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                 posix_unblock_lock(&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                         INIT_LIST_HEAD(&nbl->nbl_list);
 256                         INIT_LIST_HEAD(&nbl->nbl_lru);
 257                         fh_copy_shallow(&nbl->nbl_fh, fh);
 258                         locks_init_lock(&nbl->nbl_lock);
 259                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
 260                                         &nfsd4_cb_notify_lock_ops,
 261                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
 262                 }
 263         }
 264         return nbl;
 265 }
 266
 267 static void
 268 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
 269 {
 270         locks_release_private(&nbl->nbl_lock);
 271         kfree(nbl);
 272 }
 273
 274 static void
 275 remove_blocked_locks(struct nfs4_lockowner *lo)
 276 {
 277         struct nfs4_client *clp = lo->lo_owner.so_client;
 278         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 279         struct nfsd4_blocked_lock *nbl;
 280         LIST_HEAD(reaplist);
 281
 282         /* Dequeue all blocked locks */
 283         spin_lock(&nn->blocked_locks_lock);
 284         while (!list_empty(&lo->lo_blocked)) {
 285                 nbl = list_first_entry(&lo->lo_blocked,
 286                                         struct nfsd4_blocked_lock,
 287                                         nbl_list);
 288                 list_del_init(&nbl->nbl_list);
 289                 list_move(&nbl->nbl_lru, &reaplist);
 290         }
 291         spin_unlock(&nn->blocked_locks_lock);
 292
 293         /* Now free them */
 294         while (!list_empty(&reaplist)) {
 295                 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
 296                                         nbl_lru);
 297                 list_del_init(&nbl->nbl_lru);
 298                 posix_unblock_lock(&nbl->nbl_lock);
 299                 free_blocked_lock(nbl);
 300         }
 301 }
 302
 303 static int
 304 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
 305 {
 306         /*
 307          * Since this is just an optimization, we don't try very hard if it
 308          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
 309          * just quit trying on anything else.
 310          */
 311         switch (task->tk_status) {
 312         case -NFS4ERR_DELAY:
 313                 rpc_delay(task, 1 * HZ);
 314                 return 0;
 315         default:
 316                 return 1;
 317         }
 318 }
 319
 320 static void
 321 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
 322 {
 323         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 324                                                 struct nfsd4_blocked_lock, nbl_cb);
 325
 326         free_blocked_lock(nbl);
 327 }
 328
 329 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
 330         .done           = nfsd4_cb_notify_lock_done,
 331         .release        = nfsd4_cb_notify_lock_release,
 332 };
 333
 334 static inline struct nfs4_stateowner *
 335 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 336 {
 337         atomic_inc(&sop->so_count);
 338         return sop;
 339 }
 340
 341 static int
 342 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 343 {
 344         return (sop->so_owner.len == owner->len) &&
 345                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 346 }
 347
 348 static struct nfs4_openowner *
 349 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 350                         struct nfs4_client *clp)
 351 {
 352         struct nfs4_stateowner *so;
 353
 354         lockdep_assert_held(&clp->cl_lock);
 355
 356         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 357                             so_strhash) {
 358                 if (!so->so_is_open_owner)
 359                         continue;
 360                 if (same_owner_str(so, &open->op_owner))
 361                         return openowner(nfs4_get_stateowner(so));
 362         }
 363         return NULL;
 364 }
 365
 366 static struct nfs4_openowner *
 367 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 368                         struct nfs4_client *clp)
 369 {
 370         struct nfs4_openowner *oo;
 371
 372         spin_lock(&clp->cl_lock);
 373         oo = find_openstateowner_str_locked(hashval, open, clp);
 374         spin_unlock(&clp->cl_lock);
 375         return oo;
 376 }
 377
 378 static inline u32
 379 opaque_hashval(const void *ptr, int nbytes)
 380 {
 381         unsigned char *cptr = (unsigned char *) ptr;
 382
 383         u32 x = 0;
 384         while (nbytes--) {
 385                 x *= 37;
 386                 x += *cptr++;
 387         }
 388         return x;
 389 }
 390
 391 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 392 {
 393         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 394
 395         kmem_cache_free(file_slab, fp);
 396 }
 397
 398 void
 399 put_nfs4_file(struct nfs4_file *fi)
 400 {
 401         might_lock(&state_lock);
 402
 403         if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
 404                 hlist_del_rcu(&fi->fi_hash);
 405                 spin_unlock(&state_lock);
 406                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 407                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 408                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 409         }
 410 }
 411
 412 static struct file *
 413 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 414 {
 415         if (f->fi_fds[oflag])
 416                 return get_file(f->fi_fds[oflag]);
 417         return NULL;
 418 }
 419
 420 static struct file *
 421 find_writeable_file_locked(struct nfs4_file *f)
 422 {
 423         struct file *ret;
 424
 425         lockdep_assert_held(&f->fi_lock);
 426
 427         ret = __nfs4_get_fd(f, O_WRONLY);
 428         if (!ret)
 429                 ret = __nfs4_get_fd(f, O_RDWR);
 430         return ret;
 431 }
 432
 433 static struct file *
 434 find_writeable_file(struct nfs4_file *f)
 435 {
 436         struct file *ret;
 437
 438         spin_lock(&f->fi_lock);
 439         ret = find_writeable_file_locked(f);
 440         spin_unlock(&f->fi_lock);
 441
 442         return ret;
 443 }
 444
 445 static struct file *find_readable_file_locked(struct nfs4_file *f)
 446 {
 447         struct file *ret;
 448
 449         lockdep_assert_held(&f->fi_lock);
 450
 451         ret = __nfs4_get_fd(f, O_RDONLY);
 452         if (!ret)
 453                 ret = __nfs4_get_fd(f, O_RDWR);
 454         return ret;
 455 }
 456
 457 static struct file *
 458 find_readable_file(struct nfs4_file *f)
 459 {
 460         struct file *ret;
 461
 462         spin_lock(&f->fi_lock);
 463         ret = find_readable_file_locked(f);
 464         spin_unlock(&f->fi_lock);
 465
 466         return ret;
 467 }
 468
 469 struct file *
 470 find_any_file(struct nfs4_file *f)
 471 {
 472         struct file *ret;
 473
 474         if (!f)
 475                 return NULL;
 476         spin_lock(&f->fi_lock);
 477         ret = __nfs4_get_fd(f, O_RDWR);
 478         if (!ret) {
 479                 ret = __nfs4_get_fd(f, O_WRONLY);
 480                 if (!ret)
 481                         ret = __nfs4_get_fd(f, O_RDONLY);
 482         }
 483         spin_unlock(&f->fi_lock);
 484         return ret;
 485 }
 486
 487 static atomic_long_t num_delegations;
 488 unsigned long max_delegations;
 489
 490 /*
 491  * Open owner state (share locks)
 492  */
 493
 494 /* hash tables for lock and open owners */
 495 #define OWNER_HASH_BITS              8
 496 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 497 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 498
 499 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 500 {
 501         unsigned int ret;
 502
 503         ret = opaque_hashval(ownername->data, ownername->len);
 504         return ret & OWNER_HASH_MASK;
 505 }
 506
 507 /* hash table for nfs4_file */
 508 #define FILE_HASH_BITS                   8
 509 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 510
 511 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 512 {
 513         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 514 }
 515
 516 static unsigned int file_hashval(struct knfsd_fh *fh)
 517 {
 518         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 519 }
 520
 521 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 522
 523 static void
 524 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 525 {
 526         lockdep_assert_held(&fp->fi_lock);
 527
 528         if (access & NFS4_SHARE_ACCESS_WRITE)
 529                 atomic_inc(&fp->fi_access[O_WRONLY]);
 530         if (access & NFS4_SHARE_ACCESS_READ)
 531                 atomic_inc(&fp->fi_access[O_RDONLY]);
 532 }
 533
 534 static __be32
 535 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 536 {
 537         lockdep_assert_held(&fp->fi_lock);
 538
 539         /* Does this access mode make sense? */
 540         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 541                 return nfserr_inval;
 542
 543         /* Does it conflict with a deny mode already set? */
 544         if ((access & fp->fi_share_deny) != 0)
 545                 return nfserr_share_denied;
 546
 547         __nfs4_file_get_access(fp, access);
 548         return nfs_ok;
 549 }
 550
 551 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 552 {
 553         /* Common case is that there is no deny mode. */
 554         if (deny) {
 555                 /* Does this deny mode make sense? */
 556                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 557                         return nfserr_inval;
 558
 559                 if ((deny & NFS4_SHARE_DENY_READ) &&
 560                     atomic_read(&fp->fi_access[O_RDONLY]))
 561                         return nfserr_share_denied;
 562
 563                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 564                     atomic_read(&fp->fi_access[O_WRONLY]))
 565                         return nfserr_share_denied;
 566         }
 567         return nfs_ok;
 568 }
 569
 570 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 571 {
 572         might_lock(&fp->fi_lock);
 573
 574         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 575                 struct file *f1 = NULL;
 576                 struct file *f2 = NULL;
 577
 578                 swap(f1, fp->fi_fds[oflag]);
 579                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 580                         swap(f2, fp->fi_fds[O_RDWR]);
 581                 spin_unlock(&fp->fi_lock);
 582                 if (f1)
 583                         fput(f1);
 584                 if (f2)
 585                         fput(f2);
 586         }
 587 }
 588
 589 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 590 {
 591         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 592
 593         if (access & NFS4_SHARE_ACCESS_WRITE)
 594                 __nfs4_file_put_access(fp, O_WRONLY);
 595         if (access & NFS4_SHARE_ACCESS_READ)
 596                 __nfs4_file_put_access(fp, O_RDONLY);
 597 }
 598
 599 /*
 600  * Allocate a new open/delegation state counter. This is needed for
 601  * pNFS for proper return on close semantics.
 602  *
 603  * Note that we only allocate it for pNFS-enabled exports, otherwise
 604  * all pointers to struct nfs4_clnt_odstate are always NULL.
 605  */
 606 static struct nfs4_clnt_odstate *
 607 alloc_clnt_odstate(struct nfs4_client *clp)
 608 {
 609         struct nfs4_clnt_odstate *co;
 610
 611         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 612         if (co) {
 613                 co->co_client = clp;
 614                 refcount_set(&co->co_odcount, 1);
 615         }
 616         return co;
 617 }
 618
 619 static void
 620 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 621 {
 622         struct nfs4_file *fp = co->co_file;
 623
 624         lockdep_assert_held(&fp->fi_lock);
 625         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 626 }
 627
 628 static inline void
 629 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 630 {
 631         if (co)
 632                 refcount_inc(&co->co_odcount);
 633 }
 634
 635 static void
 636 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 637 {
 638         struct nfs4_file *fp;
 639
 640         if (!co)
 641                 return;
 642
 643         fp = co->co_file;
 644         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 645                 list_del(&co->co_perfile);
 646                 spin_unlock(&fp->fi_lock);
 647
 648                 nfsd4_return_all_file_layouts(co->co_client, fp);
 649                 kmem_cache_free(odstate_slab, co);
 650         }
 651 }
 652
 653 static struct nfs4_clnt_odstate *
 654 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 655 {
 656         struct nfs4_clnt_odstate *co;
 657         struct nfs4_client *cl;
 658
 659         if (!new)
 660                 return NULL;
 661
 662         cl = new->co_client;
 663
 664         spin_lock(&fp->fi_lock);
 665         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 666                 if (co->co_client == cl) {
 667                         get_clnt_odstate(co);
 668                         goto out;
 669                 }
 670         }
 671         co = new;
 672         co->co_file = fp;
 673         hash_clnt_odstate_locked(new);
 674 out:
 675         spin_unlock(&fp->fi_lock);
 676         return co;
 677 }
 678
 679 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 680                                   void (*sc_free)(struct nfs4_stid *))
 681 {
 682         struct nfs4_stid *stid;
 683         int new_id;
 684
 685         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 686         if (!stid)
 687                 return NULL;
 688
 689         idr_preload(GFP_KERNEL);
 690         spin_lock(&cl->cl_lock);
 691         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
 692         spin_unlock(&cl->cl_lock);
 693         idr_preload_end();
 694         if (new_id < 0)
 695                 goto out_free;
 696
 697         stid->sc_free = sc_free;
 698         stid->sc_client = cl;
 699         stid->sc_stateid.si_opaque.so_id = new_id;
 700         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 701         /* Will be incremented before return to client: */
 702         refcount_set(&stid->sc_count, 1);
 703         spin_lock_init(&stid->sc_lock);
 704
 705         /*
 706          * It shouldn't be a problem to reuse an opaque stateid value.
 707          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 708          * example, a stray write retransmission could be accepted by
 709          * the server when it should have been rejected.  Therefore,
 710          * adopt a trick from the sctp code to attempt to maximize the
 711          * amount of time until an id is reused, by ensuring they always
 712          * "increase" (mod INT_MAX):
 713          */
 714         return stid;
 715 out_free:
 716         kmem_cache_free(slab, stid);
 717         return NULL;
 718 }
 719
 720 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 721 {
 722         struct nfs4_stid *stid;
 723
 724         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
 725         if (!stid)
 726                 return NULL;
 727
 728         return openlockstateid(stid);
 729 }
 730
 731 static void nfs4_free_deleg(struct nfs4_stid *stid)
 732 {
 733         kmem_cache_free(deleg_slab, stid);
 734         atomic_long_dec(&num_delegations);
 735 }
 736
 737 /*
 738  * When we recall a delegation, we should be careful not to hand it
 739  * out again straight away.
 740  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 741  * in which the filehandles of recalled delegations are "stored".
 742  * If a filehandle appear in either filter, a delegation is blocked.
 743  * When a delegation is recalled, the filehandle is stored in the "new"
 744  * filter.
 745  * Every 30 seconds we swap the filters and clear the "new" one,
 746  * unless both are empty of course.
 747  *
 748  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 749  * low 3 bytes as hash-table indices.
 750  *
 751  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 752  * is used to manage concurrent access.  Testing does not need the lock
 753  * except when swapping the two filters.
 754  */
 755 static DEFINE_SPINLOCK(blocked_delegations_lock);
 756 static struct bloom_pair {
 757         int     entries, old_entries;
 758         time_t  swap_time;
 759         int     new; /* index into 'set' */
 760         DECLARE_BITMAP(set[2], 256);
 761 } blocked_delegations;
 762
 763 static int delegation_blocked(struct knfsd_fh *fh)
 764 {
 765         u32 hash;
 766         struct bloom_pair *bd = &blocked_delegations;
 767
 768         if (bd->entries == 0)
 769                 return 0;
 770         if (seconds_since_boot() - bd->swap_time > 30) {
 771                 spin_lock(&blocked_delegations_lock);
 772                 if (seconds_since_boot() - bd->swap_time > 30) {
 773                         bd->entries -= bd->old_entries;
 774                         bd->old_entries = bd->entries;
 775                         memset(bd->set[bd->new], 0,
 776                                sizeof(bd->set[0]));
 777                         bd->new = 1-bd->new;
 778                         bd->swap_time = seconds_since_boot();
 779                 }
 780                 spin_unlock(&blocked_delegations_lock);
 781         }
 782         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 783         if (test_bit(hash&255, bd->set[0]) &&
 784             test_bit((hash>>8)&255, bd->set[0]) &&
 785             test_bit((hash>>16)&255, bd->set[0]))
 786                 return 1;
 787
 788         if (test_bit(hash&255, bd->set[1]) &&
 789             test_bit((hash>>8)&255, bd->set[1]) &&
 790             test_bit((hash>>16)&255, bd->set[1]))
 791                 return 1;
 792
 793         return 0;
 794 }
 795
 796 static void block_delegations(struct knfsd_fh *fh)
 797 {
 798         u32 hash;
 799         struct bloom_pair *bd = &blocked_delegations;
 800
 801         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 802
 803         spin_lock(&blocked_delegations_lock);
 804         __set_bit(hash&255, bd->set[bd->new]);
 805         __set_bit((hash>>8)&255, bd->set[bd->new]);
 806         __set_bit((hash>>16)&255, bd->set[bd->new]);
 807         if (bd->entries == 0)
 808                 bd->swap_time = seconds_since_boot();
 809         bd->entries += 1;
 810         spin_unlock(&blocked_delegations_lock);
 811 }
 812
 813 static struct nfs4_delegation *
 814 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
 815                  struct svc_fh *current_fh,
 816                  struct nfs4_clnt_odstate *odstate)
 817 {
 818         struct nfs4_delegation *dp;
 819         long n;
 820
 821         dprintk("NFSD alloc_init_deleg\n");
 822         n = atomic_long_inc_return(&num_delegations);
 823         if (n < 0 || n > max_delegations)
 824                 goto out_dec;
 825         if (delegation_blocked(&current_fh->fh_handle))
 826                 goto out_dec;
 827         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
 828         if (dp == NULL)
 829                 goto out_dec;
 830
 831         /*
 832          * delegation seqid's are never incremented.  The 4.1 special
 833          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 834          * 0 anyway just for consistency and use 1:
 835          */
 836         dp->dl_stid.sc_stateid.si_generation = 1;
 837         INIT_LIST_HEAD(&dp->dl_perfile);
 838         INIT_LIST_HEAD(&dp->dl_perclnt);
 839         INIT_LIST_HEAD(&dp->dl_recall_lru);
 840         dp->dl_clnt_odstate = odstate;
 841         get_clnt_odstate(odstate);
 842         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 843         dp->dl_retries = 1;
 844         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 845                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 846         get_nfs4_file(fp);
 847         dp->dl_stid.sc_file = fp;
 848         return dp;
 849 out_dec:
 850         atomic_long_dec(&num_delegations);
 851         return NULL;
 852 }
 853
 854 void
 855 nfs4_put_stid(struct nfs4_stid *s)
 856 {
 857         struct nfs4_file *fp = s->sc_file;
 858         struct nfs4_client *clp = s->sc_client;
 859
 860         might_lock(&clp->cl_lock);
 861
 862         if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 863                 wake_up_all(&close_wq);
 864                 return;
 865         }
 866         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 867         spin_unlock(&clp->cl_lock);
 868         s->sc_free(s);
 869         if (fp)
 870                 put_nfs4_file(fp);
 871 }
 872
 873 void
 874 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
 875 {
 876         stateid_t *src = &stid->sc_stateid;
 877
 878         spin_lock(&stid->sc_lock);
 879         if (unlikely(++src->si_generation == 0))
 880                 src->si_generation = 1;
 881         memcpy(dst, src, sizeof(*dst));
 882         spin_unlock(&stid->sc_lock);
 883 }
 884
 885 static void put_deleg_file(struct nfs4_file *fp)
 886 {
 887         struct file *filp = NULL;
 888
 889         spin_lock(&fp->fi_lock);
 890         if (--fp->fi_delegees == 0)
 891                 swap(filp, fp->fi_deleg_file);
 892         spin_unlock(&fp->fi_lock);
 893
 894         if (filp)
 895                 fput(filp);
 896 }
 897
 898 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
 899 {
 900         struct nfs4_file *fp = dp->dl_stid.sc_file;
 901         struct file *filp = fp->fi_deleg_file;
 902
 903         WARN_ON_ONCE(!fp->fi_delegees);
 904
 905         vfs_setlease(filp, F_UNLCK, NULL, (void **)&dp);
 906         put_deleg_file(fp);
 907 }
 908
 909 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
 910 {
 911         put_clnt_odstate(dp->dl_clnt_odstate);
 912         nfs4_unlock_deleg_lease(dp);
 913         nfs4_put_stid(&dp->dl_stid);
 914 }
 915
 916 void nfs4_unhash_stid(struct nfs4_stid *s)
 917 {
 918         s->sc_type = 0;
 919 }
 920
 921 /**
 922  * nfs4_delegation_exists - Discover if this delegation already exists
 923  * @clp:     a pointer to the nfs4_client we're granting a delegation to
 924  * @fp:      a pointer to the nfs4_file we're granting a delegation on
 925  *
 926  * Return:
 927  *      On success: true iff an existing delegation is found
 928  */
 929
 930 static bool
 931 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
 932 {
 933         struct nfs4_delegation *searchdp = NULL;
 934         struct nfs4_client *searchclp = NULL;
 935
 936         lockdep_assert_held(&state_lock);
 937         lockdep_assert_held(&fp->fi_lock);
 938
 939         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
 940                 searchclp = searchdp->dl_stid.sc_client;
 941                 if (clp == searchclp) {
 942                         return true;
 943                 }
 944         }
 945         return false;
 946 }
 947
 948 /**
 949  * hash_delegation_locked - Add a delegation to the appropriate lists
 950  * @dp:     a pointer to the nfs4_delegation we are adding.
 951  * @fp:     a pointer to the nfs4_file we're granting a delegation on
 952  *
 953  * Return:
 954  *      On success: NULL if the delegation was successfully hashed.
 955  *
 956  *      On error: -EAGAIN if one was previously granted to this
 957  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
 958  *
 959  */
 960
 961 static int
 962 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
 963 {
 964         struct nfs4_client *clp = dp->dl_stid.sc_client;
 965
 966         lockdep_assert_held(&state_lock);
 967         lockdep_assert_held(&fp->fi_lock);
 968
 969         if (nfs4_delegation_exists(clp, fp))
 970                 return -EAGAIN;
 971         refcount_inc(&dp->dl_stid.sc_count);
 972         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 973         list_add(&dp->dl_perfile, &fp->fi_delegations);
 974         list_add(&dp->dl_perclnt, &clp->cl_delegations);
 975         return 0;
 976 }
 977
 978 static bool
 979 unhash_delegation_locked(struct nfs4_delegation *dp)
 980 {
 981         struct nfs4_file *fp = dp->dl_stid.sc_file;
 982
 983         lockdep_assert_held(&state_lock);
 984
 985         if (list_empty(&dp->dl_perfile))
 986                 return false;
 987
 988         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
 989         /* Ensure that deleg break won't try to requeue it */
 990         ++dp->dl_time;
 991         spin_lock(&fp->fi_lock);
 992         list_del_init(&dp->dl_perclnt);
 993         list_del_init(&dp->dl_recall_lru);
 994         list_del_init(&dp->dl_perfile);
 995         spin_unlock(&fp->fi_lock);
 996         return true;
 997 }
 998
 999 static void destroy_delegation(struct nfs4_delegation *dp)
1000 {
1001         bool unhashed;
1002
1003         spin_lock(&state_lock);
1004         unhashed = unhash_delegation_locked(dp);
1005         spin_unlock(&state_lock);
1006         if (unhashed)
1007                 destroy_unhashed_deleg(dp);
1008 }
1009
1010 static void revoke_delegation(struct nfs4_delegation *dp)
1011 {
1012         struct nfs4_client *clp = dp->dl_stid.sc_client;
1013
1014         WARN_ON(!list_empty(&dp->dl_recall_lru));
1015
1016         if (clp->cl_minorversion) {
1017                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1018                 refcount_inc(&dp->dl_stid.sc_count);
1019                 spin_lock(&clp->cl_lock);
1020                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1021                 spin_unlock(&clp->cl_lock);
1022         }
1023         destroy_unhashed_deleg(dp);
1024 }
1025
1026 /*
1027  * SETCLIENTID state
1028  */
1029
1030 static unsigned int clientid_hashval(u32 id)
1031 {
1032         return id & CLIENT_HASH_MASK;
1033 }
1034
1035 static unsigned int clientstr_hashval(const char *name)
1036 {
1037         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
1038 }
1039
1040 /*
1041  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1042  * st_{access,deny}_bmap field of the stateid, in order to track not
1043  * only what share bits are currently in force, but also what
1044  * combinations of share bits previous opens have used.  This allows us
1045  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1046  * return an error if the client attempt to downgrade to a combination
1047  * of share bits not explicable by closing some of its previous opens.
1048  *
1049  * XXX: This enforcement is actually incomplete, since we don't keep
1050  * track of access/deny bit combinations; so, e.g., we allow:
1051  *
1052  *      OPEN allow read, deny write
1053  *      OPEN allow both, deny none
1054  *      DOWNGRADE allow read, deny none
1055  *
1056  * which we should reject.
1057  */
1058 static unsigned int
1059 bmap_to_share_mode(unsigned long bmap) {
1060         int i;
1061         unsigned int access = 0;
1062
1063         for (i = 1; i < 4; i++) {
1064                 if (test_bit(i, &bmap))
1065                         access |= i;
1066         }
1067         return access;
1068 }
1069
1070 /* set share access for a given stateid */
1071 static inline void
1072 set_access(u32 access, struct nfs4_ol_stateid *stp)
1073 {
1074         unsigned char mask = 1 << access;
1075
1076         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1077         stp->st_access_bmap |= mask;
1078 }
1079
1080 /* clear share access for a given stateid */
1081 static inline void
1082 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1083 {
1084         unsigned char mask = 1 << access;
1085
1086         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1087         stp->st_access_bmap &= ~mask;
1088 }
1089
1090 /* test whether a given stateid has access */
1091 static inline bool
1092 test_access(u32 access, struct nfs4_ol_stateid *stp)
1093 {
1094         unsigned char mask = 1 << access;
1095
1096         return (bool)(stp->st_access_bmap & mask);
1097 }
1098
1099 /* set share deny for a given stateid */
1100 static inline void
1101 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1102 {
1103         unsigned char mask = 1 << deny;
1104
1105         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1106         stp->st_deny_bmap |= mask;
1107 }
1108
1109 /* clear share deny for a given stateid */
1110 static inline void
1111 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1112 {
1113         unsigned char mask = 1 << deny;
1114
1115         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1116         stp->st_deny_bmap &= ~mask;
1117 }
1118
1119 /* test whether a given stateid is denying specific access */
1120 static inline bool
1121 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1122 {
1123         unsigned char mask = 1 << deny;
1124
1125         return (bool)(stp->st_deny_bmap & mask);
1126 }
1127
1128 static int nfs4_access_to_omode(u32 access)
1129 {
1130         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1131         case NFS4_SHARE_ACCESS_READ:
1132                 return O_RDONLY;
1133         case NFS4_SHARE_ACCESS_WRITE:
1134                 return O_WRONLY;
1135         case NFS4_SHARE_ACCESS_BOTH:
1136                 return O_RDWR;
1137         }
1138         WARN_ON_ONCE(1);
1139         return O_RDONLY;
1140 }
1141
1142 /*
1143  * A stateid that had a deny mode associated with it is being released
1144  * or downgraded. Recalculate the deny mode on the file.
1145  */
1146 static void
1147 recalculate_deny_mode(struct nfs4_file *fp)
1148 {
1149         struct nfs4_ol_stateid *stp;
1150
1151         spin_lock(&fp->fi_lock);
1152         fp->fi_share_deny = 0;
1153         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1154                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1155         spin_unlock(&fp->fi_lock);
1156 }
1157
1158 static void
1159 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1160 {
1161         int i;
1162         bool change = false;
1163
1164         for (i = 1; i < 4; i++) {
1165                 if ((i & deny) != i) {
1166                         change = true;
1167                         clear_deny(i, stp);
1168                 }
1169         }
1170
1171         /* Recalculate per-file deny mode if there was a change */
1172         if (change)
1173                 recalculate_deny_mode(stp->st_stid.sc_file);
1174 }
1175
1176 /* release all access and file references for a given stateid */
1177 static void
1178 release_all_access(struct nfs4_ol_stateid *stp)
1179 {
1180         int i;
1181         struct nfs4_file *fp = stp->st_stid.sc_file;
1182
1183         if (fp && stp->st_deny_bmap != 0)
1184                 recalculate_deny_mode(fp);
1185
1186         for (i = 1; i < 4; i++) {
1187                 if (test_access(i, stp))
1188                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1189                 clear_access(i, stp);
1190         }
1191 }
1192
1193 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1194 {
1195         kfree(sop->so_owner.data);
1196         sop->so_ops->so_free(sop);
1197 }
1198
1199 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1200 {
1201         struct nfs4_client *clp = sop->so_client;
1202
1203         might_lock(&clp->cl_lock);
1204
1205         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1206                 return;
1207         sop->so_ops->so_unhash(sop);
1208         spin_unlock(&clp->cl_lock);
1209         nfs4_free_stateowner(sop);
1210 }
1211
1212 static bool
1213 nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1214 {
1215         return list_empty(&stp->st_perfile);
1216 }
1217
1218 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1219 {
1220         struct nfs4_file *fp = stp->st_stid.sc_file;
1221
1222         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1223
1224         if (list_empty(&stp->st_perfile))
1225                 return false;
1226
1227         spin_lock(&fp->fi_lock);
1228         list_del_init(&stp->st_perfile);
1229         spin_unlock(&fp->fi_lock);
1230         list_del(&stp->st_perstateowner);
1231         return true;
1232 }
1233
1234 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1235 {
1236         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1237
1238         put_clnt_odstate(stp->st_clnt_odstate);
1239         release_all_access(stp);
1240         if (stp->st_stateowner)
1241                 nfs4_put_stateowner(stp->st_stateowner);
1242         kmem_cache_free(stateid_slab, stid);
1243 }
1244
1245 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1246 {
1247         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1248         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1249         struct file *file;
1250
1251         file = find_any_file(stp->st_stid.sc_file);
1252         if (file)
1253                 filp_close(file, (fl_owner_t)lo);
1254         nfs4_free_ol_stateid(stid);
1255 }
1256
1257 /*
1258  * Put the persistent reference to an already unhashed generic stateid, while
1259  * holding the cl_lock. If it's the last reference, then put it onto the
1260  * reaplist for later destruction.
1261  */
1262 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1263                                        struct list_head *reaplist)
1264 {
1265         struct nfs4_stid *s = &stp->st_stid;
1266         struct nfs4_client *clp = s->sc_client;
1267
1268         lockdep_assert_held(&clp->cl_lock);
1269
1270         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1271
1272         if (!refcount_dec_and_test(&s->sc_count)) {
1273                 wake_up_all(&close_wq);
1274                 return;
1275         }
1276
1277         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1278         list_add(&stp->st_locks, reaplist);
1279 }
1280
1281 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1282 {
1283         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1284
1285         if (!unhash_ol_stateid(stp))
1286                 return false;
1287         list_del_init(&stp->st_locks);
1288         nfs4_unhash_stid(&stp->st_stid);
1289         return true;
1290 }
1291
1292 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1293 {
1294         struct nfs4_client *clp = stp->st_stid.sc_client;
1295         bool unhashed;
1296
1297         spin_lock(&clp->cl_lock);
1298         unhashed = unhash_lock_stateid(stp);
1299         spin_unlock(&clp->cl_lock);
1300         if (unhashed)
1301                 nfs4_put_stid(&stp->st_stid);
1302 }
1303
1304 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1305 {
1306         struct nfs4_client *clp = lo->lo_owner.so_client;
1307
1308         lockdep_assert_held(&clp->cl_lock);
1309
1310         list_del_init(&lo->lo_owner.so_strhash);
1311 }
1312
1313 /*
1314  * Free a list of generic stateids that were collected earlier after being
1315  * fully unhashed.
1316  */
1317 static void
1318 free_ol_stateid_reaplist(struct list_head *reaplist)
1319 {
1320         struct nfs4_ol_stateid *stp;
1321         struct nfs4_file *fp;
1322
1323         might_sleep();
1324
1325         while (!list_empty(reaplist)) {
1326                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1327                                        st_locks);
1328                 list_del(&stp->st_locks);
1329                 fp = stp->st_stid.sc_file;
1330                 stp->st_stid.sc_free(&stp->st_stid);
1331                 if (fp)
1332                         put_nfs4_file(fp);
1333         }
1334 }
1335
1336 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1337                                        struct list_head *reaplist)
1338 {
1339         struct nfs4_ol_stateid *stp;
1340
1341         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1342
1343         while (!list_empty(&open_stp->st_locks)) {
1344                 stp = list_entry(open_stp->st_locks.next,
1345                                 struct nfs4_ol_stateid, st_locks);
1346                 WARN_ON(!unhash_lock_stateid(stp));
1347                 put_ol_stateid_locked(stp, reaplist);
1348         }
1349 }
1350
1351 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1352                                 struct list_head *reaplist)
1353 {
1354         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1355
1356         if (!unhash_ol_stateid(stp))
1357                 return false;
1358         release_open_stateid_locks(stp, reaplist);
1359         return true;
1360 }
1361
1362 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1363 {
1364         LIST_HEAD(reaplist);
1365
1366         spin_lock(&stp->st_stid.sc_client->cl_lock);
1367         if (unhash_open_stateid(stp, &reaplist))
1368                 put_ol_stateid_locked(stp, &reaplist);
1369         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1370         free_ol_stateid_reaplist(&reaplist);
1371 }
1372
1373 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1374 {
1375         struct nfs4_client *clp = oo->oo_owner.so_client;
1376
1377         lockdep_assert_held(&clp->cl_lock);
1378
1379         list_del_init(&oo->oo_owner.so_strhash);
1380         list_del_init(&oo->oo_perclient);
1381 }
1382
1383 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1384 {
1385         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1386                                           nfsd_net_id);
1387         struct nfs4_ol_stateid *s;
1388
1389         spin_lock(&nn->client_lock);
1390         s = oo->oo_last_closed_stid;
1391         if (s) {
1392                 list_del_init(&oo->oo_close_lru);
1393                 oo->oo_last_closed_stid = NULL;
1394         }
1395         spin_unlock(&nn->client_lock);
1396         if (s)
1397                 nfs4_put_stid(&s->st_stid);
1398 }
1399
1400 static void release_openowner(struct nfs4_openowner *oo)
1401 {
1402         struct nfs4_ol_stateid *stp;
1403         struct nfs4_client *clp = oo->oo_owner.so_client;
1404         struct list_head reaplist;
1405
1406         INIT_LIST_HEAD(&reaplist);
1407
1408         spin_lock(&clp->cl_lock);
1409         unhash_openowner_locked(oo);
1410         while (!list_empty(&oo->oo_owner.so_stateids)) {
1411                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1412                                 struct nfs4_ol_stateid, st_perstateowner);
1413                 if (unhash_open_stateid(stp, &reaplist))
1414                         put_ol_stateid_locked(stp, &reaplist);
1415         }
1416         spin_unlock(&clp->cl_lock);
1417         free_ol_stateid_reaplist(&reaplist);
1418         release_last_closed_stateid(oo);
1419         nfs4_put_stateowner(&oo->oo_owner);
1420 }
1421
1422 static inline int
1423 hash_sessionid(struct nfs4_sessionid *sessionid)
1424 {
1425         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1426
1427         return sid->sequence % SESSION_HASH_SIZE;
1428 }
1429
1430 #ifdef CONFIG_SUNRPC_DEBUG
1431 static inline void
1432 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1433 {
1434         u32 *ptr = (u32 *)(&sessionid->data[0]);
1435         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1436 }
1437 #else
1438 static inline void
1439 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1440 {
1441 }
1442 #endif
1443
1444 /*
1445  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1446  * won't be used for replay.
1447  */
1448 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1449 {
1450         struct nfs4_stateowner *so = cstate->replay_owner;
1451
1452         if (nfserr == nfserr_replay_me)
1453                 return;
1454
1455         if (!seqid_mutating_err(ntohl(nfserr))) {
1456                 nfsd4_cstate_clear_replay(cstate);
1457                 return;
1458         }
1459         if (!so)
1460                 return;
1461         if (so->so_is_open_owner)
1462                 release_last_closed_stateid(openowner(so));
1463         so->so_seqid++;
1464         return;
1465 }
1466
1467 static void
1468 gen_sessionid(struct nfsd4_session *ses)
1469 {
1470         struct nfs4_client *clp = ses->se_client;
1471         struct nfsd4_sessionid *sid;
1472
1473         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1474         sid->clientid = clp->cl_clientid;
1475         sid->sequence = current_sessionid++;
1476         sid->reserved = 0;
1477 }
1478
1479 /*
1480  * The protocol defines ca_maxresponssize_cached to include the size of
1481  * the rpc header, but all we need to cache is the data starting after
1482  * the end of the initial SEQUENCE operation--the rest we regenerate
1483  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1484  * value that is the number of bytes in our cache plus a few additional
1485  * bytes.  In order to stay on the safe side, and not promise more than
1486  * we can cache, those additional bytes must be the minimum possible: 24
1487  * bytes of rpc header (xid through accept state, with AUTH_NULL
1488  * verifier), 12 for the compound header (with zero-length tag), and 44
1489  * for the SEQUENCE op response:
1490  */
1491 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1492
1493 static void
1494 free_session_slots(struct nfsd4_session *ses)
1495 {
1496         int i;
1497
1498         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1499                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1500                 kfree(ses->se_slots[i]);
1501         }
1502 }
1503
1504 /*
1505  * We don't actually need to cache the rpc and session headers, so we
1506  * can allocate a little less for each slot:
1507  */
1508 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1509 {
1510         u32 size;
1511
1512         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1513                 size = 0;
1514         else
1515                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1516         return size + sizeof(struct nfsd4_slot);
1517 }
1518
1519 /*
1520  * XXX: If we run out of reserved DRC memory we could (up to a point)
1521  * re-negotiate active sessions and reduce their slot usage to make
1522  * room for new connections. For now we just fail the create session.
1523  */
1524 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1525 {
1526         u32 slotsize = slot_bytes(ca);
1527         u32 num = ca->maxreqs;
1528         unsigned long avail, total_avail;
1529
1530         spin_lock(&nfsd_drc_lock);
1531         total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1532         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1533         /*
1534          * Never use more than a third of the remaining memory,
1535          * unless it's the only way to give this client a slot:
1536          */
1537         avail = clamp_t(unsigned long, avail, slotsize, total_avail/3);
1538         num = min_t(int, num, avail / slotsize);
1539         nfsd_drc_mem_used += num * slotsize;
1540         spin_unlock(&nfsd_drc_lock);
1541
1542         return num;
1543 }
1544
1545 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1546 {
1547         int slotsize = slot_bytes(ca);
1548
1549         spin_lock(&nfsd_drc_lock);
1550         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1551         spin_unlock(&nfsd_drc_lock);
1552 }
1553
1554 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1555                                            struct nfsd4_channel_attrs *battrs)
1556 {
1557         int numslots = fattrs->maxreqs;
1558         int slotsize = slot_bytes(fattrs);
1559         struct nfsd4_session *new;
1560         int mem, i;
1561
1562         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1563                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1564         mem = numslots * sizeof(struct nfsd4_slot *);
1565
1566         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1567         if (!new)
1568                 return NULL;
1569         /* allocate each struct nfsd4_slot and data cache in one piece */
1570         for (i = 0; i < numslots; i++) {
1571                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1572                 if (!new->se_slots[i])
1573                         goto out_free;
1574         }
1575
1576         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1577         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1578
1579         return new;
1580 out_free:
1581         while (i--)
1582                 kfree(new->se_slots[i]);
1583         kfree(new);
1584         return NULL;
1585 }
1586
1587 static void free_conn(struct nfsd4_conn *c)
1588 {
1589         svc_xprt_put(c->cn_xprt);
1590         kfree(c);
1591 }
1592
1593 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1594 {
1595         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1596         struct nfs4_client *clp = c->cn_session->se_client;
1597
1598         spin_lock(&clp->cl_lock);
1599         if (!list_empty(&c->cn_persession)) {
1600                 list_del(&c->cn_persession);
1601                 free_conn(c);
1602         }
1603         nfsd4_probe_callback(clp);
1604         spin_unlock(&clp->cl_lock);
1605 }
1606
1607 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1608 {
1609         struct nfsd4_conn *conn;
1610
1611         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1612         if (!conn)
1613                 return NULL;
1614         svc_xprt_get(rqstp->rq_xprt);
1615         conn->cn_xprt = rqstp->rq_xprt;
1616         conn->cn_flags = flags;
1617         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1618         return conn;
1619 }
1620
1621 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1622 {
1623         conn->cn_session = ses;
1624         list_add(&conn->cn_persession, &ses->se_conns);
1625 }
1626
1627 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1628 {
1629         struct nfs4_client *clp = ses->se_client;
1630
1631         spin_lock(&clp->cl_lock);
1632         __nfsd4_hash_conn(conn, ses);
1633         spin_unlock(&clp->cl_lock);
1634 }
1635
1636 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1637 {
1638         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1639         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1640 }
1641
1642 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1643 {
1644         int ret;
1645
1646         nfsd4_hash_conn(conn, ses);
1647         ret = nfsd4_register_conn(conn);
1648         if (ret)
1649                 /* oops; xprt is already down: */
1650                 nfsd4_conn_lost(&conn->cn_xpt_user);
1651         /* We may have gained or lost a callback channel: */
1652         nfsd4_probe_callback_sync(ses->se_client);
1653 }
1654
1655 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1656 {
1657         u32 dir = NFS4_CDFC4_FORE;
1658
1659         if (cses->flags & SESSION4_BACK_CHAN)
1660                 dir |= NFS4_CDFC4_BACK;
1661         return alloc_conn(rqstp, dir);
1662 }
1663
1664 /* must be called under client_lock */
1665 static void nfsd4_del_conns(struct nfsd4_session *s)
1666 {
1667         struct nfs4_client *clp = s->se_client;
1668         struct nfsd4_conn *c;
1669
1670         spin_lock(&clp->cl_lock);
1671         while (!list_empty(&s->se_conns)) {
1672                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1673                 list_del_init(&c->cn_persession);
1674                 spin_unlock(&clp->cl_lock);
1675
1676                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1677                 free_conn(c);
1678
1679                 spin_lock(&clp->cl_lock);
1680         }
1681         spin_unlock(&clp->cl_lock);
1682 }
1683
1684 static void __free_session(struct nfsd4_session *ses)
1685 {
1686         free_session_slots(ses);
1687         kfree(ses);
1688 }
1689
1690 static void free_session(struct nfsd4_session *ses)
1691 {
1692         nfsd4_del_conns(ses);
1693         nfsd4_put_drc_mem(&ses->se_fchannel);
1694         __free_session(ses);
1695 }
1696
1697 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1698 {
1699         int idx;
1700         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1701
1702         new->se_client = clp;
1703         gen_sessionid(new);
1704
1705         INIT_LIST_HEAD(&new->se_conns);
1706
1707         new->se_cb_seq_nr = 1;
1708         new->se_flags = cses->flags;
1709         new->se_cb_prog = cses->callback_prog;
1710         new->se_cb_sec = cses->cb_sec;
1711         atomic_set(&new->se_ref, 0);
1712         idx = hash_sessionid(&new->se_sessionid);
1713         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1714         spin_lock(&clp->cl_lock);
1715         list_add(&new->se_perclnt, &clp->cl_sessions);
1716         spin_unlock(&clp->cl_lock);
1717
1718         {
1719                 struct sockaddr *sa = svc_addr(rqstp);
1720                 /*
1721                  * This is a little silly; with sessions there's no real
1722                  * use for the callback address.  Use the peer address
1723                  * as a reasonable default for now, but consider fixing
1724                  * the rpc client not to require an address in the
1725                  * future:
1726                  */
1727                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1728                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1729         }
1730 }
1731
1732 /* caller must hold client_lock */
1733 static struct nfsd4_session *
1734 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1735 {
1736         struct nfsd4_session *elem;
1737         int idx;
1738         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1739
1740         lockdep_assert_held(&nn->client_lock);
1741
1742         dump_sessionid(__func__, sessionid);
1743         idx = hash_sessionid(sessionid);
1744         /* Search in the appropriate list */
1745         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1746                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1747                             NFS4_MAX_SESSIONID_LEN)) {
1748                         return elem;
1749                 }
1750         }
1751
1752         dprintk("%s: session not found\n", __func__);
1753         return NULL;
1754 }
1755
1756 static struct nfsd4_session *
1757 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1758                 __be32 *ret)
1759 {
1760         struct nfsd4_session *session;
1761         __be32 status = nfserr_badsession;
1762
1763         session = __find_in_sessionid_hashtbl(sessionid, net);
1764         if (!session)
1765                 goto out;
1766         status = nfsd4_get_session_locked(session);
1767         if (status)
1768                 session = NULL;
1769 out:
1770         *ret = status;
1771         return session;
1772 }
1773
1774 /* caller must hold client_lock */
1775 static void
1776 unhash_session(struct nfsd4_session *ses)
1777 {
1778         struct nfs4_client *clp = ses->se_client;
1779         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1780
1781         lockdep_assert_held(&nn->client_lock);
1782
1783         list_del(&ses->se_hash);
1784         spin_lock(&ses->se_client->cl_lock);
1785         list_del(&ses->se_perclnt);
1786         spin_unlock(&ses->se_client->cl_lock);
1787 }
1788
1789 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1790 static int
1791 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1792 {
1793         /*
1794          * We're assuming the clid was not given out from a boot
1795          * precisely 2^32 (about 136 years) before this one.  That seems
1796          * a safe assumption:
1797          */
1798         if (clid->cl_boot == (u32)nn->boot_time)
1799                 return 0;
1800         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1801                 clid->cl_boot, clid->cl_id, nn->boot_time);
1802         return 1;
1803 }
1804
1805 /*
1806  * XXX Should we use a slab cache ?
1807  * This type of memory management is somewhat inefficient, but we use it
1808  * anyway since SETCLIENTID is not a common operation.
1809  */
1810 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1811 {
1812         struct nfs4_client *clp;
1813         int i;
1814
1815         clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
1816         if (clp == NULL)
1817                 return NULL;
1818         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1819         if (clp->cl_name.data == NULL)
1820                 goto err_no_name;
1821         clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
1822                                                  sizeof(struct list_head),
1823                                                  GFP_KERNEL);
1824         if (!clp->cl_ownerstr_hashtbl)
1825                 goto err_no_hashtbl;
1826         for (i = 0; i < OWNER_HASH_SIZE; i++)
1827                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1828         clp->cl_name.len = name.len;
1829         INIT_LIST_HEAD(&clp->cl_sessions);
1830         idr_init(&clp->cl_stateids);
1831         atomic_set(&clp->cl_refcount, 0);
1832         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1833         INIT_LIST_HEAD(&clp->cl_idhash);
1834         INIT_LIST_HEAD(&clp->cl_openowners);
1835         INIT_LIST_HEAD(&clp->cl_delegations);
1836         INIT_LIST_HEAD(&clp->cl_lru);
1837         INIT_LIST_HEAD(&clp->cl_revoked);
1838 #ifdef CONFIG_NFSD_PNFS
1839         INIT_LIST_HEAD(&clp->cl_lo_states);
1840 #endif
1841         spin_lock_init(&clp->cl_lock);
1842         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1843         return clp;
1844 err_no_hashtbl:
1845         kfree(clp->cl_name.data);
1846 err_no_name:
1847         kmem_cache_free(client_slab, clp);
1848         return NULL;
1849 }
1850
1851 static void
1852 free_client(struct nfs4_client *clp)
1853 {
1854         while (!list_empty(&clp->cl_sessions)) {
1855                 struct nfsd4_session *ses;
1856                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1857                                 se_perclnt);
1858                 list_del(&ses->se_perclnt);
1859                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1860                 free_session(ses);
1861         }
1862         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1863         free_svc_cred(&clp->cl_cred);
1864         kfree(clp->cl_ownerstr_hashtbl);
1865         kfree(clp->cl_name.data);
1866         idr_destroy(&clp->cl_stateids);
1867         kmem_cache_free(client_slab, clp);
1868 }
1869
1870 /* must be called under the client_lock */
1871 static void
1872 unhash_client_locked(struct nfs4_client *clp)
1873 {
1874         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1875         struct nfsd4_session *ses;
1876
1877         lockdep_assert_held(&nn->client_lock);
1878
1879         /* Mark the client as expired! */
1880         clp->cl_time = 0;
1881         /* Make it invisible */
1882         if (!list_empty(&clp->cl_idhash)) {
1883                 list_del_init(&clp->cl_idhash);
1884                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1885                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1886                 else
1887                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1888         }
1889         list_del_init(&clp->cl_lru);
1890         spin_lock(&clp->cl_lock);
1891         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1892                 list_del_init(&ses->se_hash);
1893         spin_unlock(&clp->cl_lock);
1894 }
1895
1896 static void
1897 unhash_client(struct nfs4_client *clp)
1898 {
1899         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1900
1901         spin_lock(&nn->client_lock);
1902         unhash_client_locked(clp);
1903         spin_unlock(&nn->client_lock);
1904 }
1905
1906 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1907 {
1908         if (atomic_read(&clp->cl_refcount))
1909                 return nfserr_jukebox;
1910         unhash_client_locked(clp);
1911         return nfs_ok;
1912 }
1913
1914 static void
1915 __destroy_client(struct nfs4_client *clp)
1916 {
1917         int i;
1918         struct nfs4_openowner *oo;
1919         struct nfs4_delegation *dp;
1920         struct list_head reaplist;
1921
1922         INIT_LIST_HEAD(&reaplist);
1923         spin_lock(&state_lock);
1924         while (!list_empty(&clp->cl_delegations)) {
1925                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1926                 WARN_ON(!unhash_delegation_locked(dp));
1927                 list_add(&dp->dl_recall_lru, &reaplist);
1928         }
1929         spin_unlock(&state_lock);
1930         while (!list_empty(&reaplist)) {
1931                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1932                 list_del_init(&dp->dl_recall_lru);
1933                 destroy_unhashed_deleg(dp);
1934         }
1935         while (!list_empty(&clp->cl_revoked)) {
1936                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1937                 list_del_init(&dp->dl_recall_lru);
1938                 nfs4_put_stid(&dp->dl_stid);
1939         }
1940         while (!list_empty(&clp->cl_openowners)) {
1941                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1942                 nfs4_get_stateowner(&oo->oo_owner);
1943                 release_openowner(oo);
1944         }
1945         for (i = 0; i < OWNER_HASH_SIZE; i++) {
1946                 struct nfs4_stateowner *so, *tmp;
1947
1948                 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
1949                                          so_strhash) {
1950                         /* Should be no openowners at this point */
1951                         WARN_ON_ONCE(so->so_is_open_owner);
1952                         remove_blocked_locks(lockowner(so));
1953                 }
1954         }
1955         nfsd4_return_all_client_layouts(clp);
1956         nfsd4_shutdown_callback(clp);
1957         if (clp->cl_cb_conn.cb_xprt)
1958                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1959         free_client(clp);
1960 }
1961
1962 static void
1963 destroy_client(struct nfs4_client *clp)
1964 {
1965         unhash_client(clp);
1966         __destroy_client(clp);
1967 }
1968
1969 static void expire_client(struct nfs4_client *clp)
1970 {
1971         unhash_client(clp);
1972         nfsd4_client_record_remove(clp);
1973         __destroy_client(clp);
1974 }
1975
1976 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1977 {
1978         memcpy(target->cl_verifier.data, source->data,
1979                         sizeof(target->cl_verifier.data));
1980 }
1981
1982 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1983 {
1984         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1985         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1986 }
1987
1988 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1989 {
1990         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
1991         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
1992                                                                 GFP_KERNEL);
1993         target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
1994         if ((source->cr_principal && !target->cr_principal) ||
1995             (source->cr_raw_principal && !target->cr_raw_principal) ||
1996             (source->cr_targ_princ && !target->cr_targ_princ))
1997                 return -ENOMEM;
1998
1999         target->cr_flavor = source->cr_flavor;
2000         target->cr_uid = source->cr_uid;
2001         target->cr_gid = source->cr_gid;
2002         target->cr_group_info = source->cr_group_info;
2003         get_group_info(target->cr_group_info);
2004         target->cr_gss_mech = source->cr_gss_mech;
2005         if (source->cr_gss_mech)
2006                 gss_mech_get(source->cr_gss_mech);
2007         return 0;
2008 }
2009
2010 static int
2011 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2012 {
2013         if (o1->len < o2->len)
2014                 return -1;
2015         if (o1->len > o2->len)
2016                 return 1;
2017         return memcmp(o1->data, o2->data, o1->len);
2018 }
2019
2020 static int same_name(const char *n1, const char *n2)
2021 {
2022         return 0 == memcmp(n1, n2, HEXDIR_LEN);
2023 }
2024
2025 static int
2026 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2027 {
2028         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2029 }
2030
2031 static int
2032 same_clid(clientid_t *cl1, clientid_t *cl2)
2033 {
2034         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2035 }
2036
2037 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2038 {
2039         int i;
2040
2041         if (g1->ngroups != g2->ngroups)
2042                 return false;
2043         for (i=0; i<g1->ngroups; i++)
2044                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2045                         return false;
2046         return true;
2047 }
2048
2049 /*
2050  * RFC 3530 language requires clid_inuse be returned when the
2051  * "principal" associated with a requests differs from that previously
2052  * used.  We use uid, gid's, and gss principal string as our best
2053  * approximation.  We also don't want to allow non-gss use of a client
2054  * established using gss: in theory cr_principal should catch that
2055  * change, but in practice cr_principal can be null even in the gss case
2056  * since gssd doesn't always pass down a principal string.
2057  */
2058 static bool is_gss_cred(struct svc_cred *cr)
2059 {
2060         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2061         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2062 }
2063
2064
2065 static bool
2066 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2067 {
2068         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2069                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2070                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2071                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2072                 return false;
2073         /* XXX: check that cr_targ_princ fields match ? */
2074         if (cr1->cr_principal == cr2->cr_principal)
2075                 return true;
2076         if (!cr1->cr_principal || !cr2->cr_principal)
2077                 return false;
2078         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2079 }
2080
2081 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2082 {
2083         struct svc_cred *cr = &rqstp->rq_cred;
2084         u32 service;
2085
2086         if (!cr->cr_gss_mech)
2087                 return false;
2088         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2089         return service == RPC_GSS_SVC_INTEGRITY ||
2090                service == RPC_GSS_SVC_PRIVACY;
2091 }
2092
2093 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2094 {
2095         struct svc_cred *cr = &rqstp->rq_cred;
2096
2097         if (!cl->cl_mach_cred)
2098                 return true;
2099         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2100                 return false;
2101         if (!svc_rqst_integrity_protected(rqstp))
2102                 return false;
2103         if (cl->cl_cred.cr_raw_principal)
2104                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2105                                                 cr->cr_raw_principal);
2106         if (!cr->cr_principal)
2107                 return false;
2108         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2109 }
2110
2111 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2112 {
2113         __be32 verf[2];
2114
2115         /*
2116          * This is opaque to client, so no need to byte-swap. Use
2117          * __force to keep sparse happy
2118          */
2119         verf[0] = (__force __be32)get_seconds();
2120         verf[1] = (__force __be32)nn->clverifier_counter++;
2121         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2122 }
2123
2124 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2125 {
2126         clp->cl_clientid.cl_boot = nn->boot_time;
2127         clp->cl_clientid.cl_id = nn->clientid_counter++;
2128         gen_confirm(clp, nn);
2129 }
2130
2131 static struct nfs4_stid *
2132 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2133 {
2134         struct nfs4_stid *ret;
2135
2136         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2137         if (!ret || !ret->sc_type)
2138                 return NULL;
2139         return ret;
2140 }
2141
2142 static struct nfs4_stid *
2143 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2144 {
2145         struct nfs4_stid *s;
2146
2147         spin_lock(&cl->cl_lock);
2148         s = find_stateid_locked(cl, t);
2149         if (s != NULL) {
2150                 if (typemask & s->sc_type)
2151                         refcount_inc(&s->sc_count);
2152                 else
2153                         s = NULL;
2154         }
2155         spin_unlock(&cl->cl_lock);
2156         return s;
2157 }
2158
2159 static struct nfs4_client *create_client(struct xdr_netobj name,
2160                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2161 {
2162         struct nfs4_client *clp;
2163         struct sockaddr *sa = svc_addr(rqstp);
2164         int ret;
2165         struct net *net = SVC_NET(rqstp);
2166
2167         clp = alloc_client(name);
2168         if (clp == NULL)
2169                 return NULL;
2170
2171         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2172         if (ret) {
2173                 free_client(clp);
2174                 return NULL;
2175         }
2176         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2177         clp->cl_time = get_seconds();
2178         clear_bit(0, &clp->cl_cb_slot_busy);
2179         copy_verf(clp, verf);
2180         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2181         clp->cl_cb_session = NULL;
2182         clp->net = net;
2183         return clp;
2184 }
2185
2186 static void
2187 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2188 {
2189         struct rb_node **new = &(root->rb_node), *parent = NULL;
2190         struct nfs4_client *clp;
2191
2192         while (*new) {
2193                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2194                 parent = *new;
2195
2196                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2197                         new = &((*new)->rb_left);
2198                 else
2199                         new = &((*new)->rb_right);
2200         }
2201
2202         rb_link_node(&new_clp->cl_namenode, parent, new);
2203         rb_insert_color(&new_clp->cl_namenode, root);
2204 }
2205
2206 static struct nfs4_client *
2207 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2208 {
2209         int cmp;
2210         struct rb_node *node = root->rb_node;
2211         struct nfs4_client *clp;
2212
2213         while (node) {
2214                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2215                 cmp = compare_blob(&clp->cl_name, name);
2216                 if (cmp > 0)
2217                         node = node->rb_left;
2218                 else if (cmp < 0)
2219                         node = node->rb_right;
2220                 else
2221                         return clp;
2222         }
2223         return NULL;
2224 }
2225
2226 static void
2227 add_to_unconfirmed(struct nfs4_client *clp)
2228 {
2229         unsigned int idhashval;
2230         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2231
2232         lockdep_assert_held(&nn->client_lock);
2233
2234         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2235         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2236         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2237         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2238         renew_client_locked(clp);
2239 }
2240
2241 static void
2242 move_to_confirmed(struct nfs4_client *clp)
2243 {
2244         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2245         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2246
2247         lockdep_assert_held(&nn->client_lock);
2248
2249         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2250         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2251         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2252         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2253         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2254         renew_client_locked(clp);
2255 }
2256
2257 static struct nfs4_client *
2258 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2259 {
2260         struct nfs4_client *clp;
2261         unsigned int idhashval = clientid_hashval(clid->cl_id);
2262
2263         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2264                 if (same_clid(&clp->cl_clientid, clid)) {
2265                         if ((bool)clp->cl_minorversion != sessions)
2266                                 return NULL;
2267                         renew_client_locked(clp);
2268                         return clp;
2269                 }
2270         }
2271         return NULL;
2272 }
2273
2274 static struct nfs4_client *
2275 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2276 {
2277         struct list_head *tbl = nn->conf_id_hashtbl;
2278
2279         lockdep_assert_held(&nn->client_lock);
2280         return find_client_in_id_table(tbl, clid, sessions);
2281 }
2282
2283 static struct nfs4_client *
2284 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2285 {
2286         struct list_head *tbl = nn->unconf_id_hashtbl;
2287
2288         lockdep_assert_held(&nn->client_lock);
2289         return find_client_in_id_table(tbl, clid, sessions);
2290 }
2291
2292 static bool clp_used_exchangeid(struct nfs4_client *clp)
2293 {
2294         return clp->cl_exchange_flags != 0;
2295 }
2296
2297 static struct nfs4_client *
2298 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2299 {
2300         lockdep_assert_held(&nn->client_lock);
2301         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2302 }
2303
2304 static struct nfs4_client *
2305 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2306 {
2307         lockdep_assert_held(&nn->client_lock);
2308         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2309 }
2310
2311 static void
2312 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2313 {
2314         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2315         struct sockaddr *sa = svc_addr(rqstp);
2316         u32 scopeid = rpc_get_scope_id(sa);
2317         unsigned short expected_family;
2318
2319         /* Currently, we only support tcp and tcp6 for the callback channel */
2320         if (se->se_callback_netid_len == 3 &&
2321             !memcmp(se->se_callback_netid_val, "tcp", 3))
2322                 expected_family = AF_INET;
2323         else if (se->se_callback_netid_len == 4 &&
2324                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2325                 expected_family = AF_INET6;
2326         else
2327                 goto out_err;
2328
2329         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2330                                             se->se_callback_addr_len,
2331                                             (struct sockaddr *)&conn->cb_addr,
2332                                             sizeof(conn->cb_addr));
2333
2334         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2335                 goto out_err;
2336
2337         if (conn->cb_addr.ss_family == AF_INET6)
2338                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2339
2340         conn->cb_prog = se->se_callback_prog;
2341         conn->cb_ident = se->se_callback_ident;
2342         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2343         return;
2344 out_err:
2345         conn->cb_addr.ss_family = AF_UNSPEC;
2346         conn->cb_addrlen = 0;
2347         dprintk("NFSD: this client (clientid %08x/%08x) "
2348                 "will not receive delegations\n",
2349                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2350
2351         return;
2352 }
2353
2354 /*
2355  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2356  */
2357 static void
2358 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2359 {
2360         struct xdr_buf *buf = resp->xdr.buf;
2361         struct nfsd4_slot *slot = resp->cstate.slot;
2362         unsigned int base;
2363
2364         dprintk("--> %s slot %p\n", __func__, slot);
2365
2366         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2367         slot->sl_opcnt = resp->opcnt;
2368         slot->sl_status = resp->cstate.status;
2369         free_svc_cred(&slot->sl_cred);
2370         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2371
2372         if (!nfsd4_cache_this(resp)) {
2373                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2374                 return;
2375         }
2376         slot->sl_flags |= NFSD4_SLOT_CACHED;
2377
2378         base = resp->cstate.data_offset;
2379         slot->sl_datalen = buf->len - base;
2380         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2381                 WARN(1, "%s: sessions DRC could not cache compound\n",
2382                      __func__);
2383         return;
2384 }
2385
2386 /*
2387  * Encode the replay sequence operation from the slot values.
2388  * If cachethis is FALSE encode the uncached rep error on the next
2389  * operation which sets resp->p and increments resp->opcnt for
2390  * nfs4svc_encode_compoundres.
2391  *
2392  */
2393 static __be32
2394 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2395                           struct nfsd4_compoundres *resp)
2396 {
2397         struct nfsd4_op *op;
2398         struct nfsd4_slot *slot = resp->cstate.slot;
2399
2400         /* Encode the replayed sequence operation */
2401         op = &args->ops[resp->opcnt - 1];
2402         nfsd4_encode_operation(resp, op);
2403
2404         if (slot->sl_flags & NFSD4_SLOT_CACHED)
2405                 return op->status;
2406         if (args->opcnt == 1) {
2407                 /*
2408                  * The original operation wasn't a solo sequence--we
2409                  * always cache those--so this retry must not match the
2410                  * original:
2411                  */
2412                 op->status = nfserr_seq_false_retry;
2413         } else {
2414                 op = &args->ops[resp->opcnt++];
2415                 op->status = nfserr_retry_uncached_rep;
2416                 nfsd4_encode_operation(resp, op);
2417         }
2418         return op->status;
2419 }
2420
2421 /*
2422  * The sequence operation is not cached because we can use the slot and
2423  * session values.
2424  */
2425 static __be32
2426 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2427                          struct nfsd4_sequence *seq)
2428 {
2429         struct nfsd4_slot *slot = resp->cstate.slot;
2430         struct xdr_stream *xdr = &resp->xdr;
2431         __be32 *p;
2432         __be32 status;
2433
2434         dprintk("--> %s slot %p\n", __func__, slot);
2435
2436         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2437         if (status)
2438                 return status;
2439
2440         p = xdr_reserve_space(xdr, slot->sl_datalen);
2441         if (!p) {
2442                 WARN_ON_ONCE(1);
2443                 return nfserr_serverfault;
2444         }
2445         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2446         xdr_commit_encode(xdr);
2447
2448         resp->opcnt = slot->sl_opcnt;
2449         return slot->sl_status;
2450 }
2451
2452 /*
2453  * Set the exchange_id flags returned by the server.
2454  */
2455 static void
2456 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2457 {
2458 #ifdef CONFIG_NFSD_PNFS
2459         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2460 #else
2461         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2462 #endif
2463
2464         /* Referrals are supported, Migration is not. */
2465         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2466
2467         /* set the wire flags to return to client. */
2468         clid->flags = new->cl_exchange_flags;
2469 }
2470
2471 static bool client_has_openowners(struct nfs4_client *clp)
2472 {
2473         struct nfs4_openowner *oo;
2474
2475         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2476                 if (!list_empty(&oo->oo_owner.so_stateids))
2477                         return true;
2478         }
2479         return false;
2480 }
2481
2482 static bool client_has_state(struct nfs4_client *clp)
2483 {
2484         return client_has_openowners(clp)
2485 #ifdef CONFIG_NFSD_PNFS
2486                 || !list_empty(&clp->cl_lo_states)
2487 #endif
2488                 || !list_empty(&clp->cl_delegations)
2489                 || !list_empty(&clp->cl_sessions);
2490 }
2491
2492 __be32
2493 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2494                 union nfsd4_op_u *u)
2495 {
2496         struct nfsd4_exchange_id *exid = &u->exchange_id;
2497         struct nfs4_client *conf, *new;
2498         struct nfs4_client *unconf = NULL;
2499         __be32 status;
2500         char                    addr_str[INET6_ADDRSTRLEN];
2501         nfs4_verifier           verf = exid->verifier;
2502         struct sockaddr         *sa = svc_addr(rqstp);
2503         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2504         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2505
2506         rpc_ntop(sa, addr_str, sizeof(addr_str));
2507         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2508                 "ip_addr=%s flags %x, spa_how %d\n",
2509                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2510                 addr_str, exid->flags, exid->spa_how);
2511
2512         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2513                 return nfserr_inval;
2514
2515         new = create_client(exid->clname, rqstp, &verf);
2516         if (new == NULL)
2517                 return nfserr_jukebox;
2518
2519         switch (exid->spa_how) {
2520         case SP4_MACH_CRED:
2521                 exid->spo_must_enforce[0] = 0;
2522                 exid->spo_must_enforce[1] = (
2523                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
2524                         1 << (OP_EXCHANGE_ID - 32) |
2525                         1 << (OP_CREATE_SESSION - 32) |
2526                         1 << (OP_DESTROY_SESSION - 32) |
2527                         1 << (OP_DESTROY_CLIENTID - 32));
2528
2529                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2530                                         1 << (OP_OPEN_DOWNGRADE) |
2531                                         1 << (OP_LOCKU) |
2532                                         1 << (OP_DELEGRETURN));
2533
2534                 exid->spo_must_allow[1] &= (
2535                                         1 << (OP_TEST_STATEID - 32) |
2536                                         1 << (OP_FREE_STATEID - 32));
2537                 if (!svc_rqst_integrity_protected(rqstp)) {
2538                         status = nfserr_inval;
2539                         goto out_nolock;
2540                 }
2541                 /*
2542                  * Sometimes userspace doesn't give us a principal.
2543                  * Which is a bug, really.  Anyway, we can't enforce
2544                  * MACH_CRED in that case, better to give up now:
2545                  */
2546                 if (!new->cl_cred.cr_principal &&
2547                                         !new->cl_cred.cr_raw_principal) {
2548                         status = nfserr_serverfault;
2549                         goto out_nolock;
2550                 }
2551                 new->cl_mach_cred = true;
2552         case SP4_NONE:
2553                 break;
2554         default:                                /* checked by xdr code */
2555                 WARN_ON_ONCE(1);
2556         case SP4_SSV:
2557                 status = nfserr_encr_alg_unsupp;
2558                 goto out_nolock;
2559         }
2560
2561         /* Cases below refer to rfc 5661 section 18.35.4: */
2562         spin_lock(&nn->client_lock);
2563         conf = find_confirmed_client_by_name(&exid->clname, nn);
2564         if (conf) {
2565                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2566                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2567
2568                 if (update) {
2569                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2570                                 status = nfserr_inval;
2571                                 goto out;
2572                         }
2573                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
2574                                 status = nfserr_wrong_cred;
2575                                 goto out;
2576                         }
2577                         if (!creds_match) { /* case 9 */
2578                                 status = nfserr_perm;
2579                                 goto out;
2580                         }
2581                         if (!verfs_match) { /* case 8 */
2582                                 status = nfserr_not_same;
2583                                 goto out;
2584                         }
2585                         /* case 6 */
2586                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2587                         goto out_copy;
2588                 }
2589                 if (!creds_match) { /* case 3 */
2590                         if (client_has_state(conf)) {
2591                                 status = nfserr_clid_inuse;
2592                                 goto out;
2593                         }
2594                         goto out_new;
2595                 }
2596                 if (verfs_match) { /* case 2 */
2597                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2598                         goto out_copy;
2599                 }
2600                 /* case 5, client reboot */
2601                 conf = NULL;
2602                 goto out_new;
2603         }
2604
2605         if (update) { /* case 7 */
2606                 status = nfserr_noent;
2607                 goto out;
2608         }
2609
2610         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2611         if (unconf) /* case 4, possible retry or client restart */
2612                 unhash_client_locked(unconf);
2613
2614         /* case 1 (normal case) */
2615 out_new:
2616         if (conf) {
2617                 status = mark_client_expired_locked(conf);
2618                 if (status)
2619                         goto out;
2620         }
2621         new->cl_minorversion = cstate->minorversion;
2622         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2623         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2624
2625         gen_clid(new, nn);
2626         add_to_unconfirmed(new);
2627         swap(new, conf);
2628 out_copy:
2629         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2630         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2631
2632         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2633         nfsd4_set_ex_flags(conf, exid);
2634
2635         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2636                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2637         status = nfs_ok;
2638
2639 out:
2640         spin_unlock(&nn->client_lock);
2641 out_nolock:
2642         if (new)
2643                 expire_client(new);
2644         if (unconf)
2645                 expire_client(unconf);
2646         return status;
2647 }
2648
2649 static __be32
2650 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2651 {
2652         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2653                 slot_seqid);
2654
2655         /* The slot is in use, and no response has been sent. */
2656         if (slot_inuse) {
2657                 if (seqid == slot_seqid)
2658                         return nfserr_jukebox;
2659                 else
2660                         return nfserr_seq_misordered;
2661         }
2662         /* Note unsigned 32-bit arithmetic handles wraparound: */
2663         if (likely(seqid == slot_seqid + 1))
2664                 return nfs_ok;
2665         if (seqid == slot_seqid)
2666                 return nfserr_replay_cache;
2667         return nfserr_seq_misordered;
2668 }
2669
2670 /*
2671  * Cache the create session result into the create session single DRC
2672  * slot cache by saving the xdr structure. sl_seqid has been set.
2673  * Do this for solo or embedded create session operations.
2674  */
2675 static void
2676 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2677                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2678 {
2679         slot->sl_status = nfserr;
2680         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2681 }
2682
2683 static __be32
2684 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2685                             struct nfsd4_clid_slot *slot)
2686 {
2687         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2688         return slot->sl_status;
2689 }
2690
2691 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2692                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2693                         1 +     /* MIN tag is length with zero, only length */ \
2694                         3 +     /* version, opcount, opcode */ \
2695                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2696                                 /* seqid, slotID, slotID, cache */ \
2697                         4 ) * sizeof(__be32))
2698
2699 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2700                         2 +     /* verifier: AUTH_NULL, length 0 */\
2701                         1 +     /* status */ \
2702                         1 +     /* MIN tag is length with zero, only length */ \
2703                         3 +     /* opcount, opcode, opstatus*/ \
2704                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2705                                 /* seqid, slotID, slotID, slotID, status */ \
2706                         5 ) * sizeof(__be32))
2707
2708 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2709 {
2710         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2711
2712         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2713                 return nfserr_toosmall;
2714         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2715                 return nfserr_toosmall;
2716         ca->headerpadsz = 0;
2717         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2718         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2719         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2720         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2721                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2722         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2723         /*
2724          * Note decreasing slot size below client's request may make it
2725          * difficult for client to function correctly, whereas
2726          * decreasing the number of slots will (just?) affect
2727          * performance.  When short on memory we therefore prefer to
2728          * decrease number of slots instead of their size.  Clients that
2729          * request larger slots than they need will get poor results:
2730          */
2731         ca->maxreqs = nfsd4_get_drc_mem(ca);
2732         if (!ca->maxreqs)
2733                 return nfserr_jukebox;
2734
2735         return nfs_ok;
2736 }
2737
2738 /*
2739  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2740  * These are based on similar macros in linux/sunrpc/msg_prot.h .
2741  */
2742 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2743         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2744
2745 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2746         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2747
2748 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2749                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2750 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2751                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2752                                  sizeof(__be32))
2753
2754 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2755 {
2756         ca->headerpadsz = 0;
2757
2758         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2759                 return nfserr_toosmall;
2760         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2761                 return nfserr_toosmall;
2762         ca->maxresp_cached = 0;
2763         if (ca->maxops < 2)
2764                 return nfserr_toosmall;
2765
2766         return nfs_ok;
2767 }
2768
2769 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2770 {
2771         switch (cbs->flavor) {
2772         case RPC_AUTH_NULL:
2773         case RPC_AUTH_UNIX:
2774                 return nfs_ok;
2775         default:
2776                 /*
2777                  * GSS case: the spec doesn't allow us to return this
2778                  * error.  But it also doesn't allow us not to support
2779                  * GSS.
2780                  * I'd rather this fail hard than return some error the
2781                  * client might think it can already handle:
2782                  */
2783                 return nfserr_encr_alg_unsupp;
2784         }
2785 }
2786
2787 __be32
2788 nfsd4_create_session(struct svc_rqst *rqstp,
2789                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
2790 {
2791         struct nfsd4_create_session *cr_ses = &u->create_session;
2792         struct sockaddr *sa = svc_addr(rqstp);
2793         struct nfs4_client *conf, *unconf;
2794         struct nfs4_client *old = NULL;
2795         struct nfsd4_session *new;
2796         struct nfsd4_conn *conn;
2797         struct nfsd4_clid_slot *cs_slot = NULL;
2798         __be32 status = 0;
2799         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2800
2801         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2802                 return nfserr_inval;
2803         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2804         if (status)
2805                 return status;
2806         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2807         if (status)
2808                 return status;
2809         status = check_backchannel_attrs(&cr_ses->back_channel);
2810         if (status)
2811                 goto out_release_drc_mem;
2812         status = nfserr_jukebox;
2813         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2814         if (!new)
2815                 goto out_release_drc_mem;
2816         conn = alloc_conn_from_crses(rqstp, cr_ses);
2817         if (!conn)
2818                 goto out_free_session;
2819
2820         spin_lock(&nn->client_lock);
2821         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2822         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2823         WARN_ON_ONCE(conf && unconf);
2824
2825         if (conf) {
2826                 status = nfserr_wrong_cred;
2827                 if (!nfsd4_mach_creds_match(conf, rqstp))
2828                         goto out_free_conn;
2829                 cs_slot = &conf->cl_cs_slot;
2830                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2831                 if (status) {
2832                         if (status == nfserr_replay_cache)
2833                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2834                         goto out_free_conn;
2835                 }
2836         } else if (unconf) {
2837                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2838                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2839                         status = nfserr_clid_inuse;
2840                         goto out_free_conn;
2841                 }
2842                 status = nfserr_wrong_cred;
2843                 if (!nfsd4_mach_creds_match(unconf, rqstp))
2844                         goto out_free_conn;
2845                 cs_slot = &unconf->cl_cs_slot;
2846                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2847                 if (status) {
2848                         /* an unconfirmed replay returns misordered */
2849                         status = nfserr_seq_misordered;
2850                         goto out_free_conn;
2851                 }
2852                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2853                 if (old) {
2854                         status = mark_client_expired_locked(old);
2855                         if (status) {
2856                                 old = NULL;
2857                                 goto out_free_conn;
2858                         }
2859                 }
2860                 move_to_confirmed(unconf);
2861                 conf = unconf;
2862         } else {
2863                 status = nfserr_stale_clientid;
2864                 goto out_free_conn;
2865         }
2866         status = nfs_ok;
2867         /* Persistent sessions are not supported */
2868         cr_ses->flags &= ~SESSION4_PERSIST;
2869         /* Upshifting from TCP to RDMA is not supported */
2870         cr_ses->flags &= ~SESSION4_RDMA;
2871
2872         init_session(rqstp, new, conf, cr_ses);
2873         nfsd4_get_session_locked(new);
2874
2875         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2876                NFS4_MAX_SESSIONID_LEN);
2877         cs_slot->sl_seqid++;
2878         cr_ses->seqid = cs_slot->sl_seqid;
2879
2880         /* cache solo and embedded create sessions under the client_lock */
2881         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2882         spin_unlock(&nn->client_lock);
2883         /* init connection and backchannel */
2884         nfsd4_init_conn(rqstp, conn, new);
2885         nfsd4_put_session(new);
2886         if (old)
2887                 expire_client(old);
2888         return status;
2889 out_free_conn:
2890         spin_unlock(&nn->client_lock);
2891         free_conn(conn);
2892         if (old)
2893                 expire_client(old);
2894 out_free_session:
2895         __free_session(new);
2896 out_release_drc_mem:
2897         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2898         return status;
2899 }
2900
2901 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2902 {
2903         switch (*dir) {
2904         case NFS4_CDFC4_FORE:
2905         case NFS4_CDFC4_BACK:
2906                 return nfs_ok;
2907         case NFS4_CDFC4_FORE_OR_BOTH:
2908         case NFS4_CDFC4_BACK_OR_BOTH:
2909                 *dir = NFS4_CDFC4_BOTH;
2910                 return nfs_ok;
2911         };
2912         return nfserr_inval;
2913 }
2914
2915 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
2916                 struct nfsd4_compound_state *cstate,
2917                 union nfsd4_op_u *u)
2918 {
2919         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
2920         struct nfsd4_session *session = cstate->session;
2921         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2922         __be32 status;
2923
2924         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2925         if (status)
2926                 return status;
2927         spin_lock(&nn->client_lock);
2928         session->se_cb_prog = bc->bc_cb_program;
2929         session->se_cb_sec = bc->bc_cb_sec;
2930         spin_unlock(&nn->client_lock);
2931
2932         nfsd4_probe_callback(session->se_client);
2933
2934         return nfs_ok;
2935 }
2936
2937 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2938                      struct nfsd4_compound_state *cstate,
2939                      union nfsd4_op_u *u)
2940 {
2941         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
2942         __be32 status;
2943         struct nfsd4_conn *conn;
2944         struct nfsd4_session *session;
2945         struct net *net = SVC_NET(rqstp);
2946         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2947
2948         if (!nfsd4_last_compound_op(rqstp))
2949                 return nfserr_not_only_op;
2950         spin_lock(&nn->client_lock);
2951         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2952         spin_unlock(&nn->client_lock);
2953         if (!session)
2954                 goto out_no_session;
2955         status = nfserr_wrong_cred;
2956         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2957                 goto out;
2958         status = nfsd4_map_bcts_dir(&bcts->dir);
2959         if (status)
2960                 goto out;
2961         conn = alloc_conn(rqstp, bcts->dir);
2962         status = nfserr_jukebox;
2963         if (!conn)
2964                 goto out;
2965         nfsd4_init_conn(rqstp, conn, session);
2966         status = nfs_ok;
2967 out:
2968         nfsd4_put_session(session);
2969 out_no_session:
2970         return status;
2971 }
2972
2973 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
2974 {
2975         if (!cstate->session)
2976                 return false;
2977         return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
2978 }
2979
2980 __be32
2981 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
2982                 union nfsd4_op_u *u)
2983 {
2984         struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
2985         struct nfsd4_session *ses;
2986         __be32 status;
2987         int ref_held_by_me = 0;
2988         struct net *net = SVC_NET(r);
2989         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2990
2991         status = nfserr_not_only_op;
2992         if (nfsd4_compound_in_session(cstate, sessionid)) {
2993                 if (!nfsd4_last_compound_op(r))
2994                         goto out;
2995                 ref_held_by_me++;
2996         }
2997         dump_sessionid(__func__, sessionid);
2998         spin_lock(&nn->client_lock);
2999         ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3000         if (!ses)
3001                 goto out_client_lock;
3002         status = nfserr_wrong_cred;
3003         if (!nfsd4_mach_creds_match(ses->se_client, r))
3004                 goto out_put_session;
3005         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3006         if (status)
3007                 goto out_put_session;
3008         unhash_session(ses);
3009         spin_unlock(&nn->client_lock);
3010
3011         nfsd4_probe_callback_sync(ses->se_client);
3012
3013         spin_lock(&nn->client_lock);
3014         status = nfs_ok;
3015 out_put_session:
3016         nfsd4_put_session_locked(ses);
3017 out_client_lock:
3018         spin_unlock(&nn->client_lock);
3019 out:
3020         return status;
3021 }
3022
3023 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3024 {
3025         struct nfsd4_conn *c;
3026
3027         list_for_each_entry(c, &s->se_conns, cn_persession) {
3028                 if (c->cn_xprt == xpt) {
3029                         return c;
3030                 }
3031         }
3032         return NULL;
3033 }
3034
3035 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3036 {
3037         struct nfs4_client *clp = ses->se_client;
3038         struct nfsd4_conn *c;
3039         __be32 status = nfs_ok;
3040         int ret;
3041
3042         spin_lock(&clp->cl_lock);
3043         c = __nfsd4_find_conn(new->cn_xprt, ses);
3044         if (c)
3045                 goto out_free;
3046         status = nfserr_conn_not_bound_to_session;
3047         if (clp->cl_mach_cred)
3048                 goto out_free;
3049         __nfsd4_hash_conn(new, ses);
3050         spin_unlock(&clp->cl_lock);
3051         ret = nfsd4_register_conn(new);
3052         if (ret)
3053                 /* oops; xprt is already down: */
3054                 nfsd4_conn_lost(&new->cn_xpt_user);
3055         return nfs_ok;
3056 out_free:
3057         spin_unlock(&clp->cl_lock);
3058         free_conn(new);
3059         return status;
3060 }
3061
3062 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3063 {
3064         struct nfsd4_compoundargs *args = rqstp->rq_argp;
3065
3066         return args->opcnt > session->se_fchannel.maxops;
3067 }
3068
3069 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3070                                   struct nfsd4_session *session)
3071 {
3072         struct xdr_buf *xb = &rqstp->rq_arg;
3073
3074         return xb->len > session->se_fchannel.maxreq_sz;
3075 }
3076
3077 static bool replay_matches_cache(struct svc_rqst *rqstp,
3078                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3079 {
3080         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3081
3082         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3083             (bool)seq->cachethis)
3084                 return false;
3085         /*
3086          * If there's an error then the reply can have fewer ops than
3087          * the call.
3088          */
3089         if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
3090                 return false;
3091         /*
3092          * But if we cached a reply with *more* ops than the call you're
3093          * sending us now, then this new call is clearly not really a
3094          * replay of the old one:
3095          */
3096         if (slot->sl_opcnt > argp->opcnt)
3097                 return false;
3098         /* This is the only check explicitly called by spec: */
3099         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3100                 return false;
3101         /*
3102          * There may be more comparisons we could actually do, but the
3103          * spec doesn't require us to catch every case where the calls
3104          * don't match (that would require caching the call as well as
3105          * the reply), so we don't bother.
3106          */
3107         return true;
3108 }
3109
3110 __be32
3111 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3112                 union nfsd4_op_u *u)
3113 {
3114         struct nfsd4_sequence *seq = &u->sequence;
3115         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3116         struct xdr_stream *xdr = &resp->xdr;
3117         struct nfsd4_session *session;
3118         struct nfs4_client *clp;
3119         struct nfsd4_slot *slot;
3120         struct nfsd4_conn *conn;
3121         __be32 status;
3122         int buflen;
3123         struct net *net = SVC_NET(rqstp);
3124         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3125
3126         if (resp->opcnt != 1)
3127                 return nfserr_sequence_pos;
3128
3129         /*
3130          * Will be either used or freed by nfsd4_sequence_check_conn
3131          * below.
3132          */
3133         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3134         if (!conn)
3135                 return nfserr_jukebox;
3136
3137         spin_lock(&nn->client_lock);
3138         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3139         if (!session)
3140                 goto out_no_session;
3141         clp = session->se_client;
3142
3143         status = nfserr_too_many_ops;
3144         if (nfsd4_session_too_many_ops(rqstp, session))
3145                 goto out_put_session;
3146
3147         status = nfserr_req_too_big;
3148         if (nfsd4_request_too_big(rqstp, session))
3149                 goto out_put_session;
3150
3151         status = nfserr_badslot;
3152         if (seq->slotid >= session->se_fchannel.maxreqs)
3153                 goto out_put_session;
3154
3155         slot = session->se_slots[seq->slotid];
3156         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3157
3158         /* We do not negotiate the number of slots yet, so set the
3159          * maxslots to the session maxreqs which is used to encode
3160          * sr_highest_slotid and the sr_target_slot id to maxslots */
3161         seq->maxslots = session->se_fchannel.maxreqs;
3162
3163         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3164                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3165         if (status == nfserr_replay_cache) {
3166                 status = nfserr_seq_misordered;
3167                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3168                         goto out_put_session;
3169                 status = nfserr_seq_false_retry;
3170                 if (!replay_matches_cache(rqstp, seq, slot))
3171                         goto out_put_session;
3172                 cstate->slot = slot;
3173                 cstate->session = session;
3174                 cstate->clp = clp;
3175                 /* Return the cached reply status and set cstate->status
3176                  * for nfsd4_proc_compound processing */
3177                 status = nfsd4_replay_cache_entry(resp, seq);
3178                 cstate->status = nfserr_replay_cache;
3179                 goto out;
3180         }
3181         if (status)
3182                 goto out_put_session;
3183
3184         status = nfsd4_sequence_check_conn(conn, session);
3185         conn = NULL;
3186         if (status)
3187                 goto out_put_session;
3188
3189         buflen = (seq->cachethis) ?
3190                         session->se_fchannel.maxresp_cached :
3191                         session->se_fchannel.maxresp_sz;
3192         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3193                                     nfserr_rep_too_big;
3194         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3195                 goto out_put_session;
3196         svc_reserve(rqstp, buflen);
3197
3198         status = nfs_ok;
3199         /* Success! bump slot seqid */
3200         slot->sl_seqid = seq->seqid;
3201         slot->sl_flags |= NFSD4_SLOT_INUSE;
3202         if (seq->cachethis)
3203                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3204         else
3205                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3206
3207         cstate->slot = slot;
3208         cstate->session = session;
3209         cstate->clp = clp;
3210
3211 out:
3212         switch (clp->cl_cb_state) {
3213         case NFSD4_CB_DOWN:
3214                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3215                 break;
3216         case NFSD4_CB_FAULT:
3217                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3218                 break;
3219         default:
3220                 seq->status_flags = 0;
3221         }
3222         if (!list_empty(&clp->cl_revoked))
3223                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3224 out_no_session:
3225         if (conn)
3226                 free_conn(conn);
3227         spin_unlock(&nn->client_lock);
3228         return status;
3229 out_put_session:
3230         nfsd4_put_session_locked(session);
3231         goto out_no_session;
3232 }
3233
3234 void
3235 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3236 {
3237         struct nfsd4_compound_state *cs = &resp->cstate;
3238
3239         if (nfsd4_has_session(cs)) {
3240                 if (cs->status != nfserr_replay_cache) {
3241                         nfsd4_store_cache_entry(resp);
3242                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3243                 }
3244                 /* Drop session reference that was taken in nfsd4_sequence() */
3245                 nfsd4_put_session(cs->session);
3246         } else if (cs->clp)
3247                 put_client_renew(cs->clp);
3248 }
3249
3250 __be32
3251 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3252                 struct nfsd4_compound_state *cstate,
3253                 union nfsd4_op_u *u)
3254 {
3255         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3256         struct nfs4_client *conf, *unconf;
3257         struct nfs4_client *clp = NULL;
3258         __be32 status = 0;
3259         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3260
3261         spin_lock(&nn->client_lock);
3262         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3263         conf = find_confirmed_client(&dc->clientid, true, nn);
3264         WARN_ON_ONCE(conf && unconf);
3265
3266         if (conf) {
3267                 if (client_has_state(conf)) {
3268                         status = nfserr_clientid_busy;
3269                         goto out;
3270                 }
3271                 status = mark_client_expired_locked(conf);
3272                 if (status)
3273                         goto out;
3274                 clp = conf;
3275         } else if (unconf)
3276                 clp = unconf;
3277         else {
3278                 status = nfserr_stale_clientid;
3279                 goto out;
3280         }
3281         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3282                 clp = NULL;
3283                 status = nfserr_wrong_cred;
3284                 goto out;
3285         }
3286         unhash_client_locked(clp);
3287 out:
3288         spin_unlock(&nn->client_lock);
3289         if (clp)
3290                 expire_client(clp);
3291         return status;
3292 }
3293
3294 __be32
3295 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3296                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3297 {
3298         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3299         __be32 status = 0;
3300
3301         if (rc->rca_one_fs) {
3302                 if (!cstate->current_fh.fh_dentry)
3303                         return nfserr_nofilehandle;
3304                 /*
3305                  * We don't take advantage of the rca_one_fs case.
3306                  * That's OK, it's optional, we can safely ignore it.
3307                  */
3308                 return nfs_ok;
3309         }
3310
3311         status = nfserr_complete_already;
3312         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3313                              &cstate->session->se_client->cl_flags))
3314                 goto out;
3315
3316         status = nfserr_stale_clientid;
3317         if (is_client_expired(cstate->session->se_client))
3318                 /*
3319                  * The following error isn't really legal.
3320                  * But we only get here if the client just explicitly
3321                  * destroyed the client.  Surely it no longer cares what
3322                  * error it gets back on an operation for the dead
3323                  * client.
3324                  */
3325                 goto out;
3326
3327         status = nfs_ok;
3328         nfsd4_client_record_create(cstate->session->se_client);
3329 out:
3330         return status;
3331 }
3332
3333 __be32
3334 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3335                   union nfsd4_op_u *u)
3336 {
3337         struct nfsd4_setclientid *setclid = &u->setclientid;
3338         struct xdr_netobj       clname = setclid->se_name;
3339         nfs4_verifier           clverifier = setclid->se_verf;
3340         struct nfs4_client      *conf, *new;
3341         struct nfs4_client      *unconf = NULL;
3342         __be32                  status;
3343         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3344
3345         new = create_client(clname, rqstp, &clverifier);
3346         if (new == NULL)
3347                 return nfserr_jukebox;
3348         /* Cases below refer to rfc 3530 section 14.2.33: */
3349         spin_lock(&nn->client_lock);
3350         conf = find_confirmed_client_by_name(&clname, nn);
3351         if (conf && client_has_state(conf)) {
3352                 /* case 0: */
3353                 status = nfserr_clid_inuse;
3354                 if (clp_used_exchangeid(conf))
3355                         goto out;
3356                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3357                         char addr_str[INET6_ADDRSTRLEN];
3358                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3359                                  sizeof(addr_str));
3360                         dprintk("NFSD: setclientid: string in use by client "
3361                                 "at %s\n", addr_str);
3362                         goto out;
3363                 }
3364         }
3365         unconf = find_unconfirmed_client_by_name(&clname, nn);
3366         if (unconf)
3367                 unhash_client_locked(unconf);
3368         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3369                 /* case 1: probable callback update */
3370                 copy_clid(new, conf);
3371                 gen_confirm(new, nn);
3372         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3373                 gen_clid(new, nn);
3374         new->cl_minorversion = 0;
3375         gen_callback(new, setclid, rqstp);
3376         add_to_unconfirmed(new);
3377         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3378         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3379         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3380         new = NULL;
3381         status = nfs_ok;
3382 out:
3383         spin_unlock(&nn->client_lock);
3384         if (new)
3385                 free_client(new);
3386         if (unconf)
3387                 expire_client(unconf);
3388         return status;
3389 }
3390
3391
3392 __be32
3393 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3394                         struct nfsd4_compound_state *cstate,
3395                         union nfsd4_op_u *u)
3396 {
3397         struct nfsd4_setclientid_confirm *setclientid_confirm =
3398                         &u->setclientid_confirm;
3399         struct nfs4_client *conf, *unconf;
3400         struct nfs4_client *old = NULL;
3401         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3402         clientid_t * clid = &setclientid_confirm->sc_clientid;
3403         __be32 status;
3404         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3405
3406         if (STALE_CLIENTID(clid, nn))
3407                 return nfserr_stale_clientid;
3408
3409         spin_lock(&nn->client_lock);
3410         conf = find_confirmed_client(clid, false, nn);
3411         unconf = find_unconfirmed_client(clid, false, nn);
3412         /*
3413          * We try hard to give out unique clientid's, so if we get an
3414          * attempt to confirm the same clientid with a different cred,
3415          * the client may be buggy; this should never happen.
3416          *
3417          * Nevertheless, RFC 7530 recommends INUSE for this case:
3418          */
3419         status = nfserr_clid_inuse;
3420         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3421                 goto out;
3422         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3423                 goto out;
3424         /* cases below refer to rfc 3530 section 14.2.34: */
3425         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3426                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3427                         /* case 2: probable retransmit */
3428                         status = nfs_ok;
3429                 } else /* case 4: client hasn't noticed we rebooted yet? */
3430                         status = nfserr_stale_clientid;
3431                 goto out;
3432         }
3433         status = nfs_ok;
3434         if (conf) { /* case 1: callback update */
3435                 old = unconf;
3436                 unhash_client_locked(old);
3437                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3438         } else { /* case 3: normal case; new or rebooted client */
3439                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3440                 if (old) {
3441                         status = nfserr_clid_inuse;
3442                         if (client_has_state(old)
3443                                         && !same_creds(&unconf->cl_cred,
3444                                                         &old->cl_cred))
3445                                 goto out;
3446                         status = mark_client_expired_locked(old);
3447                         if (status) {
3448                                 old = NULL;
3449                                 goto out;
3450                         }
3451                 }
3452                 move_to_confirmed(unconf);
3453                 conf = unconf;
3454         }
3455         get_client_locked(conf);
3456         spin_unlock(&nn->client_lock);
3457         nfsd4_probe_callback(conf);
3458         spin_lock(&nn->client_lock);
3459         put_client_renew_locked(conf);
3460 out:
3461         spin_unlock(&nn->client_lock);
3462         if (old)
3463                 expire_client(old);
3464         return status;
3465 }
3466
3467 static struct nfs4_file *nfsd4_alloc_file(void)
3468 {
3469         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3470 }
3471
3472 /* OPEN Share state helper functions */
3473 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3474                                 struct nfs4_file *fp)
3475 {
3476         lockdep_assert_held(&state_lock);
3477
3478         refcount_set(&fp->fi_ref, 1);
3479         spin_lock_init(&fp->fi_lock);
3480         INIT_LIST_HEAD(&fp->fi_stateids);
3481         INIT_LIST_HEAD(&fp->fi_delegations);
3482         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3483         fh_copy_shallow(&fp->fi_fhandle, fh);
3484         fp->fi_deleg_file = NULL;
3485         fp->fi_had_conflict = false;
3486         fp->fi_share_deny = 0;
3487         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3488         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3489 #ifdef CONFIG_NFSD_PNFS
3490         INIT_LIST_HEAD(&fp->fi_lo_states);
3491         atomic_set(&fp->fi_lo_recalls, 0);
3492 #endif
3493         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3494 }
3495
3496 void
3497 nfsd4_free_slabs(void)
3498 {
3499         kmem_cache_destroy(client_slab);
3500         kmem_cache_destroy(openowner_slab);
3501         kmem_cache_destroy(lockowner_slab);
3502         kmem_cache_destroy(file_slab);
3503         kmem_cache_destroy(stateid_slab);
3504         kmem_cache_destroy(deleg_slab);
3505         kmem_cache_destroy(odstate_slab);
3506 }
3507
3508 int
3509 nfsd4_init_slabs(void)
3510 {
3511         client_slab = kmem_cache_create("nfsd4_clients",
3512                         sizeof(struct nfs4_client), 0, 0, NULL);
3513         if (client_slab == NULL)
3514                 goto out;
3515         openowner_slab = kmem_cache_create("nfsd4_openowners",
3516                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3517         if (openowner_slab == NULL)
3518                 goto out_free_client_slab;
3519         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3520                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3521         if (lockowner_slab == NULL)
3522                 goto out_free_openowner_slab;
3523         file_slab = kmem_cache_create("nfsd4_files",
3524                         sizeof(struct nfs4_file), 0, 0, NULL);
3525         if (file_slab == NULL)
3526                 goto out_free_lockowner_slab;
3527         stateid_slab = kmem_cache_create("nfsd4_stateids",
3528                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3529         if (stateid_slab == NULL)
3530                 goto out_free_file_slab;
3531         deleg_slab = kmem_cache_create("nfsd4_delegations",
3532                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3533         if (deleg_slab == NULL)
3534                 goto out_free_stateid_slab;
3535         odstate_slab = kmem_cache_create("nfsd4_odstate",
3536                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3537         if (odstate_slab == NULL)
3538                 goto out_free_deleg_slab;
3539         return 0;
3540
3541 out_free_deleg_slab:
3542         kmem_cache_destroy(deleg_slab);
3543 out_free_stateid_slab:
3544         kmem_cache_destroy(stateid_slab);
3545 out_free_file_slab:
3546         kmem_cache_destroy(file_slab);
3547 out_free_lockowner_slab:
3548         kmem_cache_destroy(lockowner_slab);
3549 out_free_openowner_slab:
3550         kmem_cache_destroy(openowner_slab);
3551 out_free_client_slab:
3552         kmem_cache_destroy(client_slab);
3553 out:
3554         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3555         return -ENOMEM;
3556 }
3557
3558 static void init_nfs4_replay(struct nfs4_replay *rp)
3559 {
3560         rp->rp_status = nfserr_serverfault;
3561         rp->rp_buflen = 0;
3562         rp->rp_buf = rp->rp_ibuf;
3563         mutex_init(&rp->rp_mutex);
3564 }
3565
3566 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3567                 struct nfs4_stateowner *so)
3568 {
3569         if (!nfsd4_has_session(cstate)) {
3570                 mutex_lock(&so->so_replay.rp_mutex);
3571                 cstate->replay_owner = nfs4_get_stateowner(so);
3572         }
3573 }
3574
3575 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3576 {
3577         struct nfs4_stateowner *so = cstate->replay_owner;
3578
3579         if (so != NULL) {
3580                 cstate->replay_owner = NULL;
3581                 mutex_unlock(&so->so_replay.rp_mutex);
3582                 nfs4_put_stateowner(so);
3583         }
3584 }
3585
3586 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3587 {
3588         struct nfs4_stateowner *sop;
3589
3590         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3591         if (!sop)
3592                 return NULL;
3593
3594         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3595         if (!sop->so_owner.data) {
3596                 kmem_cache_free(slab, sop);
3597                 return NULL;
3598         }
3599         sop->so_owner.len = owner->len;
3600
3601         INIT_LIST_HEAD(&sop->so_stateids);
3602         sop->so_client = clp;
3603         init_nfs4_replay(&sop->so_replay);
3604         atomic_set(&sop->so_count, 1);
3605         return sop;
3606 }
3607
3608 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3609 {
3610         lockdep_assert_held(&clp->cl_lock);
3611
3612         list_add(&oo->oo_owner.so_strhash,
3613                  &clp->cl_ownerstr_hashtbl[strhashval]);
3614         list_add(&oo->oo_perclient, &clp->cl_openowners);
3615 }
3616
3617 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3618 {
3619         unhash_openowner_locked(openowner(so));
3620 }
3621
3622 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3623 {
3624         struct nfs4_openowner *oo = openowner(so);
3625
3626         kmem_cache_free(openowner_slab, oo);
3627 }
3628
3629 static const struct nfs4_stateowner_operations openowner_ops = {
3630         .so_unhash =    nfs4_unhash_openowner,
3631         .so_free =      nfs4_free_openowner,
3632 };
3633
3634 static struct nfs4_ol_stateid *
3635 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3636 {
3637         struct nfs4_ol_stateid *local, *ret = NULL;
3638         struct nfs4_openowner *oo = open->op_openowner;
3639
3640         lockdep_assert_held(&fp->fi_lock);
3641
3642         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3643                 /* ignore lock owners */
3644                 if (local->st_stateowner->so_is_open_owner == 0)
3645                         continue;
3646                 if (local->st_stateowner != &oo->oo_owner)
3647                         continue;
3648                 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
3649                         ret = local;
3650                         refcount_inc(&ret->st_stid.sc_count);
3651                         break;
3652                 }
3653         }
3654         return ret;
3655 }
3656
3657 static __be32
3658 nfsd4_verify_open_stid(struct nfs4_stid *s)
3659 {
3660         __be32 ret = nfs_ok;
3661
3662         switch (s->sc_type) {
3663         default:
3664                 break;
3665         case 0:
3666         case NFS4_CLOSED_STID:
3667         case NFS4_CLOSED_DELEG_STID:
3668                 ret = nfserr_bad_stateid;
3669                 break;
3670         case NFS4_REVOKED_DELEG_STID:
3671                 ret = nfserr_deleg_revoked;
3672         }
3673         return ret;
3674 }
3675
3676 /* Lock the stateid st_mutex, and deal with races with CLOSE */
3677 static __be32
3678 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
3679 {
3680         __be32 ret;
3681
3682         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
3683         ret = nfsd4_verify_open_stid(&stp->st_stid);
3684         if (ret != nfs_ok)
3685                 mutex_unlock(&stp->st_mutex);
3686         return ret;
3687 }
3688
3689 static struct nfs4_ol_stateid *
3690 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3691 {
3692         struct nfs4_ol_stateid *stp;
3693         for (;;) {
3694                 spin_lock(&fp->fi_lock);
3695                 stp = nfsd4_find_existing_open(fp, open);
3696                 spin_unlock(&fp->fi_lock);
3697                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
3698                         break;
3699                 nfs4_put_stid(&stp->st_stid);
3700         }
3701         return stp;
3702 }
3703
3704 static struct nfs4_openowner *
3705 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3706                            struct nfsd4_compound_state *cstate)
3707 {
3708         struct nfs4_client *clp = cstate->clp;
3709         struct nfs4_openowner *oo, *ret;
3710
3711         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3712         if (!oo)
3713                 return NULL;
3714         oo->oo_owner.so_ops = &openowner_ops;
3715         oo->oo_owner.so_is_open_owner = 1;
3716         oo->oo_owner.so_seqid = open->op_seqid;
3717         oo->oo_flags = 0;
3718         if (nfsd4_has_session(cstate))
3719                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3720         oo->oo_time = 0;
3721         oo->oo_last_closed_stid = NULL;
3722         INIT_LIST_HEAD(&oo->oo_close_lru);
3723         spin_lock(&clp->cl_lock);
3724         ret = find_openstateowner_str_locked(strhashval, open, clp);
3725         if (ret == NULL) {
3726                 hash_openowner(oo, clp, strhashval);
3727                 ret = oo;
3728         } else
3729                 nfs4_free_stateowner(&oo->oo_owner);
3730
3731         spin_unlock(&clp->cl_lock);
3732         return ret;
3733 }
3734
3735 static struct nfs4_ol_stateid *
3736 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3737 {
3738
3739         struct nfs4_openowner *oo = open->op_openowner;
3740         struct nfs4_ol_stateid *retstp = NULL;
3741         struct nfs4_ol_stateid *stp;
3742
3743         stp = open->op_stp;
3744         /* We are moving these outside of the spinlocks to avoid the warnings */
3745         mutex_init(&stp->st_mutex);
3746         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
3747
3748 retry:
3749         spin_lock(&oo->oo_owner.so_client->cl_lock);
3750         spin_lock(&fp->fi_lock);
3751
3752         retstp = nfsd4_find_existing_open(fp, open);
3753         if (retstp)
3754                 goto out_unlock;
3755
3756         open->op_stp = NULL;
3757         refcount_inc(&stp->st_stid.sc_count);
3758         stp->st_stid.sc_type = NFS4_OPEN_STID;
3759         INIT_LIST_HEAD(&stp->st_locks);
3760         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3761         get_nfs4_file(fp);
3762         stp->st_stid.sc_file = fp;
3763         stp->st_access_bmap = 0;
3764         stp->st_deny_bmap = 0;
3765         stp->st_openstp = NULL;
3766         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3767         list_add(&stp->st_perfile, &fp->fi_stateids);
3768
3769 out_unlock:
3770         spin_unlock(&fp->fi_lock);
3771         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3772         if (retstp) {
3773                 /* Handle races with CLOSE */
3774                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
3775                         nfs4_put_stid(&retstp->st_stid);
3776                         goto retry;
3777                 }
3778                 /* To keep mutex tracking happy */
3779                 mutex_unlock(&stp->st_mutex);
3780                 stp = retstp;
3781         }
3782         return stp;
3783 }
3784
3785 /*
3786  * In the 4.0 case we need to keep the owners around a little while to handle
3787  * CLOSE replay. We still do need to release any file access that is held by
3788  * them before returning however.
3789  */
3790 static void
3791 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3792 {
3793         struct nfs4_ol_stateid *last;
3794         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3795         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3796                                                 nfsd_net_id);
3797
3798         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3799
3800         /*
3801          * We know that we hold one reference via nfsd4_close, and another
3802          * "persistent" reference for the client. If the refcount is higher
3803          * than 2, then there are still calls in progress that are using this
3804          * stateid. We can't put the sc_file reference until they are finished.
3805          * Wait for the refcount to drop to 2. Since it has been unhashed,
3806          * there should be no danger of the refcount going back up again at
3807          * this point.
3808          */
3809         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
3810
3811         release_all_access(s);
3812         if (s->st_stid.sc_file) {
3813                 put_nfs4_file(s->st_stid.sc_file);
3814                 s->st_stid.sc_file = NULL;
3815         }
3816
3817         spin_lock(&nn->client_lock);
3818         last = oo->oo_last_closed_stid;
3819         oo->oo_last_closed_stid = s;
3820         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3821         oo->oo_time = get_seconds();
3822         spin_unlock(&nn->client_lock);
3823         if (last)
3824                 nfs4_put_stid(&last->st_stid);
3825 }
3826
3827 /* search file_hashtbl[] for file */
3828 static struct nfs4_file *
3829 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3830 {
3831         struct nfs4_file *fp;
3832
3833         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3834                 if (fh_match(&fp->fi_fhandle, fh)) {
3835                         if (refcount_inc_not_zero(&fp->fi_ref))
3836                                 return fp;
3837                 }
3838         }
3839         return NULL;
3840 }
3841
3842 struct nfs4_file *
3843 find_file(struct knfsd_fh *fh)
3844 {
3845         struct nfs4_file *fp;
3846         unsigned int hashval = file_hashval(fh);
3847
3848         rcu_read_lock();
3849         fp = find_file_locked(fh, hashval);
3850         rcu_read_unlock();
3851         return fp;
3852 }
3853
3854 static struct nfs4_file *
3855 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3856 {
3857         struct nfs4_file *fp;
3858         unsigned int hashval = file_hashval(fh);
3859
3860         rcu_read_lock();
3861         fp = find_file_locked(fh, hashval);
3862         rcu_read_unlock();
3863         if (fp)
3864                 return fp;
3865
3866         spin_lock(&state_lock);
3867         fp = find_file_locked(fh, hashval);
3868         if (likely(fp == NULL)) {
3869                 nfsd4_init_file(fh, hashval, new);
3870                 fp = new;
3871         }
3872         spin_unlock(&state_lock);
3873
3874         return fp;
3875 }
3876
3877 /*
3878  * Called to check deny when READ with all zero stateid or
3879  * WRITE with all zero or all one stateid
3880  */
3881 static __be32
3882 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3883 {
3884         struct nfs4_file *fp;
3885         __be32 ret = nfs_ok;
3886
3887         fp = find_file(&current_fh->fh_handle);
3888         if (!fp)
3889                 return ret;
3890         /* Check for conflicting share reservations */
3891         spin_lock(&fp->fi_lock);
3892         if (fp->fi_share_deny & deny_type)
3893                 ret = nfserr_locked;
3894         spin_unlock(&fp->fi_lock);
3895         put_nfs4_file(fp);
3896         return ret;
3897 }
3898
3899 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3900 {
3901         struct nfs4_delegation *dp = cb_to_delegation(cb);
3902         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3903                                           nfsd_net_id);
3904
3905         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3906
3907         /*
3908          * We can't do this in nfsd_break_deleg_cb because it is
3909          * already holding inode->i_lock.
3910          *
3911          * If the dl_time != 0, then we know that it has already been
3912          * queued for a lease break. Don't queue it again.
3913          */
3914         spin_lock(&state_lock);
3915         if (dp->dl_time == 0) {
3916                 dp->dl_time = get_seconds();
3917                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3918         }
3919         spin_unlock(&state_lock);
3920 }
3921
3922 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3923                 struct rpc_task *task)
3924 {
3925         struct nfs4_delegation *dp = cb_to_delegation(cb);
3926
3927         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3928                 return 1;
3929
3930         switch (task->tk_status) {
3931         case 0:
3932                 return 1;
3933         case -EBADHANDLE:
3934         case -NFS4ERR_BAD_STATEID:
3935                 /*
3936                  * Race: client probably got cb_recall before open reply
3937                  * granting delegation.
3938                  */
3939                 if (dp->dl_retries--) {
3940                         rpc_delay(task, 2 * HZ);
3941                         return 0;
3942                 }
3943                 /*FALLTHRU*/
3944         default:
3945                 return -1;
3946         }
3947 }
3948
3949 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3950 {
3951         struct nfs4_delegation *dp = cb_to_delegation(cb);
3952
3953         nfs4_put_stid(&dp->dl_stid);
3954 }
3955
3956 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3957         .prepare        = nfsd4_cb_recall_prepare,
3958         .done           = nfsd4_cb_recall_done,
3959         .release        = nfsd4_cb_recall_release,
3960 };
3961
3962 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3963 {
3964         /*
3965          * We're assuming the state code never drops its reference
3966          * without first removing the lease.  Since we're in this lease
3967          * callback (and since the lease code is serialized by the
3968          * i_lock) we know the server hasn't removed the lease yet, and
3969          * we know it's safe to take a reference.
3970          */
3971         refcount_inc(&dp->dl_stid.sc_count);
3972         nfsd4_run_cb(&dp->dl_recall);
3973 }
3974
3975 /* Called from break_lease() with i_lock held. */
3976 static bool
3977 nfsd_break_deleg_cb(struct file_lock *fl)
3978 {
3979         bool ret = false;
3980         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
3981         struct nfs4_file *fp = dp->dl_stid.sc_file;
3982
3983         /*
3984          * We don't want the locks code to timeout the lease for us;
3985          * we'll remove it ourself if a delegation isn't returned
3986          * in time:
3987          */
3988         fl->fl_break_time = 0;
3989
3990         spin_lock(&fp->fi_lock);
3991         fp->fi_had_conflict = true;
3992         nfsd_break_one_deleg(dp);
3993         spin_unlock(&fp->fi_lock);
3994         return ret;
3995 }
3996
3997 static int
3998 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3999                      struct list_head *dispose)
4000 {
4001         if (arg & F_UNLCK)
4002                 return lease_modify(onlist, arg, dispose);
4003         else
4004                 return -EAGAIN;
4005 }
4006
4007 static const struct lock_manager_operations nfsd_lease_mng_ops = {
4008         .lm_break = nfsd_break_deleg_cb,
4009         .lm_change = nfsd_change_deleg_cb,
4010 };
4011
4012 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4013 {
4014         if (nfsd4_has_session(cstate))
4015                 return nfs_ok;
4016         if (seqid == so->so_seqid - 1)
4017                 return nfserr_replay_me;
4018         if (seqid == so->so_seqid)
4019                 return nfs_ok;
4020         return nfserr_bad_seqid;
4021 }
4022
4023 static __be32 lookup_clientid(clientid_t *clid,
4024                 struct nfsd4_compound_state *cstate,
4025                 struct nfsd_net *nn)
4026 {
4027         struct nfs4_client *found;
4028
4029         if (cstate->clp) {
4030                 found = cstate->clp;
4031                 if (!same_clid(&found->cl_clientid, clid))
4032                         return nfserr_stale_clientid;
4033                 return nfs_ok;
4034         }
4035
4036         if (STALE_CLIENTID(clid, nn))
4037                 return nfserr_stale_clientid;
4038
4039         /*
4040          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4041          * cached already then we know this is for is for v4.0 and "sessions"
4042          * will be false.
4043          */
4044         WARN_ON_ONCE(cstate->session);
4045         spin_lock(&nn->client_lock);
4046         found = find_confirmed_client(clid, false, nn);
4047         if (!found) {
4048                 spin_unlock(&nn->client_lock);
4049                 return nfserr_expired;
4050         }
4051         atomic_inc(&found->cl_refcount);
4052         spin_unlock(&nn->client_lock);
4053
4054         /* Cache the nfs4_client in cstate! */
4055         cstate->clp = found;
4056         return nfs_ok;
4057 }
4058
4059 __be32
4060 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4061                     struct nfsd4_open *open, struct nfsd_net *nn)
4062 {
4063         clientid_t *clientid = &open->op_clientid;
4064         struct nfs4_client *clp = NULL;
4065         unsigned int strhashval;
4066         struct nfs4_openowner *oo = NULL;
4067         __be32 status;
4068
4069         if (STALE_CLIENTID(&open->op_clientid, nn))
4070                 return nfserr_stale_clientid;
4071         /*
4072          * In case we need it later, after we've already created the
4073          * file and don't want to risk a further failure:
4074          */
4075         open->op_file = nfsd4_alloc_file();
4076         if (open->op_file == NULL)
4077                 return nfserr_jukebox;
4078
4079         status = lookup_clientid(clientid, cstate, nn);
4080         if (status)
4081                 return status;
4082         clp = cstate->clp;
4083
4084         strhashval = ownerstr_hashval(&open->op_owner);
4085         oo = find_openstateowner_str(strhashval, open, clp);
4086         open->op_openowner = oo;
4087         if (!oo) {
4088                 goto new_owner;
4089         }
4090         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4091                 /* Replace unconfirmed owners without checking for replay. */
4092                 release_openowner(oo);
4093                 open->op_openowner = NULL;
4094                 goto new_owner;
4095         }
4096         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4097         if (status)
4098                 return status;
4099         goto alloc_stateid;
4100 new_owner:
4101         oo = alloc_init_open_stateowner(strhashval, open, cstate);
4102         if (oo == NULL)
4103                 return nfserr_jukebox;
4104         open->op_openowner = oo;
4105 alloc_stateid:
4106         open->op_stp = nfs4_alloc_open_stateid(clp);
4107         if (!open->op_stp)
4108                 return nfserr_jukebox;
4109
4110         if (nfsd4_has_session(cstate) &&
4111             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4112                 open->op_odstate = alloc_clnt_odstate(clp);
4113                 if (!open->op_odstate)
4114                         return nfserr_jukebox;
4115         }
4116
4117         return nfs_ok;
4118 }
4119
4120 static inline __be32
4121 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4122 {
4123         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4124                 return nfserr_openmode;
4125         else
4126                 return nfs_ok;
4127 }
4128
4129 static int share_access_to_flags(u32 share_access)
4130 {
4131         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4132 }
4133
4134 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4135 {
4136         struct nfs4_stid *ret;
4137
4138         ret = find_stateid_by_type(cl, s,
4139                                 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4140         if (!ret)
4141                 return NULL;
4142         return delegstateid(ret);
4143 }
4144
4145 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4146 {
4147         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4148                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4149 }
4150
4151 static __be32
4152 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4153                 struct nfs4_delegation **dp)
4154 {
4155         int flags;
4156         __be32 status = nfserr_bad_stateid;
4157         struct nfs4_delegation *deleg;
4158
4159         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4160         if (deleg == NULL)
4161                 goto out;
4162         if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4163                 nfs4_put_stid(&deleg->dl_stid);
4164                 if (cl->cl_minorversion)
4165                         status = nfserr_deleg_revoked;
4166                 goto out;
4167         }
4168         flags = share_access_to_flags(open->op_share_access);
4169         status = nfs4_check_delegmode(deleg, flags);
4170         if (status) {
4171                 nfs4_put_stid(&deleg->dl_stid);
4172                 goto out;
4173         }
4174         *dp = deleg;
4175 out:
4176         if (!nfsd4_is_deleg_cur(open))
4177                 return nfs_ok;
4178         if (status)
4179                 return status;
4180         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4181         return nfs_ok;
4182 }
4183
4184 static inline int nfs4_access_to_access(u32 nfs4_access)
4185 {
4186         int flags = 0;
4187
4188         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4189                 flags |= NFSD_MAY_READ;
4190         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4191                 flags |= NFSD_MAY_WRITE;
4192         return flags;
4193 }
4194
4195 static inline __be32
4196 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4197                 struct nfsd4_open *open)
4198 {
4199         struct iattr iattr = {
4200                 .ia_valid = ATTR_SIZE,
4201                 .ia_size = 0,
4202         };
4203         if (!open->op_truncate)
4204                 return 0;
4205         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4206                 return nfserr_inval;
4207         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4208 }
4209
4210 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4211                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4212                 struct nfsd4_open *open)
4213 {
4214         struct file *filp = NULL;
4215         __be32 status;
4216         int oflag = nfs4_access_to_omode(open->op_share_access);
4217         int access = nfs4_access_to_access(open->op_share_access);
4218         unsigned char old_access_bmap, old_deny_bmap;
4219
4220         spin_lock(&fp->fi_lock);
4221
4222         /*
4223          * Are we trying to set a deny mode that would conflict with
4224          * current access?
4225          */
4226         status = nfs4_file_check_deny(fp, open->op_share_deny);
4227         if (status != nfs_ok) {
4228                 spin_unlock(&fp->fi_lock);
4229                 goto out;
4230         }
4231
4232         /* set access to the file */
4233         status = nfs4_file_get_access(fp, open->op_share_access);
4234         if (status != nfs_ok) {
4235                 spin_unlock(&fp->fi_lock);
4236                 goto out;
4237         }
4238
4239         /* Set access bits in stateid */
4240         old_access_bmap = stp->st_access_bmap;
4241         set_access(open->op_share_access, stp);
4242
4243         /* Set new deny mask */
4244         old_deny_bmap = stp->st_deny_bmap;
4245         set_deny(open->op_share_deny, stp);
4246         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4247
4248         if (!fp->fi_fds[oflag]) {
4249                 spin_unlock(&fp->fi_lock);
4250                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4251                 if (status)
4252                         goto out_put_access;
4253                 spin_lock(&fp->fi_lock);
4254                 if (!fp->fi_fds[oflag]) {
4255                         fp->fi_fds[oflag] = filp;
4256                         filp = NULL;
4257                 }
4258         }
4259         spin_unlock(&fp->fi_lock);
4260         if (filp)
4261                 fput(filp);
4262
4263         status = nfsd4_truncate(rqstp, cur_fh, open);
4264         if (status)
4265                 goto out_put_access;
4266 out:
4267         return status;
4268 out_put_access:
4269         stp->st_access_bmap = old_access_bmap;
4270         nfs4_file_put_access(fp, open->op_share_access);
4271         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4272         goto out;
4273 }
4274
4275 static __be32
4276 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)
4277 {
4278         __be32 status;
4279         unsigned char old_deny_bmap = stp->st_deny_bmap;
4280
4281         if (!test_access(open->op_share_access, stp))
4282                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4283
4284         /* test and set deny mode */
4285         spin_lock(&fp->fi_lock);
4286         status = nfs4_file_check_deny(fp, open->op_share_deny);
4287         if (status == nfs_ok) {
4288                 set_deny(open->op_share_deny, stp);
4289                 fp->fi_share_deny |=
4290                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4291         }
4292         spin_unlock(&fp->fi_lock);
4293
4294         if (status != nfs_ok)
4295                 return status;
4296
4297         status = nfsd4_truncate(rqstp, cur_fh, open);
4298         if (status != nfs_ok)
4299                 reset_union_bmap_deny(old_deny_bmap, stp);
4300         return status;
4301 }
4302
4303 /* Should we give out recallable state?: */
4304 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4305 {
4306         if (clp->cl_cb_state == NFSD4_CB_UP)
4307                 return true;
4308         /*
4309          * In the sessions case, since we don't have to establish a
4310          * separate connection for callbacks, we assume it's OK
4311          * until we hear otherwise:
4312          */
4313         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4314 }
4315
4316 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
4317                                                 int flag)
4318 {
4319         struct file_lock *fl;
4320
4321         fl = locks_alloc_lock();
4322         if (!fl)
4323                 return NULL;
4324         fl->fl_lmops = &nfsd_lease_mng_ops;
4325         fl->fl_flags = FL_DELEG;
4326         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4327         fl->fl_end = OFFSET_MAX;
4328         fl->fl_owner = (fl_owner_t)dp;
4329         fl->fl_pid = current->tgid;
4330         fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file;
4331         return fl;
4332 }
4333
4334 static struct nfs4_delegation *
4335 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4336                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4337 {
4338         int status = 0;
4339         struct nfs4_delegation *dp;
4340         struct file *filp;
4341         struct file_lock *fl;
4342
4343         /*
4344          * The fi_had_conflict and nfs_get_existing_delegation checks
4345          * here are just optimizations; we'll need to recheck them at
4346          * the end:
4347          */
4348         if (fp->fi_had_conflict)
4349                 return ERR_PTR(-EAGAIN);
4350
4351         filp = find_readable_file(fp);
4352         if (!filp) {
4353                 /* We should always have a readable file here */
4354                 WARN_ON_ONCE(1);
4355                 return ERR_PTR(-EBADF);
4356         }
4357         spin_lock(&state_lock);
4358         spin_lock(&fp->fi_lock);
4359         if (nfs4_delegation_exists(clp, fp))
4360                 status = -EAGAIN;
4361         else if (!fp->fi_deleg_file) {
4362                 fp->fi_deleg_file = filp;
4363                 /* increment early to prevent fi_deleg_file from being
4364                  * cleared */
4365                 fp->fi_delegees = 1;
4366                 filp = NULL;
4367         } else
4368                 fp->fi_delegees++;
4369         spin_unlock(&fp->fi_lock);
4370         spin_unlock(&state_lock);
4371         if (filp)
4372                 fput(filp);
4373         if (status)
4374                 return ERR_PTR(status);
4375
4376         status = -ENOMEM;
4377         dp = alloc_init_deleg(clp, fp, fh, odstate);
4378         if (!dp)
4379                 goto out_delegees;
4380
4381         fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
4382         if (!fl)
4383                 goto out_clnt_odstate;
4384
4385         status = vfs_setlease(fp->fi_deleg_file, fl->fl_type, &fl, NULL);
4386         if (fl)
4387                 locks_free_lock(fl);
4388         if (status)
4389                 goto out_clnt_odstate;
4390
4391         spin_lock(&state_lock);
4392         spin_lock(&fp->fi_lock);
4393         if (fp->fi_had_conflict)
4394                 status = -EAGAIN;
4395         else
4396                 status = hash_delegation_locked(dp, fp);
4397         spin_unlock(&fp->fi_lock);
4398         spin_unlock(&state_lock);
4399
4400         if (status)
4401                 goto out_unlock;
4402
4403         return dp;
4404 out_unlock:
4405         vfs_setlease(fp->fi_deleg_file, F_UNLCK, NULL, (void **)&dp);
4406 out_clnt_odstate:
4407         put_clnt_odstate(dp->dl_clnt_odstate);
4408         nfs4_put_stid(&dp->dl_stid);
4409 out_delegees:
4410         put_deleg_file(fp);
4411         return ERR_PTR(status);
4412 }
4413
4414 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4415 {
4416         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4417         if (status == -EAGAIN)
4418                 open->op_why_no_deleg = WND4_CONTENTION;
4419         else {
4420                 open->op_why_no_deleg = WND4_RESOURCE;
4421                 switch (open->op_deleg_want) {
4422                 case NFS4_SHARE_WANT_READ_DELEG:
4423                 case NFS4_SHARE_WANT_WRITE_DELEG:
4424                 case NFS4_SHARE_WANT_ANY_DELEG:
4425                         break;
4426                 case NFS4_SHARE_WANT_CANCEL:
4427                         open->op_why_no_deleg = WND4_CANCELLED;
4428                         break;
4429                 case NFS4_SHARE_WANT_NO_DELEG:
4430                         WARN_ON_ONCE(1);
4431                 }
4432         }
4433 }
4434
4435 /*
4436  * Attempt to hand out a delegation.
4437  *
4438  * Note we don't support write delegations, and won't until the vfs has
4439  * proper support for them.
4440  */
4441 static void
4442 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4443                         struct nfs4_ol_stateid *stp)
4444 {
4445         struct nfs4_delegation *dp;
4446         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4447         struct nfs4_client *clp = stp->st_stid.sc_client;
4448         int cb_up;
4449         int status = 0;
4450
4451         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4452         open->op_recall = 0;
4453         switch (open->op_claim_type) {
4454                 case NFS4_OPEN_CLAIM_PREVIOUS:
4455                         if (!cb_up)
4456                                 open->op_recall = 1;
4457                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4458                                 goto out_no_deleg;
4459                         break;
4460                 case NFS4_OPEN_CLAIM_NULL:
4461                 case NFS4_OPEN_CLAIM_FH:
4462                         /*
4463                          * Let's not give out any delegations till everyone's
4464                          * had the chance to reclaim theirs, *and* until
4465                          * NLM locks have all been reclaimed:
4466                          */
4467                         if (locks_in_grace(clp->net))
4468                                 goto out_no_deleg;
4469                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4470                                 goto out_no_deleg;
4471                         /*
4472                          * Also, if the file was opened for write or
4473                          * create, there's a good chance the client's
4474                          * about to write to it, resulting in an
4475                          * immediate recall (since we don't support
4476                          * write delegations):
4477                          */
4478                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4479                                 goto out_no_deleg;
4480                         if (open->op_create == NFS4_OPEN_CREATE)
4481                                 goto out_no_deleg;
4482                         break;
4483                 default:
4484                         goto out_no_deleg;
4485         }
4486         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4487         if (IS_ERR(dp))
4488                 goto out_no_deleg;
4489
4490         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4491
4492         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4493                 STATEID_VAL(&dp->dl_stid.sc_stateid));
4494         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4495         nfs4_put_stid(&dp->dl_stid);
4496         return;
4497 out_no_deleg:
4498         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4499         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4500             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4501                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4502                 open->op_recall = 1;
4503         }
4504
4505         /* 4.1 client asking for a delegation? */
4506         if (open->op_deleg_want)
4507                 nfsd4_open_deleg_none_ext(open, status);
4508         return;
4509 }
4510
4511 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4512                                         struct nfs4_delegation *dp)
4513 {
4514         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4515             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4516                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4517                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4518         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4519                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4520                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4521                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4522         }
4523         /* Otherwise the client must be confused wanting a delegation
4524          * it already has, therefore we don't return
4525          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4526          */
4527 }
4528
4529 __be32
4530 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4531 {
4532         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4533         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4534         struct nfs4_file *fp = NULL;
4535         struct nfs4_ol_stateid *stp = NULL;
4536         struct nfs4_delegation *dp = NULL;
4537         __be32 status;
4538         bool new_stp = false;
4539
4540         /*
4541          * Lookup file; if found, lookup stateid and check open request,
4542          * and check for delegations in the process of being recalled.
4543          * If not found, create the nfs4_file struct
4544          */
4545         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4546         if (fp != open->op_file) {
4547                 status = nfs4_check_deleg(cl, open, &dp);
4548                 if (status)
4549                         goto out;
4550                 stp = nfsd4_find_and_lock_existing_open(fp, open);
4551         } else {
4552                 open->op_file = NULL;
4553                 status = nfserr_bad_stateid;
4554                 if (nfsd4_is_deleg_cur(open))
4555                         goto out;
4556         }
4557
4558         if (!stp) {
4559                 stp = init_open_stateid(fp, open);
4560                 if (!open->op_stp)
4561                         new_stp = true;
4562         }
4563
4564         /*
4565          * OPEN the file, or upgrade an existing OPEN.
4566          * If truncate fails, the OPEN fails.
4567          *
4568          * stp is already locked.
4569          */
4570         if (!new_stp) {
4571                 /* Stateid was found, this is an OPEN upgrade */
4572                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4573                 if (status) {
4574                         mutex_unlock(&stp->st_mutex);
4575                         goto out;
4576                 }
4577         } else {
4578                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4579                 if (status) {
4580                         stp->st_stid.sc_type = NFS4_CLOSED_STID;
4581                         release_open_stateid(stp);
4582                         mutex_unlock(&stp->st_mutex);
4583                         goto out;
4584                 }
4585
4586                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4587                                                         open->op_odstate);
4588                 if (stp->st_clnt_odstate == open->op_odstate)
4589                         open->op_odstate = NULL;
4590         }
4591
4592         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4593         mutex_unlock(&stp->st_mutex);
4594
4595         if (nfsd4_has_session(&resp->cstate)) {
4596                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4597                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4598                         open->op_why_no_deleg = WND4_NOT_WANTED;
4599                         goto nodeleg;
4600                 }
4601         }
4602
4603         /*
4604         * Attempt to hand out a delegation. No error return, because the
4605         * OPEN succeeds even if we fail.
4606         */
4607         nfs4_open_delegation(current_fh, open, stp);
4608 nodeleg:
4609         status = nfs_ok;
4610
4611         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4612                 STATEID_VAL(&stp->st_stid.sc_stateid));
4613 out:
4614         /* 4.1 client trying to upgrade/downgrade delegation? */
4615         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4616             open->op_deleg_want)
4617                 nfsd4_deleg_xgrade_none_ext(open, dp);
4618
4619         if (fp)
4620                 put_nfs4_file(fp);
4621         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4622                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4623         /*
4624         * To finish the open response, we just need to set the rflags.
4625         */
4626         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4627         if (nfsd4_has_session(&resp->cstate))
4628                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
4629         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
4630                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4631
4632         if (dp)
4633                 nfs4_put_stid(&dp->dl_stid);
4634         if (stp)
4635                 nfs4_put_stid(&stp->st_stid);
4636
4637         return status;
4638 }
4639
4640 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4641                               struct nfsd4_open *open)
4642 {
4643         if (open->op_openowner) {
4644                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4645
4646                 nfsd4_cstate_assign_replay(cstate, so);
4647                 nfs4_put_stateowner(so);
4648         }
4649         if (open->op_file)
4650                 kmem_cache_free(file_slab, open->op_file);
4651         if (open->op_stp)
4652                 nfs4_put_stid(&open->op_stp->st_stid);
4653         if (open->op_odstate)
4654                 kmem_cache_free(odstate_slab, open->op_odstate);
4655 }
4656
4657 __be32
4658 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4659             union nfsd4_op_u *u)
4660 {
4661         clientid_t *clid = &u->renew;
4662         struct nfs4_client *clp;
4663         __be32 status;
4664         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4665
4666         dprintk("process_renew(%08x/%08x): starting\n",
4667                         clid->cl_boot, clid->cl_id);
4668         status = lookup_clientid(clid, cstate, nn);
4669         if (status)
4670                 goto out;
4671         clp = cstate->clp;
4672         status = nfserr_cb_path_down;
4673         if (!list_empty(&clp->cl_delegations)
4674                         && clp->cl_cb_state != NFSD4_CB_UP)
4675                 goto out;
4676         status = nfs_ok;
4677 out:
4678         return status;
4679 }
4680
4681 void
4682 nfsd4_end_grace(struct nfsd_net *nn)
4683 {
4684         /* do nothing if grace period already ended */
4685         if (nn->grace_ended)
4686                 return;
4687
4688         dprintk("NFSD: end of grace period\n");
4689         nn->grace_ended = true;
4690         /*
4691          * If the server goes down again right now, an NFSv4
4692          * client will still be allowed to reclaim after it comes back up,
4693          * even if it hasn't yet had a chance to reclaim state this time.
4694          *
4695          */
4696         nfsd4_record_grace_done(nn);
4697         /*
4698          * At this point, NFSv4 clients can still reclaim.  But if the
4699          * server crashes, any that have not yet reclaimed will be out
4700          * of luck on the next boot.
4701          *
4702          * (NFSv4.1+ clients are considered to have reclaimed once they
4703          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
4704          * have reclaimed after their first OPEN.)
4705          */
4706         locks_end_grace(&nn->nfsd4_manager);
4707         /*
4708          * At this point, and once lockd and/or any other containers
4709          * exit their grace period, further reclaims will fail and
4710          * regular locking can resume.
4711          */
4712 }
4713
4714 /*
4715  * If we've waited a lease period but there are still clients trying to
4716  * reclaim, wait a little longer to give them a chance to finish.
4717  */
4718 static bool clients_still_reclaiming(struct nfsd_net *nn)
4719 {
4720         unsigned long now = get_seconds();
4721         unsigned long double_grace_period_end = nn->boot_time +
4722                                                 2 * nn->nfsd4_lease;
4723
4724         if (!nn->somebody_reclaimed)
4725                 return false;
4726         nn->somebody_reclaimed = false;
4727         /*
4728          * If we've given them *two* lease times to reclaim, and they're
4729          * still not done, give up:
4730          */
4731         if (time_after(now, double_grace_period_end))
4732                 return false;
4733         return true;
4734 }
4735
4736 static time_t
4737 nfs4_laundromat(struct nfsd_net *nn)
4738 {
4739         struct nfs4_client *clp;
4740         struct nfs4_openowner *oo;
4741         struct nfs4_delegation *dp;
4742         struct nfs4_ol_stateid *stp;
4743         struct nfsd4_blocked_lock *nbl;
4744         struct list_head *pos, *next, reaplist;
4745         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4746         time_t t, new_timeo = nn->nfsd4_lease;
4747
4748         dprintk("NFSD: laundromat service - starting\n");
4749
4750         if (clients_still_reclaiming(nn)) {
4751                 new_timeo = 0;
4752                 goto out;
4753         }
4754         nfsd4_end_grace(nn);
4755         INIT_LIST_HEAD(&reaplist);
4756         spin_lock(&nn->client_lock);
4757         list_for_each_safe(pos, next, &nn->client_lru) {
4758                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4759                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4760                         t = clp->cl_time - cutoff;
4761                         new_timeo = min(new_timeo, t);
4762                         break;
4763                 }
4764                 if (mark_client_expired_locked(clp)) {
4765                         dprintk("NFSD: client in use (clientid %08x)\n",
4766                                 clp->cl_clientid.cl_id);
4767                         continue;
4768                 }
4769                 list_add(&clp->cl_lru, &reaplist);
4770         }
4771         spin_unlock(&nn->client_lock);
4772         list_for_each_safe(pos, next, &reaplist) {
4773                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4774                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4775                         clp->cl_clientid.cl_id);
4776                 list_del_init(&clp->cl_lru);
4777                 expire_client(clp);
4778         }
4779         spin_lock(&state_lock);
4780         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4781                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4782                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4783                         t = dp->dl_time - cutoff;
4784                         new_timeo = min(new_timeo, t);
4785                         break;
4786                 }
4787                 WARN_ON(!unhash_delegation_locked(dp));
4788                 list_add(&dp->dl_recall_lru, &reaplist);
4789         }
4790         spin_unlock(&state_lock);
4791         while (!list_empty(&reaplist)) {
4792                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4793                                         dl_recall_lru);
4794                 list_del_init(&dp->dl_recall_lru);
4795                 revoke_delegation(dp);
4796         }
4797
4798         spin_lock(&nn->client_lock);
4799         while (!list_empty(&nn->close_lru)) {
4800                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4801                                         oo_close_lru);
4802                 if (time_after((unsigned long)oo->oo_time,
4803                                (unsigned long)cutoff)) {
4804                         t = oo->oo_time - cutoff;
4805                         new_timeo = min(new_timeo, t);
4806                         break;
4807                 }
4808                 list_del_init(&oo->oo_close_lru);
4809                 stp = oo->oo_last_closed_stid;
4810                 oo->oo_last_closed_stid = NULL;
4811                 spin_unlock(&nn->client_lock);
4812                 nfs4_put_stid(&stp->st_stid);
4813                 spin_lock(&nn->client_lock);
4814         }
4815         spin_unlock(&nn->client_lock);
4816
4817         /*
4818          * It's possible for a client to try and acquire an already held lock
4819          * that is being held for a long time, and then lose interest in it.
4820          * So, we clean out any un-revisited request after a lease period
4821          * under the assumption that the client is no longer interested.
4822          *
4823          * RFC5661, sec. 9.6 states that the client must not rely on getting
4824          * notifications and must continue to poll for locks, even when the
4825          * server supports them. Thus this shouldn't lead to clients blocking
4826          * indefinitely once the lock does become free.
4827          */
4828         BUG_ON(!list_empty(&reaplist));
4829         spin_lock(&nn->blocked_locks_lock);
4830         while (!list_empty(&nn->blocked_locks_lru)) {
4831                 nbl = list_first_entry(&nn->blocked_locks_lru,
4832                                         struct nfsd4_blocked_lock, nbl_lru);
4833                 if (time_after((unsigned long)nbl->nbl_time,
4834                                (unsigned long)cutoff)) {
4835                         t = nbl->nbl_time - cutoff;
4836                         new_timeo = min(new_timeo, t);
4837                         break;
4838                 }
4839                 list_move(&nbl->nbl_lru, &reaplist);
4840                 list_del_init(&nbl->nbl_list);
4841         }
4842         spin_unlock(&nn->blocked_locks_lock);
4843
4844         while (!list_empty(&reaplist)) {
4845                 nbl = list_first_entry(&reaplist,
4846                                         struct nfsd4_blocked_lock, nbl_lru);
4847                 list_del_init(&nbl->nbl_lru);
4848                 posix_unblock_lock(&nbl->nbl_lock);
4849                 free_blocked_lock(nbl);
4850         }
4851 out:
4852         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4853         return new_timeo;
4854 }
4855
4856 static struct workqueue_struct *laundry_wq;
4857 static void laundromat_main(struct work_struct *);
4858
4859 static void
4860 laundromat_main(struct work_struct *laundry)
4861 {
4862         time_t t;
4863         struct delayed_work *dwork = to_delayed_work(laundry);
4864         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4865                                            laundromat_work);
4866
4867         t = nfs4_laundromat(nn);
4868         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4869         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4870 }
4871
4872 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4873 {
4874         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4875                 return nfserr_bad_stateid;
4876         return nfs_ok;
4877 }
4878
4879 static inline int
4880 access_permit_read(struct nfs4_ol_stateid *stp)
4881 {
4882         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4883                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4884                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4885 }
4886
4887 static inline int
4888 access_permit_write(struct nfs4_ol_stateid *stp)
4889 {
4890         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4891                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4892 }
4893
4894 static
4895 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4896 {
4897         __be32 status = nfserr_openmode;
4898
4899         /* For lock stateid's, we test the parent open, not the lock: */
4900         if (stp->st_openstp)
4901                 stp = stp->st_openstp;
4902         if ((flags & WR_STATE) && !access_permit_write(stp))
4903                 goto out;
4904         if ((flags & RD_STATE) && !access_permit_read(stp))
4905                 goto out;
4906         status = nfs_ok;
4907 out:
4908         return status;
4909 }
4910
4911 static inline __be32
4912 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4913 {
4914         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4915                 return nfs_ok;
4916         else if (opens_in_grace(net)) {
4917                 /* Answer in remaining cases depends on existence of
4918                  * conflicting state; so we must wait out the grace period. */
4919                 return nfserr_grace;
4920         } else if (flags & WR_STATE)
4921                 return nfs4_share_conflict(current_fh,
4922                                 NFS4_SHARE_DENY_WRITE);
4923         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4924                 return nfs4_share_conflict(current_fh,
4925                                 NFS4_SHARE_DENY_READ);
4926 }
4927
4928 /*
4929  * Allow READ/WRITE during grace period on recovered state only for files
4930  * that are not able to provide mandatory locking.
4931  */
4932 static inline int
4933 grace_disallows_io(struct net *net, struct inode *inode)
4934 {
4935         return opens_in_grace(net) && mandatory_lock(inode);
4936 }
4937
4938 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4939 {
4940         /*
4941          * When sessions are used the stateid generation number is ignored
4942          * when it is zero.
4943          */
4944         if (has_session && in->si_generation == 0)
4945                 return nfs_ok;
4946
4947         if (in->si_generation == ref->si_generation)
4948                 return nfs_ok;
4949
4950         /* If the client sends us a stateid from the future, it's buggy: */
4951         if (nfsd4_stateid_generation_after(in, ref))
4952                 return nfserr_bad_stateid;
4953         /*
4954          * However, we could see a stateid from the past, even from a
4955          * non-buggy client.  For example, if the client sends a lock
4956          * while some IO is outstanding, the lock may bump si_generation
4957          * while the IO is still in flight.  The client could avoid that
4958          * situation by waiting for responses on all the IO requests,
4959          * but better performance may result in retrying IO that
4960          * receives an old_stateid error if requests are rarely
4961          * reordered in flight:
4962          */
4963         return nfserr_old_stateid;
4964 }
4965
4966 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
4967 {
4968         __be32 ret;
4969
4970         spin_lock(&s->sc_lock);
4971         ret = nfsd4_verify_open_stid(s);
4972         if (ret == nfs_ok)
4973                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
4974         spin_unlock(&s->sc_lock);
4975         return ret;
4976 }
4977
4978 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4979 {
4980         if (ols->st_stateowner->so_is_open_owner &&
4981             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4982                 return nfserr_bad_stateid;
4983         return nfs_ok;
4984 }
4985
4986 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4987 {
4988         struct nfs4_stid *s;
4989         __be32 status = nfserr_bad_stateid;
4990
4991         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
4992                 CLOSE_STATEID(stateid))
4993                 return status;
4994         /* Client debugging aid. */
4995         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4996                 char addr_str[INET6_ADDRSTRLEN];
4997                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4998                                  sizeof(addr_str));
4999                 pr_warn_ratelimited("NFSD: client %s testing state ID "
5000                                         "with incorrect client ID\n", addr_str);
5001                 return status;
5002         }
5003         spin_lock(&cl->cl_lock);
5004         s = find_stateid_locked(cl, stateid);
5005         if (!s)
5006                 goto out_unlock;
5007         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
5008         if (status)
5009                 goto out_unlock;
5010         switch (s->sc_type) {
5011         case NFS4_DELEG_STID:
5012                 status = nfs_ok;
5013                 break;
5014         case NFS4_REVOKED_DELEG_STID:
5015                 status = nfserr_deleg_revoked;
5016                 break;
5017         case NFS4_OPEN_STID:
5018         case NFS4_LOCK_STID:
5019                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5020                 break;
5021         default:
5022                 printk("unknown stateid type %x\n", s->sc_type);
5023                 /* Fallthrough */
5024         case NFS4_CLOSED_STID:
5025         case NFS4_CLOSED_DELEG_STID:
5026                 status = nfserr_bad_stateid;
5027         }
5028 out_unlock:
5029         spin_unlock(&cl->cl_lock);
5030         return status;
5031 }
5032
5033 __be32
5034 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5035                      stateid_t *stateid, unsigned char typemask,
5036                      struct nfs4_stid **s, struct nfsd_net *nn)
5037 {
5038         __be32 status;
5039         bool return_revoked = false;
5040
5041         /*
5042          *  only return revoked delegations if explicitly asked.
5043          *  otherwise we report revoked or bad_stateid status.
5044          */
5045         if (typemask & NFS4_REVOKED_DELEG_STID)
5046                 return_revoked = true;
5047         else if (typemask & NFS4_DELEG_STID)
5048                 typemask |= NFS4_REVOKED_DELEG_STID;
5049
5050         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5051                 CLOSE_STATEID(stateid))
5052                 return nfserr_bad_stateid;
5053         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
5054         if (status == nfserr_stale_clientid) {
5055                 if (cstate->session)
5056                         return nfserr_bad_stateid;
5057                 return nfserr_stale_stateid;
5058         }
5059         if (status)
5060                 return status;
5061         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5062         if (!*s)
5063                 return nfserr_bad_stateid;
5064         if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5065                 nfs4_put_stid(*s);
5066                 if (cstate->minorversion)
5067                         return nfserr_deleg_revoked;
5068                 return nfserr_bad_stateid;
5069         }
5070         return nfs_ok;
5071 }
5072
5073 static struct file *
5074 nfs4_find_file(struct nfs4_stid *s, int flags)
5075 {
5076         if (!s)
5077                 return NULL;
5078
5079         switch (s->sc_type) {
5080         case NFS4_DELEG_STID:
5081                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5082                         return NULL;
5083                 return get_file(s->sc_file->fi_deleg_file);
5084         case NFS4_OPEN_STID:
5085         case NFS4_LOCK_STID:
5086                 if (flags & RD_STATE)
5087                         return find_readable_file(s->sc_file);
5088                 else
5089                         return find_writeable_file(s->sc_file);
5090                 break;
5091         }
5092
5093         return NULL;
5094 }
5095
5096 static __be32
5097 nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
5098 {
5099         __be32 status;
5100
5101         status = nfsd4_check_openowner_confirmed(ols);
5102         if (status)
5103                 return status;
5104         return nfs4_check_openmode(ols, flags);
5105 }
5106
5107 static __be32
5108 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5109                 struct file **filpp, bool *tmp_file, int flags)
5110 {
5111         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5112         struct file *file;
5113         __be32 status;
5114
5115         file = nfs4_find_file(s, flags);
5116         if (file) {
5117                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5118                                 acc | NFSD_MAY_OWNER_OVERRIDE);
5119                 if (status) {
5120                         fput(file);
5121                         return status;
5122                 }
5123
5124                 *filpp = file;
5125         } else {
5126                 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
5127                 if (status)
5128                         return status;
5129
5130                 if (tmp_file)
5131                         *tmp_file = true;
5132         }
5133
5134         return 0;
5135 }
5136
5137 /*
5138  * Checks for stateid operations
5139  */
5140 __be32
5141 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5142                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5143                 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
5144 {
5145         struct inode *ino = d_inode(fhp->fh_dentry);
5146         struct net *net = SVC_NET(rqstp);
5147         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5148         struct nfs4_stid *s = NULL;
5149         __be32 status;
5150
5151         if (filpp)
5152                 *filpp = NULL;
5153         if (tmp_file)
5154                 *tmp_file = false;
5155
5156         if (grace_disallows_io(net, ino))
5157                 return nfserr_grace;
5158
5159         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5160                 status = check_special_stateids(net, fhp, stateid, flags);
5161                 goto done;
5162         }
5163
5164         status = nfsd4_lookup_stateid(cstate, stateid,
5165                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5166                                 &s, nn);
5167         if (status)
5168                 return status;
5169         status = nfsd4_stid_check_stateid_generation(stateid, s,
5170                         nfsd4_has_session(cstate));
5171         if (status)
5172                 goto out;
5173
5174         switch (s->sc_type) {
5175         case NFS4_DELEG_STID:
5176                 status = nfs4_check_delegmode(delegstateid(s), flags);
5177                 break;
5178         case NFS4_OPEN_STID:
5179         case NFS4_LOCK_STID:
5180                 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
5181                 break;
5182         default:
5183                 status = nfserr_bad_stateid;
5184                 break;
5185         }
5186         if (status)
5187                 goto out;
5188         status = nfs4_check_fh(fhp, s);
5189
5190 done:
5191         if (!status && filpp)
5192                 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
5193 out:
5194         if (s)
5195                 nfs4_put_stid(s);
5196         return status;
5197 }
5198
5199 /*
5200  * Test if the stateid is valid
5201  */
5202 __be32
5203 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5204                    union nfsd4_op_u *u)
5205 {
5206         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5207         struct nfsd4_test_stateid_id *stateid;
5208         struct nfs4_client *cl = cstate->session->se_client;
5209
5210         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5211                 stateid->ts_id_status =
5212                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5213
5214         return nfs_ok;
5215 }
5216
5217 static __be32
5218 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5219 {
5220         struct nfs4_ol_stateid *stp = openlockstateid(s);
5221         __be32 ret;
5222
5223         ret = nfsd4_lock_ol_stateid(stp);
5224         if (ret)
5225                 goto out_put_stid;
5226
5227         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5228         if (ret)
5229                 goto out;
5230
5231         ret = nfserr_locks_held;
5232         if (check_for_locks(stp->st_stid.sc_file,
5233                             lockowner(stp->st_stateowner)))
5234                 goto out;
5235
5236         release_lock_stateid(stp);
5237         ret = nfs_ok;
5238
5239 out:
5240         mutex_unlock(&stp->st_mutex);
5241 out_put_stid:
5242         nfs4_put_stid(s);
5243         return ret;
5244 }
5245
5246 __be32
5247 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5248                    union nfsd4_op_u *u)
5249 {
5250         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5251         stateid_t *stateid = &free_stateid->fr_stateid;
5252         struct nfs4_stid *s;
5253         struct nfs4_delegation *dp;
5254         struct nfs4_client *cl = cstate->session->se_client;
5255         __be32 ret = nfserr_bad_stateid;
5256
5257         spin_lock(&cl->cl_lock);
5258         s = find_stateid_locked(cl, stateid);
5259         if (!s)
5260                 goto out_unlock;
5261         spin_lock(&s->sc_lock);
5262         switch (s->sc_type) {
5263         case NFS4_DELEG_STID:
5264                 ret = nfserr_locks_held;
5265                 break;
5266         case NFS4_OPEN_STID:
5267                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5268                 if (ret)
5269                         break;
5270                 ret = nfserr_locks_held;
5271                 break;
5272         case NFS4_LOCK_STID:
5273                 spin_unlock(&s->sc_lock);
5274                 refcount_inc(&s->sc_count);
5275                 spin_unlock(&cl->cl_lock);
5276                 ret = nfsd4_free_lock_stateid(stateid, s);
5277                 goto out;
5278         case NFS4_REVOKED_DELEG_STID:
5279                 spin_unlock(&s->sc_lock);
5280                 dp = delegstateid(s);
5281                 list_del_init(&dp->dl_recall_lru);
5282                 spin_unlock(&cl->cl_lock);
5283                 nfs4_put_stid(s);
5284                 ret = nfs_ok;
5285                 goto out;
5286         /* Default falls through and returns nfserr_bad_stateid */
5287         }
5288         spin_unlock(&s->sc_lock);
5289 out_unlock:
5290         spin_unlock(&cl->cl_lock);
5291 out:
5292         return ret;
5293 }
5294
5295 static inline int
5296 setlkflg (int type)
5297 {
5298         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5299                 RD_STATE : WR_STATE;
5300 }
5301
5302 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5303 {
5304         struct svc_fh *current_fh = &cstate->current_fh;
5305         struct nfs4_stateowner *sop = stp->st_stateowner;
5306         __be32 status;
5307
5308         status = nfsd4_check_seqid(cstate, sop, seqid);
5309         if (status)
5310                 return status;
5311         status = nfsd4_lock_ol_stateid(stp);
5312         if (status != nfs_ok)
5313                 return status;
5314         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5315         if (status == nfs_ok)
5316                 status = nfs4_check_fh(current_fh, &stp->st_stid);
5317         if (status != nfs_ok)
5318                 mutex_unlock(&stp->st_mutex);
5319         return status;
5320 }
5321
5322 /*
5323  * Checks for sequence id mutating operations.
5324  */
5325 static __be32
5326 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5327                          stateid_t *stateid, char typemask,
5328                          struct nfs4_ol_stateid **stpp,
5329                          struct nfsd_net *nn)
5330 {
5331         __be32 status;
5332         struct nfs4_stid *s;
5333         struct nfs4_ol_stateid *stp = NULL;
5334
5335         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5336                 seqid, STATEID_VAL(stateid));
5337
5338         *stpp = NULL;
5339         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5340         if (status)
5341                 return status;
5342         stp = openlockstateid(s);
5343         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5344
5345         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5346         if (!status)
5347                 *stpp = stp;
5348         else
5349                 nfs4_put_stid(&stp->st_stid);
5350         return status;
5351 }
5352
5353 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5354                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5355 {
5356         __be32 status;
5357         struct nfs4_openowner *oo;
5358         struct nfs4_ol_stateid *stp;
5359
5360         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5361                                                 NFS4_OPEN_STID, &stp, nn);
5362         if (status)
5363                 return status;
5364         oo = openowner(stp->st_stateowner);
5365         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5366                 mutex_unlock(&stp->st_mutex);
5367                 nfs4_put_stid(&stp->st_stid);
5368                 return nfserr_bad_stateid;
5369         }
5370         *stpp = stp;
5371         return nfs_ok;
5372 }
5373
5374 __be32
5375 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5376                    union nfsd4_op_u *u)
5377 {
5378         struct nfsd4_open_confirm *oc = &u->open_confirm;
5379         __be32 status;
5380         struct nfs4_openowner *oo;
5381         struct nfs4_ol_stateid *stp;
5382         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5383
5384         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5385                         cstate->current_fh.fh_dentry);
5386
5387         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5388         if (status)
5389                 return status;
5390
5391         status = nfs4_preprocess_seqid_op(cstate,
5392                                         oc->oc_seqid, &oc->oc_req_stateid,
5393                                         NFS4_OPEN_STID, &stp, nn);
5394         if (status)
5395                 goto out;
5396         oo = openowner(stp->st_stateowner);
5397         status = nfserr_bad_stateid;
5398         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5399                 mutex_unlock(&stp->st_mutex);
5400                 goto put_stateid;
5401         }
5402         oo->oo_flags |= NFS4_OO_CONFIRMED;
5403         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5404         mutex_unlock(&stp->st_mutex);
5405         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5406                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5407
5408         nfsd4_client_record_create(oo->oo_owner.so_client);
5409         status = nfs_ok;
5410 put_stateid:
5411         nfs4_put_stid(&stp->st_stid);
5412 out:
5413         nfsd4_bump_seqid(cstate, status);
5414         return status;
5415 }
5416
5417 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5418 {
5419         if (!test_access(access, stp))
5420                 return;
5421         nfs4_file_put_access(stp->st_stid.sc_file, access);
5422         clear_access(access, stp);
5423 }
5424
5425 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5426 {
5427         switch (to_access) {
5428         case NFS4_SHARE_ACCESS_READ:
5429                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5430                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5431                 break;
5432         case NFS4_SHARE_ACCESS_WRITE:
5433                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5434                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5435                 break;
5436         case NFS4_SHARE_ACCESS_BOTH:
5437                 break;
5438         default:
5439                 WARN_ON_ONCE(1);
5440         }
5441 }
5442
5443 __be32
5444 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5445                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
5446 {
5447         struct nfsd4_open_downgrade *od = &u->open_downgrade;
5448         __be32 status;
5449         struct nfs4_ol_stateid *stp;
5450         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5451
5452         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5453                         cstate->current_fh.fh_dentry);
5454
5455         /* We don't yet support WANT bits: */
5456         if (od->od_deleg_want)
5457                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5458                         od->od_deleg_want);
5459
5460         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5461                                         &od->od_stateid, &stp, nn);
5462         if (status)
5463                 goto out;
5464         status = nfserr_inval;
5465         if (!test_access(od->od_share_access, stp)) {
5466                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5467                         stp->st_access_bmap, od->od_share_access);
5468                 goto put_stateid;
5469         }
5470         if (!test_deny(od->od_share_deny, stp)) {
5471                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5472                         stp->st_deny_bmap, od->od_share_deny);
5473                 goto put_stateid;
5474         }
5475         nfs4_stateid_downgrade(stp, od->od_share_access);
5476         reset_union_bmap_deny(od->od_share_deny, stp);
5477         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5478         status = nfs_ok;
5479 put_stateid:
5480         mutex_unlock(&stp->st_mutex);
5481         nfs4_put_stid(&stp->st_stid);
5482 out:
5483         nfsd4_bump_seqid(cstate, status);
5484         return status;
5485 }
5486
5487 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5488 {
5489         struct nfs4_client *clp = s->st_stid.sc_client;
5490         bool unhashed;
5491         LIST_HEAD(reaplist);
5492
5493         spin_lock(&clp->cl_lock);
5494         unhashed = unhash_open_stateid(s, &reaplist);
5495
5496         if (clp->cl_minorversion) {
5497                 if (unhashed)
5498                         put_ol_stateid_locked(s, &reaplist);
5499                 spin_unlock(&clp->cl_lock);
5500                 free_ol_stateid_reaplist(&reaplist);
5501         } else {
5502                 spin_unlock(&clp->cl_lock);
5503                 free_ol_stateid_reaplist(&reaplist);
5504                 if (unhashed)
5505                         move_to_close_lru(s, clp->net);
5506         }
5507 }
5508
5509 /*
5510  * nfs4_unlock_state() called after encode
5511  */
5512 __be32
5513 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5514                 union nfsd4_op_u *u)
5515 {
5516         struct nfsd4_close *close = &u->close;
5517         __be32 status;
5518         struct nfs4_ol_stateid *stp;
5519         struct net *net = SVC_NET(rqstp);
5520         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5521
5522         dprintk("NFSD: nfsd4_close on file %pd\n",
5523                         cstate->current_fh.fh_dentry);
5524
5525         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5526                                         &close->cl_stateid,
5527                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
5528                                         &stp, nn);
5529         nfsd4_bump_seqid(cstate, status);
5530         if (status)
5531                 goto out;
5532
5533         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5534
5535         /*
5536          * Technically we don't _really_ have to increment or copy it, since
5537          * it should just be gone after this operation and we clobber the
5538          * copied value below, but we continue to do so here just to ensure
5539          * that racing ops see that there was a state change.
5540          */
5541         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5542
5543         nfsd4_close_open_stateid(stp);
5544         mutex_unlock(&stp->st_mutex);
5545
5546         /* v4.1+ suggests that we send a special stateid in here, since the
5547          * clients should just ignore this anyway. Since this is not useful
5548          * for v4.0 clients either, we set it to the special close_stateid
5549          * universally.
5550          *
5551          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
5552          */
5553         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
5554
5555         /* put reference from nfs4_preprocess_seqid_op */
5556         nfs4_put_stid(&stp->st_stid);
5557 out:
5558         return status;
5559 }
5560
5561 __be32
5562 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5563                   union nfsd4_op_u *u)
5564 {
5565         struct nfsd4_delegreturn *dr = &u->delegreturn;
5566         struct nfs4_delegation *dp;
5567         stateid_t *stateid = &dr->dr_stateid;
5568         struct nfs4_stid *s;
5569         __be32 status;
5570         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5571
5572         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5573                 return status;
5574
5575         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5576         if (status)
5577                 goto out;
5578         dp = delegstateid(s);
5579         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
5580         if (status)
5581                 goto put_stateid;
5582
5583         destroy_delegation(dp);
5584 put_stateid:
5585         nfs4_put_stid(&dp->dl_stid);
5586 out:
5587         return status;
5588 }
5589
5590 static inline u64
5591 end_offset(u64 start, u64 len)
5592 {
5593         u64 end;
5594
5595         end = start + len;
5596         return end >= start ? end: NFS4_MAX_UINT64;
5597 }
5598
5599 /* last octet in a range */
5600 static inline u64
5601 last_byte_offset(u64 start, u64 len)
5602 {
5603         u64 end;
5604
5605         WARN_ON_ONCE(!len);
5606         end = start + len;
5607         return end > start ? end - 1: NFS4_MAX_UINT64;
5608 }
5609
5610 /*
5611  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5612  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5613  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
5614  * locking, this prevents us from being completely protocol-compliant.  The
5615  * real solution to this problem is to start using unsigned file offsets in
5616  * the VFS, but this is a very deep change!
5617  */
5618 static inline void
5619 nfs4_transform_lock_offset(struct file_lock *lock)
5620 {
5621         if (lock->fl_start < 0)
5622                 lock->fl_start = OFFSET_MAX;
5623         if (lock->fl_end < 0)
5624                 lock->fl_end = OFFSET_MAX;
5625 }
5626
5627 static fl_owner_t
5628 nfsd4_fl_get_owner(fl_owner_t owner)
5629 {
5630         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5631
5632         nfs4_get_stateowner(&lo->lo_owner);
5633         return owner;
5634 }
5635
5636 static void
5637 nfsd4_fl_put_owner(fl_owner_t owner)
5638 {
5639         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5640
5641         if (lo)
5642                 nfs4_put_stateowner(&lo->lo_owner);
5643 }
5644
5645 static void
5646 nfsd4_lm_notify(struct file_lock *fl)
5647 {
5648         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
5649         struct net                      *net = lo->lo_owner.so_client->net;
5650         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
5651         struct nfsd4_blocked_lock       *nbl = container_of(fl,
5652                                                 struct nfsd4_blocked_lock, nbl_lock);
5653         bool queue = false;
5654
5655         /* An empty list means that something else is going to be using it */
5656         spin_lock(&nn->blocked_locks_lock);
5657         if (!list_empty(&nbl->nbl_list)) {
5658                 list_del_init(&nbl->nbl_list);
5659                 list_del_init(&nbl->nbl_lru);
5660                 queue = true;
5661         }
5662         spin_unlock(&nn->blocked_locks_lock);
5663
5664         if (queue)
5665                 nfsd4_run_cb(&nbl->nbl_cb);
5666 }
5667
5668 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
5669         .lm_notify = nfsd4_lm_notify,
5670         .lm_get_owner = nfsd4_fl_get_owner,
5671         .lm_put_owner = nfsd4_fl_put_owner,
5672 };
5673
5674 static inline void
5675 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5676 {
5677         struct nfs4_lockowner *lo;
5678
5679         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5680                 lo = (struct nfs4_lockowner *) fl->fl_owner;
5681                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5682                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
5683                 if (!deny->ld_owner.data)
5684                         /* We just don't care that much */
5685                         goto nevermind;
5686                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5687                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5688         } else {
5689 nevermind:
5690                 deny->ld_owner.len = 0;
5691                 deny->ld_owner.data = NULL;
5692                 deny->ld_clientid.cl_boot = 0;
5693                 deny->ld_clientid.cl_id = 0;
5694         }
5695         deny->ld_start = fl->fl_start;
5696         deny->ld_length = NFS4_MAX_UINT64;
5697         if (fl->fl_end != NFS4_MAX_UINT64)
5698                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5699         deny->ld_type = NFS4_READ_LT;
5700         if (fl->fl_type != F_RDLCK)
5701                 deny->ld_type = NFS4_WRITE_LT;
5702 }
5703
5704 static struct nfs4_lockowner *
5705 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5706 {
5707         unsigned int strhashval = ownerstr_hashval(owner);
5708         struct nfs4_stateowner *so;
5709
5710         lockdep_assert_held(&clp->cl_lock);
5711
5712         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5713                             so_strhash) {
5714                 if (so->so_is_open_owner)
5715                         continue;
5716                 if (same_owner_str(so, owner))
5717                         return lockowner(nfs4_get_stateowner(so));
5718         }
5719         return NULL;
5720 }
5721
5722 static struct nfs4_lockowner *
5723 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5724 {
5725         struct nfs4_lockowner *lo;
5726
5727         spin_lock(&clp->cl_lock);
5728         lo = find_lockowner_str_locked(clp, owner);
5729         spin_unlock(&clp->cl_lock);
5730         return lo;
5731 }
5732
5733 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5734 {
5735         unhash_lockowner_locked(lockowner(sop));
5736 }
5737
5738 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5739 {
5740         struct nfs4_lockowner *lo = lockowner(sop);
5741
5742         kmem_cache_free(lockowner_slab, lo);
5743 }
5744
5745 static const struct nfs4_stateowner_operations lockowner_ops = {
5746         .so_unhash =    nfs4_unhash_lockowner,
5747         .so_free =      nfs4_free_lockowner,
5748 };
5749
5750 /*
5751  * Alloc a lock owner structure.
5752  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5753  * occurred.
5754  *
5755  * strhashval = ownerstr_hashval
5756  */
5757 static struct nfs4_lockowner *
5758 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5759                            struct nfs4_ol_stateid *open_stp,
5760                            struct nfsd4_lock *lock)
5761 {
5762         struct nfs4_lockowner *lo, *ret;
5763
5764         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5765         if (!lo)
5766                 return NULL;
5767         INIT_LIST_HEAD(&lo->lo_blocked);
5768         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5769         lo->lo_owner.so_is_open_owner = 0;
5770         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5771         lo->lo_owner.so_ops = &lockowner_ops;
5772         spin_lock(&clp->cl_lock);
5773         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5774         if (ret == NULL) {
5775                 list_add(&lo->lo_owner.so_strhash,
5776                          &clp->cl_ownerstr_hashtbl[strhashval]);
5777                 ret = lo;
5778         } else
5779                 nfs4_free_stateowner(&lo->lo_owner);
5780
5781         spin_unlock(&clp->cl_lock);
5782         return ret;
5783 }
5784
5785 static struct nfs4_ol_stateid *
5786 find_lock_stateid(const struct nfs4_lockowner *lo,
5787                   const struct nfs4_ol_stateid *ost)
5788 {
5789         struct nfs4_ol_stateid *lst;
5790
5791         lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
5792
5793         /* If ost is not hashed, ost->st_locks will not be valid */
5794         if (!nfs4_ol_stateid_unhashed(ost))
5795                 list_for_each_entry(lst, &ost->st_locks, st_locks) {
5796                         if (lst->st_stateowner == &lo->lo_owner) {
5797                                 refcount_inc(&lst->st_stid.sc_count);
5798                                 return lst;
5799                         }
5800                 }
5801         return NULL;
5802 }
5803
5804 static struct nfs4_ol_stateid *
5805 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5806                   struct nfs4_file *fp, struct inode *inode,
5807                   struct nfs4_ol_stateid *open_stp)
5808 {
5809         struct nfs4_client *clp = lo->lo_owner.so_client;
5810         struct nfs4_ol_stateid *retstp;
5811
5812         mutex_init(&stp->st_mutex);
5813         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
5814 retry:
5815         spin_lock(&clp->cl_lock);
5816         if (nfs4_ol_stateid_unhashed(open_stp))
5817                 goto out_close;
5818         retstp = find_lock_stateid(lo, open_stp);
5819         if (retstp)
5820                 goto out_found;
5821         refcount_inc(&stp->st_stid.sc_count);
5822         stp->st_stid.sc_type = NFS4_LOCK_STID;
5823         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5824         get_nfs4_file(fp);
5825         stp->st_stid.sc_file = fp;
5826         stp->st_access_bmap = 0;
5827         stp->st_deny_bmap = open_stp->st_deny_bmap;
5828         stp->st_openstp = open_stp;
5829         spin_lock(&fp->fi_lock);
5830         list_add(&stp->st_locks, &open_stp->st_locks);
5831         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5832         list_add(&stp->st_perfile, &fp->fi_stateids);
5833         spin_unlock(&fp->fi_lock);
5834         spin_unlock(&clp->cl_lock);
5835         return stp;
5836 out_found:
5837         spin_unlock(&clp->cl_lock);
5838         if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
5839                 nfs4_put_stid(&retstp->st_stid);
5840                 goto retry;
5841         }
5842         /* To keep mutex tracking happy */
5843         mutex_unlock(&stp->st_mutex);
5844         return retstp;
5845 out_close:
5846         spin_unlock(&clp->cl_lock);
5847         mutex_unlock(&stp->st_mutex);
5848         return NULL;
5849 }
5850
5851 static struct nfs4_ol_stateid *
5852 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5853                             struct inode *inode, struct nfs4_ol_stateid *ost,
5854                             bool *new)
5855 {
5856         struct nfs4_stid *ns = NULL;
5857         struct nfs4_ol_stateid *lst;
5858         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5859         struct nfs4_client *clp = oo->oo_owner.so_client;
5860
5861         *new = false;
5862         spin_lock(&clp->cl_lock);
5863         lst = find_lock_stateid(lo, ost);
5864         spin_unlock(&clp->cl_lock);
5865         if (lst != NULL) {
5866                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
5867                         goto out;
5868                 nfs4_put_stid(&lst->st_stid);
5869         }
5870         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
5871         if (ns == NULL)
5872                 return NULL;
5873
5874         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
5875         if (lst == openlockstateid(ns))
5876                 *new = true;
5877         else
5878                 nfs4_put_stid(ns);
5879 out:
5880         return lst;
5881 }
5882
5883 static int
5884 check_lock_length(u64 offset, u64 length)
5885 {
5886         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5887                 (length > ~offset)));
5888 }
5889
5890 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5891 {
5892         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5893
5894         lockdep_assert_held(&fp->fi_lock);
5895
5896         if (test_access(access, lock_stp))
5897                 return;
5898         __nfs4_file_get_access(fp, access);
5899         set_access(access, lock_stp);
5900 }
5901
5902 static __be32
5903 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5904                             struct nfs4_ol_stateid *ost,
5905                             struct nfsd4_lock *lock,
5906                             struct nfs4_ol_stateid **plst, bool *new)
5907 {
5908         __be32 status;
5909         struct nfs4_file *fi = ost->st_stid.sc_file;
5910         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5911         struct nfs4_client *cl = oo->oo_owner.so_client;
5912         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5913         struct nfs4_lockowner *lo;
5914         struct nfs4_ol_stateid *lst;
5915         unsigned int strhashval;
5916
5917         lo = find_lockowner_str(cl, &lock->lk_new_owner);
5918         if (!lo) {
5919                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5920                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5921                 if (lo == NULL)
5922                         return nfserr_jukebox;
5923         } else {
5924                 /* with an existing lockowner, seqids must be the same */
5925                 status = nfserr_bad_seqid;
5926                 if (!cstate->minorversion &&
5927                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5928                         goto out;
5929         }
5930
5931         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5932         if (lst == NULL) {
5933                 status = nfserr_jukebox;
5934                 goto out;
5935         }
5936
5937         status = nfs_ok;
5938         *plst = lst;
5939 out:
5940         nfs4_put_stateowner(&lo->lo_owner);
5941         return status;
5942 }
5943
5944 /*
5945  *  LOCK operation
5946  */
5947 __be32
5948 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5949            union nfsd4_op_u *u)
5950 {
5951         struct nfsd4_lock *lock = &u->lock;
5952         struct nfs4_openowner *open_sop = NULL;
5953         struct nfs4_lockowner *lock_sop = NULL;
5954         struct nfs4_ol_stateid *lock_stp = NULL;
5955         struct nfs4_ol_stateid *open_stp = NULL;
5956         struct nfs4_file *fp;
5957         struct file *filp = NULL;
5958         struct nfsd4_blocked_lock *nbl = NULL;
5959         struct file_lock *file_lock = NULL;
5960         struct file_lock *conflock = NULL;
5961         __be32 status = 0;
5962         int lkflg;
5963         int err;
5964         bool new = false;
5965         unsigned char fl_type;
5966         unsigned int fl_flags = FL_POSIX;
5967         struct net *net = SVC_NET(rqstp);
5968         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5969
5970         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5971                 (long long) lock->lk_offset,
5972                 (long long) lock->lk_length);
5973
5974         if (check_lock_length(lock->lk_offset, lock->lk_length))
5975                  return nfserr_inval;
5976
5977         if ((status = fh_verify(rqstp, &cstate->current_fh,
5978                                 S_IFREG, NFSD_MAY_LOCK))) {
5979                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5980                 return status;
5981         }
5982
5983         if (lock->lk_is_new) {
5984                 if (nfsd4_has_session(cstate))
5985                         /* See rfc 5661 18.10.3: given clientid is ignored: */
5986                         memcpy(&lock->lk_new_clientid,
5987                                 &cstate->session->se_client->cl_clientid,
5988                                 sizeof(clientid_t));
5989
5990                 status = nfserr_stale_clientid;
5991                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5992                         goto out;
5993
5994                 /* validate and update open stateid and open seqid */
5995                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5996                                         lock->lk_new_open_seqid,
5997                                         &lock->lk_new_open_stateid,
5998                                         &open_stp, nn);
5999                 if (status)
6000                         goto out;
6001                 mutex_unlock(&open_stp->st_mutex);
6002                 open_sop = openowner(open_stp->st_stateowner);
6003                 status = nfserr_bad_stateid;
6004                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
6005                                                 &lock->lk_new_clientid))
6006                         goto out;
6007                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
6008                                                         &lock_stp, &new);
6009         } else {
6010                 status = nfs4_preprocess_seqid_op(cstate,
6011                                        lock->lk_old_lock_seqid,
6012                                        &lock->lk_old_lock_stateid,
6013                                        NFS4_LOCK_STID, &lock_stp, nn);
6014         }
6015         if (status)
6016                 goto out;
6017         lock_sop = lockowner(lock_stp->st_stateowner);
6018
6019         lkflg = setlkflg(lock->lk_type);
6020         status = nfs4_check_openmode(lock_stp, lkflg);
6021         if (status)
6022                 goto out;
6023
6024         status = nfserr_grace;
6025         if (locks_in_grace(net) && !lock->lk_reclaim)
6026                 goto out;
6027         status = nfserr_no_grace;
6028         if (!locks_in_grace(net) && lock->lk_reclaim)
6029                 goto out;
6030
6031         fp = lock_stp->st_stid.sc_file;
6032         switch (lock->lk_type) {
6033                 case NFS4_READW_LT:
6034                         if (nfsd4_has_session(cstate))
6035                                 fl_flags |= FL_SLEEP;
6036                         /* Fallthrough */
6037                 case NFS4_READ_LT:
6038                         spin_lock(&fp->fi_lock);
6039                         filp = find_readable_file_locked(fp);
6040                         if (filp)
6041                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
6042                         spin_unlock(&fp->fi_lock);
6043                         fl_type = F_RDLCK;
6044                         break;
6045                 case NFS4_WRITEW_LT:
6046                         if (nfsd4_has_session(cstate))
6047                                 fl_flags |= FL_SLEEP;
6048                         /* Fallthrough */
6049                 case NFS4_WRITE_LT:
6050                         spin_lock(&fp->fi_lock);
6051                         filp = find_writeable_file_locked(fp);
6052                         if (filp)
6053                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6054                         spin_unlock(&fp->fi_lock);
6055                         fl_type = F_WRLCK;
6056                         break;
6057                 default:
6058                         status = nfserr_inval;
6059                 goto out;
6060         }
6061
6062         if (!filp) {
6063                 status = nfserr_openmode;
6064                 goto out;
6065         }
6066
6067         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6068         if (!nbl) {
6069                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6070                 status = nfserr_jukebox;
6071                 goto out;
6072         }
6073
6074         file_lock = &nbl->nbl_lock;
6075         file_lock->fl_type = fl_type;
6076         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6077         file_lock->fl_pid = current->tgid;
6078         file_lock->fl_file = filp;
6079         file_lock->fl_flags = fl_flags;
6080         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6081         file_lock->fl_start = lock->lk_offset;
6082         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6083         nfs4_transform_lock_offset(file_lock);
6084
6085         conflock = locks_alloc_lock();
6086         if (!conflock) {
6087                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6088                 status = nfserr_jukebox;
6089                 goto out;
6090         }
6091
6092         if (fl_flags & FL_SLEEP) {
6093                 nbl->nbl_time = get_seconds();
6094                 spin_lock(&nn->blocked_locks_lock);
6095                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6096                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6097                 spin_unlock(&nn->blocked_locks_lock);
6098         }
6099
6100         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
6101         switch (err) {
6102         case 0: /* success! */
6103                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6104                 status = 0;
6105                 if (lock->lk_reclaim)
6106                         nn->somebody_reclaimed = true;
6107                 break;
6108         case FILE_LOCK_DEFERRED:
6109                 nbl = NULL;
6110                 /* Fallthrough */
6111         case -EAGAIN:           /* conflock holds conflicting lock */
6112                 status = nfserr_denied;
6113                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6114                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6115                 break;
6116         case -EDEADLK:
6117                 status = nfserr_deadlock;
6118                 break;
6119         default:
6120                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6121                 status = nfserrno(err);
6122                 break;
6123         }
6124 out:
6125         if (nbl) {
6126                 /* dequeue it if we queued it before */
6127                 if (fl_flags & FL_SLEEP) {
6128                         spin_lock(&nn->blocked_locks_lock);
6129                         list_del_init(&nbl->nbl_list);
6130                         list_del_init(&nbl->nbl_lru);
6131                         spin_unlock(&nn->blocked_locks_lock);
6132                 }
6133                 free_blocked_lock(nbl);
6134         }
6135         if (filp)
6136                 fput(filp);
6137         if (lock_stp) {
6138                 /* Bump seqid manually if the 4.0 replay owner is openowner */
6139                 if (cstate->replay_owner &&
6140                     cstate->replay_owner != &lock_sop->lo_owner &&
6141                     seqid_mutating_err(ntohl(status)))
6142                         lock_sop->lo_owner.so_seqid++;
6143
6144                 /*
6145                  * If this is a new, never-before-used stateid, and we are
6146                  * returning an error, then just go ahead and release it.
6147                  */
6148                 if (status && new)
6149                         release_lock_stateid(lock_stp);
6150
6151                 mutex_unlock(&lock_stp->st_mutex);
6152
6153                 nfs4_put_stid(&lock_stp->st_stid);
6154         }
6155         if (open_stp)
6156                 nfs4_put_stid(&open_stp->st_stid);
6157         nfsd4_bump_seqid(cstate, status);
6158         if (conflock)
6159                 locks_free_lock(conflock);
6160         return status;
6161 }
6162
6163 /*
6164  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6165  * so we do a temporary open here just to get an open file to pass to
6166  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
6167  * inode operation.)
6168  */
6169 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6170 {
6171         struct file *file;
6172         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
6173         if (!err) {
6174                 err = nfserrno(vfs_test_lock(file, lock));
6175                 fput(file);
6176         }
6177         return err;
6178 }
6179
6180 /*
6181  * LOCKT operation
6182  */
6183 __be32
6184 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6185             union nfsd4_op_u *u)
6186 {
6187         struct nfsd4_lockt *lockt = &u->lockt;
6188         struct file_lock *file_lock = NULL;
6189         struct nfs4_lockowner *lo = NULL;
6190         __be32 status;
6191         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6192
6193         if (locks_in_grace(SVC_NET(rqstp)))
6194                 return nfserr_grace;
6195
6196         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6197                  return nfserr_inval;
6198
6199         if (!nfsd4_has_session(cstate)) {
6200                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
6201                 if (status)
6202                         goto out;
6203         }
6204
6205         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6206                 goto out;
6207
6208         file_lock = locks_alloc_lock();
6209         if (!file_lock) {
6210                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6211                 status = nfserr_jukebox;
6212                 goto out;
6213         }
6214
6215         switch (lockt->lt_type) {
6216                 case NFS4_READ_LT:
6217                 case NFS4_READW_LT:
6218                         file_lock->fl_type = F_RDLCK;
6219                 break;
6220                 case NFS4_WRITE_LT:
6221                 case NFS4_WRITEW_LT:
6222                         file_lock->fl_type = F_WRLCK;
6223                 break;
6224                 default:
6225                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6226                         status = nfserr_inval;
6227                 goto out;
6228         }
6229
6230         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6231         if (lo)
6232                 file_lock->fl_owner = (fl_owner_t)lo;
6233         file_lock->fl_pid = current->tgid;
6234         file_lock->fl_flags = FL_POSIX;
6235
6236         file_lock->fl_start = lockt->lt_offset;
6237         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6238
6239         nfs4_transform_lock_offset(file_lock);
6240
6241         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6242         if (status)
6243                 goto out;
6244
6245         if (file_lock->fl_type != F_UNLCK) {
6246                 status = nfserr_denied;
6247                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6248         }
6249 out:
6250         if (lo)
6251                 nfs4_put_stateowner(&lo->lo_owner);
6252         if (file_lock)
6253                 locks_free_lock(file_lock);
6254         return status;
6255 }
6256
6257 __be32
6258 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6259             union nfsd4_op_u *u)
6260 {
6261         struct nfsd4_locku *locku = &u->locku;
6262         struct nfs4_ol_stateid *stp;
6263         struct file *filp = NULL;
6264         struct file_lock *file_lock = NULL;
6265         __be32 status;
6266         int err;
6267         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6268
6269         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6270                 (long long) locku->lu_offset,
6271                 (long long) locku->lu_length);
6272
6273         if (check_lock_length(locku->lu_offset, locku->lu_length))
6274                  return nfserr_inval;
6275
6276         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6277                                         &locku->lu_stateid, NFS4_LOCK_STID,
6278                                         &stp, nn);
6279         if (status)
6280                 goto out;
6281         filp = find_any_file(stp->st_stid.sc_file);
6282         if (!filp) {
6283                 status = nfserr_lock_range;
6284                 goto put_stateid;
6285         }
6286         file_lock = locks_alloc_lock();
6287         if (!file_lock) {
6288                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6289                 status = nfserr_jukebox;
6290                 goto fput;
6291         }
6292
6293         file_lock->fl_type = F_UNLCK;
6294         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6295         file_lock->fl_pid = current->tgid;
6296         file_lock->fl_file = filp;
6297         file_lock->fl_flags = FL_POSIX;
6298         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6299         file_lock->fl_start = locku->lu_offset;
6300
6301         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6302                                                 locku->lu_length);
6303         nfs4_transform_lock_offset(file_lock);
6304
6305         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6306         if (err) {
6307                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6308                 goto out_nfserr;
6309         }
6310         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6311 fput:
6312         fput(filp);
6313 put_stateid:
6314         mutex_unlock(&stp->st_mutex);
6315         nfs4_put_stid(&stp->st_stid);
6316 out:
6317         nfsd4_bump_seqid(cstate, status);
6318         if (file_lock)
6319                 locks_free_lock(file_lock);
6320         return status;
6321
6322 out_nfserr:
6323         status = nfserrno(err);
6324         goto fput;
6325 }
6326
6327 /*
6328  * returns
6329  *      true:  locks held by lockowner
6330  *      false: no locks held by lockowner
6331  */
6332 static bool
6333 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6334 {
6335         struct file_lock *fl;
6336         int status = false;
6337         struct file *filp = find_any_file(fp);
6338         struct inode *inode;
6339         struct file_lock_context *flctx;
6340
6341         if (!filp) {
6342                 /* Any valid lock stateid should have some sort of access */
6343                 WARN_ON_ONCE(1);
6344                 return status;
6345         }
6346
6347         inode = locks_inode(filp);
6348         flctx = inode->i_flctx;
6349
6350         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6351                 spin_lock(&flctx->flc_lock);
6352                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6353                         if (fl->fl_owner == (fl_owner_t)lowner) {
6354                                 status = true;
6355                                 break;
6356                         }
6357                 }
6358                 spin_unlock(&flctx->flc_lock);
6359         }
6360         fput(filp);
6361         return status;
6362 }
6363
6364 __be32
6365 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6366                         struct nfsd4_compound_state *cstate,
6367                         union nfsd4_op_u *u)
6368 {
6369         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6370         clientid_t *clid = &rlockowner->rl_clientid;
6371         struct nfs4_stateowner *sop;
6372         struct nfs4_lockowner *lo = NULL;
6373         struct nfs4_ol_stateid *stp;
6374         struct xdr_netobj *owner = &rlockowner->rl_owner;
6375         unsigned int hashval = ownerstr_hashval(owner);
6376         __be32 status;
6377         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6378         struct nfs4_client *clp;
6379         LIST_HEAD (reaplist);
6380
6381         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6382                 clid->cl_boot, clid->cl_id);
6383
6384         status = lookup_clientid(clid, cstate, nn);
6385         if (status)
6386                 return status;
6387
6388         clp = cstate->clp;
6389         /* Find the matching lock stateowner */
6390         spin_lock(&clp->cl_lock);
6391         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6392                             so_strhash) {
6393
6394                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6395                         continue;
6396
6397                 /* see if there are still any locks associated with it */
6398                 lo = lockowner(sop);
6399                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6400                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
6401                                 status = nfserr_locks_held;
6402                                 spin_unlock(&clp->cl_lock);
6403                                 return status;
6404                         }
6405                 }
6406
6407                 nfs4_get_stateowner(sop);
6408                 break;
6409         }
6410         if (!lo) {
6411                 spin_unlock(&clp->cl_lock);
6412                 return status;
6413         }
6414
6415         unhash_lockowner_locked(lo);
6416         while (!list_empty(&lo->lo_owner.so_stateids)) {
6417                 stp = list_first_entry(&lo->lo_owner.so_stateids,
6418                                        struct nfs4_ol_stateid,
6419                                        st_perstateowner);
6420                 WARN_ON(!unhash_lock_stateid(stp));
6421                 put_ol_stateid_locked(stp, &reaplist);
6422         }
6423         spin_unlock(&clp->cl_lock);
6424         free_ol_stateid_reaplist(&reaplist);
6425         remove_blocked_locks(lo);
6426         nfs4_put_stateowner(&lo->lo_owner);
6427
6428         return status;
6429 }
6430
6431 static inline struct nfs4_client_reclaim *
6432 alloc_reclaim(void)
6433 {
6434         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6435 }
6436
6437 bool
6438 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
6439 {
6440         struct nfs4_client_reclaim *crp;
6441
6442         crp = nfsd4_find_reclaim_client(name, nn);
6443         return (crp && crp->cr_clp);
6444 }
6445
6446 /*
6447  * failure => all reset bets are off, nfserr_no_grace...
6448  */
6449 struct nfs4_client_reclaim *
6450 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6451 {
6452         unsigned int strhashval;
6453         struct nfs4_client_reclaim *crp;
6454
6455         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6456         crp = alloc_reclaim();
6457         if (crp) {
6458                 strhashval = clientstr_hashval(name);
6459                 INIT_LIST_HEAD(&crp->cr_strhash);
6460                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6461                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6462                 crp->cr_clp = NULL;
6463                 nn->reclaim_str_hashtbl_size++;
6464         }
6465         return crp;
6466 }
6467
6468 void
6469 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6470 {
6471         list_del(&crp->cr_strhash);
6472         kfree(crp);
6473         nn->reclaim_str_hashtbl_size--;
6474 }
6475
6476 void
6477 nfs4_release_reclaim(struct nfsd_net *nn)
6478 {
6479         struct nfs4_client_reclaim *crp = NULL;
6480         int i;
6481
6482         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6483                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6484                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6485                                         struct nfs4_client_reclaim, cr_strhash);
6486                         nfs4_remove_reclaim_record(crp, nn);
6487                 }
6488         }
6489         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6490 }
6491
6492 /*
6493  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6494 struct nfs4_client_reclaim *
6495 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6496 {
6497         unsigned int strhashval;
6498         struct nfs4_client_reclaim *crp = NULL;
6499
6500         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6501
6502         strhashval = clientstr_hashval(recdir);
6503         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6504                 if (same_name(crp->cr_recdir, recdir)) {
6505                         return crp;
6506                 }
6507         }
6508         return NULL;
6509 }
6510
6511 /*
6512 * Called from OPEN. Look for clientid in reclaim list.
6513 */
6514 __be32
6515 nfs4_check_open_reclaim(clientid_t *clid,
6516                 struct nfsd4_compound_state *cstate,
6517                 struct nfsd_net *nn)
6518 {
6519         __be32 status;
6520
6521         /* find clientid in conf_id_hashtbl */
6522         status = lookup_clientid(clid, cstate, nn);
6523         if (status)
6524                 return nfserr_reclaim_bad;
6525
6526         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6527                 return nfserr_no_grace;
6528
6529         if (nfsd4_client_record_check(cstate->clp))
6530                 return nfserr_reclaim_bad;
6531
6532         return nfs_ok;
6533 }
6534
6535 #ifdef CONFIG_NFSD_FAULT_INJECTION
6536 static inline void
6537 put_client(struct nfs4_client *clp)
6538 {
6539         atomic_dec(&clp->cl_refcount);
6540 }
6541
6542 static struct nfs4_client *
6543 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6544 {
6545         struct nfs4_client *clp;
6546         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6547                                           nfsd_net_id);
6548
6549         if (!nfsd_netns_ready(nn))
6550                 return NULL;
6551
6552         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6553                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6554                         return clp;
6555         }
6556         return NULL;
6557 }
6558
6559 u64
6560 nfsd_inject_print_clients(void)
6561 {
6562         struct nfs4_client *clp;
6563         u64 count = 0;
6564         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6565                                           nfsd_net_id);
6566         char buf[INET6_ADDRSTRLEN];
6567
6568         if (!nfsd_netns_ready(nn))
6569                 return 0;
6570
6571         spin_lock(&nn->client_lock);
6572         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6573                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6574                 pr_info("NFS Client: %s\n", buf);
6575                 ++count;
6576         }
6577         spin_unlock(&nn->client_lock);
6578
6579         return count;
6580 }
6581
6582 u64
6583 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6584 {
6585         u64 count = 0;
6586         struct nfs4_client *clp;
6587         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6588                                           nfsd_net_id);
6589
6590         if (!nfsd_netns_ready(nn))
6591                 return count;
6592
6593         spin_lock(&nn->client_lock);
6594         clp = nfsd_find_client(addr, addr_size);
6595         if (clp) {
6596                 if (mark_client_expired_locked(clp) == nfs_ok)
6597                         ++count;
6598                 else
6599                         clp = NULL;
6600         }
6601         spin_unlock(&nn->client_lock);
6602
6603         if (clp)
6604                 expire_client(clp);
6605
6606         return count;
6607 }
6608
6609 u64
6610 nfsd_inject_forget_clients(u64 max)
6611 {
6612         u64 count = 0;
6613         struct nfs4_client *clp, *next;
6614         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6615                                                 nfsd_net_id);
6616         LIST_HEAD(reaplist);
6617
6618         if (!nfsd_netns_ready(nn))
6619                 return count;
6620
6621         spin_lock(&nn->client_lock);
6622         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6623                 if (mark_client_expired_locked(clp) == nfs_ok) {
6624                         list_add(&clp->cl_lru, &reaplist);
6625                         if (max != 0 && ++count >= max)
6626                                 break;
6627                 }
6628         }
6629         spin_unlock(&nn->client_lock);
6630
6631         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6632                 expire_client(clp);
6633
6634         return count;
6635 }
6636
6637 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6638                              const char *type)
6639 {
6640         char buf[INET6_ADDRSTRLEN];
6641         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6642         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6643 }
6644
6645 static void
6646 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6647                              struct list_head *collect)
6648 {
6649         struct nfs4_client *clp = lst->st_stid.sc_client;
6650         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6651                                           nfsd_net_id);
6652
6653         if (!collect)
6654                 return;
6655
6656         lockdep_assert_held(&nn->client_lock);
6657         atomic_inc(&clp->cl_refcount);
6658         list_add(&lst->st_locks, collect);
6659 }
6660
6661 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6662                                     struct list_head *collect,
6663                                     bool (*func)(struct nfs4_ol_stateid *))
6664 {
6665         struct nfs4_openowner *oop;
6666         struct nfs4_ol_stateid *stp, *st_next;
6667         struct nfs4_ol_stateid *lst, *lst_next;
6668         u64 count = 0;
6669
6670         spin_lock(&clp->cl_lock);
6671         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6672                 list_for_each_entry_safe(stp, st_next,
6673                                 &oop->oo_owner.so_stateids, st_perstateowner) {
6674                         list_for_each_entry_safe(lst, lst_next,
6675                                         &stp->st_locks, st_locks) {
6676                                 if (func) {
6677                                         if (func(lst))
6678                                                 nfsd_inject_add_lock_to_list(lst,
6679                                                                         collect);
6680                                 }
6681                                 ++count;
6682                                 /*
6683                                  * Despite the fact that these functions deal
6684                                  * with 64-bit integers for "count", we must
6685                                  * ensure that it doesn't blow up the
6686                                  * clp->cl_refcount. Throw a warning if we
6687                                  * start to approach INT_MAX here.
6688                                  */
6689                                 WARN_ON_ONCE(count == (INT_MAX / 2));
6690                                 if (count == max)
6691                                         goto out;
6692                         }
6693                 }
6694         }
6695 out:
6696         spin_unlock(&clp->cl_lock);
6697
6698         return count;
6699 }
6700
6701 static u64
6702 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6703                           u64 max)
6704 {
6705         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6706 }
6707
6708 static u64
6709 nfsd_print_client_locks(struct nfs4_client *clp)
6710 {
6711         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6712         nfsd_print_count(clp, count, "locked files");
6713         return count;
6714 }
6715
6716 u64
6717 nfsd_inject_print_locks(void)
6718 {
6719         struct nfs4_client *clp;
6720         u64 count = 0;
6721         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6722                                                 nfsd_net_id);
6723
6724         if (!nfsd_netns_ready(nn))
6725                 return 0;
6726
6727         spin_lock(&nn->client_lock);
6728         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6729                 count += nfsd_print_client_locks(clp);
6730         spin_unlock(&nn->client_lock);
6731
6732         return count;
6733 }
6734
6735 static void
6736 nfsd_reap_locks(struct list_head *reaplist)
6737 {
6738         struct nfs4_client *clp;
6739         struct nfs4_ol_stateid *stp, *next;
6740
6741         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6742                 list_del_init(&stp->st_locks);
6743                 clp = stp->st_stid.sc_client;
6744                 nfs4_put_stid(&stp->st_stid);
6745                 put_client(clp);
6746         }
6747 }
6748
6749 u64
6750 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6751 {
6752         unsigned int count = 0;
6753         struct nfs4_client *clp;
6754         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6755                                                 nfsd_net_id);
6756         LIST_HEAD(reaplist);
6757
6758         if (!nfsd_netns_ready(nn))
6759                 return count;
6760
6761         spin_lock(&nn->client_lock);
6762         clp = nfsd_find_client(addr, addr_size);
6763         if (clp)
6764                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6765         spin_unlock(&nn->client_lock);
6766         nfsd_reap_locks(&reaplist);
6767         return count;
6768 }
6769
6770 u64
6771 nfsd_inject_forget_locks(u64 max)
6772 {
6773         u64 count = 0;
6774         struct nfs4_client *clp;
6775         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6776                                                 nfsd_net_id);
6777         LIST_HEAD(reaplist);
6778
6779         if (!nfsd_netns_ready(nn))
6780                 return count;
6781
6782         spin_lock(&nn->client_lock);
6783         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6784                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6785                 if (max != 0 && count >= max)
6786                         break;
6787         }
6788         spin_unlock(&nn->client_lock);
6789         nfsd_reap_locks(&reaplist);
6790         return count;
6791 }
6792
6793 static u64
6794 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6795                               struct list_head *collect,
6796                               void (*func)(struct nfs4_openowner *))
6797 {
6798         struct nfs4_openowner *oop, *next;
6799         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6800                                                 nfsd_net_id);
6801         u64 count = 0;
6802
6803         lockdep_assert_held(&nn->client_lock);
6804
6805         spin_lock(&clp->cl_lock);
6806         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6807                 if (func) {
6808                         func(oop);
6809                         if (collect) {
6810                                 atomic_inc(&clp->cl_refcount);
6811                                 list_add(&oop->oo_perclient, collect);
6812                         }
6813                 }
6814                 ++count;
6815                 /*
6816                  * Despite the fact that these functions deal with
6817                  * 64-bit integers for "count", we must ensure that
6818                  * it doesn't blow up the clp->cl_refcount. Throw a
6819                  * warning if we start to approach INT_MAX here.
6820                  */
6821                 WARN_ON_ONCE(count == (INT_MAX / 2));
6822                 if (count == max)
6823                         break;
6824         }
6825         spin_unlock(&clp->cl_lock);
6826
6827         return count;
6828 }
6829
6830 static u64
6831 nfsd_print_client_openowners(struct nfs4_client *clp)
6832 {
6833         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6834
6835         nfsd_print_count(clp, count, "openowners");
6836         return count;
6837 }
6838
6839 static u64
6840 nfsd_collect_client_openowners(struct nfs4_client *clp,
6841                                struct list_head *collect, u64 max)
6842 {
6843         return nfsd_foreach_client_openowner(clp, max, collect,
6844                                                 unhash_openowner_locked);
6845 }
6846
6847 u64
6848 nfsd_inject_print_openowners(void)
6849 {
6850         struct nfs4_client *clp;
6851         u64 count = 0;
6852         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6853                                                 nfsd_net_id);
6854
6855         if (!nfsd_netns_ready(nn))
6856                 return 0;
6857
6858         spin_lock(&nn->client_lock);
6859         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6860                 count += nfsd_print_client_openowners(clp);
6861         spin_unlock(&nn->client_lock);
6862
6863         return count;
6864 }
6865
6866 static void
6867 nfsd_reap_openowners(struct list_head *reaplist)
6868 {
6869         struct nfs4_client *clp;
6870         struct nfs4_openowner *oop, *next;
6871
6872         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6873                 list_del_init(&oop->oo_perclient);
6874                 clp = oop->oo_owner.so_client;
6875                 release_openowner(oop);
6876                 put_client(clp);
6877         }
6878 }
6879
6880 u64
6881 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6882                                      size_t addr_size)
6883 {
6884         unsigned int count = 0;
6885         struct nfs4_client *clp;
6886         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6887                                                 nfsd_net_id);
6888         LIST_HEAD(reaplist);
6889
6890         if (!nfsd_netns_ready(nn))
6891                 return count;
6892
6893         spin_lock(&nn->client_lock);
6894         clp = nfsd_find_client(addr, addr_size);
6895         if (clp)
6896                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6897         spin_unlock(&nn->client_lock);
6898         nfsd_reap_openowners(&reaplist);
6899         return count;
6900 }
6901
6902 u64
6903 nfsd_inject_forget_openowners(u64 max)
6904 {
6905         u64 count = 0;
6906         struct nfs4_client *clp;
6907         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6908                                                 nfsd_net_id);
6909         LIST_HEAD(reaplist);
6910
6911         if (!nfsd_netns_ready(nn))
6912                 return count;
6913
6914         spin_lock(&nn->client_lock);
6915         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6916                 count += nfsd_collect_client_openowners(clp, &reaplist,
6917                                                         max - count);
6918                 if (max != 0 && count >= max)
6919                         break;
6920         }
6921         spin_unlock(&nn->client_lock);
6922         nfsd_reap_openowners(&reaplist);
6923         return count;
6924 }
6925
6926 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6927                                      struct list_head *victims)
6928 {
6929         struct nfs4_delegation *dp, *next;
6930         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6931                                                 nfsd_net_id);
6932         u64 count = 0;
6933
6934         lockdep_assert_held(&nn->client_lock);
6935
6936         spin_lock(&state_lock);
6937         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6938                 if (victims) {
6939                         /*
6940                          * It's not safe to mess with delegations that have a
6941                          * non-zero dl_time. They might have already been broken
6942                          * and could be processed by the laundromat outside of
6943                          * the state_lock. Just leave them be.
6944                          */
6945                         if (dp->dl_time != 0)
6946                                 continue;
6947
6948                         atomic_inc(&clp->cl_refcount);
6949                         WARN_ON(!unhash_delegation_locked(dp));
6950                         list_add(&dp->dl_recall_lru, victims);
6951                 }
6952                 ++count;
6953                 /*
6954                  * Despite the fact that these functions deal with
6955                  * 64-bit integers for "count", we must ensure that
6956                  * it doesn't blow up the clp->cl_refcount. Throw a
6957                  * warning if we start to approach INT_MAX here.
6958                  */
6959                 WARN_ON_ONCE(count == (INT_MAX / 2));
6960                 if (count == max)
6961                         break;
6962         }
6963         spin_unlock(&state_lock);
6964         return count;
6965 }
6966
6967 static u64
6968 nfsd_print_client_delegations(struct nfs4_client *clp)
6969 {
6970         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6971
6972         nfsd_print_count(clp, count, "delegations");
6973         return count;
6974 }
6975
6976 u64
6977 nfsd_inject_print_delegations(void)
6978 {
6979         struct nfs4_client *clp;
6980         u64 count = 0;
6981         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6982                                                 nfsd_net_id);
6983
6984         if (!nfsd_netns_ready(nn))
6985                 return 0;
6986
6987         spin_lock(&nn->client_lock);
6988         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6989                 count += nfsd_print_client_delegations(clp);
6990         spin_unlock(&nn->client_lock);
6991
6992         return count;
6993 }
6994
6995 static void
6996 nfsd_forget_delegations(struct list_head *reaplist)
6997 {
6998         struct nfs4_client *clp;
6999         struct nfs4_delegation *dp, *next;
7000
7001         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7002                 list_del_init(&dp->dl_recall_lru);
7003                 clp = dp->dl_stid.sc_client;
7004                 revoke_delegation(dp);
7005                 put_client(clp);
7006         }
7007 }
7008
7009 u64
7010 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
7011                                       size_t addr_size)
7012 {
7013         u64 count = 0;
7014         struct nfs4_client *clp;
7015         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7016                                                 nfsd_net_id);
7017         LIST_HEAD(reaplist);
7018
7019         if (!nfsd_netns_ready(nn))
7020                 return count;
7021
7022         spin_lock(&nn->client_lock);
7023         clp = nfsd_find_client(addr, addr_size);
7024         if (clp)
7025                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7026         spin_unlock(&nn->client_lock);
7027
7028         nfsd_forget_delegations(&reaplist);
7029         return count;
7030 }
7031
7032 u64
7033 nfsd_inject_forget_delegations(u64 max)
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         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7046                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7047                 if (max != 0 && count >= max)
7048                         break;
7049         }
7050         spin_unlock(&nn->client_lock);
7051         nfsd_forget_delegations(&reaplist);
7052         return count;
7053 }
7054
7055 static void
7056 nfsd_recall_delegations(struct list_head *reaplist)
7057 {
7058         struct nfs4_client *clp;
7059         struct nfs4_delegation *dp, *next;
7060
7061         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7062                 list_del_init(&dp->dl_recall_lru);
7063                 clp = dp->dl_stid.sc_client;
7064                 /*
7065                  * We skipped all entries that had a zero dl_time before,
7066                  * so we can now reset the dl_time back to 0. If a delegation
7067                  * break comes in now, then it won't make any difference since
7068                  * we're recalling it either way.
7069                  */
7070                 spin_lock(&state_lock);
7071                 dp->dl_time = 0;
7072                 spin_unlock(&state_lock);
7073                 nfsd_break_one_deleg(dp);
7074                 put_client(clp);
7075         }
7076 }
7077
7078 u64
7079 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7080                                       size_t addr_size)
7081 {
7082         u64 count = 0;
7083         struct nfs4_client *clp;
7084         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7085                                                 nfsd_net_id);
7086         LIST_HEAD(reaplist);
7087
7088         if (!nfsd_netns_ready(nn))
7089                 return count;
7090
7091         spin_lock(&nn->client_lock);
7092         clp = nfsd_find_client(addr, addr_size);
7093         if (clp)
7094                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7095         spin_unlock(&nn->client_lock);
7096
7097         nfsd_recall_delegations(&reaplist);
7098         return count;
7099 }
7100
7101 u64
7102 nfsd_inject_recall_delegations(u64 max)
7103 {
7104         u64 count = 0;
7105         struct nfs4_client *clp, *next;
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         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7115                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7116                 if (max != 0 && ++count >= max)
7117                         break;
7118         }
7119         spin_unlock(&nn->client_lock);
7120         nfsd_recall_delegations(&reaplist);
7121         return count;
7122 }
7123 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7124
7125 /*
7126  * Since the lifetime of a delegation isn't limited to that of an open, a
7127  * client may quite reasonably hang on to a delegation as long as it has
7128  * the inode cached.  This becomes an obvious problem the first time a
7129  * client's inode cache approaches the size of the server's total memory.
7130  *
7131  * For now we avoid this problem by imposing a hard limit on the number
7132  * of delegations, which varies according to the server's memory size.
7133  */
7134 static void
7135 set_max_delegations(void)
7136 {
7137         /*
7138          * Allow at most 4 delegations per megabyte of RAM.  Quick
7139          * estimates suggest that in the worst case (where every delegation
7140          * is for a different inode), a delegation could take about 1.5K,
7141          * giving a worst case usage of about 6% of memory.
7142          */
7143         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7144 }
7145
7146 static int nfs4_state_create_net(struct net *net)
7147 {
7148         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7149         int i;
7150
7151         nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7152                                             sizeof(struct list_head),
7153                                             GFP_KERNEL);
7154         if (!nn->conf_id_hashtbl)
7155                 goto err;
7156         nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7157                                               sizeof(struct list_head),
7158                                               GFP_KERNEL);
7159         if (!nn->unconf_id_hashtbl)
7160                 goto err_unconf_id;
7161         nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7162                                               sizeof(struct list_head),
7163                                               GFP_KERNEL);
7164         if (!nn->sessionid_hashtbl)
7165                 goto err_sessionid;
7166
7167         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7168                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7169                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7170         }
7171         for (i = 0; i < SESSION_HASH_SIZE; i++)
7172                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7173         nn->conf_name_tree = RB_ROOT;
7174         nn->unconf_name_tree = RB_ROOT;
7175         nn->boot_time = get_seconds();
7176         nn->grace_ended = false;
7177         nn->nfsd4_manager.block_opens = true;
7178         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7179         INIT_LIST_HEAD(&nn->client_lru);
7180         INIT_LIST_HEAD(&nn->close_lru);
7181         INIT_LIST_HEAD(&nn->del_recall_lru);
7182         spin_lock_init(&nn->client_lock);
7183
7184         spin_lock_init(&nn->blocked_locks_lock);
7185         INIT_LIST_HEAD(&nn->blocked_locks_lru);
7186
7187         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7188         get_net(net);
7189
7190         return 0;
7191
7192 err_sessionid:
7193         kfree(nn->unconf_id_hashtbl);
7194 err_unconf_id:
7195         kfree(nn->conf_id_hashtbl);
7196 err:
7197         return -ENOMEM;
7198 }
7199
7200 static void
7201 nfs4_state_destroy_net(struct net *net)
7202 {
7203         int i;
7204         struct nfs4_client *clp = NULL;
7205         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7206
7207         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7208                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7209                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7210                         destroy_client(clp);
7211                 }
7212         }
7213
7214         WARN_ON(!list_empty(&nn->blocked_locks_lru));
7215
7216         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7217                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7218                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7219                         destroy_client(clp);
7220                 }
7221         }
7222
7223         kfree(nn->sessionid_hashtbl);
7224         kfree(nn->unconf_id_hashtbl);
7225         kfree(nn->conf_id_hashtbl);
7226         put_net(net);
7227 }
7228
7229 int
7230 nfs4_state_start_net(struct net *net)
7231 {
7232         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7233         int ret;
7234
7235         ret = nfs4_state_create_net(net);
7236         if (ret)
7237                 return ret;
7238         locks_start_grace(net, &nn->nfsd4_manager);
7239         nfsd4_client_tracking_init(net);
7240         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %x)\n",
7241                nn->nfsd4_grace, net->ns.inum);
7242         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7243         return 0;
7244 }
7245
7246 /* initialization to perform when the nfsd service is started: */
7247
7248 int
7249 nfs4_state_start(void)
7250 {
7251         int ret;
7252
7253         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7254         if (laundry_wq == NULL) {
7255                 ret = -ENOMEM;
7256                 goto out;
7257         }
7258         ret = nfsd4_create_callback_queue();
7259         if (ret)
7260                 goto out_free_laundry;
7261
7262         set_max_delegations();
7263         return 0;
7264
7265 out_free_laundry:
7266         destroy_workqueue(laundry_wq);
7267 out:
7268         return ret;
7269 }
7270
7271 void
7272 nfs4_state_shutdown_net(struct net *net)
7273 {
7274         struct nfs4_delegation *dp = NULL;
7275         struct list_head *pos, *next, reaplist;
7276         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7277
7278         cancel_delayed_work_sync(&nn->laundromat_work);
7279         locks_end_grace(&nn->nfsd4_manager);
7280
7281         INIT_LIST_HEAD(&reaplist);
7282         spin_lock(&state_lock);
7283         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7284                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7285                 WARN_ON(!unhash_delegation_locked(dp));
7286                 list_add(&dp->dl_recall_lru, &reaplist);
7287         }
7288         spin_unlock(&state_lock);
7289         list_for_each_safe(pos, next, &reaplist) {
7290                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7291                 list_del_init(&dp->dl_recall_lru);
7292                 destroy_unhashed_deleg(dp);
7293         }
7294
7295         nfsd4_client_tracking_exit(net);
7296         nfs4_state_destroy_net(net);
7297 }
7298
7299 void
7300 nfs4_state_shutdown(void)
7301 {
7302         destroy_workqueue(laundry_wq);
7303         nfsd4_destroy_callback_queue();
7304 }
7305
7306 static void
7307 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7308 {
7309         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7310                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7311 }
7312
7313 static void
7314 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7315 {
7316         if (cstate->minorversion) {
7317                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7318                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7319         }
7320 }
7321
7322 void
7323 clear_current_stateid(struct nfsd4_compound_state *cstate)
7324 {
7325         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7326 }
7327
7328 /*
7329  * functions to set current state id
7330  */
7331 void
7332 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7333                 union nfsd4_op_u *u)
7334 {
7335         put_stateid(cstate, &u->open_downgrade.od_stateid);
7336 }
7337
7338 void
7339 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7340                 union nfsd4_op_u *u)
7341 {
7342         put_stateid(cstate, &u->open.op_stateid);
7343 }
7344
7345 void
7346 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7347                 union nfsd4_op_u *u)
7348 {
7349         put_stateid(cstate, &u->close.cl_stateid);
7350 }
7351
7352 void
7353 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7354                 union nfsd4_op_u *u)
7355 {
7356         put_stateid(cstate, &u->lock.lk_resp_stateid);
7357 }
7358
7359 /*
7360  * functions to consume current state id
7361  */
7362
7363 void
7364 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7365                 union nfsd4_op_u *u)
7366 {
7367         get_stateid(cstate, &u->open_downgrade.od_stateid);
7368 }
7369
7370 void
7371 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7372                 union nfsd4_op_u *u)
7373 {
7374         get_stateid(cstate, &u->delegreturn.dr_stateid);
7375 }
7376
7377 void
7378 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7379                 union nfsd4_op_u *u)
7380 {
7381         get_stateid(cstate, &u->free_stateid.fr_stateid);
7382 }
7383
7384 void
7385 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7386                 union nfsd4_op_u *u)
7387 {
7388         get_stateid(cstate, &u->setattr.sa_stateid);
7389 }
7390
7391 void
7392 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7393                 union nfsd4_op_u *u)
7394 {
7395         get_stateid(cstate, &u->close.cl_stateid);
7396 }
7397
7398 void
7399 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7400                 union nfsd4_op_u *u)
7401 {
7402         get_stateid(cstate, &u->locku.lu_stateid);
7403 }
7404
7405 void
7406 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7407                 union nfsd4_op_u *u)
7408 {
7409         get_stateid(cstate, &u->read.rd_stateid);
7410 }
7411
7412 void
7413 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7414                 union nfsd4_op_u *u)
7415 {
7416         get_stateid(cstate, &u->write.wr_stateid);
7417 }