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
  55 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  56
  57 #define all_ones {{~0,~0},~0}
  58 static const stateid_t one_stateid = {
  59         .si_generation = ~0,
  60         .si_opaque = all_ones,
  61 };
  62 static const stateid_t zero_stateid = {
  63         /* all fields zero */
  64 };
  65 static const stateid_t currentstateid = {
  66         .si_generation = 1,
  67 };
  68 static const stateid_t close_stateid = {
  69         .si_generation = 0xffffffffU,
  70 };
  71
  72 static u64 current_sessionid = 1;
  73
  74 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  75 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  76 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  77 #define CLOSE_STATEID(stateid)  (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
  78
  79 /* forward declarations */
  80 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  81 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  82 void nfsd4_end_grace(struct nfsd_net *nn);
  83 static void _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps);
  84
  85 /* Locking: */
  86
  87 /*
  88  * Currently used for the del_recall_lru and file hash table.  In an
  89  * effort to decrease the scope of the client_mutex, this spinlock may
  90  * eventually cover more:
  91  */
  92 static DEFINE_SPINLOCK(state_lock);
  93
  94 enum nfsd4_st_mutex_lock_subclass {
  95         OPEN_STATEID_MUTEX = 0,
  96         LOCK_STATEID_MUTEX = 1,
  97 };
  98
  99 /*
 100  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
 101  * the refcount on the open stateid to drop.
 102  */
 103 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
 104
 105 /*
 106  * A waitqueue where a writer to clients/#/ctl destroying a client can
 107  * wait for cl_rpc_users to drop to 0 and then for the client to be
 108  * unhashed.
 109  */
 110 static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
 111
 112 static struct kmem_cache *client_slab;
 113 static struct kmem_cache *openowner_slab;
 114 static struct kmem_cache *lockowner_slab;
 115 static struct kmem_cache *file_slab;
 116 static struct kmem_cache *stateid_slab;
 117 static struct kmem_cache *deleg_slab;
 118 static struct kmem_cache *odstate_slab;
 119
 120 static void free_session(struct nfsd4_session *);
 121
 122 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 123 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
 124
 125 static bool is_session_dead(struct nfsd4_session *ses)
 126 {
 127         return ses->se_flags & NFS4_SESSION_DEAD;
 128 }
 129
 130 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 131 {
 132         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 133                 return nfserr_jukebox;
 134         ses->se_flags |= NFS4_SESSION_DEAD;
 135         return nfs_ok;
 136 }
 137
 138 static bool is_client_expired(struct nfs4_client *clp)
 139 {
 140         return clp->cl_time == 0;
 141 }
 142
 143 static __be32 get_client_locked(struct nfs4_client *clp)
 144 {
 145         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 146
 147         lockdep_assert_held(&nn->client_lock);
 148
 149         if (is_client_expired(clp))
 150                 return nfserr_expired;
 151         atomic_inc(&clp->cl_rpc_users);
 152         return nfs_ok;
 153 }
 154
 155 /* must be called under the client_lock */
 156 static inline void
 157 renew_client_locked(struct nfs4_client *clp)
 158 {
 159         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 160
 161         if (is_client_expired(clp)) {
 162                 WARN_ON(1);
 163                 printk("%s: client (clientid %08x/%08x) already expired\n",
 164                         __func__,
 165                         clp->cl_clientid.cl_boot,
 166                         clp->cl_clientid.cl_id);
 167                 return;
 168         }
 169
 170         dprintk("renewing client (clientid %08x/%08x)\n",
 171                         clp->cl_clientid.cl_boot,
 172                         clp->cl_clientid.cl_id);
 173         list_move_tail(&clp->cl_lru, &nn->client_lru);
 174         clp->cl_time = ktime_get_boottime_seconds();
 175 }
 176
 177 static void put_client_renew_locked(struct nfs4_client *clp)
 178 {
 179         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 180
 181         lockdep_assert_held(&nn->client_lock);
 182
 183         if (!atomic_dec_and_test(&clp->cl_rpc_users))
 184                 return;
 185         if (!is_client_expired(clp))
 186                 renew_client_locked(clp);
 187         else
 188                 wake_up_all(&expiry_wq);
 189 }
 190
 191 static void put_client_renew(struct nfs4_client *clp)
 192 {
 193         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 194
 195         if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
 196                 return;
 197         if (!is_client_expired(clp))
 198                 renew_client_locked(clp);
 199         else
 200                 wake_up_all(&expiry_wq);
 201         spin_unlock(&nn->client_lock);
 202 }
 203
 204 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 205 {
 206         __be32 status;
 207
 208         if (is_session_dead(ses))
 209                 return nfserr_badsession;
 210         status = get_client_locked(ses->se_client);
 211         if (status)
 212                 return status;
 213         atomic_inc(&ses->se_ref);
 214         return nfs_ok;
 215 }
 216
 217 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 218 {
 219         struct nfs4_client *clp = ses->se_client;
 220         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 221
 222         lockdep_assert_held(&nn->client_lock);
 223
 224         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 225                 free_session(ses);
 226         put_client_renew_locked(clp);
 227 }
 228
 229 static void nfsd4_put_session(struct nfsd4_session *ses)
 230 {
 231         struct nfs4_client *clp = ses->se_client;
 232         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 233
 234         spin_lock(&nn->client_lock);
 235         nfsd4_put_session_locked(ses);
 236         spin_unlock(&nn->client_lock);
 237 }
 238
 239 static struct nfsd4_blocked_lock *
 240 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 241                         struct nfsd_net *nn)
 242 {
 243         struct nfsd4_blocked_lock *cur, *found = NULL;
 244
 245         spin_lock(&nn->blocked_locks_lock);
 246         list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
 247                 if (fh_match(fh, &cur->nbl_fh)) {
 248                         list_del_init(&cur->nbl_list);
 249                         list_del_init(&cur->nbl_lru);
 250                         found = cur;
 251                         break;
 252                 }
 253         }
 254         spin_unlock(&nn->blocked_locks_lock);
 255         if (found)
 256                 locks_delete_block(&found->nbl_lock);
 257         return found;
 258 }
 259
 260 static struct nfsd4_blocked_lock *
 261 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 262                         struct nfsd_net *nn)
 263 {
 264         struct nfsd4_blocked_lock *nbl;
 265
 266         nbl = find_blocked_lock(lo, fh, nn);
 267         if (!nbl) {
 268                 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
 269                 if (nbl) {
 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         spin_lock(&f->fi_lock);
 498         ret = __nfs4_get_fd(f, O_RDWR);
 499         if (!ret) {
 500                 ret = __nfs4_get_fd(f, O_WRONLY);
 501                 if (!ret)
 502                         ret = __nfs4_get_fd(f, O_RDONLY);
 503         }
 504         spin_unlock(&f->fi_lock);
 505         return ret;
 506 }
 507
 508 static atomic_long_t num_delegations;
 509 unsigned long max_delegations;
 510
 511 /*
 512  * Open owner state (share locks)
 513  */
 514
 515 /* hash tables for lock and open owners */
 516 #define OWNER_HASH_BITS              8
 517 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 518 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 519
 520 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 521 {
 522         unsigned int ret;
 523
 524         ret = opaque_hashval(ownername->data, ownername->len);
 525         return ret & OWNER_HASH_MASK;
 526 }
 527
 528 /* hash table for nfs4_file */
 529 #define FILE_HASH_BITS                   8
 530 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 531
 532 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 533 {
 534         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 535 }
 536
 537 static unsigned int file_hashval(struct knfsd_fh *fh)
 538 {
 539         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 540 }
 541
 542 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 543
 544 static void
 545 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 546 {
 547         lockdep_assert_held(&fp->fi_lock);
 548
 549         if (access & NFS4_SHARE_ACCESS_WRITE)
 550                 atomic_inc(&fp->fi_access[O_WRONLY]);
 551         if (access & NFS4_SHARE_ACCESS_READ)
 552                 atomic_inc(&fp->fi_access[O_RDONLY]);
 553 }
 554
 555 static __be32
 556 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 557 {
 558         lockdep_assert_held(&fp->fi_lock);
 559
 560         /* Does this access mode make sense? */
 561         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 562                 return nfserr_inval;
 563
 564         /* Does it conflict with a deny mode already set? */
 565         if ((access & fp->fi_share_deny) != 0)
 566                 return nfserr_share_denied;
 567
 568         __nfs4_file_get_access(fp, access);
 569         return nfs_ok;
 570 }
 571
 572 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 573 {
 574         /* Common case is that there is no deny mode. */
 575         if (deny) {
 576                 /* Does this deny mode make sense? */
 577                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 578                         return nfserr_inval;
 579
 580                 if ((deny & NFS4_SHARE_DENY_READ) &&
 581                     atomic_read(&fp->fi_access[O_RDONLY]))
 582                         return nfserr_share_denied;
 583
 584                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 585                     atomic_read(&fp->fi_access[O_WRONLY]))
 586                         return nfserr_share_denied;
 587         }
 588         return nfs_ok;
 589 }
 590
 591 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 592 {
 593         might_lock(&fp->fi_lock);
 594
 595         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 596                 struct nfsd_file *f1 = NULL;
 597                 struct nfsd_file *f2 = NULL;
 598
 599                 swap(f1, fp->fi_fds[oflag]);
 600                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 601                         swap(f2, fp->fi_fds[O_RDWR]);
 602                 spin_unlock(&fp->fi_lock);
 603                 if (f1)
 604                         nfsd_file_put(f1);
 605                 if (f2)
 606                         nfsd_file_put(f2);
 607         }
 608 }
 609
 610 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 611 {
 612         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 613
 614         if (access & NFS4_SHARE_ACCESS_WRITE)
 615                 __nfs4_file_put_access(fp, O_WRONLY);
 616         if (access & NFS4_SHARE_ACCESS_READ)
 617                 __nfs4_file_put_access(fp, O_RDONLY);
 618 }
 619
 620 /*
 621  * Allocate a new open/delegation state counter. This is needed for
 622  * pNFS for proper return on close semantics.
 623  *
 624  * Note that we only allocate it for pNFS-enabled exports, otherwise
 625  * all pointers to struct nfs4_clnt_odstate are always NULL.
 626  */
 627 static struct nfs4_clnt_odstate *
 628 alloc_clnt_odstate(struct nfs4_client *clp)
 629 {
 630         struct nfs4_clnt_odstate *co;
 631
 632         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 633         if (co) {
 634                 co->co_client = clp;
 635                 refcount_set(&co->co_odcount, 1);
 636         }
 637         return co;
 638 }
 639
 640 static void
 641 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 642 {
 643         struct nfs4_file *fp = co->co_file;
 644
 645         lockdep_assert_held(&fp->fi_lock);
 646         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 647 }
 648
 649 static inline void
 650 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 651 {
 652         if (co)
 653                 refcount_inc(&co->co_odcount);
 654 }
 655
 656 static void
 657 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 658 {
 659         struct nfs4_file *fp;
 660
 661         if (!co)
 662                 return;
 663
 664         fp = co->co_file;
 665         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 666                 list_del(&co->co_perfile);
 667                 spin_unlock(&fp->fi_lock);
 668
 669                 nfsd4_return_all_file_layouts(co->co_client, fp);
 670                 kmem_cache_free(odstate_slab, co);
 671         }
 672 }
 673
 674 static struct nfs4_clnt_odstate *
 675 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 676 {
 677         struct nfs4_clnt_odstate *co;
 678         struct nfs4_client *cl;
 679
 680         if (!new)
 681                 return NULL;
 682
 683         cl = new->co_client;
 684
 685         spin_lock(&fp->fi_lock);
 686         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 687                 if (co->co_client == cl) {
 688                         get_clnt_odstate(co);
 689                         goto out;
 690                 }
 691         }
 692         co = new;
 693         co->co_file = fp;
 694         hash_clnt_odstate_locked(new);
 695 out:
 696         spin_unlock(&fp->fi_lock);
 697         return co;
 698 }
 699
 700 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 701                                   void (*sc_free)(struct nfs4_stid *))
 702 {
 703         struct nfs4_stid *stid;
 704         int new_id;
 705
 706         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 707         if (!stid)
 708                 return NULL;
 709
 710         idr_preload(GFP_KERNEL);
 711         spin_lock(&cl->cl_lock);
 712         /* Reserving 0 for start of file in nfsdfs "states" file: */
 713         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
 714         spin_unlock(&cl->cl_lock);
 715         idr_preload_end();
 716         if (new_id < 0)
 717                 goto out_free;
 718
 719         stid->sc_free = sc_free;
 720         stid->sc_client = cl;
 721         stid->sc_stateid.si_opaque.so_id = new_id;
 722         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 723         /* Will be incremented before return to client: */
 724         refcount_set(&stid->sc_count, 1);
 725         spin_lock_init(&stid->sc_lock);
 726         INIT_LIST_HEAD(&stid->sc_cp_list);
 727
 728         /*
 729          * It shouldn't be a problem to reuse an opaque stateid value.
 730          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 731          * example, a stray write retransmission could be accepted by
 732          * the server when it should have been rejected.  Therefore,
 733          * adopt a trick from the sctp code to attempt to maximize the
 734          * amount of time until an id is reused, by ensuring they always
 735          * "increase" (mod INT_MAX):
 736          */
 737         return stid;
 738 out_free:
 739         kmem_cache_free(slab, stid);
 740         return NULL;
 741 }
 742
 743 /*
 744  * Create a unique stateid_t to represent each COPY.
 745  */
 746 static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
 747                               unsigned char sc_type)
 748 {
 749         int new_id;
 750
 751         stid->stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
 752         stid->stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
 753         stid->sc_type = sc_type;
 754
 755         idr_preload(GFP_KERNEL);
 756         spin_lock(&nn->s2s_cp_lock);
 757         new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
 758         stid->stid.si_opaque.so_id = new_id;
 759         spin_unlock(&nn->s2s_cp_lock);
 760         idr_preload_end();
 761         if (new_id < 0)
 762                 return 0;
 763         return 1;
 764 }
 765
 766 int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
 767 {
 768         return nfs4_init_cp_state(nn, &copy->cp_stateid, NFS4_COPY_STID);
 769 }
 770
 771 struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
 772                                                      struct nfs4_stid *p_stid)
 773 {
 774         struct nfs4_cpntf_state *cps;
 775
 776         cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
 777         if (!cps)
 778                 return NULL;
 779         cps->cpntf_time = ktime_get_boottime_seconds();
 780         refcount_set(&cps->cp_stateid.sc_count, 1);
 781         if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
 782                 goto out_free;
 783         spin_lock(&nn->s2s_cp_lock);
 784         list_add(&cps->cp_list, &p_stid->sc_cp_list);
 785         spin_unlock(&nn->s2s_cp_lock);
 786         return cps;
 787 out_free:
 788         kfree(cps);
 789         return NULL;
 790 }
 791
 792 void nfs4_free_copy_state(struct nfsd4_copy *copy)
 793 {
 794         struct nfsd_net *nn;
 795
 796         WARN_ON_ONCE(copy->cp_stateid.sc_type != NFS4_COPY_STID);
 797         nn = net_generic(copy->cp_clp->net, nfsd_net_id);
 798         spin_lock(&nn->s2s_cp_lock);
 799         idr_remove(&nn->s2s_cp_stateids,
 800                    copy->cp_stateid.stid.si_opaque.so_id);
 801         spin_unlock(&nn->s2s_cp_lock);
 802 }
 803
 804 static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
 805 {
 806         struct nfs4_cpntf_state *cps;
 807         struct nfsd_net *nn;
 808
 809         nn = net_generic(net, nfsd_net_id);
 810         spin_lock(&nn->s2s_cp_lock);
 811         while (!list_empty(&stid->sc_cp_list)) {
 812                 cps = list_first_entry(&stid->sc_cp_list,
 813                                        struct nfs4_cpntf_state, cp_list);
 814                 _free_cpntf_state_locked(nn, cps);
 815         }
 816         spin_unlock(&nn->s2s_cp_lock);
 817 }
 818
 819 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 820 {
 821         struct nfs4_stid *stid;
 822
 823         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
 824         if (!stid)
 825                 return NULL;
 826
 827         return openlockstateid(stid);
 828 }
 829
 830 static void nfs4_free_deleg(struct nfs4_stid *stid)
 831 {
 832         kmem_cache_free(deleg_slab, stid);
 833         atomic_long_dec(&num_delegations);
 834 }
 835
 836 /*
 837  * When we recall a delegation, we should be careful not to hand it
 838  * out again straight away.
 839  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 840  * in which the filehandles of recalled delegations are "stored".
 841  * If a filehandle appear in either filter, a delegation is blocked.
 842  * When a delegation is recalled, the filehandle is stored in the "new"
 843  * filter.
 844  * Every 30 seconds we swap the filters and clear the "new" one,
 845  * unless both are empty of course.
 846  *
 847  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 848  * low 3 bytes as hash-table indices.
 849  *
 850  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 851  * is used to manage concurrent access.  Testing does not need the lock
 852  * except when swapping the two filters.
 853  */
 854 static DEFINE_SPINLOCK(blocked_delegations_lock);
 855 static struct bloom_pair {
 856         int     entries, old_entries;
 857         time64_t swap_time;
 858         int     new; /* index into 'set' */
 859         DECLARE_BITMAP(set[2], 256);
 860 } blocked_delegations;
 861
 862 static int delegation_blocked(struct knfsd_fh *fh)
 863 {
 864         u32 hash;
 865         struct bloom_pair *bd = &blocked_delegations;
 866
 867         if (bd->entries == 0)
 868                 return 0;
 869         if (ktime_get_seconds() - bd->swap_time > 30) {
 870                 spin_lock(&blocked_delegations_lock);
 871                 if (ktime_get_seconds() - bd->swap_time > 30) {
 872                         bd->entries -= bd->old_entries;
 873                         bd->old_entries = bd->entries;
 874                         memset(bd->set[bd->new], 0,
 875                                sizeof(bd->set[0]));
 876                         bd->new = 1-bd->new;
 877                         bd->swap_time = ktime_get_seconds();
 878                 }
 879                 spin_unlock(&blocked_delegations_lock);
 880         }
 881         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 882         if (test_bit(hash&255, bd->set[0]) &&
 883             test_bit((hash>>8)&255, bd->set[0]) &&
 884             test_bit((hash>>16)&255, bd->set[0]))
 885                 return 1;
 886
 887         if (test_bit(hash&255, bd->set[1]) &&
 888             test_bit((hash>>8)&255, bd->set[1]) &&
 889             test_bit((hash>>16)&255, bd->set[1]))
 890                 return 1;
 891
 892         return 0;
 893 }
 894
 895 static void block_delegations(struct knfsd_fh *fh)
 896 {
 897         u32 hash;
 898         struct bloom_pair *bd = &blocked_delegations;
 899
 900         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 901
 902         spin_lock(&blocked_delegations_lock);
 903         __set_bit(hash&255, bd->set[bd->new]);
 904         __set_bit((hash>>8)&255, bd->set[bd->new]);
 905         __set_bit((hash>>16)&255, bd->set[bd->new]);
 906         if (bd->entries == 0)
 907                 bd->swap_time = ktime_get_seconds();
 908         bd->entries += 1;
 909         spin_unlock(&blocked_delegations_lock);
 910 }
 911
 912 static struct nfs4_delegation *
 913 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
 914                  struct svc_fh *current_fh,
 915                  struct nfs4_clnt_odstate *odstate)
 916 {
 917         struct nfs4_delegation *dp;
 918         long n;
 919
 920         dprintk("NFSD alloc_init_deleg\n");
 921         n = atomic_long_inc_return(&num_delegations);
 922         if (n < 0 || n > max_delegations)
 923                 goto out_dec;
 924         if (delegation_blocked(&current_fh->fh_handle))
 925                 goto out_dec;
 926         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
 927         if (dp == NULL)
 928                 goto out_dec;
 929
 930         /*
 931          * delegation seqid's are never incremented.  The 4.1 special
 932          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 933          * 0 anyway just for consistency and use 1:
 934          */
 935         dp->dl_stid.sc_stateid.si_generation = 1;
 936         INIT_LIST_HEAD(&dp->dl_perfile);
 937         INIT_LIST_HEAD(&dp->dl_perclnt);
 938         INIT_LIST_HEAD(&dp->dl_recall_lru);
 939         dp->dl_clnt_odstate = odstate;
 940         get_clnt_odstate(odstate);
 941         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 942         dp->dl_retries = 1;
 943         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 944                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 945         get_nfs4_file(fp);
 946         dp->dl_stid.sc_file = fp;
 947         return dp;
 948 out_dec:
 949         atomic_long_dec(&num_delegations);
 950         return NULL;
 951 }
 952
 953 void
 954 nfs4_put_stid(struct nfs4_stid *s)
 955 {
 956         struct nfs4_file *fp = s->sc_file;
 957         struct nfs4_client *clp = s->sc_client;
 958
 959         might_lock(&clp->cl_lock);
 960
 961         if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 962                 wake_up_all(&close_wq);
 963                 return;
 964         }
 965         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 966         nfs4_free_cpntf_statelist(clp->net, s);
 967         spin_unlock(&clp->cl_lock);
 968         s->sc_free(s);
 969         if (fp)
 970                 put_nfs4_file(fp);
 971 }
 972
 973 void
 974 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
 975 {
 976         stateid_t *src = &stid->sc_stateid;
 977
 978         spin_lock(&stid->sc_lock);
 979         if (unlikely(++src->si_generation == 0))
 980                 src->si_generation = 1;
 981         memcpy(dst, src, sizeof(*dst));
 982         spin_unlock(&stid->sc_lock);
 983 }
 984
 985 static void put_deleg_file(struct nfs4_file *fp)
 986 {
 987         struct nfsd_file *nf = NULL;
 988
 989         spin_lock(&fp->fi_lock);
 990         if (--fp->fi_delegees == 0)
 991                 swap(nf, fp->fi_deleg_file);
 992         spin_unlock(&fp->fi_lock);
 993
 994         if (nf)
 995                 nfsd_file_put(nf);
 996 }
 997
 998 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
 999 {
1000         struct nfs4_file *fp = dp->dl_stid.sc_file;
1001         struct nfsd_file *nf = fp->fi_deleg_file;
1002
1003         WARN_ON_ONCE(!fp->fi_delegees);
1004
1005         vfs_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
1006         put_deleg_file(fp);
1007 }
1008
1009 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
1010 {
1011         put_clnt_odstate(dp->dl_clnt_odstate);
1012         nfs4_unlock_deleg_lease(dp);
1013         nfs4_put_stid(&dp->dl_stid);
1014 }
1015
1016 void nfs4_unhash_stid(struct nfs4_stid *s)
1017 {
1018         s->sc_type = 0;
1019 }
1020
1021 /**
1022  * nfs4_delegation_exists - Discover if this delegation already exists
1023  * @clp:     a pointer to the nfs4_client we're granting a delegation to
1024  * @fp:      a pointer to the nfs4_file we're granting a delegation on
1025  *
1026  * Return:
1027  *      On success: true iff an existing delegation is found
1028  */
1029
1030 static bool
1031 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
1032 {
1033         struct nfs4_delegation *searchdp = NULL;
1034         struct nfs4_client *searchclp = NULL;
1035
1036         lockdep_assert_held(&state_lock);
1037         lockdep_assert_held(&fp->fi_lock);
1038
1039         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
1040                 searchclp = searchdp->dl_stid.sc_client;
1041                 if (clp == searchclp) {
1042                         return true;
1043                 }
1044         }
1045         return false;
1046 }
1047
1048 /**
1049  * hash_delegation_locked - Add a delegation to the appropriate lists
1050  * @dp:     a pointer to the nfs4_delegation we are adding.
1051  * @fp:     a pointer to the nfs4_file we're granting a delegation on
1052  *
1053  * Return:
1054  *      On success: NULL if the delegation was successfully hashed.
1055  *
1056  *      On error: -EAGAIN if one was previously granted to this
1057  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
1058  *
1059  */
1060
1061 static int
1062 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1063 {
1064         struct nfs4_client *clp = dp->dl_stid.sc_client;
1065
1066         lockdep_assert_held(&state_lock);
1067         lockdep_assert_held(&fp->fi_lock);
1068
1069         if (nfs4_delegation_exists(clp, fp))
1070                 return -EAGAIN;
1071         refcount_inc(&dp->dl_stid.sc_count);
1072         dp->dl_stid.sc_type = NFS4_DELEG_STID;
1073         list_add(&dp->dl_perfile, &fp->fi_delegations);
1074         list_add(&dp->dl_perclnt, &clp->cl_delegations);
1075         return 0;
1076 }
1077
1078 static bool
1079 unhash_delegation_locked(struct nfs4_delegation *dp)
1080 {
1081         struct nfs4_file *fp = dp->dl_stid.sc_file;
1082
1083         lockdep_assert_held(&state_lock);
1084
1085         if (list_empty(&dp->dl_perfile))
1086                 return false;
1087
1088         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
1089         /* Ensure that deleg break won't try to requeue it */
1090         ++dp->dl_time;
1091         spin_lock(&fp->fi_lock);
1092         list_del_init(&dp->dl_perclnt);
1093         list_del_init(&dp->dl_recall_lru);
1094         list_del_init(&dp->dl_perfile);
1095         spin_unlock(&fp->fi_lock);
1096         return true;
1097 }
1098
1099 static void destroy_delegation(struct nfs4_delegation *dp)
1100 {
1101         bool unhashed;
1102
1103         spin_lock(&state_lock);
1104         unhashed = unhash_delegation_locked(dp);
1105         spin_unlock(&state_lock);
1106         if (unhashed)
1107                 destroy_unhashed_deleg(dp);
1108 }
1109
1110 static void revoke_delegation(struct nfs4_delegation *dp)
1111 {
1112         struct nfs4_client *clp = dp->dl_stid.sc_client;
1113
1114         WARN_ON(!list_empty(&dp->dl_recall_lru));
1115
1116         if (clp->cl_minorversion) {
1117                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1118                 refcount_inc(&dp->dl_stid.sc_count);
1119                 spin_lock(&clp->cl_lock);
1120                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1121                 spin_unlock(&clp->cl_lock);
1122         }
1123         destroy_unhashed_deleg(dp);
1124 }
1125
1126 /*
1127  * SETCLIENTID state
1128  */
1129
1130 static unsigned int clientid_hashval(u32 id)
1131 {
1132         return id & CLIENT_HASH_MASK;
1133 }
1134
1135 static unsigned int clientstr_hashval(struct xdr_netobj name)
1136 {
1137         return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1138 }
1139
1140 /*
1141  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1142  * st_{access,deny}_bmap field of the stateid, in order to track not
1143  * only what share bits are currently in force, but also what
1144  * combinations of share bits previous opens have used.  This allows us
1145  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1146  * return an error if the client attempt to downgrade to a combination
1147  * of share bits not explicable by closing some of its previous opens.
1148  *
1149  * XXX: This enforcement is actually incomplete, since we don't keep
1150  * track of access/deny bit combinations; so, e.g., we allow:
1151  *
1152  *      OPEN allow read, deny write
1153  *      OPEN allow both, deny none
1154  *      DOWNGRADE allow read, deny none
1155  *
1156  * which we should reject.
1157  */
1158 static unsigned int
1159 bmap_to_share_mode(unsigned long bmap) {
1160         int i;
1161         unsigned int access = 0;
1162
1163         for (i = 1; i < 4; i++) {
1164                 if (test_bit(i, &bmap))
1165                         access |= i;
1166         }
1167         return access;
1168 }
1169
1170 /* set share access for a given stateid */
1171 static inline void
1172 set_access(u32 access, struct nfs4_ol_stateid *stp)
1173 {
1174         unsigned char mask = 1 << access;
1175
1176         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1177         stp->st_access_bmap |= mask;
1178 }
1179
1180 /* clear share access for a given stateid */
1181 static inline void
1182 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1183 {
1184         unsigned char mask = 1 << access;
1185
1186         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1187         stp->st_access_bmap &= ~mask;
1188 }
1189
1190 /* test whether a given stateid has access */
1191 static inline bool
1192 test_access(u32 access, struct nfs4_ol_stateid *stp)
1193 {
1194         unsigned char mask = 1 << access;
1195
1196         return (bool)(stp->st_access_bmap & mask);
1197 }
1198
1199 /* set share deny for a given stateid */
1200 static inline void
1201 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1202 {
1203         unsigned char mask = 1 << deny;
1204
1205         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1206         stp->st_deny_bmap |= mask;
1207 }
1208
1209 /* clear share deny for a given stateid */
1210 static inline void
1211 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1212 {
1213         unsigned char mask = 1 << deny;
1214
1215         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1216         stp->st_deny_bmap &= ~mask;
1217 }
1218
1219 /* test whether a given stateid is denying specific access */
1220 static inline bool
1221 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1222 {
1223         unsigned char mask = 1 << deny;
1224
1225         return (bool)(stp->st_deny_bmap & mask);
1226 }
1227
1228 static int nfs4_access_to_omode(u32 access)
1229 {
1230         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1231         case NFS4_SHARE_ACCESS_READ:
1232                 return O_RDONLY;
1233         case NFS4_SHARE_ACCESS_WRITE:
1234                 return O_WRONLY;
1235         case NFS4_SHARE_ACCESS_BOTH:
1236                 return O_RDWR;
1237         }
1238         WARN_ON_ONCE(1);
1239         return O_RDONLY;
1240 }
1241
1242 /*
1243  * A stateid that had a deny mode associated with it is being released
1244  * or downgraded. Recalculate the deny mode on the file.
1245  */
1246 static void
1247 recalculate_deny_mode(struct nfs4_file *fp)
1248 {
1249         struct nfs4_ol_stateid *stp;
1250
1251         spin_lock(&fp->fi_lock);
1252         fp->fi_share_deny = 0;
1253         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1254                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1255         spin_unlock(&fp->fi_lock);
1256 }
1257
1258 static void
1259 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1260 {
1261         int i;
1262         bool change = false;
1263
1264         for (i = 1; i < 4; i++) {
1265                 if ((i & deny) != i) {
1266                         change = true;
1267                         clear_deny(i, stp);
1268                 }
1269         }
1270
1271         /* Recalculate per-file deny mode if there was a change */
1272         if (change)
1273                 recalculate_deny_mode(stp->st_stid.sc_file);
1274 }
1275
1276 /* release all access and file references for a given stateid */
1277 static void
1278 release_all_access(struct nfs4_ol_stateid *stp)
1279 {
1280         int i;
1281         struct nfs4_file *fp = stp->st_stid.sc_file;
1282
1283         if (fp && stp->st_deny_bmap != 0)
1284                 recalculate_deny_mode(fp);
1285
1286         for (i = 1; i < 4; i++) {
1287                 if (test_access(i, stp))
1288                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1289                 clear_access(i, stp);
1290         }
1291 }
1292
1293 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1294 {
1295         kfree(sop->so_owner.data);
1296         sop->so_ops->so_free(sop);
1297 }
1298
1299 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1300 {
1301         struct nfs4_client *clp = sop->so_client;
1302
1303         might_lock(&clp->cl_lock);
1304
1305         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1306                 return;
1307         sop->so_ops->so_unhash(sop);
1308         spin_unlock(&clp->cl_lock);
1309         nfs4_free_stateowner(sop);
1310 }
1311
1312 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1313 {
1314         struct nfs4_file *fp = stp->st_stid.sc_file;
1315
1316         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1317
1318         if (list_empty(&stp->st_perfile))
1319                 return false;
1320
1321         spin_lock(&fp->fi_lock);
1322         list_del_init(&stp->st_perfile);
1323         spin_unlock(&fp->fi_lock);
1324         list_del(&stp->st_perstateowner);
1325         return true;
1326 }
1327
1328 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1329 {
1330         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1331
1332         put_clnt_odstate(stp->st_clnt_odstate);
1333         release_all_access(stp);
1334         if (stp->st_stateowner)
1335                 nfs4_put_stateowner(stp->st_stateowner);
1336         kmem_cache_free(stateid_slab, stid);
1337 }
1338
1339 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1340 {
1341         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1342         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1343         struct nfsd_file *nf;
1344
1345         nf = find_any_file(stp->st_stid.sc_file);
1346         if (nf) {
1347                 get_file(nf->nf_file);
1348                 filp_close(nf->nf_file, (fl_owner_t)lo);
1349                 nfsd_file_put(nf);
1350         }
1351         nfs4_free_ol_stateid(stid);
1352 }
1353
1354 /*
1355  * Put the persistent reference to an already unhashed generic stateid, while
1356  * holding the cl_lock. If it's the last reference, then put it onto the
1357  * reaplist for later destruction.
1358  */
1359 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1360                                        struct list_head *reaplist)
1361 {
1362         struct nfs4_stid *s = &stp->st_stid;
1363         struct nfs4_client *clp = s->sc_client;
1364
1365         lockdep_assert_held(&clp->cl_lock);
1366
1367         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1368
1369         if (!refcount_dec_and_test(&s->sc_count)) {
1370                 wake_up_all(&close_wq);
1371                 return;
1372         }
1373
1374         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1375         list_add(&stp->st_locks, reaplist);
1376 }
1377
1378 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1379 {
1380         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1381
1382         list_del_init(&stp->st_locks);
1383         nfs4_unhash_stid(&stp->st_stid);
1384         return unhash_ol_stateid(stp);
1385 }
1386
1387 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1388 {
1389         struct nfs4_client *clp = stp->st_stid.sc_client;
1390         bool unhashed;
1391
1392         spin_lock(&clp->cl_lock);
1393         unhashed = unhash_lock_stateid(stp);
1394         spin_unlock(&clp->cl_lock);
1395         if (unhashed)
1396                 nfs4_put_stid(&stp->st_stid);
1397 }
1398
1399 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1400 {
1401         struct nfs4_client *clp = lo->lo_owner.so_client;
1402
1403         lockdep_assert_held(&clp->cl_lock);
1404
1405         list_del_init(&lo->lo_owner.so_strhash);
1406 }
1407
1408 /*
1409  * Free a list of generic stateids that were collected earlier after being
1410  * fully unhashed.
1411  */
1412 static void
1413 free_ol_stateid_reaplist(struct list_head *reaplist)
1414 {
1415         struct nfs4_ol_stateid *stp;
1416         struct nfs4_file *fp;
1417
1418         might_sleep();
1419
1420         while (!list_empty(reaplist)) {
1421                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1422                                        st_locks);
1423                 list_del(&stp->st_locks);
1424                 fp = stp->st_stid.sc_file;
1425                 stp->st_stid.sc_free(&stp->st_stid);
1426                 if (fp)
1427                         put_nfs4_file(fp);
1428         }
1429 }
1430
1431 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1432                                        struct list_head *reaplist)
1433 {
1434         struct nfs4_ol_stateid *stp;
1435
1436         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1437
1438         while (!list_empty(&open_stp->st_locks)) {
1439                 stp = list_entry(open_stp->st_locks.next,
1440                                 struct nfs4_ol_stateid, st_locks);
1441                 WARN_ON(!unhash_lock_stateid(stp));
1442                 put_ol_stateid_locked(stp, reaplist);
1443         }
1444 }
1445
1446 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1447                                 struct list_head *reaplist)
1448 {
1449         bool unhashed;
1450
1451         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1452
1453         unhashed = unhash_ol_stateid(stp);
1454         release_open_stateid_locks(stp, reaplist);
1455         return unhashed;
1456 }
1457
1458 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1459 {
1460         LIST_HEAD(reaplist);
1461
1462         spin_lock(&stp->st_stid.sc_client->cl_lock);
1463         if (unhash_open_stateid(stp, &reaplist))
1464                 put_ol_stateid_locked(stp, &reaplist);
1465         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1466         free_ol_stateid_reaplist(&reaplist);
1467 }
1468
1469 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1470 {
1471         struct nfs4_client *clp = oo->oo_owner.so_client;
1472
1473         lockdep_assert_held(&clp->cl_lock);
1474
1475         list_del_init(&oo->oo_owner.so_strhash);
1476         list_del_init(&oo->oo_perclient);
1477 }
1478
1479 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1480 {
1481         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1482                                           nfsd_net_id);
1483         struct nfs4_ol_stateid *s;
1484
1485         spin_lock(&nn->client_lock);
1486         s = oo->oo_last_closed_stid;
1487         if (s) {
1488                 list_del_init(&oo->oo_close_lru);
1489                 oo->oo_last_closed_stid = NULL;
1490         }
1491         spin_unlock(&nn->client_lock);
1492         if (s)
1493                 nfs4_put_stid(&s->st_stid);
1494 }
1495
1496 static void release_openowner(struct nfs4_openowner *oo)
1497 {
1498         struct nfs4_ol_stateid *stp;
1499         struct nfs4_client *clp = oo->oo_owner.so_client;
1500         struct list_head reaplist;
1501
1502         INIT_LIST_HEAD(&reaplist);
1503
1504         spin_lock(&clp->cl_lock);
1505         unhash_openowner_locked(oo);
1506         while (!list_empty(&oo->oo_owner.so_stateids)) {
1507                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1508                                 struct nfs4_ol_stateid, st_perstateowner);
1509                 if (unhash_open_stateid(stp, &reaplist))
1510                         put_ol_stateid_locked(stp, &reaplist);
1511         }
1512         spin_unlock(&clp->cl_lock);
1513         free_ol_stateid_reaplist(&reaplist);
1514         release_last_closed_stateid(oo);
1515         nfs4_put_stateowner(&oo->oo_owner);
1516 }
1517
1518 static inline int
1519 hash_sessionid(struct nfs4_sessionid *sessionid)
1520 {
1521         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1522
1523         return sid->sequence % SESSION_HASH_SIZE;
1524 }
1525
1526 #ifdef CONFIG_SUNRPC_DEBUG
1527 static inline void
1528 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1529 {
1530         u32 *ptr = (u32 *)(&sessionid->data[0]);
1531         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1532 }
1533 #else
1534 static inline void
1535 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1536 {
1537 }
1538 #endif
1539
1540 /*
1541  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1542  * won't be used for replay.
1543  */
1544 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1545 {
1546         struct nfs4_stateowner *so = cstate->replay_owner;
1547
1548         if (nfserr == nfserr_replay_me)
1549                 return;
1550
1551         if (!seqid_mutating_err(ntohl(nfserr))) {
1552                 nfsd4_cstate_clear_replay(cstate);
1553                 return;
1554         }
1555         if (!so)
1556                 return;
1557         if (so->so_is_open_owner)
1558                 release_last_closed_stateid(openowner(so));
1559         so->so_seqid++;
1560         return;
1561 }
1562
1563 static void
1564 gen_sessionid(struct nfsd4_session *ses)
1565 {
1566         struct nfs4_client *clp = ses->se_client;
1567         struct nfsd4_sessionid *sid;
1568
1569         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1570         sid->clientid = clp->cl_clientid;
1571         sid->sequence = current_sessionid++;
1572         sid->reserved = 0;
1573 }
1574
1575 /*
1576  * The protocol defines ca_maxresponssize_cached to include the size of
1577  * the rpc header, but all we need to cache is the data starting after
1578  * the end of the initial SEQUENCE operation--the rest we regenerate
1579  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1580  * value that is the number of bytes in our cache plus a few additional
1581  * bytes.  In order to stay on the safe side, and not promise more than
1582  * we can cache, those additional bytes must be the minimum possible: 24
1583  * bytes of rpc header (xid through accept state, with AUTH_NULL
1584  * verifier), 12 for the compound header (with zero-length tag), and 44
1585  * for the SEQUENCE op response:
1586  */
1587 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1588
1589 static void
1590 free_session_slots(struct nfsd4_session *ses)
1591 {
1592         int i;
1593
1594         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1595                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1596                 kfree(ses->se_slots[i]);
1597         }
1598 }
1599
1600 /*
1601  * We don't actually need to cache the rpc and session headers, so we
1602  * can allocate a little less for each slot:
1603  */
1604 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1605 {
1606         u32 size;
1607
1608         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1609                 size = 0;
1610         else
1611                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1612         return size + sizeof(struct nfsd4_slot);
1613 }
1614
1615 /*
1616  * XXX: If we run out of reserved DRC memory we could (up to a point)
1617  * re-negotiate active sessions and reduce their slot usage to make
1618  * room for new connections. For now we just fail the create session.
1619  */
1620 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1621 {
1622         u32 slotsize = slot_bytes(ca);
1623         u32 num = ca->maxreqs;
1624         unsigned long avail, total_avail;
1625         unsigned int scale_factor;
1626
1627         spin_lock(&nfsd_drc_lock);
1628         if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1629                 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1630         else
1631                 /* We have handed out more space than we chose in
1632                  * set_max_drc() to allow.  That isn't really a
1633                  * problem as long as that doesn't make us think we
1634                  * have lots more due to integer overflow.
1635                  */
1636                 total_avail = 0;
1637         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1638         /*
1639          * Never use more than a fraction of the remaining memory,
1640          * unless it's the only way to give this client a slot.
1641          * The chosen fraction is either 1/8 or 1/number of threads,
1642          * whichever is smaller.  This ensures there are adequate
1643          * slots to support multiple clients per thread.
1644          * Give the client one slot even if that would require
1645          * over-allocation--it is better than failure.
1646          */
1647         scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1648
1649         avail = clamp_t(unsigned long, avail, slotsize,
1650                         total_avail/scale_factor);
1651         num = min_t(int, num, avail / slotsize);
1652         num = max_t(int, num, 1);
1653         nfsd_drc_mem_used += num * slotsize;
1654         spin_unlock(&nfsd_drc_lock);
1655
1656         return num;
1657 }
1658
1659 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1660 {
1661         int slotsize = slot_bytes(ca);
1662
1663         spin_lock(&nfsd_drc_lock);
1664         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1665         spin_unlock(&nfsd_drc_lock);
1666 }
1667
1668 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1669                                            struct nfsd4_channel_attrs *battrs)
1670 {
1671         int numslots = fattrs->maxreqs;
1672         int slotsize = slot_bytes(fattrs);
1673         struct nfsd4_session *new;
1674         int mem, i;
1675
1676         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1677                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1678         mem = numslots * sizeof(struct nfsd4_slot *);
1679
1680         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1681         if (!new)
1682                 return NULL;
1683         /* allocate each struct nfsd4_slot and data cache in one piece */
1684         for (i = 0; i < numslots; i++) {
1685                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1686                 if (!new->se_slots[i])
1687                         goto out_free;
1688         }
1689
1690         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1691         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1692
1693         return new;
1694 out_free:
1695         while (i--)
1696                 kfree(new->se_slots[i]);
1697         kfree(new);
1698         return NULL;
1699 }
1700
1701 static void free_conn(struct nfsd4_conn *c)
1702 {
1703         svc_xprt_put(c->cn_xprt);
1704         kfree(c);
1705 }
1706
1707 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1708 {
1709         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1710         struct nfs4_client *clp = c->cn_session->se_client;
1711
1712         spin_lock(&clp->cl_lock);
1713         if (!list_empty(&c->cn_persession)) {
1714                 list_del(&c->cn_persession);
1715                 free_conn(c);
1716         }
1717         nfsd4_probe_callback(clp);
1718         spin_unlock(&clp->cl_lock);
1719 }
1720
1721 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1722 {
1723         struct nfsd4_conn *conn;
1724
1725         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1726         if (!conn)
1727                 return NULL;
1728         svc_xprt_get(rqstp->rq_xprt);
1729         conn->cn_xprt = rqstp->rq_xprt;
1730         conn->cn_flags = flags;
1731         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1732         return conn;
1733 }
1734
1735 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1736 {
1737         conn->cn_session = ses;
1738         list_add(&conn->cn_persession, &ses->se_conns);
1739 }
1740
1741 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1742 {
1743         struct nfs4_client *clp = ses->se_client;
1744
1745         spin_lock(&clp->cl_lock);
1746         __nfsd4_hash_conn(conn, ses);
1747         spin_unlock(&clp->cl_lock);
1748 }
1749
1750 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1751 {
1752         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1753         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1754 }
1755
1756 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1757 {
1758         int ret;
1759
1760         nfsd4_hash_conn(conn, ses);
1761         ret = nfsd4_register_conn(conn);
1762         if (ret)
1763                 /* oops; xprt is already down: */
1764                 nfsd4_conn_lost(&conn->cn_xpt_user);
1765         /* We may have gained or lost a callback channel: */
1766         nfsd4_probe_callback_sync(ses->se_client);
1767 }
1768
1769 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1770 {
1771         u32 dir = NFS4_CDFC4_FORE;
1772
1773         if (cses->flags & SESSION4_BACK_CHAN)
1774                 dir |= NFS4_CDFC4_BACK;
1775         return alloc_conn(rqstp, dir);
1776 }
1777
1778 /* must be called under client_lock */
1779 static void nfsd4_del_conns(struct nfsd4_session *s)
1780 {
1781         struct nfs4_client *clp = s->se_client;
1782         struct nfsd4_conn *c;
1783
1784         spin_lock(&clp->cl_lock);
1785         while (!list_empty(&s->se_conns)) {
1786                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1787                 list_del_init(&c->cn_persession);
1788                 spin_unlock(&clp->cl_lock);
1789
1790                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1791                 free_conn(c);
1792
1793                 spin_lock(&clp->cl_lock);
1794         }
1795         spin_unlock(&clp->cl_lock);
1796 }
1797
1798 static void __free_session(struct nfsd4_session *ses)
1799 {
1800         free_session_slots(ses);
1801         kfree(ses);
1802 }
1803
1804 static void free_session(struct nfsd4_session *ses)
1805 {
1806         nfsd4_del_conns(ses);
1807         nfsd4_put_drc_mem(&ses->se_fchannel);
1808         __free_session(ses);
1809 }
1810
1811 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1812 {
1813         int idx;
1814         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1815
1816         new->se_client = clp;
1817         gen_sessionid(new);
1818
1819         INIT_LIST_HEAD(&new->se_conns);
1820
1821         new->se_cb_seq_nr = 1;
1822         new->se_flags = cses->flags;
1823         new->se_cb_prog = cses->callback_prog;
1824         new->se_cb_sec = cses->cb_sec;
1825         atomic_set(&new->se_ref, 0);
1826         idx = hash_sessionid(&new->se_sessionid);
1827         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1828         spin_lock(&clp->cl_lock);
1829         list_add(&new->se_perclnt, &clp->cl_sessions);
1830         spin_unlock(&clp->cl_lock);
1831
1832         {
1833                 struct sockaddr *sa = svc_addr(rqstp);
1834                 /*
1835                  * This is a little silly; with sessions there's no real
1836                  * use for the callback address.  Use the peer address
1837                  * as a reasonable default for now, but consider fixing
1838                  * the rpc client not to require an address in the
1839                  * future:
1840                  */
1841                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1842                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1843         }
1844 }
1845
1846 /* caller must hold client_lock */
1847 static struct nfsd4_session *
1848 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1849 {
1850         struct nfsd4_session *elem;
1851         int idx;
1852         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1853
1854         lockdep_assert_held(&nn->client_lock);
1855
1856         dump_sessionid(__func__, sessionid);
1857         idx = hash_sessionid(sessionid);
1858         /* Search in the appropriate list */
1859         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1860                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1861                             NFS4_MAX_SESSIONID_LEN)) {
1862                         return elem;
1863                 }
1864         }
1865
1866         dprintk("%s: session not found\n", __func__);
1867         return NULL;
1868 }
1869
1870 static struct nfsd4_session *
1871 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1872                 __be32 *ret)
1873 {
1874         struct nfsd4_session *session;
1875         __be32 status = nfserr_badsession;
1876
1877         session = __find_in_sessionid_hashtbl(sessionid, net);
1878         if (!session)
1879                 goto out;
1880         status = nfsd4_get_session_locked(session);
1881         if (status)
1882                 session = NULL;
1883 out:
1884         *ret = status;
1885         return session;
1886 }
1887
1888 /* caller must hold client_lock */
1889 static void
1890 unhash_session(struct nfsd4_session *ses)
1891 {
1892         struct nfs4_client *clp = ses->se_client;
1893         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1894
1895         lockdep_assert_held(&nn->client_lock);
1896
1897         list_del(&ses->se_hash);
1898         spin_lock(&ses->se_client->cl_lock);
1899         list_del(&ses->se_perclnt);
1900         spin_unlock(&ses->se_client->cl_lock);
1901 }
1902
1903 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1904 static int
1905 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1906 {
1907         /*
1908          * We're assuming the clid was not given out from a boot
1909          * precisely 2^32 (about 136 years) before this one.  That seems
1910          * a safe assumption:
1911          */
1912         if (clid->cl_boot == (u32)nn->boot_time)
1913                 return 0;
1914         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08llx\n",
1915                 clid->cl_boot, clid->cl_id, nn->boot_time);
1916         return 1;
1917 }
1918
1919 /*
1920  * XXX Should we use a slab cache ?
1921  * This type of memory management is somewhat inefficient, but we use it
1922  * anyway since SETCLIENTID is not a common operation.
1923  */
1924 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1925 {
1926         struct nfs4_client *clp;
1927         int i;
1928
1929         clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
1930         if (clp == NULL)
1931                 return NULL;
1932         xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
1933         if (clp->cl_name.data == NULL)
1934                 goto err_no_name;
1935         clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
1936                                                  sizeof(struct list_head),
1937                                                  GFP_KERNEL);
1938         if (!clp->cl_ownerstr_hashtbl)
1939                 goto err_no_hashtbl;
1940         for (i = 0; i < OWNER_HASH_SIZE; i++)
1941                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1942         INIT_LIST_HEAD(&clp->cl_sessions);
1943         idr_init(&clp->cl_stateids);
1944         atomic_set(&clp->cl_rpc_users, 0);
1945         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1946         INIT_LIST_HEAD(&clp->cl_idhash);
1947         INIT_LIST_HEAD(&clp->cl_openowners);
1948         INIT_LIST_HEAD(&clp->cl_delegations);
1949         INIT_LIST_HEAD(&clp->cl_lru);
1950         INIT_LIST_HEAD(&clp->cl_revoked);
1951 #ifdef CONFIG_NFSD_PNFS
1952         INIT_LIST_HEAD(&clp->cl_lo_states);
1953 #endif
1954         INIT_LIST_HEAD(&clp->async_copies);
1955         spin_lock_init(&clp->async_lock);
1956         spin_lock_init(&clp->cl_lock);
1957         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1958         return clp;
1959 err_no_hashtbl:
1960         kfree(clp->cl_name.data);
1961 err_no_name:
1962         kmem_cache_free(client_slab, clp);
1963         return NULL;
1964 }
1965
1966 static void __free_client(struct kref *k)
1967 {
1968         struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
1969         struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
1970
1971         free_svc_cred(&clp->cl_cred);
1972         kfree(clp->cl_ownerstr_hashtbl);
1973         kfree(clp->cl_name.data);
1974         kfree(clp->cl_nii_domain.data);
1975         kfree(clp->cl_nii_name.data);
1976         idr_destroy(&clp->cl_stateids);
1977         kmem_cache_free(client_slab, clp);
1978 }
1979
1980 static void drop_client(struct nfs4_client *clp)
1981 {
1982         kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
1983 }
1984
1985 static void
1986 free_client(struct nfs4_client *clp)
1987 {
1988         while (!list_empty(&clp->cl_sessions)) {
1989                 struct nfsd4_session *ses;
1990                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1991                                 se_perclnt);
1992                 list_del(&ses->se_perclnt);
1993                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1994                 free_session(ses);
1995         }
1996         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1997         if (clp->cl_nfsd_dentry) {
1998                 nfsd_client_rmdir(clp->cl_nfsd_dentry);
1999                 clp->cl_nfsd_dentry = NULL;
2000                 wake_up_all(&expiry_wq);
2001         }
2002         drop_client(clp);
2003 }
2004
2005 /* must be called under the client_lock */
2006 static void
2007 unhash_client_locked(struct nfs4_client *clp)
2008 {
2009         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2010         struct nfsd4_session *ses;
2011
2012         lockdep_assert_held(&nn->client_lock);
2013
2014         /* Mark the client as expired! */
2015         clp->cl_time = 0;
2016         /* Make it invisible */
2017         if (!list_empty(&clp->cl_idhash)) {
2018                 list_del_init(&clp->cl_idhash);
2019                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2020                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
2021                 else
2022                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2023         }
2024         list_del_init(&clp->cl_lru);
2025         spin_lock(&clp->cl_lock);
2026         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2027                 list_del_init(&ses->se_hash);
2028         spin_unlock(&clp->cl_lock);
2029 }
2030
2031 static void
2032 unhash_client(struct nfs4_client *clp)
2033 {
2034         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2035
2036         spin_lock(&nn->client_lock);
2037         unhash_client_locked(clp);
2038         spin_unlock(&nn->client_lock);
2039 }
2040
2041 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
2042 {
2043         if (atomic_read(&clp->cl_rpc_users))
2044                 return nfserr_jukebox;
2045         unhash_client_locked(clp);
2046         return nfs_ok;
2047 }
2048
2049 static void
2050 __destroy_client(struct nfs4_client *clp)
2051 {
2052         int i;
2053         struct nfs4_openowner *oo;
2054         struct nfs4_delegation *dp;
2055         struct list_head reaplist;
2056
2057         INIT_LIST_HEAD(&reaplist);
2058         spin_lock(&state_lock);
2059         while (!list_empty(&clp->cl_delegations)) {
2060                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2061                 WARN_ON(!unhash_delegation_locked(dp));
2062                 list_add(&dp->dl_recall_lru, &reaplist);
2063         }
2064         spin_unlock(&state_lock);
2065         while (!list_empty(&reaplist)) {
2066                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2067                 list_del_init(&dp->dl_recall_lru);
2068                 destroy_unhashed_deleg(dp);
2069         }
2070         while (!list_empty(&clp->cl_revoked)) {
2071                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2072                 list_del_init(&dp->dl_recall_lru);
2073                 nfs4_put_stid(&dp->dl_stid);
2074         }
2075         while (!list_empty(&clp->cl_openowners)) {
2076                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2077                 nfs4_get_stateowner(&oo->oo_owner);
2078                 release_openowner(oo);
2079         }
2080         for (i = 0; i < OWNER_HASH_SIZE; i++) {
2081                 struct nfs4_stateowner *so, *tmp;
2082
2083                 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2084                                          so_strhash) {
2085                         /* Should be no openowners at this point */
2086                         WARN_ON_ONCE(so->so_is_open_owner);
2087                         remove_blocked_locks(lockowner(so));
2088                 }
2089         }
2090         nfsd4_return_all_client_layouts(clp);
2091         nfsd4_shutdown_copy(clp);
2092         nfsd4_shutdown_callback(clp);
2093         if (clp->cl_cb_conn.cb_xprt)
2094                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2095         free_client(clp);
2096         wake_up_all(&expiry_wq);
2097 }
2098
2099 static void
2100 destroy_client(struct nfs4_client *clp)
2101 {
2102         unhash_client(clp);
2103         __destroy_client(clp);
2104 }
2105
2106 static void inc_reclaim_complete(struct nfs4_client *clp)
2107 {
2108         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2109
2110         if (!nn->track_reclaim_completes)
2111                 return;
2112         if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2113                 return;
2114         if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2115                         nn->reclaim_str_hashtbl_size) {
2116                 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2117                                 clp->net->ns.inum);
2118                 nfsd4_end_grace(nn);
2119         }
2120 }
2121
2122 static void expire_client(struct nfs4_client *clp)
2123 {
2124         unhash_client(clp);
2125         nfsd4_client_record_remove(clp);
2126         __destroy_client(clp);
2127 }
2128
2129 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2130 {
2131         memcpy(target->cl_verifier.data, source->data,
2132                         sizeof(target->cl_verifier.data));
2133 }
2134
2135 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2136 {
2137         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2138         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2139 }
2140
2141 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2142 {
2143         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2144         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2145                                                                 GFP_KERNEL);
2146         target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2147         if ((source->cr_principal && !target->cr_principal) ||
2148             (source->cr_raw_principal && !target->cr_raw_principal) ||
2149             (source->cr_targ_princ && !target->cr_targ_princ))
2150                 return -ENOMEM;
2151
2152         target->cr_flavor = source->cr_flavor;
2153         target->cr_uid = source->cr_uid;
2154         target->cr_gid = source->cr_gid;
2155         target->cr_group_info = source->cr_group_info;
2156         get_group_info(target->cr_group_info);
2157         target->cr_gss_mech = source->cr_gss_mech;
2158         if (source->cr_gss_mech)
2159                 gss_mech_get(source->cr_gss_mech);
2160         return 0;
2161 }
2162
2163 static int
2164 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2165 {
2166         if (o1->len < o2->len)
2167                 return -1;
2168         if (o1->len > o2->len)
2169                 return 1;
2170         return memcmp(o1->data, o2->data, o1->len);
2171 }
2172
2173 static int
2174 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2175 {
2176         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2177 }
2178
2179 static int
2180 same_clid(clientid_t *cl1, clientid_t *cl2)
2181 {
2182         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2183 }
2184
2185 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2186 {
2187         int i;
2188
2189         if (g1->ngroups != g2->ngroups)
2190                 return false;
2191         for (i=0; i<g1->ngroups; i++)
2192                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2193                         return false;
2194         return true;
2195 }
2196
2197 /*
2198  * RFC 3530 language requires clid_inuse be returned when the
2199  * "principal" associated with a requests differs from that previously
2200  * used.  We use uid, gid's, and gss principal string as our best
2201  * approximation.  We also don't want to allow non-gss use of a client
2202  * established using gss: in theory cr_principal should catch that
2203  * change, but in practice cr_principal can be null even in the gss case
2204  * since gssd doesn't always pass down a principal string.
2205  */
2206 static bool is_gss_cred(struct svc_cred *cr)
2207 {
2208         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2209         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2210 }
2211
2212
2213 static bool
2214 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2215 {
2216         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2217                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2218                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2219                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2220                 return false;
2221         /* XXX: check that cr_targ_princ fields match ? */
2222         if (cr1->cr_principal == cr2->cr_principal)
2223                 return true;
2224         if (!cr1->cr_principal || !cr2->cr_principal)
2225                 return false;
2226         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2227 }
2228
2229 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2230 {
2231         struct svc_cred *cr = &rqstp->rq_cred;
2232         u32 service;
2233
2234         if (!cr->cr_gss_mech)
2235                 return false;
2236         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2237         return service == RPC_GSS_SVC_INTEGRITY ||
2238                service == RPC_GSS_SVC_PRIVACY;
2239 }
2240
2241 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2242 {
2243         struct svc_cred *cr = &rqstp->rq_cred;
2244
2245         if (!cl->cl_mach_cred)
2246                 return true;
2247         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2248                 return false;
2249         if (!svc_rqst_integrity_protected(rqstp))
2250                 return false;
2251         if (cl->cl_cred.cr_raw_principal)
2252                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2253                                                 cr->cr_raw_principal);
2254         if (!cr->cr_principal)
2255                 return false;
2256         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2257 }
2258
2259 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2260 {
2261         __be32 verf[2];
2262
2263         /*
2264          * This is opaque to client, so no need to byte-swap. Use
2265          * __force to keep sparse happy
2266          */
2267         verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2268         verf[1] = (__force __be32)nn->clverifier_counter++;
2269         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2270 }
2271
2272 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2273 {
2274         clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2275         clp->cl_clientid.cl_id = nn->clientid_counter++;
2276         gen_confirm(clp, nn);
2277 }
2278
2279 static struct nfs4_stid *
2280 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2281 {
2282         struct nfs4_stid *ret;
2283
2284         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2285         if (!ret || !ret->sc_type)
2286                 return NULL;
2287         return ret;
2288 }
2289
2290 static struct nfs4_stid *
2291 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2292 {
2293         struct nfs4_stid *s;
2294
2295         spin_lock(&cl->cl_lock);
2296         s = find_stateid_locked(cl, t);
2297         if (s != NULL) {
2298                 if (typemask & s->sc_type)
2299                         refcount_inc(&s->sc_count);
2300                 else
2301                         s = NULL;
2302         }
2303         spin_unlock(&cl->cl_lock);
2304         return s;
2305 }
2306
2307 static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2308 {
2309         struct nfsdfs_client *nc;
2310         nc = get_nfsdfs_client(inode);
2311         if (!nc)
2312                 return NULL;
2313         return container_of(nc, struct nfs4_client, cl_nfsdfs);
2314 }
2315
2316 static void seq_quote_mem(struct seq_file *m, char *data, int len)
2317 {
2318         seq_printf(m, "\"");
2319         seq_escape_mem_ascii(m, data, len);
2320         seq_printf(m, "\"");
2321 }
2322
2323 static int client_info_show(struct seq_file *m, void *v)
2324 {
2325         struct inode *inode = m->private;
2326         struct nfs4_client *clp;
2327         u64 clid;
2328
2329         clp = get_nfsdfs_clp(inode);
2330         if (!clp)
2331                 return -ENXIO;
2332         memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2333         seq_printf(m, "clientid: 0x%llx\n", clid);
2334         seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2335         seq_printf(m, "name: ");
2336         seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2337         seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2338         if (clp->cl_nii_domain.data) {
2339                 seq_printf(m, "Implementation domain: ");
2340                 seq_quote_mem(m, clp->cl_nii_domain.data,
2341                                         clp->cl_nii_domain.len);
2342                 seq_printf(m, "\nImplementation name: ");
2343                 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2344                 seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2345                         clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2346         }
2347         drop_client(clp);
2348
2349         return 0;
2350 }
2351
2352 static int client_info_open(struct inode *inode, struct file *file)
2353 {
2354         return single_open(file, client_info_show, inode);
2355 }
2356
2357 static const struct file_operations client_info_fops = {
2358         .open           = client_info_open,
2359         .read           = seq_read,
2360         .llseek         = seq_lseek,
2361         .release        = single_release,
2362 };
2363
2364 static void *states_start(struct seq_file *s, loff_t *pos)
2365         __acquires(&clp->cl_lock)
2366 {
2367         struct nfs4_client *clp = s->private;
2368         unsigned long id = *pos;
2369         void *ret;
2370
2371         spin_lock(&clp->cl_lock);
2372         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2373         *pos = id;
2374         return ret;
2375 }
2376
2377 static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2378 {
2379         struct nfs4_client *clp = s->private;
2380         unsigned long id = *pos;
2381         void *ret;
2382
2383         id = *pos;
2384         id++;
2385         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2386         *pos = id;
2387         return ret;
2388 }
2389
2390 static void states_stop(struct seq_file *s, void *v)
2391         __releases(&clp->cl_lock)
2392 {
2393         struct nfs4_client *clp = s->private;
2394
2395         spin_unlock(&clp->cl_lock);
2396 }
2397
2398 static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2399 {
2400         struct inode *inode = f->nf_inode;
2401
2402         seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2403                                         MAJOR(inode->i_sb->s_dev),
2404                                          MINOR(inode->i_sb->s_dev),
2405                                          inode->i_ino);
2406 }
2407
2408 static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2409 {
2410         seq_printf(s, "owner: ");
2411         seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2412 }
2413
2414 static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2415 {
2416         struct nfs4_ol_stateid *ols;
2417         struct nfs4_file *nf;
2418         struct nfsd_file *file;
2419         struct nfs4_stateowner *oo;
2420         unsigned int access, deny;
2421
2422         if (st->sc_type != NFS4_OPEN_STID && st->sc_type != NFS4_LOCK_STID)
2423                 return 0; /* XXX: or SEQ_SKIP? */
2424         ols = openlockstateid(st);
2425         oo = ols->st_stateowner;
2426         nf = st->sc_file;
2427         file = find_any_file(nf);
2428
2429         seq_printf(s, "- 0x%16phN: { type: open, ", &st->sc_stateid);
2430
2431         access = bmap_to_share_mode(ols->st_access_bmap);
2432         deny   = bmap_to_share_mode(ols->st_deny_bmap);
2433
2434         seq_printf(s, "access: %s%s, ",
2435                 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2436                 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2437         seq_printf(s, "deny: %s%s, ",
2438                 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2439                 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2440
2441         nfs4_show_superblock(s, file);
2442         seq_printf(s, ", ");
2443         nfs4_show_owner(s, oo);
2444         seq_printf(s, " }\n");
2445         nfsd_file_put(file);
2446
2447         return 0;
2448 }
2449
2450 static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2451 {
2452         struct nfs4_ol_stateid *ols;
2453         struct nfs4_file *nf;
2454         struct nfsd_file *file;
2455         struct nfs4_stateowner *oo;
2456
2457         ols = openlockstateid(st);
2458         oo = ols->st_stateowner;
2459         nf = st->sc_file;
2460         file = find_any_file(nf);
2461
2462         seq_printf(s, "- 0x%16phN: { type: lock, ", &st->sc_stateid);
2463
2464         /*
2465          * Note: a lock stateid isn't really the same thing as a lock,
2466          * it's the locking state held by one owner on a file, and there
2467          * may be multiple (or no) lock ranges associated with it.
2468          * (Same for the matter is true of open stateids.)
2469          */
2470
2471         nfs4_show_superblock(s, file);
2472         /* XXX: open stateid? */
2473         seq_printf(s, ", ");
2474         nfs4_show_owner(s, oo);
2475         seq_printf(s, " }\n");
2476         nfsd_file_put(file);
2477
2478         return 0;
2479 }
2480
2481 static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2482 {
2483         struct nfs4_delegation *ds;
2484         struct nfs4_file *nf;
2485         struct nfsd_file *file;
2486
2487         ds = delegstateid(st);
2488         nf = st->sc_file;
2489         file = nf->fi_deleg_file;
2490
2491         seq_printf(s, "- 0x%16phN: { type: deleg, ", &st->sc_stateid);
2492
2493         /* Kinda dead code as long as we only support read delegs: */
2494         seq_printf(s, "access: %s, ",
2495                 ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2496
2497         /* XXX: lease time, whether it's being recalled. */
2498
2499         nfs4_show_superblock(s, file);
2500         seq_printf(s, " }\n");
2501
2502         return 0;
2503 }
2504
2505 static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2506 {
2507         struct nfs4_layout_stateid *ls;
2508         struct nfsd_file *file;
2509
2510         ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2511         file = ls->ls_file;
2512
2513         seq_printf(s, "- 0x%16phN: { type: layout, ", &st->sc_stateid);
2514
2515         /* XXX: What else would be useful? */
2516
2517         nfs4_show_superblock(s, file);
2518         seq_printf(s, " }\n");
2519
2520         return 0;
2521 }
2522
2523 static int states_show(struct seq_file *s, void *v)
2524 {
2525         struct nfs4_stid *st = v;
2526
2527         switch (st->sc_type) {
2528         case NFS4_OPEN_STID:
2529                 return nfs4_show_open(s, st);
2530         case NFS4_LOCK_STID:
2531                 return nfs4_show_lock(s, st);
2532         case NFS4_DELEG_STID:
2533                 return nfs4_show_deleg(s, st);
2534         case NFS4_LAYOUT_STID:
2535                 return nfs4_show_layout(s, st);
2536         default:
2537                 return 0; /* XXX: or SEQ_SKIP? */
2538         }
2539         /* XXX: copy stateids? */
2540 }
2541
2542 static struct seq_operations states_seq_ops = {
2543         .start = states_start,
2544         .next = states_next,
2545         .stop = states_stop,
2546         .show = states_show
2547 };
2548
2549 static int client_states_open(struct inode *inode, struct file *file)
2550 {
2551         struct seq_file *s;
2552         struct nfs4_client *clp;
2553         int ret;
2554
2555         clp = get_nfsdfs_clp(inode);
2556         if (!clp)
2557                 return -ENXIO;
2558
2559         ret = seq_open(file, &states_seq_ops);
2560         if (ret)
2561                 return ret;
2562         s = file->private_data;
2563         s->private = clp;
2564         return 0;
2565 }
2566
2567 static int client_opens_release(struct inode *inode, struct file *file)
2568 {
2569         struct seq_file *m = file->private_data;
2570         struct nfs4_client *clp = m->private;
2571
2572         /* XXX: alternatively, we could get/drop in seq start/stop */
2573         drop_client(clp);
2574         return 0;
2575 }
2576
2577 static const struct file_operations client_states_fops = {
2578         .open           = client_states_open,
2579         .read           = seq_read,
2580         .llseek         = seq_lseek,
2581         .release        = client_opens_release,
2582 };
2583
2584 /*
2585  * Normally we refuse to destroy clients that are in use, but here the
2586  * administrator is telling us to just do it.  We also want to wait
2587  * so the caller has a guarantee that the client's locks are gone by
2588  * the time the write returns:
2589  */
2590 static void force_expire_client(struct nfs4_client *clp)
2591 {
2592         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2593         bool already_expired;
2594
2595         spin_lock(&clp->cl_lock);
2596         clp->cl_time = 0;
2597         spin_unlock(&clp->cl_lock);
2598
2599         wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2600         spin_lock(&nn->client_lock);
2601         already_expired = list_empty(&clp->cl_lru);
2602         if (!already_expired)
2603                 unhash_client_locked(clp);
2604         spin_unlock(&nn->client_lock);
2605
2606         if (!already_expired)
2607                 expire_client(clp);
2608         else
2609                 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2610 }
2611
2612 static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2613                                    size_t size, loff_t *pos)
2614 {
2615         char *data;
2616         struct nfs4_client *clp;
2617
2618         data = simple_transaction_get(file, buf, size);
2619         if (IS_ERR(data))
2620                 return PTR_ERR(data);
2621         if (size != 7 || 0 != memcmp(data, "expire\n", 7))
2622                 return -EINVAL;
2623         clp = get_nfsdfs_clp(file_inode(file));
2624         if (!clp)
2625                 return -ENXIO;
2626         force_expire_client(clp);
2627         drop_client(clp);
2628         return 7;
2629 }
2630
2631 static const struct file_operations client_ctl_fops = {
2632         .write          = client_ctl_write,
2633         .release        = simple_transaction_release,
2634 };
2635
2636 static const struct tree_descr client_files[] = {
2637         [0] = {"info", &client_info_fops, S_IRUSR},
2638         [1] = {"states", &client_states_fops, S_IRUSR},
2639         [2] = {"ctl", &client_ctl_fops, S_IRUSR|S_IWUSR},
2640         [3] = {""},
2641 };
2642
2643 static struct nfs4_client *create_client(struct xdr_netobj name,
2644                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2645 {
2646         struct nfs4_client *clp;
2647         struct sockaddr *sa = svc_addr(rqstp);
2648         int ret;
2649         struct net *net = SVC_NET(rqstp);
2650         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2651
2652         clp = alloc_client(name);
2653         if (clp == NULL)
2654                 return NULL;
2655
2656         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2657         if (ret) {
2658                 free_client(clp);
2659                 return NULL;
2660         }
2661         gen_clid(clp, nn);
2662         kref_init(&clp->cl_nfsdfs.cl_ref);
2663         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2664         clp->cl_time = ktime_get_boottime_seconds();
2665         clear_bit(0, &clp->cl_cb_slot_busy);
2666         copy_verf(clp, verf);
2667         memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
2668         clp->cl_cb_session = NULL;
2669         clp->net = net;
2670         clp->cl_nfsd_dentry = nfsd_client_mkdir(nn, &clp->cl_nfsdfs,
2671                         clp->cl_clientid.cl_id - nn->clientid_base,
2672                         client_files);
2673         if (!clp->cl_nfsd_dentry) {
2674                 free_client(clp);
2675                 return NULL;
2676         }
2677         return clp;
2678 }
2679
2680 static void
2681 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2682 {
2683         struct rb_node **new = &(root->rb_node), *parent = NULL;
2684         struct nfs4_client *clp;
2685
2686         while (*new) {
2687                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2688                 parent = *new;
2689
2690                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2691                         new = &((*new)->rb_left);
2692                 else
2693                         new = &((*new)->rb_right);
2694         }
2695
2696         rb_link_node(&new_clp->cl_namenode, parent, new);
2697         rb_insert_color(&new_clp->cl_namenode, root);
2698 }
2699
2700 static struct nfs4_client *
2701 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2702 {
2703         int cmp;
2704         struct rb_node *node = root->rb_node;
2705         struct nfs4_client *clp;
2706
2707         while (node) {
2708                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2709                 cmp = compare_blob(&clp->cl_name, name);
2710                 if (cmp > 0)
2711                         node = node->rb_left;
2712                 else if (cmp < 0)
2713                         node = node->rb_right;
2714                 else
2715                         return clp;
2716         }
2717         return NULL;
2718 }
2719
2720 static void
2721 add_to_unconfirmed(struct nfs4_client *clp)
2722 {
2723         unsigned int idhashval;
2724         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2725
2726         lockdep_assert_held(&nn->client_lock);
2727
2728         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2729         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2730         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2731         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2732         renew_client_locked(clp);
2733 }
2734
2735 static void
2736 move_to_confirmed(struct nfs4_client *clp)
2737 {
2738         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2739         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2740
2741         lockdep_assert_held(&nn->client_lock);
2742
2743         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2744         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2745         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2746         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2747         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2748         renew_client_locked(clp);
2749 }
2750
2751 static struct nfs4_client *
2752 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2753 {
2754         struct nfs4_client *clp;
2755         unsigned int idhashval = clientid_hashval(clid->cl_id);
2756
2757         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2758                 if (same_clid(&clp->cl_clientid, clid)) {
2759                         if ((bool)clp->cl_minorversion != sessions)
2760                                 return NULL;
2761                         renew_client_locked(clp);
2762                         return clp;
2763                 }
2764         }
2765         return NULL;
2766 }
2767
2768 static struct nfs4_client *
2769 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2770 {
2771         struct list_head *tbl = nn->conf_id_hashtbl;
2772
2773         lockdep_assert_held(&nn->client_lock);
2774         return find_client_in_id_table(tbl, clid, sessions);
2775 }
2776
2777 static struct nfs4_client *
2778 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2779 {
2780         struct list_head *tbl = nn->unconf_id_hashtbl;
2781
2782         lockdep_assert_held(&nn->client_lock);
2783         return find_client_in_id_table(tbl, clid, sessions);
2784 }
2785
2786 static bool clp_used_exchangeid(struct nfs4_client *clp)
2787 {
2788         return clp->cl_exchange_flags != 0;
2789 }
2790
2791 static struct nfs4_client *
2792 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2793 {
2794         lockdep_assert_held(&nn->client_lock);
2795         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2796 }
2797
2798 static struct nfs4_client *
2799 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2800 {
2801         lockdep_assert_held(&nn->client_lock);
2802         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2803 }
2804
2805 static void
2806 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2807 {
2808         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2809         struct sockaddr *sa = svc_addr(rqstp);
2810         u32 scopeid = rpc_get_scope_id(sa);
2811         unsigned short expected_family;
2812
2813         /* Currently, we only support tcp and tcp6 for the callback channel */
2814         if (se->se_callback_netid_len == 3 &&
2815             !memcmp(se->se_callback_netid_val, "tcp", 3))
2816                 expected_family = AF_INET;
2817         else if (se->se_callback_netid_len == 4 &&
2818                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2819                 expected_family = AF_INET6;
2820         else
2821                 goto out_err;
2822
2823         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2824                                             se->se_callback_addr_len,
2825                                             (struct sockaddr *)&conn->cb_addr,
2826                                             sizeof(conn->cb_addr));
2827
2828         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2829                 goto out_err;
2830
2831         if (conn->cb_addr.ss_family == AF_INET6)
2832                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2833
2834         conn->cb_prog = se->se_callback_prog;
2835         conn->cb_ident = se->se_callback_ident;
2836         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2837         return;
2838 out_err:
2839         conn->cb_addr.ss_family = AF_UNSPEC;
2840         conn->cb_addrlen = 0;
2841         dprintk("NFSD: this client (clientid %08x/%08x) "
2842                 "will not receive delegations\n",
2843                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2844
2845         return;
2846 }
2847
2848 /*
2849  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2850  */
2851 static void
2852 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2853 {
2854         struct xdr_buf *buf = resp->xdr.buf;
2855         struct nfsd4_slot *slot = resp->cstate.slot;
2856         unsigned int base;
2857
2858         dprintk("--> %s slot %p\n", __func__, slot);
2859
2860         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2861         slot->sl_opcnt = resp->opcnt;
2862         slot->sl_status = resp->cstate.status;
2863         free_svc_cred(&slot->sl_cred);
2864         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2865
2866         if (!nfsd4_cache_this(resp)) {
2867                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2868                 return;
2869         }
2870         slot->sl_flags |= NFSD4_SLOT_CACHED;
2871
2872         base = resp->cstate.data_offset;
2873         slot->sl_datalen = buf->len - base;
2874         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2875                 WARN(1, "%s: sessions DRC could not cache compound\n",
2876                      __func__);
2877         return;
2878 }
2879
2880 /*
2881  * Encode the replay sequence operation from the slot values.
2882  * If cachethis is FALSE encode the uncached rep error on the next
2883  * operation which sets resp->p and increments resp->opcnt for
2884  * nfs4svc_encode_compoundres.
2885  *
2886  */
2887 static __be32
2888 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2889                           struct nfsd4_compoundres *resp)
2890 {
2891         struct nfsd4_op *op;
2892         struct nfsd4_slot *slot = resp->cstate.slot;
2893
2894         /* Encode the replayed sequence operation */
2895         op = &args->ops[resp->opcnt - 1];
2896         nfsd4_encode_operation(resp, op);
2897
2898         if (slot->sl_flags & NFSD4_SLOT_CACHED)
2899                 return op->status;
2900         if (args->opcnt == 1) {
2901                 /*
2902                  * The original operation wasn't a solo sequence--we
2903                  * always cache those--so this retry must not match the
2904                  * original:
2905                  */
2906                 op->status = nfserr_seq_false_retry;
2907         } else {
2908                 op = &args->ops[resp->opcnt++];
2909                 op->status = nfserr_retry_uncached_rep;
2910                 nfsd4_encode_operation(resp, op);
2911         }
2912         return op->status;
2913 }
2914
2915 /*
2916  * The sequence operation is not cached because we can use the slot and
2917  * session values.
2918  */
2919 static __be32
2920 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2921                          struct nfsd4_sequence *seq)
2922 {
2923         struct nfsd4_slot *slot = resp->cstate.slot;
2924         struct xdr_stream *xdr = &resp->xdr;
2925         __be32 *p;
2926         __be32 status;
2927
2928         dprintk("--> %s slot %p\n", __func__, slot);
2929
2930         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2931         if (status)
2932                 return status;
2933
2934         p = xdr_reserve_space(xdr, slot->sl_datalen);
2935         if (!p) {
2936                 WARN_ON_ONCE(1);
2937                 return nfserr_serverfault;
2938         }
2939         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2940         xdr_commit_encode(xdr);
2941
2942         resp->opcnt = slot->sl_opcnt;
2943         return slot->sl_status;
2944 }
2945
2946 /*
2947  * Set the exchange_id flags returned by the server.
2948  */
2949 static void
2950 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2951 {
2952 #ifdef CONFIG_NFSD_PNFS
2953         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2954 #else
2955         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2956 #endif
2957
2958         /* Referrals are supported, Migration is not. */
2959         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2960
2961         /* set the wire flags to return to client. */
2962         clid->flags = new->cl_exchange_flags;
2963 }
2964
2965 static bool client_has_openowners(struct nfs4_client *clp)
2966 {
2967         struct nfs4_openowner *oo;
2968
2969         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2970                 if (!list_empty(&oo->oo_owner.so_stateids))
2971                         return true;
2972         }
2973         return false;
2974 }
2975
2976 static bool client_has_state(struct nfs4_client *clp)
2977 {
2978         return client_has_openowners(clp)
2979 #ifdef CONFIG_NFSD_PNFS
2980                 || !list_empty(&clp->cl_lo_states)
2981 #endif
2982                 || !list_empty(&clp->cl_delegations)
2983                 || !list_empty(&clp->cl_sessions)
2984                 || !list_empty(&clp->async_copies);
2985 }
2986
2987 static __be32 copy_impl_id(struct nfs4_client *clp,
2988                                 struct nfsd4_exchange_id *exid)
2989 {
2990         if (!exid->nii_domain.data)
2991                 return 0;
2992         xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
2993         if (!clp->cl_nii_domain.data)
2994                 return nfserr_jukebox;
2995         xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
2996         if (!clp->cl_nii_name.data)
2997                 return nfserr_jukebox;
2998         clp->cl_nii_time = exid->nii_time;
2999         return 0;
3000 }
3001
3002 __be32
3003 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3004                 union nfsd4_op_u *u)
3005 {
3006         struct nfsd4_exchange_id *exid = &u->exchange_id;
3007         struct nfs4_client *conf, *new;
3008         struct nfs4_client *unconf = NULL;
3009         __be32 status;
3010         char                    addr_str[INET6_ADDRSTRLEN];
3011         nfs4_verifier           verf = exid->verifier;
3012         struct sockaddr         *sa = svc_addr(rqstp);
3013         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3014         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3015
3016         rpc_ntop(sa, addr_str, sizeof(addr_str));
3017         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3018                 "ip_addr=%s flags %x, spa_how %d\n",
3019                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
3020                 addr_str, exid->flags, exid->spa_how);
3021
3022         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3023                 return nfserr_inval;
3024
3025         new = create_client(exid->clname, rqstp, &verf);
3026         if (new == NULL)
3027                 return nfserr_jukebox;
3028         status = copy_impl_id(new, exid);
3029         if (status)
3030                 goto out_nolock;
3031
3032         switch (exid->spa_how) {
3033         case SP4_MACH_CRED:
3034                 exid->spo_must_enforce[0] = 0;
3035                 exid->spo_must_enforce[1] = (
3036                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
3037                         1 << (OP_EXCHANGE_ID - 32) |
3038                         1 << (OP_CREATE_SESSION - 32) |
3039                         1 << (OP_DESTROY_SESSION - 32) |
3040                         1 << (OP_DESTROY_CLIENTID - 32));
3041
3042                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3043                                         1 << (OP_OPEN_DOWNGRADE) |
3044                                         1 << (OP_LOCKU) |
3045                                         1 << (OP_DELEGRETURN));
3046
3047                 exid->spo_must_allow[1] &= (
3048                                         1 << (OP_TEST_STATEID - 32) |
3049                                         1 << (OP_FREE_STATEID - 32));
3050                 if (!svc_rqst_integrity_protected(rqstp)) {
3051                         status = nfserr_inval;
3052                         goto out_nolock;
3053                 }
3054                 /*
3055                  * Sometimes userspace doesn't give us a principal.
3056                  * Which is a bug, really.  Anyway, we can't enforce
3057                  * MACH_CRED in that case, better to give up now:
3058                  */
3059                 if (!new->cl_cred.cr_principal &&
3060                                         !new->cl_cred.cr_raw_principal) {
3061                         status = nfserr_serverfault;
3062                         goto out_nolock;
3063                 }
3064                 new->cl_mach_cred = true;
3065         case SP4_NONE:
3066                 break;
3067         default:                                /* checked by xdr code */
3068                 WARN_ON_ONCE(1);
3069                 /* fall through */
3070         case SP4_SSV:
3071                 status = nfserr_encr_alg_unsupp;
3072                 goto out_nolock;
3073         }
3074
3075         /* Cases below refer to rfc 5661 section 18.35.4: */
3076         spin_lock(&nn->client_lock);
3077         conf = find_confirmed_client_by_name(&exid->clname, nn);
3078         if (conf) {
3079                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3080                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3081
3082                 if (update) {
3083                         if (!clp_used_exchangeid(conf)) { /* buggy client */
3084                                 status = nfserr_inval;
3085                                 goto out;
3086                         }
3087                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
3088                                 status = nfserr_wrong_cred;
3089                                 goto out;
3090                         }
3091                         if (!creds_match) { /* case 9 */
3092                                 status = nfserr_perm;
3093                                 goto out;
3094                         }
3095                         if (!verfs_match) { /* case 8 */
3096                                 status = nfserr_not_same;
3097                                 goto out;
3098                         }
3099                         /* case 6 */
3100                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3101                         goto out_copy;
3102                 }
3103                 if (!creds_match) { /* case 3 */
3104                         if (client_has_state(conf)) {
3105                                 status = nfserr_clid_inuse;
3106                                 goto out;
3107                         }
3108                         goto out_new;
3109                 }
3110                 if (verfs_match) { /* case 2 */
3111                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3112                         goto out_copy;
3113                 }
3114                 /* case 5, client reboot */
3115                 conf = NULL;
3116                 goto out_new;
3117         }
3118
3119         if (update) { /* case 7 */
3120                 status = nfserr_noent;
3121                 goto out;
3122         }
3123
3124         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
3125         if (unconf) /* case 4, possible retry or client restart */
3126                 unhash_client_locked(unconf);
3127
3128         /* case 1 (normal case) */
3129 out_new:
3130         if (conf) {
3131                 status = mark_client_expired_locked(conf);
3132                 if (status)
3133                         goto out;
3134         }
3135         new->cl_minorversion = cstate->minorversion;
3136         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3137         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3138
3139         add_to_unconfirmed(new);
3140         swap(new, conf);
3141 out_copy:
3142         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3143         exid->clientid.cl_id = conf->cl_clientid.cl_id;
3144
3145         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3146         nfsd4_set_ex_flags(conf, exid);
3147
3148         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3149                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3150         status = nfs_ok;
3151
3152 out:
3153         spin_unlock(&nn->client_lock);
3154 out_nolock:
3155         if (new)
3156                 expire_client(new);
3157         if (unconf)
3158                 expire_client(unconf);
3159         return status;
3160 }
3161
3162 static __be32
3163 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3164 {
3165         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3166                 slot_seqid);
3167
3168         /* The slot is in use, and no response has been sent. */
3169         if (slot_inuse) {
3170                 if (seqid == slot_seqid)
3171                         return nfserr_jukebox;
3172                 else
3173                         return nfserr_seq_misordered;
3174         }
3175         /* Note unsigned 32-bit arithmetic handles wraparound: */
3176         if (likely(seqid == slot_seqid + 1))
3177                 return nfs_ok;
3178         if (seqid == slot_seqid)
3179                 return nfserr_replay_cache;
3180         return nfserr_seq_misordered;
3181 }
3182
3183 /*
3184  * Cache the create session result into the create session single DRC
3185  * slot cache by saving the xdr structure. sl_seqid has been set.
3186  * Do this for solo or embedded create session operations.
3187  */
3188 static void
3189 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3190                            struct nfsd4_clid_slot *slot, __be32 nfserr)
3191 {
3192         slot->sl_status = nfserr;
3193         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3194 }
3195
3196 static __be32
3197 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3198                             struct nfsd4_clid_slot *slot)
3199 {
3200         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3201         return slot->sl_status;
3202 }
3203
3204 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3205                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3206                         1 +     /* MIN tag is length with zero, only length */ \
3207                         3 +     /* version, opcount, opcode */ \
3208                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3209                                 /* seqid, slotID, slotID, cache */ \
3210                         4 ) * sizeof(__be32))
3211
3212 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3213                         2 +     /* verifier: AUTH_NULL, length 0 */\
3214                         1 +     /* status */ \
3215                         1 +     /* MIN tag is length with zero, only length */ \
3216                         3 +     /* opcount, opcode, opstatus*/ \
3217                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3218                                 /* seqid, slotID, slotID, slotID, status */ \
3219                         5 ) * sizeof(__be32))
3220
3221 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3222 {
3223         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3224
3225         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3226                 return nfserr_toosmall;
3227         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3228                 return nfserr_toosmall;
3229         ca->headerpadsz = 0;
3230         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3231         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3232         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3233         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3234                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3235         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3236         /*
3237          * Note decreasing slot size below client's request may make it
3238          * difficult for client to function correctly, whereas
3239          * decreasing the number of slots will (just?) affect
3240          * performance.  When short on memory we therefore prefer to
3241          * decrease number of slots instead of their size.  Clients that
3242          * request larger slots than they need will get poor results:
3243          * Note that we always allow at least one slot, because our
3244          * accounting is soft and provides no guarantees either way.
3245          */
3246         ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3247
3248         return nfs_ok;
3249 }
3250
3251 /*
3252  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3253  * These are based on similar macros in linux/sunrpc/msg_prot.h .
3254  */
3255 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3256         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3257
3258 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3259         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3260
3261 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
3262                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3263 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
3264                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3265                                  sizeof(__be32))
3266
3267 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3268 {
3269         ca->headerpadsz = 0;
3270
3271         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3272                 return nfserr_toosmall;
3273         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3274                 return nfserr_toosmall;
3275         ca->maxresp_cached = 0;
3276         if (ca->maxops < 2)
3277                 return nfserr_toosmall;
3278
3279         return nfs_ok;
3280 }
3281
3282 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3283 {
3284         switch (cbs->flavor) {
3285         case RPC_AUTH_NULL:
3286         case RPC_AUTH_UNIX:
3287                 return nfs_ok;
3288         default:
3289                 /*
3290                  * GSS case: the spec doesn't allow us to return this
3291                  * error.  But it also doesn't allow us not to support
3292                  * GSS.
3293                  * I'd rather this fail hard than return some error the
3294                  * client might think it can already handle:
3295                  */
3296                 return nfserr_encr_alg_unsupp;
3297         }
3298 }
3299
3300 __be32
3301 nfsd4_create_session(struct svc_rqst *rqstp,
3302                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3303 {
3304         struct nfsd4_create_session *cr_ses = &u->create_session;
3305         struct sockaddr *sa = svc_addr(rqstp);
3306         struct nfs4_client *conf, *unconf;
3307         struct nfs4_client *old = NULL;
3308         struct nfsd4_session *new;
3309         struct nfsd4_conn *conn;
3310         struct nfsd4_clid_slot *cs_slot = NULL;
3311         __be32 status = 0;
3312         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3313
3314         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3315                 return nfserr_inval;
3316         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3317         if (status)
3318                 return status;
3319         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3320         if (status)
3321                 return status;
3322         status = check_backchannel_attrs(&cr_ses->back_channel);
3323         if (status)
3324                 goto out_release_drc_mem;
3325         status = nfserr_jukebox;
3326         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3327         if (!new)
3328                 goto out_release_drc_mem;
3329         conn = alloc_conn_from_crses(rqstp, cr_ses);
3330         if (!conn)
3331                 goto out_free_session;
3332
3333         spin_lock(&nn->client_lock);
3334         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3335         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3336         WARN_ON_ONCE(conf && unconf);
3337
3338         if (conf) {
3339                 status = nfserr_wrong_cred;
3340                 if (!nfsd4_mach_creds_match(conf, rqstp))
3341                         goto out_free_conn;
3342                 cs_slot = &conf->cl_cs_slot;
3343                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3344                 if (status) {
3345                         if (status == nfserr_replay_cache)
3346                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3347                         goto out_free_conn;
3348                 }
3349         } else if (unconf) {
3350                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3351                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3352                         status = nfserr_clid_inuse;
3353                         goto out_free_conn;
3354                 }
3355                 status = nfserr_wrong_cred;
3356                 if (!nfsd4_mach_creds_match(unconf, rqstp))
3357                         goto out_free_conn;
3358                 cs_slot = &unconf->cl_cs_slot;
3359                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3360                 if (status) {
3361                         /* an unconfirmed replay returns misordered */
3362                         status = nfserr_seq_misordered;
3363                         goto out_free_conn;
3364                 }
3365                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3366                 if (old) {
3367                         status = mark_client_expired_locked(old);
3368                         if (status) {
3369                                 old = NULL;
3370                                 goto out_free_conn;
3371                         }
3372                 }
3373                 move_to_confirmed(unconf);
3374                 conf = unconf;
3375         } else {
3376                 status = nfserr_stale_clientid;
3377                 goto out_free_conn;
3378         }
3379         status = nfs_ok;
3380         /* Persistent sessions are not supported */
3381         cr_ses->flags &= ~SESSION4_PERSIST;
3382         /* Upshifting from TCP to RDMA is not supported */
3383         cr_ses->flags &= ~SESSION4_RDMA;
3384
3385         init_session(rqstp, new, conf, cr_ses);
3386         nfsd4_get_session_locked(new);
3387
3388         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3389                NFS4_MAX_SESSIONID_LEN);
3390         cs_slot->sl_seqid++;
3391         cr_ses->seqid = cs_slot->sl_seqid;
3392
3393         /* cache solo and embedded create sessions under the client_lock */
3394         nfsd4_cache_create_session(cr_ses, cs_slot, status);
3395         spin_unlock(&nn->client_lock);
3396         /* init connection and backchannel */
3397         nfsd4_init_conn(rqstp, conn, new);
3398         nfsd4_put_session(new);
3399         if (old)
3400                 expire_client(old);
3401         return status;
3402 out_free_conn:
3403         spin_unlock(&nn->client_lock);
3404         free_conn(conn);
3405         if (old)
3406                 expire_client(old);
3407 out_free_session:
3408         __free_session(new);
3409 out_release_drc_mem:
3410         nfsd4_put_drc_mem(&cr_ses->fore_channel);
3411         return status;
3412 }
3413
3414 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3415 {
3416         switch (*dir) {
3417         case NFS4_CDFC4_FORE:
3418         case NFS4_CDFC4_BACK:
3419                 return nfs_ok;
3420         case NFS4_CDFC4_FORE_OR_BOTH:
3421         case NFS4_CDFC4_BACK_OR_BOTH:
3422                 *dir = NFS4_CDFC4_BOTH;
3423                 return nfs_ok;
3424         }
3425         return nfserr_inval;
3426 }
3427
3428 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3429                 struct nfsd4_compound_state *cstate,
3430                 union nfsd4_op_u *u)
3431 {
3432         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3433         struct nfsd4_session *session = cstate->session;
3434         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3435         __be32 status;
3436
3437         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3438         if (status)
3439                 return status;
3440         spin_lock(&nn->client_lock);
3441         session->se_cb_prog = bc->bc_cb_program;
3442         session->se_cb_sec = bc->bc_cb_sec;
3443         spin_unlock(&nn->client_lock);
3444
3445         nfsd4_probe_callback(session->se_client);
3446
3447         return nfs_ok;
3448 }
3449
3450 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
3451                      struct nfsd4_compound_state *cstate,
3452                      union nfsd4_op_u *u)
3453 {
3454         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
3455         __be32 status;
3456         struct nfsd4_conn *conn;
3457         struct nfsd4_session *session;
3458         struct net *net = SVC_NET(rqstp);
3459         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3460
3461         if (!nfsd4_last_compound_op(rqstp))
3462                 return nfserr_not_only_op;
3463         spin_lock(&nn->client_lock);
3464         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
3465         spin_unlock(&nn->client_lock);
3466         if (!session)
3467                 goto out_no_session;
3468         status = nfserr_wrong_cred;
3469         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
3470                 goto out;
3471         status = nfsd4_map_bcts_dir(&bcts->dir);
3472         if (status)
3473                 goto out;
3474         conn = alloc_conn(rqstp, bcts->dir);
3475         status = nfserr_jukebox;
3476         if (!conn)
3477                 goto out;
3478         nfsd4_init_conn(rqstp, conn, session);
3479         status = nfs_ok;
3480 out:
3481         nfsd4_put_session(session);
3482 out_no_session:
3483         return status;
3484 }
3485
3486 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3487 {
3488         if (!cstate->session)
3489                 return false;
3490         return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3491 }
3492
3493 __be32
3494 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3495                 union nfsd4_op_u *u)
3496 {
3497         struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3498         struct nfsd4_session *ses;
3499         __be32 status;
3500         int ref_held_by_me = 0;
3501         struct net *net = SVC_NET(r);
3502         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3503
3504         status = nfserr_not_only_op;
3505         if (nfsd4_compound_in_session(cstate, sessionid)) {
3506                 if (!nfsd4_last_compound_op(r))
3507                         goto out;
3508                 ref_held_by_me++;
3509         }
3510         dump_sessionid(__func__, sessionid);
3511         spin_lock(&nn->client_lock);
3512         ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3513         if (!ses)
3514                 goto out_client_lock;
3515         status = nfserr_wrong_cred;
3516         if (!nfsd4_mach_creds_match(ses->se_client, r))
3517                 goto out_put_session;
3518         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3519         if (status)
3520                 goto out_put_session;
3521         unhash_session(ses);
3522         spin_unlock(&nn->client_lock);
3523
3524         nfsd4_probe_callback_sync(ses->se_client);
3525
3526         spin_lock(&nn->client_lock);
3527         status = nfs_ok;
3528 out_put_session:
3529         nfsd4_put_session_locked(ses);
3530 out_client_lock:
3531         spin_unlock(&nn->client_lock);
3532 out:
3533         return status;
3534 }
3535
3536 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3537 {
3538         struct nfsd4_conn *c;
3539
3540         list_for_each_entry(c, &s->se_conns, cn_persession) {
3541                 if (c->cn_xprt == xpt) {
3542                         return c;
3543                 }
3544         }
3545         return NULL;
3546 }
3547
3548 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3549 {
3550         struct nfs4_client *clp = ses->se_client;
3551         struct nfsd4_conn *c;
3552         __be32 status = nfs_ok;
3553         int ret;
3554
3555         spin_lock(&clp->cl_lock);
3556         c = __nfsd4_find_conn(new->cn_xprt, ses);
3557         if (c)
3558                 goto out_free;
3559         status = nfserr_conn_not_bound_to_session;
3560         if (clp->cl_mach_cred)
3561                 goto out_free;
3562         __nfsd4_hash_conn(new, ses);
3563         spin_unlock(&clp->cl_lock);
3564         ret = nfsd4_register_conn(new);
3565         if (ret)
3566                 /* oops; xprt is already down: */
3567                 nfsd4_conn_lost(&new->cn_xpt_user);
3568         return nfs_ok;
3569 out_free:
3570         spin_unlock(&clp->cl_lock);
3571         free_conn(new);
3572         return status;
3573 }
3574
3575 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3576 {
3577         struct nfsd4_compoundargs *args = rqstp->rq_argp;
3578
3579         return args->opcnt > session->se_fchannel.maxops;
3580 }
3581
3582 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3583                                   struct nfsd4_session *session)
3584 {
3585         struct xdr_buf *xb = &rqstp->rq_arg;
3586
3587         return xb->len > session->se_fchannel.maxreq_sz;
3588 }
3589
3590 static bool replay_matches_cache(struct svc_rqst *rqstp,
3591                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3592 {
3593         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3594
3595         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3596             (bool)seq->cachethis)
3597                 return false;
3598         /*
3599          * If there's an error then the reply can have fewer ops than
3600          * the call.
3601          */
3602         if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
3603                 return false;
3604         /*
3605          * But if we cached a reply with *more* ops than the call you're
3606          * sending us now, then this new call is clearly not really a
3607          * replay of the old one:
3608          */
3609         if (slot->sl_opcnt > argp->opcnt)
3610                 return false;
3611         /* This is the only check explicitly called by spec: */
3612         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3613                 return false;
3614         /*
3615          * There may be more comparisons we could actually do, but the
3616          * spec doesn't require us to catch every case where the calls
3617          * don't match (that would require caching the call as well as
3618          * the reply), so we don't bother.
3619          */
3620         return true;
3621 }
3622
3623 __be32
3624 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3625                 union nfsd4_op_u *u)
3626 {
3627         struct nfsd4_sequence *seq = &u->sequence;
3628         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3629         struct xdr_stream *xdr = &resp->xdr;
3630         struct nfsd4_session *session;
3631         struct nfs4_client *clp;
3632         struct nfsd4_slot *slot;
3633         struct nfsd4_conn *conn;
3634         __be32 status;
3635         int buflen;
3636         struct net *net = SVC_NET(rqstp);
3637         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3638
3639         if (resp->opcnt != 1)
3640                 return nfserr_sequence_pos;
3641
3642         /*
3643          * Will be either used or freed by nfsd4_sequence_check_conn
3644          * below.
3645          */
3646         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3647         if (!conn)
3648                 return nfserr_jukebox;
3649
3650         spin_lock(&nn->client_lock);
3651         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3652         if (!session)
3653                 goto out_no_session;
3654         clp = session->se_client;
3655
3656         status = nfserr_too_many_ops;
3657         if (nfsd4_session_too_many_ops(rqstp, session))
3658                 goto out_put_session;
3659
3660         status = nfserr_req_too_big;
3661         if (nfsd4_request_too_big(rqstp, session))
3662                 goto out_put_session;
3663
3664         status = nfserr_badslot;
3665         if (seq->slotid >= session->se_fchannel.maxreqs)
3666                 goto out_put_session;
3667
3668         slot = session->se_slots[seq->slotid];
3669         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3670
3671         /* We do not negotiate the number of slots yet, so set the
3672          * maxslots to the session maxreqs which is used to encode
3673          * sr_highest_slotid and the sr_target_slot id to maxslots */
3674         seq->maxslots = session->se_fchannel.maxreqs;
3675
3676         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3677                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3678         if (status == nfserr_replay_cache) {
3679                 status = nfserr_seq_misordered;
3680                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3681                         goto out_put_session;
3682                 status = nfserr_seq_false_retry;
3683                 if (!replay_matches_cache(rqstp, seq, slot))
3684                         goto out_put_session;
3685                 cstate->slot = slot;
3686                 cstate->session = session;
3687                 cstate->clp = clp;
3688                 /* Return the cached reply status and set cstate->status
3689                  * for nfsd4_proc_compound processing */
3690                 status = nfsd4_replay_cache_entry(resp, seq);
3691                 cstate->status = nfserr_replay_cache;
3692                 goto out;
3693         }
3694         if (status)
3695                 goto out_put_session;
3696
3697         status = nfsd4_sequence_check_conn(conn, session);
3698         conn = NULL;
3699         if (status)
3700                 goto out_put_session;
3701
3702         buflen = (seq->cachethis) ?
3703                         session->se_fchannel.maxresp_cached :
3704                         session->se_fchannel.maxresp_sz;
3705         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3706                                     nfserr_rep_too_big;
3707         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3708                 goto out_put_session;
3709         svc_reserve(rqstp, buflen);
3710
3711         status = nfs_ok;
3712         /* Success! bump slot seqid */
3713         slot->sl_seqid = seq->seqid;
3714         slot->sl_flags |= NFSD4_SLOT_INUSE;
3715         if (seq->cachethis)
3716                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3717         else
3718                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3719
3720         cstate->slot = slot;
3721         cstate->session = session;
3722         cstate->clp = clp;
3723
3724 out:
3725         switch (clp->cl_cb_state) {
3726         case NFSD4_CB_DOWN:
3727                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3728                 break;
3729         case NFSD4_CB_FAULT:
3730                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3731                 break;
3732         default:
3733                 seq->status_flags = 0;
3734         }
3735         if (!list_empty(&clp->cl_revoked))
3736                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3737 out_no_session:
3738         if (conn)
3739                 free_conn(conn);
3740         spin_unlock(&nn->client_lock);
3741         return status;
3742 out_put_session:
3743         nfsd4_put_session_locked(session);
3744         goto out_no_session;
3745 }
3746
3747 void
3748 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3749 {
3750         struct nfsd4_compound_state *cs = &resp->cstate;
3751
3752         if (nfsd4_has_session(cs)) {
3753                 if (cs->status != nfserr_replay_cache) {
3754                         nfsd4_store_cache_entry(resp);
3755                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3756                 }
3757                 /* Drop session reference that was taken in nfsd4_sequence() */
3758                 nfsd4_put_session(cs->session);
3759         } else if (cs->clp)
3760                 put_client_renew(cs->clp);
3761 }
3762
3763 __be32
3764 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3765                 struct nfsd4_compound_state *cstate,
3766                 union nfsd4_op_u *u)
3767 {
3768         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3769         struct nfs4_client *conf, *unconf;
3770         struct nfs4_client *clp = NULL;
3771         __be32 status = 0;
3772         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3773
3774         spin_lock(&nn->client_lock);
3775         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3776         conf = find_confirmed_client(&dc->clientid, true, nn);
3777         WARN_ON_ONCE(conf && unconf);
3778
3779         if (conf) {
3780                 if (client_has_state(conf)) {
3781                         status = nfserr_clientid_busy;
3782                         goto out;
3783                 }
3784                 status = mark_client_expired_locked(conf);
3785                 if (status)
3786                         goto out;
3787                 clp = conf;
3788         } else if (unconf)
3789                 clp = unconf;
3790         else {
3791                 status = nfserr_stale_clientid;
3792                 goto out;
3793         }
3794         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3795                 clp = NULL;
3796                 status = nfserr_wrong_cred;
3797                 goto out;
3798         }
3799         unhash_client_locked(clp);
3800 out:
3801         spin_unlock(&nn->client_lock);
3802         if (clp)
3803                 expire_client(clp);
3804         return status;
3805 }
3806
3807 __be32
3808 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3809                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3810 {
3811         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3812         __be32 status = 0;
3813
3814         if (rc->rca_one_fs) {
3815                 if (!cstate->current_fh.fh_dentry)
3816                         return nfserr_nofilehandle;
3817                 /*
3818                  * We don't take advantage of the rca_one_fs case.
3819                  * That's OK, it's optional, we can safely ignore it.
3820                  */
3821                 return nfs_ok;
3822         }
3823
3824         status = nfserr_complete_already;
3825         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3826                              &cstate->session->se_client->cl_flags))
3827                 goto out;
3828
3829         status = nfserr_stale_clientid;
3830         if (is_client_expired(cstate->session->se_client))
3831                 /*
3832                  * The following error isn't really legal.
3833                  * But we only get here if the client just explicitly
3834                  * destroyed the client.  Surely it no longer cares what
3835                  * error it gets back on an operation for the dead
3836                  * client.
3837                  */
3838                 goto out;
3839
3840         status = nfs_ok;
3841         nfsd4_client_record_create(cstate->session->se_client);
3842         inc_reclaim_complete(cstate->session->se_client);
3843 out:
3844         return status;
3845 }
3846
3847 __be32
3848 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3849                   union nfsd4_op_u *u)
3850 {
3851         struct nfsd4_setclientid *setclid = &u->setclientid;
3852         struct xdr_netobj       clname = setclid->se_name;
3853         nfs4_verifier           clverifier = setclid->se_verf;
3854         struct nfs4_client      *conf, *new;
3855         struct nfs4_client      *unconf = NULL;
3856         __be32                  status;
3857         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3858
3859         new = create_client(clname, rqstp, &clverifier);
3860         if (new == NULL)
3861                 return nfserr_jukebox;
3862         /* Cases below refer to rfc 3530 section 14.2.33: */
3863         spin_lock(&nn->client_lock);
3864         conf = find_confirmed_client_by_name(&clname, nn);
3865         if (conf && client_has_state(conf)) {
3866                 /* case 0: */
3867                 status = nfserr_clid_inuse;
3868                 if (clp_used_exchangeid(conf))
3869                         goto out;
3870                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3871                         char addr_str[INET6_ADDRSTRLEN];
3872                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3873                                  sizeof(addr_str));
3874                         dprintk("NFSD: setclientid: string in use by client "
3875                                 "at %s\n", addr_str);
3876                         goto out;
3877                 }
3878         }
3879         unconf = find_unconfirmed_client_by_name(&clname, nn);
3880         if (unconf)
3881                 unhash_client_locked(unconf);
3882         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3883                 /* case 1: probable callback update */
3884                 copy_clid(new, conf);
3885                 gen_confirm(new, nn);
3886         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3887                 ;
3888         new->cl_minorversion = 0;
3889         gen_callback(new, setclid, rqstp);
3890         add_to_unconfirmed(new);
3891         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3892         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3893         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3894         new = NULL;
3895         status = nfs_ok;
3896 out:
3897         spin_unlock(&nn->client_lock);
3898         if (new)
3899                 free_client(new);
3900         if (unconf)
3901                 expire_client(unconf);
3902         return status;
3903 }
3904
3905
3906 __be32
3907 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3908                         struct nfsd4_compound_state *cstate,
3909                         union nfsd4_op_u *u)
3910 {
3911         struct nfsd4_setclientid_confirm *setclientid_confirm =
3912                         &u->setclientid_confirm;
3913         struct nfs4_client *conf, *unconf;
3914         struct nfs4_client *old = NULL;
3915         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3916         clientid_t * clid = &setclientid_confirm->sc_clientid;
3917         __be32 status;
3918         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3919
3920         if (STALE_CLIENTID(clid, nn))
3921                 return nfserr_stale_clientid;
3922
3923         spin_lock(&nn->client_lock);
3924         conf = find_confirmed_client(clid, false, nn);
3925         unconf = find_unconfirmed_client(clid, false, nn);
3926         /*
3927          * We try hard to give out unique clientid's, so if we get an
3928          * attempt to confirm the same clientid with a different cred,
3929          * the client may be buggy; this should never happen.
3930          *
3931          * Nevertheless, RFC 7530 recommends INUSE for this case:
3932          */
3933         status = nfserr_clid_inuse;
3934         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3935                 goto out;
3936         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3937                 goto out;
3938         /* cases below refer to rfc 3530 section 14.2.34: */
3939         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3940                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3941                         /* case 2: probable retransmit */
3942                         status = nfs_ok;
3943                 } else /* case 4: client hasn't noticed we rebooted yet? */
3944                         status = nfserr_stale_clientid;
3945                 goto out;
3946         }
3947         status = nfs_ok;
3948         if (conf) { /* case 1: callback update */
3949                 old = unconf;
3950                 unhash_client_locked(old);
3951                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3952         } else { /* case 3: normal case; new or rebooted client */
3953                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3954                 if (old) {
3955                         status = nfserr_clid_inuse;
3956                         if (client_has_state(old)
3957                                         && !same_creds(&unconf->cl_cred,
3958                                                         &old->cl_cred))
3959                                 goto out;
3960                         status = mark_client_expired_locked(old);
3961                         if (status) {
3962                                 old = NULL;
3963                                 goto out;
3964                         }
3965                 }
3966                 move_to_confirmed(unconf);
3967                 conf = unconf;
3968         }
3969         get_client_locked(conf);
3970         spin_unlock(&nn->client_lock);
3971         nfsd4_probe_callback(conf);
3972         spin_lock(&nn->client_lock);
3973         put_client_renew_locked(conf);
3974 out:
3975         spin_unlock(&nn->client_lock);
3976         if (old)
3977                 expire_client(old);
3978         return status;
3979 }
3980
3981 static struct nfs4_file *nfsd4_alloc_file(void)
3982 {
3983         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3984 }
3985
3986 /* OPEN Share state helper functions */
3987 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3988                                 struct nfs4_file *fp)
3989 {
3990         lockdep_assert_held(&state_lock);
3991
3992         refcount_set(&fp->fi_ref, 1);
3993         spin_lock_init(&fp->fi_lock);
3994         INIT_LIST_HEAD(&fp->fi_stateids);
3995         INIT_LIST_HEAD(&fp->fi_delegations);
3996         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3997         fh_copy_shallow(&fp->fi_fhandle, fh);
3998         fp->fi_deleg_file = NULL;
3999         fp->fi_had_conflict = false;
4000         fp->fi_share_deny = 0;
4001         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4002         memset(fp->fi_access, 0, sizeof(fp->fi_access));
4003 #ifdef CONFIG_NFSD_PNFS
4004         INIT_LIST_HEAD(&fp->fi_lo_states);
4005         atomic_set(&fp->fi_lo_recalls, 0);
4006 #endif
4007         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
4008 }
4009
4010 void
4011 nfsd4_free_slabs(void)
4012 {
4013         kmem_cache_destroy(client_slab);
4014         kmem_cache_destroy(openowner_slab);
4015         kmem_cache_destroy(lockowner_slab);
4016         kmem_cache_destroy(file_slab);
4017         kmem_cache_destroy(stateid_slab);
4018         kmem_cache_destroy(deleg_slab);
4019         kmem_cache_destroy(odstate_slab);
4020 }
4021
4022 int
4023 nfsd4_init_slabs(void)
4024 {
4025         client_slab = kmem_cache_create("nfsd4_clients",
4026                         sizeof(struct nfs4_client), 0, 0, NULL);
4027         if (client_slab == NULL)
4028                 goto out;
4029         openowner_slab = kmem_cache_create("nfsd4_openowners",
4030                         sizeof(struct nfs4_openowner), 0, 0, NULL);
4031         if (openowner_slab == NULL)
4032                 goto out_free_client_slab;
4033         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
4034                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
4035         if (lockowner_slab == NULL)
4036                 goto out_free_openowner_slab;
4037         file_slab = kmem_cache_create("nfsd4_files",
4038                         sizeof(struct nfs4_file), 0, 0, NULL);
4039         if (file_slab == NULL)
4040                 goto out_free_lockowner_slab;
4041         stateid_slab = kmem_cache_create("nfsd4_stateids",
4042                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
4043         if (stateid_slab == NULL)
4044                 goto out_free_file_slab;
4045         deleg_slab = kmem_cache_create("nfsd4_delegations",
4046                         sizeof(struct nfs4_delegation), 0, 0, NULL);
4047         if (deleg_slab == NULL)
4048                 goto out_free_stateid_slab;
4049         odstate_slab = kmem_cache_create("nfsd4_odstate",
4050                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
4051         if (odstate_slab == NULL)
4052                 goto out_free_deleg_slab;
4053         return 0;
4054
4055 out_free_deleg_slab:
4056         kmem_cache_destroy(deleg_slab);
4057 out_free_stateid_slab:
4058         kmem_cache_destroy(stateid_slab);
4059 out_free_file_slab:
4060         kmem_cache_destroy(file_slab);
4061 out_free_lockowner_slab:
4062         kmem_cache_destroy(lockowner_slab);
4063 out_free_openowner_slab:
4064         kmem_cache_destroy(openowner_slab);
4065 out_free_client_slab:
4066         kmem_cache_destroy(client_slab);
4067 out:
4068         dprintk("nfsd4: out of memory while initializing nfsv4\n");
4069         return -ENOMEM;
4070 }
4071
4072 static void init_nfs4_replay(struct nfs4_replay *rp)
4073 {
4074         rp->rp_status = nfserr_serverfault;
4075         rp->rp_buflen = 0;
4076         rp->rp_buf = rp->rp_ibuf;
4077         mutex_init(&rp->rp_mutex);
4078 }
4079
4080 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4081                 struct nfs4_stateowner *so)
4082 {
4083         if (!nfsd4_has_session(cstate)) {
4084                 mutex_lock(&so->so_replay.rp_mutex);
4085                 cstate->replay_owner = nfs4_get_stateowner(so);
4086         }
4087 }
4088
4089 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4090 {
4091         struct nfs4_stateowner *so = cstate->replay_owner;
4092
4093         if (so != NULL) {
4094                 cstate->replay_owner = NULL;
4095                 mutex_unlock(&so->so_replay.rp_mutex);
4096                 nfs4_put_stateowner(so);
4097         }
4098 }
4099
4100 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4101 {
4102         struct nfs4_stateowner *sop;
4103
4104         sop = kmem_cache_alloc(slab, GFP_KERNEL);
4105         if (!sop)
4106                 return NULL;
4107
4108         xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4109         if (!sop->so_owner.data) {
4110                 kmem_cache_free(slab, sop);
4111                 return NULL;
4112         }
4113
4114         INIT_LIST_HEAD(&sop->so_stateids);
4115         sop->so_client = clp;
4116         init_nfs4_replay(&sop->so_replay);
4117         atomic_set(&sop->so_count, 1);
4118         return sop;
4119 }
4120
4121 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4122 {
4123         lockdep_assert_held(&clp->cl_lock);
4124
4125         list_add(&oo->oo_owner.so_strhash,
4126                  &clp->cl_ownerstr_hashtbl[strhashval]);
4127         list_add(&oo->oo_perclient, &clp->cl_openowners);
4128 }
4129
4130 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4131 {
4132         unhash_openowner_locked(openowner(so));
4133 }
4134
4135 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4136 {
4137         struct nfs4_openowner *oo = openowner(so);
4138
4139         kmem_cache_free(openowner_slab, oo);
4140 }
4141
4142 static const struct nfs4_stateowner_operations openowner_ops = {
4143         .so_unhash =    nfs4_unhash_openowner,
4144         .so_free =      nfs4_free_openowner,
4145 };
4146
4147 static struct nfs4_ol_stateid *
4148 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4149 {
4150         struct nfs4_ol_stateid *local, *ret = NULL;
4151         struct nfs4_openowner *oo = open->op_openowner;
4152
4153         lockdep_assert_held(&fp->fi_lock);
4154
4155         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4156                 /* ignore lock owners */
4157                 if (local->st_stateowner->so_is_open_owner == 0)
4158                         continue;
4159                 if (local->st_stateowner != &oo->oo_owner)
4160                         continue;
4161                 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
4162                         ret = local;
4163                         refcount_inc(&ret->st_stid.sc_count);
4164                         break;
4165                 }
4166         }
4167         return ret;
4168 }
4169
4170 static __be32
4171 nfsd4_verify_open_stid(struct nfs4_stid *s)
4172 {
4173         __be32 ret = nfs_ok;
4174
4175         switch (s->sc_type) {
4176         default:
4177                 break;
4178         case 0:
4179         case NFS4_CLOSED_STID:
4180         case NFS4_CLOSED_DELEG_STID:
4181                 ret = nfserr_bad_stateid;
4182                 break;
4183         case NFS4_REVOKED_DELEG_STID:
4184                 ret = nfserr_deleg_revoked;
4185         }
4186         return ret;
4187 }
4188
4189 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4190 static __be32
4191 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4192 {
4193         __be32 ret;
4194
4195         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4196         ret = nfsd4_verify_open_stid(&stp->st_stid);
4197         if (ret != nfs_ok)
4198                 mutex_unlock(&stp->st_mutex);
4199         return ret;
4200 }
4201
4202 static struct nfs4_ol_stateid *
4203 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4204 {
4205         struct nfs4_ol_stateid *stp;
4206         for (;;) {
4207                 spin_lock(&fp->fi_lock);
4208                 stp = nfsd4_find_existing_open(fp, open);
4209                 spin_unlock(&fp->fi_lock);
4210                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4211                         break;
4212                 nfs4_put_stid(&stp->st_stid);
4213         }
4214         return stp;
4215 }
4216
4217 static struct nfs4_openowner *
4218 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4219                            struct nfsd4_compound_state *cstate)
4220 {
4221         struct nfs4_client *clp = cstate->clp;
4222         struct nfs4_openowner *oo, *ret;
4223
4224         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4225         if (!oo)
4226                 return NULL;
4227         oo->oo_owner.so_ops = &openowner_ops;
4228         oo->oo_owner.so_is_open_owner = 1;
4229         oo->oo_owner.so_seqid = open->op_seqid;
4230         oo->oo_flags = 0;
4231         if (nfsd4_has_session(cstate))
4232                 oo->oo_flags |= NFS4_OO_CONFIRMED;
4233         oo->oo_time = 0;
4234         oo->oo_last_closed_stid = NULL;
4235         INIT_LIST_HEAD(&oo->oo_close_lru);
4236         spin_lock(&clp->cl_lock);
4237         ret = find_openstateowner_str_locked(strhashval, open, clp);
4238         if (ret == NULL) {
4239                 hash_openowner(oo, clp, strhashval);
4240                 ret = oo;
4241         } else
4242                 nfs4_free_stateowner(&oo->oo_owner);
4243
4244         spin_unlock(&clp->cl_lock);
4245         return ret;
4246 }
4247
4248 static struct nfs4_ol_stateid *
4249 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4250 {
4251
4252         struct nfs4_openowner *oo = open->op_openowner;
4253         struct nfs4_ol_stateid *retstp = NULL;
4254         struct nfs4_ol_stateid *stp;
4255
4256         stp = open->op_stp;
4257         /* We are moving these outside of the spinlocks to avoid the warnings */
4258         mutex_init(&stp->st_mutex);
4259         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4260
4261 retry:
4262         spin_lock(&oo->oo_owner.so_client->cl_lock);
4263         spin_lock(&fp->fi_lock);
4264
4265         retstp = nfsd4_find_existing_open(fp, open);
4266         if (retstp)
4267                 goto out_unlock;
4268
4269         open->op_stp = NULL;
4270         refcount_inc(&stp->st_stid.sc_count);
4271         stp->st_stid.sc_type = NFS4_OPEN_STID;
4272         INIT_LIST_HEAD(&stp->st_locks);
4273         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4274         get_nfs4_file(fp);
4275         stp->st_stid.sc_file = fp;
4276         stp->st_access_bmap = 0;
4277         stp->st_deny_bmap = 0;
4278         stp->st_openstp = NULL;
4279         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4280         list_add(&stp->st_perfile, &fp->fi_stateids);
4281
4282 out_unlock:
4283         spin_unlock(&fp->fi_lock);
4284         spin_unlock(&oo->oo_owner.so_client->cl_lock);
4285         if (retstp) {
4286                 /* Handle races with CLOSE */
4287                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4288                         nfs4_put_stid(&retstp->st_stid);
4289                         goto retry;
4290                 }
4291                 /* To keep mutex tracking happy */
4292                 mutex_unlock(&stp->st_mutex);
4293                 stp = retstp;
4294         }
4295         return stp;
4296 }
4297
4298 /*
4299  * In the 4.0 case we need to keep the owners around a little while to handle
4300  * CLOSE replay. We still do need to release any file access that is held by
4301  * them before returning however.
4302  */
4303 static void
4304 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4305 {
4306         struct nfs4_ol_stateid *last;
4307         struct nfs4_openowner *oo = openowner(s->st_stateowner);
4308         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4309                                                 nfsd_net_id);
4310
4311         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4312
4313         /*
4314          * We know that we hold one reference via nfsd4_close, and another
4315          * "persistent" reference for the client. If the refcount is higher
4316          * than 2, then there are still calls in progress that are using this
4317          * stateid. We can't put the sc_file reference until they are finished.
4318          * Wait for the refcount to drop to 2. Since it has been unhashed,
4319          * there should be no danger of the refcount going back up again at
4320          * this point.
4321          */
4322         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4323
4324         release_all_access(s);
4325         if (s->st_stid.sc_file) {
4326                 put_nfs4_file(s->st_stid.sc_file);
4327                 s->st_stid.sc_file = NULL;
4328         }
4329
4330         spin_lock(&nn->client_lock);
4331         last = oo->oo_last_closed_stid;
4332         oo->oo_last_closed_stid = s;
4333         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4334         oo->oo_time = ktime_get_boottime_seconds();
4335         spin_unlock(&nn->client_lock);
4336         if (last)
4337                 nfs4_put_stid(&last->st_stid);
4338 }
4339
4340 /* search file_hashtbl[] for file */
4341 static struct nfs4_file *
4342 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
4343 {
4344         struct nfs4_file *fp;
4345
4346         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
4347                 if (fh_match(&fp->fi_fhandle, fh)) {
4348                         if (refcount_inc_not_zero(&fp->fi_ref))
4349                                 return fp;
4350                 }
4351         }
4352         return NULL;
4353 }
4354
4355 struct nfs4_file *
4356 find_file(struct knfsd_fh *fh)
4357 {
4358         struct nfs4_file *fp;
4359         unsigned int hashval = file_hashval(fh);
4360
4361         rcu_read_lock();
4362         fp = find_file_locked(fh, hashval);
4363         rcu_read_unlock();
4364         return fp;
4365 }
4366
4367 static struct nfs4_file *
4368 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
4369 {
4370         struct nfs4_file *fp;
4371         unsigned int hashval = file_hashval(fh);
4372
4373         rcu_read_lock();
4374         fp = find_file_locked(fh, hashval);
4375         rcu_read_unlock();
4376         if (fp)
4377                 return fp;
4378
4379         spin_lock(&state_lock);
4380         fp = find_file_locked(fh, hashval);
4381         if (likely(fp == NULL)) {
4382                 nfsd4_init_file(fh, hashval, new);
4383                 fp = new;
4384         }
4385         spin_unlock(&state_lock);
4386
4387         return fp;
4388 }
4389
4390 /*
4391  * Called to check deny when READ with all zero stateid or
4392  * WRITE with all zero or all one stateid
4393  */
4394 static __be32
4395 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
4396 {
4397         struct nfs4_file *fp;
4398         __be32 ret = nfs_ok;
4399
4400         fp = find_file(&current_fh->fh_handle);
4401         if (!fp)
4402                 return ret;
4403         /* Check for conflicting share reservations */
4404         spin_lock(&fp->fi_lock);
4405         if (fp->fi_share_deny & deny_type)
4406                 ret = nfserr_locked;
4407         spin_unlock(&fp->fi_lock);
4408         put_nfs4_file(fp);
4409         return ret;
4410 }
4411
4412 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
4413 {
4414         struct nfs4_delegation *dp = cb_to_delegation(cb);
4415         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
4416                                           nfsd_net_id);
4417
4418         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
4419
4420         /*
4421          * We can't do this in nfsd_break_deleg_cb because it is
4422          * already holding inode->i_lock.
4423          *
4424          * If the dl_time != 0, then we know that it has already been
4425          * queued for a lease break. Don't queue it again.
4426          */
4427         spin_lock(&state_lock);
4428         if (dp->dl_time == 0) {
4429                 dp->dl_time = ktime_get_boottime_seconds();
4430                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
4431         }
4432         spin_unlock(&state_lock);
4433 }
4434
4435 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
4436                 struct rpc_task *task)
4437 {
4438         struct nfs4_delegation *dp = cb_to_delegation(cb);
4439
4440         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
4441                 return 1;
4442
4443         switch (task->tk_status) {
4444         case 0:
4445                 return 1;
4446         case -NFS4ERR_DELAY:
4447                 rpc_delay(task, 2 * HZ);
4448                 return 0;
4449         case -EBADHANDLE:
4450         case -NFS4ERR_BAD_STATEID:
4451                 /*
4452                  * Race: client probably got cb_recall before open reply
4453                  * granting delegation.
4454                  */
4455                 if (dp->dl_retries--) {
4456                         rpc_delay(task, 2 * HZ);
4457                         return 0;
4458                 }
4459                 /*FALLTHRU*/
4460         default:
4461                 return 1;
4462         }
4463 }
4464
4465 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
4466 {
4467         struct nfs4_delegation *dp = cb_to_delegation(cb);
4468
4469         nfs4_put_stid(&dp->dl_stid);
4470 }
4471
4472 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
4473         .prepare        = nfsd4_cb_recall_prepare,
4474         .done           = nfsd4_cb_recall_done,
4475         .release        = nfsd4_cb_recall_release,
4476 };
4477
4478 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
4479 {
4480         /*
4481          * We're assuming the state code never drops its reference
4482          * without first removing the lease.  Since we're in this lease
4483          * callback (and since the lease code is serialized by the
4484          * i_lock) we know the server hasn't removed the lease yet, and
4485          * we know it's safe to take a reference.
4486          */
4487         refcount_inc(&dp->dl_stid.sc_count);
4488         nfsd4_run_cb(&dp->dl_recall);
4489 }
4490
4491 /* Called from break_lease() with i_lock held. */
4492 static bool
4493 nfsd_break_deleg_cb(struct file_lock *fl)
4494 {
4495         bool ret = false;
4496         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4497         struct nfs4_file *fp = dp->dl_stid.sc_file;
4498
4499         /*
4500          * We don't want the locks code to timeout the lease for us;
4501          * we'll remove it ourself if a delegation isn't returned
4502          * in time:
4503          */
4504         fl->fl_break_time = 0;
4505
4506         spin_lock(&fp->fi_lock);
4507         fp->fi_had_conflict = true;
4508         nfsd_break_one_deleg(dp);
4509         spin_unlock(&fp->fi_lock);
4510         return ret;
4511 }
4512
4513 static int
4514 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4515                      struct list_head *dispose)
4516 {
4517         if (arg & F_UNLCK)
4518                 return lease_modify(onlist, arg, dispose);
4519         else
4520                 return -EAGAIN;
4521 }
4522
4523 static const struct lock_manager_operations nfsd_lease_mng_ops = {
4524         .lm_break = nfsd_break_deleg_cb,
4525         .lm_change = nfsd_change_deleg_cb,
4526 };
4527
4528 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4529 {
4530         if (nfsd4_has_session(cstate))
4531                 return nfs_ok;
4532         if (seqid == so->so_seqid - 1)
4533                 return nfserr_replay_me;
4534         if (seqid == so->so_seqid)
4535                 return nfs_ok;
4536         return nfserr_bad_seqid;
4537 }
4538
4539 static __be32 lookup_clientid(clientid_t *clid,
4540                 struct nfsd4_compound_state *cstate,
4541                 struct nfsd_net *nn,
4542                 bool sessions)
4543 {
4544         struct nfs4_client *found;
4545
4546         if (cstate->clp) {
4547                 found = cstate->clp;
4548                 if (!same_clid(&found->cl_clientid, clid))
4549                         return nfserr_stale_clientid;
4550                 return nfs_ok;
4551         }
4552
4553         if (STALE_CLIENTID(clid, nn))
4554                 return nfserr_stale_clientid;
4555
4556         /*
4557          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4558          * cached already then we know this is for is for v4.0 and "sessions"
4559          * will be false.
4560          */
4561         WARN_ON_ONCE(cstate->session);
4562         spin_lock(&nn->client_lock);
4563         found = find_confirmed_client(clid, sessions, nn);
4564         if (!found) {
4565                 spin_unlock(&nn->client_lock);
4566                 return nfserr_expired;
4567         }
4568         atomic_inc(&found->cl_rpc_users);
4569         spin_unlock(&nn->client_lock);
4570
4571         /* Cache the nfs4_client in cstate! */
4572         cstate->clp = found;
4573         return nfs_ok;
4574 }
4575
4576 __be32
4577 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4578                     struct nfsd4_open *open, struct nfsd_net *nn)
4579 {
4580         clientid_t *clientid = &open->op_clientid;
4581         struct nfs4_client *clp = NULL;
4582         unsigned int strhashval;
4583         struct nfs4_openowner *oo = NULL;
4584         __be32 status;
4585
4586         if (STALE_CLIENTID(&open->op_clientid, nn))
4587                 return nfserr_stale_clientid;
4588         /*
4589          * In case we need it later, after we've already created the
4590          * file and don't want to risk a further failure:
4591          */
4592         open->op_file = nfsd4_alloc_file();
4593         if (open->op_file == NULL)
4594                 return nfserr_jukebox;
4595
4596         status = lookup_clientid(clientid, cstate, nn, false);
4597         if (status)
4598                 return status;
4599         clp = cstate->clp;
4600
4601         strhashval = ownerstr_hashval(&open->op_owner);
4602         oo = find_openstateowner_str(strhashval, open, clp);
4603         open->op_openowner = oo;
4604         if (!oo) {
4605                 goto new_owner;
4606         }
4607         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4608                 /* Replace unconfirmed owners without checking for replay. */
4609                 release_openowner(oo);
4610                 open->op_openowner = NULL;
4611                 goto new_owner;
4612         }
4613         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4614         if (status)
4615                 return status;
4616         goto alloc_stateid;
4617 new_owner:
4618         oo = alloc_init_open_stateowner(strhashval, open, cstate);
4619         if (oo == NULL)
4620                 return nfserr_jukebox;
4621         open->op_openowner = oo;
4622 alloc_stateid:
4623         open->op_stp = nfs4_alloc_open_stateid(clp);
4624         if (!open->op_stp)
4625                 return nfserr_jukebox;
4626
4627         if (nfsd4_has_session(cstate) &&
4628             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4629                 open->op_odstate = alloc_clnt_odstate(clp);
4630                 if (!open->op_odstate)
4631                         return nfserr_jukebox;
4632         }
4633
4634         return nfs_ok;
4635 }
4636
4637 static inline __be32
4638 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4639 {
4640         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4641                 return nfserr_openmode;
4642         else
4643                 return nfs_ok;
4644 }
4645
4646 static int share_access_to_flags(u32 share_access)
4647 {
4648         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4649 }
4650
4651 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4652 {
4653         struct nfs4_stid *ret;
4654
4655         ret = find_stateid_by_type(cl, s,
4656                                 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4657         if (!ret)
4658                 return NULL;
4659         return delegstateid(ret);
4660 }
4661
4662 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4663 {
4664         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4665                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4666 }
4667
4668 static __be32
4669 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4670                 struct nfs4_delegation **dp)
4671 {
4672         int flags;
4673         __be32 status = nfserr_bad_stateid;
4674         struct nfs4_delegation *deleg;
4675
4676         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4677         if (deleg == NULL)
4678                 goto out;
4679         if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4680                 nfs4_put_stid(&deleg->dl_stid);
4681                 if (cl->cl_minorversion)
4682                         status = nfserr_deleg_revoked;
4683                 goto out;
4684         }
4685         flags = share_access_to_flags(open->op_share_access);
4686         status = nfs4_check_delegmode(deleg, flags);
4687         if (status) {
4688                 nfs4_put_stid(&deleg->dl_stid);
4689                 goto out;
4690         }
4691         *dp = deleg;
4692 out:
4693         if (!nfsd4_is_deleg_cur(open))
4694                 return nfs_ok;
4695         if (status)
4696                 return status;
4697         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4698         return nfs_ok;
4699 }
4700
4701 static inline int nfs4_access_to_access(u32 nfs4_access)
4702 {
4703         int flags = 0;
4704
4705         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4706                 flags |= NFSD_MAY_READ;
4707         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4708                 flags |= NFSD_MAY_WRITE;
4709         return flags;
4710 }
4711
4712 static inline __be32
4713 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4714                 struct nfsd4_open *open)
4715 {
4716         struct iattr iattr = {
4717                 .ia_valid = ATTR_SIZE,
4718                 .ia_size = 0,
4719         };
4720         if (!open->op_truncate)
4721                 return 0;
4722         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4723                 return nfserr_inval;
4724         return nfsd_setattr(rqstp, fh, &iattr, 0, (time64_t)0);
4725 }
4726
4727 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4728                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4729                 struct nfsd4_open *open)
4730 {
4731         struct nfsd_file *nf = NULL;
4732         __be32 status;
4733         int oflag = nfs4_access_to_omode(open->op_share_access);
4734         int access = nfs4_access_to_access(open->op_share_access);
4735         unsigned char old_access_bmap, old_deny_bmap;
4736
4737         spin_lock(&fp->fi_lock);
4738
4739         /*
4740          * Are we trying to set a deny mode that would conflict with
4741          * current access?
4742          */
4743         status = nfs4_file_check_deny(fp, open->op_share_deny);
4744         if (status != nfs_ok) {
4745                 spin_unlock(&fp->fi_lock);
4746                 goto out;
4747         }
4748
4749         /* set access to the file */
4750         status = nfs4_file_get_access(fp, open->op_share_access);
4751         if (status != nfs_ok) {
4752                 spin_unlock(&fp->fi_lock);
4753                 goto out;
4754         }
4755
4756         /* Set access bits in stateid */
4757         old_access_bmap = stp->st_access_bmap;
4758         set_access(open->op_share_access, stp);
4759
4760         /* Set new deny mask */
4761         old_deny_bmap = stp->st_deny_bmap;
4762         set_deny(open->op_share_deny, stp);
4763         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4764
4765         if (!fp->fi_fds[oflag]) {
4766                 spin_unlock(&fp->fi_lock);
4767                 status = nfsd_file_acquire(rqstp, cur_fh, access, &nf);
4768                 if (status)
4769                         goto out_put_access;
4770                 spin_lock(&fp->fi_lock);
4771                 if (!fp->fi_fds[oflag]) {
4772                         fp->fi_fds[oflag] = nf;
4773                         nf = NULL;
4774                 }
4775         }
4776         spin_unlock(&fp->fi_lock);
4777         if (nf)
4778                 nfsd_file_put(nf);
4779
4780         status = nfsd4_truncate(rqstp, cur_fh, open);
4781         if (status)
4782                 goto out_put_access;
4783 out:
4784         return status;
4785 out_put_access:
4786         stp->st_access_bmap = old_access_bmap;
4787         nfs4_file_put_access(fp, open->op_share_access);
4788         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4789         goto out;
4790 }
4791
4792 static __be32
4793 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)
4794 {
4795         __be32 status;
4796         unsigned char old_deny_bmap = stp->st_deny_bmap;
4797
4798         if (!test_access(open->op_share_access, stp))
4799                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4800
4801         /* test and set deny mode */
4802         spin_lock(&fp->fi_lock);
4803         status = nfs4_file_check_deny(fp, open->op_share_deny);
4804         if (status == nfs_ok) {
4805                 set_deny(open->op_share_deny, stp);
4806                 fp->fi_share_deny |=
4807                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4808         }
4809         spin_unlock(&fp->fi_lock);
4810
4811         if (status != nfs_ok)
4812                 return status;
4813
4814         status = nfsd4_truncate(rqstp, cur_fh, open);
4815         if (status != nfs_ok)
4816                 reset_union_bmap_deny(old_deny_bmap, stp);
4817         return status;
4818 }
4819
4820 /* Should we give out recallable state?: */
4821 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4822 {
4823         if (clp->cl_cb_state == NFSD4_CB_UP)
4824                 return true;
4825         /*
4826          * In the sessions case, since we don't have to establish a
4827          * separate connection for callbacks, we assume it's OK
4828          * until we hear otherwise:
4829          */
4830         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4831 }
4832
4833 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
4834                                                 int flag)
4835 {
4836         struct file_lock *fl;
4837
4838         fl = locks_alloc_lock();
4839         if (!fl)
4840                 return NULL;
4841         fl->fl_lmops = &nfsd_lease_mng_ops;
4842         fl->fl_flags = FL_DELEG;
4843         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4844         fl->fl_end = OFFSET_MAX;
4845         fl->fl_owner = (fl_owner_t)dp;
4846         fl->fl_pid = current->tgid;
4847         fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
4848         return fl;
4849 }
4850
4851 static struct nfs4_delegation *
4852 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4853                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4854 {
4855         int status = 0;
4856         struct nfs4_delegation *dp;
4857         struct nfsd_file *nf;
4858         struct file_lock *fl;
4859
4860         /*
4861          * The fi_had_conflict and nfs_get_existing_delegation checks
4862          * here are just optimizations; we'll need to recheck them at
4863          * the end:
4864          */
4865         if (fp->fi_had_conflict)
4866                 return ERR_PTR(-EAGAIN);
4867
4868         nf = find_readable_file(fp);
4869         if (!nf) {
4870                 /* We should always have a readable file here */
4871                 WARN_ON_ONCE(1);
4872                 return ERR_PTR(-EBADF);
4873         }
4874         spin_lock(&state_lock);
4875         spin_lock(&fp->fi_lock);
4876         if (nfs4_delegation_exists(clp, fp))
4877                 status = -EAGAIN;
4878         else if (!fp->fi_deleg_file) {
4879                 fp->fi_deleg_file = nf;
4880                 /* increment early to prevent fi_deleg_file from being
4881                  * cleared */
4882                 fp->fi_delegees = 1;
4883                 nf = NULL;
4884         } else
4885                 fp->fi_delegees++;
4886         spin_unlock(&fp->fi_lock);
4887         spin_unlock(&state_lock);
4888         if (nf)
4889                 nfsd_file_put(nf);
4890         if (status)
4891                 return ERR_PTR(status);
4892
4893         status = -ENOMEM;
4894         dp = alloc_init_deleg(clp, fp, fh, odstate);
4895         if (!dp)
4896                 goto out_delegees;
4897
4898         fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
4899         if (!fl)
4900                 goto out_clnt_odstate;
4901
4902         status = vfs_setlease(fp->fi_deleg_file->nf_file, fl->fl_type, &fl, NULL);
4903         if (fl)
4904                 locks_free_lock(fl);
4905         if (status)
4906                 goto out_clnt_odstate;
4907
4908         spin_lock(&state_lock);
4909         spin_lock(&fp->fi_lock);
4910         if (fp->fi_had_conflict)
4911                 status = -EAGAIN;
4912         else
4913                 status = hash_delegation_locked(dp, fp);
4914         spin_unlock(&fp->fi_lock);
4915         spin_unlock(&state_lock);
4916
4917         if (status)
4918                 goto out_unlock;
4919
4920         return dp;
4921 out_unlock:
4922         vfs_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
4923 out_clnt_odstate:
4924         put_clnt_odstate(dp->dl_clnt_odstate);
4925         nfs4_put_stid(&dp->dl_stid);
4926 out_delegees:
4927         put_deleg_file(fp);
4928         return ERR_PTR(status);
4929 }
4930
4931 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4932 {
4933         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4934         if (status == -EAGAIN)
4935                 open->op_why_no_deleg = WND4_CONTENTION;
4936         else {
4937                 open->op_why_no_deleg = WND4_RESOURCE;
4938                 switch (open->op_deleg_want) {
4939                 case NFS4_SHARE_WANT_READ_DELEG:
4940                 case NFS4_SHARE_WANT_WRITE_DELEG:
4941                 case NFS4_SHARE_WANT_ANY_DELEG:
4942                         break;
4943                 case NFS4_SHARE_WANT_CANCEL:
4944                         open->op_why_no_deleg = WND4_CANCELLED;
4945                         break;
4946                 case NFS4_SHARE_WANT_NO_DELEG:
4947                         WARN_ON_ONCE(1);
4948                 }
4949         }
4950 }
4951
4952 /*
4953  * Attempt to hand out a delegation.
4954  *
4955  * Note we don't support write delegations, and won't until the vfs has
4956  * proper support for them.
4957  */
4958 static void
4959 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4960                         struct nfs4_ol_stateid *stp)
4961 {
4962         struct nfs4_delegation *dp;
4963         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4964         struct nfs4_client *clp = stp->st_stid.sc_client;
4965         int cb_up;
4966         int status = 0;
4967
4968         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4969         open->op_recall = 0;
4970         switch (open->op_claim_type) {
4971                 case NFS4_OPEN_CLAIM_PREVIOUS:
4972                         if (!cb_up)
4973                                 open->op_recall = 1;
4974                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4975                                 goto out_no_deleg;
4976                         break;
4977                 case NFS4_OPEN_CLAIM_NULL:
4978                 case NFS4_OPEN_CLAIM_FH:
4979                         /*
4980                          * Let's not give out any delegations till everyone's
4981                          * had the chance to reclaim theirs, *and* until
4982                          * NLM locks have all been reclaimed:
4983                          */
4984                         if (locks_in_grace(clp->net))
4985                                 goto out_no_deleg;
4986                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4987                                 goto out_no_deleg;
4988                         /*
4989                          * Also, if the file was opened for write or
4990                          * create, there's a good chance the client's
4991                          * about to write to it, resulting in an
4992                          * immediate recall (since we don't support
4993                          * write delegations):
4994                          */
4995                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4996                                 goto out_no_deleg;
4997                         if (open->op_create == NFS4_OPEN_CREATE)
4998                                 goto out_no_deleg;
4999                         break;
5000                 default:
5001                         goto out_no_deleg;
5002         }
5003         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
5004         if (IS_ERR(dp))
5005                 goto out_no_deleg;
5006
5007         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5008
5009         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
5010                 STATEID_VAL(&dp->dl_stid.sc_stateid));
5011         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
5012         nfs4_put_stid(&dp->dl_stid);
5013         return;
5014 out_no_deleg:
5015         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
5016         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
5017             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
5018                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
5019                 open->op_recall = 1;
5020         }
5021
5022         /* 4.1 client asking for a delegation? */
5023         if (open->op_deleg_want)
5024                 nfsd4_open_deleg_none_ext(open, status);
5025         return;
5026 }
5027
5028 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
5029                                         struct nfs4_delegation *dp)
5030 {
5031         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
5032             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5033                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5034                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
5035         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
5036                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5037                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5038                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
5039         }
5040         /* Otherwise the client must be confused wanting a delegation
5041          * it already has, therefore we don't return
5042          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
5043          */
5044 }
5045
5046 __be32
5047 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
5048 {
5049         struct nfsd4_compoundres *resp = rqstp->rq_resp;
5050         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
5051         struct nfs4_file *fp = NULL;
5052         struct nfs4_ol_stateid *stp = NULL;
5053         struct nfs4_delegation *dp = NULL;
5054         __be32 status;
5055         bool new_stp = false;
5056
5057         /*
5058          * Lookup file; if found, lookup stateid and check open request,
5059          * and check for delegations in the process of being recalled.
5060          * If not found, create the nfs4_file struct
5061          */
5062         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
5063         if (fp != open->op_file) {
5064                 status = nfs4_check_deleg(cl, open, &dp);
5065                 if (status)
5066                         goto out;
5067                 stp = nfsd4_find_and_lock_existing_open(fp, open);
5068         } else {
5069                 open->op_file = NULL;
5070                 status = nfserr_bad_stateid;
5071                 if (nfsd4_is_deleg_cur(open))
5072                         goto out;
5073         }
5074
5075         if (!stp) {
5076                 stp = init_open_stateid(fp, open);
5077                 if (!open->op_stp)
5078                         new_stp = true;
5079         }
5080
5081         /*
5082          * OPEN the file, or upgrade an existing OPEN.
5083          * If truncate fails, the OPEN fails.
5084          *
5085          * stp is already locked.
5086          */
5087         if (!new_stp) {
5088                 /* Stateid was found, this is an OPEN upgrade */
5089                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
5090                 if (status) {
5091                         mutex_unlock(&stp->st_mutex);
5092                         goto out;
5093                 }
5094         } else {
5095                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
5096                 if (status) {
5097                         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5098                         release_open_stateid(stp);
5099                         mutex_unlock(&stp->st_mutex);
5100                         goto out;
5101                 }
5102
5103                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
5104                                                         open->op_odstate);
5105                 if (stp->st_clnt_odstate == open->op_odstate)
5106                         open->op_odstate = NULL;
5107         }
5108
5109         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
5110         mutex_unlock(&stp->st_mutex);
5111
5112         if (nfsd4_has_session(&resp->cstate)) {
5113                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
5114                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5115                         open->op_why_no_deleg = WND4_NOT_WANTED;
5116                         goto nodeleg;
5117                 }
5118         }
5119
5120         /*
5121         * Attempt to hand out a delegation. No error return, because the
5122         * OPEN succeeds even if we fail.
5123         */
5124         nfs4_open_delegation(current_fh, open, stp);
5125 nodeleg:
5126         status = nfs_ok;
5127
5128         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
5129                 STATEID_VAL(&stp->st_stid.sc_stateid));
5130 out:
5131         /* 4.1 client trying to upgrade/downgrade delegation? */
5132         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
5133             open->op_deleg_want)
5134                 nfsd4_deleg_xgrade_none_ext(open, dp);
5135
5136         if (fp)
5137                 put_nfs4_file(fp);
5138         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
5139                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5140         /*
5141         * To finish the open response, we just need to set the rflags.
5142         */
5143         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
5144         if (nfsd4_has_session(&resp->cstate))
5145                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
5146         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
5147                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
5148
5149         if (dp)
5150                 nfs4_put_stid(&dp->dl_stid);
5151         if (stp)
5152                 nfs4_put_stid(&stp->st_stid);
5153
5154         return status;
5155 }
5156
5157 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
5158                               struct nfsd4_open *open)
5159 {
5160         if (open->op_openowner) {
5161                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
5162
5163                 nfsd4_cstate_assign_replay(cstate, so);
5164                 nfs4_put_stateowner(so);
5165         }
5166         if (open->op_file)
5167                 kmem_cache_free(file_slab, open->op_file);
5168         if (open->op_stp)
5169                 nfs4_put_stid(&open->op_stp->st_stid);
5170         if (open->op_odstate)
5171                 kmem_cache_free(odstate_slab, open->op_odstate);
5172 }
5173
5174 __be32
5175 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5176             union nfsd4_op_u *u)
5177 {
5178         clientid_t *clid = &u->renew;
5179         struct nfs4_client *clp;
5180         __be32 status;
5181         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5182
5183         dprintk("process_renew(%08x/%08x): starting\n",
5184                         clid->cl_boot, clid->cl_id);
5185         status = lookup_clientid(clid, cstate, nn, false);
5186         if (status)
5187                 goto out;
5188         clp = cstate->clp;
5189         status = nfserr_cb_path_down;
5190         if (!list_empty(&clp->cl_delegations)
5191                         && clp->cl_cb_state != NFSD4_CB_UP)
5192                 goto out;
5193         status = nfs_ok;
5194 out:
5195         return status;
5196 }
5197
5198 void
5199 nfsd4_end_grace(struct nfsd_net *nn)
5200 {
5201         /* do nothing if grace period already ended */
5202         if (nn->grace_ended)
5203                 return;
5204
5205         nn->grace_ended = true;
5206         /*
5207          * If the server goes down again right now, an NFSv4
5208          * client will still be allowed to reclaim after it comes back up,
5209          * even if it hasn't yet had a chance to reclaim state this time.
5210          *
5211          */
5212         nfsd4_record_grace_done(nn);
5213         /*
5214          * At this point, NFSv4 clients can still reclaim.  But if the
5215          * server crashes, any that have not yet reclaimed will be out
5216          * of luck on the next boot.
5217          *
5218          * (NFSv4.1+ clients are considered to have reclaimed once they
5219          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
5220          * have reclaimed after their first OPEN.)
5221          */
5222         locks_end_grace(&nn->nfsd4_manager);
5223         /*
5224          * At this point, and once lockd and/or any other containers
5225          * exit their grace period, further reclaims will fail and
5226          * regular locking can resume.
5227          */
5228 }
5229
5230 /*
5231  * If we've waited a lease period but there are still clients trying to
5232  * reclaim, wait a little longer to give them a chance to finish.
5233  */
5234 static bool clients_still_reclaiming(struct nfsd_net *nn)
5235 {
5236         time64_t double_grace_period_end = nn->boot_time +
5237                                            2 * nn->nfsd4_lease;
5238
5239         if (nn->track_reclaim_completes &&
5240                         atomic_read(&nn->nr_reclaim_complete) ==
5241                         nn->reclaim_str_hashtbl_size)
5242                 return false;
5243         if (!nn->somebody_reclaimed)
5244                 return false;
5245         nn->somebody_reclaimed = false;
5246         /*
5247          * If we've given them *two* lease times to reclaim, and they're
5248          * still not done, give up:
5249          */
5250         if (ktime_get_boottime_seconds() > double_grace_period_end)
5251                 return false;
5252         return true;
5253 }
5254
5255 static time64_t
5256 nfs4_laundromat(struct nfsd_net *nn)
5257 {
5258         struct nfs4_client *clp;
5259         struct nfs4_openowner *oo;
5260         struct nfs4_delegation *dp;
5261         struct nfs4_ol_stateid *stp;
5262         struct nfsd4_blocked_lock *nbl;
5263         struct list_head *pos, *next, reaplist;
5264         time64_t cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease;
5265         time64_t t, new_timeo = nn->nfsd4_lease;
5266         struct nfs4_cpntf_state *cps;
5267         copy_stateid_t *cps_t;
5268         int i;
5269
5270         dprintk("NFSD: laundromat service - starting\n");
5271
5272         if (clients_still_reclaiming(nn)) {
5273                 new_timeo = 0;
5274                 goto out;
5275         }
5276         dprintk("NFSD: end of grace period\n");
5277         nfsd4_end_grace(nn);
5278         INIT_LIST_HEAD(&reaplist);
5279
5280         spin_lock(&nn->s2s_cp_lock);
5281         idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
5282                 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
5283                 if (cps->cp_stateid.sc_type == NFS4_COPYNOTIFY_STID &&
5284                                 cps->cpntf_time > cutoff)
5285                         _free_cpntf_state_locked(nn, cps);
5286         }
5287         spin_unlock(&nn->s2s_cp_lock);
5288
5289         spin_lock(&nn->client_lock);
5290         list_for_each_safe(pos, next, &nn->client_lru) {
5291                 clp = list_entry(pos, struct nfs4_client, cl_lru);
5292                 if (clp->cl_time > cutoff) {
5293                         t = clp->cl_time - cutoff;
5294                         new_timeo = min(new_timeo, t);
5295                         break;
5296                 }
5297                 if (mark_client_expired_locked(clp)) {
5298                         dprintk("NFSD: client in use (clientid %08x)\n",
5299                                 clp->cl_clientid.cl_id);
5300                         continue;
5301                 }
5302                 list_add(&clp->cl_lru, &reaplist);
5303         }
5304         spin_unlock(&nn->client_lock);
5305         list_for_each_safe(pos, next, &reaplist) {
5306                 clp = list_entry(pos, struct nfs4_client, cl_lru);
5307                 dprintk("NFSD: purging unused client (clientid %08x)\n",
5308                         clp->cl_clientid.cl_id);
5309                 list_del_init(&clp->cl_lru);
5310                 expire_client(clp);
5311         }
5312         spin_lock(&state_lock);
5313         list_for_each_safe(pos, next, &nn->del_recall_lru) {
5314                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5315                 if (dp->dl_time > cutoff) {
5316                         t = dp->dl_time - cutoff;
5317                         new_timeo = min(new_timeo, t);
5318                         break;
5319                 }
5320                 WARN_ON(!unhash_delegation_locked(dp));
5321                 list_add(&dp->dl_recall_lru, &reaplist);
5322         }
5323         spin_unlock(&state_lock);
5324         while (!list_empty(&reaplist)) {
5325                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
5326                                         dl_recall_lru);
5327                 list_del_init(&dp->dl_recall_lru);
5328                 revoke_delegation(dp);
5329         }
5330
5331         spin_lock(&nn->client_lock);
5332         while (!list_empty(&nn->close_lru)) {
5333                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
5334                                         oo_close_lru);
5335                 if (oo->oo_time > cutoff) {
5336                         t = oo->oo_time - cutoff;
5337                         new_timeo = min(new_timeo, t);
5338                         break;
5339                 }
5340                 list_del_init(&oo->oo_close_lru);
5341                 stp = oo->oo_last_closed_stid;
5342                 oo->oo_last_closed_stid = NULL;
5343                 spin_unlock(&nn->client_lock);
5344                 nfs4_put_stid(&stp->st_stid);
5345                 spin_lock(&nn->client_lock);
5346         }
5347         spin_unlock(&nn->client_lock);
5348
5349         /*
5350          * It's possible for a client to try and acquire an already held lock
5351          * that is being held for a long time, and then lose interest in it.
5352          * So, we clean out any un-revisited request after a lease period
5353          * under the assumption that the client is no longer interested.
5354          *
5355          * RFC5661, sec. 9.6 states that the client must not rely on getting
5356          * notifications and must continue to poll for locks, even when the
5357          * server supports them. Thus this shouldn't lead to clients blocking
5358          * indefinitely once the lock does become free.
5359          */
5360         BUG_ON(!list_empty(&reaplist));
5361         spin_lock(&nn->blocked_locks_lock);
5362         while (!list_empty(&nn->blocked_locks_lru)) {
5363                 nbl = list_first_entry(&nn->blocked_locks_lru,
5364                                         struct nfsd4_blocked_lock, nbl_lru);
5365                 if (nbl->nbl_time > cutoff) {
5366                         t = nbl->nbl_time - cutoff;
5367                         new_timeo = min(new_timeo, t);
5368                         break;
5369                 }
5370                 list_move(&nbl->nbl_lru, &reaplist);
5371                 list_del_init(&nbl->nbl_list);
5372         }
5373         spin_unlock(&nn->blocked_locks_lock);
5374
5375         while (!list_empty(&reaplist)) {
5376                 nbl = list_first_entry(&reaplist,
5377                                         struct nfsd4_blocked_lock, nbl_lru);
5378                 list_del_init(&nbl->nbl_lru);
5379                 free_blocked_lock(nbl);
5380         }
5381 out:
5382         new_timeo = max_t(time64_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
5383         return new_timeo;
5384 }
5385
5386 static struct workqueue_struct *laundry_wq;
5387 static void laundromat_main(struct work_struct *);
5388
5389 static void
5390 laundromat_main(struct work_struct *laundry)
5391 {
5392         time64_t t;
5393         struct delayed_work *dwork = to_delayed_work(laundry);
5394         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
5395                                            laundromat_work);
5396
5397         t = nfs4_laundromat(nn);
5398         dprintk("NFSD: laundromat_main - sleeping for %lld seconds\n", t);
5399         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
5400 }
5401
5402 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
5403 {
5404         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
5405                 return nfserr_bad_stateid;
5406         return nfs_ok;
5407 }
5408
5409 static inline int
5410 access_permit_read(struct nfs4_ol_stateid *stp)
5411 {
5412         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
5413                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
5414                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
5415 }
5416
5417 static inline int
5418 access_permit_write(struct nfs4_ol_stateid *stp)
5419 {
5420         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
5421                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
5422 }
5423
5424 static
5425 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
5426 {
5427         __be32 status = nfserr_openmode;
5428
5429         /* For lock stateid's, we test the parent open, not the lock: */
5430         if (stp->st_openstp)
5431                 stp = stp->st_openstp;
5432         if ((flags & WR_STATE) && !access_permit_write(stp))
5433                 goto out;
5434         if ((flags & RD_STATE) && !access_permit_read(stp))
5435                 goto out;
5436         status = nfs_ok;
5437 out:
5438         return status;
5439 }
5440
5441 static inline __be32
5442 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
5443 {
5444         if (ONE_STATEID(stateid) && (flags & RD_STATE))
5445                 return nfs_ok;
5446         else if (opens_in_grace(net)) {
5447                 /* Answer in remaining cases depends on existence of
5448                  * conflicting state; so we must wait out the grace period. */
5449                 return nfserr_grace;
5450         } else if (flags & WR_STATE)
5451                 return nfs4_share_conflict(current_fh,
5452                                 NFS4_SHARE_DENY_WRITE);
5453         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
5454                 return nfs4_share_conflict(current_fh,
5455                                 NFS4_SHARE_DENY_READ);
5456 }
5457
5458 /*
5459  * Allow READ/WRITE during grace period on recovered state only for files
5460  * that are not able to provide mandatory locking.
5461  */
5462 static inline int
5463 grace_disallows_io(struct net *net, struct inode *inode)
5464 {
5465         return opens_in_grace(net) && mandatory_lock(inode);
5466 }
5467
5468 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
5469 {
5470         /*
5471          * When sessions are used the stateid generation number is ignored
5472          * when it is zero.
5473          */
5474         if (has_session && in->si_generation == 0)
5475                 return nfs_ok;
5476
5477         if (in->si_generation == ref->si_generation)
5478                 return nfs_ok;
5479
5480         /* If the client sends us a stateid from the future, it's buggy: */
5481         if (nfsd4_stateid_generation_after(in, ref))
5482                 return nfserr_bad_stateid;
5483         /*
5484          * However, we could see a stateid from the past, even from a
5485          * non-buggy client.  For example, if the client sends a lock
5486          * while some IO is outstanding, the lock may bump si_generation
5487          * while the IO is still in flight.  The client could avoid that
5488          * situation by waiting for responses on all the IO requests,
5489          * but better performance may result in retrying IO that
5490          * receives an old_stateid error if requests are rarely
5491          * reordered in flight:
5492          */
5493         return nfserr_old_stateid;
5494 }
5495
5496 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
5497 {
5498         __be32 ret;
5499
5500         spin_lock(&s->sc_lock);
5501         ret = nfsd4_verify_open_stid(s);
5502         if (ret == nfs_ok)
5503                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
5504         spin_unlock(&s->sc_lock);
5505         return ret;
5506 }
5507
5508 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
5509 {
5510         if (ols->st_stateowner->so_is_open_owner &&
5511             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
5512                 return nfserr_bad_stateid;
5513         return nfs_ok;
5514 }
5515
5516 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
5517 {
5518         struct nfs4_stid *s;
5519         __be32 status = nfserr_bad_stateid;
5520
5521         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5522                 CLOSE_STATEID(stateid))
5523                 return status;
5524         /* Client debugging aid. */
5525         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
5526                 char addr_str[INET6_ADDRSTRLEN];
5527                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
5528                                  sizeof(addr_str));
5529                 pr_warn_ratelimited("NFSD: client %s testing state ID "
5530                                         "with incorrect client ID\n", addr_str);
5531                 return status;
5532         }
5533         spin_lock(&cl->cl_lock);
5534         s = find_stateid_locked(cl, stateid);
5535         if (!s)
5536                 goto out_unlock;
5537         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
5538         if (status)
5539                 goto out_unlock;
5540         switch (s->sc_type) {
5541         case NFS4_DELEG_STID:
5542                 status = nfs_ok;
5543                 break;
5544         case NFS4_REVOKED_DELEG_STID:
5545                 status = nfserr_deleg_revoked;
5546                 break;
5547         case NFS4_OPEN_STID:
5548         case NFS4_LOCK_STID:
5549                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5550                 break;
5551         default:
5552                 printk("unknown stateid type %x\n", s->sc_type);
5553                 /* Fallthrough */
5554         case NFS4_CLOSED_STID:
5555         case NFS4_CLOSED_DELEG_STID:
5556                 status = nfserr_bad_stateid;
5557         }
5558 out_unlock:
5559         spin_unlock(&cl->cl_lock);
5560         return status;
5561 }
5562
5563 __be32
5564 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5565                      stateid_t *stateid, unsigned char typemask,
5566                      struct nfs4_stid **s, struct nfsd_net *nn)
5567 {
5568         __be32 status;
5569         bool return_revoked = false;
5570
5571         /*
5572          *  only return revoked delegations if explicitly asked.
5573          *  otherwise we report revoked or bad_stateid status.
5574          */
5575         if (typemask & NFS4_REVOKED_DELEG_STID)
5576                 return_revoked = true;
5577         else if (typemask & NFS4_DELEG_STID)
5578                 typemask |= NFS4_REVOKED_DELEG_STID;
5579
5580         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5581                 CLOSE_STATEID(stateid))
5582                 return nfserr_bad_stateid;
5583         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn,
5584                                  false);
5585         if (status == nfserr_stale_clientid) {
5586                 if (cstate->session)
5587                         return nfserr_bad_stateid;
5588                 return nfserr_stale_stateid;
5589         }
5590         if (status)
5591                 return status;
5592         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5593         if (!*s)
5594                 return nfserr_bad_stateid;
5595         if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5596                 nfs4_put_stid(*s);
5597                 if (cstate->minorversion)
5598                         return nfserr_deleg_revoked;
5599                 return nfserr_bad_stateid;
5600         }
5601         return nfs_ok;
5602 }
5603
5604 static struct nfsd_file *
5605 nfs4_find_file(struct nfs4_stid *s, int flags)
5606 {
5607         if (!s)
5608                 return NULL;
5609
5610         switch (s->sc_type) {
5611         case NFS4_DELEG_STID:
5612                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5613                         return NULL;
5614                 return nfsd_file_get(s->sc_file->fi_deleg_file);
5615         case NFS4_OPEN_STID:
5616         case NFS4_LOCK_STID:
5617                 if (flags & RD_STATE)
5618                         return find_readable_file(s->sc_file);
5619                 else
5620                         return find_writeable_file(s->sc_file);
5621                 break;
5622         }
5623
5624         return NULL;
5625 }
5626
5627 static __be32
5628 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
5629 {
5630         __be32 status;
5631
5632         status = nfsd4_check_openowner_confirmed(ols);
5633         if (status)
5634                 return status;
5635         return nfs4_check_openmode(ols, flags);
5636 }
5637
5638 static __be32
5639 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5640                 struct nfsd_file **nfp, int flags)
5641 {
5642         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5643         struct nfsd_file *nf;
5644         __be32 status;
5645
5646         nf = nfs4_find_file(s, flags);
5647         if (nf) {
5648                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5649                                 acc | NFSD_MAY_OWNER_OVERRIDE);
5650                 if (status) {
5651                         nfsd_file_put(nf);
5652                         goto out;
5653                 }
5654         } else {
5655                 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
5656                 if (status)
5657                         return status;
5658         }
5659         *nfp = nf;
5660 out:
5661         return status;
5662 }
5663 static void
5664 _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
5665 {
5666         WARN_ON_ONCE(cps->cp_stateid.sc_type != NFS4_COPYNOTIFY_STID);
5667         if (!refcount_dec_and_test(&cps->cp_stateid.sc_count))
5668                 return;
5669         list_del(&cps->cp_list);
5670         idr_remove(&nn->s2s_cp_stateids,
5671                    cps->cp_stateid.stid.si_opaque.so_id);
5672         kfree(cps);
5673 }
5674 /*
5675  * A READ from an inter server to server COPY will have a
5676  * copy stateid. Look up the copy notify stateid from the
5677  * idr structure and take a reference on it.
5678  */
5679 __be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
5680                           struct nfs4_client *clp,
5681                           struct nfs4_cpntf_state **cps)
5682 {
5683         copy_stateid_t *cps_t;
5684         struct nfs4_cpntf_state *state = NULL;
5685
5686         if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
5687                 return nfserr_bad_stateid;
5688         spin_lock(&nn->s2s_cp_lock);
5689         cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
5690         if (cps_t) {
5691                 state = container_of(cps_t, struct nfs4_cpntf_state,
5692                                      cp_stateid);
5693                 if (state->cp_stateid.sc_type != NFS4_COPYNOTIFY_STID) {
5694                         state = NULL;
5695                         goto unlock;
5696                 }
5697                 if (!clp)
5698                         refcount_inc(&state->cp_stateid.sc_count);
5699                 else
5700                         _free_cpntf_state_locked(nn, state);
5701         }
5702 unlock:
5703         spin_unlock(&nn->s2s_cp_lock);
5704         if (!state)
5705                 return nfserr_bad_stateid;
5706         if (!clp && state)
5707                 *cps = state;
5708         return 0;
5709 }
5710
5711 static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
5712                                struct nfs4_stid **stid)
5713 {
5714         __be32 status;
5715         struct nfs4_cpntf_state *cps = NULL;
5716         struct nfsd4_compound_state cstate;
5717
5718         status = manage_cpntf_state(nn, st, NULL, &cps);
5719         if (status)
5720                 return status;
5721
5722         cps->cpntf_time = ktime_get_boottime_seconds();
5723         memset(&cstate, 0, sizeof(cstate));
5724         status = lookup_clientid(&cps->cp_p_clid, &cstate, nn, true);
5725         if (status)
5726                 goto out;
5727         status = nfsd4_lookup_stateid(&cstate, &cps->cp_p_stateid,
5728                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5729                                 stid, nn);
5730         put_client_renew(cstate.clp);
5731 out:
5732         nfs4_put_cpntf_state(nn, cps);
5733         return status;
5734 }
5735
5736 void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
5737 {
5738         spin_lock(&nn->s2s_cp_lock);
5739         _free_cpntf_state_locked(nn, cps);
5740         spin_unlock(&nn->s2s_cp_lock);
5741 }
5742
5743 /*
5744  * Checks for stateid operations
5745  */
5746 __be32
5747 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5748                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5749                 stateid_t *stateid, int flags, struct nfsd_file **nfp,
5750                 struct nfs4_stid **cstid)
5751 {
5752         struct inode *ino = d_inode(fhp->fh_dentry);
5753         struct net *net = SVC_NET(rqstp);
5754         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5755         struct nfs4_stid *s = NULL;
5756         __be32 status;
5757
5758         if (nfp)
5759                 *nfp = NULL;
5760
5761         if (grace_disallows_io(net, ino))
5762                 return nfserr_grace;
5763
5764         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5765                 status = check_special_stateids(net, fhp, stateid, flags);
5766                 goto done;
5767         }
5768
5769         status = nfsd4_lookup_stateid(cstate, stateid,
5770                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5771                                 &s, nn);
5772         if (status == nfserr_bad_stateid)
5773                 status = find_cpntf_state(nn, stateid, &s);
5774         if (status)
5775                 return status;
5776         status = nfsd4_stid_check_stateid_generation(stateid, s,
5777                         nfsd4_has_session(cstate));
5778         if (status)
5779                 goto out;
5780
5781         switch (s->sc_type) {
5782         case NFS4_DELEG_STID:
5783                 status = nfs4_check_delegmode(delegstateid(s), flags);
5784                 break;
5785         case NFS4_OPEN_STID:
5786         case NFS4_LOCK_STID:
5787                 status = nfs4_check_olstateid(openlockstateid(s), flags);
5788                 break;
5789         default:
5790                 status = nfserr_bad_stateid;
5791                 break;
5792         }
5793         if (status)
5794                 goto out;
5795         status = nfs4_check_fh(fhp, s);
5796
5797 done:
5798         if (status == nfs_ok && nfp)
5799                 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
5800 out:
5801         if (s) {
5802                 if (!status && cstid)
5803                         *cstid = s;
5804                 else
5805                         nfs4_put_stid(s);
5806         }
5807         return status;
5808 }
5809
5810 /*
5811  * Test if the stateid is valid
5812  */
5813 __be32
5814 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5815                    union nfsd4_op_u *u)
5816 {
5817         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5818         struct nfsd4_test_stateid_id *stateid;
5819         struct nfs4_client *cl = cstate->session->se_client;
5820
5821         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5822                 stateid->ts_id_status =
5823                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5824
5825         return nfs_ok;
5826 }
5827
5828 static __be32
5829 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5830 {
5831         struct nfs4_ol_stateid *stp = openlockstateid(s);
5832         __be32 ret;
5833
5834         ret = nfsd4_lock_ol_stateid(stp);
5835         if (ret)
5836                 goto out_put_stid;
5837
5838         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5839         if (ret)
5840                 goto out;
5841
5842         ret = nfserr_locks_held;
5843         if (check_for_locks(stp->st_stid.sc_file,
5844                             lockowner(stp->st_stateowner)))
5845                 goto out;
5846
5847         release_lock_stateid(stp);
5848         ret = nfs_ok;
5849
5850 out:
5851         mutex_unlock(&stp->st_mutex);
5852 out_put_stid:
5853         nfs4_put_stid(s);
5854         return ret;
5855 }
5856
5857 __be32
5858 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5859                    union nfsd4_op_u *u)
5860 {
5861         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5862         stateid_t *stateid = &free_stateid->fr_stateid;
5863         struct nfs4_stid *s;
5864         struct nfs4_delegation *dp;
5865         struct nfs4_client *cl = cstate->session->se_client;
5866         __be32 ret = nfserr_bad_stateid;
5867
5868         spin_lock(&cl->cl_lock);
5869         s = find_stateid_locked(cl, stateid);
5870         if (!s)
5871                 goto out_unlock;
5872         spin_lock(&s->sc_lock);
5873         switch (s->sc_type) {
5874         case NFS4_DELEG_STID:
5875                 ret = nfserr_locks_held;
5876                 break;
5877         case NFS4_OPEN_STID:
5878                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5879                 if (ret)
5880                         break;
5881                 ret = nfserr_locks_held;
5882                 break;
5883         case NFS4_LOCK_STID:
5884                 spin_unlock(&s->sc_lock);
5885                 refcount_inc(&s->sc_count);
5886                 spin_unlock(&cl->cl_lock);
5887                 ret = nfsd4_free_lock_stateid(stateid, s);
5888                 goto out;
5889         case NFS4_REVOKED_DELEG_STID:
5890                 spin_unlock(&s->sc_lock);
5891                 dp = delegstateid(s);
5892                 list_del_init(&dp->dl_recall_lru);
5893                 spin_unlock(&cl->cl_lock);
5894                 nfs4_put_stid(s);
5895                 ret = nfs_ok;
5896                 goto out;
5897         /* Default falls through and returns nfserr_bad_stateid */
5898         }
5899         spin_unlock(&s->sc_lock);
5900 out_unlock:
5901         spin_unlock(&cl->cl_lock);
5902 out:
5903         return ret;
5904 }
5905
5906 static inline int
5907 setlkflg (int type)
5908 {
5909         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5910                 RD_STATE : WR_STATE;
5911 }
5912
5913 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5914 {
5915         struct svc_fh *current_fh = &cstate->current_fh;
5916         struct nfs4_stateowner *sop = stp->st_stateowner;
5917         __be32 status;
5918
5919         status = nfsd4_check_seqid(cstate, sop, seqid);
5920         if (status)
5921                 return status;
5922         status = nfsd4_lock_ol_stateid(stp);
5923         if (status != nfs_ok)
5924                 return status;
5925         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5926         if (status == nfs_ok)
5927                 status = nfs4_check_fh(current_fh, &stp->st_stid);
5928         if (status != nfs_ok)
5929                 mutex_unlock(&stp->st_mutex);
5930         return status;
5931 }
5932
5933 /*
5934  * Checks for sequence id mutating operations.
5935  */
5936 static __be32
5937 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5938                          stateid_t *stateid, char typemask,
5939                          struct nfs4_ol_stateid **stpp,
5940                          struct nfsd_net *nn)
5941 {
5942         __be32 status;
5943         struct nfs4_stid *s;
5944         struct nfs4_ol_stateid *stp = NULL;
5945
5946         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5947                 seqid, STATEID_VAL(stateid));
5948
5949         *stpp = NULL;
5950         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5951         if (status)
5952                 return status;
5953         stp = openlockstateid(s);
5954         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5955
5956         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5957         if (!status)
5958                 *stpp = stp;
5959         else
5960                 nfs4_put_stid(&stp->st_stid);
5961         return status;
5962 }
5963
5964 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5965                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5966 {
5967         __be32 status;
5968         struct nfs4_openowner *oo;
5969         struct nfs4_ol_stateid *stp;
5970
5971         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5972                                                 NFS4_OPEN_STID, &stp, nn);
5973         if (status)
5974                 return status;
5975         oo = openowner(stp->st_stateowner);
5976         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5977                 mutex_unlock(&stp->st_mutex);
5978                 nfs4_put_stid(&stp->st_stid);
5979                 return nfserr_bad_stateid;
5980         }
5981         *stpp = stp;
5982         return nfs_ok;
5983 }
5984
5985 __be32
5986 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5987                    union nfsd4_op_u *u)
5988 {
5989         struct nfsd4_open_confirm *oc = &u->open_confirm;
5990         __be32 status;
5991         struct nfs4_openowner *oo;
5992         struct nfs4_ol_stateid *stp;
5993         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5994
5995         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5996                         cstate->current_fh.fh_dentry);
5997
5998         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5999         if (status)
6000                 return status;
6001
6002         status = nfs4_preprocess_seqid_op(cstate,
6003                                         oc->oc_seqid, &oc->oc_req_stateid,
6004                                         NFS4_OPEN_STID, &stp, nn);
6005         if (status)
6006                 goto out;
6007         oo = openowner(stp->st_stateowner);
6008         status = nfserr_bad_stateid;
6009         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
6010                 mutex_unlock(&stp->st_mutex);
6011                 goto put_stateid;
6012         }
6013         oo->oo_flags |= NFS4_OO_CONFIRMED;
6014         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
6015         mutex_unlock(&stp->st_mutex);
6016         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
6017                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
6018
6019         nfsd4_client_record_create(oo->oo_owner.so_client);
6020         status = nfs_ok;
6021 put_stateid:
6022         nfs4_put_stid(&stp->st_stid);
6023 out:
6024         nfsd4_bump_seqid(cstate, status);
6025         return status;
6026 }
6027
6028 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
6029 {
6030         if (!test_access(access, stp))
6031                 return;
6032         nfs4_file_put_access(stp->st_stid.sc_file, access);
6033         clear_access(access, stp);
6034 }
6035
6036 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
6037 {
6038         switch (to_access) {
6039         case NFS4_SHARE_ACCESS_READ:
6040                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
6041                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6042                 break;
6043         case NFS4_SHARE_ACCESS_WRITE:
6044                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
6045                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6046                 break;
6047         case NFS4_SHARE_ACCESS_BOTH:
6048                 break;
6049         default:
6050                 WARN_ON_ONCE(1);
6051         }
6052 }
6053
6054 __be32
6055 nfsd4_open_downgrade(struct svc_rqst *rqstp,
6056                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
6057 {
6058         struct nfsd4_open_downgrade *od = &u->open_downgrade;
6059         __be32 status;
6060         struct nfs4_ol_stateid *stp;
6061         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6062
6063         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
6064                         cstate->current_fh.fh_dentry);
6065
6066         /* We don't yet support WANT bits: */
6067         if (od->od_deleg_want)
6068                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
6069                         od->od_deleg_want);
6070
6071         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
6072                                         &od->od_stateid, &stp, nn);
6073         if (status)
6074                 goto out;
6075         status = nfserr_inval;
6076         if (!test_access(od->od_share_access, stp)) {
6077                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
6078                         stp->st_access_bmap, od->od_share_access);
6079                 goto put_stateid;
6080         }
6081         if (!test_deny(od->od_share_deny, stp)) {
6082                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
6083                         stp->st_deny_bmap, od->od_share_deny);
6084                 goto put_stateid;
6085         }
6086         nfs4_stateid_downgrade(stp, od->od_share_access);
6087         reset_union_bmap_deny(od->od_share_deny, stp);
6088         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
6089         status = nfs_ok;
6090 put_stateid:
6091         mutex_unlock(&stp->st_mutex);
6092         nfs4_put_stid(&stp->st_stid);
6093 out:
6094         nfsd4_bump_seqid(cstate, status);
6095         return status;
6096 }
6097
6098 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
6099 {
6100         struct nfs4_client *clp = s->st_stid.sc_client;
6101         bool unhashed;
6102         LIST_HEAD(reaplist);
6103
6104         spin_lock(&clp->cl_lock);
6105         unhashed = unhash_open_stateid(s, &reaplist);
6106
6107         if (clp->cl_minorversion) {
6108                 if (unhashed)
6109                         put_ol_stateid_locked(s, &reaplist);
6110                 spin_unlock(&clp->cl_lock);
6111                 free_ol_stateid_reaplist(&reaplist);
6112         } else {
6113                 spin_unlock(&clp->cl_lock);
6114                 free_ol_stateid_reaplist(&reaplist);
6115                 if (unhashed)
6116                         move_to_close_lru(s, clp->net);
6117         }
6118 }
6119
6120 /*
6121  * nfs4_unlock_state() called after encode
6122  */
6123 __be32
6124 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6125                 union nfsd4_op_u *u)
6126 {
6127         struct nfsd4_close *close = &u->close;
6128         __be32 status;
6129         struct nfs4_ol_stateid *stp;
6130         struct net *net = SVC_NET(rqstp);
6131         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6132
6133         dprintk("NFSD: nfsd4_close on file %pd\n",
6134                         cstate->current_fh.fh_dentry);
6135
6136         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
6137                                         &close->cl_stateid,
6138                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
6139                                         &stp, nn);
6140         nfsd4_bump_seqid(cstate, status);
6141         if (status)
6142                 goto out;
6143
6144         stp->st_stid.sc_type = NFS4_CLOSED_STID;
6145
6146         /*
6147          * Technically we don't _really_ have to increment or copy it, since
6148          * it should just be gone after this operation and we clobber the
6149          * copied value below, but we continue to do so here just to ensure
6150          * that racing ops see that there was a state change.
6151          */
6152         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
6153
6154         nfsd4_close_open_stateid(stp);
6155         mutex_unlock(&stp->st_mutex);
6156
6157         /* v4.1+ suggests that we send a special stateid in here, since the
6158          * clients should just ignore this anyway. Since this is not useful
6159          * for v4.0 clients either, we set it to the special close_stateid
6160          * universally.
6161          *
6162          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
6163          */
6164         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
6165
6166         /* put reference from nfs4_preprocess_seqid_op */
6167         nfs4_put_stid(&stp->st_stid);
6168 out:
6169         return status;
6170 }
6171
6172 __be32
6173 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6174                   union nfsd4_op_u *u)
6175 {
6176         struct nfsd4_delegreturn *dr = &u->delegreturn;
6177         struct nfs4_delegation *dp;
6178         stateid_t *stateid = &dr->dr_stateid;
6179         struct nfs4_stid *s;
6180         __be32 status;
6181         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6182
6183         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6184                 return status;
6185
6186         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
6187         if (status)
6188                 goto out;
6189         dp = delegstateid(s);
6190         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
6191         if (status)
6192                 goto put_stateid;
6193
6194         destroy_delegation(dp);
6195 put_stateid:
6196         nfs4_put_stid(&dp->dl_stid);
6197 out:
6198         return status;
6199 }
6200
6201 static inline u64
6202 end_offset(u64 start, u64 len)
6203 {
6204         u64 end;
6205
6206         end = start + len;
6207         return end >= start ? end: NFS4_MAX_UINT64;
6208 }
6209
6210 /* last octet in a range */
6211 static inline u64
6212 last_byte_offset(u64 start, u64 len)
6213 {
6214         u64 end;
6215
6216         WARN_ON_ONCE(!len);
6217         end = start + len;
6218         return end > start ? end - 1: NFS4_MAX_UINT64;
6219 }
6220
6221 /*
6222  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
6223  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
6224  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
6225  * locking, this prevents us from being completely protocol-compliant.  The
6226  * real solution to this problem is to start using unsigned file offsets in
6227  * the VFS, but this is a very deep change!
6228  */
6229 static inline void
6230 nfs4_transform_lock_offset(struct file_lock *lock)
6231 {
6232         if (lock->fl_start < 0)
6233                 lock->fl_start = OFFSET_MAX;
6234         if (lock->fl_end < 0)
6235                 lock->fl_end = OFFSET_MAX;
6236 }
6237
6238 static fl_owner_t
6239 nfsd4_fl_get_owner(fl_owner_t owner)
6240 {
6241         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6242
6243         nfs4_get_stateowner(&lo->lo_owner);
6244         return owner;
6245 }
6246
6247 static void
6248 nfsd4_fl_put_owner(fl_owner_t owner)
6249 {
6250         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6251
6252         if (lo)
6253                 nfs4_put_stateowner(&lo->lo_owner);
6254 }
6255
6256 static void
6257 nfsd4_lm_notify(struct file_lock *fl)
6258 {
6259         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
6260         struct net                      *net = lo->lo_owner.so_client->net;
6261         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
6262         struct nfsd4_blocked_lock       *nbl = container_of(fl,
6263                                                 struct nfsd4_blocked_lock, nbl_lock);
6264         bool queue = false;
6265
6266         /* An empty list means that something else is going to be using it */
6267         spin_lock(&nn->blocked_locks_lock);
6268         if (!list_empty(&nbl->nbl_list)) {
6269                 list_del_init(&nbl->nbl_list);
6270                 list_del_init(&nbl->nbl_lru);
6271                 queue = true;
6272         }
6273         spin_unlock(&nn->blocked_locks_lock);
6274
6275         if (queue)
6276                 nfsd4_run_cb(&nbl->nbl_cb);
6277 }
6278
6279 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
6280         .lm_notify = nfsd4_lm_notify,
6281         .lm_get_owner = nfsd4_fl_get_owner,
6282         .lm_put_owner = nfsd4_fl_put_owner,
6283 };
6284
6285 static inline void
6286 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
6287 {
6288         struct nfs4_lockowner *lo;
6289
6290         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
6291                 lo = (struct nfs4_lockowner *) fl->fl_owner;
6292                 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
6293                                                 GFP_KERNEL);
6294                 if (!deny->ld_owner.data)
6295                         /* We just don't care that much */
6296                         goto nevermind;
6297                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
6298         } else {
6299 nevermind:
6300                 deny->ld_owner.len = 0;
6301                 deny->ld_owner.data = NULL;
6302                 deny->ld_clientid.cl_boot = 0;
6303                 deny->ld_clientid.cl_id = 0;
6304         }
6305         deny->ld_start = fl->fl_start;
6306         deny->ld_length = NFS4_MAX_UINT64;
6307         if (fl->fl_end != NFS4_MAX_UINT64)
6308                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
6309         deny->ld_type = NFS4_READ_LT;
6310         if (fl->fl_type != F_RDLCK)
6311                 deny->ld_type = NFS4_WRITE_LT;
6312 }
6313
6314 static struct nfs4_lockowner *
6315 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
6316 {
6317         unsigned int strhashval = ownerstr_hashval(owner);
6318         struct nfs4_stateowner *so;
6319
6320         lockdep_assert_held(&clp->cl_lock);
6321
6322         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
6323                             so_strhash) {
6324                 if (so->so_is_open_owner)
6325                         continue;
6326                 if (same_owner_str(so, owner))
6327                         return lockowner(nfs4_get_stateowner(so));
6328         }
6329         return NULL;
6330 }
6331
6332 static struct nfs4_lockowner *
6333 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
6334 {
6335         struct nfs4_lockowner *lo;
6336
6337         spin_lock(&clp->cl_lock);
6338         lo = find_lockowner_str_locked(clp, owner);
6339         spin_unlock(&clp->cl_lock);
6340         return lo;
6341 }
6342
6343 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
6344 {
6345         unhash_lockowner_locked(lockowner(sop));
6346 }
6347
6348 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
6349 {
6350         struct nfs4_lockowner *lo = lockowner(sop);
6351
6352         kmem_cache_free(lockowner_slab, lo);
6353 }
6354
6355 static const struct nfs4_stateowner_operations lockowner_ops = {
6356         .so_unhash =    nfs4_unhash_lockowner,
6357         .so_free =      nfs4_free_lockowner,
6358 };
6359
6360 /*
6361  * Alloc a lock owner structure.
6362  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
6363  * occurred.
6364  *
6365  * strhashval = ownerstr_hashval
6366  */
6367 static struct nfs4_lockowner *
6368 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
6369                            struct nfs4_ol_stateid *open_stp,
6370                            struct nfsd4_lock *lock)
6371 {
6372         struct nfs4_lockowner *lo, *ret;
6373
6374         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
6375         if (!lo)
6376                 return NULL;
6377         INIT_LIST_HEAD(&lo->lo_blocked);
6378         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
6379         lo->lo_owner.so_is_open_owner = 0;
6380         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
6381         lo->lo_owner.so_ops = &lockowner_ops;
6382         spin_lock(&clp->cl_lock);
6383         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
6384         if (ret == NULL) {
6385                 list_add(&lo->lo_owner.so_strhash,
6386                          &clp->cl_ownerstr_hashtbl[strhashval]);
6387                 ret = lo;
6388         } else
6389                 nfs4_free_stateowner(&lo->lo_owner);
6390
6391         spin_unlock(&clp->cl_lock);
6392         return ret;
6393 }
6394
6395 static struct nfs4_ol_stateid *
6396 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
6397 {
6398         struct nfs4_ol_stateid *lst;
6399         struct nfs4_client *clp = lo->lo_owner.so_client;
6400
6401         lockdep_assert_held(&clp->cl_lock);
6402
6403         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
6404                 if (lst->st_stid.sc_type != NFS4_LOCK_STID)
6405                         continue;
6406                 if (lst->st_stid.sc_file == fp) {
6407                         refcount_inc(&lst->st_stid.sc_count);
6408                         return lst;
6409                 }
6410         }
6411         return NULL;
6412 }
6413
6414 static struct nfs4_ol_stateid *
6415 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
6416                   struct nfs4_file *fp, struct inode *inode,
6417                   struct nfs4_ol_stateid *open_stp)
6418 {
6419         struct nfs4_client *clp = lo->lo_owner.so_client;
6420         struct nfs4_ol_stateid *retstp;
6421
6422         mutex_init(&stp->st_mutex);
6423         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
6424 retry:
6425         spin_lock(&clp->cl_lock);
6426         spin_lock(&fp->fi_lock);
6427         retstp = find_lock_stateid(lo, fp);
6428         if (retstp)
6429                 goto out_unlock;
6430
6431         refcount_inc(&stp->st_stid.sc_count);
6432         stp->st_stid.sc_type = NFS4_LOCK_STID;
6433         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
6434         get_nfs4_file(fp);
6435         stp->st_stid.sc_file = fp;
6436         stp->st_access_bmap = 0;
6437         stp->st_deny_bmap = open_stp->st_deny_bmap;
6438         stp->st_openstp = open_stp;
6439         list_add(&stp->st_locks, &open_stp->st_locks);
6440         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
6441         list_add(&stp->st_perfile, &fp->fi_stateids);
6442 out_unlock:
6443         spin_unlock(&fp->fi_lock);
6444         spin_unlock(&clp->cl_lock);
6445         if (retstp) {
6446                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
6447                         nfs4_put_stid(&retstp->st_stid);
6448                         goto retry;
6449                 }
6450                 /* To keep mutex tracking happy */
6451                 mutex_unlock(&stp->st_mutex);
6452                 stp = retstp;
6453         }
6454         return stp;
6455 }
6456
6457 static struct nfs4_ol_stateid *
6458 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
6459                             struct inode *inode, struct nfs4_ol_stateid *ost,
6460                             bool *new)
6461 {
6462         struct nfs4_stid *ns = NULL;
6463         struct nfs4_ol_stateid *lst;
6464         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6465         struct nfs4_client *clp = oo->oo_owner.so_client;
6466
6467         *new = false;
6468         spin_lock(&clp->cl_lock);
6469         lst = find_lock_stateid(lo, fi);
6470         spin_unlock(&clp->cl_lock);
6471         if (lst != NULL) {
6472                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
6473                         goto out;
6474                 nfs4_put_stid(&lst->st_stid);
6475         }
6476         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
6477         if (ns == NULL)
6478                 return NULL;
6479
6480         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
6481         if (lst == openlockstateid(ns))
6482                 *new = true;
6483         else
6484                 nfs4_put_stid(ns);
6485 out:
6486         return lst;
6487 }
6488
6489 static int
6490 check_lock_length(u64 offset, u64 length)
6491 {
6492         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
6493                 (length > ~offset)));
6494 }
6495
6496 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
6497 {
6498         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
6499
6500         lockdep_assert_held(&fp->fi_lock);
6501
6502         if (test_access(access, lock_stp))
6503                 return;
6504         __nfs4_file_get_access(fp, access);
6505         set_access(access, lock_stp);
6506 }
6507
6508 static __be32
6509 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
6510                             struct nfs4_ol_stateid *ost,
6511                             struct nfsd4_lock *lock,
6512                             struct nfs4_ol_stateid **plst, bool *new)
6513 {
6514         __be32 status;
6515         struct nfs4_file *fi = ost->st_stid.sc_file;
6516         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6517         struct nfs4_client *cl = oo->oo_owner.so_client;
6518         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
6519         struct nfs4_lockowner *lo;
6520         struct nfs4_ol_stateid *lst;
6521         unsigned int strhashval;
6522
6523         lo = find_lockowner_str(cl, &lock->lk_new_owner);
6524         if (!lo) {
6525                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
6526                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
6527                 if (lo == NULL)
6528                         return nfserr_jukebox;
6529         } else {
6530                 /* with an existing lockowner, seqids must be the same */
6531                 status = nfserr_bad_seqid;
6532                 if (!cstate->minorversion &&
6533                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
6534                         goto out;
6535         }
6536
6537         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
6538         if (lst == NULL) {
6539                 status = nfserr_jukebox;
6540                 goto out;
6541         }
6542
6543         status = nfs_ok;
6544         *plst = lst;
6545 out:
6546         nfs4_put_stateowner(&lo->lo_owner);
6547         return status;
6548 }
6549
6550 /*
6551  *  LOCK operation
6552  */
6553 __be32
6554 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6555            union nfsd4_op_u *u)
6556 {
6557         struct nfsd4_lock *lock = &u->lock;
6558         struct nfs4_openowner *open_sop = NULL;
6559         struct nfs4_lockowner *lock_sop = NULL;
6560         struct nfs4_ol_stateid *lock_stp = NULL;
6561         struct nfs4_ol_stateid *open_stp = NULL;
6562         struct nfs4_file *fp;
6563         struct nfsd_file *nf = NULL;
6564         struct nfsd4_blocked_lock *nbl = NULL;
6565         struct file_lock *file_lock = NULL;
6566         struct file_lock *conflock = NULL;
6567         __be32 status = 0;
6568         int lkflg;
6569         int err;
6570         bool new = false;
6571         unsigned char fl_type;
6572         unsigned int fl_flags = FL_POSIX;
6573         struct net *net = SVC_NET(rqstp);
6574         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6575
6576         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
6577                 (long long) lock->lk_offset,
6578                 (long long) lock->lk_length);
6579
6580         if (check_lock_length(lock->lk_offset, lock->lk_length))
6581                  return nfserr_inval;
6582
6583         if ((status = fh_verify(rqstp, &cstate->current_fh,
6584                                 S_IFREG, NFSD_MAY_LOCK))) {
6585                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
6586                 return status;
6587         }
6588
6589         if (lock->lk_is_new) {
6590                 if (nfsd4_has_session(cstate))
6591                         /* See rfc 5661 18.10.3: given clientid is ignored: */
6592                         memcpy(&lock->lk_new_clientid,
6593                                 &cstate->session->se_client->cl_clientid,
6594                                 sizeof(clientid_t));
6595
6596                 status = nfserr_stale_clientid;
6597                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
6598                         goto out;
6599
6600                 /* validate and update open stateid and open seqid */
6601                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
6602                                         lock->lk_new_open_seqid,
6603                                         &lock->lk_new_open_stateid,
6604                                         &open_stp, nn);
6605                 if (status)
6606                         goto out;
6607                 mutex_unlock(&open_stp->st_mutex);
6608                 open_sop = openowner(open_stp->st_stateowner);
6609                 status = nfserr_bad_stateid;
6610                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
6611                                                 &lock->lk_new_clientid))
6612                         goto out;
6613                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
6614                                                         &lock_stp, &new);
6615         } else {
6616                 status = nfs4_preprocess_seqid_op(cstate,
6617                                        lock->lk_old_lock_seqid,
6618                                        &lock->lk_old_lock_stateid,
6619                                        NFS4_LOCK_STID, &lock_stp, nn);
6620         }
6621         if (status)
6622                 goto out;
6623         lock_sop = lockowner(lock_stp->st_stateowner);
6624
6625         lkflg = setlkflg(lock->lk_type);
6626         status = nfs4_check_openmode(lock_stp, lkflg);
6627         if (status)
6628                 goto out;
6629
6630         status = nfserr_grace;
6631         if (locks_in_grace(net) && !lock->lk_reclaim)
6632                 goto out;
6633         status = nfserr_no_grace;
6634         if (!locks_in_grace(net) && lock->lk_reclaim)
6635                 goto out;
6636
6637         fp = lock_stp->st_stid.sc_file;
6638         switch (lock->lk_type) {
6639                 case NFS4_READW_LT:
6640                         if (nfsd4_has_session(cstate))
6641                                 fl_flags |= FL_SLEEP;
6642                         /* Fallthrough */
6643                 case NFS4_READ_LT:
6644                         spin_lock(&fp->fi_lock);
6645                         nf = find_readable_file_locked(fp);
6646                         if (nf)
6647                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
6648                         spin_unlock(&fp->fi_lock);
6649                         fl_type = F_RDLCK;
6650                         break;
6651                 case NFS4_WRITEW_LT:
6652                         if (nfsd4_has_session(cstate))
6653                                 fl_flags |= FL_SLEEP;
6654                         /* Fallthrough */
6655                 case NFS4_WRITE_LT:
6656                         spin_lock(&fp->fi_lock);
6657                         nf = find_writeable_file_locked(fp);
6658                         if (nf)
6659                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6660                         spin_unlock(&fp->fi_lock);
6661                         fl_type = F_WRLCK;
6662                         break;
6663                 default:
6664                         status = nfserr_inval;
6665                 goto out;
6666         }
6667
6668         if (!nf) {
6669                 status = nfserr_openmode;
6670                 goto out;
6671         }
6672
6673         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6674         if (!nbl) {
6675                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6676                 status = nfserr_jukebox;
6677                 goto out;
6678         }
6679
6680         file_lock = &nbl->nbl_lock;
6681         file_lock->fl_type = fl_type;
6682         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6683         file_lock->fl_pid = current->tgid;
6684         file_lock->fl_file = nf->nf_file;
6685         file_lock->fl_flags = fl_flags;
6686         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6687         file_lock->fl_start = lock->lk_offset;
6688         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6689         nfs4_transform_lock_offset(file_lock);
6690
6691         conflock = locks_alloc_lock();
6692         if (!conflock) {
6693                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6694                 status = nfserr_jukebox;
6695                 goto out;
6696         }
6697
6698         if (fl_flags & FL_SLEEP) {
6699                 nbl->nbl_time = ktime_get_boottime_seconds();
6700                 spin_lock(&nn->blocked_locks_lock);
6701                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6702                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6703                 spin_unlock(&nn->blocked_locks_lock);
6704         }
6705
6706         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
6707         switch (err) {
6708         case 0: /* success! */
6709                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6710                 status = 0;
6711                 if (lock->lk_reclaim)
6712                         nn->somebody_reclaimed = true;
6713                 break;
6714         case FILE_LOCK_DEFERRED:
6715                 nbl = NULL;
6716                 /* Fallthrough */
6717         case -EAGAIN:           /* conflock holds conflicting lock */
6718                 status = nfserr_denied;
6719                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6720                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6721                 break;
6722         case -EDEADLK:
6723                 status = nfserr_deadlock;
6724                 break;
6725         default:
6726                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6727                 status = nfserrno(err);
6728                 break;
6729         }
6730 out:
6731         if (nbl) {
6732                 /* dequeue it if we queued it before */
6733                 if (fl_flags & FL_SLEEP) {
6734                         spin_lock(&nn->blocked_locks_lock);
6735                         list_del_init(&nbl->nbl_list);
6736                         list_del_init(&nbl->nbl_lru);
6737                         spin_unlock(&nn->blocked_locks_lock);
6738                 }
6739                 free_blocked_lock(nbl);
6740         }
6741         if (nf)
6742                 nfsd_file_put(nf);
6743         if (lock_stp) {
6744                 /* Bump seqid manually if the 4.0 replay owner is openowner */
6745                 if (cstate->replay_owner &&
6746                     cstate->replay_owner != &lock_sop->lo_owner &&
6747                     seqid_mutating_err(ntohl(status)))
6748                         lock_sop->lo_owner.so_seqid++;
6749
6750                 /*
6751                  * If this is a new, never-before-used stateid, and we are
6752                  * returning an error, then just go ahead and release it.
6753                  */
6754                 if (status && new)
6755                         release_lock_stateid(lock_stp);
6756
6757                 mutex_unlock(&lock_stp->st_mutex);
6758
6759                 nfs4_put_stid(&lock_stp->st_stid);
6760         }
6761         if (open_stp)
6762                 nfs4_put_stid(&open_stp->st_stid);
6763         nfsd4_bump_seqid(cstate, status);
6764         if (conflock)
6765                 locks_free_lock(conflock);
6766         return status;
6767 }
6768
6769 /*
6770  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6771  * so we do a temporary open here just to get an open file to pass to
6772  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
6773  * inode operation.)
6774  */
6775 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6776 {
6777         struct nfsd_file *nf;
6778         __be32 err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
6779         if (!err) {
6780                 err = nfserrno(vfs_test_lock(nf->nf_file, lock));
6781                 nfsd_file_put(nf);
6782         }
6783         return err;
6784 }
6785
6786 /*
6787  * LOCKT operation
6788  */
6789 __be32
6790 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6791             union nfsd4_op_u *u)
6792 {
6793         struct nfsd4_lockt *lockt = &u->lockt;
6794         struct file_lock *file_lock = NULL;
6795         struct nfs4_lockowner *lo = NULL;
6796         __be32 status;
6797         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6798
6799         if (locks_in_grace(SVC_NET(rqstp)))
6800                 return nfserr_grace;
6801
6802         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6803                  return nfserr_inval;
6804
6805         if (!nfsd4_has_session(cstate)) {
6806                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn,
6807                                          false);
6808                 if (status)
6809                         goto out;
6810         }
6811
6812         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6813                 goto out;
6814
6815         file_lock = locks_alloc_lock();
6816         if (!file_lock) {
6817                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6818                 status = nfserr_jukebox;
6819                 goto out;
6820         }
6821
6822         switch (lockt->lt_type) {
6823                 case NFS4_READ_LT:
6824                 case NFS4_READW_LT:
6825                         file_lock->fl_type = F_RDLCK;
6826                         break;
6827                 case NFS4_WRITE_LT:
6828                 case NFS4_WRITEW_LT:
6829                         file_lock->fl_type = F_WRLCK;
6830                         break;
6831                 default:
6832                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6833                         status = nfserr_inval;
6834                         goto out;
6835         }
6836
6837         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6838         if (lo)
6839                 file_lock->fl_owner = (fl_owner_t)lo;
6840         file_lock->fl_pid = current->tgid;
6841         file_lock->fl_flags = FL_POSIX;
6842
6843         file_lock->fl_start = lockt->lt_offset;
6844         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6845
6846         nfs4_transform_lock_offset(file_lock);
6847
6848         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6849         if (status)
6850                 goto out;
6851
6852         if (file_lock->fl_type != F_UNLCK) {
6853                 status = nfserr_denied;
6854                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6855         }
6856 out:
6857         if (lo)
6858                 nfs4_put_stateowner(&lo->lo_owner);
6859         if (file_lock)
6860                 locks_free_lock(file_lock);
6861         return status;
6862 }
6863
6864 __be32
6865 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6866             union nfsd4_op_u *u)
6867 {
6868         struct nfsd4_locku *locku = &u->locku;
6869         struct nfs4_ol_stateid *stp;
6870         struct nfsd_file *nf = NULL;
6871         struct file_lock *file_lock = NULL;
6872         __be32 status;
6873         int err;
6874         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6875
6876         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6877                 (long long) locku->lu_offset,
6878                 (long long) locku->lu_length);
6879
6880         if (check_lock_length(locku->lu_offset, locku->lu_length))
6881                  return nfserr_inval;
6882
6883         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6884                                         &locku->lu_stateid, NFS4_LOCK_STID,
6885                                         &stp, nn);
6886         if (status)
6887                 goto out;
6888         nf = find_any_file(stp->st_stid.sc_file);
6889         if (!nf) {
6890                 status = nfserr_lock_range;
6891                 goto put_stateid;
6892         }
6893         file_lock = locks_alloc_lock();
6894         if (!file_lock) {
6895                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6896                 status = nfserr_jukebox;
6897                 goto put_file;
6898         }
6899
6900         file_lock->fl_type = F_UNLCK;
6901         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6902         file_lock->fl_pid = current->tgid;
6903         file_lock->fl_file = nf->nf_file;
6904         file_lock->fl_flags = FL_POSIX;
6905         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6906         file_lock->fl_start = locku->lu_offset;
6907
6908         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6909                                                 locku->lu_length);
6910         nfs4_transform_lock_offset(file_lock);
6911
6912         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
6913         if (err) {
6914                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6915                 goto out_nfserr;
6916         }
6917         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6918 put_file:
6919         nfsd_file_put(nf);
6920 put_stateid:
6921         mutex_unlock(&stp->st_mutex);
6922         nfs4_put_stid(&stp->st_stid);
6923 out:
6924         nfsd4_bump_seqid(cstate, status);
6925         if (file_lock)
6926                 locks_free_lock(file_lock);
6927         return status;
6928
6929 out_nfserr:
6930         status = nfserrno(err);
6931         goto put_file;
6932 }
6933
6934 /*
6935  * returns
6936  *      true:  locks held by lockowner
6937  *      false: no locks held by lockowner
6938  */
6939 static bool
6940 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6941 {
6942         struct file_lock *fl;
6943         int status = false;
6944         struct nfsd_file *nf = find_any_file(fp);
6945         struct inode *inode;
6946         struct file_lock_context *flctx;
6947
6948         if (!nf) {
6949                 /* Any valid lock stateid should have some sort of access */
6950                 WARN_ON_ONCE(1);
6951                 return status;
6952         }
6953
6954         inode = locks_inode(nf->nf_file);
6955         flctx = inode->i_flctx;
6956
6957         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6958                 spin_lock(&flctx->flc_lock);
6959                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6960                         if (fl->fl_owner == (fl_owner_t)lowner) {
6961                                 status = true;
6962                                 break;
6963                         }
6964                 }
6965                 spin_unlock(&flctx->flc_lock);
6966         }
6967         nfsd_file_put(nf);
6968         return status;
6969 }
6970
6971 __be32
6972 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6973                         struct nfsd4_compound_state *cstate,
6974                         union nfsd4_op_u *u)
6975 {
6976         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6977         clientid_t *clid = &rlockowner->rl_clientid;
6978         struct nfs4_stateowner *sop;
6979         struct nfs4_lockowner *lo = NULL;
6980         struct nfs4_ol_stateid *stp;
6981         struct xdr_netobj *owner = &rlockowner->rl_owner;
6982         unsigned int hashval = ownerstr_hashval(owner);
6983         __be32 status;
6984         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6985         struct nfs4_client *clp;
6986         LIST_HEAD (reaplist);
6987
6988         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6989                 clid->cl_boot, clid->cl_id);
6990
6991         status = lookup_clientid(clid, cstate, nn, false);
6992         if (status)
6993                 return status;
6994
6995         clp = cstate->clp;
6996         /* Find the matching lock stateowner */
6997         spin_lock(&clp->cl_lock);
6998         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6999                             so_strhash) {
7000
7001                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
7002                         continue;
7003
7004                 /* see if there are still any locks associated with it */
7005                 lo = lockowner(sop);
7006                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
7007                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
7008                                 status = nfserr_locks_held;
7009                                 spin_unlock(&clp->cl_lock);
7010                                 return status;
7011                         }
7012                 }
7013
7014                 nfs4_get_stateowner(sop);
7015                 break;
7016         }
7017         if (!lo) {
7018                 spin_unlock(&clp->cl_lock);
7019                 return status;
7020         }
7021
7022         unhash_lockowner_locked(lo);
7023         while (!list_empty(&lo->lo_owner.so_stateids)) {
7024                 stp = list_first_entry(&lo->lo_owner.so_stateids,
7025                                        struct nfs4_ol_stateid,
7026                                        st_perstateowner);
7027                 WARN_ON(!unhash_lock_stateid(stp));
7028                 put_ol_stateid_locked(stp, &reaplist);
7029         }
7030         spin_unlock(&clp->cl_lock);
7031         free_ol_stateid_reaplist(&reaplist);
7032         remove_blocked_locks(lo);
7033         nfs4_put_stateowner(&lo->lo_owner);
7034
7035         return status;
7036 }
7037
7038 static inline struct nfs4_client_reclaim *
7039 alloc_reclaim(void)
7040 {
7041         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
7042 }
7043
7044 bool
7045 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
7046 {
7047         struct nfs4_client_reclaim *crp;
7048
7049         crp = nfsd4_find_reclaim_client(name, nn);
7050         return (crp && crp->cr_clp);
7051 }
7052
7053 /*
7054  * failure => all reset bets are off, nfserr_no_grace...
7055  *
7056  * The caller is responsible for freeing name.data if NULL is returned (it
7057  * will be freed in nfs4_remove_reclaim_record in the normal case).
7058  */
7059 struct nfs4_client_reclaim *
7060 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
7061                 struct nfsd_net *nn)
7062 {
7063         unsigned int strhashval;
7064         struct nfs4_client_reclaim *crp;
7065
7066         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", name.len, name.data);
7067         crp = alloc_reclaim();
7068         if (crp) {
7069                 strhashval = clientstr_hashval(name);
7070                 INIT_LIST_HEAD(&crp->cr_strhash);
7071                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
7072                 crp->cr_name.data = name.data;
7073                 crp->cr_name.len = name.len;
7074                 crp->cr_princhash.data = princhash.data;
7075                 crp->cr_princhash.len = princhash.len;
7076                 crp->cr_clp = NULL;
7077                 nn->reclaim_str_hashtbl_size++;
7078         }
7079         return crp;
7080 }
7081
7082 void
7083 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
7084 {
7085         list_del(&crp->cr_strhash);
7086         kfree(crp->cr_name.data);
7087         kfree(crp->cr_princhash.data);
7088         kfree(crp);
7089         nn->reclaim_str_hashtbl_size--;
7090 }
7091
7092 void
7093 nfs4_release_reclaim(struct nfsd_net *nn)
7094 {
7095         struct nfs4_client_reclaim *crp = NULL;
7096         int i;
7097
7098         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7099                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
7100                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
7101                                         struct nfs4_client_reclaim, cr_strhash);
7102                         nfs4_remove_reclaim_record(crp, nn);
7103                 }
7104         }
7105         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
7106 }
7107
7108 /*
7109  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
7110 struct nfs4_client_reclaim *
7111 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
7112 {
7113         unsigned int strhashval;
7114         struct nfs4_client_reclaim *crp = NULL;
7115
7116         dprintk("NFSD: nfs4_find_reclaim_client for name %.*s\n", name.len, name.data);
7117
7118         strhashval = clientstr_hashval(name);
7119         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
7120                 if (compare_blob(&crp->cr_name, &name) == 0) {
7121                         return crp;
7122                 }
7123         }
7124         return NULL;
7125 }
7126
7127 /*
7128 * Called from OPEN. Look for clientid in reclaim list.
7129 */
7130 __be32
7131 nfs4_check_open_reclaim(clientid_t *clid,
7132                 struct nfsd4_compound_state *cstate,
7133                 struct nfsd_net *nn)
7134 {
7135         __be32 status;
7136
7137         /* find clientid in conf_id_hashtbl */
7138         status = lookup_clientid(clid, cstate, nn, false);
7139         if (status)
7140                 return nfserr_reclaim_bad;
7141
7142         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
7143                 return nfserr_no_grace;
7144
7145         if (nfsd4_client_record_check(cstate->clp))
7146                 return nfserr_reclaim_bad;
7147
7148         return nfs_ok;
7149 }
7150
7151 #ifdef CONFIG_NFSD_FAULT_INJECTION
7152 static inline void
7153 put_client(struct nfs4_client *clp)
7154 {
7155         atomic_dec(&clp->cl_rpc_users);
7156 }
7157
7158 static struct nfs4_client *
7159 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
7160 {
7161         struct nfs4_client *clp;
7162         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7163                                           nfsd_net_id);
7164
7165         if (!nfsd_netns_ready(nn))
7166                 return NULL;
7167
7168         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7169                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
7170                         return clp;
7171         }
7172         return NULL;
7173 }
7174
7175 u64
7176 nfsd_inject_print_clients(void)
7177 {
7178         struct nfs4_client *clp;
7179         u64 count = 0;
7180         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7181                                           nfsd_net_id);
7182         char buf[INET6_ADDRSTRLEN];
7183
7184         if (!nfsd_netns_ready(nn))
7185                 return 0;
7186
7187         spin_lock(&nn->client_lock);
7188         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7189                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
7190                 pr_info("NFS Client: %s\n", buf);
7191                 ++count;
7192         }
7193         spin_unlock(&nn->client_lock);
7194
7195         return count;
7196 }
7197
7198 u64
7199 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
7200 {
7201         u64 count = 0;
7202         struct nfs4_client *clp;
7203         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7204                                           nfsd_net_id);
7205
7206         if (!nfsd_netns_ready(nn))
7207                 return count;
7208
7209         spin_lock(&nn->client_lock);
7210         clp = nfsd_find_client(addr, addr_size);
7211         if (clp) {
7212                 if (mark_client_expired_locked(clp) == nfs_ok)
7213                         ++count;
7214                 else
7215                         clp = NULL;
7216         }
7217         spin_unlock(&nn->client_lock);
7218
7219         if (clp)
7220                 expire_client(clp);
7221
7222         return count;
7223 }
7224
7225 u64
7226 nfsd_inject_forget_clients(u64 max)
7227 {
7228         u64 count = 0;
7229         struct nfs4_client *clp, *next;
7230         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7231                                                 nfsd_net_id);
7232         LIST_HEAD(reaplist);
7233
7234         if (!nfsd_netns_ready(nn))
7235                 return count;
7236
7237         spin_lock(&nn->client_lock);
7238         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7239                 if (mark_client_expired_locked(clp) == nfs_ok) {
7240                         list_add(&clp->cl_lru, &reaplist);
7241                         if (max != 0 && ++count >= max)
7242                                 break;
7243                 }
7244         }
7245         spin_unlock(&nn->client_lock);
7246
7247         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
7248                 expire_client(clp);
7249
7250         return count;
7251 }
7252
7253 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
7254                              const char *type)
7255 {
7256         char buf[INET6_ADDRSTRLEN];
7257         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
7258         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
7259 }
7260
7261 static void
7262 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
7263                              struct list_head *collect)
7264 {
7265         struct nfs4_client *clp = lst->st_stid.sc_client;
7266         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7267                                           nfsd_net_id);
7268
7269         if (!collect)
7270                 return;
7271
7272         lockdep_assert_held(&nn->client_lock);
7273         atomic_inc(&clp->cl_rpc_users);
7274         list_add(&lst->st_locks, collect);
7275 }
7276
7277 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
7278                                     struct list_head *collect,
7279                                     bool (*func)(struct nfs4_ol_stateid *))
7280 {
7281         struct nfs4_openowner *oop;
7282         struct nfs4_ol_stateid *stp, *st_next;
7283         struct nfs4_ol_stateid *lst, *lst_next;
7284         u64 count = 0;
7285
7286         spin_lock(&clp->cl_lock);
7287         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
7288                 list_for_each_entry_safe(stp, st_next,
7289                                 &oop->oo_owner.so_stateids, st_perstateowner) {
7290                         list_for_each_entry_safe(lst, lst_next,
7291                                         &stp->st_locks, st_locks) {
7292                                 if (func) {
7293                                         if (func(lst))
7294                                                 nfsd_inject_add_lock_to_list(lst,
7295                                                                         collect);
7296                                 }
7297                                 ++count;
7298                                 /*
7299                                  * Despite the fact that these functions deal
7300                                  * with 64-bit integers for "count", we must
7301                                  * ensure that it doesn't blow up the
7302                                  * clp->cl_rpc_users. Throw a warning if we
7303                                  * start to approach INT_MAX here.
7304                                  */
7305                                 WARN_ON_ONCE(count == (INT_MAX / 2));
7306                                 if (count == max)
7307                                         goto out;
7308                         }
7309                 }
7310         }
7311 out:
7312         spin_unlock(&clp->cl_lock);
7313
7314         return count;
7315 }
7316
7317 static u64
7318 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
7319                           u64 max)
7320 {
7321         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
7322 }
7323
7324 static u64
7325 nfsd_print_client_locks(struct nfs4_client *clp)
7326 {
7327         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
7328         nfsd_print_count(clp, count, "locked files");
7329         return count;
7330 }
7331
7332 u64
7333 nfsd_inject_print_locks(void)
7334 {
7335         struct nfs4_client *clp;
7336         u64 count = 0;
7337         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7338                                                 nfsd_net_id);
7339
7340         if (!nfsd_netns_ready(nn))
7341                 return 0;
7342
7343         spin_lock(&nn->client_lock);
7344         list_for_each_entry(clp, &nn->client_lru, cl_lru)
7345                 count += nfsd_print_client_locks(clp);
7346         spin_unlock(&nn->client_lock);
7347
7348         return count;
7349 }
7350
7351 static void
7352 nfsd_reap_locks(struct list_head *reaplist)
7353 {
7354         struct nfs4_client *clp;
7355         struct nfs4_ol_stateid *stp, *next;
7356
7357         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
7358                 list_del_init(&stp->st_locks);
7359                 clp = stp->st_stid.sc_client;
7360                 nfs4_put_stid(&stp->st_stid);
7361                 put_client(clp);
7362         }
7363 }
7364
7365 u64
7366 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
7367 {
7368         unsigned int count = 0;
7369         struct nfs4_client *clp;
7370         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7371                                                 nfsd_net_id);
7372         LIST_HEAD(reaplist);
7373
7374         if (!nfsd_netns_ready(nn))
7375                 return count;
7376
7377         spin_lock(&nn->client_lock);
7378         clp = nfsd_find_client(addr, addr_size);
7379         if (clp)
7380                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
7381         spin_unlock(&nn->client_lock);
7382         nfsd_reap_locks(&reaplist);
7383         return count;
7384 }
7385
7386 u64
7387 nfsd_inject_forget_locks(u64 max)
7388 {
7389         u64 count = 0;
7390         struct nfs4_client *clp;
7391         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7392                                                 nfsd_net_id);
7393         LIST_HEAD(reaplist);
7394
7395         if (!nfsd_netns_ready(nn))
7396                 return count;
7397
7398         spin_lock(&nn->client_lock);
7399         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7400                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
7401                 if (max != 0 && count >= max)
7402                         break;
7403         }
7404         spin_unlock(&nn->client_lock);
7405         nfsd_reap_locks(&reaplist);
7406         return count;
7407 }
7408
7409 static u64
7410 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
7411                               struct list_head *collect,
7412                               void (*func)(struct nfs4_openowner *))
7413 {
7414         struct nfs4_openowner *oop, *next;
7415         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7416                                                 nfsd_net_id);
7417         u64 count = 0;
7418
7419         lockdep_assert_held(&nn->client_lock);
7420
7421         spin_lock(&clp->cl_lock);
7422         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
7423                 if (func) {
7424                         func(oop);
7425                         if (collect) {
7426                                 atomic_inc(&clp->cl_rpc_users);
7427                                 list_add(&oop->oo_perclient, collect);
7428                         }
7429                 }
7430                 ++count;
7431                 /*
7432                  * Despite the fact that these functions deal with
7433                  * 64-bit integers for "count", we must ensure that
7434                  * it doesn't blow up the clp->cl_rpc_users. Throw a
7435                  * warning if we start to approach INT_MAX here.
7436                  */
7437                 WARN_ON_ONCE(count == (INT_MAX / 2));
7438                 if (count == max)
7439                         break;
7440         }
7441         spin_unlock(&clp->cl_lock);
7442
7443         return count;
7444 }
7445
7446 static u64
7447 nfsd_print_client_openowners(struct nfs4_client *clp)
7448 {
7449         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
7450
7451         nfsd_print_count(clp, count, "openowners");
7452         return count;
7453 }
7454
7455 static u64
7456 nfsd_collect_client_openowners(struct nfs4_client *clp,
7457                                struct list_head *collect, u64 max)
7458 {
7459         return nfsd_foreach_client_openowner(clp, max, collect,
7460                                                 unhash_openowner_locked);
7461 }
7462
7463 u64
7464 nfsd_inject_print_openowners(void)
7465 {
7466         struct nfs4_client *clp;
7467         u64 count = 0;
7468         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7469                                                 nfsd_net_id);
7470
7471         if (!nfsd_netns_ready(nn))
7472                 return 0;
7473
7474         spin_lock(&nn->client_lock);
7475         list_for_each_entry(clp, &nn->client_lru, cl_lru)
7476                 count += nfsd_print_client_openowners(clp);
7477         spin_unlock(&nn->client_lock);
7478
7479         return count;
7480 }
7481
7482 static void
7483 nfsd_reap_openowners(struct list_head *reaplist)
7484 {
7485         struct nfs4_client *clp;
7486         struct nfs4_openowner *oop, *next;
7487
7488         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
7489                 list_del_init(&oop->oo_perclient);
7490                 clp = oop->oo_owner.so_client;
7491                 release_openowner(oop);
7492                 put_client(clp);
7493         }
7494 }
7495
7496 u64
7497 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
7498                                      size_t addr_size)
7499 {
7500         unsigned int count = 0;
7501         struct nfs4_client *clp;
7502         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7503                                                 nfsd_net_id);
7504         LIST_HEAD(reaplist);
7505
7506         if (!nfsd_netns_ready(nn))
7507                 return count;
7508
7509         spin_lock(&nn->client_lock);
7510         clp = nfsd_find_client(addr, addr_size);
7511         if (clp)
7512                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
7513         spin_unlock(&nn->client_lock);
7514         nfsd_reap_openowners(&reaplist);
7515         return count;
7516 }
7517
7518 u64
7519 nfsd_inject_forget_openowners(u64 max)
7520 {
7521         u64 count = 0;
7522         struct nfs4_client *clp;
7523         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7524                                                 nfsd_net_id);
7525         LIST_HEAD(reaplist);
7526
7527         if (!nfsd_netns_ready(nn))
7528                 return count;
7529
7530         spin_lock(&nn->client_lock);
7531         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7532                 count += nfsd_collect_client_openowners(clp, &reaplist,
7533                                                         max - count);
7534                 if (max != 0 && count >= max)
7535                         break;
7536         }
7537         spin_unlock(&nn->client_lock);
7538         nfsd_reap_openowners(&reaplist);
7539         return count;
7540 }
7541
7542 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
7543                                      struct list_head *victims)
7544 {
7545         struct nfs4_delegation *dp, *next;
7546         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7547                                                 nfsd_net_id);
7548         u64 count = 0;
7549
7550         lockdep_assert_held(&nn->client_lock);
7551
7552         spin_lock(&state_lock);
7553         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
7554                 if (victims) {
7555                         /*
7556                          * It's not safe to mess with delegations that have a
7557                          * non-zero dl_time. They might have already been broken
7558                          * and could be processed by the laundromat outside of
7559                          * the state_lock. Just leave them be.
7560                          */
7561                         if (dp->dl_time != 0)
7562                                 continue;
7563
7564                         atomic_inc(&clp->cl_rpc_users);
7565                         WARN_ON(!unhash_delegation_locked(dp));
7566                         list_add(&dp->dl_recall_lru, victims);
7567                 }
7568                 ++count;
7569                 /*
7570                  * Despite the fact that these functions deal with
7571                  * 64-bit integers for "count", we must ensure that
7572                  * it doesn't blow up the clp->cl_rpc_users. Throw a
7573                  * warning if we start to approach INT_MAX here.
7574                  */
7575                 WARN_ON_ONCE(count == (INT_MAX / 2));
7576                 if (count == max)
7577                         break;
7578         }
7579         spin_unlock(&state_lock);
7580         return count;
7581 }
7582
7583 static u64
7584 nfsd_print_client_delegations(struct nfs4_client *clp)
7585 {
7586         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
7587
7588         nfsd_print_count(clp, count, "delegations");
7589         return count;
7590 }
7591
7592 u64
7593 nfsd_inject_print_delegations(void)
7594 {
7595         struct nfs4_client *clp;
7596         u64 count = 0;
7597         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7598                                                 nfsd_net_id);
7599
7600         if (!nfsd_netns_ready(nn))
7601                 return 0;
7602
7603         spin_lock(&nn->client_lock);
7604         list_for_each_entry(clp, &nn->client_lru, cl_lru)
7605                 count += nfsd_print_client_delegations(clp);
7606         spin_unlock(&nn->client_lock);
7607
7608         return count;
7609 }
7610
7611 static void
7612 nfsd_forget_delegations(struct list_head *reaplist)
7613 {
7614         struct nfs4_client *clp;
7615         struct nfs4_delegation *dp, *next;
7616
7617         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7618                 list_del_init(&dp->dl_recall_lru);
7619                 clp = dp->dl_stid.sc_client;
7620                 revoke_delegation(dp);
7621                 put_client(clp);
7622         }
7623 }
7624
7625 u64
7626 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
7627                                       size_t addr_size)
7628 {
7629         u64 count = 0;
7630         struct nfs4_client *clp;
7631         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7632                                                 nfsd_net_id);
7633         LIST_HEAD(reaplist);
7634
7635         if (!nfsd_netns_ready(nn))
7636                 return count;
7637
7638         spin_lock(&nn->client_lock);
7639         clp = nfsd_find_client(addr, addr_size);
7640         if (clp)
7641                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7642         spin_unlock(&nn->client_lock);
7643
7644         nfsd_forget_delegations(&reaplist);
7645         return count;
7646 }
7647
7648 u64
7649 nfsd_inject_forget_delegations(u64 max)
7650 {
7651         u64 count = 0;
7652         struct nfs4_client *clp;
7653         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7654                                                 nfsd_net_id);
7655         LIST_HEAD(reaplist);
7656
7657         if (!nfsd_netns_ready(nn))
7658                 return count;
7659
7660         spin_lock(&nn->client_lock);
7661         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7662                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7663                 if (max != 0 && count >= max)
7664                         break;
7665         }
7666         spin_unlock(&nn->client_lock);
7667         nfsd_forget_delegations(&reaplist);
7668         return count;
7669 }
7670
7671 static void
7672 nfsd_recall_delegations(struct list_head *reaplist)
7673 {
7674         struct nfs4_client *clp;
7675         struct nfs4_delegation *dp, *next;
7676
7677         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7678                 list_del_init(&dp->dl_recall_lru);
7679                 clp = dp->dl_stid.sc_client;
7680                 /*
7681                  * We skipped all entries that had a zero dl_time before,
7682                  * so we can now reset the dl_time back to 0. If a delegation
7683                  * break comes in now, then it won't make any difference since
7684                  * we're recalling it either way.
7685                  */
7686                 spin_lock(&state_lock);
7687                 dp->dl_time = 0;
7688                 spin_unlock(&state_lock);
7689                 nfsd_break_one_deleg(dp);
7690                 put_client(clp);
7691         }
7692 }
7693
7694 u64
7695 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7696                                       size_t addr_size)
7697 {
7698         u64 count = 0;
7699         struct nfs4_client *clp;
7700         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7701                                                 nfsd_net_id);
7702         LIST_HEAD(reaplist);
7703
7704         if (!nfsd_netns_ready(nn))
7705                 return count;
7706
7707         spin_lock(&nn->client_lock);
7708         clp = nfsd_find_client(addr, addr_size);
7709         if (clp)
7710                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7711         spin_unlock(&nn->client_lock);
7712
7713         nfsd_recall_delegations(&reaplist);
7714         return count;
7715 }
7716
7717 u64
7718 nfsd_inject_recall_delegations(u64 max)
7719 {
7720         u64 count = 0;
7721         struct nfs4_client *clp, *next;
7722         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7723                                                 nfsd_net_id);
7724         LIST_HEAD(reaplist);
7725
7726         if (!nfsd_netns_ready(nn))
7727                 return count;
7728
7729         spin_lock(&nn->client_lock);
7730         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7731                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7732                 if (max != 0 && ++count >= max)
7733                         break;
7734         }
7735         spin_unlock(&nn->client_lock);
7736         nfsd_recall_delegations(&reaplist);
7737         return count;
7738 }
7739 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7740
7741 /*
7742  * Since the lifetime of a delegation isn't limited to that of an open, a
7743  * client may quite reasonably hang on to a delegation as long as it has
7744  * the inode cached.  This becomes an obvious problem the first time a
7745  * client's inode cache approaches the size of the server's total memory.
7746  *
7747  * For now we avoid this problem by imposing a hard limit on the number
7748  * of delegations, which varies according to the server's memory size.
7749  */
7750 static void
7751 set_max_delegations(void)
7752 {
7753         /*
7754          * Allow at most 4 delegations per megabyte of RAM.  Quick
7755          * estimates suggest that in the worst case (where every delegation
7756          * is for a different inode), a delegation could take about 1.5K,
7757          * giving a worst case usage of about 6% of memory.
7758          */
7759         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7760 }
7761
7762 static int nfs4_state_create_net(struct net *net)
7763 {
7764         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7765         int i;
7766
7767         nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7768                                             sizeof(struct list_head),
7769                                             GFP_KERNEL);
7770         if (!nn->conf_id_hashtbl)
7771                 goto err;
7772         nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7773                                               sizeof(struct list_head),
7774                                               GFP_KERNEL);
7775         if (!nn->unconf_id_hashtbl)
7776                 goto err_unconf_id;
7777         nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7778                                               sizeof(struct list_head),
7779                                               GFP_KERNEL);
7780         if (!nn->sessionid_hashtbl)
7781                 goto err_sessionid;
7782
7783         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7784                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7785                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7786         }
7787         for (i = 0; i < SESSION_HASH_SIZE; i++)
7788                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7789         nn->conf_name_tree = RB_ROOT;
7790         nn->unconf_name_tree = RB_ROOT;
7791         nn->boot_time = ktime_get_real_seconds();
7792         nn->grace_ended = false;
7793         nn->nfsd4_manager.block_opens = true;
7794         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7795         INIT_LIST_HEAD(&nn->client_lru);
7796         INIT_LIST_HEAD(&nn->close_lru);
7797         INIT_LIST_HEAD(&nn->del_recall_lru);
7798         spin_lock_init(&nn->client_lock);
7799         spin_lock_init(&nn->s2s_cp_lock);
7800         idr_init(&nn->s2s_cp_stateids);
7801
7802         spin_lock_init(&nn->blocked_locks_lock);
7803         INIT_LIST_HEAD(&nn->blocked_locks_lru);
7804
7805         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7806         get_net(net);
7807
7808         return 0;
7809
7810 err_sessionid:
7811         kfree(nn->unconf_id_hashtbl);
7812 err_unconf_id:
7813         kfree(nn->conf_id_hashtbl);
7814 err:
7815         return -ENOMEM;
7816 }
7817
7818 static void
7819 nfs4_state_destroy_net(struct net *net)
7820 {
7821         int i;
7822         struct nfs4_client *clp = NULL;
7823         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7824
7825         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7826                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7827                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7828                         destroy_client(clp);
7829                 }
7830         }
7831
7832         WARN_ON(!list_empty(&nn->blocked_locks_lru));
7833
7834         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7835                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7836                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7837                         destroy_client(clp);
7838                 }
7839         }
7840
7841         kfree(nn->sessionid_hashtbl);
7842         kfree(nn->unconf_id_hashtbl);
7843         kfree(nn->conf_id_hashtbl);
7844         put_net(net);
7845 }
7846
7847 int
7848 nfs4_state_start_net(struct net *net)
7849 {
7850         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7851         int ret;
7852
7853         ret = nfs4_state_create_net(net);
7854         if (ret)
7855                 return ret;
7856         locks_start_grace(net, &nn->nfsd4_manager);
7857         nfsd4_client_tracking_init(net);
7858         if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
7859                 goto skip_grace;
7860         printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
7861                nn->nfsd4_grace, net->ns.inum);
7862         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7863         return 0;
7864
7865 skip_grace:
7866         printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
7867                         net->ns.inum);
7868         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
7869         nfsd4_end_grace(nn);
7870         return 0;
7871 }
7872
7873 /* initialization to perform when the nfsd service is started: */
7874
7875 int
7876 nfs4_state_start(void)
7877 {
7878         int ret;
7879
7880         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7881         if (laundry_wq == NULL) {
7882                 ret = -ENOMEM;
7883                 goto out;
7884         }
7885         ret = nfsd4_create_callback_queue();
7886         if (ret)
7887                 goto out_free_laundry;
7888
7889         set_max_delegations();
7890         return 0;
7891
7892 out_free_laundry:
7893         destroy_workqueue(laundry_wq);
7894 out:
7895         return ret;
7896 }
7897
7898 void
7899 nfs4_state_shutdown_net(struct net *net)
7900 {
7901         struct nfs4_delegation *dp = NULL;
7902         struct list_head *pos, *next, reaplist;
7903         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7904
7905         cancel_delayed_work_sync(&nn->laundromat_work);
7906         locks_end_grace(&nn->nfsd4_manager);
7907
7908         INIT_LIST_HEAD(&reaplist);
7909         spin_lock(&state_lock);
7910         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7911                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7912                 WARN_ON(!unhash_delegation_locked(dp));
7913                 list_add(&dp->dl_recall_lru, &reaplist);
7914         }
7915         spin_unlock(&state_lock);
7916         list_for_each_safe(pos, next, &reaplist) {
7917                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7918                 list_del_init(&dp->dl_recall_lru);
7919                 destroy_unhashed_deleg(dp);
7920         }
7921
7922         nfsd4_client_tracking_exit(net);
7923         nfs4_state_destroy_net(net);
7924 }
7925
7926 void
7927 nfs4_state_shutdown(void)
7928 {
7929         destroy_workqueue(laundry_wq);
7930         nfsd4_destroy_callback_queue();
7931 }
7932
7933 static void
7934 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7935 {
7936         if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
7937             CURRENT_STATEID(stateid))
7938                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7939 }
7940
7941 static void
7942 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7943 {
7944         if (cstate->minorversion) {
7945                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7946                 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
7947         }
7948 }
7949
7950 void
7951 clear_current_stateid(struct nfsd4_compound_state *cstate)
7952 {
7953         CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
7954 }
7955
7956 /*
7957  * functions to set current state id
7958  */
7959 void
7960 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7961                 union nfsd4_op_u *u)
7962 {
7963         put_stateid(cstate, &u->open_downgrade.od_stateid);
7964 }
7965
7966 void
7967 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7968                 union nfsd4_op_u *u)
7969 {
7970         put_stateid(cstate, &u->open.op_stateid);
7971 }
7972
7973 void
7974 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7975                 union nfsd4_op_u *u)
7976 {
7977         put_stateid(cstate, &u->close.cl_stateid);
7978 }
7979
7980 void
7981 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7982                 union nfsd4_op_u *u)
7983 {
7984         put_stateid(cstate, &u->lock.lk_resp_stateid);
7985 }
7986
7987 /*
7988  * functions to consume current state id
7989  */
7990
7991 void
7992 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7993                 union nfsd4_op_u *u)
7994 {
7995         get_stateid(cstate, &u->open_downgrade.od_stateid);
7996 }
7997
7998 void
7999 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
8000                 union nfsd4_op_u *u)
8001 {
8002         get_stateid(cstate, &u->delegreturn.dr_stateid);
8003 }
8004
8005 void
8006 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
8007                 union nfsd4_op_u *u)
8008 {
8009         get_stateid(cstate, &u->free_stateid.fr_stateid);
8010 }
8011
8012 void
8013 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
8014                 union nfsd4_op_u *u)
8015 {
8016         get_stateid(cstate, &u->setattr.sa_stateid);
8017 }
8018
8019 void
8020 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
8021                 union nfsd4_op_u *u)
8022 {
8023         get_stateid(cstate, &u->close.cl_stateid);
8024 }
8025
8026 void
8027 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
8028                 union nfsd4_op_u *u)
8029 {
8030         get_stateid(cstate, &u->locku.lu_stateid);
8031 }
8032
8033 void
8034 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
8035                 union nfsd4_op_u *u)
8036 {
8037         get_stateid(cstate, &u->read.rd_stateid);
8038 }
8039
8040 void
8041 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
8042                 union nfsd4_op_u *u)
8043 {
8044         get_stateid(cstate, &u->write.wr_stateid);
8045 }