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 <linux/string_helpers.h>
  46 #include "xdr4.h"
  47 #include "xdr4cb.h"
  48 #include "vfs.h"
  49 #include "current_stateid.h"
  50
  51 #include "netns.h"
  52 #include "pnfs.h"
  53 #include "filecache.h"
  54 #include "trace.h"
  55
  56 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  57
  58 #define all_ones {{~0,~0},~0}
  59 static const stateid_t one_stateid = {
  60         .si_generation = ~0,
  61         .si_opaque = all_ones,
  62 };
  63 static const stateid_t zero_stateid = {
  64         /* all fields zero */
  65 };
  66 static const stateid_t currentstateid = {
  67         .si_generation = 1,
  68 };
  69 static const stateid_t close_stateid = {
  70         .si_generation = 0xffffffffU,
  71 };
  72
  73 static u64 current_sessionid = 1;
  74
  75 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  76 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  77 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  78 #define CLOSE_STATEID(stateid)  (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
  79
  80 /* forward declarations */
  81 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  82 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  83 void nfsd4_end_grace(struct nfsd_net *nn);
  84 static void _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps);
  85
  86 /* Locking: */
  87
  88 /*
  89  * Currently used for the del_recall_lru and file hash table.  In an
  90  * effort to decrease the scope of the client_mutex, this spinlock may
  91  * eventually cover more:
  92  */
  93 static DEFINE_SPINLOCK(state_lock);
  94
  95 enum nfsd4_st_mutex_lock_subclass {
  96         OPEN_STATEID_MUTEX = 0,
  97         LOCK_STATEID_MUTEX = 1,
  98 };
  99
 100 /*
 101  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
 102  * the refcount on the open stateid to drop.
 103  */
 104 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
 105
 106 /*
 107  * A waitqueue where a writer to clients/#/ctl destroying a client can
 108  * wait for cl_rpc_users to drop to 0 and then for the client to be
 109  * unhashed.
 110  */
 111 static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
 112
 113 static struct kmem_cache *client_slab;
 114 static struct kmem_cache *openowner_slab;
 115 static struct kmem_cache *lockowner_slab;
 116 static struct kmem_cache *file_slab;
 117 static struct kmem_cache *stateid_slab;
 118 static struct kmem_cache *deleg_slab;
 119 static struct kmem_cache *odstate_slab;
 120
 121 static void free_session(struct nfsd4_session *);
 122
 123 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 124 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
 125
 126 static bool is_session_dead(struct nfsd4_session *ses)
 127 {
 128         return ses->se_flags & NFS4_SESSION_DEAD;
 129 }
 130
 131 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 132 {
 133         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 134                 return nfserr_jukebox;
 135         ses->se_flags |= NFS4_SESSION_DEAD;
 136         return nfs_ok;
 137 }
 138
 139 static bool is_client_expired(struct nfs4_client *clp)
 140 {
 141         return clp->cl_time == 0;
 142 }
 143
 144 static __be32 get_client_locked(struct nfs4_client *clp)
 145 {
 146         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 147
 148         lockdep_assert_held(&nn->client_lock);
 149
 150         if (is_client_expired(clp))
 151                 return nfserr_expired;
 152         atomic_inc(&clp->cl_rpc_users);
 153         return nfs_ok;
 154 }
 155
 156 /* must be called under the client_lock */
 157 static inline void
 158 renew_client_locked(struct nfs4_client *clp)
 159 {
 160         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 161
 162         if (is_client_expired(clp)) {
 163                 WARN_ON(1);
 164                 printk("%s: client (clientid %08x/%08x) already expired\n",
 165                         __func__,
 166                         clp->cl_clientid.cl_boot,
 167                         clp->cl_clientid.cl_id);
 168                 return;
 169         }
 170
 171         list_move_tail(&clp->cl_lru, &nn->client_lru);
 172         clp->cl_time = ktime_get_boottime_seconds();
 173 }
 174
 175 static void put_client_renew_locked(struct nfs4_client *clp)
 176 {
 177         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 178
 179         lockdep_assert_held(&nn->client_lock);
 180
 181         if (!atomic_dec_and_test(&clp->cl_rpc_users))
 182                 return;
 183         if (!is_client_expired(clp))
 184                 renew_client_locked(clp);
 185         else
 186                 wake_up_all(&expiry_wq);
 187 }
 188
 189 static void put_client_renew(struct nfs4_client *clp)
 190 {
 191         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 192
 193         if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
 194                 return;
 195         if (!is_client_expired(clp))
 196                 renew_client_locked(clp);
 197         else
 198                 wake_up_all(&expiry_wq);
 199         spin_unlock(&nn->client_lock);
 200 }
 201
 202 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 203 {
 204         __be32 status;
 205
 206         if (is_session_dead(ses))
 207                 return nfserr_badsession;
 208         status = get_client_locked(ses->se_client);
 209         if (status)
 210                 return status;
 211         atomic_inc(&ses->se_ref);
 212         return nfs_ok;
 213 }
 214
 215 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 216 {
 217         struct nfs4_client *clp = ses->se_client;
 218         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 219
 220         lockdep_assert_held(&nn->client_lock);
 221
 222         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 223                 free_session(ses);
 224         put_client_renew_locked(clp);
 225 }
 226
 227 static void nfsd4_put_session(struct nfsd4_session *ses)
 228 {
 229         struct nfs4_client *clp = ses->se_client;
 230         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 231
 232         spin_lock(&nn->client_lock);
 233         nfsd4_put_session_locked(ses);
 234         spin_unlock(&nn->client_lock);
 235 }
 236
 237 static struct nfsd4_blocked_lock *
 238 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 239                         struct nfsd_net *nn)
 240 {
 241         struct nfsd4_blocked_lock *cur, *found = NULL;
 242
 243         spin_lock(&nn->blocked_locks_lock);
 244         list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
 245                 if (fh_match(fh, &cur->nbl_fh)) {
 246                         list_del_init(&cur->nbl_list);
 247                         list_del_init(&cur->nbl_lru);
 248                         found = cur;
 249                         break;
 250                 }
 251         }
 252         spin_unlock(&nn->blocked_locks_lock);
 253         if (found)
 254                 locks_delete_block(&found->nbl_lock);
 255         return found;
 256 }
 257
 258 static struct nfsd4_blocked_lock *
 259 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 260                         struct nfsd_net *nn)
 261 {
 262         struct nfsd4_blocked_lock *nbl;
 263
 264         nbl = find_blocked_lock(lo, fh, nn);
 265         if (!nbl) {
 266                 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
 267                 if (nbl) {
 268                         INIT_LIST_HEAD(&nbl->nbl_list);
 269                         INIT_LIST_HEAD(&nbl->nbl_lru);
 270                         fh_copy_shallow(&nbl->nbl_fh, fh);
 271                         locks_init_lock(&nbl->nbl_lock);
 272                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
 273                                         &nfsd4_cb_notify_lock_ops,
 274                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
 275                 }
 276         }
 277         return nbl;
 278 }
 279
 280 static void
 281 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
 282 {
 283         locks_delete_block(&nbl->nbl_lock);
 284         locks_release_private(&nbl->nbl_lock);
 285         kfree(nbl);
 286 }
 287
 288 static void
 289 remove_blocked_locks(struct nfs4_lockowner *lo)
 290 {
 291         struct nfs4_client *clp = lo->lo_owner.so_client;
 292         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 293         struct nfsd4_blocked_lock *nbl;
 294         LIST_HEAD(reaplist);
 295
 296         /* Dequeue all blocked locks */
 297         spin_lock(&nn->blocked_locks_lock);
 298         while (!list_empty(&lo->lo_blocked)) {
 299                 nbl = list_first_entry(&lo->lo_blocked,
 300                                         struct nfsd4_blocked_lock,
 301                                         nbl_list);
 302                 list_del_init(&nbl->nbl_list);
 303                 list_move(&nbl->nbl_lru, &reaplist);
 304         }
 305         spin_unlock(&nn->blocked_locks_lock);
 306
 307         /* Now free them */
 308         while (!list_empty(&reaplist)) {
 309                 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
 310                                         nbl_lru);
 311                 list_del_init(&nbl->nbl_lru);
 312                 free_blocked_lock(nbl);
 313         }
 314 }
 315
 316 static void
 317 nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
 318 {
 319         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 320                                                 struct nfsd4_blocked_lock, nbl_cb);
 321         locks_delete_block(&nbl->nbl_lock);
 322 }
 323
 324 static int
 325 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
 326 {
 327         /*
 328          * Since this is just an optimization, we don't try very hard if it
 329          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
 330          * just quit trying on anything else.
 331          */
 332         switch (task->tk_status) {
 333         case -NFS4ERR_DELAY:
 334                 rpc_delay(task, 1 * HZ);
 335                 return 0;
 336         default:
 337                 return 1;
 338         }
 339 }
 340
 341 static void
 342 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
 343 {
 344         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 345                                                 struct nfsd4_blocked_lock, nbl_cb);
 346
 347         free_blocked_lock(nbl);
 348 }
 349
 350 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
 351         .prepare        = nfsd4_cb_notify_lock_prepare,
 352         .done           = nfsd4_cb_notify_lock_done,
 353         .release        = nfsd4_cb_notify_lock_release,
 354 };
 355
 356 static inline struct nfs4_stateowner *
 357 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 358 {
 359         atomic_inc(&sop->so_count);
 360         return sop;
 361 }
 362
 363 static int
 364 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 365 {
 366         return (sop->so_owner.len == owner->len) &&
 367                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 368 }
 369
 370 static struct nfs4_openowner *
 371 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 372                         struct nfs4_client *clp)
 373 {
 374         struct nfs4_stateowner *so;
 375
 376         lockdep_assert_held(&clp->cl_lock);
 377
 378         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 379                             so_strhash) {
 380                 if (!so->so_is_open_owner)
 381                         continue;
 382                 if (same_owner_str(so, &open->op_owner))
 383                         return openowner(nfs4_get_stateowner(so));
 384         }
 385         return NULL;
 386 }
 387
 388 static struct nfs4_openowner *
 389 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 390                         struct nfs4_client *clp)
 391 {
 392         struct nfs4_openowner *oo;
 393
 394         spin_lock(&clp->cl_lock);
 395         oo = find_openstateowner_str_locked(hashval, open, clp);
 396         spin_unlock(&clp->cl_lock);
 397         return oo;
 398 }
 399
 400 static inline u32
 401 opaque_hashval(const void *ptr, int nbytes)
 402 {
 403         unsigned char *cptr = (unsigned char *) ptr;
 404
 405         u32 x = 0;
 406         while (nbytes--) {
 407                 x *= 37;
 408                 x += *cptr++;
 409         }
 410         return x;
 411 }
 412
 413 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 414 {
 415         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 416
 417         kmem_cache_free(file_slab, fp);
 418 }
 419
 420 void
 421 put_nfs4_file(struct nfs4_file *fi)
 422 {
 423         might_lock(&state_lock);
 424
 425         if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
 426                 hlist_del_rcu(&fi->fi_hash);
 427                 spin_unlock(&state_lock);
 428                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 429                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 430                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 431         }
 432 }
 433
 434 static struct nfsd_file *
 435 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 436 {
 437         if (f->fi_fds[oflag])
 438                 return nfsd_file_get(f->fi_fds[oflag]);
 439         return NULL;
 440 }
 441
 442 static struct nfsd_file *
 443 find_writeable_file_locked(struct nfs4_file *f)
 444 {
 445         struct nfsd_file *ret;
 446
 447         lockdep_assert_held(&f->fi_lock);
 448
 449         ret = __nfs4_get_fd(f, O_WRONLY);
 450         if (!ret)
 451                 ret = __nfs4_get_fd(f, O_RDWR);
 452         return ret;
 453 }
 454
 455 static struct nfsd_file *
 456 find_writeable_file(struct nfs4_file *f)
 457 {
 458         struct nfsd_file *ret;
 459
 460         spin_lock(&f->fi_lock);
 461         ret = find_writeable_file_locked(f);
 462         spin_unlock(&f->fi_lock);
 463
 464         return ret;
 465 }
 466
 467 static struct nfsd_file *
 468 find_readable_file_locked(struct nfs4_file *f)
 469 {
 470         struct nfsd_file *ret;
 471
 472         lockdep_assert_held(&f->fi_lock);
 473
 474         ret = __nfs4_get_fd(f, O_RDONLY);
 475         if (!ret)
 476                 ret = __nfs4_get_fd(f, O_RDWR);
 477         return ret;
 478 }
 479
 480 static struct nfsd_file *
 481 find_readable_file(struct nfs4_file *f)
 482 {
 483         struct nfsd_file *ret;
 484
 485         spin_lock(&f->fi_lock);
 486         ret = find_readable_file_locked(f);
 487         spin_unlock(&f->fi_lock);
 488
 489         return ret;
 490 }
 491
 492 struct nfsd_file *
 493 find_any_file(struct nfs4_file *f)
 494 {
 495         struct nfsd_file *ret;
 496
 497         if (!f)
 498                 return NULL;
 499         spin_lock(&f->fi_lock);
 500         ret = __nfs4_get_fd(f, O_RDWR);
 501         if (!ret) {
 502                 ret = __nfs4_get_fd(f, O_WRONLY);
 503                 if (!ret)
 504                         ret = __nfs4_get_fd(f, O_RDONLY);
 505         }
 506         spin_unlock(&f->fi_lock);
 507         return ret;
 508 }
 509
 510 static struct nfsd_file *find_deleg_file(struct nfs4_file *f)
 511 {
 512         struct nfsd_file *ret = NULL;
 513
 514         spin_lock(&f->fi_lock);
 515         if (f->fi_deleg_file)
 516                 ret = nfsd_file_get(f->fi_deleg_file);
 517         spin_unlock(&f->fi_lock);
 518         return ret;
 519 }
 520
 521 static atomic_long_t num_delegations;
 522 unsigned long max_delegations;
 523
 524 /*
 525  * Open owner state (share locks)
 526  */
 527
 528 /* hash tables for lock and open owners */
 529 #define OWNER_HASH_BITS              8
 530 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 531 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 532
 533 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 534 {
 535         unsigned int ret;
 536
 537         ret = opaque_hashval(ownername->data, ownername->len);
 538         return ret & OWNER_HASH_MASK;
 539 }
 540
 541 /* hash table for nfs4_file */
 542 #define FILE_HASH_BITS                   8
 543 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 544
 545 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 546 {
 547         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 548 }
 549
 550 static unsigned int file_hashval(struct knfsd_fh *fh)
 551 {
 552         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 553 }
 554
 555 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 556
 557 static void
 558 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 559 {
 560         lockdep_assert_held(&fp->fi_lock);
 561
 562         if (access & NFS4_SHARE_ACCESS_WRITE)
 563                 atomic_inc(&fp->fi_access[O_WRONLY]);
 564         if (access & NFS4_SHARE_ACCESS_READ)
 565                 atomic_inc(&fp->fi_access[O_RDONLY]);
 566 }
 567
 568 static __be32
 569 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 570 {
 571         lockdep_assert_held(&fp->fi_lock);
 572
 573         /* Does this access mode make sense? */
 574         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 575                 return nfserr_inval;
 576
 577         /* Does it conflict with a deny mode already set? */
 578         if ((access & fp->fi_share_deny) != 0)
 579                 return nfserr_share_denied;
 580
 581         __nfs4_file_get_access(fp, access);
 582         return nfs_ok;
 583 }
 584
 585 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 586 {
 587         /* Common case is that there is no deny mode. */
 588         if (deny) {
 589                 /* Does this deny mode make sense? */
 590                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 591                         return nfserr_inval;
 592
 593                 if ((deny & NFS4_SHARE_DENY_READ) &&
 594                     atomic_read(&fp->fi_access[O_RDONLY]))
 595                         return nfserr_share_denied;
 596
 597                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 598                     atomic_read(&fp->fi_access[O_WRONLY]))
 599                         return nfserr_share_denied;
 600         }
 601         return nfs_ok;
 602 }
 603
 604 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 605 {
 606         might_lock(&fp->fi_lock);
 607
 608         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 609                 struct nfsd_file *f1 = NULL;
 610                 struct nfsd_file *f2 = NULL;
 611
 612                 swap(f1, fp->fi_fds[oflag]);
 613                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 614                         swap(f2, fp->fi_fds[O_RDWR]);
 615                 spin_unlock(&fp->fi_lock);
 616                 if (f1)
 617                         nfsd_file_put(f1);
 618                 if (f2)
 619                         nfsd_file_put(f2);
 620         }
 621 }
 622
 623 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 624 {
 625         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 626
 627         if (access & NFS4_SHARE_ACCESS_WRITE)
 628                 __nfs4_file_put_access(fp, O_WRONLY);
 629         if (access & NFS4_SHARE_ACCESS_READ)
 630                 __nfs4_file_put_access(fp, O_RDONLY);
 631 }
 632
 633 /*
 634  * Allocate a new open/delegation state counter. This is needed for
 635  * pNFS for proper return on close semantics.
 636  *
 637  * Note that we only allocate it for pNFS-enabled exports, otherwise
 638  * all pointers to struct nfs4_clnt_odstate are always NULL.
 639  */
 640 static struct nfs4_clnt_odstate *
 641 alloc_clnt_odstate(struct nfs4_client *clp)
 642 {
 643         struct nfs4_clnt_odstate *co;
 644
 645         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 646         if (co) {
 647                 co->co_client = clp;
 648                 refcount_set(&co->co_odcount, 1);
 649         }
 650         return co;
 651 }
 652
 653 static void
 654 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 655 {
 656         struct nfs4_file *fp = co->co_file;
 657
 658         lockdep_assert_held(&fp->fi_lock);
 659         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 660 }
 661
 662 static inline void
 663 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 664 {
 665         if (co)
 666                 refcount_inc(&co->co_odcount);
 667 }
 668
 669 static void
 670 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 671 {
 672         struct nfs4_file *fp;
 673
 674         if (!co)
 675                 return;
 676
 677         fp = co->co_file;
 678         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 679                 list_del(&co->co_perfile);
 680                 spin_unlock(&fp->fi_lock);
 681
 682                 nfsd4_return_all_file_layouts(co->co_client, fp);
 683                 kmem_cache_free(odstate_slab, co);
 684         }
 685 }
 686
 687 static struct nfs4_clnt_odstate *
 688 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 689 {
 690         struct nfs4_clnt_odstate *co;
 691         struct nfs4_client *cl;
 692
 693         if (!new)
 694                 return NULL;
 695
 696         cl = new->co_client;
 697
 698         spin_lock(&fp->fi_lock);
 699         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 700                 if (co->co_client == cl) {
 701                         get_clnt_odstate(co);
 702                         goto out;
 703                 }
 704         }
 705         co = new;
 706         co->co_file = fp;
 707         hash_clnt_odstate_locked(new);
 708 out:
 709         spin_unlock(&fp->fi_lock);
 710         return co;
 711 }
 712
 713 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 714                                   void (*sc_free)(struct nfs4_stid *))
 715 {
 716         struct nfs4_stid *stid;
 717         int new_id;
 718
 719         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 720         if (!stid)
 721                 return NULL;
 722
 723         idr_preload(GFP_KERNEL);
 724         spin_lock(&cl->cl_lock);
 725         /* Reserving 0 for start of file in nfsdfs "states" file: */
 726         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
 727         spin_unlock(&cl->cl_lock);
 728         idr_preload_end();
 729         if (new_id < 0)
 730                 goto out_free;
 731
 732         stid->sc_free = sc_free;
 733         stid->sc_client = cl;
 734         stid->sc_stateid.si_opaque.so_id = new_id;
 735         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 736         /* Will be incremented before return to client: */
 737         refcount_set(&stid->sc_count, 1);
 738         spin_lock_init(&stid->sc_lock);
 739         INIT_LIST_HEAD(&stid->sc_cp_list);
 740
 741         /*
 742          * It shouldn't be a problem to reuse an opaque stateid value.
 743          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 744          * example, a stray write retransmission could be accepted by
 745          * the server when it should have been rejected.  Therefore,
 746          * adopt a trick from the sctp code to attempt to maximize the
 747          * amount of time until an id is reused, by ensuring they always
 748          * "increase" (mod INT_MAX):
 749          */
 750         return stid;
 751 out_free:
 752         kmem_cache_free(slab, stid);
 753         return NULL;
 754 }
 755
 756 /*
 757  * Create a unique stateid_t to represent each COPY.
 758  */
 759 static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
 760                               unsigned char sc_type)
 761 {
 762         int new_id;
 763
 764         stid->stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
 765         stid->stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
 766         stid->sc_type = sc_type;
 767
 768         idr_preload(GFP_KERNEL);
 769         spin_lock(&nn->s2s_cp_lock);
 770         new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
 771         stid->stid.si_opaque.so_id = new_id;
 772         spin_unlock(&nn->s2s_cp_lock);
 773         idr_preload_end();
 774         if (new_id < 0)
 775                 return 0;
 776         return 1;
 777 }
 778
 779 int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
 780 {
 781         return nfs4_init_cp_state(nn, &copy->cp_stateid, NFS4_COPY_STID);
 782 }
 783
 784 struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
 785                                                      struct nfs4_stid *p_stid)
 786 {
 787         struct nfs4_cpntf_state *cps;
 788
 789         cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
 790         if (!cps)
 791                 return NULL;
 792         cps->cpntf_time = ktime_get_boottime_seconds();
 793         refcount_set(&cps->cp_stateid.sc_count, 1);
 794         if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
 795                 goto out_free;
 796         spin_lock(&nn->s2s_cp_lock);
 797         list_add(&cps->cp_list, &p_stid->sc_cp_list);
 798         spin_unlock(&nn->s2s_cp_lock);
 799         return cps;
 800 out_free:
 801         kfree(cps);
 802         return NULL;
 803 }
 804
 805 void nfs4_free_copy_state(struct nfsd4_copy *copy)
 806 {
 807         struct nfsd_net *nn;
 808
 809         WARN_ON_ONCE(copy->cp_stateid.sc_type != NFS4_COPY_STID);
 810         nn = net_generic(copy->cp_clp->net, nfsd_net_id);
 811         spin_lock(&nn->s2s_cp_lock);
 812         idr_remove(&nn->s2s_cp_stateids,
 813                    copy->cp_stateid.stid.si_opaque.so_id);
 814         spin_unlock(&nn->s2s_cp_lock);
 815 }
 816
 817 static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
 818 {
 819         struct nfs4_cpntf_state *cps;
 820         struct nfsd_net *nn;
 821
 822         nn = net_generic(net, nfsd_net_id);
 823         spin_lock(&nn->s2s_cp_lock);
 824         while (!list_empty(&stid->sc_cp_list)) {
 825                 cps = list_first_entry(&stid->sc_cp_list,
 826                                        struct nfs4_cpntf_state, cp_list);
 827                 _free_cpntf_state_locked(nn, cps);
 828         }
 829         spin_unlock(&nn->s2s_cp_lock);
 830 }
 831
 832 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 833 {
 834         struct nfs4_stid *stid;
 835
 836         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
 837         if (!stid)
 838                 return NULL;
 839
 840         return openlockstateid(stid);
 841 }
 842
 843 static void nfs4_free_deleg(struct nfs4_stid *stid)
 844 {
 845         kmem_cache_free(deleg_slab, stid);
 846         atomic_long_dec(&num_delegations);
 847 }
 848
 849 /*
 850  * When we recall a delegation, we should be careful not to hand it
 851  * out again straight away.
 852  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 853  * in which the filehandles of recalled delegations are "stored".
 854  * If a filehandle appear in either filter, a delegation is blocked.
 855  * When a delegation is recalled, the filehandle is stored in the "new"
 856  * filter.
 857  * Every 30 seconds we swap the filters and clear the "new" one,
 858  * unless both are empty of course.
 859  *
 860  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 861  * low 3 bytes as hash-table indices.
 862  *
 863  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 864  * is used to manage concurrent access.  Testing does not need the lock
 865  * except when swapping the two filters.
 866  */
 867 static DEFINE_SPINLOCK(blocked_delegations_lock);
 868 static struct bloom_pair {
 869         int     entries, old_entries;
 870         time64_t swap_time;
 871         int     new; /* index into 'set' */
 872         DECLARE_BITMAP(set[2], 256);
 873 } blocked_delegations;
 874
 875 static int delegation_blocked(struct knfsd_fh *fh)
 876 {
 877         u32 hash;
 878         struct bloom_pair *bd = &blocked_delegations;
 879
 880         if (bd->entries == 0)
 881                 return 0;
 882         if (ktime_get_seconds() - bd->swap_time > 30) {
 883                 spin_lock(&blocked_delegations_lock);
 884                 if (ktime_get_seconds() - bd->swap_time > 30) {
 885                         bd->entries -= bd->old_entries;
 886                         bd->old_entries = bd->entries;
 887                         memset(bd->set[bd->new], 0,
 888                                sizeof(bd->set[0]));
 889                         bd->new = 1-bd->new;
 890                         bd->swap_time = ktime_get_seconds();
 891                 }
 892                 spin_unlock(&blocked_delegations_lock);
 893         }
 894         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 895         if (test_bit(hash&255, bd->set[0]) &&
 896             test_bit((hash>>8)&255, bd->set[0]) &&
 897             test_bit((hash>>16)&255, bd->set[0]))
 898                 return 1;
 899
 900         if (test_bit(hash&255, bd->set[1]) &&
 901             test_bit((hash>>8)&255, bd->set[1]) &&
 902             test_bit((hash>>16)&255, bd->set[1]))
 903                 return 1;
 904
 905         return 0;
 906 }
 907
 908 static void block_delegations(struct knfsd_fh *fh)
 909 {
 910         u32 hash;
 911         struct bloom_pair *bd = &blocked_delegations;
 912
 913         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 914
 915         spin_lock(&blocked_delegations_lock);
 916         __set_bit(hash&255, bd->set[bd->new]);
 917         __set_bit((hash>>8)&255, bd->set[bd->new]);
 918         __set_bit((hash>>16)&255, bd->set[bd->new]);
 919         if (bd->entries == 0)
 920                 bd->swap_time = ktime_get_seconds();
 921         bd->entries += 1;
 922         spin_unlock(&blocked_delegations_lock);
 923 }
 924
 925 static struct nfs4_delegation *
 926 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
 927                  struct svc_fh *current_fh,
 928                  struct nfs4_clnt_odstate *odstate)
 929 {
 930         struct nfs4_delegation *dp;
 931         long n;
 932
 933         dprintk("NFSD alloc_init_deleg\n");
 934         n = atomic_long_inc_return(&num_delegations);
 935         if (n < 0 || n > max_delegations)
 936                 goto out_dec;
 937         if (delegation_blocked(&current_fh->fh_handle))
 938                 goto out_dec;
 939         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
 940         if (dp == NULL)
 941                 goto out_dec;
 942
 943         /*
 944          * delegation seqid's are never incremented.  The 4.1 special
 945          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 946          * 0 anyway just for consistency and use 1:
 947          */
 948         dp->dl_stid.sc_stateid.si_generation = 1;
 949         INIT_LIST_HEAD(&dp->dl_perfile);
 950         INIT_LIST_HEAD(&dp->dl_perclnt);
 951         INIT_LIST_HEAD(&dp->dl_recall_lru);
 952         dp->dl_clnt_odstate = odstate;
 953         get_clnt_odstate(odstate);
 954         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 955         dp->dl_retries = 1;
 956         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 957                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 958         get_nfs4_file(fp);
 959         dp->dl_stid.sc_file = fp;
 960         return dp;
 961 out_dec:
 962         atomic_long_dec(&num_delegations);
 963         return NULL;
 964 }
 965
 966 void
 967 nfs4_put_stid(struct nfs4_stid *s)
 968 {
 969         struct nfs4_file *fp = s->sc_file;
 970         struct nfs4_client *clp = s->sc_client;
 971
 972         might_lock(&clp->cl_lock);
 973
 974         if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 975                 wake_up_all(&close_wq);
 976                 return;
 977         }
 978         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 979         nfs4_free_cpntf_statelist(clp->net, s);
 980         spin_unlock(&clp->cl_lock);
 981         s->sc_free(s);
 982         if (fp)
 983                 put_nfs4_file(fp);
 984 }
 985
 986 void
 987 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
 988 {
 989         stateid_t *src = &stid->sc_stateid;
 990
 991         spin_lock(&stid->sc_lock);
 992         if (unlikely(++src->si_generation == 0))
 993                 src->si_generation = 1;
 994         memcpy(dst, src, sizeof(*dst));
 995         spin_unlock(&stid->sc_lock);
 996 }
 997
 998 static void put_deleg_file(struct nfs4_file *fp)
 999 {
1000         struct nfsd_file *nf = NULL;
1001
1002         spin_lock(&fp->fi_lock);
1003         if (--fp->fi_delegees == 0)
1004                 swap(nf, fp->fi_deleg_file);
1005         spin_unlock(&fp->fi_lock);
1006
1007         if (nf)
1008                 nfsd_file_put(nf);
1009 }
1010
1011 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
1012 {
1013         struct nfs4_file *fp = dp->dl_stid.sc_file;
1014         struct nfsd_file *nf = fp->fi_deleg_file;
1015
1016         WARN_ON_ONCE(!fp->fi_delegees);
1017
1018         vfs_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
1019         put_deleg_file(fp);
1020 }
1021
1022 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
1023 {
1024         put_clnt_odstate(dp->dl_clnt_odstate);
1025         nfs4_unlock_deleg_lease(dp);
1026         nfs4_put_stid(&dp->dl_stid);
1027 }
1028
1029 void nfs4_unhash_stid(struct nfs4_stid *s)
1030 {
1031         s->sc_type = 0;
1032 }
1033
1034 /**
1035  * nfs4_delegation_exists - Discover if this delegation already exists
1036  * @clp:     a pointer to the nfs4_client we're granting a delegation to
1037  * @fp:      a pointer to the nfs4_file we're granting a delegation on
1038  *
1039  * Return:
1040  *      On success: true iff an existing delegation is found
1041  */
1042
1043 static bool
1044 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
1045 {
1046         struct nfs4_delegation *searchdp = NULL;
1047         struct nfs4_client *searchclp = NULL;
1048
1049         lockdep_assert_held(&state_lock);
1050         lockdep_assert_held(&fp->fi_lock);
1051
1052         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
1053                 searchclp = searchdp->dl_stid.sc_client;
1054                 if (clp == searchclp) {
1055                         return true;
1056                 }
1057         }
1058         return false;
1059 }
1060
1061 /**
1062  * hash_delegation_locked - Add a delegation to the appropriate lists
1063  * @dp:     a pointer to the nfs4_delegation we are adding.
1064  * @fp:     a pointer to the nfs4_file we're granting a delegation on
1065  *
1066  * Return:
1067  *      On success: NULL if the delegation was successfully hashed.
1068  *
1069  *      On error: -EAGAIN if one was previously granted to this
1070  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
1071  *
1072  */
1073
1074 static int
1075 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1076 {
1077         struct nfs4_client *clp = dp->dl_stid.sc_client;
1078
1079         lockdep_assert_held(&state_lock);
1080         lockdep_assert_held(&fp->fi_lock);
1081
1082         if (nfs4_delegation_exists(clp, fp))
1083                 return -EAGAIN;
1084         refcount_inc(&dp->dl_stid.sc_count);
1085         dp->dl_stid.sc_type = NFS4_DELEG_STID;
1086         list_add(&dp->dl_perfile, &fp->fi_delegations);
1087         list_add(&dp->dl_perclnt, &clp->cl_delegations);
1088         return 0;
1089 }
1090
1091 static bool
1092 unhash_delegation_locked(struct nfs4_delegation *dp)
1093 {
1094         struct nfs4_file *fp = dp->dl_stid.sc_file;
1095
1096         lockdep_assert_held(&state_lock);
1097
1098         if (list_empty(&dp->dl_perfile))
1099                 return false;
1100
1101         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
1102         /* Ensure that deleg break won't try to requeue it */
1103         ++dp->dl_time;
1104         spin_lock(&fp->fi_lock);
1105         list_del_init(&dp->dl_perclnt);
1106         list_del_init(&dp->dl_recall_lru);
1107         list_del_init(&dp->dl_perfile);
1108         spin_unlock(&fp->fi_lock);
1109         return true;
1110 }
1111
1112 static void destroy_delegation(struct nfs4_delegation *dp)
1113 {
1114         bool unhashed;
1115
1116         spin_lock(&state_lock);
1117         unhashed = unhash_delegation_locked(dp);
1118         spin_unlock(&state_lock);
1119         if (unhashed)
1120                 destroy_unhashed_deleg(dp);
1121 }
1122
1123 static void revoke_delegation(struct nfs4_delegation *dp)
1124 {
1125         struct nfs4_client *clp = dp->dl_stid.sc_client;
1126
1127         WARN_ON(!list_empty(&dp->dl_recall_lru));
1128
1129         if (clp->cl_minorversion) {
1130                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1131                 refcount_inc(&dp->dl_stid.sc_count);
1132                 spin_lock(&clp->cl_lock);
1133                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1134                 spin_unlock(&clp->cl_lock);
1135         }
1136         destroy_unhashed_deleg(dp);
1137 }
1138
1139 /*
1140  * SETCLIENTID state
1141  */
1142
1143 static unsigned int clientid_hashval(u32 id)
1144 {
1145         return id & CLIENT_HASH_MASK;
1146 }
1147
1148 static unsigned int clientstr_hashval(struct xdr_netobj name)
1149 {
1150         return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1151 }
1152
1153 /*
1154  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1155  * st_{access,deny}_bmap field of the stateid, in order to track not
1156  * only what share bits are currently in force, but also what
1157  * combinations of share bits previous opens have used.  This allows us
1158  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1159  * return an error if the client attempt to downgrade to a combination
1160  * of share bits not explicable by closing some of its previous opens.
1161  *
1162  * XXX: This enforcement is actually incomplete, since we don't keep
1163  * track of access/deny bit combinations; so, e.g., we allow:
1164  *
1165  *      OPEN allow read, deny write
1166  *      OPEN allow both, deny none
1167  *      DOWNGRADE allow read, deny none
1168  *
1169  * which we should reject.
1170  */
1171 static unsigned int
1172 bmap_to_share_mode(unsigned long bmap) {
1173         int i;
1174         unsigned int access = 0;
1175
1176         for (i = 1; i < 4; i++) {
1177                 if (test_bit(i, &bmap))
1178                         access |= i;
1179         }
1180         return access;
1181 }
1182
1183 /* set share access for a given stateid */
1184 static inline void
1185 set_access(u32 access, struct nfs4_ol_stateid *stp)
1186 {
1187         unsigned char mask = 1 << access;
1188
1189         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1190         stp->st_access_bmap |= mask;
1191 }
1192
1193 /* clear share access for a given stateid */
1194 static inline void
1195 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1196 {
1197         unsigned char mask = 1 << access;
1198
1199         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1200         stp->st_access_bmap &= ~mask;
1201 }
1202
1203 /* test whether a given stateid has access */
1204 static inline bool
1205 test_access(u32 access, struct nfs4_ol_stateid *stp)
1206 {
1207         unsigned char mask = 1 << access;
1208
1209         return (bool)(stp->st_access_bmap & mask);
1210 }
1211
1212 /* set share deny for a given stateid */
1213 static inline void
1214 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1215 {
1216         unsigned char mask = 1 << deny;
1217
1218         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1219         stp->st_deny_bmap |= mask;
1220 }
1221
1222 /* clear share deny for a given stateid */
1223 static inline void
1224 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1225 {
1226         unsigned char mask = 1 << deny;
1227
1228         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1229         stp->st_deny_bmap &= ~mask;
1230 }
1231
1232 /* test whether a given stateid is denying specific access */
1233 static inline bool
1234 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1235 {
1236         unsigned char mask = 1 << deny;
1237
1238         return (bool)(stp->st_deny_bmap & mask);
1239 }
1240
1241 static int nfs4_access_to_omode(u32 access)
1242 {
1243         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1244         case NFS4_SHARE_ACCESS_READ:
1245                 return O_RDONLY;
1246         case NFS4_SHARE_ACCESS_WRITE:
1247                 return O_WRONLY;
1248         case NFS4_SHARE_ACCESS_BOTH:
1249                 return O_RDWR;
1250         }
1251         WARN_ON_ONCE(1);
1252         return O_RDONLY;
1253 }
1254
1255 /*
1256  * A stateid that had a deny mode associated with it is being released
1257  * or downgraded. Recalculate the deny mode on the file.
1258  */
1259 static void
1260 recalculate_deny_mode(struct nfs4_file *fp)
1261 {
1262         struct nfs4_ol_stateid *stp;
1263
1264         spin_lock(&fp->fi_lock);
1265         fp->fi_share_deny = 0;
1266         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1267                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1268         spin_unlock(&fp->fi_lock);
1269 }
1270
1271 static void
1272 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1273 {
1274         int i;
1275         bool change = false;
1276
1277         for (i = 1; i < 4; i++) {
1278                 if ((i & deny) != i) {
1279                         change = true;
1280                         clear_deny(i, stp);
1281                 }
1282         }
1283
1284         /* Recalculate per-file deny mode if there was a change */
1285         if (change)
1286                 recalculate_deny_mode(stp->st_stid.sc_file);
1287 }
1288
1289 /* release all access and file references for a given stateid */
1290 static void
1291 release_all_access(struct nfs4_ol_stateid *stp)
1292 {
1293         int i;
1294         struct nfs4_file *fp = stp->st_stid.sc_file;
1295
1296         if (fp && stp->st_deny_bmap != 0)
1297                 recalculate_deny_mode(fp);
1298
1299         for (i = 1; i < 4; i++) {
1300                 if (test_access(i, stp))
1301                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1302                 clear_access(i, stp);
1303         }
1304 }
1305
1306 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1307 {
1308         kfree(sop->so_owner.data);
1309         sop->so_ops->so_free(sop);
1310 }
1311
1312 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1313 {
1314         struct nfs4_client *clp = sop->so_client;
1315
1316         might_lock(&clp->cl_lock);
1317
1318         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1319                 return;
1320         sop->so_ops->so_unhash(sop);
1321         spin_unlock(&clp->cl_lock);
1322         nfs4_free_stateowner(sop);
1323 }
1324
1325 static bool
1326 nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1327 {
1328         return list_empty(&stp->st_perfile);
1329 }
1330
1331 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1332 {
1333         struct nfs4_file *fp = stp->st_stid.sc_file;
1334
1335         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1336
1337         if (list_empty(&stp->st_perfile))
1338                 return false;
1339
1340         spin_lock(&fp->fi_lock);
1341         list_del_init(&stp->st_perfile);
1342         spin_unlock(&fp->fi_lock);
1343         list_del(&stp->st_perstateowner);
1344         return true;
1345 }
1346
1347 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1348 {
1349         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1350
1351         put_clnt_odstate(stp->st_clnt_odstate);
1352         release_all_access(stp);
1353         if (stp->st_stateowner)
1354                 nfs4_put_stateowner(stp->st_stateowner);
1355         kmem_cache_free(stateid_slab, stid);
1356 }
1357
1358 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1359 {
1360         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1361         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1362         struct nfsd_file *nf;
1363
1364         nf = find_any_file(stp->st_stid.sc_file);
1365         if (nf) {
1366                 get_file(nf->nf_file);
1367                 filp_close(nf->nf_file, (fl_owner_t)lo);
1368                 nfsd_file_put(nf);
1369         }
1370         nfs4_free_ol_stateid(stid);
1371 }
1372
1373 /*
1374  * Put the persistent reference to an already unhashed generic stateid, while
1375  * holding the cl_lock. If it's the last reference, then put it onto the
1376  * reaplist for later destruction.
1377  */
1378 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1379                                        struct list_head *reaplist)
1380 {
1381         struct nfs4_stid *s = &stp->st_stid;
1382         struct nfs4_client *clp = s->sc_client;
1383
1384         lockdep_assert_held(&clp->cl_lock);
1385
1386         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1387
1388         if (!refcount_dec_and_test(&s->sc_count)) {
1389                 wake_up_all(&close_wq);
1390                 return;
1391         }
1392
1393         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1394         list_add(&stp->st_locks, reaplist);
1395 }
1396
1397 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1398 {
1399         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1400
1401         if (!unhash_ol_stateid(stp))
1402                 return false;
1403         list_del_init(&stp->st_locks);
1404         nfs4_unhash_stid(&stp->st_stid);
1405         return true;
1406 }
1407
1408 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1409 {
1410         struct nfs4_client *clp = stp->st_stid.sc_client;
1411         bool unhashed;
1412
1413         spin_lock(&clp->cl_lock);
1414         unhashed = unhash_lock_stateid(stp);
1415         spin_unlock(&clp->cl_lock);
1416         if (unhashed)
1417                 nfs4_put_stid(&stp->st_stid);
1418 }
1419
1420 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1421 {
1422         struct nfs4_client *clp = lo->lo_owner.so_client;
1423
1424         lockdep_assert_held(&clp->cl_lock);
1425
1426         list_del_init(&lo->lo_owner.so_strhash);
1427 }
1428
1429 /*
1430  * Free a list of generic stateids that were collected earlier after being
1431  * fully unhashed.
1432  */
1433 static void
1434 free_ol_stateid_reaplist(struct list_head *reaplist)
1435 {
1436         struct nfs4_ol_stateid *stp;
1437         struct nfs4_file *fp;
1438
1439         might_sleep();
1440
1441         while (!list_empty(reaplist)) {
1442                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1443                                        st_locks);
1444                 list_del(&stp->st_locks);
1445                 fp = stp->st_stid.sc_file;
1446                 stp->st_stid.sc_free(&stp->st_stid);
1447                 if (fp)
1448                         put_nfs4_file(fp);
1449         }
1450 }
1451
1452 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1453                                        struct list_head *reaplist)
1454 {
1455         struct nfs4_ol_stateid *stp;
1456
1457         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1458
1459         while (!list_empty(&open_stp->st_locks)) {
1460                 stp = list_entry(open_stp->st_locks.next,
1461                                 struct nfs4_ol_stateid, st_locks);
1462                 WARN_ON(!unhash_lock_stateid(stp));
1463                 put_ol_stateid_locked(stp, reaplist);
1464         }
1465 }
1466
1467 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1468                                 struct list_head *reaplist)
1469 {
1470         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1471
1472         if (!unhash_ol_stateid(stp))
1473                 return false;
1474         release_open_stateid_locks(stp, reaplist);
1475         return true;
1476 }
1477
1478 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1479 {
1480         LIST_HEAD(reaplist);
1481
1482         spin_lock(&stp->st_stid.sc_client->cl_lock);
1483         if (unhash_open_stateid(stp, &reaplist))
1484                 put_ol_stateid_locked(stp, &reaplist);
1485         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1486         free_ol_stateid_reaplist(&reaplist);
1487 }
1488
1489 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1490 {
1491         struct nfs4_client *clp = oo->oo_owner.so_client;
1492
1493         lockdep_assert_held(&clp->cl_lock);
1494
1495         list_del_init(&oo->oo_owner.so_strhash);
1496         list_del_init(&oo->oo_perclient);
1497 }
1498
1499 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1500 {
1501         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1502                                           nfsd_net_id);
1503         struct nfs4_ol_stateid *s;
1504
1505         spin_lock(&nn->client_lock);
1506         s = oo->oo_last_closed_stid;
1507         if (s) {
1508                 list_del_init(&oo->oo_close_lru);
1509                 oo->oo_last_closed_stid = NULL;
1510         }
1511         spin_unlock(&nn->client_lock);
1512         if (s)
1513                 nfs4_put_stid(&s->st_stid);
1514 }
1515
1516 static void release_openowner(struct nfs4_openowner *oo)
1517 {
1518         struct nfs4_ol_stateid *stp;
1519         struct nfs4_client *clp = oo->oo_owner.so_client;
1520         struct list_head reaplist;
1521
1522         INIT_LIST_HEAD(&reaplist);
1523
1524         spin_lock(&clp->cl_lock);
1525         unhash_openowner_locked(oo);
1526         while (!list_empty(&oo->oo_owner.so_stateids)) {
1527                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1528                                 struct nfs4_ol_stateid, st_perstateowner);
1529                 if (unhash_open_stateid(stp, &reaplist))
1530                         put_ol_stateid_locked(stp, &reaplist);
1531         }
1532         spin_unlock(&clp->cl_lock);
1533         free_ol_stateid_reaplist(&reaplist);
1534         release_last_closed_stateid(oo);
1535         nfs4_put_stateowner(&oo->oo_owner);
1536 }
1537
1538 static inline int
1539 hash_sessionid(struct nfs4_sessionid *sessionid)
1540 {
1541         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1542
1543         return sid->sequence % SESSION_HASH_SIZE;
1544 }
1545
1546 #ifdef CONFIG_SUNRPC_DEBUG
1547 static inline void
1548 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1549 {
1550         u32 *ptr = (u32 *)(&sessionid->data[0]);
1551         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1552 }
1553 #else
1554 static inline void
1555 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1556 {
1557 }
1558 #endif
1559
1560 /*
1561  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1562  * won't be used for replay.
1563  */
1564 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1565 {
1566         struct nfs4_stateowner *so = cstate->replay_owner;
1567
1568         if (nfserr == nfserr_replay_me)
1569                 return;
1570
1571         if (!seqid_mutating_err(ntohl(nfserr))) {
1572                 nfsd4_cstate_clear_replay(cstate);
1573                 return;
1574         }
1575         if (!so)
1576                 return;
1577         if (so->so_is_open_owner)
1578                 release_last_closed_stateid(openowner(so));
1579         so->so_seqid++;
1580         return;
1581 }
1582
1583 static void
1584 gen_sessionid(struct nfsd4_session *ses)
1585 {
1586         struct nfs4_client *clp = ses->se_client;
1587         struct nfsd4_sessionid *sid;
1588
1589         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1590         sid->clientid = clp->cl_clientid;
1591         sid->sequence = current_sessionid++;
1592         sid->reserved = 0;
1593 }
1594
1595 /*
1596  * The protocol defines ca_maxresponssize_cached to include the size of
1597  * the rpc header, but all we need to cache is the data starting after
1598  * the end of the initial SEQUENCE operation--the rest we regenerate
1599  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1600  * value that is the number of bytes in our cache plus a few additional
1601  * bytes.  In order to stay on the safe side, and not promise more than
1602  * we can cache, those additional bytes must be the minimum possible: 24
1603  * bytes of rpc header (xid through accept state, with AUTH_NULL
1604  * verifier), 12 for the compound header (with zero-length tag), and 44
1605  * for the SEQUENCE op response:
1606  */
1607 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1608
1609 static void
1610 free_session_slots(struct nfsd4_session *ses)
1611 {
1612         int i;
1613
1614         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1615                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1616                 kfree(ses->se_slots[i]);
1617         }
1618 }
1619
1620 /*
1621  * We don't actually need to cache the rpc and session headers, so we
1622  * can allocate a little less for each slot:
1623  */
1624 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1625 {
1626         u32 size;
1627
1628         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1629                 size = 0;
1630         else
1631                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1632         return size + sizeof(struct nfsd4_slot);
1633 }
1634
1635 /*
1636  * XXX: If we run out of reserved DRC memory we could (up to a point)
1637  * re-negotiate active sessions and reduce their slot usage to make
1638  * room for new connections. For now we just fail the create session.
1639  */
1640 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1641 {
1642         u32 slotsize = slot_bytes(ca);
1643         u32 num = ca->maxreqs;
1644         unsigned long avail, total_avail;
1645         unsigned int scale_factor;
1646
1647         spin_lock(&nfsd_drc_lock);
1648         if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1649                 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1650         else
1651                 /* We have handed out more space than we chose in
1652                  * set_max_drc() to allow.  That isn't really a
1653                  * problem as long as that doesn't make us think we
1654                  * have lots more due to integer overflow.
1655                  */
1656                 total_avail = 0;
1657         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1658         /*
1659          * Never use more than a fraction of the remaining memory,
1660          * unless it's the only way to give this client a slot.
1661          * The chosen fraction is either 1/8 or 1/number of threads,
1662          * whichever is smaller.  This ensures there are adequate
1663          * slots to support multiple clients per thread.
1664          * Give the client one slot even if that would require
1665          * over-allocation--it is better than failure.
1666          */
1667         scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1668
1669         avail = clamp_t(unsigned long, avail, slotsize,
1670                         total_avail/scale_factor);
1671         num = min_t(int, num, avail / slotsize);
1672         num = max_t(int, num, 1);
1673         nfsd_drc_mem_used += num * slotsize;
1674         spin_unlock(&nfsd_drc_lock);
1675
1676         return num;
1677 }
1678
1679 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1680 {
1681         int slotsize = slot_bytes(ca);
1682
1683         spin_lock(&nfsd_drc_lock);
1684         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1685         spin_unlock(&nfsd_drc_lock);
1686 }
1687
1688 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1689                                            struct nfsd4_channel_attrs *battrs)
1690 {
1691         int numslots = fattrs->maxreqs;
1692         int slotsize = slot_bytes(fattrs);
1693         struct nfsd4_session *new;
1694         int mem, i;
1695
1696         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1697                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1698         mem = numslots * sizeof(struct nfsd4_slot *);
1699
1700         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1701         if (!new)
1702                 return NULL;
1703         /* allocate each struct nfsd4_slot and data cache in one piece */
1704         for (i = 0; i < numslots; i++) {
1705                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1706                 if (!new->se_slots[i])
1707                         goto out_free;
1708         }
1709
1710         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1711         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1712
1713         return new;
1714 out_free:
1715         while (i--)
1716                 kfree(new->se_slots[i]);
1717         kfree(new);
1718         return NULL;
1719 }
1720
1721 static void free_conn(struct nfsd4_conn *c)
1722 {
1723         svc_xprt_put(c->cn_xprt);
1724         kfree(c);
1725 }
1726
1727 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1728 {
1729         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1730         struct nfs4_client *clp = c->cn_session->se_client;
1731
1732         spin_lock(&clp->cl_lock);
1733         if (!list_empty(&c->cn_persession)) {
1734                 list_del(&c->cn_persession);
1735                 free_conn(c);
1736         }
1737         nfsd4_probe_callback(clp);
1738         spin_unlock(&clp->cl_lock);
1739 }
1740
1741 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1742 {
1743         struct nfsd4_conn *conn;
1744
1745         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1746         if (!conn)
1747                 return NULL;
1748         svc_xprt_get(rqstp->rq_xprt);
1749         conn->cn_xprt = rqstp->rq_xprt;
1750         conn->cn_flags = flags;
1751         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1752         return conn;
1753 }
1754
1755 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1756 {
1757         conn->cn_session = ses;
1758         list_add(&conn->cn_persession, &ses->se_conns);
1759 }
1760
1761 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1762 {
1763         struct nfs4_client *clp = ses->se_client;
1764
1765         spin_lock(&clp->cl_lock);
1766         __nfsd4_hash_conn(conn, ses);
1767         spin_unlock(&clp->cl_lock);
1768 }
1769
1770 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1771 {
1772         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1773         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1774 }
1775
1776 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1777 {
1778         int ret;
1779
1780         nfsd4_hash_conn(conn, ses);
1781         ret = nfsd4_register_conn(conn);
1782         if (ret)
1783                 /* oops; xprt is already down: */
1784                 nfsd4_conn_lost(&conn->cn_xpt_user);
1785         /* We may have gained or lost a callback channel: */
1786         nfsd4_probe_callback_sync(ses->se_client);
1787 }
1788
1789 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1790 {
1791         u32 dir = NFS4_CDFC4_FORE;
1792
1793         if (cses->flags & SESSION4_BACK_CHAN)
1794                 dir |= NFS4_CDFC4_BACK;
1795         return alloc_conn(rqstp, dir);
1796 }
1797
1798 /* must be called under client_lock */
1799 static void nfsd4_del_conns(struct nfsd4_session *s)
1800 {
1801         struct nfs4_client *clp = s->se_client;
1802         struct nfsd4_conn *c;
1803
1804         spin_lock(&clp->cl_lock);
1805         while (!list_empty(&s->se_conns)) {
1806                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1807                 list_del_init(&c->cn_persession);
1808                 spin_unlock(&clp->cl_lock);
1809
1810                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1811                 free_conn(c);
1812
1813                 spin_lock(&clp->cl_lock);
1814         }
1815         spin_unlock(&clp->cl_lock);
1816 }
1817
1818 static void __free_session(struct nfsd4_session *ses)
1819 {
1820         free_session_slots(ses);
1821         kfree(ses);
1822 }
1823
1824 static void free_session(struct nfsd4_session *ses)
1825 {
1826         nfsd4_del_conns(ses);
1827         nfsd4_put_drc_mem(&ses->se_fchannel);
1828         __free_session(ses);
1829 }
1830
1831 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1832 {
1833         int idx;
1834         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1835
1836         new->se_client = clp;
1837         gen_sessionid(new);
1838
1839         INIT_LIST_HEAD(&new->se_conns);
1840
1841         new->se_cb_seq_nr = 1;
1842         new->se_flags = cses->flags;
1843         new->se_cb_prog = cses->callback_prog;
1844         new->se_cb_sec = cses->cb_sec;
1845         atomic_set(&new->se_ref, 0);
1846         idx = hash_sessionid(&new->se_sessionid);
1847         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1848         spin_lock(&clp->cl_lock);
1849         list_add(&new->se_perclnt, &clp->cl_sessions);
1850         spin_unlock(&clp->cl_lock);
1851
1852         {
1853                 struct sockaddr *sa = svc_addr(rqstp);
1854                 /*
1855                  * This is a little silly; with sessions there's no real
1856                  * use for the callback address.  Use the peer address
1857                  * as a reasonable default for now, but consider fixing
1858                  * the rpc client not to require an address in the
1859                  * future:
1860                  */
1861                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1862                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1863         }
1864 }
1865
1866 /* caller must hold client_lock */
1867 static struct nfsd4_session *
1868 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1869 {
1870         struct nfsd4_session *elem;
1871         int idx;
1872         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1873
1874         lockdep_assert_held(&nn->client_lock);
1875
1876         dump_sessionid(__func__, sessionid);
1877         idx = hash_sessionid(sessionid);
1878         /* Search in the appropriate list */
1879         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1880                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1881                             NFS4_MAX_SESSIONID_LEN)) {
1882                         return elem;
1883                 }
1884         }
1885
1886         dprintk("%s: session not found\n", __func__);
1887         return NULL;
1888 }
1889
1890 static struct nfsd4_session *
1891 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1892                 __be32 *ret)
1893 {
1894         struct nfsd4_session *session;
1895         __be32 status = nfserr_badsession;
1896
1897         session = __find_in_sessionid_hashtbl(sessionid, net);
1898         if (!session)
1899                 goto out;
1900         status = nfsd4_get_session_locked(session);
1901         if (status)
1902                 session = NULL;
1903 out:
1904         *ret = status;
1905         return session;
1906 }
1907
1908 /* caller must hold client_lock */
1909 static void
1910 unhash_session(struct nfsd4_session *ses)
1911 {
1912         struct nfs4_client *clp = ses->se_client;
1913         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1914
1915         lockdep_assert_held(&nn->client_lock);
1916
1917         list_del(&ses->se_hash);
1918         spin_lock(&ses->se_client->cl_lock);
1919         list_del(&ses->se_perclnt);
1920         spin_unlock(&ses->se_client->cl_lock);
1921 }
1922
1923 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1924 static int
1925 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1926 {
1927         /*
1928          * We're assuming the clid was not given out from a boot
1929          * precisely 2^32 (about 136 years) before this one.  That seems
1930          * a safe assumption:
1931          */
1932         if (clid->cl_boot == (u32)nn->boot_time)
1933                 return 0;
1934         trace_nfsd_clid_stale(clid);
1935         return 1;
1936 }
1937
1938 /*
1939  * XXX Should we use a slab cache ?
1940  * This type of memory management is somewhat inefficient, but we use it
1941  * anyway since SETCLIENTID is not a common operation.
1942  */
1943 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1944 {
1945         struct nfs4_client *clp;
1946         int i;
1947
1948         clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
1949         if (clp == NULL)
1950                 return NULL;
1951         xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
1952         if (clp->cl_name.data == NULL)
1953                 goto err_no_name;
1954         clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
1955                                                  sizeof(struct list_head),
1956                                                  GFP_KERNEL);
1957         if (!clp->cl_ownerstr_hashtbl)
1958                 goto err_no_hashtbl;
1959         for (i = 0; i < OWNER_HASH_SIZE; i++)
1960                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1961         INIT_LIST_HEAD(&clp->cl_sessions);
1962         idr_init(&clp->cl_stateids);
1963         atomic_set(&clp->cl_rpc_users, 0);
1964         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1965         INIT_LIST_HEAD(&clp->cl_idhash);
1966         INIT_LIST_HEAD(&clp->cl_openowners);
1967         INIT_LIST_HEAD(&clp->cl_delegations);
1968         INIT_LIST_HEAD(&clp->cl_lru);
1969         INIT_LIST_HEAD(&clp->cl_revoked);
1970 #ifdef CONFIG_NFSD_PNFS
1971         INIT_LIST_HEAD(&clp->cl_lo_states);
1972 #endif
1973         INIT_LIST_HEAD(&clp->async_copies);
1974         spin_lock_init(&clp->async_lock);
1975         spin_lock_init(&clp->cl_lock);
1976         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1977         return clp;
1978 err_no_hashtbl:
1979         kfree(clp->cl_name.data);
1980 err_no_name:
1981         kmem_cache_free(client_slab, clp);
1982         return NULL;
1983 }
1984
1985 static void __free_client(struct kref *k)
1986 {
1987         struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
1988         struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
1989
1990         free_svc_cred(&clp->cl_cred);
1991         kfree(clp->cl_ownerstr_hashtbl);
1992         kfree(clp->cl_name.data);
1993         kfree(clp->cl_nii_domain.data);
1994         kfree(clp->cl_nii_name.data);
1995         idr_destroy(&clp->cl_stateids);
1996         kmem_cache_free(client_slab, clp);
1997 }
1998
1999 static void drop_client(struct nfs4_client *clp)
2000 {
2001         kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
2002 }
2003
2004 static void
2005 free_client(struct nfs4_client *clp)
2006 {
2007         while (!list_empty(&clp->cl_sessions)) {
2008                 struct nfsd4_session *ses;
2009                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
2010                                 se_perclnt);
2011                 list_del(&ses->se_perclnt);
2012                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
2013                 free_session(ses);
2014         }
2015         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
2016         if (clp->cl_nfsd_dentry) {
2017                 nfsd_client_rmdir(clp->cl_nfsd_dentry);
2018                 clp->cl_nfsd_dentry = NULL;
2019                 wake_up_all(&expiry_wq);
2020         }
2021         drop_client(clp);
2022 }
2023
2024 /* must be called under the client_lock */
2025 static void
2026 unhash_client_locked(struct nfs4_client *clp)
2027 {
2028         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2029         struct nfsd4_session *ses;
2030
2031         lockdep_assert_held(&nn->client_lock);
2032
2033         /* Mark the client as expired! */
2034         clp->cl_time = 0;
2035         /* Make it invisible */
2036         if (!list_empty(&clp->cl_idhash)) {
2037                 list_del_init(&clp->cl_idhash);
2038                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2039                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
2040                 else
2041                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2042         }
2043         list_del_init(&clp->cl_lru);
2044         spin_lock(&clp->cl_lock);
2045         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2046                 list_del_init(&ses->se_hash);
2047         spin_unlock(&clp->cl_lock);
2048 }
2049
2050 static void
2051 unhash_client(struct nfs4_client *clp)
2052 {
2053         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2054
2055         spin_lock(&nn->client_lock);
2056         unhash_client_locked(clp);
2057         spin_unlock(&nn->client_lock);
2058 }
2059
2060 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
2061 {
2062         if (atomic_read(&clp->cl_rpc_users))
2063                 return nfserr_jukebox;
2064         unhash_client_locked(clp);
2065         return nfs_ok;
2066 }
2067
2068 static void
2069 __destroy_client(struct nfs4_client *clp)
2070 {
2071         int i;
2072         struct nfs4_openowner *oo;
2073         struct nfs4_delegation *dp;
2074         struct list_head reaplist;
2075
2076         INIT_LIST_HEAD(&reaplist);
2077         spin_lock(&state_lock);
2078         while (!list_empty(&clp->cl_delegations)) {
2079                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2080                 WARN_ON(!unhash_delegation_locked(dp));
2081                 list_add(&dp->dl_recall_lru, &reaplist);
2082         }
2083         spin_unlock(&state_lock);
2084         while (!list_empty(&reaplist)) {
2085                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2086                 list_del_init(&dp->dl_recall_lru);
2087                 destroy_unhashed_deleg(dp);
2088         }
2089         while (!list_empty(&clp->cl_revoked)) {
2090                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2091                 list_del_init(&dp->dl_recall_lru);
2092                 nfs4_put_stid(&dp->dl_stid);
2093         }
2094         while (!list_empty(&clp->cl_openowners)) {
2095                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2096                 nfs4_get_stateowner(&oo->oo_owner);
2097                 release_openowner(oo);
2098         }
2099         for (i = 0; i < OWNER_HASH_SIZE; i++) {
2100                 struct nfs4_stateowner *so, *tmp;
2101
2102                 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2103                                          so_strhash) {
2104                         /* Should be no openowners at this point */
2105                         WARN_ON_ONCE(so->so_is_open_owner);
2106                         remove_blocked_locks(lockowner(so));
2107                 }
2108         }
2109         nfsd4_return_all_client_layouts(clp);
2110         nfsd4_shutdown_copy(clp);
2111         nfsd4_shutdown_callback(clp);
2112         if (clp->cl_cb_conn.cb_xprt)
2113                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2114         free_client(clp);
2115         wake_up_all(&expiry_wq);
2116 }
2117
2118 static void
2119 destroy_client(struct nfs4_client *clp)
2120 {
2121         unhash_client(clp);
2122         __destroy_client(clp);
2123 }
2124
2125 static void inc_reclaim_complete(struct nfs4_client *clp)
2126 {
2127         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2128
2129         if (!nn->track_reclaim_completes)
2130                 return;
2131         if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2132                 return;
2133         if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2134                         nn->reclaim_str_hashtbl_size) {
2135                 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2136                                 clp->net->ns.inum);
2137                 nfsd4_end_grace(nn);
2138         }
2139 }
2140
2141 static void expire_client(struct nfs4_client *clp)
2142 {
2143         unhash_client(clp);
2144         nfsd4_client_record_remove(clp);
2145         __destroy_client(clp);
2146 }
2147
2148 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2149 {
2150         memcpy(target->cl_verifier.data, source->data,
2151                         sizeof(target->cl_verifier.data));
2152 }
2153
2154 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2155 {
2156         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2157         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2158 }
2159
2160 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2161 {
2162         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2163         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2164                                                                 GFP_KERNEL);
2165         target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2166         if ((source->cr_principal && !target->cr_principal) ||
2167             (source->cr_raw_principal && !target->cr_raw_principal) ||
2168             (source->cr_targ_princ && !target->cr_targ_princ))
2169                 return -ENOMEM;
2170
2171         target->cr_flavor = source->cr_flavor;
2172         target->cr_uid = source->cr_uid;
2173         target->cr_gid = source->cr_gid;
2174         target->cr_group_info = source->cr_group_info;
2175         get_group_info(target->cr_group_info);
2176         target->cr_gss_mech = source->cr_gss_mech;
2177         if (source->cr_gss_mech)
2178                 gss_mech_get(source->cr_gss_mech);
2179         return 0;
2180 }
2181
2182 static int
2183 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2184 {
2185         if (o1->len < o2->len)
2186                 return -1;
2187         if (o1->len > o2->len)
2188                 return 1;
2189         return memcmp(o1->data, o2->data, o1->len);
2190 }
2191
2192 static int
2193 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2194 {
2195         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2196 }
2197
2198 static int
2199 same_clid(clientid_t *cl1, clientid_t *cl2)
2200 {
2201         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2202 }
2203
2204 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2205 {
2206         int i;
2207
2208         if (g1->ngroups != g2->ngroups)
2209                 return false;
2210         for (i=0; i<g1->ngroups; i++)
2211                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2212                         return false;
2213         return true;
2214 }
2215
2216 /*
2217  * RFC 3530 language requires clid_inuse be returned when the
2218  * "principal" associated with a requests differs from that previously
2219  * used.  We use uid, gid's, and gss principal string as our best
2220  * approximation.  We also don't want to allow non-gss use of a client
2221  * established using gss: in theory cr_principal should catch that
2222  * change, but in practice cr_principal can be null even in the gss case
2223  * since gssd doesn't always pass down a principal string.
2224  */
2225 static bool is_gss_cred(struct svc_cred *cr)
2226 {
2227         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2228         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2229 }
2230
2231
2232 static bool
2233 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2234 {
2235         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2236                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2237                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2238                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2239                 return false;
2240         /* XXX: check that cr_targ_princ fields match ? */
2241         if (cr1->cr_principal == cr2->cr_principal)
2242                 return true;
2243         if (!cr1->cr_principal || !cr2->cr_principal)
2244                 return false;
2245         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2246 }
2247
2248 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2249 {
2250         struct svc_cred *cr = &rqstp->rq_cred;
2251         u32 service;
2252
2253         if (!cr->cr_gss_mech)
2254                 return false;
2255         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2256         return service == RPC_GSS_SVC_INTEGRITY ||
2257                service == RPC_GSS_SVC_PRIVACY;
2258 }
2259
2260 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2261 {
2262         struct svc_cred *cr = &rqstp->rq_cred;
2263
2264         if (!cl->cl_mach_cred)
2265                 return true;
2266         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2267                 return false;
2268         if (!svc_rqst_integrity_protected(rqstp))
2269                 return false;
2270         if (cl->cl_cred.cr_raw_principal)
2271                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2272                                                 cr->cr_raw_principal);
2273         if (!cr->cr_principal)
2274                 return false;
2275         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2276 }
2277
2278 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2279 {
2280         __be32 verf[2];
2281
2282         /*
2283          * This is opaque to client, so no need to byte-swap. Use
2284          * __force to keep sparse happy
2285          */
2286         verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2287         verf[1] = (__force __be32)nn->clverifier_counter++;
2288         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2289 }
2290
2291 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2292 {
2293         clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2294         clp->cl_clientid.cl_id = nn->clientid_counter++;
2295         gen_confirm(clp, nn);
2296 }
2297
2298 static struct nfs4_stid *
2299 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2300 {
2301         struct nfs4_stid *ret;
2302
2303         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2304         if (!ret || !ret->sc_type)
2305                 return NULL;
2306         return ret;
2307 }
2308
2309 static struct nfs4_stid *
2310 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2311 {
2312         struct nfs4_stid *s;
2313
2314         spin_lock(&cl->cl_lock);
2315         s = find_stateid_locked(cl, t);
2316         if (s != NULL) {
2317                 if (typemask & s->sc_type)
2318                         refcount_inc(&s->sc_count);
2319                 else
2320                         s = NULL;
2321         }
2322         spin_unlock(&cl->cl_lock);
2323         return s;
2324 }
2325
2326 static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2327 {
2328         struct nfsdfs_client *nc;
2329         nc = get_nfsdfs_client(inode);
2330         if (!nc)
2331                 return NULL;
2332         return container_of(nc, struct nfs4_client, cl_nfsdfs);
2333 }
2334
2335 static void seq_quote_mem(struct seq_file *m, char *data, int len)
2336 {
2337         seq_printf(m, "\"");
2338         seq_escape_mem_ascii(m, data, len);
2339         seq_printf(m, "\"");
2340 }
2341
2342 static int client_info_show(struct seq_file *m, void *v)
2343 {
2344         struct inode *inode = m->private;
2345         struct nfs4_client *clp;
2346         u64 clid;
2347
2348         clp = get_nfsdfs_clp(inode);
2349         if (!clp)
2350                 return -ENXIO;
2351         memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2352         seq_printf(m, "clientid: 0x%llx\n", clid);
2353         seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2354         seq_printf(m, "name: ");
2355         seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2356         seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2357         if (clp->cl_nii_domain.data) {
2358                 seq_printf(m, "Implementation domain: ");
2359                 seq_quote_mem(m, clp->cl_nii_domain.data,
2360                                         clp->cl_nii_domain.len);
2361                 seq_printf(m, "\nImplementation name: ");
2362                 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2363                 seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2364                         clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2365         }
2366         drop_client(clp);
2367
2368         return 0;
2369 }
2370
2371 static int client_info_open(struct inode *inode, struct file *file)
2372 {
2373         return single_open(file, client_info_show, inode);
2374 }
2375
2376 static const struct file_operations client_info_fops = {
2377         .open           = client_info_open,
2378         .read           = seq_read,
2379         .llseek         = seq_lseek,
2380         .release        = single_release,
2381 };
2382
2383 static void *states_start(struct seq_file *s, loff_t *pos)
2384         __acquires(&clp->cl_lock)
2385 {
2386         struct nfs4_client *clp = s->private;
2387         unsigned long id = *pos;
2388         void *ret;
2389
2390         spin_lock(&clp->cl_lock);
2391         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2392         *pos = id;
2393         return ret;
2394 }
2395
2396 static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2397 {
2398         struct nfs4_client *clp = s->private;
2399         unsigned long id = *pos;
2400         void *ret;
2401
2402         id = *pos;
2403         id++;
2404         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2405         *pos = id;
2406         return ret;
2407 }
2408
2409 static void states_stop(struct seq_file *s, void *v)
2410         __releases(&clp->cl_lock)
2411 {
2412         struct nfs4_client *clp = s->private;
2413
2414         spin_unlock(&clp->cl_lock);
2415 }
2416
2417 static void nfs4_show_fname(struct seq_file *s, struct nfsd_file *f)
2418 {
2419          seq_printf(s, "filename: \"%pD2\"", f->nf_file);
2420 }
2421
2422 static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2423 {
2424         struct inode *inode = f->nf_inode;
2425
2426         seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2427                                         MAJOR(inode->i_sb->s_dev),
2428                                          MINOR(inode->i_sb->s_dev),
2429                                          inode->i_ino);
2430 }
2431
2432 static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2433 {
2434         seq_printf(s, "owner: ");
2435         seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2436 }
2437
2438 static void nfs4_show_stateid(struct seq_file *s, stateid_t *stid)
2439 {
2440         seq_printf(s, "0x%.8x", stid->si_generation);
2441         seq_printf(s, "%12phN", &stid->si_opaque);
2442 }
2443
2444 static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2445 {
2446         struct nfs4_ol_stateid *ols;
2447         struct nfs4_file *nf;
2448         struct nfsd_file *file;
2449         struct nfs4_stateowner *oo;
2450         unsigned int access, deny;
2451
2452         if (st->sc_type != NFS4_OPEN_STID && st->sc_type != NFS4_LOCK_STID)
2453                 return 0; /* XXX: or SEQ_SKIP? */
2454         ols = openlockstateid(st);
2455         oo = ols->st_stateowner;
2456         nf = st->sc_file;
2457         file = find_any_file(nf);
2458         if (!file)
2459                 return 0;
2460
2461         seq_printf(s, "- ");
2462         nfs4_show_stateid(s, &st->sc_stateid);
2463         seq_printf(s, ": { type: open, ");
2464
2465         access = bmap_to_share_mode(ols->st_access_bmap);
2466         deny   = bmap_to_share_mode(ols->st_deny_bmap);
2467
2468         seq_printf(s, "access: %s%s, ",
2469                 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2470                 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2471         seq_printf(s, "deny: %s%s, ",
2472                 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2473                 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2474
2475         nfs4_show_superblock(s, file);
2476         seq_printf(s, ", ");
2477         nfs4_show_fname(s, file);
2478         seq_printf(s, ", ");
2479         nfs4_show_owner(s, oo);
2480         seq_printf(s, " }\n");
2481         nfsd_file_put(file);
2482
2483         return 0;
2484 }
2485
2486 static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2487 {
2488         struct nfs4_ol_stateid *ols;
2489         struct nfs4_file *nf;
2490         struct nfsd_file *file;
2491         struct nfs4_stateowner *oo;
2492
2493         ols = openlockstateid(st);
2494         oo = ols->st_stateowner;
2495         nf = st->sc_file;
2496         file = find_any_file(nf);
2497         if (!file)
2498                 return 0;
2499
2500         seq_printf(s, "- ");
2501         nfs4_show_stateid(s, &st->sc_stateid);
2502         seq_printf(s, ": { type: lock, ");
2503
2504         /*
2505          * Note: a lock stateid isn't really the same thing as a lock,
2506          * it's the locking state held by one owner on a file, and there
2507          * may be multiple (or no) lock ranges associated with it.
2508          * (Same for the matter is true of open stateids.)
2509          */
2510
2511         nfs4_show_superblock(s, file);
2512         /* XXX: open stateid? */
2513         seq_printf(s, ", ");
2514         nfs4_show_fname(s, file);
2515         seq_printf(s, ", ");
2516         nfs4_show_owner(s, oo);
2517         seq_printf(s, " }\n");
2518         nfsd_file_put(file);
2519
2520         return 0;
2521 }
2522
2523 static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2524 {
2525         struct nfs4_delegation *ds;
2526         struct nfs4_file *nf;
2527         struct nfsd_file *file;
2528
2529         ds = delegstateid(st);
2530         nf = st->sc_file;
2531         file = find_deleg_file(nf);
2532         if (!file)
2533                 return 0;
2534
2535         seq_printf(s, "- ");
2536         nfs4_show_stateid(s, &st->sc_stateid);
2537         seq_printf(s, ": { type: deleg, ");
2538
2539         /* Kinda dead code as long as we only support read delegs: */
2540         seq_printf(s, "access: %s, ",
2541                 ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2542
2543         /* XXX: lease time, whether it's being recalled. */
2544
2545         nfs4_show_superblock(s, file);
2546         seq_printf(s, ", ");
2547         nfs4_show_fname(s, file);
2548         seq_printf(s, " }\n");
2549         nfsd_file_put(file);
2550
2551         return 0;
2552 }
2553
2554 static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2555 {
2556         struct nfs4_layout_stateid *ls;
2557         struct nfsd_file *file;
2558
2559         ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2560         file = ls->ls_file;
2561
2562         seq_printf(s, "- ");
2563         nfs4_show_stateid(s, &st->sc_stateid);
2564         seq_printf(s, ": { type: layout, ");
2565
2566         /* XXX: What else would be useful? */
2567
2568         nfs4_show_superblock(s, file);
2569         seq_printf(s, ", ");
2570         nfs4_show_fname(s, file);
2571         seq_printf(s, " }\n");
2572
2573         return 0;
2574 }
2575
2576 static int states_show(struct seq_file *s, void *v)
2577 {
2578         struct nfs4_stid *st = v;
2579
2580         switch (st->sc_type) {
2581         case NFS4_OPEN_STID:
2582                 return nfs4_show_open(s, st);
2583         case NFS4_LOCK_STID:
2584                 return nfs4_show_lock(s, st);
2585         case NFS4_DELEG_STID:
2586                 return nfs4_show_deleg(s, st);
2587         case NFS4_LAYOUT_STID:
2588                 return nfs4_show_layout(s, st);
2589         default:
2590                 return 0; /* XXX: or SEQ_SKIP? */
2591         }
2592         /* XXX: copy stateids? */
2593 }
2594
2595 static struct seq_operations states_seq_ops = {
2596         .start = states_start,
2597         .next = states_next,
2598         .stop = states_stop,
2599         .show = states_show
2600 };
2601
2602 static int client_states_open(struct inode *inode, struct file *file)
2603 {
2604         struct seq_file *s;
2605         struct nfs4_client *clp;
2606         int ret;
2607
2608         clp = get_nfsdfs_clp(inode);
2609         if (!clp)
2610                 return -ENXIO;
2611
2612         ret = seq_open(file, &states_seq_ops);
2613         if (ret)
2614                 return ret;
2615         s = file->private_data;
2616         s->private = clp;
2617         return 0;
2618 }
2619
2620 static int client_opens_release(struct inode *inode, struct file *file)
2621 {
2622         struct seq_file *m = file->private_data;
2623         struct nfs4_client *clp = m->private;
2624
2625         /* XXX: alternatively, we could get/drop in seq start/stop */
2626         drop_client(clp);
2627         return 0;
2628 }
2629
2630 static const struct file_operations client_states_fops = {
2631         .open           = client_states_open,
2632         .read           = seq_read,
2633         .llseek         = seq_lseek,
2634         .release        = client_opens_release,
2635 };
2636
2637 /*
2638  * Normally we refuse to destroy clients that are in use, but here the
2639  * administrator is telling us to just do it.  We also want to wait
2640  * so the caller has a guarantee that the client's locks are gone by
2641  * the time the write returns:
2642  */
2643 static void force_expire_client(struct nfs4_client *clp)
2644 {
2645         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2646         bool already_expired;
2647
2648         spin_lock(&clp->cl_lock);
2649         clp->cl_time = 0;
2650         spin_unlock(&clp->cl_lock);
2651
2652         wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2653         spin_lock(&nn->client_lock);
2654         already_expired = list_empty(&clp->cl_lru);
2655         if (!already_expired)
2656                 unhash_client_locked(clp);
2657         spin_unlock(&nn->client_lock);
2658
2659         if (!already_expired)
2660                 expire_client(clp);
2661         else
2662                 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2663 }
2664
2665 static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2666                                    size_t size, loff_t *pos)
2667 {
2668         char *data;
2669         struct nfs4_client *clp;
2670
2671         data = simple_transaction_get(file, buf, size);
2672         if (IS_ERR(data))
2673                 return PTR_ERR(data);
2674         if (size != 7 || 0 != memcmp(data, "expire\n", 7))
2675                 return -EINVAL;
2676         clp = get_nfsdfs_clp(file_inode(file));
2677         if (!clp)
2678                 return -ENXIO;
2679         force_expire_client(clp);
2680         drop_client(clp);
2681         return 7;
2682 }
2683
2684 static const struct file_operations client_ctl_fops = {
2685         .write          = client_ctl_write,
2686         .release        = simple_transaction_release,
2687 };
2688
2689 static const struct tree_descr client_files[] = {
2690         [0] = {"info", &client_info_fops, S_IRUSR},
2691         [1] = {"states", &client_states_fops, S_IRUSR},
2692         [2] = {"ctl", &client_ctl_fops, S_IWUSR},
2693         [3] = {""},
2694 };
2695
2696 static struct nfs4_client *create_client(struct xdr_netobj name,
2697                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2698 {
2699         struct nfs4_client *clp;
2700         struct sockaddr *sa = svc_addr(rqstp);
2701         int ret;
2702         struct net *net = SVC_NET(rqstp);
2703         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2704
2705         clp = alloc_client(name);
2706         if (clp == NULL)
2707                 return NULL;
2708
2709         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2710         if (ret) {
2711                 free_client(clp);
2712                 return NULL;
2713         }
2714         gen_clid(clp, nn);
2715         kref_init(&clp->cl_nfsdfs.cl_ref);
2716         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2717         clp->cl_time = ktime_get_boottime_seconds();
2718         clear_bit(0, &clp->cl_cb_slot_busy);
2719         copy_verf(clp, verf);
2720         memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
2721         clp->cl_cb_session = NULL;
2722         clp->net = net;
2723         clp->cl_nfsd_dentry = nfsd_client_mkdir(nn, &clp->cl_nfsdfs,
2724                         clp->cl_clientid.cl_id - nn->clientid_base,
2725                         client_files);
2726         if (!clp->cl_nfsd_dentry) {
2727                 free_client(clp);
2728                 return NULL;
2729         }
2730         return clp;
2731 }
2732
2733 static void
2734 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2735 {
2736         struct rb_node **new = &(root->rb_node), *parent = NULL;
2737         struct nfs4_client *clp;
2738
2739         while (*new) {
2740                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2741                 parent = *new;
2742
2743                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2744                         new = &((*new)->rb_left);
2745                 else
2746                         new = &((*new)->rb_right);
2747         }
2748
2749         rb_link_node(&new_clp->cl_namenode, parent, new);
2750         rb_insert_color(&new_clp->cl_namenode, root);
2751 }
2752
2753 static struct nfs4_client *
2754 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2755 {
2756         int cmp;
2757         struct rb_node *node = root->rb_node;
2758         struct nfs4_client *clp;
2759
2760         while (node) {
2761                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2762                 cmp = compare_blob(&clp->cl_name, name);
2763                 if (cmp > 0)
2764                         node = node->rb_left;
2765                 else if (cmp < 0)
2766                         node = node->rb_right;
2767                 else
2768                         return clp;
2769         }
2770         return NULL;
2771 }
2772
2773 static void
2774 add_to_unconfirmed(struct nfs4_client *clp)
2775 {
2776         unsigned int idhashval;
2777         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2778
2779         lockdep_assert_held(&nn->client_lock);
2780
2781         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2782         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2783         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2784         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2785         renew_client_locked(clp);
2786 }
2787
2788 static void
2789 move_to_confirmed(struct nfs4_client *clp)
2790 {
2791         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2792         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2793
2794         lockdep_assert_held(&nn->client_lock);
2795
2796         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2797         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2798         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2799         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2800         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2801         renew_client_locked(clp);
2802 }
2803
2804 static struct nfs4_client *
2805 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2806 {
2807         struct nfs4_client *clp;
2808         unsigned int idhashval = clientid_hashval(clid->cl_id);
2809
2810         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2811                 if (same_clid(&clp->cl_clientid, clid)) {
2812                         if ((bool)clp->cl_minorversion != sessions)
2813                                 return NULL;
2814                         renew_client_locked(clp);
2815                         return clp;
2816                 }
2817         }
2818         return NULL;
2819 }
2820
2821 static struct nfs4_client *
2822 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2823 {
2824         struct list_head *tbl = nn->conf_id_hashtbl;
2825
2826         lockdep_assert_held(&nn->client_lock);
2827         return find_client_in_id_table(tbl, clid, sessions);
2828 }
2829
2830 static struct nfs4_client *
2831 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2832 {
2833         struct list_head *tbl = nn->unconf_id_hashtbl;
2834
2835         lockdep_assert_held(&nn->client_lock);
2836         return find_client_in_id_table(tbl, clid, sessions);
2837 }
2838
2839 static bool clp_used_exchangeid(struct nfs4_client *clp)
2840 {
2841         return clp->cl_exchange_flags != 0;
2842 }
2843
2844 static struct nfs4_client *
2845 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2846 {
2847         lockdep_assert_held(&nn->client_lock);
2848         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2849 }
2850
2851 static struct nfs4_client *
2852 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2853 {
2854         lockdep_assert_held(&nn->client_lock);
2855         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2856 }
2857
2858 static void
2859 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2860 {
2861         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2862         struct sockaddr *sa = svc_addr(rqstp);
2863         u32 scopeid = rpc_get_scope_id(sa);
2864         unsigned short expected_family;
2865
2866         /* Currently, we only support tcp and tcp6 for the callback channel */
2867         if (se->se_callback_netid_len == 3 &&
2868             !memcmp(se->se_callback_netid_val, "tcp", 3))
2869                 expected_family = AF_INET;
2870         else if (se->se_callback_netid_len == 4 &&
2871                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2872                 expected_family = AF_INET6;
2873         else
2874                 goto out_err;
2875
2876         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2877                                             se->se_callback_addr_len,
2878                                             (struct sockaddr *)&conn->cb_addr,
2879                                             sizeof(conn->cb_addr));
2880
2881         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2882                 goto out_err;
2883
2884         if (conn->cb_addr.ss_family == AF_INET6)
2885                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2886
2887         conn->cb_prog = se->se_callback_prog;
2888         conn->cb_ident = se->se_callback_ident;
2889         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2890         trace_nfsd_cb_args(clp, conn);
2891         return;
2892 out_err:
2893         conn->cb_addr.ss_family = AF_UNSPEC;
2894         conn->cb_addrlen = 0;
2895         trace_nfsd_cb_nodelegs(clp);
2896         return;
2897 }
2898
2899 /*
2900  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2901  */
2902 static void
2903 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2904 {
2905         struct xdr_buf *buf = resp->xdr.buf;
2906         struct nfsd4_slot *slot = resp->cstate.slot;
2907         unsigned int base;
2908
2909         dprintk("--> %s slot %p\n", __func__, slot);
2910
2911         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2912         slot->sl_opcnt = resp->opcnt;
2913         slot->sl_status = resp->cstate.status;
2914         free_svc_cred(&slot->sl_cred);
2915         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2916
2917         if (!nfsd4_cache_this(resp)) {
2918                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2919                 return;
2920         }
2921         slot->sl_flags |= NFSD4_SLOT_CACHED;
2922
2923         base = resp->cstate.data_offset;
2924         slot->sl_datalen = buf->len - base;
2925         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2926                 WARN(1, "%s: sessions DRC could not cache compound\n",
2927                      __func__);
2928         return;
2929 }
2930
2931 /*
2932  * Encode the replay sequence operation from the slot values.
2933  * If cachethis is FALSE encode the uncached rep error on the next
2934  * operation which sets resp->p and increments resp->opcnt for
2935  * nfs4svc_encode_compoundres.
2936  *
2937  */
2938 static __be32
2939 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2940                           struct nfsd4_compoundres *resp)
2941 {
2942         struct nfsd4_op *op;
2943         struct nfsd4_slot *slot = resp->cstate.slot;
2944
2945         /* Encode the replayed sequence operation */
2946         op = &args->ops[resp->opcnt - 1];
2947         nfsd4_encode_operation(resp, op);
2948
2949         if (slot->sl_flags & NFSD4_SLOT_CACHED)
2950                 return op->status;
2951         if (args->opcnt == 1) {
2952                 /*
2953                  * The original operation wasn't a solo sequence--we
2954                  * always cache those--so this retry must not match the
2955                  * original:
2956                  */
2957                 op->status = nfserr_seq_false_retry;
2958         } else {
2959                 op = &args->ops[resp->opcnt++];
2960                 op->status = nfserr_retry_uncached_rep;
2961                 nfsd4_encode_operation(resp, op);
2962         }
2963         return op->status;
2964 }
2965
2966 /*
2967  * The sequence operation is not cached because we can use the slot and
2968  * session values.
2969  */
2970 static __be32
2971 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2972                          struct nfsd4_sequence *seq)
2973 {
2974         struct nfsd4_slot *slot = resp->cstate.slot;
2975         struct xdr_stream *xdr = &resp->xdr;
2976         __be32 *p;
2977         __be32 status;
2978
2979         dprintk("--> %s slot %p\n", __func__, slot);
2980
2981         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2982         if (status)
2983                 return status;
2984
2985         p = xdr_reserve_space(xdr, slot->sl_datalen);
2986         if (!p) {
2987                 WARN_ON_ONCE(1);
2988                 return nfserr_serverfault;
2989         }
2990         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2991         xdr_commit_encode(xdr);
2992
2993         resp->opcnt = slot->sl_opcnt;
2994         return slot->sl_status;
2995 }
2996
2997 /*
2998  * Set the exchange_id flags returned by the server.
2999  */
3000 static void
3001 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
3002 {
3003 #ifdef CONFIG_NFSD_PNFS
3004         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
3005 #else
3006         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
3007 #endif
3008
3009         /* Referrals are supported, Migration is not. */
3010         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
3011
3012         /* set the wire flags to return to client. */
3013         clid->flags = new->cl_exchange_flags;
3014 }
3015
3016 static bool client_has_openowners(struct nfs4_client *clp)
3017 {
3018         struct nfs4_openowner *oo;
3019
3020         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
3021                 if (!list_empty(&oo->oo_owner.so_stateids))
3022                         return true;
3023         }
3024         return false;
3025 }
3026
3027 static bool client_has_state(struct nfs4_client *clp)
3028 {
3029         return client_has_openowners(clp)
3030 #ifdef CONFIG_NFSD_PNFS
3031                 || !list_empty(&clp->cl_lo_states)
3032 #endif
3033                 || !list_empty(&clp->cl_delegations)
3034                 || !list_empty(&clp->cl_sessions)
3035                 || !list_empty(&clp->async_copies);
3036 }
3037
3038 static __be32 copy_impl_id(struct nfs4_client *clp,
3039                                 struct nfsd4_exchange_id *exid)
3040 {
3041         if (!exid->nii_domain.data)
3042                 return 0;
3043         xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
3044         if (!clp->cl_nii_domain.data)
3045                 return nfserr_jukebox;
3046         xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
3047         if (!clp->cl_nii_name.data)
3048                 return nfserr_jukebox;
3049         clp->cl_nii_time = exid->nii_time;
3050         return 0;
3051 }
3052
3053 __be32
3054 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3055                 union nfsd4_op_u *u)
3056 {
3057         struct nfsd4_exchange_id *exid = &u->exchange_id;
3058         struct nfs4_client *conf, *new;
3059         struct nfs4_client *unconf = NULL;
3060         __be32 status;
3061         char                    addr_str[INET6_ADDRSTRLEN];
3062         nfs4_verifier           verf = exid->verifier;
3063         struct sockaddr         *sa = svc_addr(rqstp);
3064         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3065         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3066
3067         rpc_ntop(sa, addr_str, sizeof(addr_str));
3068         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3069                 "ip_addr=%s flags %x, spa_how %d\n",
3070                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
3071                 addr_str, exid->flags, exid->spa_how);
3072
3073         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3074                 return nfserr_inval;
3075
3076         new = create_client(exid->clname, rqstp, &verf);
3077         if (new == NULL)
3078                 return nfserr_jukebox;
3079         status = copy_impl_id(new, exid);
3080         if (status)
3081                 goto out_nolock;
3082
3083         switch (exid->spa_how) {
3084         case SP4_MACH_CRED:
3085                 exid->spo_must_enforce[0] = 0;
3086                 exid->spo_must_enforce[1] = (
3087                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
3088                         1 << (OP_EXCHANGE_ID - 32) |
3089                         1 << (OP_CREATE_SESSION - 32) |
3090                         1 << (OP_DESTROY_SESSION - 32) |
3091                         1 << (OP_DESTROY_CLIENTID - 32));
3092
3093                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3094                                         1 << (OP_OPEN_DOWNGRADE) |
3095                                         1 << (OP_LOCKU) |
3096                                         1 << (OP_DELEGRETURN));
3097
3098                 exid->spo_must_allow[1] &= (
3099                                         1 << (OP_TEST_STATEID - 32) |
3100                                         1 << (OP_FREE_STATEID - 32));
3101                 if (!svc_rqst_integrity_protected(rqstp)) {
3102                         status = nfserr_inval;
3103                         goto out_nolock;
3104                 }
3105                 /*
3106                  * Sometimes userspace doesn't give us a principal.
3107                  * Which is a bug, really.  Anyway, we can't enforce
3108                  * MACH_CRED in that case, better to give up now:
3109                  */
3110                 if (!new->cl_cred.cr_principal &&
3111                                         !new->cl_cred.cr_raw_principal) {
3112                         status = nfserr_serverfault;
3113                         goto out_nolock;
3114                 }
3115                 new->cl_mach_cred = true;
3116         case SP4_NONE:
3117                 break;
3118         default:                                /* checked by xdr code */
3119                 WARN_ON_ONCE(1);
3120                 fallthrough;
3121         case SP4_SSV:
3122                 status = nfserr_encr_alg_unsupp;
3123                 goto out_nolock;
3124         }
3125
3126         /* Cases below refer to rfc 5661 section 18.35.4: */
3127         spin_lock(&nn->client_lock);
3128         conf = find_confirmed_client_by_name(&exid->clname, nn);
3129         if (conf) {
3130                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3131                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3132
3133                 if (update) {
3134                         if (!clp_used_exchangeid(conf)) { /* buggy client */
3135                                 status = nfserr_inval;
3136                                 goto out;
3137                         }
3138                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
3139                                 status = nfserr_wrong_cred;
3140                                 goto out;
3141                         }
3142                         if (!creds_match) { /* case 9 */
3143                                 status = nfserr_perm;
3144                                 goto out;
3145                         }
3146                         if (!verfs_match) { /* case 8 */
3147                                 status = nfserr_not_same;
3148                                 goto out;
3149                         }
3150                         /* case 6 */
3151                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3152                         goto out_copy;
3153                 }
3154                 if (!creds_match) { /* case 3 */
3155                         if (client_has_state(conf)) {
3156                                 status = nfserr_clid_inuse;
3157                                 goto out;
3158                         }
3159                         goto out_new;
3160                 }
3161                 if (verfs_match) { /* case 2 */
3162                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3163                         goto out_copy;
3164                 }
3165                 /* case 5, client reboot */
3166                 conf = NULL;
3167                 goto out_new;
3168         }
3169
3170         if (update) { /* case 7 */
3171                 status = nfserr_noent;
3172                 goto out;
3173         }
3174
3175         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
3176         if (unconf) /* case 4, possible retry or client restart */
3177                 unhash_client_locked(unconf);
3178
3179         /* case 1 (normal case) */
3180 out_new:
3181         if (conf) {
3182                 status = mark_client_expired_locked(conf);
3183                 if (status)
3184                         goto out;
3185         }
3186         new->cl_minorversion = cstate->minorversion;
3187         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3188         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3189
3190         add_to_unconfirmed(new);
3191         swap(new, conf);
3192 out_copy:
3193         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3194         exid->clientid.cl_id = conf->cl_clientid.cl_id;
3195
3196         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3197         nfsd4_set_ex_flags(conf, exid);
3198
3199         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3200                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3201         status = nfs_ok;
3202
3203 out:
3204         spin_unlock(&nn->client_lock);
3205 out_nolock:
3206         if (new)
3207                 expire_client(new);
3208         if (unconf)
3209                 expire_client(unconf);
3210         return status;
3211 }
3212
3213 static __be32
3214 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3215 {
3216         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3217                 slot_seqid);
3218
3219         /* The slot is in use, and no response has been sent. */
3220         if (slot_inuse) {
3221                 if (seqid == slot_seqid)
3222                         return nfserr_jukebox;
3223                 else
3224                         return nfserr_seq_misordered;
3225         }
3226         /* Note unsigned 32-bit arithmetic handles wraparound: */
3227         if (likely(seqid == slot_seqid + 1))
3228                 return nfs_ok;
3229         if (seqid == slot_seqid)
3230                 return nfserr_replay_cache;
3231         return nfserr_seq_misordered;
3232 }
3233
3234 /*
3235  * Cache the create session result into the create session single DRC
3236  * slot cache by saving the xdr structure. sl_seqid has been set.
3237  * Do this for solo or embedded create session operations.
3238  */
3239 static void
3240 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3241                            struct nfsd4_clid_slot *slot, __be32 nfserr)
3242 {
3243         slot->sl_status = nfserr;
3244         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3245 }
3246
3247 static __be32
3248 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3249                             struct nfsd4_clid_slot *slot)
3250 {
3251         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3252         return slot->sl_status;
3253 }
3254
3255 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3256                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3257                         1 +     /* MIN tag is length with zero, only length */ \
3258                         3 +     /* version, opcount, opcode */ \
3259                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3260                                 /* seqid, slotID, slotID, cache */ \
3261                         4 ) * sizeof(__be32))
3262
3263 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3264                         2 +     /* verifier: AUTH_NULL, length 0 */\
3265                         1 +     /* status */ \
3266                         1 +     /* MIN tag is length with zero, only length */ \
3267                         3 +     /* opcount, opcode, opstatus*/ \
3268                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3269                                 /* seqid, slotID, slotID, slotID, status */ \
3270                         5 ) * sizeof(__be32))
3271
3272 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3273 {
3274         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3275
3276         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3277                 return nfserr_toosmall;
3278         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3279                 return nfserr_toosmall;
3280         ca->headerpadsz = 0;
3281         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3282         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3283         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3284         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3285                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3286         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3287         /*
3288          * Note decreasing slot size below client's request may make it
3289          * difficult for client to function correctly, whereas
3290          * decreasing the number of slots will (just?) affect
3291          * performance.  When short on memory we therefore prefer to
3292          * decrease number of slots instead of their size.  Clients that
3293          * request larger slots than they need will get poor results:
3294          * Note that we always allow at least one slot, because our
3295          * accounting is soft and provides no guarantees either way.
3296          */
3297         ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3298
3299         return nfs_ok;
3300 }
3301
3302 /*
3303  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3304  * These are based on similar macros in linux/sunrpc/msg_prot.h .
3305  */
3306 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3307         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3308
3309 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3310         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3311
3312 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
3313                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3314 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
3315                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3316                                  sizeof(__be32))
3317
3318 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3319 {
3320         ca->headerpadsz = 0;
3321
3322         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3323                 return nfserr_toosmall;
3324         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3325                 return nfserr_toosmall;
3326         ca->maxresp_cached = 0;
3327         if (ca->maxops < 2)
3328                 return nfserr_toosmall;
3329
3330         return nfs_ok;
3331 }
3332
3333 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3334 {
3335         switch (cbs->flavor) {
3336         case RPC_AUTH_NULL:
3337         case RPC_AUTH_UNIX:
3338                 return nfs_ok;
3339         default:
3340                 /*
3341                  * GSS case: the spec doesn't allow us to return this
3342                  * error.  But it also doesn't allow us not to support
3343                  * GSS.
3344                  * I'd rather this fail hard than return some error the
3345                  * client might think it can already handle:
3346                  */
3347                 return nfserr_encr_alg_unsupp;
3348         }
3349 }
3350
3351 __be32
3352 nfsd4_create_session(struct svc_rqst *rqstp,
3353                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3354 {
3355         struct nfsd4_create_session *cr_ses = &u->create_session;
3356         struct sockaddr *sa = svc_addr(rqstp);
3357         struct nfs4_client *conf, *unconf;
3358         struct nfs4_client *old = NULL;
3359         struct nfsd4_session *new;
3360         struct nfsd4_conn *conn;
3361         struct nfsd4_clid_slot *cs_slot = NULL;
3362         __be32 status = 0;
3363         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3364
3365         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3366                 return nfserr_inval;
3367         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3368         if (status)
3369                 return status;
3370         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3371         if (status)
3372                 return status;
3373         status = check_backchannel_attrs(&cr_ses->back_channel);
3374         if (status)
3375                 goto out_release_drc_mem;
3376         status = nfserr_jukebox;
3377         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3378         if (!new)
3379                 goto out_release_drc_mem;
3380         conn = alloc_conn_from_crses(rqstp, cr_ses);
3381         if (!conn)
3382                 goto out_free_session;
3383
3384         spin_lock(&nn->client_lock);
3385         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3386         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3387         WARN_ON_ONCE(conf && unconf);
3388
3389         if (conf) {
3390                 status = nfserr_wrong_cred;
3391                 if (!nfsd4_mach_creds_match(conf, rqstp))
3392                         goto out_free_conn;
3393                 cs_slot = &conf->cl_cs_slot;
3394                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3395                 if (status) {
3396                         if (status == nfserr_replay_cache)
3397                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3398                         goto out_free_conn;
3399                 }
3400         } else if (unconf) {
3401                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3402                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3403                         status = nfserr_clid_inuse;
3404                         goto out_free_conn;
3405                 }
3406                 status = nfserr_wrong_cred;
3407                 if (!nfsd4_mach_creds_match(unconf, rqstp))
3408                         goto out_free_conn;
3409                 cs_slot = &unconf->cl_cs_slot;
3410                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3411                 if (status) {
3412                         /* an unconfirmed replay returns misordered */
3413                         status = nfserr_seq_misordered;
3414                         goto out_free_conn;
3415                 }
3416                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3417                 if (old) {
3418                         status = mark_client_expired_locked(old);
3419                         if (status) {
3420                                 old = NULL;
3421                                 goto out_free_conn;
3422                         }
3423                 }
3424                 move_to_confirmed(unconf);
3425                 conf = unconf;
3426         } else {
3427                 status = nfserr_stale_clientid;
3428                 goto out_free_conn;
3429         }
3430         status = nfs_ok;
3431         /* Persistent sessions are not supported */
3432         cr_ses->flags &= ~SESSION4_PERSIST;
3433         /* Upshifting from TCP to RDMA is not supported */
3434         cr_ses->flags &= ~SESSION4_RDMA;
3435
3436         init_session(rqstp, new, conf, cr_ses);
3437         nfsd4_get_session_locked(new);
3438
3439         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3440                NFS4_MAX_SESSIONID_LEN);
3441         cs_slot->sl_seqid++;
3442         cr_ses->seqid = cs_slot->sl_seqid;
3443
3444         /* cache solo and embedded create sessions under the client_lock */
3445         nfsd4_cache_create_session(cr_ses, cs_slot, status);
3446         spin_unlock(&nn->client_lock);
3447         /* init connection and backchannel */
3448         nfsd4_init_conn(rqstp, conn, new);
3449         nfsd4_put_session(new);
3450         if (old)
3451                 expire_client(old);
3452         return status;
3453 out_free_conn:
3454         spin_unlock(&nn->client_lock);
3455         free_conn(conn);
3456         if (old)
3457                 expire_client(old);
3458 out_free_session:
3459         __free_session(new);
3460 out_release_drc_mem:
3461         nfsd4_put_drc_mem(&cr_ses->fore_channel);
3462         return status;
3463 }
3464
3465 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3466 {
3467         switch (*dir) {
3468         case NFS4_CDFC4_FORE:
3469         case NFS4_CDFC4_BACK:
3470                 return nfs_ok;
3471         case NFS4_CDFC4_FORE_OR_BOTH:
3472         case NFS4_CDFC4_BACK_OR_BOTH:
3473                 *dir = NFS4_CDFC4_BOTH;
3474                 return nfs_ok;
3475         }
3476         return nfserr_inval;
3477 }
3478
3479 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3480                 struct nfsd4_compound_state *cstate,
3481                 union nfsd4_op_u *u)
3482 {
3483         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3484         struct nfsd4_session *session = cstate->session;
3485         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3486         __be32 status;
3487
3488         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3489         if (status)
3490                 return status;
3491         spin_lock(&nn->client_lock);
3492         session->se_cb_prog = bc->bc_cb_program;
3493         session->se_cb_sec = bc->bc_cb_sec;
3494         spin_unlock(&nn->client_lock);
3495
3496         nfsd4_probe_callback(session->se_client);
3497
3498         return nfs_ok;
3499 }
3500
3501 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3502 {
3503         struct nfsd4_conn *c;
3504
3505         list_for_each_entry(c, &s->se_conns, cn_persession) {
3506                 if (c->cn_xprt == xpt) {
3507                         return c;
3508                 }
3509         }
3510         return NULL;
3511 }
3512
3513 static __be32 nfsd4_match_existing_connection(struct svc_rqst *rqst,
3514                                 struct nfsd4_session *session, u32 req)
3515 {
3516         struct nfs4_client *clp = session->se_client;
3517         struct svc_xprt *xpt = rqst->rq_xprt;
3518         struct nfsd4_conn *c;
3519         __be32 status;
3520
3521         /* Following the last paragraph of RFC 5661 Section 18.34.3: */
3522         spin_lock(&clp->cl_lock);
3523         c = __nfsd4_find_conn(xpt, session);
3524         if (!c)
3525                 status = nfserr_noent;
3526         else if (req == c->cn_flags)
3527                 status = nfs_ok;
3528         else if (req == NFS4_CDFC4_FORE_OR_BOTH &&
3529                                 c->cn_flags != NFS4_CDFC4_BACK)
3530                 status = nfs_ok;
3531         else if (req == NFS4_CDFC4_BACK_OR_BOTH &&
3532                                 c->cn_flags != NFS4_CDFC4_FORE)
3533                 status = nfs_ok;
3534         else
3535                 status = nfserr_inval;
3536         spin_unlock(&clp->cl_lock);
3537         return status;
3538 }
3539
3540 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
3541                      struct nfsd4_compound_state *cstate,
3542                      union nfsd4_op_u *u)
3543 {
3544         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
3545         __be32 status;
3546         struct nfsd4_conn *conn;
3547         struct nfsd4_session *session;
3548         struct net *net = SVC_NET(rqstp);
3549         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3550
3551         if (!nfsd4_last_compound_op(rqstp))
3552                 return nfserr_not_only_op;
3553         spin_lock(&nn->client_lock);
3554         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
3555         spin_unlock(&nn->client_lock);
3556         if (!session)
3557                 goto out_no_session;
3558         status = nfserr_wrong_cred;
3559         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
3560                 goto out;
3561         status = nfsd4_match_existing_connection(rqstp, session, bcts->dir);
3562         if (status == nfs_ok || status == nfserr_inval)
3563                 goto out;
3564         status = nfsd4_map_bcts_dir(&bcts->dir);
3565         if (status)
3566                 goto out;
3567         conn = alloc_conn(rqstp, bcts->dir);
3568         status = nfserr_jukebox;
3569         if (!conn)
3570                 goto out;
3571         nfsd4_init_conn(rqstp, conn, session);
3572         status = nfs_ok;
3573 out:
3574         nfsd4_put_session(session);
3575 out_no_session:
3576         return status;
3577 }
3578
3579 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3580 {
3581         if (!cstate->session)
3582                 return false;
3583         return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3584 }
3585
3586 __be32
3587 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3588                 union nfsd4_op_u *u)
3589 {
3590         struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3591         struct nfsd4_session *ses;
3592         __be32 status;
3593         int ref_held_by_me = 0;
3594         struct net *net = SVC_NET(r);
3595         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3596
3597         status = nfserr_not_only_op;
3598         if (nfsd4_compound_in_session(cstate, sessionid)) {
3599                 if (!nfsd4_last_compound_op(r))
3600                         goto out;
3601                 ref_held_by_me++;
3602         }
3603         dump_sessionid(__func__, sessionid);
3604         spin_lock(&nn->client_lock);
3605         ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3606         if (!ses)
3607                 goto out_client_lock;
3608         status = nfserr_wrong_cred;
3609         if (!nfsd4_mach_creds_match(ses->se_client, r))
3610                 goto out_put_session;
3611         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3612         if (status)
3613                 goto out_put_session;
3614         unhash_session(ses);
3615         spin_unlock(&nn->client_lock);
3616
3617         nfsd4_probe_callback_sync(ses->se_client);
3618
3619         spin_lock(&nn->client_lock);
3620         status = nfs_ok;
3621 out_put_session:
3622         nfsd4_put_session_locked(ses);
3623 out_client_lock:
3624         spin_unlock(&nn->client_lock);
3625 out:
3626         return status;
3627 }
3628
3629 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3630 {
3631         struct nfs4_client *clp = ses->se_client;
3632         struct nfsd4_conn *c;
3633         __be32 status = nfs_ok;
3634         int ret;
3635
3636         spin_lock(&clp->cl_lock);
3637         c = __nfsd4_find_conn(new->cn_xprt, ses);
3638         if (c)
3639                 goto out_free;
3640         status = nfserr_conn_not_bound_to_session;
3641         if (clp->cl_mach_cred)
3642                 goto out_free;
3643         __nfsd4_hash_conn(new, ses);
3644         spin_unlock(&clp->cl_lock);
3645         ret = nfsd4_register_conn(new);
3646         if (ret)
3647                 /* oops; xprt is already down: */
3648                 nfsd4_conn_lost(&new->cn_xpt_user);
3649         return nfs_ok;
3650 out_free:
3651         spin_unlock(&clp->cl_lock);
3652         free_conn(new);
3653         return status;
3654 }
3655
3656 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3657 {
3658         struct nfsd4_compoundargs *args = rqstp->rq_argp;
3659
3660         return args->opcnt > session->se_fchannel.maxops;
3661 }
3662
3663 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3664                                   struct nfsd4_session *session)
3665 {
3666         struct xdr_buf *xb = &rqstp->rq_arg;
3667
3668         return xb->len > session->se_fchannel.maxreq_sz;
3669 }
3670
3671 static bool replay_matches_cache(struct svc_rqst *rqstp,
3672                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3673 {
3674         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3675
3676         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3677             (bool)seq->cachethis)
3678                 return false;
3679         /*
3680          * If there's an error then the reply can have fewer ops than
3681          * the call.
3682          */
3683         if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
3684                 return false;
3685         /*
3686          * But if we cached a reply with *more* ops than the call you're
3687          * sending us now, then this new call is clearly not really a
3688          * replay of the old one:
3689          */
3690         if (slot->sl_opcnt > argp->opcnt)
3691                 return false;
3692         /* This is the only check explicitly called by spec: */
3693         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3694                 return false;
3695         /*
3696          * There may be more comparisons we could actually do, but the
3697          * spec doesn't require us to catch every case where the calls
3698          * don't match (that would require caching the call as well as
3699          * the reply), so we don't bother.
3700          */
3701         return true;
3702 }
3703
3704 __be32
3705 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3706                 union nfsd4_op_u *u)
3707 {
3708         struct nfsd4_sequence *seq = &u->sequence;
3709         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3710         struct xdr_stream *xdr = &resp->xdr;
3711         struct nfsd4_session *session;
3712         struct nfs4_client *clp;
3713         struct nfsd4_slot *slot;
3714         struct nfsd4_conn *conn;
3715         __be32 status;
3716         int buflen;
3717         struct net *net = SVC_NET(rqstp);
3718         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3719
3720         if (resp->opcnt != 1)
3721                 return nfserr_sequence_pos;
3722
3723         /*
3724          * Will be either used or freed by nfsd4_sequence_check_conn
3725          * below.
3726          */
3727         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3728         if (!conn)
3729                 return nfserr_jukebox;
3730
3731         spin_lock(&nn->client_lock);
3732         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3733         if (!session)
3734                 goto out_no_session;
3735         clp = session->se_client;
3736
3737         status = nfserr_too_many_ops;
3738         if (nfsd4_session_too_many_ops(rqstp, session))
3739                 goto out_put_session;
3740
3741         status = nfserr_req_too_big;
3742         if (nfsd4_request_too_big(rqstp, session))
3743                 goto out_put_session;
3744
3745         status = nfserr_badslot;
3746         if (seq->slotid >= session->se_fchannel.maxreqs)
3747                 goto out_put_session;
3748
3749         slot = session->se_slots[seq->slotid];
3750         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3751
3752         /* We do not negotiate the number of slots yet, so set the
3753          * maxslots to the session maxreqs which is used to encode
3754          * sr_highest_slotid and the sr_target_slot id to maxslots */
3755         seq->maxslots = session->se_fchannel.maxreqs;
3756
3757         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3758                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3759         if (status == nfserr_replay_cache) {
3760                 status = nfserr_seq_misordered;
3761                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3762                         goto out_put_session;
3763                 status = nfserr_seq_false_retry;
3764                 if (!replay_matches_cache(rqstp, seq, slot))
3765                         goto out_put_session;
3766                 cstate->slot = slot;
3767                 cstate->session = session;
3768                 cstate->clp = clp;
3769                 /* Return the cached reply status and set cstate->status
3770                  * for nfsd4_proc_compound processing */
3771                 status = nfsd4_replay_cache_entry(resp, seq);
3772                 cstate->status = nfserr_replay_cache;
3773                 goto out;
3774         }
3775         if (status)
3776                 goto out_put_session;
3777
3778         status = nfsd4_sequence_check_conn(conn, session);
3779         conn = NULL;
3780         if (status)
3781                 goto out_put_session;
3782
3783         buflen = (seq->cachethis) ?
3784                         session->se_fchannel.maxresp_cached :
3785                         session->se_fchannel.maxresp_sz;
3786         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3787                                     nfserr_rep_too_big;
3788         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3789                 goto out_put_session;
3790         svc_reserve(rqstp, buflen);
3791
3792         status = nfs_ok;
3793         /* Success! bump slot seqid */
3794         slot->sl_seqid = seq->seqid;
3795         slot->sl_flags |= NFSD4_SLOT_INUSE;
3796         if (seq->cachethis)
3797                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3798         else
3799                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3800
3801         cstate->slot = slot;
3802         cstate->session = session;
3803         cstate->clp = clp;
3804
3805 out:
3806         switch (clp->cl_cb_state) {
3807         case NFSD4_CB_DOWN:
3808                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3809                 break;
3810         case NFSD4_CB_FAULT:
3811                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3812                 break;
3813         default:
3814                 seq->status_flags = 0;
3815         }
3816         if (!list_empty(&clp->cl_revoked))
3817                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3818 out_no_session:
3819         if (conn)
3820                 free_conn(conn);
3821         spin_unlock(&nn->client_lock);
3822         return status;
3823 out_put_session:
3824         nfsd4_put_session_locked(session);
3825         goto out_no_session;
3826 }
3827
3828 void
3829 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3830 {
3831         struct nfsd4_compound_state *cs = &resp->cstate;
3832
3833         if (nfsd4_has_session(cs)) {
3834                 if (cs->status != nfserr_replay_cache) {
3835                         nfsd4_store_cache_entry(resp);
3836                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3837                 }
3838                 /* Drop session reference that was taken in nfsd4_sequence() */
3839                 nfsd4_put_session(cs->session);
3840         } else if (cs->clp)
3841                 put_client_renew(cs->clp);
3842 }
3843
3844 __be32
3845 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3846                 struct nfsd4_compound_state *cstate,
3847                 union nfsd4_op_u *u)
3848 {
3849         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3850         struct nfs4_client *conf, *unconf;
3851         struct nfs4_client *clp = NULL;
3852         __be32 status = 0;
3853         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3854
3855         spin_lock(&nn->client_lock);
3856         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3857         conf = find_confirmed_client(&dc->clientid, true, nn);
3858         WARN_ON_ONCE(conf && unconf);
3859
3860         if (conf) {
3861                 if (client_has_state(conf)) {
3862                         status = nfserr_clientid_busy;
3863                         goto out;
3864                 }
3865                 status = mark_client_expired_locked(conf);
3866                 if (status)
3867                         goto out;
3868                 clp = conf;
3869         } else if (unconf)
3870                 clp = unconf;
3871         else {
3872                 status = nfserr_stale_clientid;
3873                 goto out;
3874         }
3875         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3876                 clp = NULL;
3877                 status = nfserr_wrong_cred;
3878                 goto out;
3879         }
3880         unhash_client_locked(clp);
3881 out:
3882         spin_unlock(&nn->client_lock);
3883         if (clp)
3884                 expire_client(clp);
3885         return status;
3886 }
3887
3888 __be32
3889 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3890                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3891 {
3892         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3893         __be32 status = 0;
3894
3895         if (rc->rca_one_fs) {
3896                 if (!cstate->current_fh.fh_dentry)
3897                         return nfserr_nofilehandle;
3898                 /*
3899                  * We don't take advantage of the rca_one_fs case.
3900                  * That's OK, it's optional, we can safely ignore it.
3901                  */
3902                 return nfs_ok;
3903         }
3904
3905         status = nfserr_complete_already;
3906         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3907                              &cstate->session->se_client->cl_flags))
3908                 goto out;
3909
3910         status = nfserr_stale_clientid;
3911         if (is_client_expired(cstate->session->se_client))
3912                 /*
3913                  * The following error isn't really legal.
3914                  * But we only get here if the client just explicitly
3915                  * destroyed the client.  Surely it no longer cares what
3916                  * error it gets back on an operation for the dead
3917                  * client.
3918                  */
3919                 goto out;
3920
3921         status = nfs_ok;
3922         nfsd4_client_record_create(cstate->session->se_client);
3923         inc_reclaim_complete(cstate->session->se_client);
3924 out:
3925         return status;
3926 }
3927
3928 __be32
3929 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3930                   union nfsd4_op_u *u)
3931 {
3932         struct nfsd4_setclientid *setclid = &u->setclientid;
3933         struct xdr_netobj       clname = setclid->se_name;
3934         nfs4_verifier           clverifier = setclid->se_verf;
3935         struct nfs4_client      *conf, *new;
3936         struct nfs4_client      *unconf = NULL;
3937         __be32                  status;
3938         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3939
3940         new = create_client(clname, rqstp, &clverifier);
3941         if (new == NULL)
3942                 return nfserr_jukebox;
3943         /* Cases below refer to rfc 3530 section 14.2.33: */
3944         spin_lock(&nn->client_lock);
3945         conf = find_confirmed_client_by_name(&clname, nn);
3946         if (conf && client_has_state(conf)) {
3947                 /* case 0: */
3948                 status = nfserr_clid_inuse;
3949                 if (clp_used_exchangeid(conf))
3950                         goto out;
3951                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3952                         trace_nfsd_clid_inuse_err(conf);
3953                         goto out;
3954                 }
3955         }
3956         unconf = find_unconfirmed_client_by_name(&clname, nn);
3957         if (unconf)
3958                 unhash_client_locked(unconf);
3959         /* We need to handle only case 1: probable callback update */
3960         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3961                 copy_clid(new, conf);
3962                 gen_confirm(new, nn);
3963         }
3964         new->cl_minorversion = 0;
3965         gen_callback(new, setclid, rqstp);
3966         add_to_unconfirmed(new);
3967         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3968         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3969         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3970         new = NULL;
3971         status = nfs_ok;
3972 out:
3973         spin_unlock(&nn->client_lock);
3974         if (new)
3975                 free_client(new);
3976         if (unconf)
3977                 expire_client(unconf);
3978         return status;
3979 }
3980
3981
3982 __be32
3983 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3984                         struct nfsd4_compound_state *cstate,
3985                         union nfsd4_op_u *u)
3986 {
3987         struct nfsd4_setclientid_confirm *setclientid_confirm =
3988                         &u->setclientid_confirm;
3989         struct nfs4_client *conf, *unconf;
3990         struct nfs4_client *old = NULL;
3991         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3992         clientid_t * clid = &setclientid_confirm->sc_clientid;
3993         __be32 status;
3994         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3995
3996         if (STALE_CLIENTID(clid, nn))
3997                 return nfserr_stale_clientid;
3998
3999         spin_lock(&nn->client_lock);
4000         conf = find_confirmed_client(clid, false, nn);
4001         unconf = find_unconfirmed_client(clid, false, nn);
4002         /*
4003          * We try hard to give out unique clientid's, so if we get an
4004          * attempt to confirm the same clientid with a different cred,
4005          * the client may be buggy; this should never happen.
4006          *
4007          * Nevertheless, RFC 7530 recommends INUSE for this case:
4008          */
4009         status = nfserr_clid_inuse;
4010         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
4011                 goto out;
4012         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
4013                 goto out;
4014         /* cases below refer to rfc 3530 section 14.2.34: */
4015         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
4016                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
4017                         /* case 2: probable retransmit */
4018                         status = nfs_ok;
4019                 } else /* case 4: client hasn't noticed we rebooted yet? */
4020                         status = nfserr_stale_clientid;
4021                 goto out;
4022         }
4023         status = nfs_ok;
4024         if (conf) { /* case 1: callback update */
4025                 old = unconf;
4026                 unhash_client_locked(old);
4027                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
4028         } else { /* case 3: normal case; new or rebooted client */
4029                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
4030                 if (old) {
4031                         status = nfserr_clid_inuse;
4032                         if (client_has_state(old)
4033                                         && !same_creds(&unconf->cl_cred,
4034                                                         &old->cl_cred))
4035                                 goto out;
4036                         status = mark_client_expired_locked(old);
4037                         if (status) {
4038                                 old = NULL;
4039                                 goto out;
4040                         }
4041                 }
4042                 move_to_confirmed(unconf);
4043                 conf = unconf;
4044         }
4045         get_client_locked(conf);
4046         spin_unlock(&nn->client_lock);
4047         nfsd4_probe_callback(conf);
4048         spin_lock(&nn->client_lock);
4049         put_client_renew_locked(conf);
4050 out:
4051         spin_unlock(&nn->client_lock);
4052         if (old)
4053                 expire_client(old);
4054         return status;
4055 }
4056
4057 static struct nfs4_file *nfsd4_alloc_file(void)
4058 {
4059         return kmem_cache_alloc(file_slab, GFP_KERNEL);
4060 }
4061
4062 /* OPEN Share state helper functions */
4063 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
4064                                 struct nfs4_file *fp)
4065 {
4066         lockdep_assert_held(&state_lock);
4067
4068         refcount_set(&fp->fi_ref, 1);
4069         spin_lock_init(&fp->fi_lock);
4070         INIT_LIST_HEAD(&fp->fi_stateids);
4071         INIT_LIST_HEAD(&fp->fi_delegations);
4072         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4073         fh_copy_shallow(&fp->fi_fhandle, fh);
4074         fp->fi_deleg_file = NULL;
4075         fp->fi_had_conflict = false;
4076         fp->fi_share_deny = 0;
4077         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4078         memset(fp->fi_access, 0, sizeof(fp->fi_access));
4079 #ifdef CONFIG_NFSD_PNFS
4080         INIT_LIST_HEAD(&fp->fi_lo_states);
4081         atomic_set(&fp->fi_lo_recalls, 0);
4082 #endif
4083         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
4084 }
4085
4086 void
4087 nfsd4_free_slabs(void)
4088 {
4089         kmem_cache_destroy(client_slab);
4090         kmem_cache_destroy(openowner_slab);
4091         kmem_cache_destroy(lockowner_slab);
4092         kmem_cache_destroy(file_slab);
4093         kmem_cache_destroy(stateid_slab);
4094         kmem_cache_destroy(deleg_slab);
4095         kmem_cache_destroy(odstate_slab);
4096 }
4097
4098 int
4099 nfsd4_init_slabs(void)
4100 {
4101         client_slab = kmem_cache_create("nfsd4_clients",
4102                         sizeof(struct nfs4_client), 0, 0, NULL);
4103         if (client_slab == NULL)
4104                 goto out;
4105         openowner_slab = kmem_cache_create("nfsd4_openowners",
4106                         sizeof(struct nfs4_openowner), 0, 0, NULL);
4107         if (openowner_slab == NULL)
4108                 goto out_free_client_slab;
4109         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
4110                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
4111         if (lockowner_slab == NULL)
4112                 goto out_free_openowner_slab;
4113         file_slab = kmem_cache_create("nfsd4_files",
4114                         sizeof(struct nfs4_file), 0, 0, NULL);
4115         if (file_slab == NULL)
4116                 goto out_free_lockowner_slab;
4117         stateid_slab = kmem_cache_create("nfsd4_stateids",
4118                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
4119         if (stateid_slab == NULL)
4120                 goto out_free_file_slab;
4121         deleg_slab = kmem_cache_create("nfsd4_delegations",
4122                         sizeof(struct nfs4_delegation), 0, 0, NULL);
4123         if (deleg_slab == NULL)
4124                 goto out_free_stateid_slab;
4125         odstate_slab = kmem_cache_create("nfsd4_odstate",
4126                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
4127         if (odstate_slab == NULL)
4128                 goto out_free_deleg_slab;
4129         return 0;
4130
4131 out_free_deleg_slab:
4132         kmem_cache_destroy(deleg_slab);
4133 out_free_stateid_slab:
4134         kmem_cache_destroy(stateid_slab);
4135 out_free_file_slab:
4136         kmem_cache_destroy(file_slab);
4137 out_free_lockowner_slab:
4138         kmem_cache_destroy(lockowner_slab);
4139 out_free_openowner_slab:
4140         kmem_cache_destroy(openowner_slab);
4141 out_free_client_slab:
4142         kmem_cache_destroy(client_slab);
4143 out:
4144         return -ENOMEM;
4145 }
4146
4147 static void init_nfs4_replay(struct nfs4_replay *rp)
4148 {
4149         rp->rp_status = nfserr_serverfault;
4150         rp->rp_buflen = 0;
4151         rp->rp_buf = rp->rp_ibuf;
4152         mutex_init(&rp->rp_mutex);
4153 }
4154
4155 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4156                 struct nfs4_stateowner *so)
4157 {
4158         if (!nfsd4_has_session(cstate)) {
4159                 mutex_lock(&so->so_replay.rp_mutex);
4160                 cstate->replay_owner = nfs4_get_stateowner(so);
4161         }
4162 }
4163
4164 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4165 {
4166         struct nfs4_stateowner *so = cstate->replay_owner;
4167
4168         if (so != NULL) {
4169                 cstate->replay_owner = NULL;
4170                 mutex_unlock(&so->so_replay.rp_mutex);
4171                 nfs4_put_stateowner(so);
4172         }
4173 }
4174
4175 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4176 {
4177         struct nfs4_stateowner *sop;
4178
4179         sop = kmem_cache_alloc(slab, GFP_KERNEL);
4180         if (!sop)
4181                 return NULL;
4182
4183         xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4184         if (!sop->so_owner.data) {
4185                 kmem_cache_free(slab, sop);
4186                 return NULL;
4187         }
4188
4189         INIT_LIST_HEAD(&sop->so_stateids);
4190         sop->so_client = clp;
4191         init_nfs4_replay(&sop->so_replay);
4192         atomic_set(&sop->so_count, 1);
4193         return sop;
4194 }
4195
4196 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4197 {
4198         lockdep_assert_held(&clp->cl_lock);
4199
4200         list_add(&oo->oo_owner.so_strhash,
4201                  &clp->cl_ownerstr_hashtbl[strhashval]);
4202         list_add(&oo->oo_perclient, &clp->cl_openowners);
4203 }
4204
4205 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4206 {
4207         unhash_openowner_locked(openowner(so));
4208 }
4209
4210 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4211 {
4212         struct nfs4_openowner *oo = openowner(so);
4213
4214         kmem_cache_free(openowner_slab, oo);
4215 }
4216
4217 static const struct nfs4_stateowner_operations openowner_ops = {
4218         .so_unhash =    nfs4_unhash_openowner,
4219         .so_free =      nfs4_free_openowner,
4220 };
4221
4222 static struct nfs4_ol_stateid *
4223 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4224 {
4225         struct nfs4_ol_stateid *local, *ret = NULL;
4226         struct nfs4_openowner *oo = open->op_openowner;
4227
4228         lockdep_assert_held(&fp->fi_lock);
4229
4230         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4231                 /* ignore lock owners */
4232                 if (local->st_stateowner->so_is_open_owner == 0)
4233                         continue;
4234                 if (local->st_stateowner != &oo->oo_owner)
4235                         continue;
4236                 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
4237                         ret = local;
4238                         refcount_inc(&ret->st_stid.sc_count);
4239                         break;
4240                 }
4241         }
4242         return ret;
4243 }
4244
4245 static __be32
4246 nfsd4_verify_open_stid(struct nfs4_stid *s)
4247 {
4248         __be32 ret = nfs_ok;
4249
4250         switch (s->sc_type) {
4251         default:
4252                 break;
4253         case 0:
4254         case NFS4_CLOSED_STID:
4255         case NFS4_CLOSED_DELEG_STID:
4256                 ret = nfserr_bad_stateid;
4257                 break;
4258         case NFS4_REVOKED_DELEG_STID:
4259                 ret = nfserr_deleg_revoked;
4260         }
4261         return ret;
4262 }
4263
4264 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4265 static __be32
4266 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4267 {
4268         __be32 ret;
4269
4270         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4271         ret = nfsd4_verify_open_stid(&stp->st_stid);
4272         if (ret != nfs_ok)
4273                 mutex_unlock(&stp->st_mutex);
4274         return ret;
4275 }
4276
4277 static struct nfs4_ol_stateid *
4278 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4279 {
4280         struct nfs4_ol_stateid *stp;
4281         for (;;) {
4282                 spin_lock(&fp->fi_lock);
4283                 stp = nfsd4_find_existing_open(fp, open);
4284                 spin_unlock(&fp->fi_lock);
4285                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4286                         break;
4287                 nfs4_put_stid(&stp->st_stid);
4288         }
4289         return stp;
4290 }
4291
4292 static struct nfs4_openowner *
4293 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4294                            struct nfsd4_compound_state *cstate)
4295 {
4296         struct nfs4_client *clp = cstate->clp;
4297         struct nfs4_openowner *oo, *ret;
4298
4299         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4300         if (!oo)
4301                 return NULL;
4302         oo->oo_owner.so_ops = &openowner_ops;
4303         oo->oo_owner.so_is_open_owner = 1;
4304         oo->oo_owner.so_seqid = open->op_seqid;
4305         oo->oo_flags = 0;
4306         if (nfsd4_has_session(cstate))
4307                 oo->oo_flags |= NFS4_OO_CONFIRMED;
4308         oo->oo_time = 0;
4309         oo->oo_last_closed_stid = NULL;
4310         INIT_LIST_HEAD(&oo->oo_close_lru);
4311         spin_lock(&clp->cl_lock);
4312         ret = find_openstateowner_str_locked(strhashval, open, clp);
4313         if (ret == NULL) {
4314                 hash_openowner(oo, clp, strhashval);
4315                 ret = oo;
4316         } else
4317                 nfs4_free_stateowner(&oo->oo_owner);
4318
4319         spin_unlock(&clp->cl_lock);
4320         return ret;
4321 }
4322
4323 static struct nfs4_ol_stateid *
4324 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4325 {
4326
4327         struct nfs4_openowner *oo = open->op_openowner;
4328         struct nfs4_ol_stateid *retstp = NULL;
4329         struct nfs4_ol_stateid *stp;
4330
4331         stp = open->op_stp;
4332         /* We are moving these outside of the spinlocks to avoid the warnings */
4333         mutex_init(&stp->st_mutex);
4334         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4335
4336 retry:
4337         spin_lock(&oo->oo_owner.so_client->cl_lock);
4338         spin_lock(&fp->fi_lock);
4339
4340         retstp = nfsd4_find_existing_open(fp, open);
4341         if (retstp)
4342                 goto out_unlock;
4343
4344         open->op_stp = NULL;
4345         refcount_inc(&stp->st_stid.sc_count);
4346         stp->st_stid.sc_type = NFS4_OPEN_STID;
4347         INIT_LIST_HEAD(&stp->st_locks);
4348         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4349         get_nfs4_file(fp);
4350         stp->st_stid.sc_file = fp;
4351         stp->st_access_bmap = 0;
4352         stp->st_deny_bmap = 0;
4353         stp->st_openstp = NULL;
4354         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4355         list_add(&stp->st_perfile, &fp->fi_stateids);
4356
4357 out_unlock:
4358         spin_unlock(&fp->fi_lock);
4359         spin_unlock(&oo->oo_owner.so_client->cl_lock);
4360         if (retstp) {
4361                 /* Handle races with CLOSE */
4362                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4363                         nfs4_put_stid(&retstp->st_stid);
4364                         goto retry;
4365                 }
4366                 /* To keep mutex tracking happy */
4367                 mutex_unlock(&stp->st_mutex);
4368                 stp = retstp;
4369         }
4370         return stp;
4371 }
4372
4373 /*
4374  * In the 4.0 case we need to keep the owners around a little while to handle
4375  * CLOSE replay. We still do need to release any file access that is held by
4376  * them before returning however.
4377  */
4378 static void
4379 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4380 {
4381         struct nfs4_ol_stateid *last;
4382         struct nfs4_openowner *oo = openowner(s->st_stateowner);
4383         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4384                                                 nfsd_net_id);
4385
4386         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4387
4388         /*
4389          * We know that we hold one reference via nfsd4_close, and another
4390          * "persistent" reference for the client. If the refcount is higher
4391          * than 2, then there are still calls in progress that are using this
4392          * stateid. We can't put the sc_file reference until they are finished.
4393          * Wait for the refcount to drop to 2. Since it has been unhashed,
4394          * there should be no danger of the refcount going back up again at
4395          * this point.
4396          */
4397         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4398
4399         release_all_access(s);
4400         if (s->st_stid.sc_file) {
4401                 put_nfs4_file(s->st_stid.sc_file);
4402                 s->st_stid.sc_file = NULL;
4403         }
4404
4405         spin_lock(&nn->client_lock);
4406         last = oo->oo_last_closed_stid;
4407         oo->oo_last_closed_stid = s;
4408         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4409         oo->oo_time = ktime_get_boottime_seconds();
4410         spin_unlock(&nn->client_lock);
4411         if (last)
4412                 nfs4_put_stid(&last->st_stid);
4413 }
4414
4415 /* search file_hashtbl[] for file */
4416 static struct nfs4_file *
4417 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
4418 {
4419         struct nfs4_file *fp;
4420
4421         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash,
4422                                 lockdep_is_held(&state_lock)) {
4423                 if (fh_match(&fp->fi_fhandle, fh)) {
4424                         if (refcount_inc_not_zero(&fp->fi_ref))
4425                                 return fp;
4426                 }
4427         }
4428         return NULL;
4429 }
4430
4431 struct nfs4_file *
4432 find_file(struct knfsd_fh *fh)
4433 {
4434         struct nfs4_file *fp;
4435         unsigned int hashval = file_hashval(fh);
4436
4437         rcu_read_lock();
4438         fp = find_file_locked(fh, hashval);
4439         rcu_read_unlock();
4440         return fp;
4441 }
4442
4443 static struct nfs4_file *
4444 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
4445 {
4446         struct nfs4_file *fp;
4447         unsigned int hashval = file_hashval(fh);
4448
4449         rcu_read_lock();
4450         fp = find_file_locked(fh, hashval);
4451         rcu_read_unlock();
4452         if (fp)
4453                 return fp;
4454
4455         spin_lock(&state_lock);
4456         fp = find_file_locked(fh, hashval);
4457         if (likely(fp == NULL)) {
4458                 nfsd4_init_file(fh, hashval, new);
4459                 fp = new;
4460         }
4461         spin_unlock(&state_lock);
4462
4463         return fp;
4464 }
4465
4466 /*
4467  * Called to check deny when READ with all zero stateid or
4468  * WRITE with all zero or all one stateid
4469  */
4470 static __be32
4471 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
4472 {
4473         struct nfs4_file *fp;
4474         __be32 ret = nfs_ok;
4475
4476         fp = find_file(&current_fh->fh_handle);
4477         if (!fp)
4478                 return ret;
4479         /* Check for conflicting share reservations */
4480         spin_lock(&fp->fi_lock);
4481         if (fp->fi_share_deny & deny_type)
4482                 ret = nfserr_locked;
4483         spin_unlock(&fp->fi_lock);
4484         put_nfs4_file(fp);
4485         return ret;
4486 }
4487
4488 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
4489 {
4490         struct nfs4_delegation *dp = cb_to_delegation(cb);
4491         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
4492                                           nfsd_net_id);
4493
4494         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
4495
4496         /*
4497          * We can't do this in nfsd_break_deleg_cb because it is
4498          * already holding inode->i_lock.
4499          *
4500          * If the dl_time != 0, then we know that it has already been
4501          * queued for a lease break. Don't queue it again.
4502          */
4503         spin_lock(&state_lock);
4504         if (dp->dl_time == 0) {
4505                 dp->dl_time = ktime_get_boottime_seconds();
4506                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
4507         }
4508         spin_unlock(&state_lock);
4509 }
4510
4511 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
4512                 struct rpc_task *task)
4513 {
4514         struct nfs4_delegation *dp = cb_to_delegation(cb);
4515
4516         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID ||
4517             dp->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4518                 return 1;
4519
4520         switch (task->tk_status) {
4521         case 0:
4522                 return 1;
4523         case -NFS4ERR_DELAY:
4524                 rpc_delay(task, 2 * HZ);
4525                 return 0;
4526         case -EBADHANDLE:
4527         case -NFS4ERR_BAD_STATEID:
4528                 /*
4529                  * Race: client probably got cb_recall before open reply
4530                  * granting delegation.
4531                  */
4532                 if (dp->dl_retries--) {
4533                         rpc_delay(task, 2 * HZ);
4534                         return 0;
4535                 }
4536                 fallthrough;
4537         default:
4538                 return 1;
4539         }
4540 }
4541
4542 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
4543 {
4544         struct nfs4_delegation *dp = cb_to_delegation(cb);
4545
4546         nfs4_put_stid(&dp->dl_stid);
4547 }
4548
4549 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
4550         .prepare        = nfsd4_cb_recall_prepare,
4551         .done           = nfsd4_cb_recall_done,
4552         .release        = nfsd4_cb_recall_release,
4553 };
4554
4555 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
4556 {
4557         /*
4558          * We're assuming the state code never drops its reference
4559          * without first removing the lease.  Since we're in this lease
4560          * callback (and since the lease code is serialized by the
4561          * i_lock) we know the server hasn't removed the lease yet, and
4562          * we know it's safe to take a reference.
4563          */
4564         refcount_inc(&dp->dl_stid.sc_count);
4565         nfsd4_run_cb(&dp->dl_recall);
4566 }
4567
4568 /* Called from break_lease() with i_lock held. */
4569 static bool
4570 nfsd_break_deleg_cb(struct file_lock *fl)
4571 {
4572         bool ret = false;
4573         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4574         struct nfs4_file *fp = dp->dl_stid.sc_file;
4575
4576         trace_nfsd_deleg_break(&dp->dl_stid.sc_stateid);
4577
4578         /*
4579          * We don't want the locks code to timeout the lease for us;
4580          * we'll remove it ourself if a delegation isn't returned
4581          * in time:
4582          */
4583         fl->fl_break_time = 0;
4584
4585         spin_lock(&fp->fi_lock);
4586         fp->fi_had_conflict = true;
4587         nfsd_break_one_deleg(dp);
4588         spin_unlock(&fp->fi_lock);
4589         return ret;
4590 }
4591
4592 static bool nfsd_breaker_owns_lease(struct file_lock *fl)
4593 {
4594         struct nfs4_delegation *dl = fl->fl_owner;
4595         struct svc_rqst *rqst;
4596         struct nfs4_client *clp;
4597
4598         if (!i_am_nfsd())
4599                 return NULL;
4600         rqst = kthread_data(current);
4601         /* Note rq_prog == NFS_ACL_PROGRAM is also possible: */
4602         if (rqst->rq_prog != NFS_PROGRAM || rqst->rq_vers < 4)
4603                 return NULL;
4604         clp = *(rqst->rq_lease_breaker);
4605         return dl->dl_stid.sc_client == clp;
4606 }
4607
4608 static int
4609 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4610                      struct list_head *dispose)
4611 {
4612         if (arg & F_UNLCK)
4613                 return lease_modify(onlist, arg, dispose);
4614         else
4615                 return -EAGAIN;
4616 }
4617
4618 static const struct lock_manager_operations nfsd_lease_mng_ops = {
4619         .lm_breaker_owns_lease = nfsd_breaker_owns_lease,
4620         .lm_break = nfsd_break_deleg_cb,
4621         .lm_change = nfsd_change_deleg_cb,
4622 };
4623
4624 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4625 {
4626         if (nfsd4_has_session(cstate))
4627                 return nfs_ok;
4628         if (seqid == so->so_seqid - 1)
4629                 return nfserr_replay_me;
4630         if (seqid == so->so_seqid)
4631                 return nfs_ok;
4632         return nfserr_bad_seqid;
4633 }
4634
4635 static __be32 lookup_clientid(clientid_t *clid,
4636                 struct nfsd4_compound_state *cstate,
4637                 struct nfsd_net *nn,
4638                 bool sessions)
4639 {
4640         struct nfs4_client *found;
4641
4642         if (cstate->clp) {
4643                 found = cstate->clp;
4644                 if (!same_clid(&found->cl_clientid, clid))
4645                         return nfserr_stale_clientid;
4646                 return nfs_ok;
4647         }
4648
4649         if (STALE_CLIENTID(clid, nn))
4650                 return nfserr_stale_clientid;
4651
4652         /*
4653          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4654          * cached already then we know this is for is for v4.0 and "sessions"
4655          * will be false.
4656          */
4657         WARN_ON_ONCE(cstate->session);
4658         spin_lock(&nn->client_lock);
4659         found = find_confirmed_client(clid, sessions, nn);
4660         if (!found) {
4661                 spin_unlock(&nn->client_lock);
4662                 return nfserr_expired;
4663         }
4664         atomic_inc(&found->cl_rpc_users);
4665         spin_unlock(&nn->client_lock);
4666
4667         /* Cache the nfs4_client in cstate! */
4668         cstate->clp = found;
4669         return nfs_ok;
4670 }
4671
4672 __be32
4673 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4674                     struct nfsd4_open *open, struct nfsd_net *nn)
4675 {
4676         clientid_t *clientid = &open->op_clientid;
4677         struct nfs4_client *clp = NULL;
4678         unsigned int strhashval;
4679         struct nfs4_openowner *oo = NULL;
4680         __be32 status;
4681
4682         if (STALE_CLIENTID(&open->op_clientid, nn))
4683                 return nfserr_stale_clientid;
4684         /*
4685          * In case we need it later, after we've already created the
4686          * file and don't want to risk a further failure:
4687          */
4688         open->op_file = nfsd4_alloc_file();
4689         if (open->op_file == NULL)
4690                 return nfserr_jukebox;
4691
4692         status = lookup_clientid(clientid, cstate, nn, false);
4693         if (status)
4694                 return status;
4695         clp = cstate->clp;
4696
4697         strhashval = ownerstr_hashval(&open->op_owner);
4698         oo = find_openstateowner_str(strhashval, open, clp);
4699         open->op_openowner = oo;
4700         if (!oo) {
4701                 goto new_owner;
4702         }
4703         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4704                 /* Replace unconfirmed owners without checking for replay. */
4705                 release_openowner(oo);
4706                 open->op_openowner = NULL;
4707                 goto new_owner;
4708         }
4709         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4710         if (status)
4711                 return status;
4712         goto alloc_stateid;
4713 new_owner:
4714         oo = alloc_init_open_stateowner(strhashval, open, cstate);
4715         if (oo == NULL)
4716                 return nfserr_jukebox;
4717         open->op_openowner = oo;
4718 alloc_stateid:
4719         open->op_stp = nfs4_alloc_open_stateid(clp);
4720         if (!open->op_stp)
4721                 return nfserr_jukebox;
4722
4723         if (nfsd4_has_session(cstate) &&
4724             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4725                 open->op_odstate = alloc_clnt_odstate(clp);
4726                 if (!open->op_odstate)
4727                         return nfserr_jukebox;
4728         }
4729
4730         return nfs_ok;
4731 }
4732
4733 static inline __be32
4734 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4735 {
4736         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4737                 return nfserr_openmode;
4738         else
4739                 return nfs_ok;
4740 }
4741
4742 static int share_access_to_flags(u32 share_access)
4743 {
4744         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4745 }
4746
4747 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4748 {
4749         struct nfs4_stid *ret;
4750
4751         ret = find_stateid_by_type(cl, s,
4752                                 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4753         if (!ret)
4754                 return NULL;
4755         return delegstateid(ret);
4756 }
4757
4758 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4759 {
4760         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4761                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4762 }
4763
4764 static __be32
4765 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4766                 struct nfs4_delegation **dp)
4767 {
4768         int flags;
4769         __be32 status = nfserr_bad_stateid;
4770         struct nfs4_delegation *deleg;
4771
4772         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4773         if (deleg == NULL)
4774                 goto out;
4775         if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4776                 nfs4_put_stid(&deleg->dl_stid);
4777                 if (cl->cl_minorversion)
4778                         status = nfserr_deleg_revoked;
4779                 goto out;
4780         }
4781         flags = share_access_to_flags(open->op_share_access);
4782         status = nfs4_check_delegmode(deleg, flags);
4783         if (status) {
4784                 nfs4_put_stid(&deleg->dl_stid);
4785                 goto out;
4786         }
4787         *dp = deleg;
4788 out:
4789         if (!nfsd4_is_deleg_cur(open))
4790                 return nfs_ok;
4791         if (status)
4792                 return status;
4793         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4794         return nfs_ok;
4795 }
4796
4797 static inline int nfs4_access_to_access(u32 nfs4_access)
4798 {
4799         int flags = 0;
4800
4801         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4802                 flags |= NFSD_MAY_READ;
4803         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4804                 flags |= NFSD_MAY_WRITE;
4805         return flags;
4806 }
4807
4808 static inline __be32
4809 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4810                 struct nfsd4_open *open)
4811 {
4812         struct iattr iattr = {
4813                 .ia_valid = ATTR_SIZE,
4814                 .ia_size = 0,
4815         };
4816         if (!open->op_truncate)
4817                 return 0;
4818         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4819                 return nfserr_inval;
4820         return nfsd_setattr(rqstp, fh, &iattr, 0, (time64_t)0);
4821 }
4822
4823 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4824                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4825                 struct nfsd4_open *open)
4826 {
4827         struct nfsd_file *nf = NULL;
4828         __be32 status;
4829         int oflag = nfs4_access_to_omode(open->op_share_access);
4830         int access = nfs4_access_to_access(open->op_share_access);
4831         unsigned char old_access_bmap, old_deny_bmap;
4832
4833         spin_lock(&fp->fi_lock);
4834
4835         /*
4836          * Are we trying to set a deny mode that would conflict with
4837          * current access?
4838          */
4839         status = nfs4_file_check_deny(fp, open->op_share_deny);
4840         if (status != nfs_ok) {
4841                 spin_unlock(&fp->fi_lock);
4842                 goto out;
4843         }
4844
4845         /* set access to the file */
4846         status = nfs4_file_get_access(fp, open->op_share_access);
4847         if (status != nfs_ok) {
4848                 spin_unlock(&fp->fi_lock);
4849                 goto out;
4850         }
4851
4852         /* Set access bits in stateid */
4853         old_access_bmap = stp->st_access_bmap;
4854         set_access(open->op_share_access, stp);
4855
4856         /* Set new deny mask */
4857         old_deny_bmap = stp->st_deny_bmap;
4858         set_deny(open->op_share_deny, stp);
4859         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4860
4861         if (!fp->fi_fds[oflag]) {
4862                 spin_unlock(&fp->fi_lock);
4863                 status = nfsd_file_acquire(rqstp, cur_fh, access, &nf);
4864                 if (status)
4865                         goto out_put_access;
4866                 spin_lock(&fp->fi_lock);
4867                 if (!fp->fi_fds[oflag]) {
4868                         fp->fi_fds[oflag] = nf;
4869                         nf = NULL;
4870                 }
4871         }
4872         spin_unlock(&fp->fi_lock);
4873         if (nf)
4874                 nfsd_file_put(nf);
4875
4876         status = nfsd4_truncate(rqstp, cur_fh, open);
4877         if (status)
4878                 goto out_put_access;
4879 out:
4880         return status;
4881 out_put_access:
4882         stp->st_access_bmap = old_access_bmap;
4883         nfs4_file_put_access(fp, open->op_share_access);
4884         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4885         goto out;
4886 }
4887
4888 static __be32
4889 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)
4890 {
4891         __be32 status;
4892         unsigned char old_deny_bmap = stp->st_deny_bmap;
4893
4894         if (!test_access(open->op_share_access, stp))
4895                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4896
4897         /* test and set deny mode */
4898         spin_lock(&fp->fi_lock);
4899         status = nfs4_file_check_deny(fp, open->op_share_deny);
4900         if (status == nfs_ok) {
4901                 set_deny(open->op_share_deny, stp);
4902                 fp->fi_share_deny |=
4903                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4904         }
4905         spin_unlock(&fp->fi_lock);
4906
4907         if (status != nfs_ok)
4908                 return status;
4909
4910         status = nfsd4_truncate(rqstp, cur_fh, open);
4911         if (status != nfs_ok)
4912                 reset_union_bmap_deny(old_deny_bmap, stp);
4913         return status;
4914 }
4915
4916 /* Should we give out recallable state?: */
4917 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4918 {
4919         if (clp->cl_cb_state == NFSD4_CB_UP)
4920                 return true;
4921         /*
4922          * In the sessions case, since we don't have to establish a
4923          * separate connection for callbacks, we assume it's OK
4924          * until we hear otherwise:
4925          */
4926         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4927 }
4928
4929 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
4930                                                 int flag)
4931 {
4932         struct file_lock *fl;
4933
4934         fl = locks_alloc_lock();
4935         if (!fl)
4936                 return NULL;
4937         fl->fl_lmops = &nfsd_lease_mng_ops;
4938         fl->fl_flags = FL_DELEG;
4939         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4940         fl->fl_end = OFFSET_MAX;
4941         fl->fl_owner = (fl_owner_t)dp;
4942         fl->fl_pid = current->tgid;
4943         fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
4944         return fl;
4945 }
4946
4947 static int nfsd4_check_conflicting_opens(struct nfs4_client *clp,
4948                                                 struct nfs4_file *fp)
4949 {
4950         struct nfs4_clnt_odstate *co;
4951         struct file *f = fp->fi_deleg_file->nf_file;
4952         struct inode *ino = locks_inode(f);
4953         int writes = atomic_read(&ino->i_writecount);
4954
4955         if (fp->fi_fds[O_WRONLY])
4956                 writes--;
4957         if (fp->fi_fds[O_RDWR])
4958                 writes--;
4959         if (writes > 0)
4960                 return -EAGAIN;
4961         spin_lock(&fp->fi_lock);
4962         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
4963                 if (co->co_client != clp) {
4964                         spin_unlock(&fp->fi_lock);
4965                         return -EAGAIN;
4966                 }
4967         }
4968         spin_unlock(&fp->fi_lock);
4969         return 0;
4970 }
4971
4972 static struct nfs4_delegation *
4973 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4974                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4975 {
4976         int status = 0;
4977         struct nfs4_delegation *dp;
4978         struct nfsd_file *nf;
4979         struct file_lock *fl;
4980
4981         /*
4982          * The fi_had_conflict and nfs_get_existing_delegation checks
4983          * here are just optimizations; we'll need to recheck them at
4984          * the end:
4985          */
4986         if (fp->fi_had_conflict)
4987                 return ERR_PTR(-EAGAIN);
4988
4989         nf = find_readable_file(fp);
4990         if (!nf) {
4991                 /*
4992                  * We probably could attempt another open and get a read
4993                  * delegation, but for now, don't bother until the
4994                  * client actually sends us one.
4995                  */
4996                 return ERR_PTR(-EAGAIN);
4997         }
4998         spin_lock(&state_lock);
4999         spin_lock(&fp->fi_lock);
5000         if (nfs4_delegation_exists(clp, fp))
5001                 status = -EAGAIN;
5002         else if (!fp->fi_deleg_file) {
5003                 fp->fi_deleg_file = nf;
5004                 /* increment early to prevent fi_deleg_file from being
5005                  * cleared */
5006                 fp->fi_delegees = 1;
5007                 nf = NULL;
5008         } else
5009                 fp->fi_delegees++;
5010         spin_unlock(&fp->fi_lock);
5011         spin_unlock(&state_lock);
5012         if (nf)
5013                 nfsd_file_put(nf);
5014         if (status)
5015                 return ERR_PTR(status);
5016
5017         status = -ENOMEM;
5018         dp = alloc_init_deleg(clp, fp, fh, odstate);
5019         if (!dp)
5020                 goto out_delegees;
5021
5022         fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
5023         if (!fl)
5024                 goto out_clnt_odstate;
5025
5026         status = nfsd4_check_conflicting_opens(clp, fp);
5027         if (status) {
5028                 locks_free_lock(fl);
5029                 goto out_clnt_odstate;
5030         }
5031         status = vfs_setlease(fp->fi_deleg_file->nf_file, fl->fl_type, &fl, NULL);
5032         if (fl)
5033                 locks_free_lock(fl);
5034         if (status)
5035                 goto out_clnt_odstate;
5036         status = nfsd4_check_conflicting_opens(clp, fp);
5037         if (status)
5038                 goto out_clnt_odstate;
5039
5040         spin_lock(&state_lock);
5041         spin_lock(&fp->fi_lock);
5042         if (fp->fi_had_conflict)
5043                 status = -EAGAIN;
5044         else
5045                 status = hash_delegation_locked(dp, fp);
5046         spin_unlock(&fp->fi_lock);
5047         spin_unlock(&state_lock);
5048
5049         if (status)
5050                 goto out_unlock;
5051
5052         return dp;
5053 out_unlock:
5054         vfs_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
5055 out_clnt_odstate:
5056         put_clnt_odstate(dp->dl_clnt_odstate);
5057         nfs4_put_stid(&dp->dl_stid);
5058 out_delegees:
5059         put_deleg_file(fp);
5060         return ERR_PTR(status);
5061 }
5062
5063 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
5064 {
5065         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5066         if (status == -EAGAIN)
5067                 open->op_why_no_deleg = WND4_CONTENTION;
5068         else {
5069                 open->op_why_no_deleg = WND4_RESOURCE;
5070                 switch (open->op_deleg_want) {
5071                 case NFS4_SHARE_WANT_READ_DELEG:
5072                 case NFS4_SHARE_WANT_WRITE_DELEG:
5073                 case NFS4_SHARE_WANT_ANY_DELEG:
5074                         break;
5075                 case NFS4_SHARE_WANT_CANCEL:
5076                         open->op_why_no_deleg = WND4_CANCELLED;
5077                         break;
5078                 case NFS4_SHARE_WANT_NO_DELEG:
5079                         WARN_ON_ONCE(1);
5080                 }
5081         }
5082 }
5083
5084 /*
5085  * Attempt to hand out a delegation.
5086  *
5087  * Note we don't support write delegations, and won't until the vfs has
5088  * proper support for them.
5089  */
5090 static void
5091 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
5092                         struct nfs4_ol_stateid *stp)
5093 {
5094         struct nfs4_delegation *dp;
5095         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
5096         struct nfs4_client *clp = stp->st_stid.sc_client;
5097         int cb_up;
5098         int status = 0;
5099
5100         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
5101         open->op_recall = 0;
5102         switch (open->op_claim_type) {
5103                 case NFS4_OPEN_CLAIM_PREVIOUS:
5104                         if (!cb_up)
5105                                 open->op_recall = 1;
5106                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
5107                                 goto out_no_deleg;
5108                         break;
5109                 case NFS4_OPEN_CLAIM_NULL:
5110                 case NFS4_OPEN_CLAIM_FH:
5111                         /*
5112                          * Let's not give out any delegations till everyone's
5113                          * had the chance to reclaim theirs, *and* until
5114                          * NLM locks have all been reclaimed:
5115                          */
5116                         if (locks_in_grace(clp->net))
5117                                 goto out_no_deleg;
5118                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
5119                                 goto out_no_deleg;
5120                         break;
5121                 default:
5122                         goto out_no_deleg;
5123         }
5124         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
5125         if (IS_ERR(dp))
5126                 goto out_no_deleg;
5127
5128         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5129
5130         trace_nfsd_deleg_read(&dp->dl_stid.sc_stateid);
5131         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
5132         nfs4_put_stid(&dp->dl_stid);
5133         return;
5134 out_no_deleg:
5135         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
5136         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
5137             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
5138                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
5139                 open->op_recall = 1;
5140         }
5141
5142         /* 4.1 client asking for a delegation? */
5143         if (open->op_deleg_want)
5144                 nfsd4_open_deleg_none_ext(open, status);
5145         return;
5146 }
5147
5148 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
5149                                         struct nfs4_delegation *dp)
5150 {
5151         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
5152             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5153                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5154                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
5155         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
5156                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5157                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5158                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
5159         }
5160         /* Otherwise the client must be confused wanting a delegation
5161          * it already has, therefore we don't return
5162          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
5163          */
5164 }
5165
5166 __be32
5167 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
5168 {
5169         struct nfsd4_compoundres *resp = rqstp->rq_resp;
5170         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
5171         struct nfs4_file *fp = NULL;
5172         struct nfs4_ol_stateid *stp = NULL;
5173         struct nfs4_delegation *dp = NULL;
5174         __be32 status;
5175         bool new_stp = false;
5176
5177         /*
5178          * Lookup file; if found, lookup stateid and check open request,
5179          * and check for delegations in the process of being recalled.
5180          * If not found, create the nfs4_file struct
5181          */
5182         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
5183         if (fp != open->op_file) {
5184                 status = nfs4_check_deleg(cl, open, &dp);
5185                 if (status)
5186                         goto out;
5187                 stp = nfsd4_find_and_lock_existing_open(fp, open);
5188         } else {
5189                 open->op_file = NULL;
5190                 status = nfserr_bad_stateid;
5191                 if (nfsd4_is_deleg_cur(open))
5192                         goto out;
5193         }
5194
5195         if (!stp) {
5196                 stp = init_open_stateid(fp, open);
5197                 if (!open->op_stp)
5198                         new_stp = true;
5199         }
5200
5201         /*
5202          * OPEN the file, or upgrade an existing OPEN.
5203          * If truncate fails, the OPEN fails.
5204          *
5205          * stp is already locked.
5206          */
5207         if (!new_stp) {
5208                 /* Stateid was found, this is an OPEN upgrade */
5209                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
5210                 if (status) {
5211                         mutex_unlock(&stp->st_mutex);
5212                         goto out;
5213                 }
5214         } else {
5215                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
5216                 if (status) {
5217                         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5218                         release_open_stateid(stp);
5219                         mutex_unlock(&stp->st_mutex);
5220                         goto out;
5221                 }
5222
5223                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
5224                                                         open->op_odstate);
5225                 if (stp->st_clnt_odstate == open->op_odstate)
5226                         open->op_odstate = NULL;
5227         }
5228
5229         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
5230         mutex_unlock(&stp->st_mutex);
5231
5232         if (nfsd4_has_session(&resp->cstate)) {
5233                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
5234                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5235                         open->op_why_no_deleg = WND4_NOT_WANTED;
5236                         goto nodeleg;
5237                 }
5238         }
5239
5240         /*
5241         * Attempt to hand out a delegation. No error return, because the
5242         * OPEN succeeds even if we fail.
5243         */
5244         nfs4_open_delegation(current_fh, open, stp);
5245 nodeleg:
5246         status = nfs_ok;
5247         trace_nfsd_open(&stp->st_stid.sc_stateid);
5248 out:
5249         /* 4.1 client trying to upgrade/downgrade delegation? */
5250         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
5251             open->op_deleg_want)
5252                 nfsd4_deleg_xgrade_none_ext(open, dp);
5253
5254         if (fp)
5255                 put_nfs4_file(fp);
5256         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
5257                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5258         /*
5259         * To finish the open response, we just need to set the rflags.
5260         */
5261         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
5262         if (nfsd4_has_session(&resp->cstate))
5263                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
5264         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
5265                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
5266
5267         if (dp)
5268                 nfs4_put_stid(&dp->dl_stid);
5269         if (stp)
5270                 nfs4_put_stid(&stp->st_stid);
5271
5272         return status;
5273 }
5274
5275 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
5276                               struct nfsd4_open *open)
5277 {
5278         if (open->op_openowner) {
5279                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
5280
5281                 nfsd4_cstate_assign_replay(cstate, so);
5282                 nfs4_put_stateowner(so);
5283         }
5284         if (open->op_file)
5285                 kmem_cache_free(file_slab, open->op_file);
5286         if (open->op_stp)
5287                 nfs4_put_stid(&open->op_stp->st_stid);
5288         if (open->op_odstate)
5289                 kmem_cache_free(odstate_slab, open->op_odstate);
5290 }
5291
5292 __be32
5293 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5294             union nfsd4_op_u *u)
5295 {
5296         clientid_t *clid = &u->renew;
5297         struct nfs4_client *clp;
5298         __be32 status;
5299         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5300
5301         trace_nfsd_clid_renew(clid);
5302         status = lookup_clientid(clid, cstate, nn, false);
5303         if (status)
5304                 goto out;
5305         clp = cstate->clp;
5306         status = nfserr_cb_path_down;
5307         if (!list_empty(&clp->cl_delegations)
5308                         && clp->cl_cb_state != NFSD4_CB_UP)
5309                 goto out;
5310         status = nfs_ok;
5311 out:
5312         return status;
5313 }
5314
5315 void
5316 nfsd4_end_grace(struct nfsd_net *nn)
5317 {
5318         /* do nothing if grace period already ended */
5319         if (nn->grace_ended)
5320                 return;
5321
5322         trace_nfsd_grace_complete(nn);
5323         nn->grace_ended = true;
5324         /*
5325          * If the server goes down again right now, an NFSv4
5326          * client will still be allowed to reclaim after it comes back up,
5327          * even if it hasn't yet had a chance to reclaim state this time.
5328          *
5329          */
5330         nfsd4_record_grace_done(nn);
5331         /*
5332          * At this point, NFSv4 clients can still reclaim.  But if the
5333          * server crashes, any that have not yet reclaimed will be out
5334          * of luck on the next boot.
5335          *
5336          * (NFSv4.1+ clients are considered to have reclaimed once they
5337          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
5338          * have reclaimed after their first OPEN.)
5339          */
5340         locks_end_grace(&nn->nfsd4_manager);
5341         /*
5342          * At this point, and once lockd and/or any other containers
5343          * exit their grace period, further reclaims will fail and
5344          * regular locking can resume.
5345          */
5346 }
5347
5348 /*
5349  * If we've waited a lease period but there are still clients trying to
5350  * reclaim, wait a little longer to give them a chance to finish.
5351  */
5352 static bool clients_still_reclaiming(struct nfsd_net *nn)
5353 {
5354         time64_t double_grace_period_end = nn->boot_time +
5355                                            2 * nn->nfsd4_lease;
5356
5357         if (nn->track_reclaim_completes &&
5358                         atomic_read(&nn->nr_reclaim_complete) ==
5359                         nn->reclaim_str_hashtbl_size)
5360                 return false;
5361         if (!nn->somebody_reclaimed)
5362                 return false;
5363         nn->somebody_reclaimed = false;
5364         /*
5365          * If we've given them *two* lease times to reclaim, and they're
5366          * still not done, give up:
5367          */
5368         if (ktime_get_boottime_seconds() > double_grace_period_end)
5369                 return false;
5370         return true;
5371 }
5372
5373 static time64_t
5374 nfs4_laundromat(struct nfsd_net *nn)
5375 {
5376         struct nfs4_client *clp;
5377         struct nfs4_openowner *oo;
5378         struct nfs4_delegation *dp;
5379         struct nfs4_ol_stateid *stp;
5380         struct nfsd4_blocked_lock *nbl;
5381         struct list_head *pos, *next, reaplist;
5382         time64_t cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease;
5383         time64_t t, new_timeo = nn->nfsd4_lease;
5384         struct nfs4_cpntf_state *cps;
5385         copy_stateid_t *cps_t;
5386         int i;
5387
5388         if (clients_still_reclaiming(nn)) {
5389                 new_timeo = 0;
5390                 goto out;
5391         }
5392         nfsd4_end_grace(nn);
5393         INIT_LIST_HEAD(&reaplist);
5394
5395         spin_lock(&nn->s2s_cp_lock);
5396         idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
5397                 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
5398                 if (cps->cp_stateid.sc_type == NFS4_COPYNOTIFY_STID &&
5399                                 cps->cpntf_time > cutoff)
5400                         _free_cpntf_state_locked(nn, cps);
5401         }
5402         spin_unlock(&nn->s2s_cp_lock);
5403
5404         spin_lock(&nn->client_lock);
5405         list_for_each_safe(pos, next, &nn->client_lru) {
5406                 clp = list_entry(pos, struct nfs4_client, cl_lru);
5407                 if (clp->cl_time > cutoff) {
5408                         t = clp->cl_time - cutoff;
5409                         new_timeo = min(new_timeo, t);
5410                         break;
5411                 }
5412                 if (mark_client_expired_locked(clp)) {
5413                         trace_nfsd_clid_expired(&clp->cl_clientid);
5414                         continue;
5415                 }
5416                 list_add(&clp->cl_lru, &reaplist);
5417         }
5418         spin_unlock(&nn->client_lock);
5419         list_for_each_safe(pos, next, &reaplist) {
5420                 clp = list_entry(pos, struct nfs4_client, cl_lru);
5421                 trace_nfsd_clid_purged(&clp->cl_clientid);
5422                 list_del_init(&clp->cl_lru);
5423                 expire_client(clp);
5424         }
5425         spin_lock(&state_lock);
5426         list_for_each_safe(pos, next, &nn->del_recall_lru) {
5427                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5428                 if (dp->dl_time > cutoff) {
5429                         t = dp->dl_time - cutoff;
5430                         new_timeo = min(new_timeo, t);
5431                         break;
5432                 }
5433                 WARN_ON(!unhash_delegation_locked(dp));
5434                 list_add(&dp->dl_recall_lru, &reaplist);
5435         }
5436         spin_unlock(&state_lock);
5437         while (!list_empty(&reaplist)) {
5438                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
5439                                         dl_recall_lru);
5440                 list_del_init(&dp->dl_recall_lru);
5441                 revoke_delegation(dp);
5442         }
5443
5444         spin_lock(&nn->client_lock);
5445         while (!list_empty(&nn->close_lru)) {
5446                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
5447                                         oo_close_lru);
5448                 if (oo->oo_time > cutoff) {
5449                         t = oo->oo_time - cutoff;
5450                         new_timeo = min(new_timeo, t);
5451                         break;
5452                 }
5453                 list_del_init(&oo->oo_close_lru);
5454                 stp = oo->oo_last_closed_stid;
5455                 oo->oo_last_closed_stid = NULL;
5456                 spin_unlock(&nn->client_lock);
5457                 nfs4_put_stid(&stp->st_stid);
5458                 spin_lock(&nn->client_lock);
5459         }
5460         spin_unlock(&nn->client_lock);
5461
5462         /*
5463          * It's possible for a client to try and acquire an already held lock
5464          * that is being held for a long time, and then lose interest in it.
5465          * So, we clean out any un-revisited request after a lease period
5466          * under the assumption that the client is no longer interested.
5467          *
5468          * RFC5661, sec. 9.6 states that the client must not rely on getting
5469          * notifications and must continue to poll for locks, even when the
5470          * server supports them. Thus this shouldn't lead to clients blocking
5471          * indefinitely once the lock does become free.
5472          */
5473         BUG_ON(!list_empty(&reaplist));
5474         spin_lock(&nn->blocked_locks_lock);
5475         while (!list_empty(&nn->blocked_locks_lru)) {
5476                 nbl = list_first_entry(&nn->blocked_locks_lru,
5477                                         struct nfsd4_blocked_lock, nbl_lru);
5478                 if (nbl->nbl_time > cutoff) {
5479                         t = nbl->nbl_time - cutoff;
5480                         new_timeo = min(new_timeo, t);
5481                         break;
5482                 }
5483                 list_move(&nbl->nbl_lru, &reaplist);
5484                 list_del_init(&nbl->nbl_list);
5485         }
5486         spin_unlock(&nn->blocked_locks_lock);
5487
5488         while (!list_empty(&reaplist)) {
5489                 nbl = list_first_entry(&reaplist,
5490                                         struct nfsd4_blocked_lock, nbl_lru);
5491                 list_del_init(&nbl->nbl_lru);
5492                 free_blocked_lock(nbl);
5493         }
5494 out:
5495         new_timeo = max_t(time64_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
5496         return new_timeo;
5497 }
5498
5499 static struct workqueue_struct *laundry_wq;
5500 static void laundromat_main(struct work_struct *);
5501
5502 static void
5503 laundromat_main(struct work_struct *laundry)
5504 {
5505         time64_t t;
5506         struct delayed_work *dwork = to_delayed_work(laundry);
5507         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
5508                                            laundromat_work);
5509
5510         t = nfs4_laundromat(nn);
5511         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
5512 }
5513
5514 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
5515 {
5516         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
5517                 return nfserr_bad_stateid;
5518         return nfs_ok;
5519 }
5520
5521 static inline int
5522 access_permit_read(struct nfs4_ol_stateid *stp)
5523 {
5524         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
5525                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
5526                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
5527 }
5528
5529 static inline int
5530 access_permit_write(struct nfs4_ol_stateid *stp)
5531 {
5532         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
5533                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
5534 }
5535
5536 static
5537 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
5538 {
5539         __be32 status = nfserr_openmode;
5540
5541         /* For lock stateid's, we test the parent open, not the lock: */
5542         if (stp->st_openstp)
5543                 stp = stp->st_openstp;
5544         if ((flags & WR_STATE) && !access_permit_write(stp))
5545                 goto out;
5546         if ((flags & RD_STATE) && !access_permit_read(stp))
5547                 goto out;
5548         status = nfs_ok;
5549 out:
5550         return status;
5551 }
5552
5553 static inline __be32
5554 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
5555 {
5556         if (ONE_STATEID(stateid) && (flags & RD_STATE))
5557                 return nfs_ok;
5558         else if (opens_in_grace(net)) {
5559                 /* Answer in remaining cases depends on existence of
5560                  * conflicting state; so we must wait out the grace period. */
5561                 return nfserr_grace;
5562         } else if (flags & WR_STATE)
5563                 return nfs4_share_conflict(current_fh,
5564                                 NFS4_SHARE_DENY_WRITE);
5565         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
5566                 return nfs4_share_conflict(current_fh,
5567                                 NFS4_SHARE_DENY_READ);
5568 }
5569
5570 /*
5571  * Allow READ/WRITE during grace period on recovered state only for files
5572  * that are not able to provide mandatory locking.
5573  */
5574 static inline int
5575 grace_disallows_io(struct net *net, struct inode *inode)
5576 {
5577         return opens_in_grace(net) && mandatory_lock(inode);
5578 }
5579
5580 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
5581 {
5582         /*
5583          * When sessions are used the stateid generation number is ignored
5584          * when it is zero.
5585          */
5586         if (has_session && in->si_generation == 0)
5587                 return nfs_ok;
5588
5589         if (in->si_generation == ref->si_generation)
5590                 return nfs_ok;
5591
5592         /* If the client sends us a stateid from the future, it's buggy: */
5593         if (nfsd4_stateid_generation_after(in, ref))
5594                 return nfserr_bad_stateid;
5595         /*
5596          * However, we could see a stateid from the past, even from a
5597          * non-buggy client.  For example, if the client sends a lock
5598          * while some IO is outstanding, the lock may bump si_generation
5599          * while the IO is still in flight.  The client could avoid that
5600          * situation by waiting for responses on all the IO requests,
5601          * but better performance may result in retrying IO that
5602          * receives an old_stateid error if requests are rarely
5603          * reordered in flight:
5604          */
5605         return nfserr_old_stateid;
5606 }
5607
5608 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
5609 {
5610         __be32 ret;
5611
5612         spin_lock(&s->sc_lock);
5613         ret = nfsd4_verify_open_stid(s);
5614         if (ret == nfs_ok)
5615                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
5616         spin_unlock(&s->sc_lock);
5617         return ret;
5618 }
5619
5620 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
5621 {
5622         if (ols->st_stateowner->so_is_open_owner &&
5623             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
5624                 return nfserr_bad_stateid;
5625         return nfs_ok;
5626 }
5627
5628 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
5629 {
5630         struct nfs4_stid *s;
5631         __be32 status = nfserr_bad_stateid;
5632
5633         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5634                 CLOSE_STATEID(stateid))
5635                 return status;
5636         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid))
5637                 return status;
5638         spin_lock(&cl->cl_lock);
5639         s = find_stateid_locked(cl, stateid);
5640         if (!s)
5641                 goto out_unlock;
5642         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
5643         if (status)
5644                 goto out_unlock;
5645         switch (s->sc_type) {
5646         case NFS4_DELEG_STID:
5647                 status = nfs_ok;
5648                 break;
5649         case NFS4_REVOKED_DELEG_STID:
5650                 status = nfserr_deleg_revoked;
5651                 break;
5652         case NFS4_OPEN_STID:
5653         case NFS4_LOCK_STID:
5654                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5655                 break;
5656         default:
5657                 printk("unknown stateid type %x\n", s->sc_type);
5658                 fallthrough;
5659         case NFS4_CLOSED_STID:
5660         case NFS4_CLOSED_DELEG_STID:
5661                 status = nfserr_bad_stateid;
5662         }
5663 out_unlock:
5664         spin_unlock(&cl->cl_lock);
5665         return status;
5666 }
5667
5668 __be32
5669 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5670                      stateid_t *stateid, unsigned char typemask,
5671                      struct nfs4_stid **s, struct nfsd_net *nn)
5672 {
5673         __be32 status;
5674         bool return_revoked = false;
5675
5676         /*
5677          *  only return revoked delegations if explicitly asked.
5678          *  otherwise we report revoked or bad_stateid status.
5679          */
5680         if (typemask & NFS4_REVOKED_DELEG_STID)
5681                 return_revoked = true;
5682         else if (typemask & NFS4_DELEG_STID)
5683                 typemask |= NFS4_REVOKED_DELEG_STID;
5684
5685         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5686                 CLOSE_STATEID(stateid))
5687                 return nfserr_bad_stateid;
5688         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn,
5689                                  false);
5690         if (status == nfserr_stale_clientid) {
5691                 if (cstate->session)
5692                         return nfserr_bad_stateid;
5693                 return nfserr_stale_stateid;
5694         }
5695         if (status)
5696                 return status;
5697         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5698         if (!*s)
5699                 return nfserr_bad_stateid;
5700         if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5701                 nfs4_put_stid(*s);
5702                 if (cstate->minorversion)
5703                         return nfserr_deleg_revoked;
5704                 return nfserr_bad_stateid;
5705         }
5706         return nfs_ok;
5707 }
5708
5709 static struct nfsd_file *
5710 nfs4_find_file(struct nfs4_stid *s, int flags)
5711 {
5712         if (!s)
5713                 return NULL;
5714
5715         switch (s->sc_type) {
5716         case NFS4_DELEG_STID:
5717                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5718                         return NULL;
5719                 return nfsd_file_get(s->sc_file->fi_deleg_file);
5720         case NFS4_OPEN_STID:
5721         case NFS4_LOCK_STID:
5722                 if (flags & RD_STATE)
5723                         return find_readable_file(s->sc_file);
5724                 else
5725                         return find_writeable_file(s->sc_file);
5726         }
5727
5728         return NULL;
5729 }
5730
5731 static __be32
5732 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
5733 {
5734         __be32 status;
5735
5736         status = nfsd4_check_openowner_confirmed(ols);
5737         if (status)
5738                 return status;
5739         return nfs4_check_openmode(ols, flags);
5740 }
5741
5742 static __be32
5743 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5744                 struct nfsd_file **nfp, int flags)
5745 {
5746         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5747         struct nfsd_file *nf;
5748         __be32 status;
5749
5750         nf = nfs4_find_file(s, flags);
5751         if (nf) {
5752                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5753                                 acc | NFSD_MAY_OWNER_OVERRIDE);
5754                 if (status) {
5755                         nfsd_file_put(nf);
5756                         goto out;
5757                 }
5758         } else {
5759                 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
5760                 if (status)
5761                         return status;
5762         }
5763         *nfp = nf;
5764 out:
5765         return status;
5766 }
5767 static void
5768 _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
5769 {
5770         WARN_ON_ONCE(cps->cp_stateid.sc_type != NFS4_COPYNOTIFY_STID);
5771         if (!refcount_dec_and_test(&cps->cp_stateid.sc_count))
5772                 return;
5773         list_del(&cps->cp_list);
5774         idr_remove(&nn->s2s_cp_stateids,
5775                    cps->cp_stateid.stid.si_opaque.so_id);
5776         kfree(cps);
5777 }
5778 /*
5779  * A READ from an inter server to server COPY will have a
5780  * copy stateid. Look up the copy notify stateid from the
5781  * idr structure and take a reference on it.
5782  */
5783 __be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
5784                           struct nfs4_client *clp,
5785                           struct nfs4_cpntf_state **cps)
5786 {
5787         copy_stateid_t *cps_t;
5788         struct nfs4_cpntf_state *state = NULL;
5789
5790         if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
5791                 return nfserr_bad_stateid;
5792         spin_lock(&nn->s2s_cp_lock);
5793         cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
5794         if (cps_t) {
5795                 state = container_of(cps_t, struct nfs4_cpntf_state,
5796                                      cp_stateid);
5797                 if (state->cp_stateid.sc_type != NFS4_COPYNOTIFY_STID) {
5798                         state = NULL;
5799                         goto unlock;
5800                 }
5801                 if (!clp)
5802                         refcount_inc(&state->cp_stateid.sc_count);
5803                 else
5804                         _free_cpntf_state_locked(nn, state);
5805         }
5806 unlock:
5807         spin_unlock(&nn->s2s_cp_lock);
5808         if (!state)
5809                 return nfserr_bad_stateid;
5810         if (!clp && state)
5811                 *cps = state;
5812         return 0;
5813 }
5814
5815 static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
5816                                struct nfs4_stid **stid)
5817 {
5818         __be32 status;
5819         struct nfs4_cpntf_state *cps = NULL;
5820         struct nfsd4_compound_state cstate;
5821
5822         status = manage_cpntf_state(nn, st, NULL, &cps);
5823         if (status)
5824                 return status;
5825
5826         cps->cpntf_time = ktime_get_boottime_seconds();
5827         memset(&cstate, 0, sizeof(cstate));
5828         status = lookup_clientid(&cps->cp_p_clid, &cstate, nn, true);
5829         if (status)
5830                 goto out;
5831         status = nfsd4_lookup_stateid(&cstate, &cps->cp_p_stateid,
5832                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5833                                 stid, nn);
5834         put_client_renew(cstate.clp);
5835 out:
5836         nfs4_put_cpntf_state(nn, cps);
5837         return status;
5838 }
5839
5840 void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
5841 {
5842         spin_lock(&nn->s2s_cp_lock);
5843         _free_cpntf_state_locked(nn, cps);
5844         spin_unlock(&nn->s2s_cp_lock);
5845 }
5846
5847 /*
5848  * Checks for stateid operations
5849  */
5850 __be32
5851 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5852                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5853                 stateid_t *stateid, int flags, struct nfsd_file **nfp,
5854                 struct nfs4_stid **cstid)
5855 {
5856         struct inode *ino = d_inode(fhp->fh_dentry);
5857         struct net *net = SVC_NET(rqstp);
5858         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5859         struct nfs4_stid *s = NULL;
5860         __be32 status;
5861
5862         if (nfp)
5863                 *nfp = NULL;
5864
5865         if (grace_disallows_io(net, ino))
5866                 return nfserr_grace;
5867
5868         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5869                 status = check_special_stateids(net, fhp, stateid, flags);
5870                 goto done;
5871         }
5872
5873         status = nfsd4_lookup_stateid(cstate, stateid,
5874                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5875                                 &s, nn);
5876         if (status == nfserr_bad_stateid)
5877                 status = find_cpntf_state(nn, stateid, &s);
5878         if (status)
5879                 return status;
5880         status = nfsd4_stid_check_stateid_generation(stateid, s,
5881                         nfsd4_has_session(cstate));
5882         if (status)
5883                 goto out;
5884
5885         switch (s->sc_type) {
5886         case NFS4_DELEG_STID:
5887                 status = nfs4_check_delegmode(delegstateid(s), flags);
5888                 break;
5889         case NFS4_OPEN_STID:
5890         case NFS4_LOCK_STID:
5891                 status = nfs4_check_olstateid(openlockstateid(s), flags);
5892                 break;
5893         default:
5894                 status = nfserr_bad_stateid;
5895                 break;
5896         }
5897         if (status)
5898                 goto out;
5899         status = nfs4_check_fh(fhp, s);
5900
5901 done:
5902         if (status == nfs_ok && nfp)
5903                 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
5904 out:
5905         if (s) {
5906                 if (!status && cstid)
5907                         *cstid = s;
5908                 else
5909                         nfs4_put_stid(s);
5910         }
5911         return status;
5912 }
5913
5914 /*
5915  * Test if the stateid is valid
5916  */
5917 __be32
5918 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5919                    union nfsd4_op_u *u)
5920 {
5921         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5922         struct nfsd4_test_stateid_id *stateid;
5923         struct nfs4_client *cl = cstate->session->se_client;
5924
5925         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5926                 stateid->ts_id_status =
5927                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5928
5929         return nfs_ok;
5930 }
5931
5932 static __be32
5933 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5934 {
5935         struct nfs4_ol_stateid *stp = openlockstateid(s);
5936         __be32 ret;
5937
5938         ret = nfsd4_lock_ol_stateid(stp);
5939         if (ret)
5940                 goto out_put_stid;
5941
5942         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5943         if (ret)
5944                 goto out;
5945
5946         ret = nfserr_locks_held;
5947         if (check_for_locks(stp->st_stid.sc_file,
5948                             lockowner(stp->st_stateowner)))
5949                 goto out;
5950
5951         release_lock_stateid(stp);
5952         ret = nfs_ok;
5953
5954 out:
5955         mutex_unlock(&stp->st_mutex);
5956 out_put_stid:
5957         nfs4_put_stid(s);
5958         return ret;
5959 }
5960
5961 __be32
5962 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5963                    union nfsd4_op_u *u)
5964 {
5965         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5966         stateid_t *stateid = &free_stateid->fr_stateid;
5967         struct nfs4_stid *s;
5968         struct nfs4_delegation *dp;
5969         struct nfs4_client *cl = cstate->session->se_client;
5970         __be32 ret = nfserr_bad_stateid;
5971
5972         spin_lock(&cl->cl_lock);
5973         s = find_stateid_locked(cl, stateid);
5974         if (!s)
5975                 goto out_unlock;
5976         spin_lock(&s->sc_lock);
5977         switch (s->sc_type) {
5978         case NFS4_DELEG_STID:
5979                 ret = nfserr_locks_held;
5980                 break;
5981         case NFS4_OPEN_STID:
5982                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5983                 if (ret)
5984                         break;
5985                 ret = nfserr_locks_held;
5986                 break;
5987         case NFS4_LOCK_STID:
5988                 spin_unlock(&s->sc_lock);
5989                 refcount_inc(&s->sc_count);
5990                 spin_unlock(&cl->cl_lock);
5991                 ret = nfsd4_free_lock_stateid(stateid, s);
5992                 goto out;
5993         case NFS4_REVOKED_DELEG_STID:
5994                 spin_unlock(&s->sc_lock);
5995                 dp = delegstateid(s);
5996                 list_del_init(&dp->dl_recall_lru);
5997                 spin_unlock(&cl->cl_lock);
5998                 nfs4_put_stid(s);
5999                 ret = nfs_ok;
6000                 goto out;
6001         /* Default falls through and returns nfserr_bad_stateid */
6002         }
6003         spin_unlock(&s->sc_lock);
6004 out_unlock:
6005         spin_unlock(&cl->cl_lock);
6006 out:
6007         return ret;
6008 }
6009
6010 static inline int
6011 setlkflg (int type)
6012 {
6013         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
6014                 RD_STATE : WR_STATE;
6015 }
6016
6017 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
6018 {
6019         struct svc_fh *current_fh = &cstate->current_fh;
6020         struct nfs4_stateowner *sop = stp->st_stateowner;
6021         __be32 status;
6022
6023         status = nfsd4_check_seqid(cstate, sop, seqid);
6024         if (status)
6025                 return status;
6026         status = nfsd4_lock_ol_stateid(stp);
6027         if (status != nfs_ok)
6028                 return status;
6029         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
6030         if (status == nfs_ok)
6031                 status = nfs4_check_fh(current_fh, &stp->st_stid);
6032         if (status != nfs_ok)
6033                 mutex_unlock(&stp->st_mutex);
6034         return status;
6035 }
6036
6037 /*
6038  * Checks for sequence id mutating operations.
6039  */
6040 static __be32
6041 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
6042                          stateid_t *stateid, char typemask,
6043                          struct nfs4_ol_stateid **stpp,
6044                          struct nfsd_net *nn)
6045 {
6046         __be32 status;
6047         struct nfs4_stid *s;
6048         struct nfs4_ol_stateid *stp = NULL;
6049
6050         trace_nfsd_preprocess(seqid, stateid);
6051
6052         *stpp = NULL;
6053         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
6054         if (status)
6055                 return status;
6056         stp = openlockstateid(s);
6057         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
6058
6059         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
6060         if (!status)
6061                 *stpp = stp;
6062         else
6063                 nfs4_put_stid(&stp->st_stid);
6064         return status;
6065 }
6066
6067 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
6068                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
6069 {
6070         __be32 status;
6071         struct nfs4_openowner *oo;
6072         struct nfs4_ol_stateid *stp;
6073
6074         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
6075                                                 NFS4_OPEN_STID, &stp, nn);
6076         if (status)
6077                 return status;
6078         oo = openowner(stp->st_stateowner);
6079         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
6080                 mutex_unlock(&stp->st_mutex);
6081                 nfs4_put_stid(&stp->st_stid);
6082                 return nfserr_bad_stateid;
6083         }
6084         *stpp = stp;
6085         return nfs_ok;
6086 }
6087
6088 __be32
6089 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6090                    union nfsd4_op_u *u)
6091 {
6092         struct nfsd4_open_confirm *oc = &u->open_confirm;
6093         __be32 status;
6094         struct nfs4_openowner *oo;
6095         struct nfs4_ol_stateid *stp;
6096         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6097
6098         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
6099                         cstate->current_fh.fh_dentry);
6100
6101         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
6102         if (status)
6103                 return status;
6104
6105         status = nfs4_preprocess_seqid_op(cstate,
6106                                         oc->oc_seqid, &oc->oc_req_stateid,
6107                                         NFS4_OPEN_STID, &stp, nn);
6108         if (status)
6109                 goto out;
6110         oo = openowner(stp->st_stateowner);
6111         status = nfserr_bad_stateid;
6112         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
6113                 mutex_unlock(&stp->st_mutex);
6114                 goto put_stateid;
6115         }
6116         oo->oo_flags |= NFS4_OO_CONFIRMED;
6117         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
6118         mutex_unlock(&stp->st_mutex);
6119         trace_nfsd_open_confirm(oc->oc_seqid, &stp->st_stid.sc_stateid);
6120         nfsd4_client_record_create(oo->oo_owner.so_client);
6121         status = nfs_ok;
6122 put_stateid:
6123         nfs4_put_stid(&stp->st_stid);
6124 out:
6125         nfsd4_bump_seqid(cstate, status);
6126         return status;
6127 }
6128
6129 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
6130 {
6131         if (!test_access(access, stp))
6132                 return;
6133         nfs4_file_put_access(stp->st_stid.sc_file, access);
6134         clear_access(access, stp);
6135 }
6136
6137 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
6138 {
6139         switch (to_access) {
6140         case NFS4_SHARE_ACCESS_READ:
6141                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
6142                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6143                 break;
6144         case NFS4_SHARE_ACCESS_WRITE:
6145                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
6146                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6147                 break;
6148         case NFS4_SHARE_ACCESS_BOTH:
6149                 break;
6150         default:
6151                 WARN_ON_ONCE(1);
6152         }
6153 }
6154
6155 __be32
6156 nfsd4_open_downgrade(struct svc_rqst *rqstp,
6157                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
6158 {
6159         struct nfsd4_open_downgrade *od = &u->open_downgrade;
6160         __be32 status;
6161         struct nfs4_ol_stateid *stp;
6162         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6163
6164         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
6165                         cstate->current_fh.fh_dentry);
6166
6167         /* We don't yet support WANT bits: */
6168         if (od->od_deleg_want)
6169                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
6170                         od->od_deleg_want);
6171
6172         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
6173                                         &od->od_stateid, &stp, nn);
6174         if (status)
6175                 goto out;
6176         status = nfserr_inval;
6177         if (!test_access(od->od_share_access, stp)) {
6178                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
6179                         stp->st_access_bmap, od->od_share_access);
6180                 goto put_stateid;
6181         }
6182         if (!test_deny(od->od_share_deny, stp)) {
6183                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
6184                         stp->st_deny_bmap, od->od_share_deny);
6185                 goto put_stateid;
6186         }
6187         nfs4_stateid_downgrade(stp, od->od_share_access);
6188         reset_union_bmap_deny(od->od_share_deny, stp);
6189         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
6190         status = nfs_ok;
6191 put_stateid:
6192         mutex_unlock(&stp->st_mutex);
6193         nfs4_put_stid(&stp->st_stid);
6194 out:
6195         nfsd4_bump_seqid(cstate, status);
6196         return status;
6197 }
6198
6199 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
6200 {
6201         struct nfs4_client *clp = s->st_stid.sc_client;
6202         bool unhashed;
6203         LIST_HEAD(reaplist);
6204
6205         spin_lock(&clp->cl_lock);
6206         unhashed = unhash_open_stateid(s, &reaplist);
6207
6208         if (clp->cl_minorversion) {
6209                 if (unhashed)
6210                         put_ol_stateid_locked(s, &reaplist);
6211                 spin_unlock(&clp->cl_lock);
6212                 free_ol_stateid_reaplist(&reaplist);
6213         } else {
6214                 spin_unlock(&clp->cl_lock);
6215                 free_ol_stateid_reaplist(&reaplist);
6216                 if (unhashed)
6217                         move_to_close_lru(s, clp->net);
6218         }
6219 }
6220
6221 /*
6222  * nfs4_unlock_state() called after encode
6223  */
6224 __be32
6225 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6226                 union nfsd4_op_u *u)
6227 {
6228         struct nfsd4_close *close = &u->close;
6229         __be32 status;
6230         struct nfs4_ol_stateid *stp;
6231         struct net *net = SVC_NET(rqstp);
6232         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6233
6234         dprintk("NFSD: nfsd4_close on file %pd\n",
6235                         cstate->current_fh.fh_dentry);
6236
6237         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
6238                                         &close->cl_stateid,
6239                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
6240                                         &stp, nn);
6241         nfsd4_bump_seqid(cstate, status);
6242         if (status)
6243                 goto out;
6244
6245         stp->st_stid.sc_type = NFS4_CLOSED_STID;
6246
6247         /*
6248          * Technically we don't _really_ have to increment or copy it, since
6249          * it should just be gone after this operation and we clobber the
6250          * copied value below, but we continue to do so here just to ensure
6251          * that racing ops see that there was a state change.
6252          */
6253         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
6254
6255         nfsd4_close_open_stateid(stp);
6256         mutex_unlock(&stp->st_mutex);
6257
6258         /* v4.1+ suggests that we send a special stateid in here, since the
6259          * clients should just ignore this anyway. Since this is not useful
6260          * for v4.0 clients either, we set it to the special close_stateid
6261          * universally.
6262          *
6263          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
6264          */
6265         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
6266
6267         /* put reference from nfs4_preprocess_seqid_op */
6268         nfs4_put_stid(&stp->st_stid);
6269 out:
6270         return status;
6271 }
6272
6273 __be32
6274 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6275                   union nfsd4_op_u *u)
6276 {
6277         struct nfsd4_delegreturn *dr = &u->delegreturn;
6278         struct nfs4_delegation *dp;
6279         stateid_t *stateid = &dr->dr_stateid;
6280         struct nfs4_stid *s;
6281         __be32 status;
6282         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6283
6284         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6285                 return status;
6286
6287         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
6288         if (status)
6289                 goto out;
6290         dp = delegstateid(s);
6291         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
6292         if (status)
6293                 goto put_stateid;
6294
6295         destroy_delegation(dp);
6296 put_stateid:
6297         nfs4_put_stid(&dp->dl_stid);
6298 out:
6299         return status;
6300 }
6301
6302 static inline u64
6303 end_offset(u64 start, u64 len)
6304 {
6305         u64 end;
6306
6307         end = start + len;
6308         return end >= start ? end: NFS4_MAX_UINT64;
6309 }
6310
6311 /* last octet in a range */
6312 static inline u64
6313 last_byte_offset(u64 start, u64 len)
6314 {
6315         u64 end;
6316
6317         WARN_ON_ONCE(!len);
6318         end = start + len;
6319         return end > start ? end - 1: NFS4_MAX_UINT64;
6320 }
6321
6322 /*
6323  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
6324  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
6325  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
6326  * locking, this prevents us from being completely protocol-compliant.  The
6327  * real solution to this problem is to start using unsigned file offsets in
6328  * the VFS, but this is a very deep change!
6329  */
6330 static inline void
6331 nfs4_transform_lock_offset(struct file_lock *lock)
6332 {
6333         if (lock->fl_start < 0)
6334                 lock->fl_start = OFFSET_MAX;
6335         if (lock->fl_end < 0)
6336                 lock->fl_end = OFFSET_MAX;
6337 }
6338
6339 static fl_owner_t
6340 nfsd4_fl_get_owner(fl_owner_t owner)
6341 {
6342         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6343
6344         nfs4_get_stateowner(&lo->lo_owner);
6345         return owner;
6346 }
6347
6348 static void
6349 nfsd4_fl_put_owner(fl_owner_t owner)
6350 {
6351         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6352
6353         if (lo)
6354                 nfs4_put_stateowner(&lo->lo_owner);
6355 }
6356
6357 static void
6358 nfsd4_lm_notify(struct file_lock *fl)
6359 {
6360         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
6361         struct net                      *net = lo->lo_owner.so_client->net;
6362         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
6363         struct nfsd4_blocked_lock       *nbl = container_of(fl,
6364                                                 struct nfsd4_blocked_lock, nbl_lock);
6365         bool queue = false;
6366
6367         /* An empty list means that something else is going to be using it */
6368         spin_lock(&nn->blocked_locks_lock);
6369         if (!list_empty(&nbl->nbl_list)) {
6370                 list_del_init(&nbl->nbl_list);
6371                 list_del_init(&nbl->nbl_lru);
6372                 queue = true;
6373         }
6374         spin_unlock(&nn->blocked_locks_lock);
6375
6376         if (queue)
6377                 nfsd4_run_cb(&nbl->nbl_cb);
6378 }
6379
6380 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
6381         .lm_notify = nfsd4_lm_notify,
6382         .lm_get_owner = nfsd4_fl_get_owner,
6383         .lm_put_owner = nfsd4_fl_put_owner,
6384 };
6385
6386 static inline void
6387 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
6388 {
6389         struct nfs4_lockowner *lo;
6390
6391         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
6392                 lo = (struct nfs4_lockowner *) fl->fl_owner;
6393                 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
6394                                                 GFP_KERNEL);
6395                 if (!deny->ld_owner.data)
6396                         /* We just don't care that much */
6397                         goto nevermind;
6398                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
6399         } else {
6400 nevermind:
6401                 deny->ld_owner.len = 0;
6402                 deny->ld_owner.data = NULL;
6403                 deny->ld_clientid.cl_boot = 0;
6404                 deny->ld_clientid.cl_id = 0;
6405         }
6406         deny->ld_start = fl->fl_start;
6407         deny->ld_length = NFS4_MAX_UINT64;
6408         if (fl->fl_end != NFS4_MAX_UINT64)
6409                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
6410         deny->ld_type = NFS4_READ_LT;
6411         if (fl->fl_type != F_RDLCK)
6412                 deny->ld_type = NFS4_WRITE_LT;
6413 }
6414
6415 static struct nfs4_lockowner *
6416 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
6417 {
6418         unsigned int strhashval = ownerstr_hashval(owner);
6419         struct nfs4_stateowner *so;
6420
6421         lockdep_assert_held(&clp->cl_lock);
6422
6423         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
6424                             so_strhash) {
6425                 if (so->so_is_open_owner)
6426                         continue;
6427                 if (same_owner_str(so, owner))
6428                         return lockowner(nfs4_get_stateowner(so));
6429         }
6430         return NULL;
6431 }
6432
6433 static struct nfs4_lockowner *
6434 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
6435 {
6436         struct nfs4_lockowner *lo;
6437
6438         spin_lock(&clp->cl_lock);
6439         lo = find_lockowner_str_locked(clp, owner);
6440         spin_unlock(&clp->cl_lock);
6441         return lo;
6442 }
6443
6444 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
6445 {
6446         unhash_lockowner_locked(lockowner(sop));
6447 }
6448
6449 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
6450 {
6451         struct nfs4_lockowner *lo = lockowner(sop);
6452
6453         kmem_cache_free(lockowner_slab, lo);
6454 }
6455
6456 static const struct nfs4_stateowner_operations lockowner_ops = {
6457         .so_unhash =    nfs4_unhash_lockowner,
6458         .so_free =      nfs4_free_lockowner,
6459 };
6460
6461 /*
6462  * Alloc a lock owner structure.
6463  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
6464  * occurred.
6465  *
6466  * strhashval = ownerstr_hashval
6467  */
6468 static struct nfs4_lockowner *
6469 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
6470                            struct nfs4_ol_stateid *open_stp,
6471                            struct nfsd4_lock *lock)
6472 {
6473         struct nfs4_lockowner *lo, *ret;
6474
6475         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
6476         if (!lo)
6477                 return NULL;
6478         INIT_LIST_HEAD(&lo->lo_blocked);
6479         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
6480         lo->lo_owner.so_is_open_owner = 0;
6481         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
6482         lo->lo_owner.so_ops = &lockowner_ops;
6483         spin_lock(&clp->cl_lock);
6484         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
6485         if (ret == NULL) {
6486                 list_add(&lo->lo_owner.so_strhash,
6487                          &clp->cl_ownerstr_hashtbl[strhashval]);
6488                 ret = lo;
6489         } else
6490                 nfs4_free_stateowner(&lo->lo_owner);
6491
6492         spin_unlock(&clp->cl_lock);
6493         return ret;
6494 }
6495
6496 static struct nfs4_ol_stateid *
6497 find_lock_stateid(const struct nfs4_lockowner *lo,
6498                   const struct nfs4_ol_stateid *ost)
6499 {
6500         struct nfs4_ol_stateid *lst;
6501
6502         lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
6503
6504         /* If ost is not hashed, ost->st_locks will not be valid */
6505         if (!nfs4_ol_stateid_unhashed(ost))
6506                 list_for_each_entry(lst, &ost->st_locks, st_locks) {
6507                         if (lst->st_stateowner == &lo->lo_owner) {
6508                                 refcount_inc(&lst->st_stid.sc_count);
6509                                 return lst;
6510                         }
6511                 }
6512         return NULL;
6513 }
6514
6515 static struct nfs4_ol_stateid *
6516 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
6517                   struct nfs4_file *fp, struct inode *inode,
6518                   struct nfs4_ol_stateid *open_stp)
6519 {
6520         struct nfs4_client *clp = lo->lo_owner.so_client;
6521         struct nfs4_ol_stateid *retstp;
6522
6523         mutex_init(&stp->st_mutex);
6524         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
6525 retry:
6526         spin_lock(&clp->cl_lock);
6527         if (nfs4_ol_stateid_unhashed(open_stp))
6528                 goto out_close;
6529         retstp = find_lock_stateid(lo, open_stp);
6530         if (retstp)
6531                 goto out_found;
6532         refcount_inc(&stp->st_stid.sc_count);
6533         stp->st_stid.sc_type = NFS4_LOCK_STID;
6534         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
6535         get_nfs4_file(fp);
6536         stp->st_stid.sc_file = fp;
6537         stp->st_access_bmap = 0;
6538         stp->st_deny_bmap = open_stp->st_deny_bmap;
6539         stp->st_openstp = open_stp;
6540         spin_lock(&fp->fi_lock);
6541         list_add(&stp->st_locks, &open_stp->st_locks);
6542         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
6543         list_add(&stp->st_perfile, &fp->fi_stateids);
6544         spin_unlock(&fp->fi_lock);
6545         spin_unlock(&clp->cl_lock);
6546         return stp;
6547 out_found:
6548         spin_unlock(&clp->cl_lock);
6549         if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
6550                 nfs4_put_stid(&retstp->st_stid);
6551                 goto retry;
6552         }
6553         /* To keep mutex tracking happy */
6554         mutex_unlock(&stp->st_mutex);
6555         return retstp;
6556 out_close:
6557         spin_unlock(&clp->cl_lock);
6558         mutex_unlock(&stp->st_mutex);
6559         return NULL;
6560 }
6561
6562 static struct nfs4_ol_stateid *
6563 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
6564                             struct inode *inode, struct nfs4_ol_stateid *ost,
6565                             bool *new)
6566 {
6567         struct nfs4_stid *ns = NULL;
6568         struct nfs4_ol_stateid *lst;
6569         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6570         struct nfs4_client *clp = oo->oo_owner.so_client;
6571
6572         *new = false;
6573         spin_lock(&clp->cl_lock);
6574         lst = find_lock_stateid(lo, ost);
6575         spin_unlock(&clp->cl_lock);
6576         if (lst != NULL) {
6577                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
6578                         goto out;
6579                 nfs4_put_stid(&lst->st_stid);
6580         }
6581         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
6582         if (ns == NULL)
6583                 return NULL;
6584
6585         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
6586         if (lst == openlockstateid(ns))
6587                 *new = true;
6588         else
6589                 nfs4_put_stid(ns);
6590 out:
6591         return lst;
6592 }
6593
6594 static int
6595 check_lock_length(u64 offset, u64 length)
6596 {
6597         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
6598                 (length > ~offset)));
6599 }
6600
6601 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
6602 {
6603         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
6604
6605         lockdep_assert_held(&fp->fi_lock);
6606
6607         if (test_access(access, lock_stp))
6608                 return;
6609         __nfs4_file_get_access(fp, access);
6610         set_access(access, lock_stp);
6611 }
6612
6613 static __be32
6614 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
6615                             struct nfs4_ol_stateid *ost,
6616                             struct nfsd4_lock *lock,
6617                             struct nfs4_ol_stateid **plst, bool *new)
6618 {
6619         __be32 status;
6620         struct nfs4_file *fi = ost->st_stid.sc_file;
6621         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6622         struct nfs4_client *cl = oo->oo_owner.so_client;
6623         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
6624         struct nfs4_lockowner *lo;
6625         struct nfs4_ol_stateid *lst;
6626         unsigned int strhashval;
6627
6628         lo = find_lockowner_str(cl, &lock->lk_new_owner);
6629         if (!lo) {
6630                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
6631                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
6632                 if (lo == NULL)
6633                         return nfserr_jukebox;
6634         } else {
6635                 /* with an existing lockowner, seqids must be the same */
6636                 status = nfserr_bad_seqid;
6637                 if (!cstate->minorversion &&
6638                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
6639                         goto out;
6640         }
6641
6642         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
6643         if (lst == NULL) {
6644                 status = nfserr_jukebox;
6645                 goto out;
6646         }
6647
6648         status = nfs_ok;
6649         *plst = lst;
6650 out:
6651         nfs4_put_stateowner(&lo->lo_owner);
6652         return status;
6653 }
6654
6655 /*
6656  *  LOCK operation
6657  */
6658 __be32
6659 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6660            union nfsd4_op_u *u)
6661 {
6662         struct nfsd4_lock *lock = &u->lock;
6663         struct nfs4_openowner *open_sop = NULL;
6664         struct nfs4_lockowner *lock_sop = NULL;
6665         struct nfs4_ol_stateid *lock_stp = NULL;
6666         struct nfs4_ol_stateid *open_stp = NULL;
6667         struct nfs4_file *fp;
6668         struct nfsd_file *nf = NULL;
6669         struct nfsd4_blocked_lock *nbl = NULL;
6670         struct file_lock *file_lock = NULL;
6671         struct file_lock *conflock = NULL;
6672         __be32 status = 0;
6673         int lkflg;
6674         int err;
6675         bool new = false;
6676         unsigned char fl_type;
6677         unsigned int fl_flags = FL_POSIX;
6678         struct net *net = SVC_NET(rqstp);
6679         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6680
6681         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
6682                 (long long) lock->lk_offset,
6683                 (long long) lock->lk_length);
6684
6685         if (check_lock_length(lock->lk_offset, lock->lk_length))
6686                  return nfserr_inval;
6687
6688         if ((status = fh_verify(rqstp, &cstate->current_fh,
6689                                 S_IFREG, NFSD_MAY_LOCK))) {
6690                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
6691                 return status;
6692         }
6693
6694         if (lock->lk_is_new) {
6695                 if (nfsd4_has_session(cstate))
6696                         /* See rfc 5661 18.10.3: given clientid is ignored: */
6697                         memcpy(&lock->lk_new_clientid,
6698                                 &cstate->session->se_client->cl_clientid,
6699                                 sizeof(clientid_t));
6700
6701                 status = nfserr_stale_clientid;
6702                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
6703                         goto out;
6704
6705                 /* validate and update open stateid and open seqid */
6706                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
6707                                         lock->lk_new_open_seqid,
6708                                         &lock->lk_new_open_stateid,
6709                                         &open_stp, nn);
6710                 if (status)
6711                         goto out;
6712                 mutex_unlock(&open_stp->st_mutex);
6713                 open_sop = openowner(open_stp->st_stateowner);
6714                 status = nfserr_bad_stateid;
6715                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
6716                                                 &lock->lk_new_clientid))
6717                         goto out;
6718                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
6719                                                         &lock_stp, &new);
6720         } else {
6721                 status = nfs4_preprocess_seqid_op(cstate,
6722                                        lock->lk_old_lock_seqid,
6723                                        &lock->lk_old_lock_stateid,
6724                                        NFS4_LOCK_STID, &lock_stp, nn);
6725         }
6726         if (status)
6727                 goto out;
6728         lock_sop = lockowner(lock_stp->st_stateowner);
6729
6730         lkflg = setlkflg(lock->lk_type);
6731         status = nfs4_check_openmode(lock_stp, lkflg);
6732         if (status)
6733                 goto out;
6734
6735         status = nfserr_grace;
6736         if (locks_in_grace(net) && !lock->lk_reclaim)
6737                 goto out;
6738         status = nfserr_no_grace;
6739         if (!locks_in_grace(net) && lock->lk_reclaim)
6740                 goto out;
6741
6742         fp = lock_stp->st_stid.sc_file;
6743         switch (lock->lk_type) {
6744                 case NFS4_READW_LT:
6745                         if (nfsd4_has_session(cstate))
6746                                 fl_flags |= FL_SLEEP;
6747                         fallthrough;
6748                 case NFS4_READ_LT:
6749                         spin_lock(&fp->fi_lock);
6750                         nf = find_readable_file_locked(fp);
6751                         if (nf)
6752                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
6753                         spin_unlock(&fp->fi_lock);
6754                         fl_type = F_RDLCK;
6755                         break;
6756                 case NFS4_WRITEW_LT:
6757                         if (nfsd4_has_session(cstate))
6758                                 fl_flags |= FL_SLEEP;
6759                         fallthrough;
6760                 case NFS4_WRITE_LT:
6761                         spin_lock(&fp->fi_lock);
6762                         nf = find_writeable_file_locked(fp);
6763                         if (nf)
6764                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6765                         spin_unlock(&fp->fi_lock);
6766                         fl_type = F_WRLCK;
6767                         break;
6768                 default:
6769                         status = nfserr_inval;
6770                 goto out;
6771         }
6772
6773         if (!nf) {
6774                 status = nfserr_openmode;
6775                 goto out;
6776         }
6777
6778         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6779         if (!nbl) {
6780                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6781                 status = nfserr_jukebox;
6782                 goto out;
6783         }
6784
6785         file_lock = &nbl->nbl_lock;
6786         file_lock->fl_type = fl_type;
6787         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6788         file_lock->fl_pid = current->tgid;
6789         file_lock->fl_file = nf->nf_file;
6790         file_lock->fl_flags = fl_flags;
6791         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6792         file_lock->fl_start = lock->lk_offset;
6793         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6794         nfs4_transform_lock_offset(file_lock);
6795
6796         conflock = locks_alloc_lock();
6797         if (!conflock) {
6798                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6799                 status = nfserr_jukebox;
6800                 goto out;
6801         }
6802
6803         if (fl_flags & FL_SLEEP) {
6804                 nbl->nbl_time = ktime_get_boottime_seconds();
6805                 spin_lock(&nn->blocked_locks_lock);
6806                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6807                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6808                 spin_unlock(&nn->blocked_locks_lock);
6809         }
6810
6811         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
6812         switch (err) {
6813         case 0: /* success! */
6814                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6815                 status = 0;
6816                 if (lock->lk_reclaim)
6817                         nn->somebody_reclaimed = true;
6818                 break;
6819         case FILE_LOCK_DEFERRED:
6820                 nbl = NULL;
6821                 fallthrough;
6822         case -EAGAIN:           /* conflock holds conflicting lock */
6823                 status = nfserr_denied;
6824                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6825                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6826                 break;
6827         case -EDEADLK:
6828                 status = nfserr_deadlock;
6829                 break;
6830         default:
6831                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6832                 status = nfserrno(err);
6833                 break;
6834         }
6835 out:
6836         if (nbl) {
6837                 /* dequeue it if we queued it before */
6838                 if (fl_flags & FL_SLEEP) {
6839                         spin_lock(&nn->blocked_locks_lock);
6840                         list_del_init(&nbl->nbl_list);
6841                         list_del_init(&nbl->nbl_lru);
6842                         spin_unlock(&nn->blocked_locks_lock);
6843                 }
6844                 free_blocked_lock(nbl);
6845         }
6846         if (nf)
6847                 nfsd_file_put(nf);
6848         if (lock_stp) {
6849                 /* Bump seqid manually if the 4.0 replay owner is openowner */
6850                 if (cstate->replay_owner &&
6851                     cstate->replay_owner != &lock_sop->lo_owner &&
6852                     seqid_mutating_err(ntohl(status)))
6853                         lock_sop->lo_owner.so_seqid++;
6854
6855                 /*
6856                  * If this is a new, never-before-used stateid, and we are
6857                  * returning an error, then just go ahead and release it.
6858                  */
6859                 if (status && new)
6860                         release_lock_stateid(lock_stp);
6861
6862                 mutex_unlock(&lock_stp->st_mutex);
6863
6864                 nfs4_put_stid(&lock_stp->st_stid);
6865         }
6866         if (open_stp)
6867                 nfs4_put_stid(&open_stp->st_stid);
6868         nfsd4_bump_seqid(cstate, status);
6869         if (conflock)
6870                 locks_free_lock(conflock);
6871         return status;
6872 }
6873
6874 /*
6875  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6876  * so we do a temporary open here just to get an open file to pass to
6877  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
6878  * inode operation.)
6879  */
6880 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6881 {
6882         struct nfsd_file *nf;
6883         __be32 err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
6884         if (!err) {
6885                 err = nfserrno(vfs_test_lock(nf->nf_file, lock));
6886                 nfsd_file_put(nf);
6887         }
6888         return err;
6889 }
6890
6891 /*
6892  * LOCKT operation
6893  */
6894 __be32
6895 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6896             union nfsd4_op_u *u)
6897 {
6898         struct nfsd4_lockt *lockt = &u->lockt;
6899         struct file_lock *file_lock = NULL;
6900         struct nfs4_lockowner *lo = NULL;
6901         __be32 status;
6902         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6903
6904         if (locks_in_grace(SVC_NET(rqstp)))
6905                 return nfserr_grace;
6906
6907         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6908                  return nfserr_inval;
6909
6910         if (!nfsd4_has_session(cstate)) {
6911                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn,
6912                                          false);
6913                 if (status)
6914                         goto out;
6915         }
6916
6917         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6918                 goto out;
6919
6920         file_lock = locks_alloc_lock();
6921         if (!file_lock) {
6922                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6923                 status = nfserr_jukebox;
6924                 goto out;
6925         }
6926
6927         switch (lockt->lt_type) {
6928                 case NFS4_READ_LT:
6929                 case NFS4_READW_LT:
6930                         file_lock->fl_type = F_RDLCK;
6931                         break;
6932                 case NFS4_WRITE_LT:
6933                 case NFS4_WRITEW_LT:
6934                         file_lock->fl_type = F_WRLCK;
6935                         break;
6936                 default:
6937                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6938                         status = nfserr_inval;
6939                         goto out;
6940         }
6941
6942         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6943         if (lo)
6944                 file_lock->fl_owner = (fl_owner_t)lo;
6945         file_lock->fl_pid = current->tgid;
6946         file_lock->fl_flags = FL_POSIX;
6947
6948         file_lock->fl_start = lockt->lt_offset;
6949         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6950
6951         nfs4_transform_lock_offset(file_lock);
6952
6953         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6954         if (status)
6955                 goto out;
6956
6957         if (file_lock->fl_type != F_UNLCK) {
6958                 status = nfserr_denied;
6959                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6960         }
6961 out:
6962         if (lo)
6963                 nfs4_put_stateowner(&lo->lo_owner);
6964         if (file_lock)
6965                 locks_free_lock(file_lock);
6966         return status;
6967 }
6968
6969 __be32
6970 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6971             union nfsd4_op_u *u)
6972 {
6973         struct nfsd4_locku *locku = &u->locku;
6974         struct nfs4_ol_stateid *stp;
6975         struct nfsd_file *nf = NULL;
6976         struct file_lock *file_lock = NULL;
6977         __be32 status;
6978         int err;
6979         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6980
6981         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6982                 (long long) locku->lu_offset,
6983                 (long long) locku->lu_length);
6984
6985         if (check_lock_length(locku->lu_offset, locku->lu_length))
6986                  return nfserr_inval;
6987
6988         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6989                                         &locku->lu_stateid, NFS4_LOCK_STID,
6990                                         &stp, nn);
6991         if (status)
6992                 goto out;
6993         nf = find_any_file(stp->st_stid.sc_file);
6994         if (!nf) {
6995                 status = nfserr_lock_range;
6996                 goto put_stateid;
6997         }
6998         file_lock = locks_alloc_lock();
6999         if (!file_lock) {
7000                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7001                 status = nfserr_jukebox;
7002                 goto put_file;
7003         }
7004
7005         file_lock->fl_type = F_UNLCK;
7006         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
7007         file_lock->fl_pid = current->tgid;
7008         file_lock->fl_file = nf->nf_file;
7009         file_lock->fl_flags = FL_POSIX;
7010         file_lock->fl_lmops = &nfsd_posix_mng_ops;
7011         file_lock->fl_start = locku->lu_offset;
7012
7013         file_lock->fl_end = last_byte_offset(locku->lu_offset,
7014                                                 locku->lu_length);
7015         nfs4_transform_lock_offset(file_lock);
7016
7017         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
7018         if (err) {
7019                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
7020                 goto out_nfserr;
7021         }
7022         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
7023 put_file:
7024         nfsd_file_put(nf);
7025 put_stateid:
7026         mutex_unlock(&stp->st_mutex);
7027         nfs4_put_stid(&stp->st_stid);
7028 out:
7029         nfsd4_bump_seqid(cstate, status);
7030         if (file_lock)
7031                 locks_free_lock(file_lock);
7032         return status;
7033
7034 out_nfserr:
7035         status = nfserrno(err);
7036         goto put_file;
7037 }
7038
7039 /*
7040  * returns
7041  *      true:  locks held by lockowner
7042  *      false: no locks held by lockowner
7043  */
7044 static bool
7045 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
7046 {
7047         struct file_lock *fl;
7048         int status = false;
7049         struct nfsd_file *nf = find_any_file(fp);
7050         struct inode *inode;
7051         struct file_lock_context *flctx;
7052
7053         if (!nf) {
7054                 /* Any valid lock stateid should have some sort of access */
7055                 WARN_ON_ONCE(1);
7056                 return status;
7057         }
7058
7059         inode = locks_inode(nf->nf_file);
7060         flctx = inode->i_flctx;
7061
7062         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
7063                 spin_lock(&flctx->flc_lock);
7064                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
7065                         if (fl->fl_owner == (fl_owner_t)lowner) {
7066                                 status = true;
7067                                 break;
7068                         }
7069                 }
7070                 spin_unlock(&flctx->flc_lock);
7071         }
7072         nfsd_file_put(nf);
7073         return status;
7074 }
7075
7076 __be32
7077 nfsd4_release_lockowner(struct svc_rqst *rqstp,
7078                         struct nfsd4_compound_state *cstate,
7079                         union nfsd4_op_u *u)
7080 {
7081         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
7082         clientid_t *clid = &rlockowner->rl_clientid;
7083         struct nfs4_stateowner *sop;
7084         struct nfs4_lockowner *lo = NULL;
7085         struct nfs4_ol_stateid *stp;
7086         struct xdr_netobj *owner = &rlockowner->rl_owner;
7087         unsigned int hashval = ownerstr_hashval(owner);
7088         __be32 status;
7089         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7090         struct nfs4_client *clp;
7091         LIST_HEAD (reaplist);
7092
7093         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
7094                 clid->cl_boot, clid->cl_id);
7095
7096         status = lookup_clientid(clid, cstate, nn, false);
7097         if (status)
7098                 return status;
7099
7100         clp = cstate->clp;
7101         /* Find the matching lock stateowner */
7102         spin_lock(&clp->cl_lock);
7103         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
7104                             so_strhash) {
7105
7106                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
7107                         continue;
7108
7109                 /* see if there are still any locks associated with it */
7110                 lo = lockowner(sop);
7111                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
7112                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
7113                                 status = nfserr_locks_held;
7114                                 spin_unlock(&clp->cl_lock);
7115                                 return status;
7116                         }
7117                 }
7118
7119                 nfs4_get_stateowner(sop);
7120                 break;
7121         }
7122         if (!lo) {
7123                 spin_unlock(&clp->cl_lock);
7124                 return status;
7125         }
7126
7127         unhash_lockowner_locked(lo);
7128         while (!list_empty(&lo->lo_owner.so_stateids)) {
7129                 stp = list_first_entry(&lo->lo_owner.so_stateids,
7130                                        struct nfs4_ol_stateid,
7131                                        st_perstateowner);
7132                 WARN_ON(!unhash_lock_stateid(stp));
7133                 put_ol_stateid_locked(stp, &reaplist);
7134         }
7135         spin_unlock(&clp->cl_lock);
7136         free_ol_stateid_reaplist(&reaplist);
7137         remove_blocked_locks(lo);
7138         nfs4_put_stateowner(&lo->lo_owner);
7139
7140         return status;
7141 }
7142
7143 static inline struct nfs4_client_reclaim *
7144 alloc_reclaim(void)
7145 {
7146         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
7147 }
7148
7149 bool
7150 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
7151 {
7152         struct nfs4_client_reclaim *crp;
7153
7154         crp = nfsd4_find_reclaim_client(name, nn);
7155         return (crp && crp->cr_clp);
7156 }
7157
7158 /*
7159  * failure => all reset bets are off, nfserr_no_grace...
7160  *
7161  * The caller is responsible for freeing name.data if NULL is returned (it
7162  * will be freed in nfs4_remove_reclaim_record in the normal case).
7163  */
7164 struct nfs4_client_reclaim *
7165 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
7166                 struct nfsd_net *nn)
7167 {
7168         unsigned int strhashval;
7169         struct nfs4_client_reclaim *crp;
7170
7171         trace_nfsd_clid_reclaim(nn, name.len, name.data);
7172         crp = alloc_reclaim();
7173         if (crp) {
7174                 strhashval = clientstr_hashval(name);
7175                 INIT_LIST_HEAD(&crp->cr_strhash);
7176                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
7177                 crp->cr_name.data = name.data;
7178                 crp->cr_name.len = name.len;
7179                 crp->cr_princhash.data = princhash.data;
7180                 crp->cr_princhash.len = princhash.len;
7181                 crp->cr_clp = NULL;
7182                 nn->reclaim_str_hashtbl_size++;
7183         }
7184         return crp;
7185 }
7186
7187 void
7188 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
7189 {
7190         list_del(&crp->cr_strhash);
7191         kfree(crp->cr_name.data);
7192         kfree(crp->cr_princhash.data);
7193         kfree(crp);
7194         nn->reclaim_str_hashtbl_size--;
7195 }
7196
7197 void
7198 nfs4_release_reclaim(struct nfsd_net *nn)
7199 {
7200         struct nfs4_client_reclaim *crp = NULL;
7201         int i;
7202
7203         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7204                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
7205                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
7206                                         struct nfs4_client_reclaim, cr_strhash);
7207                         nfs4_remove_reclaim_record(crp, nn);
7208                 }
7209         }
7210         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
7211 }
7212
7213 /*
7214  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
7215 struct nfs4_client_reclaim *
7216 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
7217 {
7218         unsigned int strhashval;
7219         struct nfs4_client_reclaim *crp = NULL;
7220
7221         trace_nfsd_clid_find(nn, name.len, name.data);
7222
7223         strhashval = clientstr_hashval(name);
7224         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
7225                 if (compare_blob(&crp->cr_name, &name) == 0) {
7226                         return crp;
7227                 }
7228         }
7229         return NULL;
7230 }
7231
7232 /*
7233 * Called from OPEN. Look for clientid in reclaim list.
7234 */
7235 __be32
7236 nfs4_check_open_reclaim(clientid_t *clid,
7237                 struct nfsd4_compound_state *cstate,
7238                 struct nfsd_net *nn)
7239 {
7240         __be32 status;
7241
7242         /* find clientid in conf_id_hashtbl */
7243         status = lookup_clientid(clid, cstate, nn, false);
7244         if (status)
7245                 return nfserr_reclaim_bad;
7246
7247         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
7248                 return nfserr_no_grace;
7249
7250         if (nfsd4_client_record_check(cstate->clp))
7251                 return nfserr_reclaim_bad;
7252
7253         return nfs_ok;
7254 }
7255
7256 /*
7257  * Since the lifetime of a delegation isn't limited to that of an open, a
7258  * client may quite reasonably hang on to a delegation as long as it has
7259  * the inode cached.  This becomes an obvious problem the first time a
7260  * client's inode cache approaches the size of the server's total memory.
7261  *
7262  * For now we avoid this problem by imposing a hard limit on the number
7263  * of delegations, which varies according to the server's memory size.
7264  */
7265 static void
7266 set_max_delegations(void)
7267 {
7268         /*
7269          * Allow at most 4 delegations per megabyte of RAM.  Quick
7270          * estimates suggest that in the worst case (where every delegation
7271          * is for a different inode), a delegation could take about 1.5K,
7272          * giving a worst case usage of about 6% of memory.
7273          */
7274         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7275 }
7276
7277 static int nfs4_state_create_net(struct net *net)
7278 {
7279         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7280         int i;
7281
7282         nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7283                                             sizeof(struct list_head),
7284                                             GFP_KERNEL);
7285         if (!nn->conf_id_hashtbl)
7286                 goto err;
7287         nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7288                                               sizeof(struct list_head),
7289                                               GFP_KERNEL);
7290         if (!nn->unconf_id_hashtbl)
7291                 goto err_unconf_id;
7292         nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7293                                               sizeof(struct list_head),
7294                                               GFP_KERNEL);
7295         if (!nn->sessionid_hashtbl)
7296                 goto err_sessionid;
7297
7298         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7299                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7300                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7301         }
7302         for (i = 0; i < SESSION_HASH_SIZE; i++)
7303                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7304         nn->conf_name_tree = RB_ROOT;
7305         nn->unconf_name_tree = RB_ROOT;
7306         nn->boot_time = ktime_get_real_seconds();
7307         nn->grace_ended = false;
7308         nn->nfsd4_manager.block_opens = true;
7309         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7310         INIT_LIST_HEAD(&nn->client_lru);
7311         INIT_LIST_HEAD(&nn->close_lru);
7312         INIT_LIST_HEAD(&nn->del_recall_lru);
7313         spin_lock_init(&nn->client_lock);
7314         spin_lock_init(&nn->s2s_cp_lock);
7315         idr_init(&nn->s2s_cp_stateids);
7316
7317         spin_lock_init(&nn->blocked_locks_lock);
7318         INIT_LIST_HEAD(&nn->blocked_locks_lru);
7319
7320         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7321         get_net(net);
7322
7323         return 0;
7324
7325 err_sessionid:
7326         kfree(nn->unconf_id_hashtbl);
7327 err_unconf_id:
7328         kfree(nn->conf_id_hashtbl);
7329 err:
7330         return -ENOMEM;
7331 }
7332
7333 static void
7334 nfs4_state_destroy_net(struct net *net)
7335 {
7336         int i;
7337         struct nfs4_client *clp = NULL;
7338         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7339
7340         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7341                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7342                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7343                         destroy_client(clp);
7344                 }
7345         }
7346
7347         WARN_ON(!list_empty(&nn->blocked_locks_lru));
7348
7349         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7350                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7351                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7352                         destroy_client(clp);
7353                 }
7354         }
7355
7356         kfree(nn->sessionid_hashtbl);
7357         kfree(nn->unconf_id_hashtbl);
7358         kfree(nn->conf_id_hashtbl);
7359         put_net(net);
7360 }
7361
7362 int
7363 nfs4_state_start_net(struct net *net)
7364 {
7365         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7366         int ret;
7367
7368         ret = get_nfsdfs(net);
7369         if (ret)
7370                 return ret;
7371         ret = nfs4_state_create_net(net);
7372         if (ret) {
7373                 mntput(nn->nfsd_mnt);
7374                 return ret;
7375         }
7376         locks_start_grace(net, &nn->nfsd4_manager);
7377         nfsd4_client_tracking_init(net);
7378         if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
7379                 goto skip_grace;
7380         printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
7381                nn->nfsd4_grace, net->ns.inum);
7382         trace_nfsd_grace_start(nn);
7383         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7384         return 0;
7385
7386 skip_grace:
7387         printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
7388                         net->ns.inum);
7389         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
7390         nfsd4_end_grace(nn);
7391         return 0;
7392 }
7393
7394 /* initialization to perform when the nfsd service is started: */
7395
7396 int
7397 nfs4_state_start(void)
7398 {
7399         int ret;
7400
7401         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7402         if (laundry_wq == NULL) {
7403                 ret = -ENOMEM;
7404                 goto out;
7405         }
7406         ret = nfsd4_create_callback_queue();
7407         if (ret)
7408                 goto out_free_laundry;
7409
7410         set_max_delegations();
7411         return 0;
7412
7413 out_free_laundry:
7414         destroy_workqueue(laundry_wq);
7415 out:
7416         return ret;
7417 }
7418
7419 void
7420 nfs4_state_shutdown_net(struct net *net)
7421 {
7422         struct nfs4_delegation *dp = NULL;
7423         struct list_head *pos, *next, reaplist;
7424         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7425
7426         cancel_delayed_work_sync(&nn->laundromat_work);
7427         locks_end_grace(&nn->nfsd4_manager);
7428
7429         INIT_LIST_HEAD(&reaplist);
7430         spin_lock(&state_lock);
7431         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7432                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7433                 WARN_ON(!unhash_delegation_locked(dp));
7434                 list_add(&dp->dl_recall_lru, &reaplist);
7435         }
7436         spin_unlock(&state_lock);
7437         list_for_each_safe(pos, next, &reaplist) {
7438                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7439                 list_del_init(&dp->dl_recall_lru);
7440                 destroy_unhashed_deleg(dp);
7441         }
7442
7443         nfsd4_client_tracking_exit(net);
7444         nfs4_state_destroy_net(net);
7445         mntput(nn->nfsd_mnt);
7446 }
7447
7448 void
7449 nfs4_state_shutdown(void)
7450 {
7451         destroy_workqueue(laundry_wq);
7452         nfsd4_destroy_callback_queue();
7453 }
7454
7455 static void
7456 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7457 {
7458         if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
7459             CURRENT_STATEID(stateid))
7460                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7461 }
7462
7463 static void
7464 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7465 {
7466         if (cstate->minorversion) {
7467                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7468                 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
7469         }
7470 }
7471
7472 void
7473 clear_current_stateid(struct nfsd4_compound_state *cstate)
7474 {
7475         CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
7476 }
7477
7478 /*
7479  * functions to set current state id
7480  */
7481 void
7482 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7483                 union nfsd4_op_u *u)
7484 {
7485         put_stateid(cstate, &u->open_downgrade.od_stateid);
7486 }
7487
7488 void
7489 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7490                 union nfsd4_op_u *u)
7491 {
7492         put_stateid(cstate, &u->open.op_stateid);
7493 }
7494
7495 void
7496 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7497                 union nfsd4_op_u *u)
7498 {
7499         put_stateid(cstate, &u->close.cl_stateid);
7500 }
7501
7502 void
7503 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7504                 union nfsd4_op_u *u)
7505 {
7506         put_stateid(cstate, &u->lock.lk_resp_stateid);
7507 }
7508
7509 /*
7510  * functions to consume current state id
7511  */
7512
7513 void
7514 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7515                 union nfsd4_op_u *u)
7516 {
7517         get_stateid(cstate, &u->open_downgrade.od_stateid);
7518 }
7519
7520 void
7521 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7522                 union nfsd4_op_u *u)
7523 {
7524         get_stateid(cstate, &u->delegreturn.dr_stateid);
7525 }
7526
7527 void
7528 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7529                 union nfsd4_op_u *u)
7530 {
7531         get_stateid(cstate, &u->free_stateid.fr_stateid);
7532 }
7533
7534 void
7535 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7536                 union nfsd4_op_u *u)
7537 {
7538         get_stateid(cstate, &u->setattr.sa_stateid);
7539 }
7540
7541 void
7542 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7543                 union nfsd4_op_u *u)
7544 {
7545         get_stateid(cstate, &u->close.cl_stateid);
7546 }
7547
7548 void
7549 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7550                 union nfsd4_op_u *u)
7551 {
7552         get_stateid(cstate, &u->locku.lu_stateid);
7553 }
7554
7555 void
7556 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7557                 union nfsd4_op_u *u)
7558 {
7559         get_stateid(cstate, &u->read.rd_stateid);
7560 }
7561
7562 void
7563 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7564                 union nfsd4_op_u *u)
7565 {
7566         get_stateid(cstate, &u->write.wr_stateid);
7567 }