         return status;
6343
6344 out_nfserr:
6345         status = nfserrno(err);
6346         goto fput;
6347 }
6348
6349 /*
6350  * returns
6351  *      true:  locks held by lockowner
6352  *      false: no locks held by lockowner
6353  */
6354 static bool
6355 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6356 {
6357         struct file_lock *fl;
6358         int status = false;
6359         struct file *filp = find_any_file(fp);
6360         struct inode *inode;
6361         struct file_lock_context *flctx;
6362
6363         if (!filp) {
6364                 /* Any valid lock stateid should have some sort of access */
6365                 WARN_ON_ONCE(1);
6366                 return status;
6367         }
6368
6369         inode = locks_inode(filp);
6370         flctx = inode->i_flctx;
6371
6372         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6373                 spin_lock(&flctx->flc_lock);
6374                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6375                         if (fl->fl_owner == (fl_owner_t)lowner) {
6376                                 status = true;
6377                                 break;
6378                         }
6379                 }
6380                 spin_unlock(&flctx->flc_lock);
6381         }
6382         fput(filp);
6383         return status;
6384 }
6385
6386 __be32
6387 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6388                         struct nfsd4_compound_state *cstate,
6389                         union nfsd4_op_u *u)
6390 {
6391         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6392         clientid_t *clid = &rlockowner->rl_clientid;
6393         struct nfs4_stateowner *sop;
6394         struct nfs4_lockowner *lo = NULL;
6395         struct nfs4_ol_stateid *stp;
6396         struct xdr_netobj *owner = &rlockowner->rl_owner;
6397         unsigned int hashval = ownerstr_hashval(owner);
6398         __be32 status;
6399         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6400         struct nfs4_client *clp;
6401         LIST_HEAD (reaplist);
6402
6403         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6404                 clid->cl_boot, clid->cl_id);
6405
6406         status = lookup_clientid(clid, cstate, nn);
6407         if (status)
6408                 return status;
6409
6410         clp = cstate->clp;
6411         /* Find the matching lock stateowner */
6412         spin_lock(&clp->cl_lock);
6413         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6414                             so_strhash) {
6415
6416                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6417                         continue;
6418
6419                 /* see if there are still any locks associated with it */
6420                 lo = lockowner(sop);
6421                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6422                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
6423                                 status = nfserr_locks_held;
6424                                 spin_unlock(&clp->cl_lock);
6425                                 return status;
6426                         }
6427                 }
6428
6429                 nfs4_get_stateowner(sop);
6430                 break;
6431         }
6432         if (!lo) {
6433                 spin_unlock(&clp->cl_lock);
6434                 return status;
6435         }
6436
6437         unhash_lockowner_locked(lo);
6438         while (!list_empty(&lo->lo_owner.so_stateids)) {
6439                 stp = list_first_entry(&lo->lo_owner.so_stateids,
6440                                        struct nfs4_ol_stateid,
6441                                        st_perstateowner);
6442                 WARN_ON(!unhash_lock_stateid(stp));
6443                 put_ol_stateid_locked(stp, &reaplist);
6444         }
6445         spin_unlock(&clp->cl_lock);
6446         free_ol_stateid_reaplist(&reaplist);
6447         remove_blocked_locks(lo);
6448         nfs4_put_stateowner(&lo->lo_owner);
6449
6450         return status;
6451 }
6452
6453 static inline struct nfs4_client_reclaim *
6454 alloc_reclaim(void)
6455 {
6456         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6457 }
6458
6459 bool
6460 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
6461 {
6462         struct nfs4_client_reclaim *crp;
6463
6464         crp = nfsd4_find_reclaim_client(name, nn);
6465         return (crp && crp->cr_clp);
6466 }
6467
6468 /*
6469  * failure => all reset bets are off, nfserr_no_grace...
6470  */
6471 struct nfs4_client_reclaim *
6472 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6473 {
6474         unsigned int strhashval;
6475         struct nfs4_client_reclaim *crp;
6476
6477         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6478         crp = alloc_reclaim();
6479         if (crp) {
6480                 strhashval = clientstr_hashval(name);
6481                 INIT_LIST_HEAD(&crp->cr_strhash);
6482                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6483                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6484                 crp->cr_clp = NULL;
6485                 nn->reclaim_str_hashtbl_size++;
6486         }
6487         return crp;
6488 }
6489
6490 void
6491 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6492 {
6493         list_del(&crp->cr_strhash);
6494         kfree(crp);
6495         nn->reclaim_str_hashtbl_size--;
6496 }
6497
6498 void
6499 nfs4_release_reclaim(struct nfsd_net *nn)
6500 {
6501         struct nfs4_client_reclaim *crp = NULL;
6502         int i;
6503
6504         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6505                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6506                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6507                                         struct nfs4_client_reclaim, cr_strhash);
6508                         nfs4_remove_reclaim_record(crp, nn);
6509                 }
6510         }
6511         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6512 }
6513
6514 /*
6515  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6516 struct nfs4_client_reclaim *
6517 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6518 {
6519         unsigned int strhashval;
6520         struct nfs4_client_reclaim *crp = NULL;
6521
6522         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6523
6524         strhashval = clientstr_hashval(recdir);
6525         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6526                 if (same_name(crp->cr_recdir, recdir)) {
6527                         return crp;
6528                 }
6529         }
6530         return NULL;
6531 }
6532
6533 /*
6534 * Called from OPEN. Look for clientid in reclaim list.
6535 */
6536 __be32
6537 nfs4_check_open_reclaim(clientid_t *clid,
6538                 struct nfsd4_compound_state *cstate,
6539                 struct nfsd_net *nn)
6540 {
6541         __be32 status;
6542
6543         /* find clientid in conf_id_hashtbl */
6544         status = lookup_clientid(clid, cstate, nn);
6545         if (status)
6546                 return nfserr_reclaim_bad;
6547
6548         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6549                 return nfserr_no_grace;
6550
6551         if (nfsd4_client_record_check(cstate->clp))
6552                 return nfserr_reclaim_bad;
6553
6554         return nfs_ok;
6555 }
6556
6557 #ifdef CONFIG_NFSD_FAULT_INJECTION
6558 static inline void
6559 put_client(struct nfs4_client *clp)
6560 {
6561         atomic_dec(&clp->cl_refcount);
6562 }
6563
6564 static struct nfs4_client *
6565 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6566 {
6567         struct nfs4_client *clp;
6568         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6569                                           nfsd_net_id);
6570
6571         if (!nfsd_netns_ready(nn))
6572                 return NULL;
6573
6574         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6575                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6576                         return clp;
6577         }
6578         return NULL;
6579 }
6580
6581 u64
6582 nfsd_inject_print_clients(void)
6583 {
6584         struct nfs4_client *clp;
6585         u64 count = 0;
6586         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6587                                           nfsd_net_id);
6588         char buf[INET6_ADDRSTRLEN];
6589
6590         if (!nfsd_netns_ready(nn))
6591                 return 0;
6592
6593         spin_lock(&nn->client_lock);
6594         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6595                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6596                 pr_info("NFS Client: %s\n", buf);
6597                 ++count;
6598         }
6599         spin_unlock(&nn->client_lock);
6600
6601         return count;
6602 }
6603
6604 u64
6605 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6606 {
6607         u64 count = 0;
6608         struct nfs4_client *clp;
6609         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6610                                           nfsd_net_id);
6611
6612         if (!nfsd_netns_ready(nn))
6613                 return count;
6614
6615         spin_lock(&nn->client_lock);
6616         clp = nfsd_find_client(addr, addr_size);
6617         if (clp) {
6618                 if (mark_client_expired_locked(clp) == nfs_ok)
6619                         ++count;
6620                 else
6621                         clp = NULL;
6622         }
6623         spin_unlock(&nn->client_lock);
6624
6625         if (clp)
6626                 expire_client(clp);
6627
6628         return count;
6629 }
6630
6631 u64
6632 nfsd_inject_forget_clients(u64 max)
6633 {
6634         u64 count = 0;
6635         struct nfs4_client *clp, *next;
6636         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6637                                                 nfsd_net_id);
6638         LIST_HEAD(reaplist);
6639
6640         if (!nfsd_netns_ready(nn))
6641                 return count;
6642
6643         spin_lock(&nn->client_lock);
6644         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6645                 if (mark_client_expired_locked(clp) == nfs_ok) {
6646                         list_add(&clp->cl_lru, &reaplist);
6647                         if (max != 0 && ++count >= max)
6648                                 break;
6649                 }
6650         }
6651         spin_unlock(&nn->client_lock);
6652
6653         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6654                 expire_client(clp);
6655
6656         return count;
6657 }
6658
6659 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6660                              const char *type)
6661 {
6662         char buf[INET6_ADDRSTRLEN];
6663         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6664         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6665 }
6666
6667 static void
6668 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6669                              struct list_head *collect)
6670 {
6671         struct nfs4_client *clp = lst->st_stid.sc_client;
6672         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6673                                           nfsd_net_id);
6674
6675         if (!collect)
6676                 return;
6677
6678         lockdep_assert_held(&nn->client_lock);
6679         atomic_inc(&clp->cl_refcount);
6680         list_add(&lst->st_locks, collect);
6681 }
6682
6683 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6684                                     struct list_head *collect,
6685                                     bool (*func)(struct nfs4_ol_stateid *))
6686 {
6687         struct nfs4_openowner *oop;
6688         struct nfs4_ol_stateid *stp, *st_next;
6689         struct nfs4_ol_stateid *lst, *lst_next;
6690         u64 count = 0;
6691
6692         spin_lock(&clp->cl_lock);
6693         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6694                 list_for_each_entry_safe(stp, st_next,
6695                                 &oop->oo_owner.so_stateids, st_perstateowner) {
6696                         list_for_each_entry_safe(lst, lst_next,
6697                                         &stp->st_locks, st_locks) {
6698                                 if (func) {
6699                                         if (func(lst))
6700                                                 nfsd_inject_add_lock_to_list(lst,
6701                                                                         collect);
6702                                 }
6703                                 ++count;
6704                                 /*
6705                                  * Despite the fact that these functions deal
6706                                  * with 64-bit integers for "count", we must
6707                                  * ensure that it doesn't blow up the
6708                                  * clp->cl_refcount. Throw a warning if we
6709                                  * start to approach INT_MAX here.
6710                                  */
6711                                 WARN_ON_ONCE(count == (INT_MAX / 2));
6712                                 if (count == max)
6713                                         goto out;
6714                         }
6715                 }
6716         }
6717 out:
6718         spin_unlock(&clp->cl_lock);
6719
6720         return count;
6721 }
6722
6723 static u64
6724 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6725                           u64 max)
6726 {
6727         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6728 }
6729
6730 static u64
6731 nfsd_print_client_locks(struct nfs4_client *clp)
6732 {
6733         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6734         nfsd_print_count(clp, count, "locked files");
6735         return count;
6736 }
6737
6738 u64
6739 nfsd_inject_print_locks(void)
6740 {
6741         struct nfs4_client *clp;
6742         u64 count = 0;
6743         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6744                                                 nfsd_net_id);
6745
6746         if (!nfsd_netns_ready(nn))
6747                 return 0;
6748
6749         spin_lock(&nn->client_lock);
6750         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6751                 count += nfsd_print_client_locks(clp);
6752         spin_unlock(&nn->client_lock);
6753
6754         return count;
6755 }
6756
6757 static void
6758 nfsd_reap_locks(struct list_head *reaplist)
6759 {
6760         struct nfs4_client *clp;
6761         struct nfs4_ol_stateid *stp, *next;
6762
6763         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6764                 list_del_init(&stp->st_locks);
6765                 clp = stp->st_stid.sc_client;
6766                 nfs4_put_stid(&stp->st_stid);
6767                 put_client(clp);
6768         }
6769 }
6770
6771 u64
6772 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6773 {
6774         unsigned int count = 0;
6775         struct nfs4_client *clp;
6776         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6777                                                 nfsd_net_id);
6778         LIST_HEAD(reaplist);
6779
6780         if (!nfsd_netns_ready(nn))
6781                 return count;
6782
6783         spin_lock(&nn->client_lock);
6784         clp = nfsd_find_client(addr, addr_size);
6785         if (clp)
6786                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6787         spin_unlock(&nn->client_lock);
6788         nfsd_reap_locks(&reaplist);
6789         return count;
6790 }
6791
6792 u64
6793 nfsd_inject_forget_locks(u64 max)
6794 {
6795         u64 count = 0;
6796         struct nfs4_client *clp;
6797         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6798                                                 nfsd_net_id);
6799         LIST_HEAD(reaplist);
6800
6801         if (!nfsd_netns_ready(nn))
6802                 return count;
6803
6804         spin_lock(&nn->client_lock);
6805         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6806                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6807                 if (max != 0 && count >= max)
6808                         break;
6809         }
6810         spin_unlock(&nn->client_lock);
6811         nfsd_reap_locks(&reaplist);
6812         return count;
6813 }
6814
6815 static u64
6816 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6817                               struct list_head *collect,
6818                               void (*func)(struct nfs4_openowner *))
6819 {
6820         struct nfs4_openowner *oop, *next;
6821         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6822                                                 nfsd_net_id);
6823         u64 count = 0;
6824
6825         lockdep_assert_held(&nn->client_lock);
6826
6827         spin_lock(&clp->cl_lock);
6828         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6829                 if (func) {
6830                         func(oop);
6831                         if (collect) {
6832                                 atomic_inc(&clp->cl_refcount);
6833                                 list_add(&oop->oo_perclient, collect);
6834                         }
6835                 }
6836                 ++count;
6837                 /*
6838                  * Despite the fact that these functions deal with
6839                  * 64-bit integers for "count", we must ensure that
6840                  * it doesn't blow up the clp->cl_refcount. Throw a
6841                  * warning if we start to approach INT_MAX here.
6842                  */
6843                 WARN_ON_ONCE(count == (INT_MAX / 2));
6844                 if (count == max)
6845                         break;
6846         }
6847         spin_unlock(&clp->cl_lock);
6848
6849         return count;
6850 }
6851
6852 static u64
6853 nfsd_print_client_openowners(struct nfs4_client *clp)
6854 {
6855         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6856
6857         nfsd_print_count(clp, count, "openowners");
6858         return count;
6859 }
6860
6861 static u64
6862 nfsd_collect_client_openowners(struct nfs4_client *clp,
6863                                struct list_head *collect, u64 max)
6864 {
6865         return nfsd_foreach_client_openowner(clp, max, collect,
6866                                                 unhash_openowner_locked);
6867 }
6868
6869 u64
6870 nfsd_inject_print_openowners(void)
6871 {
6872         struct nfs4_client *clp;
6873         u64 count = 0;
6874         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6875                                                 nfsd_net_id);
6876
6877         if (!nfsd_netns_ready(nn))
6878                 return 0;
6879
6880         spin_lock(&nn->client_lock);
6881         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6882                 count += nfsd_print_client_openowners(clp);
6883         spin_unlock(&nn->client_lock);
6884
6885         return count;
6886 }
6887
6888 static void
6889 nfsd_reap_openowners(struct list_head *reaplist)
6890 {
6891         struct nfs4_client *clp;
6892         struct nfs4_openowner *oop, *next;
6893
6894         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6895                 list_del_init(&oop->oo_perclient);
6896                 clp = oop->oo_owner.so_client;
6897                 release_openowner(oop);
6898                 put_client(clp);
6899         }
6900 }
6901
6902 u64
6903 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6904                                      size_t addr_size)
6905 {
6906         unsigned int count = 0;
6907         struct nfs4_client *clp;
6908         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6909                                                 nfsd_net_id);
6910         LIST_HEAD(reaplist);
6911
6912         if (!nfsd_netns_ready(nn))
6913                 return count;
6914
6915         spin_lock(&nn->client_lock);
6916         clp = nfsd_find_client(addr, addr_size);
6917         if (clp)
6918                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6919         spin_unlock(&nn->client_lock);
6920         nfsd_reap_openowners(&reaplist);
6921         return count;
6922 }
6923
6924 u64
6925 nfsd_inject_forget_openowners(u64 max)
6926 {
6927         u64 count = 0;
6928         struct nfs4_client *clp;
6929         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6930                                                 nfsd_net_id);
6931         LIST_HEAD(reaplist);
6932
6933         if (!nfsd_netns_ready(nn))
6934                 return count;
6935
6936         spin_lock(&nn->client_lock);
6937         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6938                 count += nfsd_collect_client_openowners(clp, &reaplist,
6939                                                         max - count);
6940                 if (max != 0 && count >= max)
6941                         break;
6942         }
6943         spin_unlock(&nn->client_lock);
6944         nfsd_reap_openowners(&reaplist);
6945         return count;
6946 }
6947
6948 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6949                                      struct list_head *victims)
6950 {
6951         struct nfs4_delegation *dp, *next;
6952         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6953                                                 nfsd_net_id);
6954         u64 count = 0;
6955
6956         lockdep_assert_held(&nn->client_lock);
6957
6958         spin_lock(&state_lock);
6959         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6960                 if (victims) {
6961                         /*
6962                          * It's not safe to mess with delegations that have a
6963                          * non-zero dl_time. They might have already been broken
6964                          * and could be processed by the laundromat outside of
6965                          * the state_lock. Just leave them be.
6966                          */
6967                         if (dp->dl_time != 0)
6968                                 continue;
6969
6970                         atomic_inc(&clp->cl_refcount);
6971                         WARN_ON(!unhash_delegation_locked(dp));
6972                         list_add(&dp->dl_recall_lru, victims);
6973                 }
6974                 ++count;
6975                 /*
6976                  * Despite the fact that these functions deal with
6977                  * 64-bit integers for "count", we must ensure that
6978                  * it doesn't blow up the clp->cl_refcount. Throw a
6979                  * warning if we start to approach INT_MAX here.
6980                  */
6981                 WARN_ON_ONCE(count == (INT_MAX / 2));
6982                 if (count == max)
6983                         break;
6984         }
6985         spin_unlock(&state_lock);
6986         return count;
6987 }
6988
6989 static u64
6990 nfsd_print_client_delegations(struct nfs4_client *clp)
6991 {
6992         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6993
6994         nfsd_print_count(clp, count, "delegations");
6995         return count;
6996 }
6997
6998 u64
6999 nfsd_inject_print_delegations(void)
7000 {
7001         struct nfs4_client *clp;
7002         u64 count = 0;
7003         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7004                                                 nfsd_net_id);
7005
7006         if (!nfsd_netns_ready(nn))
7007                 return 0;
7008
7009         spin_lock(&nn->client_lock);
7010         list_for_each_entry(clp, &nn->client_lru, cl_lru)
7011                 count += nfsd_print_client_delegations(clp);
7012         spin_unlock(&nn->client_lock);
7013
7014         return count;
7015 }
7016
7017 static void
7018 nfsd_forget_delegations(struct list_head *reaplist)
7019 {
7020         struct nfs4_client *clp;
7021         struct nfs4_delegation *dp, *next;
7022
7023         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7024                 list_del_init(&dp->dl_recall_lru);
7025                 clp = dp->dl_stid.sc_client;
7026                 revoke_delegation(dp);
7027                 put_client(clp);
7028         }
7029 }
7030
7031 u64
7032 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
7033                                       size_t addr_size)
7034 {
7035         u64 count = 0;
7036         struct nfs4_client *clp;
7037         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7038                                                 nfsd_net_id);
7039         LIST_HEAD(reaplist);
7040
7041         if (!nfsd_netns_ready(nn))
7042                 return count;
7043
7044         spin_lock(&nn->client_lock);
7045         clp = nfsd_find_client(addr, addr_size);
7046         if (clp)
7047                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7048         spin_unlock(&nn->client_lock);
7049
7050         nfsd_forget_delegations(&reaplist);
7051         return count;
7052 }
7053
7054 u64
7055 nfsd_inject_forget_delegations(u64 max)
7056 {
7057         u64 count = 0;
7058         struct nfs4_client *clp;
7059         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7060                                                 nfsd_net_id);
7061         LIST_HEAD(reaplist);
7062
7063         if (!nfsd_netns_ready(nn))
7064                 return count;
7065
7066         spin_lock(&nn->client_lock);
7067         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7068                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7069                 if (max != 0 && count >= max)
7070                         break;
7071         }
7072         spin_unlock(&nn->client_lock);
7073         nfsd_forget_delegations(&reaplist);
7074         return count;
7075 }
7076
7077 static void
7078 nfsd_recall_delegations(struct list_head *reaplist)
7079 {
7080         struct nfs4_client *clp;
7081         struct nfs4_delegation *dp, *next;
7082
7083         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7084                 list_del_init(&dp->dl_recall_lru);
7085                 clp = dp->dl_stid.sc_client;
7086                 /*
7087                  * We skipped all entries that had a zero dl_time before,
7088                  * so we can now reset the dl_time back to 0. If a delegation
7089                  * break comes in now, then it won't make any difference since
7090                  * we're recalling it either way.
7091                  */
7092                 spin_lock(&state_lock);
7093                 dp->dl_time = 0;
7094                 spin_unlock(&state_lock);
7095                 nfsd_break_one_deleg(dp);
7096                 put_client(clp);
7097         }
7098 }
7099
7100 u64
7101 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7102                                       size_t addr_size)
7103 {
7104         u64 count = 0;
7105         struct nfs4_client *clp;
7106         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7107                                                 nfsd_net_id);
7108         LIST_HEAD(reaplist);
7109
7110         if (!nfsd_netns_ready(nn))
7111                 return count;
7112
7113         spin_lock(&nn->client_lock);
7114         clp = nfsd_find_client(addr, addr_size);
7115         if (clp)
7116                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7117         spin_unlock(&nn->client_lock);
7118
7119         nfsd_recall_delegations(&reaplist);
7120         return count;
7121 }
7122
7123 u64
7124 nfsd_inject_recall_delegations(u64 max)
7125 {
7126         u64 count = 0;
7127         struct nfs4_client *clp, *next;
7128         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7129                                                 nfsd_net_id);
7130         LIST_HEAD(reaplist);
7131
7132         if (!nfsd_netns_ready(nn))
7133                 return count;
7134
7135         spin_lock(&nn->client_lock);
7136         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7137                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7138                 if (max != 0 && ++count >= max)
7139                         break;
7140         }
7141         spin_unlock(&nn->client_lock);
7142         nfsd_recall_delegations(&reaplist);
7143         return count;
7144 }
7145 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7146
7147 /*
7148  * Since the lifetime of a delegation isn't limited to that of an open, a
7149  * client may quite reasonably hang on to a delegation as long as it has
7150  * the inode cached.  This becomes an obvious problem the first time a
7151  * client's inode cache approaches the size of the server's total memory.
7152  *
7153  * For now we avoid this problem by imposing a hard limit on the number
7154  * of delegations, which varies according to the server's memory size.
7155  */
7156 static void
7157 set_max_delegations(void)
7158 {
7159         /*
7160          * Allow at most 4 delegations per megabyte of RAM.  Quick
7161          * estimates suggest that in the worst case (where every delegation
7162          * is for a different inode), a delegation could take about 1.5K,
7163          * giving a worst case usage of about 6% of memory.
7164          */
7165         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7166 }
7167
7168 static int nfs4_state_create_net(struct net *net)
7169 {
7170         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7171         int i;
7172
7173         nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7174                                             sizeof(struct list_head),
7175                                             GFP_KERNEL);
7176         if (!nn->conf_id_hashtbl)
7177                 goto err;
7178         nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7179                                               sizeof(struct list_head),
7180                                               GFP_KERNEL);
7181         if (!nn->unconf_id_hashtbl)
7182                 goto err_unconf_id;
7183         nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7184                                               sizeof(struct list_head),
7185                                               GFP_KERNEL);
7186         if (!nn->sessionid_hashtbl)
7187                 goto err_sessionid;
7188
7189         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7190                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7191                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7192         }
7193         for (i = 0; i < SESSION_HASH_SIZE; i++)
7194                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7195         nn->conf_name_tree = RB_ROOT;
7196         nn->unconf_name_tree = RB_ROOT;
7197         nn->boot_time = get_seconds();
7198         nn->grace_ended = false;
7199         nn->nfsd4_manager.block_opens = true;
7200         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7201         INIT_LIST_HEAD(&nn->client_lru);
7202         INIT_LIST_HEAD(&nn->close_lru);
7203         INIT_LIST_HEAD(&nn->del_recall_lru);
7204         spin_lock_init(&nn->client_lock);
7205         spin_lock_init(&nn->s2s_cp_lock);
7206         idr_init(&nn->s2s_cp_stateids);
7207
7208         spin_lock_init(&nn->blocked_locks_lock);
7209         INIT_LIST_HEAD(&nn->blocked_locks_lru);
7210
7211         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7212         get_net(net);
7213
7214         return 0;
7215
7216 err_sessionid:
7217         kfree(nn->unconf_id_hashtbl);
7218 err_unconf_id:
7219         kfree(nn->conf_id_hashtbl);
7220 err:
7221         return -ENOMEM;
7222 }
7223
7224 static void
7225 nfs4_state_destroy_net(struct net *net)
7226 {
7227         int i;
7228         struct nfs4_client *clp = NULL;
7229         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7230
7231         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7232                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7233                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7234                         destroy_client(clp);
7235                 }
7236         }
7237
7238         WARN_ON(!list_empty(&nn->blocked_locks_lru));
7239
7240         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7241                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7242                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7243                         destroy_client(clp);
7244                 }
7245         }
7246
7247         kfree(nn->sessionid_hashtbl);
7248         kfree(nn->unconf_id_hashtbl);
7249         kfree(nn->conf_id_hashtbl);
7250         put_net(net);
7251 }
7252
7253 int
7254 nfs4_state_start_net(struct net *net)
7255 {
7256         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7257         int ret;
7258
7259         ret = nfs4_state_create_net(net);
7260         if (ret)
7261                 return ret;
7262         locks_start_grace(net, &nn->nfsd4_manager);
7263         nfsd4_client_tracking_init(net);
7264         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %x)\n",
7265                nn->nfsd4_grace, net->ns.inum);
7266         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7267         return 0;
7268 }
7269
7270 /* initialization to perform when the nfsd service is started: */
7271
7272 int
7273 nfs4_state_start(void)
7274 {
7275         int ret;
7276
7277         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7278         if (laundry_wq == NULL) {
7279                 ret = -ENOMEM;
7280                 goto out;
7281         }
7282         ret = nfsd4_create_callback_queue();
7283         if (ret)
7284                 goto out_free_laundry;
7285
7286         set_max_delegations();
7287         return 0;
7288
7289 out_free_laundry:
7290         destroy_workqueue(laundry_wq);
7291 out:
7292         return ret;
7293 }
7294
7295 void
7296 nfs4_state_shutdown_net(struct net *net)
7297 {
7298         struct nfs4_delegation *dp = NULL;
7299         struct list_head *pos, *next, reaplist;
7300         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7301
7302         cancel_delayed_work_sync(&nn->laundromat_work);
7303         locks_end_grace(&nn->nfsd4_manager);
7304
7305         INIT_LIST_HEAD(&reaplist);
7306         spin_lock(&state_lock);
7307         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7308                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7309                 WARN_ON(!unhash_delegation_locked(dp));
7310                 list_add(&dp->dl_recall_lru, &reaplist);
7311         }
7312         spin_unlock(&state_lock);
7313         list_for_each_safe(pos, next, &reaplist) {
7314                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7315                 list_del_init(&dp->dl_recall_lru);
7316                 destroy_unhashed_deleg(dp);
7317         }
7318
7319         nfsd4_client_tracking_exit(net);
7320         nfs4_state_destroy_net(net);
7321 }
7322
7323 void
7324 nfs4_state_shutdown(void)
7325 {
7326         destroy_workqueue(laundry_wq);
7327         nfsd4_destroy_callback_queue();
7328 }
7329
7330 static void
7331 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7332 {
7333         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7334                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7335 }
7336
7337 static void
7338 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7339 {
7340         if (cstate->minorversion) {
7341                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7342                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7343         }
7344 }
7345
7346 void
7347 clear_current_stateid(struct nfsd4_compound_state *cstate)
7348 {
7349         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7350 }
7351
7352 /*
7353  * functions to set current state id
7354  */
7355 void
7356 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7357                 union nfsd4_op_u *u)
7358 {
7359         put_stateid(cstate, &u->open_downgrade.od_stateid);
7360 }
7361
7362 void
7363 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7364                 union nfsd4_op_u *u)
7365 {
7366         put_stateid(cstate, &u->open.op_stateid);
7367 }
7368
7369 void
7370 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7371                 union nfsd4_op_u *u)
7372 {
7373         put_stateid(cstate, &u->close.cl_stateid);
7374 }
7375
7376 void
7377 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7378                 union nfsd4_op_u *u)
7379 {
7380         put_stateid(cstate, &u->lock.lk_resp_stateid);
7381 }
7382
7383 /*
7384  * functions to consume current state id
7385  */
7386
7387 void
7388 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7389                 union nfsd4_op_u *u)
7390 {
7391         get_stateid(cstate, &u->open_downgrade.od_stateid);
7392 }
7393
7394 void
7395 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7396                 union nfsd4_op_u *u)
7397 {
7398         get_stateid(cstate, &u->delegreturn.dr_stateid);
7399 }
7400
7401 void
7402 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7403                 union nfsd4_op_u *u)
7404 {
7405         get_stateid(cstate, &u->free_stateid.fr_stateid);
7406 }
7407
7408 void
7409 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7410                 union nfsd4_op_u *u)
7411 {
7412         get_stateid(cstate, &u->setattr.sa_stateid);
7413 }
7414
7415 void
7416 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7417                 union nfsd4_op_u *u)
7418 {
7419         get_stateid(cstate, &u->close.cl_stateid);
7420 }
7421
7422 void
7423 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7424                 union nfsd4_op_u *u)
7425 {
7426         get_stateid(cstate, &u->locku.lu_stateid);
7427 }
7428
7429 void
7430 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7431                 union nfsd4_op_u *u)
7432 {
7433         get_stateid(cstate, &u->read.rd_stateid);
7434 }
7435
7436 void
7437 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7438                 union nfsd4_op_u *u)
7439 {
7440         get_stateid(cstate, &u->write.wr_stateid);
7441 }