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 <linux/fsnotify.h>
  47 #include <linux/rhashtable.h>
  48 #include <linux/nfs_ssc.h>
  49
  50 #include "xdr4.h"
  51 #include "xdr4cb.h"
  52 #include "vfs.h"
  53 #include "current_stateid.h"
  54
  55 #include "netns.h"
  56 #include "pnfs.h"
  57 #include "filecache.h"
  58 #include "trace.h"
  59
  60 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  61
  62 #define all_ones {{ ~0, ~0}, ~0}
  63 static const stateid_t one_stateid = {
  64         .si_generation = ~0,
  65         .si_opaque = all_ones,
  66 };
  67 static const stateid_t zero_stateid = {
  68         /* all fields zero */
  69 };
  70 static const stateid_t currentstateid = {
  71         .si_generation = 1,
  72 };
  73 static const stateid_t close_stateid = {
  74         .si_generation = 0xffffffffU,
  75 };
  76
  77 static u64 current_sessionid = 1;
  78
  79 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  80 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  81 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  82 #define CLOSE_STATEID(stateid)  (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
  83
  84 /* forward declarations */
  85 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  86 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  87 void nfsd4_end_grace(struct nfsd_net *nn);
  88 static void _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps);
  89 static void nfsd4_file_hash_remove(struct nfs4_file *fi);
  90 static void deleg_reaper(struct nfsd_net *nn);
  91
  92 /* Locking: */
  93
  94 /*
  95  * Currently used for the del_recall_lru and file hash table.  In an
  96  * effort to decrease the scope of the client_mutex, this spinlock may
  97  * eventually cover more:
  98  */
  99 static DEFINE_SPINLOCK(state_lock);
 100
 101 enum nfsd4_st_mutex_lock_subclass {
 102         OPEN_STATEID_MUTEX = 0,
 103         LOCK_STATEID_MUTEX = 1,
 104 };
 105
 106 /*
 107  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
 108  * the refcount on the open stateid to drop.
 109  */
 110 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
 111
 112 /*
 113  * A waitqueue where a writer to clients/#/ctl destroying a client can
 114  * wait for cl_rpc_users to drop to 0 and then for the client to be
 115  * unhashed.
 116  */
 117 static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
 118
 119 static struct kmem_cache *client_slab;
 120 static struct kmem_cache *openowner_slab;
 121 static struct kmem_cache *lockowner_slab;
 122 static struct kmem_cache *file_slab;
 123 static struct kmem_cache *stateid_slab;
 124 static struct kmem_cache *deleg_slab;
 125 static struct kmem_cache *odstate_slab;
 126
 127 static void free_session(struct nfsd4_session *);
 128
 129 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 130 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
 131 static const struct nfsd4_callback_ops nfsd4_cb_getattr_ops;
 132
 133 static struct workqueue_struct *laundry_wq;
 134
 135 int nfsd4_create_laundry_wq(void)
 136 {
 137         int rc = 0;
 138
 139         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
 140         if (laundry_wq == NULL)
 141                 rc = -ENOMEM;
 142         return rc;
 143 }
 144
 145 void nfsd4_destroy_laundry_wq(void)
 146 {
 147         destroy_workqueue(laundry_wq);
 148 }
 149
 150 static bool is_session_dead(struct nfsd4_session *ses)
 151 {
 152         return ses->se_flags & NFS4_SESSION_DEAD;
 153 }
 154
 155 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 156 {
 157         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 158                 return nfserr_jukebox;
 159         ses->se_flags |= NFS4_SESSION_DEAD;
 160         return nfs_ok;
 161 }
 162
 163 static bool is_client_expired(struct nfs4_client *clp)
 164 {
 165         return clp->cl_time == 0;
 166 }
 167
 168 static void nfsd4_dec_courtesy_client_count(struct nfsd_net *nn,
 169                                         struct nfs4_client *clp)
 170 {
 171         if (clp->cl_state != NFSD4_ACTIVE)
 172                 atomic_add_unless(&nn->nfsd_courtesy_clients, -1, 0);
 173 }
 174
 175 static __be32 get_client_locked(struct nfs4_client *clp)
 176 {
 177         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 178
 179         lockdep_assert_held(&nn->client_lock);
 180
 181         if (is_client_expired(clp))
 182                 return nfserr_expired;
 183         atomic_inc(&clp->cl_rpc_users);
 184         nfsd4_dec_courtesy_client_count(nn, clp);
 185         clp->cl_state = NFSD4_ACTIVE;
 186         return nfs_ok;
 187 }
 188
 189 /* must be called under the client_lock */
 190 static inline void
 191 renew_client_locked(struct nfs4_client *clp)
 192 {
 193         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 194
 195         if (is_client_expired(clp)) {
 196                 WARN_ON(1);
 197                 printk("%s: client (clientid %08x/%08x) already expired\n",
 198                         __func__,
 199                         clp->cl_clientid.cl_boot,
 200                         clp->cl_clientid.cl_id);
 201                 return;
 202         }
 203
 204         list_move_tail(&clp->cl_lru, &nn->client_lru);
 205         clp->cl_time = ktime_get_boottime_seconds();
 206         nfsd4_dec_courtesy_client_count(nn, clp);
 207         clp->cl_state = NFSD4_ACTIVE;
 208 }
 209
 210 static void put_client_renew_locked(struct nfs4_client *clp)
 211 {
 212         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 213
 214         lockdep_assert_held(&nn->client_lock);
 215
 216         if (!atomic_dec_and_test(&clp->cl_rpc_users))
 217                 return;
 218         if (!is_client_expired(clp))
 219                 renew_client_locked(clp);
 220         else
 221                 wake_up_all(&expiry_wq);
 222 }
 223
 224 static void put_client_renew(struct nfs4_client *clp)
 225 {
 226         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 227
 228         if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
 229                 return;
 230         if (!is_client_expired(clp))
 231                 renew_client_locked(clp);
 232         else
 233                 wake_up_all(&expiry_wq);
 234         spin_unlock(&nn->client_lock);
 235 }
 236
 237 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 238 {
 239         __be32 status;
 240
 241         if (is_session_dead(ses))
 242                 return nfserr_badsession;
 243         status = get_client_locked(ses->se_client);
 244         if (status)
 245                 return status;
 246         atomic_inc(&ses->se_ref);
 247         return nfs_ok;
 248 }
 249
 250 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 251 {
 252         struct nfs4_client *clp = ses->se_client;
 253         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 254
 255         lockdep_assert_held(&nn->client_lock);
 256
 257         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 258                 free_session(ses);
 259         put_client_renew_locked(clp);
 260 }
 261
 262 static void nfsd4_put_session(struct nfsd4_session *ses)
 263 {
 264         struct nfs4_client *clp = ses->se_client;
 265         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 266
 267         spin_lock(&nn->client_lock);
 268         nfsd4_put_session_locked(ses);
 269         spin_unlock(&nn->client_lock);
 270 }
 271
 272 static struct nfsd4_blocked_lock *
 273 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 274                         struct nfsd_net *nn)
 275 {
 276         struct nfsd4_blocked_lock *cur, *found = NULL;
 277
 278         spin_lock(&nn->blocked_locks_lock);
 279         list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
 280                 if (fh_match(fh, &cur->nbl_fh)) {
 281                         list_del_init(&cur->nbl_list);
 282                         WARN_ON(list_empty(&cur->nbl_lru));
 283                         list_del_init(&cur->nbl_lru);
 284                         found = cur;
 285                         break;
 286                 }
 287         }
 288         spin_unlock(&nn->blocked_locks_lock);
 289         if (found)
 290                 locks_delete_block(&found->nbl_lock);
 291         return found;
 292 }
 293
 294 static struct nfsd4_blocked_lock *
 295 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 296                         struct nfsd_net *nn)
 297 {
 298         struct nfsd4_blocked_lock *nbl;
 299
 300         nbl = find_blocked_lock(lo, fh, nn);
 301         if (!nbl) {
 302                 nbl = kmalloc(sizeof(*nbl), GFP_KERNEL);
 303                 if (nbl) {
 304                         INIT_LIST_HEAD(&nbl->nbl_list);
 305                         INIT_LIST_HEAD(&nbl->nbl_lru);
 306                         fh_copy_shallow(&nbl->nbl_fh, fh);
 307                         locks_init_lock(&nbl->nbl_lock);
 308                         kref_init(&nbl->nbl_kref);
 309                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
 310                                         &nfsd4_cb_notify_lock_ops,
 311                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
 312                 }
 313         }
 314         return nbl;
 315 }
 316
 317 static void
 318 free_nbl(struct kref *kref)
 319 {
 320         struct nfsd4_blocked_lock *nbl;
 321
 322         nbl = container_of(kref, struct nfsd4_blocked_lock, nbl_kref);
 323         locks_release_private(&nbl->nbl_lock);
 324         kfree(nbl);
 325 }
 326
 327 static void
 328 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
 329 {
 330         locks_delete_block(&nbl->nbl_lock);
 331         kref_put(&nbl->nbl_kref, free_nbl);
 332 }
 333
 334 static void
 335 remove_blocked_locks(struct nfs4_lockowner *lo)
 336 {
 337         struct nfs4_client *clp = lo->lo_owner.so_client;
 338         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 339         struct nfsd4_blocked_lock *nbl;
 340         LIST_HEAD(reaplist);
 341
 342         /* Dequeue all blocked locks */
 343         spin_lock(&nn->blocked_locks_lock);
 344         while (!list_empty(&lo->lo_blocked)) {
 345                 nbl = list_first_entry(&lo->lo_blocked,
 346                                         struct nfsd4_blocked_lock,
 347                                         nbl_list);
 348                 list_del_init(&nbl->nbl_list);
 349                 WARN_ON(list_empty(&nbl->nbl_lru));
 350                 list_move(&nbl->nbl_lru, &reaplist);
 351         }
 352         spin_unlock(&nn->blocked_locks_lock);
 353
 354         /* Now free them */
 355         while (!list_empty(&reaplist)) {
 356                 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
 357                                         nbl_lru);
 358                 list_del_init(&nbl->nbl_lru);
 359                 free_blocked_lock(nbl);
 360         }
 361 }
 362
 363 static void
 364 nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
 365 {
 366         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 367                                                 struct nfsd4_blocked_lock, nbl_cb);
 368         locks_delete_block(&nbl->nbl_lock);
 369 }
 370
 371 static int
 372 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
 373 {
 374         trace_nfsd_cb_notify_lock_done(&zero_stateid, task);
 375
 376         /*
 377          * Since this is just an optimization, we don't try very hard if it
 378          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
 379          * just quit trying on anything else.
 380          */
 381         switch (task->tk_status) {
 382         case -NFS4ERR_DELAY:
 383                 rpc_delay(task, 1 * HZ);
 384                 return 0;
 385         default:
 386                 return 1;
 387         }
 388 }
 389
 390 static void
 391 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
 392 {
 393         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 394                                                 struct nfsd4_blocked_lock, nbl_cb);
 395
 396         free_blocked_lock(nbl);
 397 }
 398
 399 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
 400         .prepare        = nfsd4_cb_notify_lock_prepare,
 401         .done           = nfsd4_cb_notify_lock_done,
 402         .release        = nfsd4_cb_notify_lock_release,
 403 };
 404
 405 /*
 406  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 407  * st_{access,deny}_bmap field of the stateid, in order to track not
 408  * only what share bits are currently in force, but also what
 409  * combinations of share bits previous opens have used.  This allows us
 410  * to enforce the recommendation in
 411  * https://datatracker.ietf.org/doc/html/rfc7530#section-16.19.4 that
 412  * the server return an error if the client attempt to downgrade to a
 413  * combination of share bits not explicable by closing some of its
 414  * previous opens.
 415  *
 416  * This enforcement is arguably incomplete, since we don't keep
 417  * track of access/deny bit combinations; so, e.g., we allow:
 418  *
 419  *      OPEN allow read, deny write
 420  *      OPEN allow both, deny none
 421  *      DOWNGRADE allow read, deny none
 422  *
 423  * which we should reject.
 424  *
 425  * But you could also argue that our current code is already overkill,
 426  * since it only exists to return NFS4ERR_INVAL on incorrect client
 427  * behavior.
 428  */
 429 static unsigned int
 430 bmap_to_share_mode(unsigned long bmap)
 431 {
 432         int i;
 433         unsigned int access = 0;
 434
 435         for (i = 1; i < 4; i++) {
 436                 if (test_bit(i, &bmap))
 437                         access |= i;
 438         }
 439         return access;
 440 }
 441
 442 /* set share access for a given stateid */
 443 static inline void
 444 set_access(u32 access, struct nfs4_ol_stateid *stp)
 445 {
 446         unsigned char mask = 1 << access;
 447
 448         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 449         stp->st_access_bmap |= mask;
 450 }
 451
 452 /* clear share access for a given stateid */
 453 static inline void
 454 clear_access(u32 access, struct nfs4_ol_stateid *stp)
 455 {
 456         unsigned char mask = 1 << access;
 457
 458         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 459         stp->st_access_bmap &= ~mask;
 460 }
 461
 462 /* test whether a given stateid has access */
 463 static inline bool
 464 test_access(u32 access, struct nfs4_ol_stateid *stp)
 465 {
 466         unsigned char mask = 1 << access;
 467
 468         return (bool)(stp->st_access_bmap & mask);
 469 }
 470
 471 /* set share deny for a given stateid */
 472 static inline void
 473 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
 474 {
 475         unsigned char mask = 1 << deny;
 476
 477         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 478         stp->st_deny_bmap |= mask;
 479 }
 480
 481 /* clear share deny for a given stateid */
 482 static inline void
 483 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
 484 {
 485         unsigned char mask = 1 << deny;
 486
 487         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 488         stp->st_deny_bmap &= ~mask;
 489 }
 490
 491 /* test whether a given stateid is denying specific access */
 492 static inline bool
 493 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
 494 {
 495         unsigned char mask = 1 << deny;
 496
 497         return (bool)(stp->st_deny_bmap & mask);
 498 }
 499
 500 static int nfs4_access_to_omode(u32 access)
 501 {
 502         switch (access & NFS4_SHARE_ACCESS_BOTH) {
 503         case NFS4_SHARE_ACCESS_READ:
 504                 return O_RDONLY;
 505         case NFS4_SHARE_ACCESS_WRITE:
 506                 return O_WRONLY;
 507         case NFS4_SHARE_ACCESS_BOTH:
 508                 return O_RDWR;
 509         }
 510         WARN_ON_ONCE(1);
 511         return O_RDONLY;
 512 }
 513
 514 static inline int
 515 access_permit_read(struct nfs4_ol_stateid *stp)
 516 {
 517         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
 518                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
 519                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
 520 }
 521
 522 static inline int
 523 access_permit_write(struct nfs4_ol_stateid *stp)
 524 {
 525         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
 526                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
 527 }
 528
 529 static inline struct nfs4_stateowner *
 530 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 531 {
 532         atomic_inc(&sop->so_count);
 533         return sop;
 534 }
 535
 536 static int
 537 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 538 {
 539         return (sop->so_owner.len == owner->len) &&
 540                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 541 }
 542
 543 static struct nfs4_openowner *
 544 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 545                         struct nfs4_client *clp)
 546 {
 547         struct nfs4_stateowner *so;
 548
 549         lockdep_assert_held(&clp->cl_lock);
 550
 551         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 552                             so_strhash) {
 553                 if (!so->so_is_open_owner)
 554                         continue;
 555                 if (same_owner_str(so, &open->op_owner))
 556                         return openowner(nfs4_get_stateowner(so));
 557         }
 558         return NULL;
 559 }
 560
 561 static inline u32
 562 opaque_hashval(const void *ptr, int nbytes)
 563 {
 564         unsigned char *cptr = (unsigned char *) ptr;
 565
 566         u32 x = 0;
 567         while (nbytes--) {
 568                 x *= 37;
 569                 x += *cptr++;
 570         }
 571         return x;
 572 }
 573
 574 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 575 {
 576         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 577
 578         kmem_cache_free(file_slab, fp);
 579 }
 580
 581 void
 582 put_nfs4_file(struct nfs4_file *fi)
 583 {
 584         if (refcount_dec_and_test(&fi->fi_ref)) {
 585                 nfsd4_file_hash_remove(fi);
 586                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 587                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 588                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 589         }
 590 }
 591
 592 static struct nfsd_file *
 593 find_writeable_file_locked(struct nfs4_file *f)
 594 {
 595         struct nfsd_file *ret;
 596
 597         lockdep_assert_held(&f->fi_lock);
 598
 599         ret = nfsd_file_get(f->fi_fds[O_WRONLY]);
 600         if (!ret)
 601                 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
 602         return ret;
 603 }
 604
 605 static struct nfsd_file *
 606 find_writeable_file(struct nfs4_file *f)
 607 {
 608         struct nfsd_file *ret;
 609
 610         spin_lock(&f->fi_lock);
 611         ret = find_writeable_file_locked(f);
 612         spin_unlock(&f->fi_lock);
 613
 614         return ret;
 615 }
 616
 617 static struct nfsd_file *
 618 find_readable_file_locked(struct nfs4_file *f)
 619 {
 620         struct nfsd_file *ret;
 621
 622         lockdep_assert_held(&f->fi_lock);
 623
 624         ret = nfsd_file_get(f->fi_fds[O_RDONLY]);
 625         if (!ret)
 626                 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
 627         return ret;
 628 }
 629
 630 static struct nfsd_file *
 631 find_readable_file(struct nfs4_file *f)
 632 {
 633         struct nfsd_file *ret;
 634
 635         spin_lock(&f->fi_lock);
 636         ret = find_readable_file_locked(f);
 637         spin_unlock(&f->fi_lock);
 638
 639         return ret;
 640 }
 641
 642 static struct nfsd_file *
 643 find_rw_file(struct nfs4_file *f)
 644 {
 645         struct nfsd_file *ret;
 646
 647         spin_lock(&f->fi_lock);
 648         ret = nfsd_file_get(f->fi_fds[O_RDWR]);
 649         spin_unlock(&f->fi_lock);
 650
 651         return ret;
 652 }
 653
 654 struct nfsd_file *
 655 find_any_file(struct nfs4_file *f)
 656 {
 657         struct nfsd_file *ret;
 658
 659         if (!f)
 660                 return NULL;
 661         spin_lock(&f->fi_lock);
 662         ret = nfsd_file_get(f->fi_fds[O_RDWR]);
 663         if (!ret) {
 664                 ret = nfsd_file_get(f->fi_fds[O_WRONLY]);
 665                 if (!ret)
 666                         ret = nfsd_file_get(f->fi_fds[O_RDONLY]);
 667         }
 668         spin_unlock(&f->fi_lock);
 669         return ret;
 670 }
 671
 672 static struct nfsd_file *find_any_file_locked(struct nfs4_file *f)
 673 {
 674         lockdep_assert_held(&f->fi_lock);
 675
 676         if (f->fi_fds[O_RDWR])
 677                 return f->fi_fds[O_RDWR];
 678         if (f->fi_fds[O_WRONLY])
 679                 return f->fi_fds[O_WRONLY];
 680         if (f->fi_fds[O_RDONLY])
 681                 return f->fi_fds[O_RDONLY];
 682         return NULL;
 683 }
 684
 685 static atomic_long_t num_delegations;
 686 unsigned long max_delegations;
 687
 688 /*
 689  * Open owner state (share locks)
 690  */
 691
 692 /* hash tables for lock and open owners */
 693 #define OWNER_HASH_BITS              8
 694 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 695 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 696
 697 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 698 {
 699         unsigned int ret;
 700
 701         ret = opaque_hashval(ownername->data, ownername->len);
 702         return ret & OWNER_HASH_MASK;
 703 }
 704
 705 static struct rhltable nfs4_file_rhltable ____cacheline_aligned_in_smp;
 706
 707 static const struct rhashtable_params nfs4_file_rhash_params = {
 708         .key_len                = sizeof_field(struct nfs4_file, fi_inode),
 709         .key_offset             = offsetof(struct nfs4_file, fi_inode),
 710         .head_offset            = offsetof(struct nfs4_file, fi_rlist),
 711
 712         /*
 713          * Start with a single page hash table to reduce resizing churn
 714          * on light workloads.
 715          */
 716         .min_size               = 256,
 717         .automatic_shrinking    = true,
 718 };
 719
 720 /*
 721  * Check if courtesy clients have conflicting access and resolve it if possible
 722  *
 723  * access:  is op_share_access if share_access is true.
 724  *          Check if access mode, op_share_access, would conflict with
 725  *          the current deny mode of the file 'fp'.
 726  * access:  is op_share_deny if share_access is false.
 727  *          Check if the deny mode, op_share_deny, would conflict with
 728  *          current access of the file 'fp'.
 729  * stp:     skip checking this entry.
 730  * new_stp: normal open, not open upgrade.
 731  *
 732  * Function returns:
 733  *      false - access/deny mode conflict with normal client.
 734  *      true  - no conflict or conflict with courtesy client(s) is resolved.
 735  */
 736 static bool
 737 nfs4_resolve_deny_conflicts_locked(struct nfs4_file *fp, bool new_stp,
 738                 struct nfs4_ol_stateid *stp, u32 access, bool share_access)
 739 {
 740         struct nfs4_ol_stateid *st;
 741         bool resolvable = true;
 742         unsigned char bmap;
 743         struct nfsd_net *nn;
 744         struct nfs4_client *clp;
 745
 746         lockdep_assert_held(&fp->fi_lock);
 747         list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
 748                 /* ignore lock stateid */
 749                 if (st->st_openstp)
 750                         continue;
 751                 if (st == stp && new_stp)
 752                         continue;
 753                 /* check file access against deny mode or vice versa */
 754                 bmap = share_access ? st->st_deny_bmap : st->st_access_bmap;
 755                 if (!(access & bmap_to_share_mode(bmap)))
 756                         continue;
 757                 clp = st->st_stid.sc_client;
 758                 if (try_to_expire_client(clp))
 759                         continue;
 760                 resolvable = false;
 761                 break;
 762         }
 763         if (resolvable) {
 764                 clp = stp->st_stid.sc_client;
 765                 nn = net_generic(clp->net, nfsd_net_id);
 766                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
 767         }
 768         return resolvable;
 769 }
 770
 771 static void
 772 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 773 {
 774         lockdep_assert_held(&fp->fi_lock);
 775
 776         if (access & NFS4_SHARE_ACCESS_WRITE)
 777                 atomic_inc(&fp->fi_access[O_WRONLY]);
 778         if (access & NFS4_SHARE_ACCESS_READ)
 779                 atomic_inc(&fp->fi_access[O_RDONLY]);
 780 }
 781
 782 static __be32
 783 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 784 {
 785         lockdep_assert_held(&fp->fi_lock);
 786
 787         /* Does this access mode make sense? */
 788         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 789                 return nfserr_inval;
 790
 791         /* Does it conflict with a deny mode already set? */
 792         if ((access & fp->fi_share_deny) != 0)
 793                 return nfserr_share_denied;
 794
 795         __nfs4_file_get_access(fp, access);
 796         return nfs_ok;
 797 }
 798
 799 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 800 {
 801         /* Common case is that there is no deny mode. */
 802         if (deny) {
 803                 /* Does this deny mode make sense? */
 804                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 805                         return nfserr_inval;
 806
 807                 if ((deny & NFS4_SHARE_DENY_READ) &&
 808                     atomic_read(&fp->fi_access[O_RDONLY]))
 809                         return nfserr_share_denied;
 810
 811                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 812                     atomic_read(&fp->fi_access[O_WRONLY]))
 813                         return nfserr_share_denied;
 814         }
 815         return nfs_ok;
 816 }
 817
 818 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 819 {
 820         might_lock(&fp->fi_lock);
 821
 822         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 823                 struct nfsd_file *f1 = NULL;
 824                 struct nfsd_file *f2 = NULL;
 825
 826                 swap(f1, fp->fi_fds[oflag]);
 827                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 828                         swap(f2, fp->fi_fds[O_RDWR]);
 829                 spin_unlock(&fp->fi_lock);
 830                 if (f1)
 831                         nfsd_file_put(f1);
 832                 if (f2)
 833                         nfsd_file_put(f2);
 834         }
 835 }
 836
 837 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 838 {
 839         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 840
 841         if (access & NFS4_SHARE_ACCESS_WRITE)
 842                 __nfs4_file_put_access(fp, O_WRONLY);
 843         if (access & NFS4_SHARE_ACCESS_READ)
 844                 __nfs4_file_put_access(fp, O_RDONLY);
 845 }
 846
 847 /*
 848  * Allocate a new open/delegation state counter. This is needed for
 849  * pNFS for proper return on close semantics.
 850  *
 851  * Note that we only allocate it for pNFS-enabled exports, otherwise
 852  * all pointers to struct nfs4_clnt_odstate are always NULL.
 853  */
 854 static struct nfs4_clnt_odstate *
 855 alloc_clnt_odstate(struct nfs4_client *clp)
 856 {
 857         struct nfs4_clnt_odstate *co;
 858
 859         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 860         if (co) {
 861                 co->co_client = clp;
 862                 refcount_set(&co->co_odcount, 1);
 863         }
 864         return co;
 865 }
 866
 867 static void
 868 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 869 {
 870         struct nfs4_file *fp = co->co_file;
 871
 872         lockdep_assert_held(&fp->fi_lock);
 873         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 874 }
 875
 876 static inline void
 877 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 878 {
 879         if (co)
 880                 refcount_inc(&co->co_odcount);
 881 }
 882
 883 static void
 884 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 885 {
 886         struct nfs4_file *fp;
 887
 888         if (!co)
 889                 return;
 890
 891         fp = co->co_file;
 892         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 893                 list_del(&co->co_perfile);
 894                 spin_unlock(&fp->fi_lock);
 895
 896                 nfsd4_return_all_file_layouts(co->co_client, fp);
 897                 kmem_cache_free(odstate_slab, co);
 898         }
 899 }
 900
 901 static struct nfs4_clnt_odstate *
 902 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 903 {
 904         struct nfs4_clnt_odstate *co;
 905         struct nfs4_client *cl;
 906
 907         if (!new)
 908                 return NULL;
 909
 910         cl = new->co_client;
 911
 912         spin_lock(&fp->fi_lock);
 913         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 914                 if (co->co_client == cl) {
 915                         get_clnt_odstate(co);
 916                         goto out;
 917                 }
 918         }
 919         co = new;
 920         co->co_file = fp;
 921         hash_clnt_odstate_locked(new);
 922 out:
 923         spin_unlock(&fp->fi_lock);
 924         return co;
 925 }
 926
 927 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 928                                   void (*sc_free)(struct nfs4_stid *))
 929 {
 930         struct nfs4_stid *stid;
 931         int new_id;
 932
 933         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 934         if (!stid)
 935                 return NULL;
 936
 937         idr_preload(GFP_KERNEL);
 938         spin_lock(&cl->cl_lock);
 939         /* Reserving 0 for start of file in nfsdfs "states" file: */
 940         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
 941         spin_unlock(&cl->cl_lock);
 942         idr_preload_end();
 943         if (new_id < 0)
 944                 goto out_free;
 945
 946         stid->sc_free = sc_free;
 947         stid->sc_client = cl;
 948         stid->sc_stateid.si_opaque.so_id = new_id;
 949         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 950         /* Will be incremented before return to client: */
 951         refcount_set(&stid->sc_count, 1);
 952         spin_lock_init(&stid->sc_lock);
 953         INIT_LIST_HEAD(&stid->sc_cp_list);
 954
 955         /*
 956          * It shouldn't be a problem to reuse an opaque stateid value.
 957          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 958          * example, a stray write retransmission could be accepted by
 959          * the server when it should have been rejected.  Therefore,
 960          * adopt a trick from the sctp code to attempt to maximize the
 961          * amount of time until an id is reused, by ensuring they always
 962          * "increase" (mod INT_MAX):
 963          */
 964         return stid;
 965 out_free:
 966         kmem_cache_free(slab, stid);
 967         return NULL;
 968 }
 969
 970 /*
 971  * Create a unique stateid_t to represent each COPY.
 972  */
 973 static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
 974                               unsigned char cs_type)
 975 {
 976         int new_id;
 977
 978         stid->cs_stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
 979         stid->cs_stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
 980
 981         idr_preload(GFP_KERNEL);
 982         spin_lock(&nn->s2s_cp_lock);
 983         new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
 984         stid->cs_stid.si_opaque.so_id = new_id;
 985         stid->cs_stid.si_generation = 1;
 986         spin_unlock(&nn->s2s_cp_lock);
 987         idr_preload_end();
 988         if (new_id < 0)
 989                 return 0;
 990         stid->cs_type = cs_type;
 991         return 1;
 992 }
 993
 994 int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
 995 {
 996         return nfs4_init_cp_state(nn, &copy->cp_stateid, NFS4_COPY_STID);
 997 }
 998
 999 struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
1000                                                      struct nfs4_stid *p_stid)
1001 {
1002         struct nfs4_cpntf_state *cps;
1003
1004         cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
1005         if (!cps)
1006                 return NULL;
1007         cps->cpntf_time = ktime_get_boottime_seconds();
1008         refcount_set(&cps->cp_stateid.cs_count, 1);
1009         if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
1010                 goto out_free;
1011         spin_lock(&nn->s2s_cp_lock);
1012         list_add(&cps->cp_list, &p_stid->sc_cp_list);
1013         spin_unlock(&nn->s2s_cp_lock);
1014         return cps;
1015 out_free:
1016         kfree(cps);
1017         return NULL;
1018 }
1019
1020 void nfs4_free_copy_state(struct nfsd4_copy *copy)
1021 {
1022         struct nfsd_net *nn;
1023
1024         if (copy->cp_stateid.cs_type != NFS4_COPY_STID)
1025                 return;
1026         nn = net_generic(copy->cp_clp->net, nfsd_net_id);
1027         spin_lock(&nn->s2s_cp_lock);
1028         idr_remove(&nn->s2s_cp_stateids,
1029                    copy->cp_stateid.cs_stid.si_opaque.so_id);
1030         spin_unlock(&nn->s2s_cp_lock);
1031 }
1032
1033 static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
1034 {
1035         struct nfs4_cpntf_state *cps;
1036         struct nfsd_net *nn;
1037
1038         nn = net_generic(net, nfsd_net_id);
1039         spin_lock(&nn->s2s_cp_lock);
1040         while (!list_empty(&stid->sc_cp_list)) {
1041                 cps = list_first_entry(&stid->sc_cp_list,
1042                                        struct nfs4_cpntf_state, cp_list);
1043                 _free_cpntf_state_locked(nn, cps);
1044         }
1045         spin_unlock(&nn->s2s_cp_lock);
1046 }
1047
1048 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
1049 {
1050         struct nfs4_stid *stid;
1051
1052         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
1053         if (!stid)
1054                 return NULL;
1055
1056         return openlockstateid(stid);
1057 }
1058
1059 static void nfs4_free_deleg(struct nfs4_stid *stid)
1060 {
1061         struct nfs4_delegation *dp = delegstateid(stid);
1062
1063         WARN_ON_ONCE(!list_empty(&stid->sc_cp_list));
1064         WARN_ON_ONCE(!list_empty(&dp->dl_perfile));
1065         WARN_ON_ONCE(!list_empty(&dp->dl_perclnt));
1066         WARN_ON_ONCE(!list_empty(&dp->dl_recall_lru));
1067         kmem_cache_free(deleg_slab, stid);
1068         atomic_long_dec(&num_delegations);
1069 }
1070
1071 /*
1072  * When we recall a delegation, we should be careful not to hand it
1073  * out again straight away.
1074  * To ensure this we keep a pair of bloom filters ('new' and 'old')
1075  * in which the filehandles of recalled delegations are "stored".
1076  * If a filehandle appear in either filter, a delegation is blocked.
1077  * When a delegation is recalled, the filehandle is stored in the "new"
1078  * filter.
1079  * Every 30 seconds we swap the filters and clear the "new" one,
1080  * unless both are empty of course.
1081  *
1082  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
1083  * low 3 bytes as hash-table indices.
1084  *
1085  * 'blocked_delegations_lock', which is always taken in block_delegations(),
1086  * is used to manage concurrent access.  Testing does not need the lock
1087  * except when swapping the two filters.
1088  */
1089 static DEFINE_SPINLOCK(blocked_delegations_lock);
1090 static struct bloom_pair {
1091         int     entries, old_entries;
1092         time64_t swap_time;
1093         int     new; /* index into 'set' */
1094         DECLARE_BITMAP(set[2], 256);
1095 } blocked_delegations;
1096
1097 static int delegation_blocked(struct knfsd_fh *fh)
1098 {
1099         u32 hash;
1100         struct bloom_pair *bd = &blocked_delegations;
1101
1102         if (bd->entries == 0)
1103                 return 0;
1104         if (ktime_get_seconds() - bd->swap_time > 30) {
1105                 spin_lock(&blocked_delegations_lock);
1106                 if (ktime_get_seconds() - bd->swap_time > 30) {
1107                         bd->entries -= bd->old_entries;
1108                         bd->old_entries = bd->entries;
1109                         memset(bd->set[bd->new], 0,
1110                                sizeof(bd->set[0]));
1111                         bd->new = 1-bd->new;
1112                         bd->swap_time = ktime_get_seconds();
1113                 }
1114                 spin_unlock(&blocked_delegations_lock);
1115         }
1116         hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1117         if (test_bit(hash&255, bd->set[0]) &&
1118             test_bit((hash>>8)&255, bd->set[0]) &&
1119             test_bit((hash>>16)&255, bd->set[0]))
1120                 return 1;
1121
1122         if (test_bit(hash&255, bd->set[1]) &&
1123             test_bit((hash>>8)&255, bd->set[1]) &&
1124             test_bit((hash>>16)&255, bd->set[1]))
1125                 return 1;
1126
1127         return 0;
1128 }
1129
1130 static void block_delegations(struct knfsd_fh *fh)
1131 {
1132         u32 hash;
1133         struct bloom_pair *bd = &blocked_delegations;
1134
1135         hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1136
1137         spin_lock(&blocked_delegations_lock);
1138         __set_bit(hash&255, bd->set[bd->new]);
1139         __set_bit((hash>>8)&255, bd->set[bd->new]);
1140         __set_bit((hash>>16)&255, bd->set[bd->new]);
1141         if (bd->entries == 0)
1142                 bd->swap_time = ktime_get_seconds();
1143         bd->entries += 1;
1144         spin_unlock(&blocked_delegations_lock);
1145 }
1146
1147 static struct nfs4_delegation *
1148 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
1149                  struct nfs4_clnt_odstate *odstate, u32 dl_type)
1150 {
1151         struct nfs4_delegation *dp;
1152         struct nfs4_stid *stid;
1153         long n;
1154
1155         dprintk("NFSD alloc_init_deleg\n");
1156         n = atomic_long_inc_return(&num_delegations);
1157         if (n < 0 || n > max_delegations)
1158                 goto out_dec;
1159         if (delegation_blocked(&fp->fi_fhandle))
1160                 goto out_dec;
1161         stid = nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg);
1162         if (stid == NULL)
1163                 goto out_dec;
1164         dp = delegstateid(stid);
1165
1166         /*
1167          * delegation seqid's are never incremented.  The 4.1 special
1168          * meaning of seqid 0 isn't meaningful, really, but let's avoid
1169          * 0 anyway just for consistency and use 1:
1170          */
1171         dp->dl_stid.sc_stateid.si_generation = 1;
1172         INIT_LIST_HEAD(&dp->dl_perfile);
1173         INIT_LIST_HEAD(&dp->dl_perclnt);
1174         INIT_LIST_HEAD(&dp->dl_recall_lru);
1175         dp->dl_clnt_odstate = odstate;
1176         get_clnt_odstate(odstate);
1177         dp->dl_type = dl_type;
1178         dp->dl_retries = 1;
1179         dp->dl_recalled = false;
1180         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
1181                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
1182         nfsd4_init_cb(&dp->dl_cb_fattr.ncf_getattr, dp->dl_stid.sc_client,
1183                         &nfsd4_cb_getattr_ops, NFSPROC4_CLNT_CB_GETATTR);
1184         dp->dl_cb_fattr.ncf_file_modified = false;
1185         dp->dl_cb_fattr.ncf_cb_bmap[0] = FATTR4_WORD0_CHANGE | FATTR4_WORD0_SIZE;
1186         get_nfs4_file(fp);
1187         dp->dl_stid.sc_file = fp;
1188         return dp;
1189 out_dec:
1190         atomic_long_dec(&num_delegations);
1191         return NULL;
1192 }
1193
1194 void
1195 nfs4_put_stid(struct nfs4_stid *s)
1196 {
1197         struct nfs4_file *fp = s->sc_file;
1198         struct nfs4_client *clp = s->sc_client;
1199
1200         might_lock(&clp->cl_lock);
1201
1202         if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
1203                 wake_up_all(&close_wq);
1204                 return;
1205         }
1206         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1207         if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
1208                 atomic_dec(&s->sc_client->cl_admin_revoked);
1209         nfs4_free_cpntf_statelist(clp->net, s);
1210         spin_unlock(&clp->cl_lock);
1211         s->sc_free(s);
1212         if (fp)
1213                 put_nfs4_file(fp);
1214 }
1215
1216 void
1217 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
1218 {
1219         stateid_t *src = &stid->sc_stateid;
1220
1221         spin_lock(&stid->sc_lock);
1222         if (unlikely(++src->si_generation == 0))
1223                 src->si_generation = 1;
1224         memcpy(dst, src, sizeof(*dst));
1225         spin_unlock(&stid->sc_lock);
1226 }
1227
1228 static void put_deleg_file(struct nfs4_file *fp)
1229 {
1230         struct nfsd_file *nf = NULL;
1231
1232         spin_lock(&fp->fi_lock);
1233         if (--fp->fi_delegees == 0)
1234                 swap(nf, fp->fi_deleg_file);
1235         spin_unlock(&fp->fi_lock);
1236
1237         if (nf)
1238                 nfsd_file_put(nf);
1239 }
1240
1241 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
1242 {
1243         struct nfs4_file *fp = dp->dl_stid.sc_file;
1244         struct nfsd_file *nf = fp->fi_deleg_file;
1245
1246         WARN_ON_ONCE(!fp->fi_delegees);
1247
1248         kernel_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
1249         put_deleg_file(fp);
1250 }
1251
1252 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
1253 {
1254         put_clnt_odstate(dp->dl_clnt_odstate);
1255         nfs4_unlock_deleg_lease(dp);
1256         nfs4_put_stid(&dp->dl_stid);
1257 }
1258
1259 /**
1260  * nfs4_delegation_exists - Discover if this delegation already exists
1261  * @clp:     a pointer to the nfs4_client we're granting a delegation to
1262  * @fp:      a pointer to the nfs4_file we're granting a delegation on
1263  *
1264  * Return:
1265  *      On success: true iff an existing delegation is found
1266  */
1267
1268 static bool
1269 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
1270 {
1271         struct nfs4_delegation *searchdp = NULL;
1272         struct nfs4_client *searchclp = NULL;
1273
1274         lockdep_assert_held(&state_lock);
1275         lockdep_assert_held(&fp->fi_lock);
1276
1277         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
1278                 searchclp = searchdp->dl_stid.sc_client;
1279                 if (clp == searchclp) {
1280                         return true;
1281                 }
1282         }
1283         return false;
1284 }
1285
1286 /**
1287  * hash_delegation_locked - Add a delegation to the appropriate lists
1288  * @dp:     a pointer to the nfs4_delegation we are adding.
1289  * @fp:     a pointer to the nfs4_file we're granting a delegation on
1290  *
1291  * Return:
1292  *      On success: NULL if the delegation was successfully hashed.
1293  *
1294  *      On error: -EAGAIN if one was previously granted to this
1295  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
1296  *
1297  */
1298
1299 static int
1300 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1301 {
1302         struct nfs4_client *clp = dp->dl_stid.sc_client;
1303
1304         lockdep_assert_held(&state_lock);
1305         lockdep_assert_held(&fp->fi_lock);
1306         lockdep_assert_held(&clp->cl_lock);
1307
1308         if (nfs4_delegation_exists(clp, fp))
1309                 return -EAGAIN;
1310         refcount_inc(&dp->dl_stid.sc_count);
1311         dp->dl_stid.sc_type = SC_TYPE_DELEG;
1312         list_add(&dp->dl_perfile, &fp->fi_delegations);
1313         list_add(&dp->dl_perclnt, &clp->cl_delegations);
1314         return 0;
1315 }
1316
1317 static bool delegation_hashed(struct nfs4_delegation *dp)
1318 {
1319         return !(list_empty(&dp->dl_perfile));
1320 }
1321
1322 static bool
1323 unhash_delegation_locked(struct nfs4_delegation *dp, unsigned short statusmask)
1324 {
1325         struct nfs4_file *fp = dp->dl_stid.sc_file;
1326
1327         lockdep_assert_held(&state_lock);
1328
1329         if (!delegation_hashed(dp))
1330                 return false;
1331
1332         if (statusmask == SC_STATUS_REVOKED &&
1333             dp->dl_stid.sc_client->cl_minorversion == 0)
1334                 statusmask = SC_STATUS_CLOSED;
1335         dp->dl_stid.sc_status |= statusmask;
1336         if (statusmask & SC_STATUS_ADMIN_REVOKED)
1337                 atomic_inc(&dp->dl_stid.sc_client->cl_admin_revoked);
1338
1339         /* Ensure that deleg break won't try to requeue it */
1340         ++dp->dl_time;
1341         spin_lock(&fp->fi_lock);
1342         list_del_init(&dp->dl_perclnt);
1343         list_del_init(&dp->dl_recall_lru);
1344         list_del_init(&dp->dl_perfile);
1345         spin_unlock(&fp->fi_lock);
1346         return true;
1347 }
1348
1349 static void destroy_delegation(struct nfs4_delegation *dp)
1350 {
1351         bool unhashed;
1352
1353         spin_lock(&state_lock);
1354         unhashed = unhash_delegation_locked(dp, SC_STATUS_CLOSED);
1355         spin_unlock(&state_lock);
1356         if (unhashed)
1357                 destroy_unhashed_deleg(dp);
1358 }
1359
1360 static void revoke_delegation(struct nfs4_delegation *dp)
1361 {
1362         struct nfs4_client *clp = dp->dl_stid.sc_client;
1363
1364         WARN_ON(!list_empty(&dp->dl_recall_lru));
1365
1366         trace_nfsd_stid_revoke(&dp->dl_stid);
1367
1368         if (dp->dl_stid.sc_status &
1369             (SC_STATUS_REVOKED | SC_STATUS_ADMIN_REVOKED)) {
1370                 spin_lock(&clp->cl_lock);
1371                 refcount_inc(&dp->dl_stid.sc_count);
1372                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1373                 spin_unlock(&clp->cl_lock);
1374         }
1375         destroy_unhashed_deleg(dp);
1376 }
1377
1378 /*
1379  * SETCLIENTID state
1380  */
1381
1382 static unsigned int clientid_hashval(u32 id)
1383 {
1384         return id & CLIENT_HASH_MASK;
1385 }
1386
1387 static unsigned int clientstr_hashval(struct xdr_netobj name)
1388 {
1389         return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1390 }
1391
1392 /*
1393  * A stateid that had a deny mode associated with it is being released
1394  * or downgraded. Recalculate the deny mode on the file.
1395  */
1396 static void
1397 recalculate_deny_mode(struct nfs4_file *fp)
1398 {
1399         struct nfs4_ol_stateid *stp;
1400         u32 old_deny;
1401
1402         spin_lock(&fp->fi_lock);
1403         old_deny = fp->fi_share_deny;
1404         fp->fi_share_deny = 0;
1405         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1406                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1407                 if (fp->fi_share_deny == old_deny)
1408                         break;
1409         }
1410         spin_unlock(&fp->fi_lock);
1411 }
1412
1413 static void
1414 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1415 {
1416         int i;
1417         bool change = false;
1418
1419         for (i = 1; i < 4; i++) {
1420                 if ((i & deny) != i) {
1421                         change = true;
1422                         clear_deny(i, stp);
1423                 }
1424         }
1425
1426         /* Recalculate per-file deny mode if there was a change */
1427         if (change)
1428                 recalculate_deny_mode(stp->st_stid.sc_file);
1429 }
1430
1431 /* release all access and file references for a given stateid */
1432 static void
1433 release_all_access(struct nfs4_ol_stateid *stp)
1434 {
1435         int i;
1436         struct nfs4_file *fp = stp->st_stid.sc_file;
1437
1438         if (fp && stp->st_deny_bmap != 0)
1439                 recalculate_deny_mode(fp);
1440
1441         for (i = 1; i < 4; i++) {
1442                 if (test_access(i, stp))
1443                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1444                 clear_access(i, stp);
1445         }
1446 }
1447
1448 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1449 {
1450         kfree(sop->so_owner.data);
1451         sop->so_ops->so_free(sop);
1452 }
1453
1454 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1455 {
1456         struct nfs4_client *clp = sop->so_client;
1457
1458         might_lock(&clp->cl_lock);
1459
1460         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1461                 return;
1462         sop->so_ops->so_unhash(sop);
1463         spin_unlock(&clp->cl_lock);
1464         nfs4_free_stateowner(sop);
1465 }
1466
1467 static bool
1468 nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1469 {
1470         return list_empty(&stp->st_perfile);
1471 }
1472
1473 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1474 {
1475         struct nfs4_file *fp = stp->st_stid.sc_file;
1476
1477         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1478
1479         if (list_empty(&stp->st_perfile))
1480                 return false;
1481
1482         spin_lock(&fp->fi_lock);
1483         list_del_init(&stp->st_perfile);
1484         spin_unlock(&fp->fi_lock);
1485         list_del(&stp->st_perstateowner);
1486         return true;
1487 }
1488
1489 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1490 {
1491         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1492
1493         put_clnt_odstate(stp->st_clnt_odstate);
1494         release_all_access(stp);
1495         if (stp->st_stateowner)
1496                 nfs4_put_stateowner(stp->st_stateowner);
1497         WARN_ON(!list_empty(&stid->sc_cp_list));
1498         kmem_cache_free(stateid_slab, stid);
1499 }
1500
1501 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1502 {
1503         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1504         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1505         struct nfsd_file *nf;
1506
1507         nf = find_any_file(stp->st_stid.sc_file);
1508         if (nf) {
1509                 get_file(nf->nf_file);
1510                 filp_close(nf->nf_file, (fl_owner_t)lo);
1511                 nfsd_file_put(nf);
1512         }
1513         nfs4_free_ol_stateid(stid);
1514 }
1515
1516 /*
1517  * Put the persistent reference to an already unhashed generic stateid, while
1518  * holding the cl_lock. If it's the last reference, then put it onto the
1519  * reaplist for later destruction.
1520  */
1521 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1522                                        struct list_head *reaplist)
1523 {
1524         struct nfs4_stid *s = &stp->st_stid;
1525         struct nfs4_client *clp = s->sc_client;
1526
1527         lockdep_assert_held(&clp->cl_lock);
1528
1529         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1530
1531         if (!refcount_dec_and_test(&s->sc_count)) {
1532                 wake_up_all(&close_wq);
1533                 return;
1534         }
1535
1536         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1537         if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
1538                 atomic_dec(&s->sc_client->cl_admin_revoked);
1539         list_add(&stp->st_locks, reaplist);
1540 }
1541
1542 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1543 {
1544         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1545
1546         if (!unhash_ol_stateid(stp))
1547                 return false;
1548         list_del_init(&stp->st_locks);
1549         stp->st_stid.sc_status |= SC_STATUS_CLOSED;
1550         return true;
1551 }
1552
1553 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1554 {
1555         struct nfs4_client *clp = stp->st_stid.sc_client;
1556         bool unhashed;
1557
1558         spin_lock(&clp->cl_lock);
1559         unhashed = unhash_lock_stateid(stp);
1560         spin_unlock(&clp->cl_lock);
1561         if (unhashed)
1562                 nfs4_put_stid(&stp->st_stid);
1563 }
1564
1565 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1566 {
1567         struct nfs4_client *clp = lo->lo_owner.so_client;
1568
1569         lockdep_assert_held(&clp->cl_lock);
1570
1571         list_del_init(&lo->lo_owner.so_strhash);
1572 }
1573
1574 /*
1575  * Free a list of generic stateids that were collected earlier after being
1576  * fully unhashed.
1577  */
1578 static void
1579 free_ol_stateid_reaplist(struct list_head *reaplist)
1580 {
1581         struct nfs4_ol_stateid *stp;
1582         struct nfs4_file *fp;
1583
1584         might_sleep();
1585
1586         while (!list_empty(reaplist)) {
1587                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1588                                        st_locks);
1589                 list_del(&stp->st_locks);
1590                 fp = stp->st_stid.sc_file;
1591                 stp->st_stid.sc_free(&stp->st_stid);
1592                 if (fp)
1593                         put_nfs4_file(fp);
1594         }
1595 }
1596
1597 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1598                                        struct list_head *reaplist)
1599 {
1600         struct nfs4_ol_stateid *stp;
1601
1602         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1603
1604         while (!list_empty(&open_stp->st_locks)) {
1605                 stp = list_entry(open_stp->st_locks.next,
1606                                 struct nfs4_ol_stateid, st_locks);
1607                 unhash_lock_stateid(stp);
1608                 put_ol_stateid_locked(stp, reaplist);
1609         }
1610 }
1611
1612 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1613                                 struct list_head *reaplist)
1614 {
1615         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1616
1617         if (!unhash_ol_stateid(stp))
1618                 return false;
1619         release_open_stateid_locks(stp, reaplist);
1620         return true;
1621 }
1622
1623 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1624 {
1625         LIST_HEAD(reaplist);
1626
1627         spin_lock(&stp->st_stid.sc_client->cl_lock);
1628         stp->st_stid.sc_status |= SC_STATUS_CLOSED;
1629         if (unhash_open_stateid(stp, &reaplist))
1630                 put_ol_stateid_locked(stp, &reaplist);
1631         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1632         free_ol_stateid_reaplist(&reaplist);
1633 }
1634
1635 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1636 {
1637         struct nfs4_client *clp = oo->oo_owner.so_client;
1638
1639         lockdep_assert_held(&clp->cl_lock);
1640
1641         list_del_init(&oo->oo_owner.so_strhash);
1642         list_del_init(&oo->oo_perclient);
1643 }
1644
1645 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1646 {
1647         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1648                                           nfsd_net_id);
1649         struct nfs4_ol_stateid *s;
1650
1651         spin_lock(&nn->client_lock);
1652         s = oo->oo_last_closed_stid;
1653         if (s) {
1654                 list_del_init(&oo->oo_close_lru);
1655                 oo->oo_last_closed_stid = NULL;
1656         }
1657         spin_unlock(&nn->client_lock);
1658         if (s)
1659                 nfs4_put_stid(&s->st_stid);
1660 }
1661
1662 static void release_openowner(struct nfs4_openowner *oo)
1663 {
1664         struct nfs4_ol_stateid *stp;
1665         struct nfs4_client *clp = oo->oo_owner.so_client;
1666         struct list_head reaplist;
1667
1668         INIT_LIST_HEAD(&reaplist);
1669
1670         spin_lock(&clp->cl_lock);
1671         unhash_openowner_locked(oo);
1672         while (!list_empty(&oo->oo_owner.so_stateids)) {
1673                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1674                                 struct nfs4_ol_stateid, st_perstateowner);
1675                 if (unhash_open_stateid(stp, &reaplist))
1676                         put_ol_stateid_locked(stp, &reaplist);
1677         }
1678         spin_unlock(&clp->cl_lock);
1679         free_ol_stateid_reaplist(&reaplist);
1680         release_last_closed_stateid(oo);
1681         nfs4_put_stateowner(&oo->oo_owner);
1682 }
1683
1684 static struct nfs4_stid *find_one_sb_stid(struct nfs4_client *clp,
1685                                           struct super_block *sb,
1686                                           unsigned int sc_types)
1687 {
1688         unsigned long id, tmp;
1689         struct nfs4_stid *stid;
1690
1691         spin_lock(&clp->cl_lock);
1692         idr_for_each_entry_ul(&clp->cl_stateids, stid, tmp, id)
1693                 if ((stid->sc_type & sc_types) &&
1694                     stid->sc_status == 0 &&
1695                     stid->sc_file->fi_inode->i_sb == sb) {
1696                         refcount_inc(&stid->sc_count);
1697                         break;
1698                 }
1699         spin_unlock(&clp->cl_lock);
1700         return stid;
1701 }
1702
1703 /**
1704  * nfsd4_revoke_states - revoke all nfsv4 states associated with given filesystem
1705  * @net:  used to identify instance of nfsd (there is one per net namespace)
1706  * @sb:   super_block used to identify target filesystem
1707  *
1708  * All nfs4 states (open, lock, delegation, layout) held by the server instance
1709  * and associated with a file on the given filesystem will be revoked resulting
1710  * in any files being closed and so all references from nfsd to the filesystem
1711  * being released.  Thus nfsd will no longer prevent the filesystem from being
1712  * unmounted.
1713  *
1714  * The clients which own the states will subsequently being notified that the
1715  * states have been "admin-revoked".
1716  */
1717 void nfsd4_revoke_states(struct net *net, struct super_block *sb)
1718 {
1719         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1720         unsigned int idhashval;
1721         unsigned int sc_types;
1722
1723         sc_types = SC_TYPE_OPEN | SC_TYPE_LOCK | SC_TYPE_DELEG | SC_TYPE_LAYOUT;
1724
1725         spin_lock(&nn->client_lock);
1726         for (idhashval = 0; idhashval < CLIENT_HASH_MASK; idhashval++) {
1727                 struct list_head *head = &nn->conf_id_hashtbl[idhashval];
1728                 struct nfs4_client *clp;
1729         retry:
1730                 list_for_each_entry(clp, head, cl_idhash) {
1731                         struct nfs4_stid *stid = find_one_sb_stid(clp, sb,
1732                                                                   sc_types);
1733                         if (stid) {
1734                                 struct nfs4_ol_stateid *stp;
1735                                 struct nfs4_delegation *dp;
1736                                 struct nfs4_layout_stateid *ls;
1737
1738                                 spin_unlock(&nn->client_lock);
1739                                 switch (stid->sc_type) {
1740                                 case SC_TYPE_OPEN:
1741                                         stp = openlockstateid(stid);
1742                                         mutex_lock_nested(&stp->st_mutex,
1743                                                           OPEN_STATEID_MUTEX);
1744
1745                                         spin_lock(&clp->cl_lock);
1746                                         if (stid->sc_status == 0) {
1747                                                 stid->sc_status |=
1748                                                         SC_STATUS_ADMIN_REVOKED;
1749                                                 atomic_inc(&clp->cl_admin_revoked);
1750                                                 spin_unlock(&clp->cl_lock);
1751                                                 release_all_access(stp);
1752                                         } else
1753                                                 spin_unlock(&clp->cl_lock);
1754                                         mutex_unlock(&stp->st_mutex);
1755                                         break;
1756                                 case SC_TYPE_LOCK:
1757                                         stp = openlockstateid(stid);
1758                                         mutex_lock_nested(&stp->st_mutex,
1759                                                           LOCK_STATEID_MUTEX);
1760                                         spin_lock(&clp->cl_lock);
1761                                         if (stid->sc_status == 0) {
1762                                                 struct nfs4_lockowner *lo =
1763                                                         lockowner(stp->st_stateowner);
1764                                                 struct nfsd_file *nf;
1765
1766                                                 stid->sc_status |=
1767                                                         SC_STATUS_ADMIN_REVOKED;
1768                                                 atomic_inc(&clp->cl_admin_revoked);
1769                                                 spin_unlock(&clp->cl_lock);
1770                                                 nf = find_any_file(stp->st_stid.sc_file);
1771                                                 if (nf) {
1772                                                         get_file(nf->nf_file);
1773                                                         filp_close(nf->nf_file,
1774                                                                    (fl_owner_t)lo);
1775                                                         nfsd_file_put(nf);
1776                                                 }
1777                                                 release_all_access(stp);
1778                                         } else
1779                                                 spin_unlock(&clp->cl_lock);
1780                                         mutex_unlock(&stp->st_mutex);
1781                                         break;
1782                                 case SC_TYPE_DELEG:
1783                                         dp = delegstateid(stid);
1784                                         spin_lock(&state_lock);
1785                                         if (!unhash_delegation_locked(
1786                                                     dp, SC_STATUS_ADMIN_REVOKED))
1787                                                 dp = NULL;
1788                                         spin_unlock(&state_lock);
1789                                         if (dp)
1790                                                 revoke_delegation(dp);
1791                                         break;
1792                                 case SC_TYPE_LAYOUT:
1793                                         ls = layoutstateid(stid);
1794                                         nfsd4_close_layout(ls);
1795                                         break;
1796                                 }
1797                                 nfs4_put_stid(stid);
1798                                 spin_lock(&nn->client_lock);
1799                                 if (clp->cl_minorversion == 0)
1800                                         /* Allow cleanup after a lease period.
1801                                          * store_release ensures cleanup will
1802                                          * see any newly revoked states if it
1803                                          * sees the time updated.
1804                                          */
1805                                         nn->nfs40_last_revoke =
1806                                                 ktime_get_boottime_seconds();
1807                                 goto retry;
1808                         }
1809                 }
1810         }
1811         spin_unlock(&nn->client_lock);
1812 }
1813
1814 static inline int
1815 hash_sessionid(struct nfs4_sessionid *sessionid)
1816 {
1817         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1818
1819         return sid->sequence % SESSION_HASH_SIZE;
1820 }
1821
1822 #ifdef CONFIG_SUNRPC_DEBUG
1823 static inline void
1824 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1825 {
1826         u32 *ptr = (u32 *)(&sessionid->data[0]);
1827         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1828 }
1829 #else
1830 static inline void
1831 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1832 {
1833 }
1834 #endif
1835
1836 /*
1837  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1838  * won't be used for replay.
1839  */
1840 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1841 {
1842         struct nfs4_stateowner *so = cstate->replay_owner;
1843
1844         if (nfserr == nfserr_replay_me)
1845                 return;
1846
1847         if (!seqid_mutating_err(ntohl(nfserr))) {
1848                 nfsd4_cstate_clear_replay(cstate);
1849                 return;
1850         }
1851         if (!so)
1852                 return;
1853         if (so->so_is_open_owner)
1854                 release_last_closed_stateid(openowner(so));
1855         so->so_seqid++;
1856         return;
1857 }
1858
1859 static void
1860 gen_sessionid(struct nfsd4_session *ses)
1861 {
1862         struct nfs4_client *clp = ses->se_client;
1863         struct nfsd4_sessionid *sid;
1864
1865         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1866         sid->clientid = clp->cl_clientid;
1867         sid->sequence = current_sessionid++;
1868         sid->reserved = 0;
1869 }
1870
1871 /*
1872  * The protocol defines ca_maxresponssize_cached to include the size of
1873  * the rpc header, but all we need to cache is the data starting after
1874  * the end of the initial SEQUENCE operation--the rest we regenerate
1875  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1876  * value that is the number of bytes in our cache plus a few additional
1877  * bytes.  In order to stay on the safe side, and not promise more than
1878  * we can cache, those additional bytes must be the minimum possible: 24
1879  * bytes of rpc header (xid through accept state, with AUTH_NULL
1880  * verifier), 12 for the compound header (with zero-length tag), and 44
1881  * for the SEQUENCE op response:
1882  */
1883 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1884
1885 static void
1886 free_session_slots(struct nfsd4_session *ses)
1887 {
1888         int i;
1889
1890         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1891                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1892                 kfree(ses->se_slots[i]);
1893         }
1894 }
1895
1896 /*
1897  * We don't actually need to cache the rpc and session headers, so we
1898  * can allocate a little less for each slot:
1899  */
1900 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1901 {
1902         u32 size;
1903
1904         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1905                 size = 0;
1906         else
1907                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1908         return size + sizeof(struct nfsd4_slot);
1909 }
1910
1911 /*
1912  * XXX: If we run out of reserved DRC memory we could (up to a point)
1913  * re-negotiate active sessions and reduce their slot usage to make
1914  * room for new connections. For now we just fail the create session.
1915  */
1916 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1917 {
1918         u32 slotsize = slot_bytes(ca);
1919         u32 num = ca->maxreqs;
1920         unsigned long avail, total_avail;
1921         unsigned int scale_factor;
1922
1923         spin_lock(&nfsd_drc_lock);
1924         if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1925                 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1926         else
1927                 /* We have handed out more space than we chose in
1928                  * set_max_drc() to allow.  That isn't really a
1929                  * problem as long as that doesn't make us think we
1930                  * have lots more due to integer overflow.
1931                  */
1932                 total_avail = 0;
1933         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1934         /*
1935          * Never use more than a fraction of the remaining memory,
1936          * unless it's the only way to give this client a slot.
1937          * The chosen fraction is either 1/8 or 1/number of threads,
1938          * whichever is smaller.  This ensures there are adequate
1939          * slots to support multiple clients per thread.
1940          * Give the client one slot even if that would require
1941          * over-allocation--it is better than failure.
1942          */
1943         scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1944
1945         avail = clamp_t(unsigned long, avail, slotsize,
1946                         total_avail/scale_factor);
1947         num = min_t(int, num, avail / slotsize);
1948         num = max_t(int, num, 1);
1949         nfsd_drc_mem_used += num * slotsize;
1950         spin_unlock(&nfsd_drc_lock);
1951
1952         return num;
1953 }
1954
1955 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1956 {
1957         int slotsize = slot_bytes(ca);
1958
1959         spin_lock(&nfsd_drc_lock);
1960         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1961         spin_unlock(&nfsd_drc_lock);
1962 }
1963
1964 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1965                                            struct nfsd4_channel_attrs *battrs)
1966 {
1967         int numslots = fattrs->maxreqs;
1968         int slotsize = slot_bytes(fattrs);
1969         struct nfsd4_session *new;
1970         int i;
1971
1972         BUILD_BUG_ON(struct_size(new, se_slots, NFSD_MAX_SLOTS_PER_SESSION)
1973                      > PAGE_SIZE);
1974
1975         new = kzalloc(struct_size(new, se_slots, numslots), GFP_KERNEL);
1976         if (!new)
1977                 return NULL;
1978         /* allocate each struct nfsd4_slot and data cache in one piece */
1979         for (i = 0; i < numslots; i++) {
1980                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1981                 if (!new->se_slots[i])
1982                         goto out_free;
1983         }
1984
1985         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1986         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1987
1988         return new;
1989 out_free:
1990         while (i--)
1991                 kfree(new->se_slots[i]);
1992         kfree(new);
1993         return NULL;
1994 }
1995
1996 static void free_conn(struct nfsd4_conn *c)
1997 {
1998         svc_xprt_put(c->cn_xprt);
1999         kfree(c);
2000 }
2001
2002 static void nfsd4_conn_lost(struct svc_xpt_user *u)
2003 {
2004         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
2005         struct nfs4_client *clp = c->cn_session->se_client;
2006
2007         trace_nfsd_cb_lost(clp);
2008
2009         spin_lock(&clp->cl_lock);
2010         if (!list_empty(&c->cn_persession)) {
2011                 list_del(&c->cn_persession);
2012                 free_conn(c);
2013         }
2014         nfsd4_probe_callback(clp);
2015         spin_unlock(&clp->cl_lock);
2016 }
2017
2018 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
2019 {
2020         struct nfsd4_conn *conn;
2021
2022         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
2023         if (!conn)
2024                 return NULL;
2025         svc_xprt_get(rqstp->rq_xprt);
2026         conn->cn_xprt = rqstp->rq_xprt;
2027         conn->cn_flags = flags;
2028         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
2029         return conn;
2030 }
2031
2032 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
2033 {
2034         conn->cn_session = ses;
2035         list_add(&conn->cn_persession, &ses->se_conns);
2036 }
2037
2038 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
2039 {
2040         struct nfs4_client *clp = ses->se_client;
2041
2042         spin_lock(&clp->cl_lock);
2043         __nfsd4_hash_conn(conn, ses);
2044         spin_unlock(&clp->cl_lock);
2045 }
2046
2047 static int nfsd4_register_conn(struct nfsd4_conn *conn)
2048 {
2049         conn->cn_xpt_user.callback = nfsd4_conn_lost;
2050         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
2051 }
2052
2053 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
2054 {
2055         int ret;
2056
2057         nfsd4_hash_conn(conn, ses);
2058         ret = nfsd4_register_conn(conn);
2059         if (ret)
2060                 /* oops; xprt is already down: */
2061                 nfsd4_conn_lost(&conn->cn_xpt_user);
2062         /* We may have gained or lost a callback channel: */
2063         nfsd4_probe_callback_sync(ses->se_client);
2064 }
2065
2066 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
2067 {
2068         u32 dir = NFS4_CDFC4_FORE;
2069
2070         if (cses->flags & SESSION4_BACK_CHAN)
2071                 dir |= NFS4_CDFC4_BACK;
2072         return alloc_conn(rqstp, dir);
2073 }
2074
2075 /* must be called under client_lock */
2076 static void nfsd4_del_conns(struct nfsd4_session *s)
2077 {
2078         struct nfs4_client *clp = s->se_client;
2079         struct nfsd4_conn *c;
2080
2081         spin_lock(&clp->cl_lock);
2082         while (!list_empty(&s->se_conns)) {
2083                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
2084                 list_del_init(&c->cn_persession);
2085                 spin_unlock(&clp->cl_lock);
2086
2087                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
2088                 free_conn(c);
2089
2090                 spin_lock(&clp->cl_lock);
2091         }
2092         spin_unlock(&clp->cl_lock);
2093 }
2094
2095 static void __free_session(struct nfsd4_session *ses)
2096 {
2097         free_session_slots(ses);
2098         kfree(ses);
2099 }
2100
2101 static void free_session(struct nfsd4_session *ses)
2102 {
2103         nfsd4_del_conns(ses);
2104         nfsd4_put_drc_mem(&ses->se_fchannel);
2105         __free_session(ses);
2106 }
2107
2108 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
2109 {
2110         int idx;
2111         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2112
2113         new->se_client = clp;
2114         gen_sessionid(new);
2115
2116         INIT_LIST_HEAD(&new->se_conns);
2117
2118         new->se_cb_seq_nr = 1;
2119         new->se_flags = cses->flags;
2120         new->se_cb_prog = cses->callback_prog;
2121         new->se_cb_sec = cses->cb_sec;
2122         atomic_set(&new->se_ref, 0);
2123         idx = hash_sessionid(&new->se_sessionid);
2124         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
2125         spin_lock(&clp->cl_lock);
2126         list_add(&new->se_perclnt, &clp->cl_sessions);
2127         spin_unlock(&clp->cl_lock);
2128
2129         {
2130                 struct sockaddr *sa = svc_addr(rqstp);
2131                 /*
2132                  * This is a little silly; with sessions there's no real
2133                  * use for the callback address.  Use the peer address
2134                  * as a reasonable default for now, but consider fixing
2135                  * the rpc client not to require an address in the
2136                  * future:
2137                  */
2138                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
2139                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
2140         }
2141 }
2142
2143 /* caller must hold client_lock */
2144 static struct nfsd4_session *
2145 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
2146 {
2147         struct nfsd4_session *elem;
2148         int idx;
2149         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2150
2151         lockdep_assert_held(&nn->client_lock);
2152
2153         dump_sessionid(__func__, sessionid);
2154         idx = hash_sessionid(sessionid);
2155         /* Search in the appropriate list */
2156         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
2157                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
2158                             NFS4_MAX_SESSIONID_LEN)) {
2159                         return elem;
2160                 }
2161         }
2162
2163         dprintk("%s: session not found\n", __func__);
2164         return NULL;
2165 }
2166
2167 static struct nfsd4_session *
2168 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
2169                 __be32 *ret)
2170 {
2171         struct nfsd4_session *session;
2172         __be32 status = nfserr_badsession;
2173
2174         session = __find_in_sessionid_hashtbl(sessionid, net);
2175         if (!session)
2176                 goto out;
2177         status = nfsd4_get_session_locked(session);
2178         if (status)
2179                 session = NULL;
2180 out:
2181         *ret = status;
2182         return session;
2183 }
2184
2185 /* caller must hold client_lock */
2186 static void
2187 unhash_session(struct nfsd4_session *ses)
2188 {
2189         struct nfs4_client *clp = ses->se_client;
2190         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2191
2192         lockdep_assert_held(&nn->client_lock);
2193
2194         list_del(&ses->se_hash);
2195         spin_lock(&ses->se_client->cl_lock);
2196         list_del(&ses->se_perclnt);
2197         spin_unlock(&ses->se_client->cl_lock);
2198 }
2199
2200 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
2201 static int
2202 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
2203 {
2204         /*
2205          * We're assuming the clid was not given out from a boot
2206          * precisely 2^32 (about 136 years) before this one.  That seems
2207          * a safe assumption:
2208          */
2209         if (clid->cl_boot == (u32)nn->boot_time)
2210                 return 0;
2211         trace_nfsd_clid_stale(clid);
2212         return 1;
2213 }
2214
2215 /*
2216  * XXX Should we use a slab cache ?
2217  * This type of memory management is somewhat inefficient, but we use it
2218  * anyway since SETCLIENTID is not a common operation.
2219  */
2220 static struct nfs4_client *alloc_client(struct xdr_netobj name,
2221                                 struct nfsd_net *nn)
2222 {
2223         struct nfs4_client *clp;
2224         int i;
2225
2226         if (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) {
2227                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
2228                 return NULL;
2229         }
2230         clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
2231         if (clp == NULL)
2232                 return NULL;
2233         xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
2234         if (clp->cl_name.data == NULL)
2235                 goto err_no_name;
2236         clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
2237                                                  sizeof(struct list_head),
2238                                                  GFP_KERNEL);
2239         if (!clp->cl_ownerstr_hashtbl)
2240                 goto err_no_hashtbl;
2241         clp->cl_callback_wq = alloc_ordered_workqueue("nfsd4_callbacks", 0);
2242         if (!clp->cl_callback_wq)
2243                 goto err_no_callback_wq;
2244
2245         for (i = 0; i < OWNER_HASH_SIZE; i++)
2246                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
2247         INIT_LIST_HEAD(&clp->cl_sessions);
2248         idr_init(&clp->cl_stateids);
2249         atomic_set(&clp->cl_rpc_users, 0);
2250         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
2251         clp->cl_state = NFSD4_ACTIVE;
2252         atomic_inc(&nn->nfs4_client_count);
2253         atomic_set(&clp->cl_delegs_in_recall, 0);
2254         INIT_LIST_HEAD(&clp->cl_idhash);
2255         INIT_LIST_HEAD(&clp->cl_openowners);
2256         INIT_LIST_HEAD(&clp->cl_delegations);
2257         INIT_LIST_HEAD(&clp->cl_lru);
2258         INIT_LIST_HEAD(&clp->cl_revoked);
2259 #ifdef CONFIG_NFSD_PNFS
2260         INIT_LIST_HEAD(&clp->cl_lo_states);
2261 #endif
2262         INIT_LIST_HEAD(&clp->async_copies);
2263         spin_lock_init(&clp->async_lock);
2264         spin_lock_init(&clp->cl_lock);
2265         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
2266         return clp;
2267 err_no_callback_wq:
2268         kfree(clp->cl_ownerstr_hashtbl);
2269 err_no_hashtbl:
2270         kfree(clp->cl_name.data);
2271 err_no_name:
2272         kmem_cache_free(client_slab, clp);
2273         return NULL;
2274 }
2275
2276 static void __free_client(struct kref *k)
2277 {
2278         struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
2279         struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
2280
2281         free_svc_cred(&clp->cl_cred);
2282         destroy_workqueue(clp->cl_callback_wq);
2283         kfree(clp->cl_ownerstr_hashtbl);
2284         kfree(clp->cl_name.data);
2285         kfree(clp->cl_nii_domain.data);
2286         kfree(clp->cl_nii_name.data);
2287         idr_destroy(&clp->cl_stateids);
2288         kfree(clp->cl_ra);
2289         kmem_cache_free(client_slab, clp);
2290 }
2291
2292 static void drop_client(struct nfs4_client *clp)
2293 {
2294         kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
2295 }
2296
2297 static void
2298 free_client(struct nfs4_client *clp)
2299 {
2300         while (!list_empty(&clp->cl_sessions)) {
2301                 struct nfsd4_session *ses;
2302                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
2303                                 se_perclnt);
2304                 list_del(&ses->se_perclnt);
2305                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
2306                 free_session(ses);
2307         }
2308         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
2309         if (clp->cl_nfsd_dentry) {
2310                 nfsd_client_rmdir(clp->cl_nfsd_dentry);
2311                 clp->cl_nfsd_dentry = NULL;
2312                 wake_up_all(&expiry_wq);
2313         }
2314         drop_client(clp);
2315 }
2316
2317 /* must be called under the client_lock */
2318 static void
2319 unhash_client_locked(struct nfs4_client *clp)
2320 {
2321         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2322         struct nfsd4_session *ses;
2323
2324         lockdep_assert_held(&nn->client_lock);
2325
2326         /* Mark the client as expired! */
2327         clp->cl_time = 0;
2328         /* Make it invisible */
2329         if (!list_empty(&clp->cl_idhash)) {
2330                 list_del_init(&clp->cl_idhash);
2331                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2332                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
2333                 else
2334                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2335         }
2336         list_del_init(&clp->cl_lru);
2337         spin_lock(&clp->cl_lock);
2338         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2339                 list_del_init(&ses->se_hash);
2340         spin_unlock(&clp->cl_lock);
2341 }
2342
2343 static void
2344 unhash_client(struct nfs4_client *clp)
2345 {
2346         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2347
2348         spin_lock(&nn->client_lock);
2349         unhash_client_locked(clp);
2350         spin_unlock(&nn->client_lock);
2351 }
2352
2353 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
2354 {
2355         int users = atomic_read(&clp->cl_rpc_users);
2356
2357         trace_nfsd_mark_client_expired(clp, users);
2358
2359         if (users)
2360                 return nfserr_jukebox;
2361         unhash_client_locked(clp);
2362         return nfs_ok;
2363 }
2364
2365 static void
2366 __destroy_client(struct nfs4_client *clp)
2367 {
2368         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2369         int i;
2370         struct nfs4_openowner *oo;
2371         struct nfs4_delegation *dp;
2372         struct list_head reaplist;
2373
2374         INIT_LIST_HEAD(&reaplist);
2375         spin_lock(&state_lock);
2376         while (!list_empty(&clp->cl_delegations)) {
2377                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2378                 unhash_delegation_locked(dp, SC_STATUS_CLOSED);
2379                 list_add(&dp->dl_recall_lru, &reaplist);
2380         }
2381         spin_unlock(&state_lock);
2382         while (!list_empty(&reaplist)) {
2383                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2384                 list_del_init(&dp->dl_recall_lru);
2385                 destroy_unhashed_deleg(dp);
2386         }
2387         while (!list_empty(&clp->cl_revoked)) {
2388                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2389                 list_del_init(&dp->dl_recall_lru);
2390                 nfs4_put_stid(&dp->dl_stid);
2391         }
2392         while (!list_empty(&clp->cl_openowners)) {
2393                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2394                 nfs4_get_stateowner(&oo->oo_owner);
2395                 release_openowner(oo);
2396         }
2397         for (i = 0; i < OWNER_HASH_SIZE; i++) {
2398                 struct nfs4_stateowner *so, *tmp;
2399
2400                 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2401                                          so_strhash) {
2402                         /* Should be no openowners at this point */
2403                         WARN_ON_ONCE(so->so_is_open_owner);
2404                         remove_blocked_locks(lockowner(so));
2405                 }
2406         }
2407         nfsd4_return_all_client_layouts(clp);
2408         nfsd4_shutdown_copy(clp);
2409         nfsd4_shutdown_callback(clp);
2410         if (clp->cl_cb_conn.cb_xprt)
2411                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2412         atomic_add_unless(&nn->nfs4_client_count, -1, 0);
2413         nfsd4_dec_courtesy_client_count(nn, clp);
2414         free_client(clp);
2415         wake_up_all(&expiry_wq);
2416 }
2417
2418 static void
2419 destroy_client(struct nfs4_client *clp)
2420 {
2421         unhash_client(clp);
2422         __destroy_client(clp);
2423 }
2424
2425 static void inc_reclaim_complete(struct nfs4_client *clp)
2426 {
2427         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2428
2429         if (!nn->track_reclaim_completes)
2430                 return;
2431         if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2432                 return;
2433         if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2434                         nn->reclaim_str_hashtbl_size) {
2435                 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2436                                 clp->net->ns.inum);
2437                 nfsd4_end_grace(nn);
2438         }
2439 }
2440
2441 static void expire_client(struct nfs4_client *clp)
2442 {
2443         unhash_client(clp);
2444         nfsd4_client_record_remove(clp);
2445         __destroy_client(clp);
2446 }
2447
2448 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2449 {
2450         memcpy(target->cl_verifier.data, source->data,
2451                         sizeof(target->cl_verifier.data));
2452 }
2453
2454 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2455 {
2456         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2457         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2458 }
2459
2460 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2461 {
2462         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2463         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2464                                                                 GFP_KERNEL);
2465         target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2466         if ((source->cr_principal && !target->cr_principal) ||
2467             (source->cr_raw_principal && !target->cr_raw_principal) ||
2468             (source->cr_targ_princ && !target->cr_targ_princ))
2469                 return -ENOMEM;
2470
2471         target->cr_flavor = source->cr_flavor;
2472         target->cr_uid = source->cr_uid;
2473         target->cr_gid = source->cr_gid;
2474         target->cr_group_info = source->cr_group_info;
2475         get_group_info(target->cr_group_info);
2476         target->cr_gss_mech = source->cr_gss_mech;
2477         if (source->cr_gss_mech)
2478                 gss_mech_get(source->cr_gss_mech);
2479         return 0;
2480 }
2481
2482 static int
2483 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2484 {
2485         if (o1->len < o2->len)
2486                 return -1;
2487         if (o1->len > o2->len)
2488                 return 1;
2489         return memcmp(o1->data, o2->data, o1->len);
2490 }
2491
2492 static int
2493 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2494 {
2495         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2496 }
2497
2498 static int
2499 same_clid(clientid_t *cl1, clientid_t *cl2)
2500 {
2501         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2502 }
2503
2504 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2505 {
2506         int i;
2507
2508         if (g1->ngroups != g2->ngroups)
2509                 return false;
2510         for (i=0; i<g1->ngroups; i++)
2511                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2512                         return false;
2513         return true;
2514 }
2515
2516 /*
2517  * RFC 3530 language requires clid_inuse be returned when the
2518  * "principal" associated with a requests differs from that previously
2519  * used.  We use uid, gid's, and gss principal string as our best
2520  * approximation.  We also don't want to allow non-gss use of a client
2521  * established using gss: in theory cr_principal should catch that
2522  * change, but in practice cr_principal can be null even in the gss case
2523  * since gssd doesn't always pass down a principal string.
2524  */
2525 static bool is_gss_cred(struct svc_cred *cr)
2526 {
2527         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2528         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2529 }
2530
2531
2532 static bool
2533 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2534 {
2535         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2536                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2537                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2538                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2539                 return false;
2540         /* XXX: check that cr_targ_princ fields match ? */
2541         if (cr1->cr_principal == cr2->cr_principal)
2542                 return true;
2543         if (!cr1->cr_principal || !cr2->cr_principal)
2544                 return false;
2545         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2546 }
2547
2548 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2549 {
2550         struct svc_cred *cr = &rqstp->rq_cred;
2551         u32 service;
2552
2553         if (!cr->cr_gss_mech)
2554                 return false;
2555         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2556         return service == RPC_GSS_SVC_INTEGRITY ||
2557                service == RPC_GSS_SVC_PRIVACY;
2558 }
2559
2560 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2561 {
2562         struct svc_cred *cr = &rqstp->rq_cred;
2563
2564         if (!cl->cl_mach_cred)
2565                 return true;
2566         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2567                 return false;
2568         if (!svc_rqst_integrity_protected(rqstp))
2569                 return false;
2570         if (cl->cl_cred.cr_raw_principal)
2571                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2572                                                 cr->cr_raw_principal);
2573         if (!cr->cr_principal)
2574                 return false;
2575         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2576 }
2577
2578 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2579 {
2580         __be32 verf[2];
2581
2582         /*
2583          * This is opaque to client, so no need to byte-swap. Use
2584          * __force to keep sparse happy
2585          */
2586         verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2587         verf[1] = (__force __be32)nn->clverifier_counter++;
2588         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2589 }
2590
2591 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2592 {
2593         clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2594         clp->cl_clientid.cl_id = nn->clientid_counter++;
2595         gen_confirm(clp, nn);
2596 }
2597
2598 static struct nfs4_stid *
2599 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2600 {
2601         struct nfs4_stid *ret;
2602
2603         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2604         if (!ret || !ret->sc_type)
2605                 return NULL;
2606         return ret;
2607 }
2608
2609 static struct nfs4_stid *
2610 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t,
2611                      unsigned short typemask, unsigned short ok_states)
2612 {
2613         struct nfs4_stid *s;
2614
2615         spin_lock(&cl->cl_lock);
2616         s = find_stateid_locked(cl, t);
2617         if (s != NULL) {
2618                 if ((s->sc_status & ~ok_states) == 0 &&
2619                     (typemask & s->sc_type))
2620                         refcount_inc(&s->sc_count);
2621                 else
2622                         s = NULL;
2623         }
2624         spin_unlock(&cl->cl_lock);
2625         return s;
2626 }
2627
2628 static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2629 {
2630         struct nfsdfs_client *nc;
2631         nc = get_nfsdfs_client(inode);
2632         if (!nc)
2633                 return NULL;
2634         return container_of(nc, struct nfs4_client, cl_nfsdfs);
2635 }
2636
2637 static void seq_quote_mem(struct seq_file *m, char *data, int len)
2638 {
2639         seq_puts(m, "\"");
2640         seq_escape_mem(m, data, len, ESCAPE_HEX | ESCAPE_NAP | ESCAPE_APPEND, "\"\\");
2641         seq_puts(m, "\"");
2642 }
2643
2644 static const char *cb_state2str(int state)
2645 {
2646         switch (state) {
2647         case NFSD4_CB_UP:
2648                 return "UP";
2649         case NFSD4_CB_UNKNOWN:
2650                 return "UNKNOWN";
2651         case NFSD4_CB_DOWN:
2652                 return "DOWN";
2653         case NFSD4_CB_FAULT:
2654                 return "FAULT";
2655         }
2656         return "UNDEFINED";
2657 }
2658
2659 static int client_info_show(struct seq_file *m, void *v)
2660 {
2661         struct inode *inode = file_inode(m->file);
2662         struct nfs4_client *clp;
2663         u64 clid;
2664
2665         clp = get_nfsdfs_clp(inode);
2666         if (!clp)
2667                 return -ENXIO;
2668         memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2669         seq_printf(m, "clientid: 0x%llx\n", clid);
2670         seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2671
2672         if (clp->cl_state == NFSD4_COURTESY)
2673                 seq_puts(m, "status: courtesy\n");
2674         else if (clp->cl_state == NFSD4_EXPIRABLE)
2675                 seq_puts(m, "status: expirable\n");
2676         else if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2677                 seq_puts(m, "status: confirmed\n");
2678         else
2679                 seq_puts(m, "status: unconfirmed\n");
2680         seq_printf(m, "seconds from last renew: %lld\n",
2681                 ktime_get_boottime_seconds() - clp->cl_time);
2682         seq_puts(m, "name: ");
2683         seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2684         seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2685         if (clp->cl_nii_domain.data) {
2686                 seq_puts(m, "Implementation domain: ");
2687                 seq_quote_mem(m, clp->cl_nii_domain.data,
2688                                         clp->cl_nii_domain.len);
2689                 seq_puts(m, "\nImplementation name: ");
2690                 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2691                 seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2692                         clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2693         }
2694         seq_printf(m, "callback state: %s\n", cb_state2str(clp->cl_cb_state));
2695         seq_printf(m, "callback address: %pISpc\n", &clp->cl_cb_conn.cb_addr);
2696         seq_printf(m, "admin-revoked states: %d\n",
2697                    atomic_read(&clp->cl_admin_revoked));
2698         drop_client(clp);
2699
2700         return 0;
2701 }
2702
2703 DEFINE_SHOW_ATTRIBUTE(client_info);
2704
2705 static void *states_start(struct seq_file *s, loff_t *pos)
2706         __acquires(&clp->cl_lock)
2707 {
2708         struct nfs4_client *clp = s->private;
2709         unsigned long id = *pos;
2710         void *ret;
2711
2712         spin_lock(&clp->cl_lock);
2713         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2714         *pos = id;
2715         return ret;
2716 }
2717
2718 static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2719 {
2720         struct nfs4_client *clp = s->private;
2721         unsigned long id = *pos;
2722         void *ret;
2723
2724         id = *pos;
2725         id++;
2726         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2727         *pos = id;
2728         return ret;
2729 }
2730
2731 static void states_stop(struct seq_file *s, void *v)
2732         __releases(&clp->cl_lock)
2733 {
2734         struct nfs4_client *clp = s->private;
2735
2736         spin_unlock(&clp->cl_lock);
2737 }
2738
2739 static void nfs4_show_fname(struct seq_file *s, struct nfsd_file *f)
2740 {
2741          seq_printf(s, "filename: \"%pD2\"", f->nf_file);
2742 }
2743
2744 static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2745 {
2746         struct inode *inode = file_inode(f->nf_file);
2747
2748         seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2749                                         MAJOR(inode->i_sb->s_dev),
2750                                          MINOR(inode->i_sb->s_dev),
2751                                          inode->i_ino);
2752 }
2753
2754 static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2755 {
2756         seq_puts(s, "owner: ");
2757         seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2758 }
2759
2760 static void nfs4_show_stateid(struct seq_file *s, stateid_t *stid)
2761 {
2762         seq_printf(s, "0x%.8x", stid->si_generation);
2763         seq_printf(s, "%12phN", &stid->si_opaque);
2764 }
2765
2766 static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2767 {
2768         struct nfs4_ol_stateid *ols;
2769         struct nfs4_file *nf;
2770         struct nfsd_file *file;
2771         struct nfs4_stateowner *oo;
2772         unsigned int access, deny;
2773
2774         ols = openlockstateid(st);
2775         oo = ols->st_stateowner;
2776         nf = st->sc_file;
2777
2778         seq_puts(s, "- ");
2779         nfs4_show_stateid(s, &st->sc_stateid);
2780         seq_puts(s, ": { type: open, ");
2781
2782         access = bmap_to_share_mode(ols->st_access_bmap);
2783         deny   = bmap_to_share_mode(ols->st_deny_bmap);
2784
2785         seq_printf(s, "access: %s%s, ",
2786                 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2787                 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2788         seq_printf(s, "deny: %s%s, ",
2789                 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2790                 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2791
2792         spin_lock(&nf->fi_lock);
2793         file = find_any_file_locked(nf);
2794         if (file) {
2795                 nfs4_show_superblock(s, file);
2796                 seq_puts(s, ", ");
2797                 nfs4_show_fname(s, file);
2798                 seq_puts(s, ", ");
2799         }
2800         spin_unlock(&nf->fi_lock);
2801         nfs4_show_owner(s, oo);
2802         if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2803                 seq_puts(s, ", admin-revoked");
2804         seq_puts(s, " }\n");
2805         return 0;
2806 }
2807
2808 static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2809 {
2810         struct nfs4_ol_stateid *ols;
2811         struct nfs4_file *nf;
2812         struct nfsd_file *file;
2813         struct nfs4_stateowner *oo;
2814
2815         ols = openlockstateid(st);
2816         oo = ols->st_stateowner;
2817         nf = st->sc_file;
2818
2819         seq_puts(s, "- ");
2820         nfs4_show_stateid(s, &st->sc_stateid);
2821         seq_puts(s, ": { type: lock, ");
2822
2823         spin_lock(&nf->fi_lock);
2824         file = find_any_file_locked(nf);
2825         if (file) {
2826                 /*
2827                  * Note: a lock stateid isn't really the same thing as a lock,
2828                  * it's the locking state held by one owner on a file, and there
2829                  * may be multiple (or no) lock ranges associated with it.
2830                  * (Same for the matter is true of open stateids.)
2831                  */
2832
2833                 nfs4_show_superblock(s, file);
2834                 /* XXX: open stateid? */
2835                 seq_puts(s, ", ");
2836                 nfs4_show_fname(s, file);
2837                 seq_puts(s, ", ");
2838         }
2839         nfs4_show_owner(s, oo);
2840         if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2841                 seq_puts(s, ", admin-revoked");
2842         seq_puts(s, " }\n");
2843         spin_unlock(&nf->fi_lock);
2844         return 0;
2845 }
2846
2847 static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2848 {
2849         struct nfs4_delegation *ds;
2850         struct nfs4_file *nf;
2851         struct nfsd_file *file;
2852
2853         ds = delegstateid(st);
2854         nf = st->sc_file;
2855
2856         seq_puts(s, "- ");
2857         nfs4_show_stateid(s, &st->sc_stateid);
2858         seq_puts(s, ": { type: deleg, ");
2859
2860         seq_printf(s, "access: %s",
2861                    ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2862
2863         /* XXX: lease time, whether it's being recalled. */
2864
2865         spin_lock(&nf->fi_lock);
2866         file = nf->fi_deleg_file;
2867         if (file) {
2868                 seq_puts(s, ", ");
2869                 nfs4_show_superblock(s, file);
2870                 seq_puts(s, ", ");
2871                 nfs4_show_fname(s, file);
2872         }
2873         spin_unlock(&nf->fi_lock);
2874         if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2875                 seq_puts(s, ", admin-revoked");
2876         seq_puts(s, " }\n");
2877         return 0;
2878 }
2879
2880 static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2881 {
2882         struct nfs4_layout_stateid *ls;
2883         struct nfsd_file *file;
2884
2885         ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2886
2887         seq_puts(s, "- ");
2888         nfs4_show_stateid(s, &st->sc_stateid);
2889         seq_puts(s, ": { type: layout");
2890
2891         /* XXX: What else would be useful? */
2892
2893         spin_lock(&ls->ls_stid.sc_file->fi_lock);
2894         file = ls->ls_file;
2895         if (file) {
2896                 seq_puts(s, ", ");
2897                 nfs4_show_superblock(s, file);
2898                 seq_puts(s, ", ");
2899                 nfs4_show_fname(s, file);
2900         }
2901         spin_unlock(&ls->ls_stid.sc_file->fi_lock);
2902         if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2903                 seq_puts(s, ", admin-revoked");
2904         seq_puts(s, " }\n");
2905
2906         return 0;
2907 }
2908
2909 static int states_show(struct seq_file *s, void *v)
2910 {
2911         struct nfs4_stid *st = v;
2912
2913         switch (st->sc_type) {
2914         case SC_TYPE_OPEN:
2915                 return nfs4_show_open(s, st);
2916         case SC_TYPE_LOCK:
2917                 return nfs4_show_lock(s, st);
2918         case SC_TYPE_DELEG:
2919                 return nfs4_show_deleg(s, st);
2920         case SC_TYPE_LAYOUT:
2921                 return nfs4_show_layout(s, st);
2922         default:
2923                 return 0; /* XXX: or SEQ_SKIP? */
2924         }
2925         /* XXX: copy stateids? */
2926 }
2927
2928 static struct seq_operations states_seq_ops = {
2929         .start = states_start,
2930         .next = states_next,
2931         .stop = states_stop,
2932         .show = states_show
2933 };
2934
2935 static int client_states_open(struct inode *inode, struct file *file)
2936 {
2937         struct seq_file *s;
2938         struct nfs4_client *clp;
2939         int ret;
2940
2941         clp = get_nfsdfs_clp(inode);
2942         if (!clp)
2943                 return -ENXIO;
2944
2945         ret = seq_open(file, &states_seq_ops);
2946         if (ret)
2947                 return ret;
2948         s = file->private_data;
2949         s->private = clp;
2950         return 0;
2951 }
2952
2953 static int client_opens_release(struct inode *inode, struct file *file)
2954 {
2955         struct seq_file *m = file->private_data;
2956         struct nfs4_client *clp = m->private;
2957
2958         /* XXX: alternatively, we could get/drop in seq start/stop */
2959         drop_client(clp);
2960         return seq_release(inode, file);
2961 }
2962
2963 static const struct file_operations client_states_fops = {
2964         .open           = client_states_open,
2965         .read           = seq_read,
2966         .llseek         = seq_lseek,
2967         .release        = client_opens_release,
2968 };
2969
2970 /*
2971  * Normally we refuse to destroy clients that are in use, but here the
2972  * administrator is telling us to just do it.  We also want to wait
2973  * so the caller has a guarantee that the client's locks are gone by
2974  * the time the write returns:
2975  */
2976 static void force_expire_client(struct nfs4_client *clp)
2977 {
2978         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2979         bool already_expired;
2980
2981         trace_nfsd_clid_admin_expired(&clp->cl_clientid);
2982
2983         spin_lock(&nn->client_lock);
2984         clp->cl_time = 0;
2985         spin_unlock(&nn->client_lock);
2986
2987         wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2988         spin_lock(&nn->client_lock);
2989         already_expired = list_empty(&clp->cl_lru);
2990         if (!already_expired)
2991                 unhash_client_locked(clp);
2992         spin_unlock(&nn->client_lock);
2993
2994         if (!already_expired)
2995                 expire_client(clp);
2996         else
2997                 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2998 }
2999
3000 static ssize_t client_ctl_write(struct file *file, const char __user *buf,
3001                                    size_t size, loff_t *pos)
3002 {
3003         char *data;
3004         struct nfs4_client *clp;
3005
3006         data = simple_transaction_get(file, buf, size);
3007         if (IS_ERR(data))
3008                 return PTR_ERR(data);
3009         if (size != 7 || 0 != memcmp(data, "expire\n", 7))
3010                 return -EINVAL;
3011         clp = get_nfsdfs_clp(file_inode(file));
3012         if (!clp)
3013                 return -ENXIO;
3014         force_expire_client(clp);
3015         drop_client(clp);
3016         return 7;
3017 }
3018
3019 static const struct file_operations client_ctl_fops = {
3020         .write          = client_ctl_write,
3021         .release        = simple_transaction_release,
3022 };
3023
3024 static const struct tree_descr client_files[] = {
3025         [0] = {"info", &client_info_fops, S_IRUSR},
3026         [1] = {"states", &client_states_fops, S_IRUSR},
3027         [2] = {"ctl", &client_ctl_fops, S_IWUSR},
3028         [3] = {""},
3029 };
3030
3031 static int
3032 nfsd4_cb_recall_any_done(struct nfsd4_callback *cb,
3033                                 struct rpc_task *task)
3034 {
3035         trace_nfsd_cb_recall_any_done(cb, task);
3036         switch (task->tk_status) {
3037         case -NFS4ERR_DELAY:
3038                 rpc_delay(task, 2 * HZ);
3039                 return 0;
3040         default:
3041                 return 1;
3042         }
3043 }
3044
3045 static void
3046 nfsd4_cb_recall_any_release(struct nfsd4_callback *cb)
3047 {
3048         struct nfs4_client *clp = cb->cb_clp;
3049
3050         clear_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
3051         drop_client(clp);
3052 }
3053
3054 static int
3055 nfsd4_cb_getattr_done(struct nfsd4_callback *cb, struct rpc_task *task)
3056 {
3057         struct nfs4_cb_fattr *ncf =
3058                         container_of(cb, struct nfs4_cb_fattr, ncf_getattr);
3059
3060         ncf->ncf_cb_status = task->tk_status;
3061         switch (task->tk_status) {
3062         case -NFS4ERR_DELAY:
3063                 rpc_delay(task, 2 * HZ);
3064                 return 0;
3065         default:
3066                 return 1;
3067         }
3068 }
3069
3070 static void
3071 nfsd4_cb_getattr_release(struct nfsd4_callback *cb)
3072 {
3073         struct nfs4_cb_fattr *ncf =
3074                         container_of(cb, struct nfs4_cb_fattr, ncf_getattr);
3075         struct nfs4_delegation *dp =
3076                         container_of(ncf, struct nfs4_delegation, dl_cb_fattr);
3077
3078         nfs4_put_stid(&dp->dl_stid);
3079         clear_bit(CB_GETATTR_BUSY, &ncf->ncf_cb_flags);
3080         wake_up_bit(&ncf->ncf_cb_flags, CB_GETATTR_BUSY);
3081 }
3082
3083 static const struct nfsd4_callback_ops nfsd4_cb_recall_any_ops = {
3084         .done           = nfsd4_cb_recall_any_done,
3085         .release        = nfsd4_cb_recall_any_release,
3086 };
3087
3088 static const struct nfsd4_callback_ops nfsd4_cb_getattr_ops = {
3089         .done           = nfsd4_cb_getattr_done,
3090         .release        = nfsd4_cb_getattr_release,
3091 };
3092
3093 static void nfs4_cb_getattr(struct nfs4_cb_fattr *ncf)
3094 {
3095         struct nfs4_delegation *dp =
3096                         container_of(ncf, struct nfs4_delegation, dl_cb_fattr);
3097
3098         if (test_and_set_bit(CB_GETATTR_BUSY, &ncf->ncf_cb_flags))
3099                 return;
3100         /* set to proper status when nfsd4_cb_getattr_done runs */
3101         ncf->ncf_cb_status = NFS4ERR_IO;
3102
3103         refcount_inc(&dp->dl_stid.sc_count);
3104         nfsd4_run_cb(&ncf->ncf_getattr);
3105 }
3106
3107 static struct nfs4_client *create_client(struct xdr_netobj name,
3108                 struct svc_rqst *rqstp, nfs4_verifier *verf)
3109 {
3110         struct nfs4_client *clp;
3111         struct sockaddr *sa = svc_addr(rqstp);
3112         int ret;
3113         struct net *net = SVC_NET(rqstp);
3114         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3115         struct dentry *dentries[ARRAY_SIZE(client_files)];
3116
3117         clp = alloc_client(name, nn);
3118         if (clp == NULL)
3119                 return NULL;
3120
3121         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
3122         if (ret) {
3123                 free_client(clp);
3124                 return NULL;
3125         }
3126         gen_clid(clp, nn);
3127         kref_init(&clp->cl_nfsdfs.cl_ref);
3128         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
3129         clp->cl_time = ktime_get_boottime_seconds();
3130         clear_bit(0, &clp->cl_cb_slot_busy);
3131         copy_verf(clp, verf);
3132         memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
3133         clp->cl_cb_session = NULL;
3134         clp->net = net;
3135         clp->cl_nfsd_dentry = nfsd_client_mkdir(
3136                 nn, &clp->cl_nfsdfs,
3137                 clp->cl_clientid.cl_id - nn->clientid_base,
3138                 client_files, dentries);
3139         clp->cl_nfsd_info_dentry = dentries[0];
3140         if (!clp->cl_nfsd_dentry) {
3141                 free_client(clp);
3142                 return NULL;
3143         }
3144         clp->cl_ra = kzalloc(sizeof(*clp->cl_ra), GFP_KERNEL);
3145         if (!clp->cl_ra) {
3146                 free_client(clp);
3147                 return NULL;
3148         }
3149         clp->cl_ra_time = 0;
3150         nfsd4_init_cb(&clp->cl_ra->ra_cb, clp, &nfsd4_cb_recall_any_ops,
3151                         NFSPROC4_CLNT_CB_RECALL_ANY);
3152         return clp;
3153 }
3154
3155 static void
3156 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
3157 {
3158         struct rb_node **new = &(root->rb_node), *parent = NULL;
3159         struct nfs4_client *clp;
3160
3161         while (*new) {
3162                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
3163                 parent = *new;
3164
3165                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
3166                         new = &((*new)->rb_left);
3167                 else
3168                         new = &((*new)->rb_right);
3169         }
3170
3171         rb_link_node(&new_clp->cl_namenode, parent, new);
3172         rb_insert_color(&new_clp->cl_namenode, root);
3173 }
3174
3175 static struct nfs4_client *
3176 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
3177 {
3178         int cmp;
3179         struct rb_node *node = root->rb_node;
3180         struct nfs4_client *clp;
3181
3182         while (node) {
3183                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
3184                 cmp = compare_blob(&clp->cl_name, name);
3185                 if (cmp > 0)
3186                         node = node->rb_left;
3187                 else if (cmp < 0)
3188                         node = node->rb_right;
3189                 else
3190                         return clp;
3191         }
3192         return NULL;
3193 }
3194
3195 static void
3196 add_to_unconfirmed(struct nfs4_client *clp)
3197 {
3198         unsigned int idhashval;
3199         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3200
3201         lockdep_assert_held(&nn->client_lock);
3202
3203         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3204         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
3205         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3206         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
3207         renew_client_locked(clp);
3208 }
3209
3210 static void
3211 move_to_confirmed(struct nfs4_client *clp)
3212 {
3213         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3214         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3215
3216         lockdep_assert_held(&nn->client_lock);
3217
3218         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
3219         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
3220         add_clp_to_name_tree(clp, &nn->conf_name_tree);
3221         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3222         trace_nfsd_clid_confirmed(&clp->cl_clientid);
3223         renew_client_locked(clp);
3224 }
3225
3226 static struct nfs4_client *
3227 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
3228 {
3229         struct nfs4_client *clp;
3230         unsigned int idhashval = clientid_hashval(clid->cl_id);
3231
3232         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
3233                 if (same_clid(&clp->cl_clientid, clid)) {
3234                         if ((bool)clp->cl_minorversion != sessions)
3235                                 return NULL;
3236                         renew_client_locked(clp);
3237                         return clp;
3238                 }
3239         }
3240         return NULL;
3241 }
3242
3243 static struct nfs4_client *
3244 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3245 {
3246         struct list_head *tbl = nn->conf_id_hashtbl;
3247
3248         lockdep_assert_held(&nn->client_lock);
3249         return find_client_in_id_table(tbl, clid, sessions);
3250 }
3251
3252 static struct nfs4_client *
3253 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3254 {
3255         struct list_head *tbl = nn->unconf_id_hashtbl;
3256
3257         lockdep_assert_held(&nn->client_lock);
3258         return find_client_in_id_table(tbl, clid, sessions);
3259 }
3260
3261 static bool clp_used_exchangeid(struct nfs4_client *clp)
3262 {
3263         return clp->cl_exchange_flags != 0;
3264 }
3265
3266 static struct nfs4_client *
3267 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3268 {
3269         lockdep_assert_held(&nn->client_lock);
3270         return find_clp_in_name_tree(name, &nn->conf_name_tree);
3271 }
3272
3273 static struct nfs4_client *
3274 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3275 {
3276         lockdep_assert_held(&nn->client_lock);
3277         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
3278 }
3279
3280 static void
3281 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
3282 {
3283         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
3284         struct sockaddr *sa = svc_addr(rqstp);
3285         u32 scopeid = rpc_get_scope_id(sa);
3286         unsigned short expected_family;
3287
3288         /* Currently, we only support tcp and tcp6 for the callback channel */
3289         if (se->se_callback_netid_len == 3 &&
3290             !memcmp(se->se_callback_netid_val, "tcp", 3))
3291                 expected_family = AF_INET;
3292         else if (se->se_callback_netid_len == 4 &&
3293                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
3294                 expected_family = AF_INET6;
3295         else
3296                 goto out_err;
3297
3298         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
3299                                             se->se_callback_addr_len,
3300                                             (struct sockaddr *)&conn->cb_addr,
3301                                             sizeof(conn->cb_addr));
3302
3303         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
3304                 goto out_err;
3305
3306         if (conn->cb_addr.ss_family == AF_INET6)
3307                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
3308
3309         conn->cb_prog = se->se_callback_prog;
3310         conn->cb_ident = se->se_callback_ident;
3311         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
3312         trace_nfsd_cb_args(clp, conn);
3313         return;
3314 out_err:
3315         conn->cb_addr.ss_family = AF_UNSPEC;
3316         conn->cb_addrlen = 0;
3317         trace_nfsd_cb_nodelegs(clp);
3318         return;
3319 }
3320
3321 /*
3322  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
3323  */
3324 static void
3325 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
3326 {
3327         struct xdr_buf *buf = resp->xdr->buf;
3328         struct nfsd4_slot *slot = resp->cstate.slot;
3329         unsigned int base;
3330
3331         dprintk("--> %s slot %p\n", __func__, slot);
3332
3333         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
3334         slot->sl_opcnt = resp->opcnt;
3335         slot->sl_status = resp->cstate.status;
3336         free_svc_cred(&slot->sl_cred);
3337         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
3338
3339         if (!nfsd4_cache_this(resp)) {
3340                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
3341                 return;
3342         }
3343         slot->sl_flags |= NFSD4_SLOT_CACHED;
3344
3345         base = resp->cstate.data_offset;
3346         slot->sl_datalen = buf->len - base;
3347         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
3348                 WARN(1, "%s: sessions DRC could not cache compound\n",
3349                      __func__);
3350         return;
3351 }
3352
3353 /*
3354  * Encode the replay sequence operation from the slot values.
3355  * If cachethis is FALSE encode the uncached rep error on the next
3356  * operation which sets resp->p and increments resp->opcnt for
3357  * nfs4svc_encode_compoundres.
3358  *
3359  */
3360 static __be32
3361 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
3362                           struct nfsd4_compoundres *resp)
3363 {
3364         struct nfsd4_op *op;
3365         struct nfsd4_slot *slot = resp->cstate.slot;
3366
3367         /* Encode the replayed sequence operation */
3368         op = &args->ops[resp->opcnt - 1];
3369         nfsd4_encode_operation(resp, op);
3370
3371         if (slot->sl_flags & NFSD4_SLOT_CACHED)
3372                 return op->status;
3373         if (args->opcnt == 1) {
3374                 /*
3375                  * The original operation wasn't a solo sequence--we
3376                  * always cache those--so this retry must not match the
3377                  * original:
3378                  */
3379                 op->status = nfserr_seq_false_retry;
3380         } else {
3381                 op = &args->ops[resp->opcnt++];
3382                 op->status = nfserr_retry_uncached_rep;
3383                 nfsd4_encode_operation(resp, op);
3384         }
3385         return op->status;
3386 }
3387
3388 /*
3389  * The sequence operation is not cached because we can use the slot and
3390  * session values.
3391  */
3392 static __be32
3393 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
3394                          struct nfsd4_sequence *seq)
3395 {
3396         struct nfsd4_slot *slot = resp->cstate.slot;
3397         struct xdr_stream *xdr = resp->xdr;
3398         __be32 *p;
3399         __be32 status;
3400
3401         dprintk("--> %s slot %p\n", __func__, slot);
3402
3403         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
3404         if (status)
3405                 return status;
3406
3407         p = xdr_reserve_space(xdr, slot->sl_datalen);
3408         if (!p) {
3409                 WARN_ON_ONCE(1);
3410                 return nfserr_serverfault;
3411         }
3412         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
3413         xdr_commit_encode(xdr);
3414
3415         resp->opcnt = slot->sl_opcnt;
3416         return slot->sl_status;
3417 }
3418
3419 /*
3420  * Set the exchange_id flags returned by the server.
3421  */
3422 static void
3423 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
3424 {
3425 #ifdef CONFIG_NFSD_PNFS
3426         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
3427 #else
3428         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
3429 #endif
3430
3431         /* Referrals are supported, Migration is not. */
3432         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
3433
3434         /* set the wire flags to return to client. */
3435         clid->flags = new->cl_exchange_flags;
3436 }
3437
3438 static bool client_has_openowners(struct nfs4_client *clp)
3439 {
3440         struct nfs4_openowner *oo;
3441
3442         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
3443                 if (!list_empty(&oo->oo_owner.so_stateids))
3444                         return true;
3445         }
3446         return false;
3447 }
3448
3449 static bool client_has_state(struct nfs4_client *clp)
3450 {
3451         return client_has_openowners(clp)
3452 #ifdef CONFIG_NFSD_PNFS
3453                 || !list_empty(&clp->cl_lo_states)
3454 #endif
3455                 || !list_empty(&clp->cl_delegations)
3456                 || !list_empty(&clp->cl_sessions)
3457                 || !list_empty(&clp->async_copies);
3458 }
3459
3460 static __be32 copy_impl_id(struct nfs4_client *clp,
3461                                 struct nfsd4_exchange_id *exid)
3462 {
3463         if (!exid->nii_domain.data)
3464                 return 0;
3465         xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
3466         if (!clp->cl_nii_domain.data)
3467                 return nfserr_jukebox;
3468         xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
3469         if (!clp->cl_nii_name.data)
3470                 return nfserr_jukebox;
3471         clp->cl_nii_time = exid->nii_time;
3472         return 0;
3473 }
3474
3475 __be32
3476 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3477                 union nfsd4_op_u *u)
3478 {
3479         struct nfsd4_exchange_id *exid = &u->exchange_id;
3480         struct nfs4_client *conf, *new;
3481         struct nfs4_client *unconf = NULL;
3482         __be32 status;
3483         char                    addr_str[INET6_ADDRSTRLEN];
3484         nfs4_verifier           verf = exid->verifier;
3485         struct sockaddr         *sa = svc_addr(rqstp);
3486         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3487         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3488
3489         rpc_ntop(sa, addr_str, sizeof(addr_str));
3490         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3491                 "ip_addr=%s flags %x, spa_how %u\n",
3492                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
3493                 addr_str, exid->flags, exid->spa_how);
3494
3495         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3496                 return nfserr_inval;
3497
3498         new = create_client(exid->clname, rqstp, &verf);
3499         if (new == NULL)
3500                 return nfserr_jukebox;
3501         status = copy_impl_id(new, exid);
3502         if (status)
3503                 goto out_nolock;
3504
3505         switch (exid->spa_how) {
3506         case SP4_MACH_CRED:
3507                 exid->spo_must_enforce[0] = 0;
3508                 exid->spo_must_enforce[1] = (
3509                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
3510                         1 << (OP_EXCHANGE_ID - 32) |
3511                         1 << (OP_CREATE_SESSION - 32) |
3512                         1 << (OP_DESTROY_SESSION - 32) |
3513                         1 << (OP_DESTROY_CLIENTID - 32));
3514
3515                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3516                                         1 << (OP_OPEN_DOWNGRADE) |
3517                                         1 << (OP_LOCKU) |
3518                                         1 << (OP_DELEGRETURN));
3519
3520                 exid->spo_must_allow[1] &= (
3521                                         1 << (OP_TEST_STATEID - 32) |
3522                                         1 << (OP_FREE_STATEID - 32));
3523                 if (!svc_rqst_integrity_protected(rqstp)) {
3524                         status = nfserr_inval;
3525                         goto out_nolock;
3526                 }
3527                 /*
3528                  * Sometimes userspace doesn't give us a principal.
3529                  * Which is a bug, really.  Anyway, we can't enforce
3530                  * MACH_CRED in that case, better to give up now:
3531                  */
3532                 if (!new->cl_cred.cr_principal &&
3533                                         !new->cl_cred.cr_raw_principal) {
3534                         status = nfserr_serverfault;
3535                         goto out_nolock;
3536                 }
3537                 new->cl_mach_cred = true;
3538                 break;
3539         case SP4_NONE:
3540                 break;
3541         default:                                /* checked by xdr code */
3542                 WARN_ON_ONCE(1);
3543                 fallthrough;
3544         case SP4_SSV:
3545                 status = nfserr_encr_alg_unsupp;
3546                 goto out_nolock;
3547         }
3548
3549         /* Cases below refer to rfc 5661 section 18.35.4: */
3550         spin_lock(&nn->client_lock);
3551         conf = find_confirmed_client_by_name(&exid->clname, nn);
3552         if (conf) {
3553                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3554                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3555
3556                 if (update) {
3557                         if (!clp_used_exchangeid(conf)) { /* buggy client */
3558                                 status = nfserr_inval;
3559                                 goto out;
3560                         }
3561                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
3562                                 status = nfserr_wrong_cred;
3563                                 goto out;
3564                         }
3565                         if (!creds_match) { /* case 9 */
3566                                 status = nfserr_perm;
3567                                 goto out;
3568                         }
3569                         if (!verfs_match) { /* case 8 */
3570                                 status = nfserr_not_same;
3571                                 goto out;
3572                         }
3573                         /* case 6 */
3574                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3575                         trace_nfsd_clid_confirmed_r(conf);
3576                         goto out_copy;
3577                 }
3578                 if (!creds_match) { /* case 3 */
3579                         if (client_has_state(conf)) {
3580                                 status = nfserr_clid_inuse;
3581                                 trace_nfsd_clid_cred_mismatch(conf, rqstp);
3582                                 goto out;
3583                         }
3584                         goto out_new;
3585                 }
3586                 if (verfs_match) { /* case 2 */
3587                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3588                         trace_nfsd_clid_confirmed_r(conf);
3589                         goto out_copy;
3590                 }
3591                 /* case 5, client reboot */
3592                 trace_nfsd_clid_verf_mismatch(conf, rqstp, &verf);
3593                 conf = NULL;
3594                 goto out_new;
3595         }
3596
3597         if (update) { /* case 7 */
3598                 status = nfserr_noent;
3599                 goto out;
3600         }
3601
3602         unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
3603         if (unconf) /* case 4, possible retry or client restart */
3604                 unhash_client_locked(unconf);
3605
3606         /* case 1, new owner ID */
3607         trace_nfsd_clid_fresh(new);
3608
3609 out_new:
3610         if (conf) {
3611                 status = mark_client_expired_locked(conf);
3612                 if (status)
3613                         goto out;
3614                 trace_nfsd_clid_replaced(&conf->cl_clientid);
3615         }
3616         new->cl_minorversion = cstate->minorversion;
3617         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3618         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3619
3620         /* Contrived initial CREATE_SESSION response */
3621         new->cl_cs_slot.sl_status = nfserr_seq_misordered;
3622
3623         add_to_unconfirmed(new);
3624         swap(new, conf);
3625 out_copy:
3626         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3627         exid->clientid.cl_id = conf->cl_clientid.cl_id;
3628
3629         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3630         nfsd4_set_ex_flags(conf, exid);
3631
3632         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3633                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3634         status = nfs_ok;
3635
3636 out:
3637         spin_unlock(&nn->client_lock);
3638 out_nolock:
3639         if (new)
3640                 expire_client(new);
3641         if (unconf) {
3642                 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
3643                 expire_client(unconf);
3644         }
3645         return status;
3646 }
3647
3648 static __be32 check_slot_seqid(u32 seqid, u32 slot_seqid, bool slot_inuse)
3649 {
3650         /* The slot is in use, and no response has been sent. */
3651         if (slot_inuse) {
3652                 if (seqid == slot_seqid)
3653                         return nfserr_jukebox;
3654                 else
3655                         return nfserr_seq_misordered;
3656         }
3657         /* Note unsigned 32-bit arithmetic handles wraparound: */
3658         if (likely(seqid == slot_seqid + 1))
3659                 return nfs_ok;
3660         if (seqid == slot_seqid)
3661                 return nfserr_replay_cache;
3662         return nfserr_seq_misordered;
3663 }
3664
3665 /*
3666  * Cache the create session result into the create session single DRC
3667  * slot cache by saving the xdr structure. sl_seqid has been set.
3668  * Do this for solo or embedded create session operations.
3669  */
3670 static void
3671 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3672                            struct nfsd4_clid_slot *slot, __be32 nfserr)
3673 {
3674         slot->sl_status = nfserr;
3675         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3676 }
3677
3678 static __be32
3679 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3680                             struct nfsd4_clid_slot *slot)
3681 {
3682         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3683         return slot->sl_status;
3684 }
3685
3686 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3687                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3688                         1 +     /* MIN tag is length with zero, only length */ \
3689                         3 +     /* version, opcount, opcode */ \
3690                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3691                                 /* seqid, slotID, slotID, cache */ \
3692                         4 ) * sizeof(__be32))
3693
3694 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3695                         2 +     /* verifier: AUTH_NULL, length 0 */\
3696                         1 +     /* status */ \
3697                         1 +     /* MIN tag is length with zero, only length */ \
3698                         3 +     /* opcount, opcode, opstatus*/ \
3699                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3700                                 /* seqid, slotID, slotID, slotID, status */ \
3701                         5 ) * sizeof(__be32))
3702
3703 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3704 {
3705         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3706
3707         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3708                 return nfserr_toosmall;
3709         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3710                 return nfserr_toosmall;
3711         ca->headerpadsz = 0;
3712         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3713         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3714         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3715         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3716                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3717         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3718         /*
3719          * Note decreasing slot size below client's request may make it
3720          * difficult for client to function correctly, whereas
3721          * decreasing the number of slots will (just?) affect
3722          * performance.  When short on memory we therefore prefer to
3723          * decrease number of slots instead of their size.  Clients that
3724          * request larger slots than they need will get poor results:
3725          * Note that we always allow at least one slot, because our
3726          * accounting is soft and provides no guarantees either way.
3727          */
3728         ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3729
3730         return nfs_ok;
3731 }
3732
3733 /*
3734  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3735  * These are based on similar macros in linux/sunrpc/msg_prot.h .
3736  */
3737 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3738         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3739
3740 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3741         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3742
3743 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
3744                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3745 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
3746                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3747                                  sizeof(__be32))
3748
3749 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3750 {
3751         ca->headerpadsz = 0;
3752
3753         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3754                 return nfserr_toosmall;
3755         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3756                 return nfserr_toosmall;
3757         ca->maxresp_cached = 0;
3758         if (ca->maxops < 2)
3759                 return nfserr_toosmall;
3760
3761         return nfs_ok;
3762 }
3763
3764 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3765 {
3766         switch (cbs->flavor) {
3767         case RPC_AUTH_NULL:
3768         case RPC_AUTH_UNIX:
3769                 return nfs_ok;
3770         default:
3771                 /*
3772                  * GSS case: the spec doesn't allow us to return this
3773                  * error.  But it also doesn't allow us not to support
3774                  * GSS.
3775                  * I'd rather this fail hard than return some error the
3776                  * client might think it can already handle:
3777                  */
3778                 return nfserr_encr_alg_unsupp;
3779         }
3780 }
3781
3782 __be32
3783 nfsd4_create_session(struct svc_rqst *rqstp,
3784                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3785 {
3786         struct nfsd4_create_session *cr_ses = &u->create_session;
3787         struct sockaddr *sa = svc_addr(rqstp);
3788         struct nfs4_client *conf, *unconf;
3789         struct nfsd4_clid_slot *cs_slot;
3790         struct nfs4_client *old = NULL;
3791         struct nfsd4_session *new;
3792         struct nfsd4_conn *conn;
3793         __be32 status = 0;
3794         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3795
3796         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3797                 return nfserr_inval;
3798         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3799         if (status)
3800                 return status;
3801         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3802         if (status)
3803                 return status;
3804         status = check_backchannel_attrs(&cr_ses->back_channel);
3805         if (status)
3806                 goto out_release_drc_mem;
3807         status = nfserr_jukebox;
3808         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3809         if (!new)
3810                 goto out_release_drc_mem;
3811         conn = alloc_conn_from_crses(rqstp, cr_ses);
3812         if (!conn)
3813                 goto out_free_session;
3814
3815         spin_lock(&nn->client_lock);
3816
3817         /* RFC 8881 Section 18.36.4 Phase 1: Client record look-up. */
3818         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3819         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3820         if (!conf && !unconf) {
3821                 status = nfserr_stale_clientid;
3822                 goto out_free_conn;
3823         }
3824
3825         /* RFC 8881 Section 18.36.4 Phase 2: Sequence ID processing. */
3826         if (conf) {
3827                 cs_slot = &conf->cl_cs_slot;
3828                 trace_nfsd_slot_seqid_conf(conf, cr_ses);
3829         } else {
3830                 cs_slot = &unconf->cl_cs_slot;
3831                 trace_nfsd_slot_seqid_unconf(unconf, cr_ses);
3832         }
3833         status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3834         switch (status) {
3835         case nfs_ok:
3836                 cs_slot->sl_seqid++;
3837                 cr_ses->seqid = cs_slot->sl_seqid;
3838                 break;
3839         case nfserr_replay_cache:
3840                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3841                 fallthrough;
3842         case nfserr_jukebox:
3843                 /* The server MUST NOT cache NFS4ERR_DELAY */
3844                 goto out_free_conn;
3845         default:
3846                 goto out_cache_error;
3847         }
3848
3849         /* RFC 8881 Section 18.36.4 Phase 3: Client ID confirmation. */
3850         if (conf) {
3851                 status = nfserr_wrong_cred;
3852                 if (!nfsd4_mach_creds_match(conf, rqstp))
3853                         goto out_cache_error;
3854         } else {
3855                 status = nfserr_clid_inuse;
3856                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3857                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3858                         trace_nfsd_clid_cred_mismatch(unconf, rqstp);
3859                         goto out_cache_error;
3860                 }
3861                 status = nfserr_wrong_cred;
3862                 if (!nfsd4_mach_creds_match(unconf, rqstp))
3863                         goto out_cache_error;
3864                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3865                 if (old) {
3866                         status = mark_client_expired_locked(old);
3867                         if (status)
3868                                 goto out_expired_error;
3869                         trace_nfsd_clid_replaced(&old->cl_clientid);
3870                 }
3871                 move_to_confirmed(unconf);
3872                 conf = unconf;
3873         }
3874
3875         /* RFC 8881 Section 18.36.4 Phase 4: Session creation. */
3876         status = nfs_ok;
3877         /* Persistent sessions are not supported */
3878         cr_ses->flags &= ~SESSION4_PERSIST;
3879         /* Upshifting from TCP to RDMA is not supported */
3880         cr_ses->flags &= ~SESSION4_RDMA;
3881
3882         init_session(rqstp, new, conf, cr_ses);
3883         nfsd4_get_session_locked(new);
3884
3885         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3886                NFS4_MAX_SESSIONID_LEN);
3887
3888         /* cache solo and embedded create sessions under the client_lock */
3889         nfsd4_cache_create_session(cr_ses, cs_slot, status);
3890         spin_unlock(&nn->client_lock);
3891         if (conf == unconf)
3892                 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
3893         /* init connection and backchannel */
3894         nfsd4_init_conn(rqstp, conn, new);
3895         nfsd4_put_session(new);
3896         if (old)
3897                 expire_client(old);
3898         return status;
3899
3900 out_expired_error:
3901         old = NULL;
3902         /*
3903          * Revert the slot seq_nr change so the server will process
3904          * the client's resend instead of returning a cached response.
3905          */
3906         if (status == nfserr_jukebox) {
3907                 cs_slot->sl_seqid--;
3908                 cr_ses->seqid = cs_slot->sl_seqid;
3909                 goto out_free_conn;
3910         }
3911 out_cache_error:
3912         nfsd4_cache_create_session(cr_ses, cs_slot, status);
3913 out_free_conn:
3914         spin_unlock(&nn->client_lock);
3915         free_conn(conn);
3916         if (old)
3917                 expire_client(old);
3918 out_free_session:
3919         __free_session(new);
3920 out_release_drc_mem:
3921         nfsd4_put_drc_mem(&cr_ses->fore_channel);
3922         return status;
3923 }
3924
3925 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3926 {
3927         switch (*dir) {
3928         case NFS4_CDFC4_FORE:
3929         case NFS4_CDFC4_BACK:
3930                 return nfs_ok;
3931         case NFS4_CDFC4_FORE_OR_BOTH:
3932         case NFS4_CDFC4_BACK_OR_BOTH:
3933                 *dir = NFS4_CDFC4_BOTH;
3934                 return nfs_ok;
3935         }
3936         return nfserr_inval;
3937 }
3938
3939 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3940                 struct nfsd4_compound_state *cstate,
3941                 union nfsd4_op_u *u)
3942 {
3943         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3944         struct nfsd4_session *session = cstate->session;
3945         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3946         __be32 status;
3947
3948         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3949         if (status)
3950                 return status;
3951         spin_lock(&nn->client_lock);
3952         session->se_cb_prog = bc->bc_cb_program;
3953         session->se_cb_sec = bc->bc_cb_sec;
3954         spin_unlock(&nn->client_lock);
3955
3956         nfsd4_probe_callback(session->se_client);
3957
3958         return nfs_ok;
3959 }
3960
3961 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3962 {
3963         struct nfsd4_conn *c;
3964
3965         list_for_each_entry(c, &s->se_conns, cn_persession) {
3966                 if (c->cn_xprt == xpt) {
3967                         return c;
3968                 }
3969         }
3970         return NULL;
3971 }
3972
3973 static __be32 nfsd4_match_existing_connection(struct svc_rqst *rqst,
3974                 struct nfsd4_session *session, u32 req, struct nfsd4_conn **conn)
3975 {
3976         struct nfs4_client *clp = session->se_client;
3977         struct svc_xprt *xpt = rqst->rq_xprt;
3978         struct nfsd4_conn *c;
3979         __be32 status;
3980
3981         /* Following the last paragraph of RFC 5661 Section 18.34.3: */
3982         spin_lock(&clp->cl_lock);
3983         c = __nfsd4_find_conn(xpt, session);
3984         if (!c)
3985                 status = nfserr_noent;
3986         else if (req == c->cn_flags)
3987                 status = nfs_ok;
3988         else if (req == NFS4_CDFC4_FORE_OR_BOTH &&
3989                                 c->cn_flags != NFS4_CDFC4_BACK)
3990                 status = nfs_ok;
3991         else if (req == NFS4_CDFC4_BACK_OR_BOTH &&
3992                                 c->cn_flags != NFS4_CDFC4_FORE)
3993                 status = nfs_ok;
3994         else
3995                 status = nfserr_inval;
3996         spin_unlock(&clp->cl_lock);
3997         if (status == nfs_ok && conn)
3998                 *conn = c;
3999         return status;
4000 }
4001
4002 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
4003                      struct nfsd4_compound_state *cstate,
4004                      union nfsd4_op_u *u)
4005 {
4006         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
4007         __be32 status;
4008         struct nfsd4_conn *conn;
4009         struct nfsd4_session *session;
4010         struct net *net = SVC_NET(rqstp);
4011         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4012
4013         if (!nfsd4_last_compound_op(rqstp))
4014                 return nfserr_not_only_op;
4015         spin_lock(&nn->client_lock);
4016         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
4017         spin_unlock(&nn->client_lock);
4018         if (!session)
4019                 goto out_no_session;
4020         status = nfserr_wrong_cred;
4021         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
4022                 goto out;
4023         status = nfsd4_match_existing_connection(rqstp, session,
4024                         bcts->dir, &conn);
4025         if (status == nfs_ok) {
4026                 if (bcts->dir == NFS4_CDFC4_FORE_OR_BOTH ||
4027                                 bcts->dir == NFS4_CDFC4_BACK)
4028                         conn->cn_flags |= NFS4_CDFC4_BACK;
4029                 nfsd4_probe_callback(session->se_client);
4030                 goto out;
4031         }
4032         if (status == nfserr_inval)
4033                 goto out;
4034         status = nfsd4_map_bcts_dir(&bcts->dir);
4035         if (status)
4036                 goto out;
4037         conn = alloc_conn(rqstp, bcts->dir);
4038         status = nfserr_jukebox;
4039         if (!conn)
4040                 goto out;
4041         nfsd4_init_conn(rqstp, conn, session);
4042         status = nfs_ok;
4043 out:
4044         nfsd4_put_session(session);
4045 out_no_session:
4046         return status;
4047 }
4048
4049 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
4050 {
4051         if (!cstate->session)
4052                 return false;
4053         return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
4054 }
4055
4056 __be32
4057 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
4058                 union nfsd4_op_u *u)
4059 {
4060         struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
4061         struct nfsd4_session *ses;
4062         __be32 status;
4063         int ref_held_by_me = 0;
4064         struct net *net = SVC_NET(r);
4065         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4066
4067         status = nfserr_not_only_op;
4068         if (nfsd4_compound_in_session(cstate, sessionid)) {
4069                 if (!nfsd4_last_compound_op(r))
4070                         goto out;
4071                 ref_held_by_me++;
4072         }
4073         dump_sessionid(__func__, sessionid);
4074         spin_lock(&nn->client_lock);
4075         ses = find_in_sessionid_hashtbl(sessionid, net, &status);
4076         if (!ses)
4077                 goto out_client_lock;
4078         status = nfserr_wrong_cred;
4079         if (!nfsd4_mach_creds_match(ses->se_client, r))
4080                 goto out_put_session;
4081         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
4082         if (status)
4083                 goto out_put_session;
4084         unhash_session(ses);
4085         spin_unlock(&nn->client_lock);
4086
4087         nfsd4_probe_callback_sync(ses->se_client);
4088
4089         spin_lock(&nn->client_lock);
4090         status = nfs_ok;
4091 out_put_session:
4092         nfsd4_put_session_locked(ses);
4093 out_client_lock:
4094         spin_unlock(&nn->client_lock);
4095 out:
4096         return status;
4097 }
4098
4099 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
4100 {
4101         struct nfs4_client *clp = ses->se_client;
4102         struct nfsd4_conn *c;
4103         __be32 status = nfs_ok;
4104         int ret;
4105
4106         spin_lock(&clp->cl_lock);
4107         c = __nfsd4_find_conn(new->cn_xprt, ses);
4108         if (c)
4109                 goto out_free;
4110         status = nfserr_conn_not_bound_to_session;
4111         if (clp->cl_mach_cred)
4112                 goto out_free;
4113         __nfsd4_hash_conn(new, ses);
4114         spin_unlock(&clp->cl_lock);
4115         ret = nfsd4_register_conn(new);
4116         if (ret)
4117                 /* oops; xprt is already down: */
4118                 nfsd4_conn_lost(&new->cn_xpt_user);
4119         return nfs_ok;
4120 out_free:
4121         spin_unlock(&clp->cl_lock);
4122         free_conn(new);
4123         return status;
4124 }
4125
4126 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
4127 {
4128         struct nfsd4_compoundargs *args = rqstp->rq_argp;
4129
4130         return args->opcnt > session->se_fchannel.maxops;
4131 }
4132
4133 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
4134                                   struct nfsd4_session *session)
4135 {
4136         struct xdr_buf *xb = &rqstp->rq_arg;
4137
4138         return xb->len > session->se_fchannel.maxreq_sz;
4139 }
4140
4141 static bool replay_matches_cache(struct svc_rqst *rqstp,
4142                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
4143 {
4144         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
4145
4146         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
4147             (bool)seq->cachethis)
4148                 return false;
4149         /*
4150          * If there's an error then the reply can have fewer ops than
4151          * the call.
4152          */
4153         if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
4154                 return false;
4155         /*
4156          * But if we cached a reply with *more* ops than the call you're
4157          * sending us now, then this new call is clearly not really a
4158          * replay of the old one:
4159          */
4160         if (slot->sl_opcnt > argp->opcnt)
4161                 return false;
4162         /* This is the only check explicitly called by spec: */
4163         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
4164                 return false;
4165         /*
4166          * There may be more comparisons we could actually do, but the
4167          * spec doesn't require us to catch every case where the calls
4168          * don't match (that would require caching the call as well as
4169          * the reply), so we don't bother.
4170          */
4171         return true;
4172 }
4173
4174 __be32
4175 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4176                 union nfsd4_op_u *u)
4177 {
4178         struct nfsd4_sequence *seq = &u->sequence;
4179         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4180         struct xdr_stream *xdr = resp->xdr;
4181         struct nfsd4_session *session;
4182         struct nfs4_client *clp;
4183         struct nfsd4_slot *slot;
4184         struct nfsd4_conn *conn;
4185         __be32 status;
4186         int buflen;
4187         struct net *net = SVC_NET(rqstp);
4188         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4189
4190         if (resp->opcnt != 1)
4191                 return nfserr_sequence_pos;
4192
4193         /*
4194          * Will be either used or freed by nfsd4_sequence_check_conn
4195          * below.
4196          */
4197         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
4198         if (!conn)
4199                 return nfserr_jukebox;
4200
4201         spin_lock(&nn->client_lock);
4202         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
4203         if (!session)
4204                 goto out_no_session;
4205         clp = session->se_client;
4206
4207         status = nfserr_too_many_ops;
4208         if (nfsd4_session_too_many_ops(rqstp, session))
4209                 goto out_put_session;
4210
4211         status = nfserr_req_too_big;
4212         if (nfsd4_request_too_big(rqstp, session))
4213                 goto out_put_session;
4214
4215         status = nfserr_badslot;
4216         if (seq->slotid >= session->se_fchannel.maxreqs)
4217                 goto out_put_session;
4218
4219         slot = session->se_slots[seq->slotid];
4220         dprintk("%s: slotid %d\n", __func__, seq->slotid);
4221
4222         /* We do not negotiate the number of slots yet, so set the
4223          * maxslots to the session maxreqs which is used to encode
4224          * sr_highest_slotid and the sr_target_slot id to maxslots */
4225         seq->maxslots = session->se_fchannel.maxreqs;
4226
4227         trace_nfsd_slot_seqid_sequence(clp, seq, slot);
4228         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
4229                                         slot->sl_flags & NFSD4_SLOT_INUSE);
4230         if (status == nfserr_replay_cache) {
4231                 status = nfserr_seq_misordered;
4232                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
4233                         goto out_put_session;
4234                 status = nfserr_seq_false_retry;
4235                 if (!replay_matches_cache(rqstp, seq, slot))
4236                         goto out_put_session;
4237                 cstate->slot = slot;
4238                 cstate->session = session;
4239                 cstate->clp = clp;
4240                 /* Return the cached reply status and set cstate->status
4241                  * for nfsd4_proc_compound processing */
4242                 status = nfsd4_replay_cache_entry(resp, seq);
4243                 cstate->status = nfserr_replay_cache;
4244                 goto out;
4245         }
4246         if (status)
4247                 goto out_put_session;
4248
4249         status = nfsd4_sequence_check_conn(conn, session);
4250         conn = NULL;
4251         if (status)
4252                 goto out_put_session;
4253
4254         buflen = (seq->cachethis) ?
4255                         session->se_fchannel.maxresp_cached :
4256                         session->se_fchannel.maxresp_sz;
4257         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
4258                                     nfserr_rep_too_big;
4259         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
4260                 goto out_put_session;
4261         svc_reserve(rqstp, buflen);
4262
4263         status = nfs_ok;
4264         /* Success! bump slot seqid */
4265         slot->sl_seqid = seq->seqid;
4266         slot->sl_flags |= NFSD4_SLOT_INUSE;
4267         if (seq->cachethis)
4268                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
4269         else
4270                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
4271
4272         cstate->slot = slot;
4273         cstate->session = session;
4274         cstate->clp = clp;
4275
4276 out:
4277         switch (clp->cl_cb_state) {
4278         case NFSD4_CB_DOWN:
4279                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
4280                 break;
4281         case NFSD4_CB_FAULT:
4282                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
4283                 break;
4284         default:
4285                 seq->status_flags = 0;
4286         }
4287         if (!list_empty(&clp->cl_revoked))
4288                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
4289         if (atomic_read(&clp->cl_admin_revoked))
4290                 seq->status_flags |= SEQ4_STATUS_ADMIN_STATE_REVOKED;
4291         trace_nfsd_seq4_status(rqstp, seq);
4292 out_no_session:
4293         if (conn)
4294                 free_conn(conn);
4295         spin_unlock(&nn->client_lock);
4296         return status;
4297 out_put_session:
4298         nfsd4_put_session_locked(session);
4299         goto out_no_session;
4300 }
4301
4302 void
4303 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
4304 {
4305         struct nfsd4_compound_state *cs = &resp->cstate;
4306
4307         if (nfsd4_has_session(cs)) {
4308                 if (cs->status != nfserr_replay_cache) {
4309                         nfsd4_store_cache_entry(resp);
4310                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
4311                 }
4312                 /* Drop session reference that was taken in nfsd4_sequence() */
4313                 nfsd4_put_session(cs->session);
4314         } else if (cs->clp)
4315                 put_client_renew(cs->clp);
4316 }
4317
4318 __be32
4319 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
4320                 struct nfsd4_compound_state *cstate,
4321                 union nfsd4_op_u *u)
4322 {
4323         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
4324         struct nfs4_client *conf, *unconf;
4325         struct nfs4_client *clp = NULL;
4326         __be32 status = 0;
4327         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4328
4329         spin_lock(&nn->client_lock);
4330         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
4331         conf = find_confirmed_client(&dc->clientid, true, nn);
4332         WARN_ON_ONCE(conf && unconf);
4333
4334         if (conf) {
4335                 if (client_has_state(conf)) {
4336                         status = nfserr_clientid_busy;
4337                         goto out;
4338                 }
4339                 status = mark_client_expired_locked(conf);
4340                 if (status)
4341                         goto out;
4342                 clp = conf;
4343         } else if (unconf)
4344                 clp = unconf;
4345         else {
4346                 status = nfserr_stale_clientid;
4347                 goto out;
4348         }
4349         if (!nfsd4_mach_creds_match(clp, rqstp)) {
4350                 clp = NULL;
4351                 status = nfserr_wrong_cred;
4352                 goto out;
4353         }
4354         trace_nfsd_clid_destroyed(&clp->cl_clientid);
4355         unhash_client_locked(clp);
4356 out:
4357         spin_unlock(&nn->client_lock);
4358         if (clp)
4359                 expire_client(clp);
4360         return status;
4361 }
4362
4363 __be32
4364 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
4365                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
4366 {
4367         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
4368         struct nfs4_client *clp = cstate->clp;
4369         __be32 status = 0;
4370
4371         if (rc->rca_one_fs) {
4372                 if (!cstate->current_fh.fh_dentry)
4373                         return nfserr_nofilehandle;
4374                 /*
4375                  * We don't take advantage of the rca_one_fs case.
4376                  * That's OK, it's optional, we can safely ignore it.
4377                  */
4378                 return nfs_ok;
4379         }
4380
4381         status = nfserr_complete_already;
4382         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
4383                 goto out;
4384
4385         status = nfserr_stale_clientid;
4386         if (is_client_expired(clp))
4387                 /*
4388                  * The following error isn't really legal.
4389                  * But we only get here if the client just explicitly
4390                  * destroyed the client.  Surely it no longer cares what
4391                  * error it gets back on an operation for the dead
4392                  * client.
4393                  */
4394                 goto out;
4395
4396         status = nfs_ok;
4397         trace_nfsd_clid_reclaim_complete(&clp->cl_clientid);
4398         nfsd4_client_record_create(clp);
4399         inc_reclaim_complete(clp);
4400 out:
4401         return status;
4402 }
4403
4404 __be32
4405 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4406                   union nfsd4_op_u *u)
4407 {
4408         struct nfsd4_setclientid *setclid = &u->setclientid;
4409         struct xdr_netobj       clname = setclid->se_name;
4410         nfs4_verifier           clverifier = setclid->se_verf;
4411         struct nfs4_client      *conf, *new;
4412         struct nfs4_client      *unconf = NULL;
4413         __be32                  status;
4414         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4415
4416         new = create_client(clname, rqstp, &clverifier);
4417         if (new == NULL)
4418                 return nfserr_jukebox;
4419         spin_lock(&nn->client_lock);
4420         conf = find_confirmed_client_by_name(&clname, nn);
4421         if (conf && client_has_state(conf)) {
4422                 status = nfserr_clid_inuse;
4423                 if (clp_used_exchangeid(conf))
4424                         goto out;
4425                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4426                         trace_nfsd_clid_cred_mismatch(conf, rqstp);
4427                         goto out;
4428                 }
4429         }
4430         unconf = find_unconfirmed_client_by_name(&clname, nn);
4431         if (unconf)
4432                 unhash_client_locked(unconf);
4433         if (conf) {
4434                 if (same_verf(&conf->cl_verifier, &clverifier)) {
4435                         copy_clid(new, conf);
4436                         gen_confirm(new, nn);
4437                 } else
4438                         trace_nfsd_clid_verf_mismatch(conf, rqstp,
4439                                                       &clverifier);
4440         } else
4441                 trace_nfsd_clid_fresh(new);
4442         new->cl_minorversion = 0;
4443         gen_callback(new, setclid, rqstp);
4444         add_to_unconfirmed(new);
4445         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
4446         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
4447         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
4448         new = NULL;
4449         status = nfs_ok;
4450 out:
4451         spin_unlock(&nn->client_lock);
4452         if (new)
4453                 free_client(new);
4454         if (unconf) {
4455                 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
4456                 expire_client(unconf);
4457         }
4458         return status;
4459 }
4460
4461 __be32
4462 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
4463                         struct nfsd4_compound_state *cstate,
4464                         union nfsd4_op_u *u)
4465 {
4466         struct nfsd4_setclientid_confirm *setclientid_confirm =
4467                         &u->setclientid_confirm;
4468         struct nfs4_client *conf, *unconf;
4469         struct nfs4_client *old = NULL;
4470         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
4471         clientid_t * clid = &setclientid_confirm->sc_clientid;
4472         __be32 status;
4473         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4474
4475         if (STALE_CLIENTID(clid, nn))
4476                 return nfserr_stale_clientid;
4477
4478         spin_lock(&nn->client_lock);
4479         conf = find_confirmed_client(clid, false, nn);
4480         unconf = find_unconfirmed_client(clid, false, nn);
4481         /*
4482          * We try hard to give out unique clientid's, so if we get an
4483          * attempt to confirm the same clientid with a different cred,
4484          * the client may be buggy; this should never happen.
4485          *
4486          * Nevertheless, RFC 7530 recommends INUSE for this case:
4487          */
4488         status = nfserr_clid_inuse;
4489         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
4490                 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
4491                 goto out;
4492         }
4493         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4494                 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4495                 goto out;
4496         }
4497         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
4498                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
4499                         status = nfs_ok;
4500                 } else
4501                         status = nfserr_stale_clientid;
4502                 goto out;
4503         }
4504         status = nfs_ok;
4505         if (conf) {
4506                 old = unconf;
4507                 unhash_client_locked(old);
4508                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
4509         } else {
4510                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
4511                 if (old) {
4512                         status = nfserr_clid_inuse;
4513                         if (client_has_state(old)
4514                                         && !same_creds(&unconf->cl_cred,
4515                                                         &old->cl_cred)) {
4516                                 old = NULL;
4517                                 goto out;
4518                         }
4519                         status = mark_client_expired_locked(old);
4520                         if (status) {
4521                                 old = NULL;
4522                                 goto out;
4523                         }
4524                         trace_nfsd_clid_replaced(&old->cl_clientid);
4525                 }
4526                 move_to_confirmed(unconf);
4527                 conf = unconf;
4528         }
4529         get_client_locked(conf);
4530         spin_unlock(&nn->client_lock);
4531         if (conf == unconf)
4532                 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
4533         nfsd4_probe_callback(conf);
4534         spin_lock(&nn->client_lock);
4535         put_client_renew_locked(conf);
4536 out:
4537         spin_unlock(&nn->client_lock);
4538         if (old)
4539                 expire_client(old);
4540         return status;
4541 }
4542
4543 static struct nfs4_file *nfsd4_alloc_file(void)
4544 {
4545         return kmem_cache_alloc(file_slab, GFP_KERNEL);
4546 }
4547
4548 /* OPEN Share state helper functions */
4549
4550 static void nfsd4_file_init(const struct svc_fh *fh, struct nfs4_file *fp)
4551 {
4552         refcount_set(&fp->fi_ref, 1);
4553         spin_lock_init(&fp->fi_lock);
4554         INIT_LIST_HEAD(&fp->fi_stateids);
4555         INIT_LIST_HEAD(&fp->fi_delegations);
4556         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4557         fh_copy_shallow(&fp->fi_fhandle, &fh->fh_handle);
4558         fp->fi_deleg_file = NULL;
4559         fp->fi_had_conflict = false;
4560         fp->fi_share_deny = 0;
4561         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4562         memset(fp->fi_access, 0, sizeof(fp->fi_access));
4563         fp->fi_aliased = false;
4564         fp->fi_inode = d_inode(fh->fh_dentry);
4565 #ifdef CONFIG_NFSD_PNFS
4566         INIT_LIST_HEAD(&fp->fi_lo_states);
4567         atomic_set(&fp->fi_lo_recalls, 0);
4568 #endif
4569 }
4570
4571 void
4572 nfsd4_free_slabs(void)
4573 {
4574         kmem_cache_destroy(client_slab);
4575         kmem_cache_destroy(openowner_slab);
4576         kmem_cache_destroy(lockowner_slab);
4577         kmem_cache_destroy(file_slab);
4578         kmem_cache_destroy(stateid_slab);
4579         kmem_cache_destroy(deleg_slab);
4580         kmem_cache_destroy(odstate_slab);
4581 }
4582
4583 int
4584 nfsd4_init_slabs(void)
4585 {
4586         client_slab = KMEM_CACHE(nfs4_client, 0);
4587         if (client_slab == NULL)
4588                 goto out;
4589         openowner_slab = KMEM_CACHE(nfs4_openowner, 0);
4590         if (openowner_slab == NULL)
4591                 goto out_free_client_slab;
4592         lockowner_slab = KMEM_CACHE(nfs4_lockowner, 0);
4593         if (lockowner_slab == NULL)
4594                 goto out_free_openowner_slab;
4595         file_slab = KMEM_CACHE(nfs4_file, 0);
4596         if (file_slab == NULL)
4597                 goto out_free_lockowner_slab;
4598         stateid_slab = KMEM_CACHE(nfs4_ol_stateid, 0);
4599         if (stateid_slab == NULL)
4600                 goto out_free_file_slab;
4601         deleg_slab = KMEM_CACHE(nfs4_delegation, 0);
4602         if (deleg_slab == NULL)
4603                 goto out_free_stateid_slab;
4604         odstate_slab = KMEM_CACHE(nfs4_clnt_odstate, 0);
4605         if (odstate_slab == NULL)
4606                 goto out_free_deleg_slab;
4607         return 0;
4608
4609 out_free_deleg_slab:
4610         kmem_cache_destroy(deleg_slab);
4611 out_free_stateid_slab:
4612         kmem_cache_destroy(stateid_slab);
4613 out_free_file_slab:
4614         kmem_cache_destroy(file_slab);
4615 out_free_lockowner_slab:
4616         kmem_cache_destroy(lockowner_slab);
4617 out_free_openowner_slab:
4618         kmem_cache_destroy(openowner_slab);
4619 out_free_client_slab:
4620         kmem_cache_destroy(client_slab);
4621 out:
4622         return -ENOMEM;
4623 }
4624
4625 static unsigned long
4626 nfsd4_state_shrinker_count(struct shrinker *shrink, struct shrink_control *sc)
4627 {
4628         int count;
4629         struct nfsd_net *nn = shrink->private_data;
4630
4631         count = atomic_read(&nn->nfsd_courtesy_clients);
4632         if (!count)
4633                 count = atomic_long_read(&num_delegations);
4634         if (count)
4635                 queue_work(laundry_wq, &nn->nfsd_shrinker_work);
4636         return (unsigned long)count;
4637 }
4638
4639 static unsigned long
4640 nfsd4_state_shrinker_scan(struct shrinker *shrink, struct shrink_control *sc)
4641 {
4642         return SHRINK_STOP;
4643 }
4644
4645 void
4646 nfsd4_init_leases_net(struct nfsd_net *nn)
4647 {
4648         struct sysinfo si;
4649         u64 max_clients;
4650
4651         nn->nfsd4_lease = 90;   /* default lease time */
4652         nn->nfsd4_grace = 90;
4653         nn->somebody_reclaimed = false;
4654         nn->track_reclaim_completes = false;
4655         nn->clverifier_counter = get_random_u32();
4656         nn->clientid_base = get_random_u32();
4657         nn->clientid_counter = nn->clientid_base + 1;
4658         nn->s2s_cp_cl_id = nn->clientid_counter++;
4659
4660         atomic_set(&nn->nfs4_client_count, 0);
4661         si_meminfo(&si);
4662         max_clients = (u64)si.totalram * si.mem_unit / (1024 * 1024 * 1024);
4663         max_clients *= NFS4_CLIENTS_PER_GB;
4664         nn->nfs4_max_clients = max_t(int, max_clients, NFS4_CLIENTS_PER_GB);
4665
4666         atomic_set(&nn->nfsd_courtesy_clients, 0);
4667 }
4668
4669 enum rp_lock {
4670         RP_UNLOCKED,
4671         RP_LOCKED,
4672         RP_UNHASHED,
4673 };
4674
4675 static void init_nfs4_replay(struct nfs4_replay *rp)
4676 {
4677         rp->rp_status = nfserr_serverfault;
4678         rp->rp_buflen = 0;
4679         rp->rp_buf = rp->rp_ibuf;
4680         atomic_set(&rp->rp_locked, RP_UNLOCKED);
4681 }
4682
4683 static int nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4684                                       struct nfs4_stateowner *so)
4685 {
4686         if (!nfsd4_has_session(cstate)) {
4687                 wait_var_event(&so->so_replay.rp_locked,
4688                                atomic_cmpxchg(&so->so_replay.rp_locked,
4689                                               RP_UNLOCKED, RP_LOCKED) != RP_LOCKED);
4690                 if (atomic_read(&so->so_replay.rp_locked) == RP_UNHASHED)
4691                         return -EAGAIN;
4692                 cstate->replay_owner = nfs4_get_stateowner(so);
4693         }
4694         return 0;
4695 }
4696
4697 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4698 {
4699         struct nfs4_stateowner *so = cstate->replay_owner;
4700
4701         if (so != NULL) {
4702                 cstate->replay_owner = NULL;
4703                 atomic_set(&so->so_replay.rp_locked, RP_UNLOCKED);
4704                 wake_up_var(&so->so_replay.rp_locked);
4705                 nfs4_put_stateowner(so);
4706         }
4707 }
4708
4709 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4710 {
4711         struct nfs4_stateowner *sop;
4712
4713         sop = kmem_cache_alloc(slab, GFP_KERNEL);
4714         if (!sop)
4715                 return NULL;
4716
4717         xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4718         if (!sop->so_owner.data) {
4719                 kmem_cache_free(slab, sop);
4720                 return NULL;
4721         }
4722
4723         INIT_LIST_HEAD(&sop->so_stateids);
4724         sop->so_client = clp;
4725         init_nfs4_replay(&sop->so_replay);
4726         atomic_set(&sop->so_count, 1);
4727         return sop;
4728 }
4729
4730 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4731 {
4732         lockdep_assert_held(&clp->cl_lock);
4733
4734         list_add(&oo->oo_owner.so_strhash,
4735                  &clp->cl_ownerstr_hashtbl[strhashval]);
4736         list_add(&oo->oo_perclient, &clp->cl_openowners);
4737 }
4738
4739 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4740 {
4741         unhash_openowner_locked(openowner(so));
4742 }
4743
4744 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4745 {
4746         struct nfs4_openowner *oo = openowner(so);
4747
4748         kmem_cache_free(openowner_slab, oo);
4749 }
4750
4751 static const struct nfs4_stateowner_operations openowner_ops = {
4752         .so_unhash =    nfs4_unhash_openowner,
4753         .so_free =      nfs4_free_openowner,
4754 };
4755
4756 static struct nfs4_ol_stateid *
4757 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4758 {
4759         struct nfs4_ol_stateid *local, *ret = NULL;
4760         struct nfs4_openowner *oo = open->op_openowner;
4761
4762         lockdep_assert_held(&fp->fi_lock);
4763
4764         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4765                 /* ignore lock owners */
4766                 if (local->st_stateowner->so_is_open_owner == 0)
4767                         continue;
4768                 if (local->st_stateowner != &oo->oo_owner)
4769                         continue;
4770                 if (local->st_stid.sc_type == SC_TYPE_OPEN &&
4771                     !local->st_stid.sc_status) {
4772                         ret = local;
4773                         refcount_inc(&ret->st_stid.sc_count);
4774                         break;
4775                 }
4776         }
4777         return ret;
4778 }
4779
4780 static void nfsd4_drop_revoked_stid(struct nfs4_stid *s)
4781         __releases(&s->sc_client->cl_lock)
4782 {
4783         struct nfs4_client *cl = s->sc_client;
4784         LIST_HEAD(reaplist);
4785         struct nfs4_ol_stateid *stp;
4786         struct nfs4_delegation *dp;
4787         bool unhashed;
4788
4789         switch (s->sc_type) {
4790         case SC_TYPE_OPEN:
4791                 stp = openlockstateid(s);
4792                 if (unhash_open_stateid(stp, &reaplist))
4793                         put_ol_stateid_locked(stp, &reaplist);
4794                 spin_unlock(&cl->cl_lock);
4795                 free_ol_stateid_reaplist(&reaplist);
4796                 break;
4797         case SC_TYPE_LOCK:
4798                 stp = openlockstateid(s);
4799                 unhashed = unhash_lock_stateid(stp);
4800                 spin_unlock(&cl->cl_lock);
4801                 if (unhashed)
4802                         nfs4_put_stid(s);
4803                 break;
4804         case SC_TYPE_DELEG:
4805                 dp = delegstateid(s);
4806                 list_del_init(&dp->dl_recall_lru);
4807                 spin_unlock(&cl->cl_lock);
4808                 nfs4_put_stid(s);
4809                 break;
4810         default:
4811                 spin_unlock(&cl->cl_lock);
4812         }
4813 }
4814
4815 static void nfsd40_drop_revoked_stid(struct nfs4_client *cl,
4816                                     stateid_t *stid)
4817 {
4818         /* NFSv4.0 has no way for the client to tell the server
4819          * that it can forget an admin-revoked stateid.
4820          * So we keep it around until the first time that the
4821          * client uses it, and drop it the first time
4822          * nfserr_admin_revoked is returned.
4823          * For v4.1 and later we wait until explicitly told
4824          * to free the stateid.
4825          */
4826         if (cl->cl_minorversion == 0) {
4827                 struct nfs4_stid *st;
4828
4829                 spin_lock(&cl->cl_lock);
4830                 st = find_stateid_locked(cl, stid);
4831                 if (st)
4832                         nfsd4_drop_revoked_stid(st);
4833                 else
4834                         spin_unlock(&cl->cl_lock);
4835         }
4836 }
4837
4838 static __be32
4839 nfsd4_verify_open_stid(struct nfs4_stid *s)
4840 {
4841         __be32 ret = nfs_ok;
4842
4843         if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
4844                 ret = nfserr_admin_revoked;
4845         else if (s->sc_status & SC_STATUS_REVOKED)
4846                 ret = nfserr_deleg_revoked;
4847         else if (s->sc_status & SC_STATUS_CLOSED)
4848                 ret = nfserr_bad_stateid;
4849         return ret;
4850 }
4851
4852 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4853 static __be32
4854 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4855 {
4856         __be32 ret;
4857
4858         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4859         ret = nfsd4_verify_open_stid(&stp->st_stid);
4860         if (ret == nfserr_admin_revoked)
4861                 nfsd40_drop_revoked_stid(stp->st_stid.sc_client,
4862                                         &stp->st_stid.sc_stateid);
4863
4864         if (ret != nfs_ok)
4865                 mutex_unlock(&stp->st_mutex);
4866         return ret;
4867 }
4868
4869 static struct nfs4_ol_stateid *
4870 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4871 {
4872         struct nfs4_ol_stateid *stp;
4873         for (;;) {
4874                 spin_lock(&fp->fi_lock);
4875                 stp = nfsd4_find_existing_open(fp, open);
4876                 spin_unlock(&fp->fi_lock);
4877                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4878                         break;
4879                 nfs4_put_stid(&stp->st_stid);
4880         }
4881         return stp;
4882 }
4883
4884 static struct nfs4_openowner *
4885 find_or_alloc_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4886                               struct nfsd4_compound_state *cstate)
4887 {
4888         struct nfs4_client *clp = cstate->clp;
4889         struct nfs4_openowner *oo, *new = NULL;
4890
4891 retry:
4892         spin_lock(&clp->cl_lock);
4893         oo = find_openstateowner_str(strhashval, open, clp);
4894         if (!oo && new) {
4895                 hash_openowner(new, clp, strhashval);
4896                 spin_unlock(&clp->cl_lock);
4897                 return new;
4898         }
4899         spin_unlock(&clp->cl_lock);
4900
4901         if (oo && !(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4902                 /* Replace unconfirmed owners without checking for replay. */
4903                 release_openowner(oo);
4904                 oo = NULL;
4905         }
4906         if (oo) {
4907                 if (new)
4908                         nfs4_free_stateowner(&new->oo_owner);
4909                 return oo;
4910         }
4911
4912         new = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4913         if (!new)
4914                 return NULL;
4915         new->oo_owner.so_ops = &openowner_ops;
4916         new->oo_owner.so_is_open_owner = 1;
4917         new->oo_owner.so_seqid = open->op_seqid;
4918         new->oo_flags = 0;
4919         if (nfsd4_has_session(cstate))
4920                 new->oo_flags |= NFS4_OO_CONFIRMED;
4921         new->oo_time = 0;
4922         new->oo_last_closed_stid = NULL;
4923         INIT_LIST_HEAD(&new->oo_close_lru);
4924         goto retry;
4925 }
4926
4927 static struct nfs4_ol_stateid *
4928 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4929 {
4930
4931         struct nfs4_openowner *oo = open->op_openowner;
4932         struct nfs4_ol_stateid *retstp = NULL;
4933         struct nfs4_ol_stateid *stp;
4934
4935         stp = open->op_stp;
4936         /* We are moving these outside of the spinlocks to avoid the warnings */
4937         mutex_init(&stp->st_mutex);
4938         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4939
4940 retry:
4941         spin_lock(&oo->oo_owner.so_client->cl_lock);
4942         spin_lock(&fp->fi_lock);
4943
4944         retstp = nfsd4_find_existing_open(fp, open);
4945         if (retstp)
4946                 goto out_unlock;
4947
4948         open->op_stp = NULL;
4949         refcount_inc(&stp->st_stid.sc_count);
4950         stp->st_stid.sc_type = SC_TYPE_OPEN;
4951         INIT_LIST_HEAD(&stp->st_locks);
4952         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4953         get_nfs4_file(fp);
4954         stp->st_stid.sc_file = fp;
4955         stp->st_access_bmap = 0;
4956         stp->st_deny_bmap = 0;
4957         stp->st_openstp = NULL;
4958         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4959         list_add(&stp->st_perfile, &fp->fi_stateids);
4960
4961 out_unlock:
4962         spin_unlock(&fp->fi_lock);
4963         spin_unlock(&oo->oo_owner.so_client->cl_lock);
4964         if (retstp) {
4965                 /* Handle races with CLOSE */
4966                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4967                         nfs4_put_stid(&retstp->st_stid);
4968                         goto retry;
4969                 }
4970                 /* To keep mutex tracking happy */
4971                 mutex_unlock(&stp->st_mutex);
4972                 stp = retstp;
4973         }
4974         return stp;
4975 }
4976
4977 /*
4978  * In the 4.0 case we need to keep the owners around a little while to handle
4979  * CLOSE replay. We still do need to release any file access that is held by
4980  * them before returning however.
4981  */
4982 static void
4983 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4984 {
4985         struct nfs4_ol_stateid *last;
4986         struct nfs4_openowner *oo = openowner(s->st_stateowner);
4987         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4988                                                 nfsd_net_id);
4989
4990         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4991
4992         /*
4993          * We know that we hold one reference via nfsd4_close, and another
4994          * "persistent" reference for the client. If the refcount is higher
4995          * than 2, then there are still calls in progress that are using this
4996          * stateid. We can't put the sc_file reference until they are finished.
4997          * Wait for the refcount to drop to 2. Since it has been unhashed,
4998          * there should be no danger of the refcount going back up again at
4999          * this point.
5000          * Some threads with a reference might be waiting for rp_locked,
5001          * so tell them to stop waiting.
5002          */
5003         atomic_set(&oo->oo_owner.so_replay.rp_locked, RP_UNHASHED);
5004         wake_up_var(&oo->oo_owner.so_replay.rp_locked);
5005         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
5006
5007         release_all_access(s);
5008         if (s->st_stid.sc_file) {
5009                 put_nfs4_file(s->st_stid.sc_file);
5010                 s->st_stid.sc_file = NULL;
5011         }
5012
5013         spin_lock(&nn->client_lock);
5014         last = oo->oo_last_closed_stid;
5015         oo->oo_last_closed_stid = s;
5016         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
5017         oo->oo_time = ktime_get_boottime_seconds();
5018         spin_unlock(&nn->client_lock);
5019         if (last)
5020                 nfs4_put_stid(&last->st_stid);
5021 }
5022
5023 static noinline_for_stack struct nfs4_file *
5024 nfsd4_file_hash_lookup(const struct svc_fh *fhp)
5025 {
5026         struct inode *inode = d_inode(fhp->fh_dentry);
5027         struct rhlist_head *tmp, *list;
5028         struct nfs4_file *fi;
5029
5030         rcu_read_lock();
5031         list = rhltable_lookup(&nfs4_file_rhltable, &inode,
5032                                nfs4_file_rhash_params);
5033         rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
5034                 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
5035                         if (refcount_inc_not_zero(&fi->fi_ref)) {
5036                                 rcu_read_unlock();
5037                                 return fi;
5038                         }
5039                 }
5040         }
5041         rcu_read_unlock();
5042         return NULL;
5043 }
5044
5045 /*
5046  * On hash insertion, identify entries with the same inode but
5047  * distinct filehandles. They will all be on the list returned
5048  * by rhltable_lookup().
5049  *
5050  * inode->i_lock prevents racing insertions from adding an entry
5051  * for the same inode/fhp pair twice.
5052  */
5053 static noinline_for_stack struct nfs4_file *
5054 nfsd4_file_hash_insert(struct nfs4_file *new, const struct svc_fh *fhp)
5055 {
5056         struct inode *inode = d_inode(fhp->fh_dentry);
5057         struct rhlist_head *tmp, *list;
5058         struct nfs4_file *ret = NULL;
5059         bool alias_found = false;
5060         struct nfs4_file *fi;
5061         int err;
5062
5063         rcu_read_lock();
5064         spin_lock(&inode->i_lock);
5065
5066         list = rhltable_lookup(&nfs4_file_rhltable, &inode,
5067                                nfs4_file_rhash_params);
5068         rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
5069                 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
5070                         if (refcount_inc_not_zero(&fi->fi_ref))
5071                                 ret = fi;
5072                 } else
5073                         fi->fi_aliased = alias_found = true;
5074         }
5075         if (ret)
5076                 goto out_unlock;
5077
5078         nfsd4_file_init(fhp, new);
5079         err = rhltable_insert(&nfs4_file_rhltable, &new->fi_rlist,
5080                               nfs4_file_rhash_params);
5081         if (err)
5082                 goto out_unlock;
5083
5084         new->fi_aliased = alias_found;
5085         ret = new;
5086
5087 out_unlock:
5088         spin_unlock(&inode->i_lock);
5089         rcu_read_unlock();
5090         return ret;
5091 }
5092
5093 static noinline_for_stack void nfsd4_file_hash_remove(struct nfs4_file *fi)
5094 {
5095         rhltable_remove(&nfs4_file_rhltable, &fi->fi_rlist,
5096                         nfs4_file_rhash_params);
5097 }
5098
5099 /*
5100  * Called to check deny when READ with all zero stateid or
5101  * WRITE with all zero or all one stateid
5102  */
5103 static __be32
5104 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
5105 {
5106         struct nfs4_file *fp;
5107         __be32 ret = nfs_ok;
5108
5109         fp = nfsd4_file_hash_lookup(current_fh);
5110         if (!fp)
5111                 return ret;
5112
5113         /* Check for conflicting share reservations */
5114         spin_lock(&fp->fi_lock);
5115         if (fp->fi_share_deny & deny_type)
5116                 ret = nfserr_locked;
5117         spin_unlock(&fp->fi_lock);
5118         put_nfs4_file(fp);
5119         return ret;
5120 }
5121
5122 static bool nfsd4_deleg_present(const struct inode *inode)
5123 {
5124         struct file_lock_context *ctx = locks_inode_context(inode);
5125
5126         return ctx && !list_empty_careful(&ctx->flc_lease);
5127 }
5128
5129 /**
5130  * nfsd_wait_for_delegreturn - wait for delegations to be returned
5131  * @rqstp: the RPC transaction being executed
5132  * @inode: in-core inode of the file being waited for
5133  *
5134  * The timeout prevents deadlock if all nfsd threads happen to be
5135  * tied up waiting for returning delegations.
5136  *
5137  * Return values:
5138  *   %true: delegation was returned
5139  *   %false: timed out waiting for delegreturn
5140  */
5141 bool nfsd_wait_for_delegreturn(struct svc_rqst *rqstp, struct inode *inode)
5142 {
5143         long __maybe_unused timeo;
5144
5145         timeo = wait_var_event_timeout(inode, !nfsd4_deleg_present(inode),
5146                                        NFSD_DELEGRETURN_TIMEOUT);
5147         trace_nfsd_delegret_wakeup(rqstp, inode, timeo);
5148         return timeo > 0;
5149 }
5150
5151 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
5152 {
5153         struct nfs4_delegation *dp = cb_to_delegation(cb);
5154         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
5155                                           nfsd_net_id);
5156
5157         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
5158
5159         /*
5160          * We can't do this in nfsd_break_deleg_cb because it is
5161          * already holding inode->i_lock.
5162          *
5163          * If the dl_time != 0, then we know that it has already been
5164          * queued for a lease break. Don't queue it again.
5165          */
5166         spin_lock(&state_lock);
5167         if (delegation_hashed(dp) && dp->dl_time == 0) {
5168                 dp->dl_time = ktime_get_boottime_seconds();
5169                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
5170         }
5171         spin_unlock(&state_lock);
5172 }
5173
5174 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
5175                 struct rpc_task *task)
5176 {
5177         struct nfs4_delegation *dp = cb_to_delegation(cb);
5178
5179         trace_nfsd_cb_recall_done(&dp->dl_stid.sc_stateid, task);
5180
5181         if (dp->dl_stid.sc_status)
5182                 /* CLOSED or REVOKED */
5183                 return 1;
5184
5185         switch (task->tk_status) {
5186         case 0:
5187                 return 1;
5188         case -NFS4ERR_DELAY:
5189                 rpc_delay(task, 2 * HZ);
5190                 return 0;
5191         case -EBADHANDLE:
5192         case -NFS4ERR_BAD_STATEID:
5193                 /*
5194                  * Race: client probably got cb_recall before open reply
5195                  * granting delegation.
5196                  */
5197                 if (dp->dl_retries--) {
5198                         rpc_delay(task, 2 * HZ);
5199                         return 0;
5200                 }
5201                 fallthrough;
5202         default:
5203                 return 1;
5204         }
5205 }
5206
5207 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
5208 {
5209         struct nfs4_delegation *dp = cb_to_delegation(cb);
5210
5211         nfs4_put_stid(&dp->dl_stid);
5212 }
5213
5214 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
5215         .prepare        = nfsd4_cb_recall_prepare,
5216         .done           = nfsd4_cb_recall_done,
5217         .release        = nfsd4_cb_recall_release,
5218 };
5219
5220 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
5221 {
5222         /*
5223          * We're assuming the state code never drops its reference
5224          * without first removing the lease.  Since we're in this lease
5225          * callback (and since the lease code is serialized by the
5226          * flc_lock) we know the server hasn't removed the lease yet, and
5227          * we know it's safe to take a reference.
5228          */
5229         refcount_inc(&dp->dl_stid.sc_count);
5230         WARN_ON_ONCE(!nfsd4_run_cb(&dp->dl_recall));
5231 }
5232
5233 /* Called from break_lease() with flc_lock held. */
5234 static bool
5235 nfsd_break_deleg_cb(struct file_lease *fl)
5236 {
5237         struct nfs4_delegation *dp = (struct nfs4_delegation *) fl->c.flc_owner;
5238         struct nfs4_file *fp = dp->dl_stid.sc_file;
5239         struct nfs4_client *clp = dp->dl_stid.sc_client;
5240         struct nfsd_net *nn;
5241
5242         trace_nfsd_cb_recall(&dp->dl_stid);
5243
5244         dp->dl_recalled = true;
5245         atomic_inc(&clp->cl_delegs_in_recall);
5246         if (try_to_expire_client(clp)) {
5247                 nn = net_generic(clp->net, nfsd_net_id);
5248                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
5249         }
5250
5251         /*
5252          * We don't want the locks code to timeout the lease for us;
5253          * we'll remove it ourself if a delegation isn't returned
5254          * in time:
5255          */
5256         fl->fl_break_time = 0;
5257
5258         fp->fi_had_conflict = true;
5259         nfsd_break_one_deleg(dp);
5260         return false;
5261 }
5262
5263 /**
5264  * nfsd_breaker_owns_lease - Check if lease conflict was resolved
5265  * @fl: Lock state to check
5266  *
5267  * Return values:
5268  *   %true: Lease conflict was resolved
5269  *   %false: Lease conflict was not resolved.
5270  */
5271 static bool nfsd_breaker_owns_lease(struct file_lease *fl)
5272 {
5273         struct nfs4_delegation *dl = fl->c.flc_owner;
5274         struct svc_rqst *rqst;
5275         struct nfs4_client *clp;
5276
5277         if (!i_am_nfsd())
5278                 return false;
5279         rqst = kthread_data(current);
5280         /* Note rq_prog == NFS_ACL_PROGRAM is also possible: */
5281         if (rqst->rq_prog != NFS_PROGRAM || rqst->rq_vers < 4)
5282                 return false;
5283         clp = *(rqst->rq_lease_breaker);
5284         return dl->dl_stid.sc_client == clp;
5285 }
5286
5287 static int
5288 nfsd_change_deleg_cb(struct file_lease *onlist, int arg,
5289                      struct list_head *dispose)
5290 {
5291         struct nfs4_delegation *dp = (struct nfs4_delegation *) onlist->c.flc_owner;
5292         struct nfs4_client *clp = dp->dl_stid.sc_client;
5293
5294         if (arg & F_UNLCK) {
5295                 if (dp->dl_recalled)
5296                         atomic_dec(&clp->cl_delegs_in_recall);
5297                 return lease_modify(onlist, arg, dispose);
5298         } else
5299                 return -EAGAIN;
5300 }
5301
5302 static const struct lease_manager_operations nfsd_lease_mng_ops = {
5303         .lm_breaker_owns_lease = nfsd_breaker_owns_lease,
5304         .lm_break = nfsd_break_deleg_cb,
5305         .lm_change = nfsd_change_deleg_cb,
5306 };
5307
5308 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
5309 {
5310         if (nfsd4_has_session(cstate))
5311                 return nfs_ok;
5312         if (seqid == so->so_seqid - 1)
5313                 return nfserr_replay_me;
5314         if (seqid == so->so_seqid)
5315                 return nfs_ok;
5316         return nfserr_bad_seqid;
5317 }
5318
5319 static struct nfs4_client *lookup_clientid(clientid_t *clid, bool sessions,
5320                                                 struct nfsd_net *nn)
5321 {
5322         struct nfs4_client *found;
5323
5324         spin_lock(&nn->client_lock);
5325         found = find_confirmed_client(clid, sessions, nn);
5326         if (found)
5327                 atomic_inc(&found->cl_rpc_users);
5328         spin_unlock(&nn->client_lock);
5329         return found;
5330 }
5331
5332 static __be32 set_client(clientid_t *clid,
5333                 struct nfsd4_compound_state *cstate,
5334                 struct nfsd_net *nn)
5335 {
5336         if (cstate->clp) {
5337                 if (!same_clid(&cstate->clp->cl_clientid, clid))
5338                         return nfserr_stale_clientid;
5339                 return nfs_ok;
5340         }
5341         if (STALE_CLIENTID(clid, nn))
5342                 return nfserr_stale_clientid;
5343         /*
5344          * We're in the 4.0 case (otherwise the SEQUENCE op would have
5345          * set cstate->clp), so session = false:
5346          */
5347         cstate->clp = lookup_clientid(clid, false, nn);
5348         if (!cstate->clp)
5349                 return nfserr_expired;
5350         return nfs_ok;
5351 }
5352
5353 __be32
5354 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
5355                     struct nfsd4_open *open, struct nfsd_net *nn)
5356 {
5357         clientid_t *clientid = &open->op_clientid;
5358         struct nfs4_client *clp = NULL;
5359         unsigned int strhashval;
5360         struct nfs4_openowner *oo = NULL;
5361         __be32 status;
5362
5363         /*
5364          * In case we need it later, after we've already created the
5365          * file and don't want to risk a further failure:
5366          */
5367         open->op_file = nfsd4_alloc_file();
5368         if (open->op_file == NULL)
5369                 return nfserr_jukebox;
5370
5371         status = set_client(clientid, cstate, nn);
5372         if (status)
5373                 return status;
5374         clp = cstate->clp;
5375
5376         strhashval = ownerstr_hashval(&open->op_owner);
5377 retry:
5378         oo = find_or_alloc_open_stateowner(strhashval, open, cstate);
5379         open->op_openowner = oo;
5380         if (!oo)
5381                 return nfserr_jukebox;
5382         if (nfsd4_cstate_assign_replay(cstate, &oo->oo_owner) == -EAGAIN) {
5383                 nfs4_put_stateowner(&oo->oo_owner);
5384                 goto retry;
5385         }
5386         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
5387         if (status)
5388                 return status;
5389
5390         open->op_stp = nfs4_alloc_open_stateid(clp);
5391         if (!open->op_stp)
5392                 return nfserr_jukebox;
5393
5394         if (nfsd4_has_session(cstate) &&
5395             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
5396                 open->op_odstate = alloc_clnt_odstate(clp);
5397                 if (!open->op_odstate)
5398                         return nfserr_jukebox;
5399         }
5400
5401         return nfs_ok;
5402 }
5403
5404 static inline __be32
5405 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
5406 {
5407         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
5408                 return nfserr_openmode;
5409         else
5410                 return nfs_ok;
5411 }
5412
5413 static int share_access_to_flags(u32 share_access)
5414 {
5415         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
5416 }
5417
5418 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl,
5419                                                   stateid_t *s)
5420 {
5421         struct nfs4_stid *ret;
5422
5423         ret = find_stateid_by_type(cl, s, SC_TYPE_DELEG, SC_STATUS_REVOKED);
5424         if (!ret)
5425                 return NULL;
5426         return delegstateid(ret);
5427 }
5428
5429 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
5430 {
5431         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
5432                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
5433 }
5434
5435 static __be32
5436 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
5437                 struct nfs4_delegation **dp)
5438 {
5439         int flags;
5440         __be32 status = nfserr_bad_stateid;
5441         struct nfs4_delegation *deleg;
5442
5443         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
5444         if (deleg == NULL)
5445                 goto out;
5446         if (deleg->dl_stid.sc_status & SC_STATUS_ADMIN_REVOKED) {
5447                 nfs4_put_stid(&deleg->dl_stid);
5448                 status = nfserr_admin_revoked;
5449                 goto out;
5450         }
5451         if (deleg->dl_stid.sc_status & SC_STATUS_REVOKED) {
5452                 nfs4_put_stid(&deleg->dl_stid);
5453                 nfsd40_drop_revoked_stid(cl, &open->op_delegate_stateid);
5454                 status = nfserr_deleg_revoked;
5455                 goto out;
5456         }
5457         flags = share_access_to_flags(open->op_share_access);
5458         status = nfs4_check_delegmode(deleg, flags);
5459         if (status) {
5460                 nfs4_put_stid(&deleg->dl_stid);
5461                 goto out;
5462         }
5463         *dp = deleg;
5464 out:
5465         if (!nfsd4_is_deleg_cur(open))
5466                 return nfs_ok;
5467         if (status)
5468                 return status;
5469         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5470         return nfs_ok;
5471 }
5472
5473 static inline int nfs4_access_to_access(u32 nfs4_access)
5474 {
5475         int flags = 0;
5476
5477         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
5478                 flags |= NFSD_MAY_READ;
5479         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
5480                 flags |= NFSD_MAY_WRITE;
5481         return flags;
5482 }
5483
5484 static inline __be32
5485 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
5486                 struct nfsd4_open *open)
5487 {
5488         struct iattr iattr = {
5489                 .ia_valid = ATTR_SIZE,
5490                 .ia_size = 0,
5491         };
5492         struct nfsd_attrs attrs = {
5493                 .na_iattr       = &iattr,
5494         };
5495         if (!open->op_truncate)
5496                 return 0;
5497         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
5498                 return nfserr_inval;
5499         return nfsd_setattr(rqstp, fh, &attrs, NULL);
5500 }
5501
5502 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
5503                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5504                 struct nfsd4_open *open, bool new_stp)
5505 {
5506         struct nfsd_file *nf = NULL;
5507         __be32 status;
5508         int oflag = nfs4_access_to_omode(open->op_share_access);
5509         int access = nfs4_access_to_access(open->op_share_access);
5510         unsigned char old_access_bmap, old_deny_bmap;
5511
5512         spin_lock(&fp->fi_lock);
5513
5514         /*
5515          * Are we trying to set a deny mode that would conflict with
5516          * current access?
5517          */
5518         status = nfs4_file_check_deny(fp, open->op_share_deny);
5519         if (status != nfs_ok) {
5520                 if (status != nfserr_share_denied) {
5521                         spin_unlock(&fp->fi_lock);
5522                         goto out;
5523                 }
5524                 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5525                                 stp, open->op_share_deny, false))
5526                         status = nfserr_jukebox;
5527                 spin_unlock(&fp->fi_lock);
5528                 goto out;
5529         }
5530
5531         /* set access to the file */
5532         status = nfs4_file_get_access(fp, open->op_share_access);
5533         if (status != nfs_ok) {
5534                 if (status != nfserr_share_denied) {
5535                         spin_unlock(&fp->fi_lock);
5536                         goto out;
5537                 }
5538                 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5539                                 stp, open->op_share_access, true))
5540                         status = nfserr_jukebox;
5541                 spin_unlock(&fp->fi_lock);
5542                 goto out;
5543         }
5544
5545         /* Set access bits in stateid */
5546         old_access_bmap = stp->st_access_bmap;
5547         set_access(open->op_share_access, stp);
5548
5549         /* Set new deny mask */
5550         old_deny_bmap = stp->st_deny_bmap;
5551         set_deny(open->op_share_deny, stp);
5552         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5553
5554         if (!fp->fi_fds[oflag]) {
5555                 spin_unlock(&fp->fi_lock);
5556
5557                 status = nfsd_file_acquire_opened(rqstp, cur_fh, access,
5558                                                   open->op_filp, &nf);
5559                 if (status != nfs_ok)
5560                         goto out_put_access;
5561
5562                 spin_lock(&fp->fi_lock);
5563                 if (!fp->fi_fds[oflag]) {
5564                         fp->fi_fds[oflag] = nf;
5565                         nf = NULL;
5566                 }
5567         }
5568         spin_unlock(&fp->fi_lock);
5569         if (nf)
5570                 nfsd_file_put(nf);
5571
5572         status = nfserrno(nfsd_open_break_lease(cur_fh->fh_dentry->d_inode,
5573                                                                 access));
5574         if (status)
5575                 goto out_put_access;
5576
5577         status = nfsd4_truncate(rqstp, cur_fh, open);
5578         if (status)
5579                 goto out_put_access;
5580 out:
5581         return status;
5582 out_put_access:
5583         stp->st_access_bmap = old_access_bmap;
5584         nfs4_file_put_access(fp, open->op_share_access);
5585         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
5586         goto out;
5587 }
5588
5589 static __be32
5590 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp,
5591                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5592                 struct nfsd4_open *open)
5593 {
5594         __be32 status;
5595         unsigned char old_deny_bmap = stp->st_deny_bmap;
5596
5597         if (!test_access(open->op_share_access, stp))
5598                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open, false);
5599
5600         /* test and set deny mode */
5601         spin_lock(&fp->fi_lock);
5602         status = nfs4_file_check_deny(fp, open->op_share_deny);
5603         switch (status) {
5604         case nfs_ok:
5605                 set_deny(open->op_share_deny, stp);
5606                 fp->fi_share_deny |=
5607                         (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5608                 break;
5609         case nfserr_share_denied:
5610                 if (nfs4_resolve_deny_conflicts_locked(fp, false,
5611                                 stp, open->op_share_deny, false))
5612                         status = nfserr_jukebox;
5613                 break;
5614         }
5615         spin_unlock(&fp->fi_lock);
5616
5617         if (status != nfs_ok)
5618                 return status;
5619
5620         status = nfsd4_truncate(rqstp, cur_fh, open);
5621         if (status != nfs_ok)
5622                 reset_union_bmap_deny(old_deny_bmap, stp);
5623         return status;
5624 }
5625
5626 /* Should we give out recallable state?: */
5627 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
5628 {
5629         if (clp->cl_cb_state == NFSD4_CB_UP)
5630                 return true;
5631         /*
5632          * In the sessions case, since we don't have to establish a
5633          * separate connection for callbacks, we assume it's OK
5634          * until we hear otherwise:
5635          */
5636         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
5637 }
5638
5639 static struct file_lease *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
5640                                                 int flag)
5641 {
5642         struct file_lease *fl;
5643
5644         fl = locks_alloc_lease();
5645         if (!fl)
5646                 return NULL;
5647         fl->fl_lmops = &nfsd_lease_mng_ops;
5648         fl->c.flc_flags = FL_DELEG;
5649         fl->c.flc_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
5650         fl->c.flc_owner = (fl_owner_t)dp;
5651         fl->c.flc_pid = current->tgid;
5652         fl->c.flc_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
5653         return fl;
5654 }
5655
5656 static int nfsd4_check_conflicting_opens(struct nfs4_client *clp,
5657                                          struct nfs4_file *fp)
5658 {
5659         struct nfs4_ol_stateid *st;
5660         struct file *f = fp->fi_deleg_file->nf_file;
5661         struct inode *ino = file_inode(f);
5662         int writes;
5663
5664         writes = atomic_read(&ino->i_writecount);
5665         if (!writes)
5666                 return 0;
5667         /*
5668          * There could be multiple filehandles (hence multiple
5669          * nfs4_files) referencing this file, but that's not too
5670          * common; let's just give up in that case rather than
5671          * trying to go look up all the clients using that other
5672          * nfs4_file as well:
5673          */
5674         if (fp->fi_aliased)
5675                 return -EAGAIN;
5676         /*
5677          * If there's a close in progress, make sure that we see it
5678          * clear any fi_fds[] entries before we see it decrement
5679          * i_writecount:
5680          */
5681         smp_mb__after_atomic();
5682
5683         if (fp->fi_fds[O_WRONLY])
5684                 writes--;
5685         if (fp->fi_fds[O_RDWR])
5686                 writes--;
5687         if (writes > 0)
5688                 return -EAGAIN; /* There may be non-NFSv4 writers */
5689         /*
5690          * It's possible there are non-NFSv4 write opens in progress,
5691          * but if they haven't incremented i_writecount yet then they
5692          * also haven't called break lease yet; so, they'll break this
5693          * lease soon enough.  So, all that's left to check for is NFSv4
5694          * opens:
5695          */
5696         spin_lock(&fp->fi_lock);
5697         list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
5698                 if (st->st_openstp == NULL /* it's an open */ &&
5699                     access_permit_write(st) &&
5700                     st->st_stid.sc_client != clp) {
5701                         spin_unlock(&fp->fi_lock);
5702                         return -EAGAIN;
5703                 }
5704         }
5705         spin_unlock(&fp->fi_lock);
5706         /*
5707          * There's a small chance that we could be racing with another
5708          * NFSv4 open.  However, any open that hasn't added itself to
5709          * the fi_stateids list also hasn't called break_lease yet; so,
5710          * they'll break this lease soon enough.
5711          */
5712         return 0;
5713 }
5714
5715 /*
5716  * It's possible that between opening the dentry and setting the delegation,
5717  * that it has been renamed or unlinked. Redo the lookup to verify that this
5718  * hasn't happened.
5719  */
5720 static int
5721 nfsd4_verify_deleg_dentry(struct nfsd4_open *open, struct nfs4_file *fp,
5722                           struct svc_fh *parent)
5723 {
5724         struct svc_export *exp;
5725         struct dentry *child;
5726         __be32 err;
5727
5728         err = nfsd_lookup_dentry(open->op_rqstp, parent,
5729                                  open->op_fname, open->op_fnamelen,
5730                                  &exp, &child);
5731
5732         if (err)
5733                 return -EAGAIN;
5734
5735         exp_put(exp);
5736         dput(child);
5737         if (child != file_dentry(fp->fi_deleg_file->nf_file))
5738                 return -EAGAIN;
5739
5740         return 0;
5741 }
5742
5743 /*
5744  * We avoid breaking delegations held by a client due to its own activity, but
5745  * clearing setuid/setgid bits on a write is an implicit activity and the client
5746  * may not notice and continue using the old mode. Avoid giving out a delegation
5747  * on setuid/setgid files when the client is requesting an open for write.
5748  */
5749 static int
5750 nfsd4_verify_setuid_write(struct nfsd4_open *open, struct nfsd_file *nf)
5751 {
5752         struct inode *inode = file_inode(nf->nf_file);
5753
5754         if ((open->op_share_access & NFS4_SHARE_ACCESS_WRITE) &&
5755             (inode->i_mode & (S_ISUID|S_ISGID)))
5756                 return -EAGAIN;
5757         return 0;
5758 }
5759
5760 static struct nfs4_delegation *
5761 nfs4_set_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5762                     struct svc_fh *parent)
5763 {
5764         int status = 0;
5765         struct nfs4_client *clp = stp->st_stid.sc_client;
5766         struct nfs4_file *fp = stp->st_stid.sc_file;
5767         struct nfs4_clnt_odstate *odstate = stp->st_clnt_odstate;
5768         struct nfs4_delegation *dp;
5769         struct nfsd_file *nf = NULL;
5770         struct file_lease *fl;
5771         u32 dl_type;
5772
5773         /*
5774          * The fi_had_conflict and nfs_get_existing_delegation checks
5775          * here are just optimizations; we'll need to recheck them at
5776          * the end:
5777          */
5778         if (fp->fi_had_conflict)
5779                 return ERR_PTR(-EAGAIN);
5780
5781         /*
5782          * Try for a write delegation first. RFC8881 section 10.4 says:
5783          *
5784          *  "An OPEN_DELEGATE_WRITE delegation allows the client to handle,
5785          *   on its own, all opens."
5786          *
5787          * Furthermore the client can use a write delegation for most READ
5788          * operations as well, so we require a O_RDWR file here.
5789          *
5790          * Offer a write delegation in the case of a BOTH open, and ensure
5791          * we get the O_RDWR descriptor.
5792          */
5793         if ((open->op_share_access & NFS4_SHARE_ACCESS_BOTH) == NFS4_SHARE_ACCESS_BOTH) {
5794                 nf = find_rw_file(fp);
5795                 dl_type = NFS4_OPEN_DELEGATE_WRITE;
5796         }
5797
5798         /*
5799          * If the file is being opened O_RDONLY or we couldn't get a O_RDWR
5800          * file for some reason, then try for a read delegation instead.
5801          */
5802         if (!nf && (open->op_share_access & NFS4_SHARE_ACCESS_READ)) {
5803                 nf = find_readable_file(fp);
5804                 dl_type = NFS4_OPEN_DELEGATE_READ;
5805         }
5806
5807         if (!nf)
5808                 return ERR_PTR(-EAGAIN);
5809
5810         spin_lock(&state_lock);
5811         spin_lock(&fp->fi_lock);
5812         if (nfs4_delegation_exists(clp, fp))
5813                 status = -EAGAIN;
5814         else if (nfsd4_verify_setuid_write(open, nf))
5815                 status = -EAGAIN;
5816         else if (!fp->fi_deleg_file) {
5817                 fp->fi_deleg_file = nf;
5818                 /* increment early to prevent fi_deleg_file from being
5819                  * cleared */
5820                 fp->fi_delegees = 1;
5821                 nf = NULL;
5822         } else
5823                 fp->fi_delegees++;
5824         spin_unlock(&fp->fi_lock);
5825         spin_unlock(&state_lock);
5826         if (nf)
5827                 nfsd_file_put(nf);
5828         if (status)
5829                 return ERR_PTR(status);
5830
5831         status = -ENOMEM;
5832         dp = alloc_init_deleg(clp, fp, odstate, dl_type);
5833         if (!dp)
5834                 goto out_delegees;
5835
5836         fl = nfs4_alloc_init_lease(dp, dl_type);
5837         if (!fl)
5838                 goto out_clnt_odstate;
5839
5840         status = kernel_setlease(fp->fi_deleg_file->nf_file,
5841                                       fl->c.flc_type, &fl, NULL);
5842         if (fl)
5843                 locks_free_lease(fl);
5844         if (status)
5845                 goto out_clnt_odstate;
5846
5847         if (parent) {
5848                 status = nfsd4_verify_deleg_dentry(open, fp, parent);
5849                 if (status)
5850                         goto out_unlock;
5851         }
5852
5853         status = nfsd4_check_conflicting_opens(clp, fp);
5854         if (status)
5855                 goto out_unlock;
5856
5857         /*
5858          * Now that the deleg is set, check again to ensure that nothing
5859          * raced in and changed the mode while we weren't lookng.
5860          */
5861         status = nfsd4_verify_setuid_write(open, fp->fi_deleg_file);
5862         if (status)
5863                 goto out_unlock;
5864
5865         status = -EAGAIN;
5866         if (fp->fi_had_conflict)
5867                 goto out_unlock;
5868
5869         spin_lock(&state_lock);
5870         spin_lock(&clp->cl_lock);
5871         spin_lock(&fp->fi_lock);
5872         status = hash_delegation_locked(dp, fp);
5873         spin_unlock(&fp->fi_lock);
5874         spin_unlock(&clp->cl_lock);
5875         spin_unlock(&state_lock);
5876
5877         if (status)
5878                 goto out_unlock;
5879
5880         return dp;
5881 out_unlock:
5882         kernel_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
5883 out_clnt_odstate:
5884         put_clnt_odstate(dp->dl_clnt_odstate);
5885         nfs4_put_stid(&dp->dl_stid);
5886 out_delegees:
5887         put_deleg_file(fp);
5888         return ERR_PTR(status);
5889 }
5890
5891 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
5892 {
5893         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5894         if (status == -EAGAIN)
5895                 open->op_why_no_deleg = WND4_CONTENTION;
5896         else {
5897                 open->op_why_no_deleg = WND4_RESOURCE;
5898                 switch (open->op_deleg_want) {
5899                 case NFS4_SHARE_WANT_READ_DELEG:
5900                 case NFS4_SHARE_WANT_WRITE_DELEG:
5901                 case NFS4_SHARE_WANT_ANY_DELEG:
5902                         break;
5903                 case NFS4_SHARE_WANT_CANCEL:
5904                         open->op_why_no_deleg = WND4_CANCELLED;
5905                         break;
5906                 case NFS4_SHARE_WANT_NO_DELEG:
5907                         WARN_ON_ONCE(1);
5908                 }
5909         }
5910 }
5911
5912 /*
5913  * The Linux NFS server does not offer write delegations to NFSv4.0
5914  * clients in order to avoid conflicts between write delegations and
5915  * GETATTRs requesting CHANGE or SIZE attributes.
5916  *
5917  * With NFSv4.1 and later minorversions, the SEQUENCE operation that
5918  * begins each COMPOUND contains a client ID. Delegation recall can
5919  * be avoided when the server recognizes the client sending a
5920  * GETATTR also holds write delegation it conflicts with.
5921  *
5922  * However, the NFSv4.0 protocol does not enable a server to
5923  * determine that a GETATTR originated from the client holding the
5924  * conflicting delegation versus coming from some other client. Per
5925  * RFC 7530 Section 16.7.5, the server must recall or send a
5926  * CB_GETATTR even when the GETATTR originates from the client that
5927  * holds the conflicting delegation.
5928  *
5929  * An NFSv4.0 client can trigger a pathological situation if it
5930  * always sends a DELEGRETURN preceded by a conflicting GETATTR in
5931  * the same COMPOUND. COMPOUND execution will always stop at the
5932  * GETATTR and the DELEGRETURN will never get executed. The server
5933  * eventually revokes the delegation, which can result in loss of
5934  * open or lock state.
5935  */
5936 static void
5937 nfs4_open_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5938                      struct svc_fh *currentfh)
5939 {
5940         struct nfs4_delegation *dp;
5941         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
5942         struct nfs4_client *clp = stp->st_stid.sc_client;
5943         struct svc_fh *parent = NULL;
5944         int cb_up;
5945         int status = 0;
5946         struct kstat stat;
5947         struct path path;
5948
5949         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
5950         open->op_recall = false;
5951         switch (open->op_claim_type) {
5952                 case NFS4_OPEN_CLAIM_PREVIOUS:
5953                         if (!cb_up)
5954                                 open->op_recall = true;
5955                         break;
5956                 case NFS4_OPEN_CLAIM_NULL:
5957                         parent = currentfh;
5958                         fallthrough;
5959                 case NFS4_OPEN_CLAIM_FH:
5960                         /*
5961                          * Let's not give out any delegations till everyone's
5962                          * had the chance to reclaim theirs, *and* until
5963                          * NLM locks have all been reclaimed:
5964                          */
5965                         if (locks_in_grace(clp->net))
5966                                 goto out_no_deleg;
5967                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
5968                                 goto out_no_deleg;
5969                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE &&
5970                                         !clp->cl_minorversion)
5971                                 goto out_no_deleg;
5972                         break;
5973                 default:
5974                         goto out_no_deleg;
5975         }
5976         dp = nfs4_set_delegation(open, stp, parent);
5977         if (IS_ERR(dp))
5978                 goto out_no_deleg;
5979
5980         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5981
5982         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE) {
5983                 open->op_delegate_type = NFS4_OPEN_DELEGATE_WRITE;
5984                 trace_nfsd_deleg_write(&dp->dl_stid.sc_stateid);
5985                 path.mnt = currentfh->fh_export->ex_path.mnt;
5986                 path.dentry = currentfh->fh_dentry;
5987                 if (vfs_getattr(&path, &stat,
5988                                 (STATX_SIZE | STATX_CTIME | STATX_CHANGE_COOKIE),
5989                                 AT_STATX_SYNC_AS_STAT)) {
5990                         nfs4_put_stid(&dp->dl_stid);
5991                         destroy_delegation(dp);
5992                         goto out_no_deleg;
5993                 }
5994                 dp->dl_cb_fattr.ncf_cur_fsize = stat.size;
5995                 dp->dl_cb_fattr.ncf_initial_cinfo =
5996                         nfsd4_change_attribute(&stat, d_inode(currentfh->fh_dentry));
5997         } else {
5998                 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
5999                 trace_nfsd_deleg_read(&dp->dl_stid.sc_stateid);
6000         }
6001         nfs4_put_stid(&dp->dl_stid);
6002         return;
6003 out_no_deleg:
6004         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
6005         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
6006             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
6007                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
6008                 open->op_recall = true;
6009         }
6010
6011         /* 4.1 client asking for a delegation? */
6012         if (open->op_deleg_want)
6013                 nfsd4_open_deleg_none_ext(open, status);
6014         return;
6015 }
6016
6017 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
6018                                         struct nfs4_delegation *dp)
6019 {
6020         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
6021             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
6022                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
6023                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
6024         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
6025                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
6026                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
6027                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
6028         }
6029         /* Otherwise the client must be confused wanting a delegation
6030          * it already has, therefore we don't return
6031          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
6032          */
6033 }
6034
6035 /**
6036  * nfsd4_process_open2 - finish open processing
6037  * @rqstp: the RPC transaction being executed
6038  * @current_fh: NFSv4 COMPOUND's current filehandle
6039  * @open: OPEN arguments
6040  *
6041  * If successful, (1) truncate the file if open->op_truncate was
6042  * set, (2) set open->op_stateid, (3) set open->op_delegation.
6043  *
6044  * Returns %nfs_ok on success; otherwise an nfs4stat value in
6045  * network byte order is returned.
6046  */
6047 __be32
6048 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
6049 {
6050         struct nfsd4_compoundres *resp = rqstp->rq_resp;
6051         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
6052         struct nfs4_file *fp = NULL;
6053         struct nfs4_ol_stateid *stp = NULL;
6054         struct nfs4_delegation *dp = NULL;
6055         __be32 status;
6056         bool new_stp = false;
6057
6058         /*
6059          * Lookup file; if found, lookup stateid and check open request,
6060          * and check for delegations in the process of being recalled.
6061          * If not found, create the nfs4_file struct
6062          */
6063         fp = nfsd4_file_hash_insert(open->op_file, current_fh);
6064         if (unlikely(!fp))
6065                 return nfserr_jukebox;
6066         if (fp != open->op_file) {
6067                 status = nfs4_check_deleg(cl, open, &dp);
6068                 if (status)
6069                         goto out;
6070                 stp = nfsd4_find_and_lock_existing_open(fp, open);
6071         } else {
6072                 open->op_file = NULL;
6073                 status = nfserr_bad_stateid;
6074                 if (nfsd4_is_deleg_cur(open))
6075                         goto out;
6076         }
6077
6078         if (!stp) {
6079                 stp = init_open_stateid(fp, open);
6080                 if (!open->op_stp)
6081                         new_stp = true;
6082         }
6083
6084         /*
6085          * OPEN the file, or upgrade an existing OPEN.
6086          * If truncate fails, the OPEN fails.
6087          *
6088          * stp is already locked.
6089          */
6090         if (!new_stp) {
6091                 /* Stateid was found, this is an OPEN upgrade */
6092                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
6093                 if (status) {
6094                         mutex_unlock(&stp->st_mutex);
6095                         goto out;
6096                 }
6097         } else {
6098                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open, true);
6099                 if (status) {
6100                         release_open_stateid(stp);
6101                         mutex_unlock(&stp->st_mutex);
6102                         goto out;
6103                 }
6104
6105                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
6106                                                         open->op_odstate);
6107                 if (stp->st_clnt_odstate == open->op_odstate)
6108                         open->op_odstate = NULL;
6109         }
6110
6111         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
6112         mutex_unlock(&stp->st_mutex);
6113
6114         if (nfsd4_has_session(&resp->cstate)) {
6115                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
6116                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
6117                         open->op_why_no_deleg = WND4_NOT_WANTED;
6118                         goto nodeleg;
6119                 }
6120         }
6121
6122         /*
6123         * Attempt to hand out a delegation. No error return, because the
6124         * OPEN succeeds even if we fail.
6125         */
6126         nfs4_open_delegation(open, stp, &resp->cstate.current_fh);
6127 nodeleg:
6128         status = nfs_ok;
6129         trace_nfsd_open(&stp->st_stid.sc_stateid);
6130 out:
6131         /* 4.1 client trying to upgrade/downgrade delegation? */
6132         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
6133             open->op_deleg_want)
6134                 nfsd4_deleg_xgrade_none_ext(open, dp);
6135
6136         if (fp)
6137                 put_nfs4_file(fp);
6138         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
6139                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
6140         /*
6141         * To finish the open response, we just need to set the rflags.
6142         */
6143         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
6144         if (nfsd4_has_session(&resp->cstate))
6145                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
6146         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
6147                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
6148
6149         if (dp)
6150                 nfs4_put_stid(&dp->dl_stid);
6151         if (stp)
6152                 nfs4_put_stid(&stp->st_stid);
6153
6154         return status;
6155 }
6156
6157 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
6158                               struct nfsd4_open *open)
6159 {
6160         if (open->op_openowner)
6161                 nfs4_put_stateowner(&open->op_openowner->oo_owner);
6162         if (open->op_file)
6163                 kmem_cache_free(file_slab, open->op_file);
6164         if (open->op_stp)
6165                 nfs4_put_stid(&open->op_stp->st_stid);
6166         if (open->op_odstate)
6167                 kmem_cache_free(odstate_slab, open->op_odstate);
6168 }
6169
6170 __be32
6171 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6172             union nfsd4_op_u *u)
6173 {
6174         clientid_t *clid = &u->renew;
6175         struct nfs4_client *clp;
6176         __be32 status;
6177         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6178
6179         trace_nfsd_clid_renew(clid);
6180         status = set_client(clid, cstate, nn);
6181         if (status)
6182                 return status;
6183         clp = cstate->clp;
6184         if (!list_empty(&clp->cl_delegations)
6185                         && clp->cl_cb_state != NFSD4_CB_UP)
6186                 return nfserr_cb_path_down;
6187         return nfs_ok;
6188 }
6189
6190 void
6191 nfsd4_end_grace(struct nfsd_net *nn)
6192 {
6193         /* do nothing if grace period already ended */
6194         if (nn->grace_ended)
6195                 return;
6196
6197         trace_nfsd_grace_complete(nn);
6198         nn->grace_ended = true;
6199         /*
6200          * If the server goes down again right now, an NFSv4
6201          * client will still be allowed to reclaim after it comes back up,
6202          * even if it hasn't yet had a chance to reclaim state this time.
6203          *
6204          */
6205         nfsd4_record_grace_done(nn);
6206         /*
6207          * At this point, NFSv4 clients can still reclaim.  But if the
6208          * server crashes, any that have not yet reclaimed will be out
6209          * of luck on the next boot.
6210          *
6211          * (NFSv4.1+ clients are considered to have reclaimed once they
6212          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
6213          * have reclaimed after their first OPEN.)
6214          */
6215         locks_end_grace(&nn->nfsd4_manager);
6216         /*
6217          * At this point, and once lockd and/or any other containers
6218          * exit their grace period, further reclaims will fail and
6219          * regular locking can resume.
6220          */
6221 }
6222
6223 /*
6224  * If we've waited a lease period but there are still clients trying to
6225  * reclaim, wait a little longer to give them a chance to finish.
6226  */
6227 static bool clients_still_reclaiming(struct nfsd_net *nn)
6228 {
6229         time64_t double_grace_period_end = nn->boot_time +
6230                                            2 * nn->nfsd4_lease;
6231
6232         if (nn->track_reclaim_completes &&
6233                         atomic_read(&nn->nr_reclaim_complete) ==
6234                         nn->reclaim_str_hashtbl_size)
6235                 return false;
6236         if (!nn->somebody_reclaimed)
6237                 return false;
6238         nn->somebody_reclaimed = false;
6239         /*
6240          * If we've given them *two* lease times to reclaim, and they're
6241          * still not done, give up:
6242          */
6243         if (ktime_get_boottime_seconds() > double_grace_period_end)
6244                 return false;
6245         return true;
6246 }
6247
6248 struct laundry_time {
6249         time64_t cutoff;
6250         time64_t new_timeo;
6251 };
6252
6253 static bool state_expired(struct laundry_time *lt, time64_t last_refresh)
6254 {
6255         time64_t time_remaining;
6256
6257         if (last_refresh < lt->cutoff)
6258                 return true;
6259         time_remaining = last_refresh - lt->cutoff;
6260         lt->new_timeo = min(lt->new_timeo, time_remaining);
6261         return false;
6262 }
6263
6264 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
6265 void nfsd4_ssc_init_umount_work(struct nfsd_net *nn)
6266 {
6267         spin_lock_init(&nn->nfsd_ssc_lock);
6268         INIT_LIST_HEAD(&nn->nfsd_ssc_mount_list);
6269         init_waitqueue_head(&nn->nfsd_ssc_waitq);
6270 }
6271 EXPORT_SYMBOL_GPL(nfsd4_ssc_init_umount_work);
6272
6273 /*
6274  * This is called when nfsd is being shutdown, after all inter_ssc
6275  * cleanup were done, to destroy the ssc delayed unmount list.
6276  */
6277 static void nfsd4_ssc_shutdown_umount(struct nfsd_net *nn)
6278 {
6279         struct nfsd4_ssc_umount_item *ni = NULL;
6280         struct nfsd4_ssc_umount_item *tmp;
6281
6282         spin_lock(&nn->nfsd_ssc_lock);
6283         list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
6284                 list_del(&ni->nsui_list);
6285                 spin_unlock(&nn->nfsd_ssc_lock);
6286                 mntput(ni->nsui_vfsmount);
6287                 kfree(ni);
6288                 spin_lock(&nn->nfsd_ssc_lock);
6289         }
6290         spin_unlock(&nn->nfsd_ssc_lock);
6291 }
6292
6293 static void nfsd4_ssc_expire_umount(struct nfsd_net *nn)
6294 {
6295         bool do_wakeup = false;
6296         struct nfsd4_ssc_umount_item *ni = NULL;
6297         struct nfsd4_ssc_umount_item *tmp;
6298
6299         spin_lock(&nn->nfsd_ssc_lock);
6300         list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
6301                 if (time_after(jiffies, ni->nsui_expire)) {
6302                         if (refcount_read(&ni->nsui_refcnt) > 1)
6303                                 continue;
6304
6305                         /* mark being unmount */
6306                         ni->nsui_busy = true;
6307                         spin_unlock(&nn->nfsd_ssc_lock);
6308                         mntput(ni->nsui_vfsmount);
6309                         spin_lock(&nn->nfsd_ssc_lock);
6310
6311                         /* waiters need to start from begin of list */
6312                         list_del(&ni->nsui_list);
6313                         kfree(ni);
6314
6315                         /* wakeup ssc_connect waiters */
6316                         do_wakeup = true;
6317                         continue;
6318                 }
6319                 break;
6320         }
6321         if (do_wakeup)
6322                 wake_up_all(&nn->nfsd_ssc_waitq);
6323         spin_unlock(&nn->nfsd_ssc_lock);
6324 }
6325 #endif
6326
6327 /* Check if any lock belonging to this lockowner has any blockers */
6328 static bool
6329 nfs4_lockowner_has_blockers(struct nfs4_lockowner *lo)
6330 {
6331         struct file_lock_context *ctx;
6332         struct nfs4_ol_stateid *stp;
6333         struct nfs4_file *nf;
6334
6335         list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
6336                 nf = stp->st_stid.sc_file;
6337                 ctx = locks_inode_context(nf->fi_inode);
6338                 if (!ctx)
6339                         continue;
6340                 if (locks_owner_has_blockers(ctx, lo))
6341                         return true;
6342         }
6343         return false;
6344 }
6345
6346 static bool
6347 nfs4_anylock_blockers(struct nfs4_client *clp)
6348 {
6349         int i;
6350         struct nfs4_stateowner *so;
6351         struct nfs4_lockowner *lo;
6352
6353         if (atomic_read(&clp->cl_delegs_in_recall))
6354                 return true;
6355         spin_lock(&clp->cl_lock);
6356         for (i = 0; i < OWNER_HASH_SIZE; i++) {
6357                 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[i],
6358                                 so_strhash) {
6359                         if (so->so_is_open_owner)
6360                                 continue;
6361                         lo = lockowner(so);
6362                         if (nfs4_lockowner_has_blockers(lo)) {
6363                                 spin_unlock(&clp->cl_lock);
6364                                 return true;
6365                         }
6366                 }
6367         }
6368         spin_unlock(&clp->cl_lock);
6369         return false;
6370 }
6371
6372 static void
6373 nfs4_get_client_reaplist(struct nfsd_net *nn, struct list_head *reaplist,
6374                                 struct laundry_time *lt)
6375 {
6376         unsigned int maxreap, reapcnt = 0;
6377         struct list_head *pos, *next;
6378         struct nfs4_client *clp;
6379
6380         maxreap = (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) ?
6381                         NFSD_CLIENT_MAX_TRIM_PER_RUN : 0;
6382         INIT_LIST_HEAD(reaplist);
6383         spin_lock(&nn->client_lock);
6384         list_for_each_safe(pos, next, &nn->client_lru) {
6385                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6386                 if (clp->cl_state == NFSD4_EXPIRABLE)
6387                         goto exp_client;
6388                 if (!state_expired(lt, clp->cl_time))
6389                         break;
6390                 if (!atomic_read(&clp->cl_rpc_users)) {
6391                         if (clp->cl_state == NFSD4_ACTIVE)
6392                                 atomic_inc(&nn->nfsd_courtesy_clients);
6393                         clp->cl_state = NFSD4_COURTESY;
6394                 }
6395                 if (!client_has_state(clp))
6396                         goto exp_client;
6397                 if (!nfs4_anylock_blockers(clp))
6398                         if (reapcnt >= maxreap)
6399                                 continue;
6400 exp_client:
6401                 if (!mark_client_expired_locked(clp)) {
6402                         list_add(&clp->cl_lru, reaplist);
6403                         reapcnt++;
6404                 }
6405         }
6406         spin_unlock(&nn->client_lock);
6407 }
6408
6409 static void
6410 nfs4_get_courtesy_client_reaplist(struct nfsd_net *nn,
6411                                 struct list_head *reaplist)
6412 {
6413         unsigned int maxreap = 0, reapcnt = 0;
6414         struct list_head *pos, *next;
6415         struct nfs4_client *clp;
6416
6417         maxreap = NFSD_CLIENT_MAX_TRIM_PER_RUN;
6418         INIT_LIST_HEAD(reaplist);
6419
6420         spin_lock(&nn->client_lock);
6421         list_for_each_safe(pos, next, &nn->client_lru) {
6422                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6423                 if (clp->cl_state == NFSD4_ACTIVE)
6424                         break;
6425                 if (reapcnt >= maxreap)
6426                         break;
6427                 if (!mark_client_expired_locked(clp)) {
6428                         list_add(&clp->cl_lru, reaplist);
6429                         reapcnt++;
6430                 }
6431         }
6432         spin_unlock(&nn->client_lock);
6433 }
6434
6435 static void
6436 nfs4_process_client_reaplist(struct list_head *reaplist)
6437 {
6438         struct list_head *pos, *next;
6439         struct nfs4_client *clp;
6440
6441         list_for_each_safe(pos, next, reaplist) {
6442                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6443                 trace_nfsd_clid_purged(&clp->cl_clientid);
6444                 list_del_init(&clp->cl_lru);
6445                 expire_client(clp);
6446         }
6447 }
6448
6449 static void nfs40_clean_admin_revoked(struct nfsd_net *nn,
6450                                       struct laundry_time *lt)
6451 {
6452         struct nfs4_client *clp;
6453
6454         spin_lock(&nn->client_lock);
6455         if (nn->nfs40_last_revoke == 0 ||
6456             nn->nfs40_last_revoke > lt->cutoff) {
6457                 spin_unlock(&nn->client_lock);
6458                 return;
6459         }
6460         nn->nfs40_last_revoke = 0;
6461
6462 retry:
6463         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6464                 unsigned long id, tmp;
6465                 struct nfs4_stid *stid;
6466
6467                 if (atomic_read(&clp->cl_admin_revoked) == 0)
6468                         continue;
6469
6470                 spin_lock(&clp->cl_lock);
6471                 idr_for_each_entry_ul(&clp->cl_stateids, stid, tmp, id)
6472                         if (stid->sc_status & SC_STATUS_ADMIN_REVOKED) {
6473                                 refcount_inc(&stid->sc_count);
6474                                 spin_unlock(&nn->client_lock);
6475                                 /* this function drops ->cl_lock */
6476                                 nfsd4_drop_revoked_stid(stid);
6477                                 nfs4_put_stid(stid);
6478                                 spin_lock(&nn->client_lock);
6479                                 goto retry;
6480                         }
6481                 spin_unlock(&clp->cl_lock);
6482         }
6483         spin_unlock(&nn->client_lock);
6484 }
6485
6486 static time64_t
6487 nfs4_laundromat(struct nfsd_net *nn)
6488 {
6489         struct nfs4_openowner *oo;
6490         struct nfs4_delegation *dp;
6491         struct nfs4_ol_stateid *stp;
6492         struct nfsd4_blocked_lock *nbl;
6493         struct list_head *pos, *next, reaplist;
6494         struct laundry_time lt = {
6495                 .cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease,
6496                 .new_timeo = nn->nfsd4_lease
6497         };
6498         struct nfs4_cpntf_state *cps;
6499         copy_stateid_t *cps_t;
6500         int i;
6501
6502         if (clients_still_reclaiming(nn)) {
6503                 lt.new_timeo = 0;
6504                 goto out;
6505         }
6506         nfsd4_end_grace(nn);
6507
6508         spin_lock(&nn->s2s_cp_lock);
6509         idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
6510                 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
6511                 if (cps->cp_stateid.cs_type == NFS4_COPYNOTIFY_STID &&
6512                                 state_expired(&lt, cps->cpntf_time))
6513                         _free_cpntf_state_locked(nn, cps);
6514         }
6515         spin_unlock(&nn->s2s_cp_lock);
6516         nfs4_get_client_reaplist(nn, &reaplist, &lt);
6517         nfs4_process_client_reaplist(&reaplist);
6518
6519         nfs40_clean_admin_revoked(nn, &lt);
6520
6521         spin_lock(&state_lock);
6522         list_for_each_safe(pos, next, &nn->del_recall_lru) {
6523                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6524                 if (!state_expired(&lt, dp->dl_time))
6525                         break;
6526                 unhash_delegation_locked(dp, SC_STATUS_REVOKED);
6527                 list_add(&dp->dl_recall_lru, &reaplist);
6528         }
6529         spin_unlock(&state_lock);
6530         while (!list_empty(&reaplist)) {
6531                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
6532                                         dl_recall_lru);
6533                 list_del_init(&dp->dl_recall_lru);
6534                 revoke_delegation(dp);
6535         }
6536
6537         spin_lock(&nn->client_lock);
6538         while (!list_empty(&nn->close_lru)) {
6539                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
6540                                         oo_close_lru);
6541                 if (!state_expired(&lt, oo->oo_time))
6542                         break;
6543                 list_del_init(&oo->oo_close_lru);
6544                 stp = oo->oo_last_closed_stid;
6545                 oo->oo_last_closed_stid = NULL;
6546                 spin_unlock(&nn->client_lock);
6547                 nfs4_put_stid(&stp->st_stid);
6548                 spin_lock(&nn->client_lock);
6549         }
6550         spin_unlock(&nn->client_lock);
6551
6552         /*
6553          * It's possible for a client to try and acquire an already held lock
6554          * that is being held for a long time, and then lose interest in it.
6555          * So, we clean out any un-revisited request after a lease period
6556          * under the assumption that the client is no longer interested.
6557          *
6558          * RFC5661, sec. 9.6 states that the client must not rely on getting
6559          * notifications and must continue to poll for locks, even when the
6560          * server supports them. Thus this shouldn't lead to clients blocking
6561          * indefinitely once the lock does become free.
6562          */
6563         BUG_ON(!list_empty(&reaplist));
6564         spin_lock(&nn->blocked_locks_lock);
6565         while (!list_empty(&nn->blocked_locks_lru)) {
6566                 nbl = list_first_entry(&nn->blocked_locks_lru,
6567                                         struct nfsd4_blocked_lock, nbl_lru);
6568                 if (!state_expired(&lt, nbl->nbl_time))
6569                         break;
6570                 list_move(&nbl->nbl_lru, &reaplist);
6571                 list_del_init(&nbl->nbl_list);
6572         }
6573         spin_unlock(&nn->blocked_locks_lock);
6574
6575         while (!list_empty(&reaplist)) {
6576                 nbl = list_first_entry(&reaplist,
6577                                         struct nfsd4_blocked_lock, nbl_lru);
6578                 list_del_init(&nbl->nbl_lru);
6579                 free_blocked_lock(nbl);
6580         }
6581 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
6582         /* service the server-to-server copy delayed unmount list */
6583         nfsd4_ssc_expire_umount(nn);
6584 #endif
6585         if (atomic_long_read(&num_delegations) >= max_delegations)
6586                 deleg_reaper(nn);
6587 out:
6588         return max_t(time64_t, lt.new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
6589 }
6590
6591 static void laundromat_main(struct work_struct *);
6592
6593 static void
6594 laundromat_main(struct work_struct *laundry)
6595 {
6596         time64_t t;
6597         struct delayed_work *dwork = to_delayed_work(laundry);
6598         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
6599                                            laundromat_work);
6600
6601         t = nfs4_laundromat(nn);
6602         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
6603 }
6604
6605 static void
6606 courtesy_client_reaper(struct nfsd_net *nn)
6607 {
6608         struct list_head reaplist;
6609
6610         nfs4_get_courtesy_client_reaplist(nn, &reaplist);
6611         nfs4_process_client_reaplist(&reaplist);
6612 }
6613
6614 static void
6615 deleg_reaper(struct nfsd_net *nn)
6616 {
6617         struct list_head *pos, *next;
6618         struct nfs4_client *clp;
6619         struct list_head cblist;
6620
6621         INIT_LIST_HEAD(&cblist);
6622         spin_lock(&nn->client_lock);
6623         list_for_each_safe(pos, next, &nn->client_lru) {
6624                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6625                 if (clp->cl_state != NFSD4_ACTIVE ||
6626                         list_empty(&clp->cl_delegations) ||
6627                         atomic_read(&clp->cl_delegs_in_recall) ||
6628                         test_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags) ||
6629                         (ktime_get_boottime_seconds() -
6630                                 clp->cl_ra_time < 5)) {
6631                         continue;
6632                 }
6633                 list_add(&clp->cl_ra_cblist, &cblist);
6634
6635                 /* release in nfsd4_cb_recall_any_release */
6636                 kref_get(&clp->cl_nfsdfs.cl_ref);
6637                 set_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
6638                 clp->cl_ra_time = ktime_get_boottime_seconds();
6639         }
6640         spin_unlock(&nn->client_lock);
6641
6642         while (!list_empty(&cblist)) {
6643                 clp = list_first_entry(&cblist, struct nfs4_client,
6644                                         cl_ra_cblist);
6645                 list_del_init(&clp->cl_ra_cblist);
6646                 clp->cl_ra->ra_keep = 0;
6647                 clp->cl_ra->ra_bmval[0] = BIT(RCA4_TYPE_MASK_RDATA_DLG);
6648                 clp->cl_ra->ra_bmval[0] = BIT(RCA4_TYPE_MASK_RDATA_DLG) |
6649                                                 BIT(RCA4_TYPE_MASK_WDATA_DLG);
6650                 trace_nfsd_cb_recall_any(clp->cl_ra);
6651                 nfsd4_run_cb(&clp->cl_ra->ra_cb);
6652         }
6653 }
6654
6655 static void
6656 nfsd4_state_shrinker_worker(struct work_struct *work)
6657 {
6658         struct nfsd_net *nn = container_of(work, struct nfsd_net,
6659                                 nfsd_shrinker_work);
6660
6661         courtesy_client_reaper(nn);
6662         deleg_reaper(nn);
6663 }
6664
6665 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
6666 {
6667         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
6668                 return nfserr_bad_stateid;
6669         return nfs_ok;
6670 }
6671
6672 static
6673 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
6674 {
6675         __be32 status = nfserr_openmode;
6676
6677         /* For lock stateid's, we test the parent open, not the lock: */
6678         if (stp->st_openstp)
6679                 stp = stp->st_openstp;
6680         if ((flags & WR_STATE) && !access_permit_write(stp))
6681                 goto out;
6682         if ((flags & RD_STATE) && !access_permit_read(stp))
6683                 goto out;
6684         status = nfs_ok;
6685 out:
6686         return status;
6687 }
6688
6689 static inline __be32
6690 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
6691 {
6692         if (ONE_STATEID(stateid) && (flags & RD_STATE))
6693                 return nfs_ok;
6694         else if (opens_in_grace(net)) {
6695                 /* Answer in remaining cases depends on existence of
6696                  * conflicting state; so we must wait out the grace period. */
6697                 return nfserr_grace;
6698         } else if (flags & WR_STATE)
6699                 return nfs4_share_conflict(current_fh,
6700                                 NFS4_SHARE_DENY_WRITE);
6701         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
6702                 return nfs4_share_conflict(current_fh,
6703                                 NFS4_SHARE_DENY_READ);
6704 }
6705
6706 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
6707 {
6708         /*
6709          * When sessions are used the stateid generation number is ignored
6710          * when it is zero.
6711          */
6712         if (has_session && in->si_generation == 0)
6713                 return nfs_ok;
6714
6715         if (in->si_generation == ref->si_generation)
6716                 return nfs_ok;
6717
6718         /* If the client sends us a stateid from the future, it's buggy: */
6719         if (nfsd4_stateid_generation_after(in, ref))
6720                 return nfserr_bad_stateid;
6721         /*
6722          * However, we could see a stateid from the past, even from a
6723          * non-buggy client.  For example, if the client sends a lock
6724          * while some IO is outstanding, the lock may bump si_generation
6725          * while the IO is still in flight.  The client could avoid that
6726          * situation by waiting for responses on all the IO requests,
6727          * but better performance may result in retrying IO that
6728          * receives an old_stateid error if requests are rarely
6729          * reordered in flight:
6730          */
6731         return nfserr_old_stateid;
6732 }
6733
6734 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
6735 {
6736         __be32 ret;
6737
6738         spin_lock(&s->sc_lock);
6739         ret = nfsd4_verify_open_stid(s);
6740         if (ret == nfs_ok)
6741                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
6742         spin_unlock(&s->sc_lock);
6743         if (ret == nfserr_admin_revoked)
6744                 nfsd40_drop_revoked_stid(s->sc_client,
6745                                         &s->sc_stateid);
6746         return ret;
6747 }
6748
6749 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
6750 {
6751         if (ols->st_stateowner->so_is_open_owner &&
6752             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
6753                 return nfserr_bad_stateid;
6754         return nfs_ok;
6755 }
6756
6757 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
6758 {
6759         struct nfs4_stid *s;
6760         __be32 status = nfserr_bad_stateid;
6761
6762         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6763                 CLOSE_STATEID(stateid))
6764                 return status;
6765         spin_lock(&cl->cl_lock);
6766         s = find_stateid_locked(cl, stateid);
6767         if (!s)
6768                 goto out_unlock;
6769         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
6770         if (status)
6771                 goto out_unlock;
6772         status = nfsd4_verify_open_stid(s);
6773         if (status)
6774                 goto out_unlock;
6775
6776         switch (s->sc_type) {
6777         case SC_TYPE_DELEG:
6778                 status = nfs_ok;
6779                 break;
6780         case SC_TYPE_OPEN:
6781         case SC_TYPE_LOCK:
6782                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
6783                 break;
6784         default:
6785                 printk("unknown stateid type %x\n", s->sc_type);
6786                 status = nfserr_bad_stateid;
6787         }
6788 out_unlock:
6789         spin_unlock(&cl->cl_lock);
6790         if (status == nfserr_admin_revoked)
6791                 nfsd40_drop_revoked_stid(cl, stateid);
6792         return status;
6793 }
6794
6795 __be32
6796 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
6797                      stateid_t *stateid,
6798                      unsigned short typemask, unsigned short statusmask,
6799                      struct nfs4_stid **s, struct nfsd_net *nn)
6800 {
6801         __be32 status;
6802         struct nfs4_stid *stid;
6803         bool return_revoked = false;
6804
6805         /*
6806          *  only return revoked delegations if explicitly asked.
6807          *  otherwise we report revoked or bad_stateid status.
6808          */
6809         if (statusmask & SC_STATUS_REVOKED)
6810                 return_revoked = true;
6811         if (typemask & SC_TYPE_DELEG)
6812                 /* Always allow REVOKED for DELEG so we can
6813                  * retturn the appropriate error.
6814                  */
6815                 statusmask |= SC_STATUS_REVOKED;
6816
6817         statusmask |= SC_STATUS_ADMIN_REVOKED;
6818
6819         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6820                 CLOSE_STATEID(stateid))
6821                 return nfserr_bad_stateid;
6822         status = set_client(&stateid->si_opaque.so_clid, cstate, nn);
6823         if (status == nfserr_stale_clientid) {
6824                 if (cstate->session)
6825                         return nfserr_bad_stateid;
6826                 return nfserr_stale_stateid;
6827         }
6828         if (status)
6829                 return status;
6830         stid = find_stateid_by_type(cstate->clp, stateid, typemask, statusmask);
6831         if (!stid)
6832                 return nfserr_bad_stateid;
6833         if ((stid->sc_status & SC_STATUS_REVOKED) && !return_revoked) {
6834                 nfs4_put_stid(stid);
6835                 return nfserr_deleg_revoked;
6836         }
6837         if (stid->sc_status & SC_STATUS_ADMIN_REVOKED) {
6838                 nfsd40_drop_revoked_stid(cstate->clp, stateid);
6839                 nfs4_put_stid(stid);
6840                 return nfserr_admin_revoked;
6841         }
6842         *s = stid;
6843         return nfs_ok;
6844 }
6845
6846 static struct nfsd_file *
6847 nfs4_find_file(struct nfs4_stid *s, int flags)
6848 {
6849         struct nfsd_file *ret = NULL;
6850
6851         if (!s || s->sc_status)
6852                 return NULL;
6853
6854         switch (s->sc_type) {
6855         case SC_TYPE_DELEG:
6856                 spin_lock(&s->sc_file->fi_lock);
6857                 ret = nfsd_file_get(s->sc_file->fi_deleg_file);
6858                 spin_unlock(&s->sc_file->fi_lock);
6859                 break;
6860         case SC_TYPE_OPEN:
6861         case SC_TYPE_LOCK:
6862                 if (flags & RD_STATE)
6863                         ret = find_readable_file(s->sc_file);
6864                 else
6865                         ret = find_writeable_file(s->sc_file);
6866         }
6867
6868         return ret;
6869 }
6870
6871 static __be32
6872 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
6873 {
6874         __be32 status;
6875
6876         status = nfsd4_check_openowner_confirmed(ols);
6877         if (status)
6878                 return status;
6879         return nfs4_check_openmode(ols, flags);
6880 }
6881
6882 static __be32
6883 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
6884                 struct nfsd_file **nfp, int flags)
6885 {
6886         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
6887         struct nfsd_file *nf;
6888         __be32 status;
6889
6890         nf = nfs4_find_file(s, flags);
6891         if (nf) {
6892                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
6893                                 acc | NFSD_MAY_OWNER_OVERRIDE);
6894                 if (status) {
6895                         nfsd_file_put(nf);
6896                         goto out;
6897                 }
6898         } else {
6899                 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
6900                 if (status)
6901                         return status;
6902         }
6903         *nfp = nf;
6904 out:
6905         return status;
6906 }
6907 static void
6908 _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6909 {
6910         WARN_ON_ONCE(cps->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID);
6911         if (!refcount_dec_and_test(&cps->cp_stateid.cs_count))
6912                 return;
6913         list_del(&cps->cp_list);
6914         idr_remove(&nn->s2s_cp_stateids,
6915                    cps->cp_stateid.cs_stid.si_opaque.so_id);
6916         kfree(cps);
6917 }
6918 /*
6919  * A READ from an inter server to server COPY will have a
6920  * copy stateid. Look up the copy notify stateid from the
6921  * idr structure and take a reference on it.
6922  */
6923 __be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6924                           struct nfs4_client *clp,
6925                           struct nfs4_cpntf_state **cps)
6926 {
6927         copy_stateid_t *cps_t;
6928         struct nfs4_cpntf_state *state = NULL;
6929
6930         if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
6931                 return nfserr_bad_stateid;
6932         spin_lock(&nn->s2s_cp_lock);
6933         cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
6934         if (cps_t) {
6935                 state = container_of(cps_t, struct nfs4_cpntf_state,
6936                                      cp_stateid);
6937                 if (state->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID) {
6938                         state = NULL;
6939                         goto unlock;
6940                 }
6941                 if (!clp)
6942                         refcount_inc(&state->cp_stateid.cs_count);
6943                 else
6944                         _free_cpntf_state_locked(nn, state);
6945         }
6946 unlock:
6947         spin_unlock(&nn->s2s_cp_lock);
6948         if (!state)
6949                 return nfserr_bad_stateid;
6950         if (!clp)
6951                 *cps = state;
6952         return 0;
6953 }
6954
6955 static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6956                                struct nfs4_stid **stid)
6957 {
6958         __be32 status;
6959         struct nfs4_cpntf_state *cps = NULL;
6960         struct nfs4_client *found;
6961
6962         status = manage_cpntf_state(nn, st, NULL, &cps);
6963         if (status)
6964                 return status;
6965
6966         cps->cpntf_time = ktime_get_boottime_seconds();
6967
6968         status = nfserr_expired;
6969         found = lookup_clientid(&cps->cp_p_clid, true, nn);
6970         if (!found)
6971                 goto out;
6972
6973         *stid = find_stateid_by_type(found, &cps->cp_p_stateid,
6974                                      SC_TYPE_DELEG|SC_TYPE_OPEN|SC_TYPE_LOCK,
6975                                      0);
6976         if (*stid)
6977                 status = nfs_ok;
6978         else
6979                 status = nfserr_bad_stateid;
6980
6981         put_client_renew(found);
6982 out:
6983         nfs4_put_cpntf_state(nn, cps);
6984         return status;
6985 }
6986
6987 void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6988 {
6989         spin_lock(&nn->s2s_cp_lock);
6990         _free_cpntf_state_locked(nn, cps);
6991         spin_unlock(&nn->s2s_cp_lock);
6992 }
6993
6994 /**
6995  * nfs4_preprocess_stateid_op - find and prep stateid for an operation
6996  * @rqstp: incoming request from client
6997  * @cstate: current compound state
6998  * @fhp: filehandle associated with requested stateid
6999  * @stateid: stateid (provided by client)
7000  * @flags: flags describing type of operation to be done
7001  * @nfp: optional nfsd_file return pointer (may be NULL)
7002  * @cstid: optional returned nfs4_stid pointer (may be NULL)
7003  *
7004  * Given info from the client, look up a nfs4_stid for the operation. On
7005  * success, it returns a reference to the nfs4_stid and/or the nfsd_file
7006  * associated with it.
7007  */
7008 __be32
7009 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
7010                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
7011                 stateid_t *stateid, int flags, struct nfsd_file **nfp,
7012                 struct nfs4_stid **cstid)
7013 {
7014         struct net *net = SVC_NET(rqstp);
7015         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7016         struct nfs4_stid *s = NULL;
7017         __be32 status;
7018
7019         if (nfp)
7020                 *nfp = NULL;
7021
7022         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
7023                 if (cstid)
7024                         status = nfserr_bad_stateid;
7025                 else
7026                         status = check_special_stateids(net, fhp, stateid,
7027                                                                         flags);
7028                 goto done;
7029         }
7030
7031         status = nfsd4_lookup_stateid(cstate, stateid,
7032                                 SC_TYPE_DELEG|SC_TYPE_OPEN|SC_TYPE_LOCK,
7033                                 0, &s, nn);
7034         if (status == nfserr_bad_stateid)
7035                 status = find_cpntf_state(nn, stateid, &s);
7036         if (status)
7037                 return status;
7038         status = nfsd4_stid_check_stateid_generation(stateid, s,
7039                         nfsd4_has_session(cstate));
7040         if (status)
7041                 goto out;
7042
7043         switch (s->sc_type) {
7044         case SC_TYPE_DELEG:
7045                 status = nfs4_check_delegmode(delegstateid(s), flags);
7046                 break;
7047         case SC_TYPE_OPEN:
7048         case SC_TYPE_LOCK:
7049                 status = nfs4_check_olstateid(openlockstateid(s), flags);
7050                 break;
7051         }
7052         if (status)
7053                 goto out;
7054         status = nfs4_check_fh(fhp, s);
7055
7056 done:
7057         if (status == nfs_ok && nfp)
7058                 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
7059 out:
7060         if (s) {
7061                 if (!status && cstid)
7062                         *cstid = s;
7063                 else
7064                         nfs4_put_stid(s);
7065         }
7066         return status;
7067 }
7068
7069 /*
7070  * Test if the stateid is valid
7071  */
7072 __be32
7073 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7074                    union nfsd4_op_u *u)
7075 {
7076         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
7077         struct nfsd4_test_stateid_id *stateid;
7078         struct nfs4_client *cl = cstate->clp;
7079
7080         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
7081                 stateid->ts_id_status =
7082                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
7083
7084         return nfs_ok;
7085 }
7086
7087 static __be32
7088 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
7089 {
7090         struct nfs4_ol_stateid *stp = openlockstateid(s);
7091         __be32 ret;
7092
7093         ret = nfsd4_lock_ol_stateid(stp);
7094         if (ret)
7095                 goto out_put_stid;
7096
7097         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
7098         if (ret)
7099                 goto out;
7100
7101         ret = nfserr_locks_held;
7102         if (check_for_locks(stp->st_stid.sc_file,
7103                             lockowner(stp->st_stateowner)))
7104                 goto out;
7105
7106         release_lock_stateid(stp);
7107         ret = nfs_ok;
7108
7109 out:
7110         mutex_unlock(&stp->st_mutex);
7111 out_put_stid:
7112         nfs4_put_stid(s);
7113         return ret;
7114 }
7115
7116 __be32
7117 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7118                    union nfsd4_op_u *u)
7119 {
7120         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
7121         stateid_t *stateid = &free_stateid->fr_stateid;
7122         struct nfs4_stid *s;
7123         struct nfs4_delegation *dp;
7124         struct nfs4_client *cl = cstate->clp;
7125         __be32 ret = nfserr_bad_stateid;
7126
7127         spin_lock(&cl->cl_lock);
7128         s = find_stateid_locked(cl, stateid);
7129         if (!s || s->sc_status & SC_STATUS_CLOSED)
7130                 goto out_unlock;
7131         if (s->sc_status & SC_STATUS_ADMIN_REVOKED) {
7132                 nfsd4_drop_revoked_stid(s);
7133                 ret = nfs_ok;
7134                 goto out;
7135         }
7136         spin_lock(&s->sc_lock);
7137         switch (s->sc_type) {
7138         case SC_TYPE_DELEG:
7139                 if (s->sc_status & SC_STATUS_REVOKED) {
7140                         spin_unlock(&s->sc_lock);
7141                         dp = delegstateid(s);
7142                         list_del_init(&dp->dl_recall_lru);
7143                         spin_unlock(&cl->cl_lock);
7144                         nfs4_put_stid(s);
7145                         ret = nfs_ok;
7146                         goto out;
7147                 }
7148                 ret = nfserr_locks_held;
7149                 break;
7150         case SC_TYPE_OPEN:
7151                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
7152                 if (ret)
7153                         break;
7154                 ret = nfserr_locks_held;
7155                 break;
7156         case SC_TYPE_LOCK:
7157                 spin_unlock(&s->sc_lock);
7158                 refcount_inc(&s->sc_count);
7159                 spin_unlock(&cl->cl_lock);
7160                 ret = nfsd4_free_lock_stateid(stateid, s);
7161                 goto out;
7162         }
7163         spin_unlock(&s->sc_lock);
7164 out_unlock:
7165         spin_unlock(&cl->cl_lock);
7166 out:
7167         return ret;
7168 }
7169
7170 static inline int
7171 setlkflg (int type)
7172 {
7173         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
7174                 RD_STATE : WR_STATE;
7175 }
7176
7177 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
7178 {
7179         struct svc_fh *current_fh = &cstate->current_fh;
7180         struct nfs4_stateowner *sop = stp->st_stateowner;
7181         __be32 status;
7182
7183         status = nfsd4_check_seqid(cstate, sop, seqid);
7184         if (status)
7185                 return status;
7186         status = nfsd4_lock_ol_stateid(stp);
7187         if (status != nfs_ok)
7188                 return status;
7189         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
7190         if (status == nfs_ok)
7191                 status = nfs4_check_fh(current_fh, &stp->st_stid);
7192         if (status != nfs_ok)
7193                 mutex_unlock(&stp->st_mutex);
7194         return status;
7195 }
7196
7197 /**
7198  * nfs4_preprocess_seqid_op - find and prep an ol_stateid for a seqid-morphing op
7199  * @cstate: compund state
7200  * @seqid: seqid (provided by client)
7201  * @stateid: stateid (provided by client)
7202  * @typemask: mask of allowable types for this operation
7203  * @statusmask: mask of allowed states: 0 or STID_CLOSED
7204  * @stpp: return pointer for the stateid found
7205  * @nn: net namespace for request
7206  *
7207  * Given a stateid+seqid from a client, look up an nfs4_ol_stateid and
7208  * return it in @stpp. On a nfs_ok return, the returned stateid will
7209  * have its st_mutex locked.
7210  */
7211 static __be32
7212 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
7213                          stateid_t *stateid,
7214                          unsigned short typemask, unsigned short statusmask,
7215                          struct nfs4_ol_stateid **stpp,
7216                          struct nfsd_net *nn)
7217 {
7218         __be32 status;
7219         struct nfs4_stid *s;
7220         struct nfs4_ol_stateid *stp = NULL;
7221
7222         trace_nfsd_preprocess(seqid, stateid);
7223
7224         *stpp = NULL;
7225 retry:
7226         status = nfsd4_lookup_stateid(cstate, stateid,
7227                                       typemask, statusmask, &s, nn);
7228         if (status)
7229                 return status;
7230         stp = openlockstateid(s);
7231         if (nfsd4_cstate_assign_replay(cstate, stp->st_stateowner) == -EAGAIN) {
7232                 nfs4_put_stateowner(stp->st_stateowner);
7233                 goto retry;
7234         }
7235
7236         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
7237         if (!status)
7238                 *stpp = stp;
7239         else
7240                 nfs4_put_stid(&stp->st_stid);
7241         return status;
7242 }
7243
7244 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
7245                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
7246 {
7247         __be32 status;
7248         struct nfs4_openowner *oo;
7249         struct nfs4_ol_stateid *stp;
7250
7251         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
7252                                           SC_TYPE_OPEN, 0, &stp, nn);
7253         if (status)
7254                 return status;
7255         oo = openowner(stp->st_stateowner);
7256         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
7257                 mutex_unlock(&stp->st_mutex);
7258                 nfs4_put_stid(&stp->st_stid);
7259                 return nfserr_bad_stateid;
7260         }
7261         *stpp = stp;
7262         return nfs_ok;
7263 }
7264
7265 __be32
7266 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7267                    union nfsd4_op_u *u)
7268 {
7269         struct nfsd4_open_confirm *oc = &u->open_confirm;
7270         __be32 status;
7271         struct nfs4_openowner *oo;
7272         struct nfs4_ol_stateid *stp;
7273         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7274
7275         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
7276                         cstate->current_fh.fh_dentry);
7277
7278         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
7279         if (status)
7280                 return status;
7281
7282         status = nfs4_preprocess_seqid_op(cstate,
7283                                           oc->oc_seqid, &oc->oc_req_stateid,
7284                                           SC_TYPE_OPEN, 0, &stp, nn);
7285         if (status)
7286                 goto out;
7287         oo = openowner(stp->st_stateowner);
7288         status = nfserr_bad_stateid;
7289         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
7290                 mutex_unlock(&stp->st_mutex);
7291                 goto put_stateid;
7292         }
7293         oo->oo_flags |= NFS4_OO_CONFIRMED;
7294         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
7295         mutex_unlock(&stp->st_mutex);
7296         trace_nfsd_open_confirm(oc->oc_seqid, &stp->st_stid.sc_stateid);
7297         nfsd4_client_record_create(oo->oo_owner.so_client);
7298         status = nfs_ok;
7299 put_stateid:
7300         nfs4_put_stid(&stp->st_stid);
7301 out:
7302         nfsd4_bump_seqid(cstate, status);
7303         return status;
7304 }
7305
7306 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
7307 {
7308         if (!test_access(access, stp))
7309                 return;
7310         nfs4_file_put_access(stp->st_stid.sc_file, access);
7311         clear_access(access, stp);
7312 }
7313
7314 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
7315 {
7316         switch (to_access) {
7317         case NFS4_SHARE_ACCESS_READ:
7318                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
7319                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
7320                 break;
7321         case NFS4_SHARE_ACCESS_WRITE:
7322                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
7323                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
7324                 break;
7325         case NFS4_SHARE_ACCESS_BOTH:
7326                 break;
7327         default:
7328                 WARN_ON_ONCE(1);
7329         }
7330 }
7331
7332 __be32
7333 nfsd4_open_downgrade(struct svc_rqst *rqstp,
7334                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
7335 {
7336         struct nfsd4_open_downgrade *od = &u->open_downgrade;
7337         __be32 status;
7338         struct nfs4_ol_stateid *stp;
7339         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7340
7341         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
7342                         cstate->current_fh.fh_dentry);
7343
7344         /* We don't yet support WANT bits: */
7345         if (od->od_deleg_want)
7346                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
7347                         od->od_deleg_want);
7348
7349         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
7350                                         &od->od_stateid, &stp, nn);
7351         if (status)
7352                 goto out;
7353         status = nfserr_inval;
7354         if (!test_access(od->od_share_access, stp)) {
7355                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
7356                         stp->st_access_bmap, od->od_share_access);
7357                 goto put_stateid;
7358         }
7359         if (!test_deny(od->od_share_deny, stp)) {
7360                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
7361                         stp->st_deny_bmap, od->od_share_deny);
7362                 goto put_stateid;
7363         }
7364         nfs4_stateid_downgrade(stp, od->od_share_access);
7365         reset_union_bmap_deny(od->od_share_deny, stp);
7366         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
7367         status = nfs_ok;
7368 put_stateid:
7369         mutex_unlock(&stp->st_mutex);
7370         nfs4_put_stid(&stp->st_stid);
7371 out:
7372         nfsd4_bump_seqid(cstate, status);
7373         return status;
7374 }
7375
7376 static bool nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
7377 {
7378         struct nfs4_client *clp = s->st_stid.sc_client;
7379         bool unhashed;
7380         LIST_HEAD(reaplist);
7381         struct nfs4_ol_stateid *stp;
7382
7383         spin_lock(&clp->cl_lock);
7384         unhashed = unhash_open_stateid(s, &reaplist);
7385
7386         if (clp->cl_minorversion) {
7387                 if (unhashed)
7388                         put_ol_stateid_locked(s, &reaplist);
7389                 spin_unlock(&clp->cl_lock);
7390                 list_for_each_entry(stp, &reaplist, st_locks)
7391                         nfs4_free_cpntf_statelist(clp->net, &stp->st_stid);
7392                 free_ol_stateid_reaplist(&reaplist);
7393                 return false;
7394         } else {
7395                 spin_unlock(&clp->cl_lock);
7396                 free_ol_stateid_reaplist(&reaplist);
7397                 return unhashed;
7398         }
7399 }
7400
7401 /*
7402  * nfs4_unlock_state() called after encode
7403  */
7404 __be32
7405 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7406                 union nfsd4_op_u *u)
7407 {
7408         struct nfsd4_close *close = &u->close;
7409         __be32 status;
7410         struct nfs4_ol_stateid *stp;
7411         struct net *net = SVC_NET(rqstp);
7412         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7413         bool need_move_to_close_list;
7414
7415         dprintk("NFSD: nfsd4_close on file %pd\n",
7416                         cstate->current_fh.fh_dentry);
7417
7418         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
7419                                           &close->cl_stateid,
7420                                           SC_TYPE_OPEN, SC_STATUS_CLOSED,
7421                                           &stp, nn);
7422         nfsd4_bump_seqid(cstate, status);
7423         if (status)
7424                 goto out;
7425
7426         spin_lock(&stp->st_stid.sc_client->cl_lock);
7427         stp->st_stid.sc_status |= SC_STATUS_CLOSED;
7428         spin_unlock(&stp->st_stid.sc_client->cl_lock);
7429
7430         /*
7431          * Technically we don't _really_ have to increment or copy it, since
7432          * it should just be gone after this operation and we clobber the
7433          * copied value below, but we continue to do so here just to ensure
7434          * that racing ops see that there was a state change.
7435          */
7436         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
7437
7438         need_move_to_close_list = nfsd4_close_open_stateid(stp);
7439         mutex_unlock(&stp->st_mutex);
7440         if (need_move_to_close_list)
7441                 move_to_close_lru(stp, net);
7442
7443         /* v4.1+ suggests that we send a special stateid in here, since the
7444          * clients should just ignore this anyway. Since this is not useful
7445          * for v4.0 clients either, we set it to the special close_stateid
7446          * universally.
7447          *
7448          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
7449          */
7450         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
7451
7452         /* put reference from nfs4_preprocess_seqid_op */
7453         nfs4_put_stid(&stp->st_stid);
7454 out:
7455         return status;
7456 }
7457
7458 __be32
7459 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7460                   union nfsd4_op_u *u)
7461 {
7462         struct nfsd4_delegreturn *dr = &u->delegreturn;
7463         struct nfs4_delegation *dp;
7464         stateid_t *stateid = &dr->dr_stateid;
7465         struct nfs4_stid *s;
7466         __be32 status;
7467         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7468
7469         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
7470                 return status;
7471
7472         status = nfsd4_lookup_stateid(cstate, stateid, SC_TYPE_DELEG, 0, &s, nn);
7473         if (status)
7474                 goto out;
7475         dp = delegstateid(s);
7476         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
7477         if (status)
7478                 goto put_stateid;
7479
7480         trace_nfsd_deleg_return(stateid);
7481         wake_up_var(d_inode(cstate->current_fh.fh_dentry));
7482         destroy_delegation(dp);
7483 put_stateid:
7484         nfs4_put_stid(&dp->dl_stid);
7485 out:
7486         return status;
7487 }
7488
7489 /* last octet in a range */
7490 static inline u64
7491 last_byte_offset(u64 start, u64 len)
7492 {
7493         u64 end;
7494
7495         WARN_ON_ONCE(!len);
7496         end = start + len;
7497         return end > start ? end - 1: NFS4_MAX_UINT64;
7498 }
7499
7500 /*
7501  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
7502  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
7503  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
7504  * locking, this prevents us from being completely protocol-compliant.  The
7505  * real solution to this problem is to start using unsigned file offsets in
7506  * the VFS, but this is a very deep change!
7507  */
7508 static inline void
7509 nfs4_transform_lock_offset(struct file_lock *lock)
7510 {
7511         if (lock->fl_start < 0)
7512                 lock->fl_start = OFFSET_MAX;
7513         if (lock->fl_end < 0)
7514                 lock->fl_end = OFFSET_MAX;
7515 }
7516
7517 static fl_owner_t
7518 nfsd4_lm_get_owner(fl_owner_t owner)
7519 {
7520         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7521
7522         nfs4_get_stateowner(&lo->lo_owner);
7523         return owner;
7524 }
7525
7526 static void
7527 nfsd4_lm_put_owner(fl_owner_t owner)
7528 {
7529         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7530
7531         if (lo)
7532                 nfs4_put_stateowner(&lo->lo_owner);
7533 }
7534
7535 /* return pointer to struct nfs4_client if client is expirable */
7536 static bool
7537 nfsd4_lm_lock_expirable(struct file_lock *cfl)
7538 {
7539         struct nfs4_lockowner *lo = (struct nfs4_lockowner *) cfl->c.flc_owner;
7540         struct nfs4_client *clp = lo->lo_owner.so_client;
7541         struct nfsd_net *nn;
7542
7543         if (try_to_expire_client(clp)) {
7544                 nn = net_generic(clp->net, nfsd_net_id);
7545                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
7546                 return true;
7547         }
7548         return false;
7549 }
7550
7551 /* schedule laundromat to run immediately and wait for it to complete */
7552 static void
7553 nfsd4_lm_expire_lock(void)
7554 {
7555         flush_workqueue(laundry_wq);
7556 }
7557
7558 static void
7559 nfsd4_lm_notify(struct file_lock *fl)
7560 {
7561         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *) fl->c.flc_owner;
7562         struct net                      *net = lo->lo_owner.so_client->net;
7563         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
7564         struct nfsd4_blocked_lock       *nbl = container_of(fl,
7565                                                 struct nfsd4_blocked_lock, nbl_lock);
7566         bool queue = false;
7567
7568         /* An empty list means that something else is going to be using it */
7569         spin_lock(&nn->blocked_locks_lock);
7570         if (!list_empty(&nbl->nbl_list)) {
7571                 list_del_init(&nbl->nbl_list);
7572                 list_del_init(&nbl->nbl_lru);
7573                 queue = true;
7574         }
7575         spin_unlock(&nn->blocked_locks_lock);
7576
7577         if (queue) {
7578                 trace_nfsd_cb_notify_lock(lo, nbl);
7579                 nfsd4_run_cb(&nbl->nbl_cb);
7580         }
7581 }
7582
7583 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
7584         .lm_mod_owner = THIS_MODULE,
7585         .lm_notify = nfsd4_lm_notify,
7586         .lm_get_owner = nfsd4_lm_get_owner,
7587         .lm_put_owner = nfsd4_lm_put_owner,
7588         .lm_lock_expirable = nfsd4_lm_lock_expirable,
7589         .lm_expire_lock = nfsd4_lm_expire_lock,
7590 };
7591
7592 static inline void
7593 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
7594 {
7595         struct nfs4_lockowner *lo;
7596
7597         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
7598                 lo = (struct nfs4_lockowner *) fl->c.flc_owner;
7599                 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
7600                                                 GFP_KERNEL);
7601                 if (!deny->ld_owner.data)
7602                         /* We just don't care that much */
7603                         goto nevermind;
7604                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
7605         } else {
7606 nevermind:
7607                 deny->ld_owner.len = 0;
7608                 deny->ld_owner.data = NULL;
7609                 deny->ld_clientid.cl_boot = 0;
7610                 deny->ld_clientid.cl_id = 0;
7611         }
7612         deny->ld_start = fl->fl_start;
7613         deny->ld_length = NFS4_MAX_UINT64;
7614         if (fl->fl_end != NFS4_MAX_UINT64)
7615                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
7616         deny->ld_type = NFS4_READ_LT;
7617         if (fl->c.flc_type != F_RDLCK)
7618                 deny->ld_type = NFS4_WRITE_LT;
7619 }
7620
7621 static struct nfs4_lockowner *
7622 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
7623 {
7624         unsigned int strhashval = ownerstr_hashval(owner);
7625         struct nfs4_stateowner *so;
7626
7627         lockdep_assert_held(&clp->cl_lock);
7628
7629         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
7630                             so_strhash) {
7631                 if (so->so_is_open_owner)
7632                         continue;
7633                 if (same_owner_str(so, owner))
7634                         return lockowner(nfs4_get_stateowner(so));
7635         }
7636         return NULL;
7637 }
7638
7639 static struct nfs4_lockowner *
7640 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
7641 {
7642         struct nfs4_lockowner *lo;
7643
7644         spin_lock(&clp->cl_lock);
7645         lo = find_lockowner_str_locked(clp, owner);
7646         spin_unlock(&clp->cl_lock);
7647         return lo;
7648 }
7649
7650 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
7651 {
7652         unhash_lockowner_locked(lockowner(sop));
7653 }
7654
7655 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
7656 {
7657         struct nfs4_lockowner *lo = lockowner(sop);
7658
7659         kmem_cache_free(lockowner_slab, lo);
7660 }
7661
7662 static const struct nfs4_stateowner_operations lockowner_ops = {
7663         .so_unhash =    nfs4_unhash_lockowner,
7664         .so_free =      nfs4_free_lockowner,
7665 };
7666
7667 /*
7668  * Alloc a lock owner structure.
7669  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
7670  * occurred.
7671  *
7672  * strhashval = ownerstr_hashval
7673  */
7674 static struct nfs4_lockowner *
7675 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
7676                            struct nfs4_ol_stateid *open_stp,
7677                            struct nfsd4_lock *lock)
7678 {
7679         struct nfs4_lockowner *lo, *ret;
7680
7681         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
7682         if (!lo)
7683                 return NULL;
7684         INIT_LIST_HEAD(&lo->lo_blocked);
7685         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
7686         lo->lo_owner.so_is_open_owner = 0;
7687         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
7688         lo->lo_owner.so_ops = &lockowner_ops;
7689         spin_lock(&clp->cl_lock);
7690         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
7691         if (ret == NULL) {
7692                 list_add(&lo->lo_owner.so_strhash,
7693                          &clp->cl_ownerstr_hashtbl[strhashval]);
7694                 ret = lo;
7695         } else
7696                 nfs4_free_stateowner(&lo->lo_owner);
7697
7698         spin_unlock(&clp->cl_lock);
7699         return ret;
7700 }
7701
7702 static struct nfs4_ol_stateid *
7703 find_lock_stateid(const struct nfs4_lockowner *lo,
7704                   const struct nfs4_ol_stateid *ost)
7705 {
7706         struct nfs4_ol_stateid *lst;
7707
7708         lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
7709
7710         /* If ost is not hashed, ost->st_locks will not be valid */
7711         if (!nfs4_ol_stateid_unhashed(ost))
7712                 list_for_each_entry(lst, &ost->st_locks, st_locks) {
7713                         if (lst->st_stateowner == &lo->lo_owner) {
7714                                 refcount_inc(&lst->st_stid.sc_count);
7715                                 return lst;
7716                         }
7717                 }
7718         return NULL;
7719 }
7720
7721 static struct nfs4_ol_stateid *
7722 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
7723                   struct nfs4_file *fp, struct inode *inode,
7724                   struct nfs4_ol_stateid *open_stp)
7725 {
7726         struct nfs4_client *clp = lo->lo_owner.so_client;
7727         struct nfs4_ol_stateid *retstp;
7728
7729         mutex_init(&stp->st_mutex);
7730         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
7731 retry:
7732         spin_lock(&clp->cl_lock);
7733         if (nfs4_ol_stateid_unhashed(open_stp))
7734                 goto out_close;
7735         retstp = find_lock_stateid(lo, open_stp);
7736         if (retstp)
7737                 goto out_found;
7738         refcount_inc(&stp->st_stid.sc_count);
7739         stp->st_stid.sc_type = SC_TYPE_LOCK;
7740         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
7741         get_nfs4_file(fp);
7742         stp->st_stid.sc_file = fp;
7743         stp->st_access_bmap = 0;
7744         stp->st_deny_bmap = open_stp->st_deny_bmap;
7745         stp->st_openstp = open_stp;
7746         spin_lock(&fp->fi_lock);
7747         list_add(&stp->st_locks, &open_stp->st_locks);
7748         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
7749         list_add(&stp->st_perfile, &fp->fi_stateids);
7750         spin_unlock(&fp->fi_lock);
7751         spin_unlock(&clp->cl_lock);
7752         return stp;
7753 out_found:
7754         spin_unlock(&clp->cl_lock);
7755         if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
7756                 nfs4_put_stid(&retstp->st_stid);
7757                 goto retry;
7758         }
7759         /* To keep mutex tracking happy */
7760         mutex_unlock(&stp->st_mutex);
7761         return retstp;
7762 out_close:
7763         spin_unlock(&clp->cl_lock);
7764         mutex_unlock(&stp->st_mutex);
7765         return NULL;
7766 }
7767
7768 static struct nfs4_ol_stateid *
7769 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
7770                             struct inode *inode, struct nfs4_ol_stateid *ost,
7771                             bool *new)
7772 {
7773         struct nfs4_stid *ns = NULL;
7774         struct nfs4_ol_stateid *lst;
7775         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7776         struct nfs4_client *clp = oo->oo_owner.so_client;
7777
7778         *new = false;
7779         spin_lock(&clp->cl_lock);
7780         lst = find_lock_stateid(lo, ost);
7781         spin_unlock(&clp->cl_lock);
7782         if (lst != NULL) {
7783                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
7784                         goto out;
7785                 nfs4_put_stid(&lst->st_stid);
7786         }
7787         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
7788         if (ns == NULL)
7789                 return NULL;
7790
7791         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
7792         if (lst == openlockstateid(ns))
7793                 *new = true;
7794         else
7795                 nfs4_put_stid(ns);
7796 out:
7797         return lst;
7798 }
7799
7800 static int
7801 check_lock_length(u64 offset, u64 length)
7802 {
7803         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
7804                 (length > ~offset)));
7805 }
7806
7807 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
7808 {
7809         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
7810
7811         lockdep_assert_held(&fp->fi_lock);
7812
7813         if (test_access(access, lock_stp))
7814                 return;
7815         __nfs4_file_get_access(fp, access);
7816         set_access(access, lock_stp);
7817 }
7818
7819 static __be32
7820 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
7821                             struct nfs4_ol_stateid *ost,
7822                             struct nfsd4_lock *lock,
7823                             struct nfs4_ol_stateid **plst, bool *new)
7824 {
7825         __be32 status;
7826         struct nfs4_file *fi = ost->st_stid.sc_file;
7827         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7828         struct nfs4_client *cl = oo->oo_owner.so_client;
7829         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
7830         struct nfs4_lockowner *lo;
7831         struct nfs4_ol_stateid *lst;
7832         unsigned int strhashval;
7833
7834         lo = find_lockowner_str(cl, &lock->lk_new_owner);
7835         if (!lo) {
7836                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
7837                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
7838                 if (lo == NULL)
7839                         return nfserr_jukebox;
7840         } else {
7841                 /* with an existing lockowner, seqids must be the same */
7842                 status = nfserr_bad_seqid;
7843                 if (!cstate->minorversion &&
7844                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
7845                         goto out;
7846         }
7847
7848         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
7849         if (lst == NULL) {
7850                 status = nfserr_jukebox;
7851                 goto out;
7852         }
7853
7854         status = nfs_ok;
7855         *plst = lst;
7856 out:
7857         nfs4_put_stateowner(&lo->lo_owner);
7858         return status;
7859 }
7860
7861 /*
7862  *  LOCK operation
7863  */
7864 __be32
7865 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7866            union nfsd4_op_u *u)
7867 {
7868         struct nfsd4_lock *lock = &u->lock;
7869         struct nfs4_openowner *open_sop = NULL;
7870         struct nfs4_lockowner *lock_sop = NULL;
7871         struct nfs4_ol_stateid *lock_stp = NULL;
7872         struct nfs4_ol_stateid *open_stp = NULL;
7873         struct nfs4_file *fp;
7874         struct nfsd_file *nf = NULL;
7875         struct nfsd4_blocked_lock *nbl = NULL;
7876         struct file_lock *file_lock = NULL;
7877         struct file_lock *conflock = NULL;
7878         struct super_block *sb;
7879         __be32 status = 0;
7880         int lkflg;
7881         int err;
7882         bool new = false;
7883         unsigned char type;
7884         unsigned int flags = FL_POSIX;
7885         struct net *net = SVC_NET(rqstp);
7886         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7887
7888         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
7889                 (long long) lock->lk_offset,
7890                 (long long) lock->lk_length);
7891
7892         if (check_lock_length(lock->lk_offset, lock->lk_length))
7893                  return nfserr_inval;
7894
7895         if ((status = fh_verify(rqstp, &cstate->current_fh,
7896                                 S_IFREG, NFSD_MAY_LOCK))) {
7897                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
7898                 return status;
7899         }
7900         sb = cstate->current_fh.fh_dentry->d_sb;
7901
7902         if (lock->lk_is_new) {
7903                 if (nfsd4_has_session(cstate))
7904                         /* See rfc 5661 18.10.3: given clientid is ignored: */
7905                         memcpy(&lock->lk_new_clientid,
7906                                 &cstate->clp->cl_clientid,
7907                                 sizeof(clientid_t));
7908
7909                 /* validate and update open stateid and open seqid */
7910                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
7911                                         lock->lk_new_open_seqid,
7912                                         &lock->lk_new_open_stateid,
7913                                         &open_stp, nn);
7914                 if (status)
7915                         goto out;
7916                 mutex_unlock(&open_stp->st_mutex);
7917                 open_sop = openowner(open_stp->st_stateowner);
7918                 status = nfserr_bad_stateid;
7919                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
7920                                                 &lock->lk_new_clientid))
7921                         goto out;
7922                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
7923                                                         &lock_stp, &new);
7924         } else {
7925                 status = nfs4_preprocess_seqid_op(cstate,
7926                                                   lock->lk_old_lock_seqid,
7927                                                   &lock->lk_old_lock_stateid,
7928                                                   SC_TYPE_LOCK, 0, &lock_stp,
7929                                                   nn);
7930         }
7931         if (status)
7932                 goto out;
7933         lock_sop = lockowner(lock_stp->st_stateowner);
7934
7935         lkflg = setlkflg(lock->lk_type);
7936         status = nfs4_check_openmode(lock_stp, lkflg);
7937         if (status)
7938                 goto out;
7939
7940         status = nfserr_grace;
7941         if (locks_in_grace(net) && !lock->lk_reclaim)
7942                 goto out;
7943         status = nfserr_no_grace;
7944         if (!locks_in_grace(net) && lock->lk_reclaim)
7945                 goto out;
7946
7947         if (lock->lk_reclaim)
7948                 flags |= FL_RECLAIM;
7949
7950         fp = lock_stp->st_stid.sc_file;
7951         switch (lock->lk_type) {
7952                 case NFS4_READW_LT:
7953                         if (nfsd4_has_session(cstate) ||
7954                             exportfs_lock_op_is_async(sb->s_export_op))
7955                                 flags |= FL_SLEEP;
7956                         fallthrough;
7957                 case NFS4_READ_LT:
7958                         spin_lock(&fp->fi_lock);
7959                         nf = find_readable_file_locked(fp);
7960                         if (nf)
7961                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
7962                         spin_unlock(&fp->fi_lock);
7963                         type = F_RDLCK;
7964                         break;
7965                 case NFS4_WRITEW_LT:
7966                         if (nfsd4_has_session(cstate) ||
7967                             exportfs_lock_op_is_async(sb->s_export_op))
7968                                 flags |= FL_SLEEP;
7969                         fallthrough;
7970                 case NFS4_WRITE_LT:
7971                         spin_lock(&fp->fi_lock);
7972                         nf = find_writeable_file_locked(fp);
7973                         if (nf)
7974                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
7975                         spin_unlock(&fp->fi_lock);
7976                         type = F_WRLCK;
7977                         break;
7978                 default:
7979                         status = nfserr_inval;
7980                 goto out;
7981         }
7982
7983         if (!nf) {
7984                 status = nfserr_openmode;
7985                 goto out;
7986         }
7987
7988         /*
7989          * Most filesystems with their own ->lock operations will block
7990          * the nfsd thread waiting to acquire the lock.  That leads to
7991          * deadlocks (we don't want every nfsd thread tied up waiting
7992          * for file locks), so don't attempt blocking lock notifications
7993          * on those filesystems:
7994          */
7995         if (!exportfs_lock_op_is_async(sb->s_export_op))
7996                 flags &= ~FL_SLEEP;
7997
7998         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
7999         if (!nbl) {
8000                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
8001                 status = nfserr_jukebox;
8002                 goto out;
8003         }
8004
8005         file_lock = &nbl->nbl_lock;
8006         file_lock->c.flc_type = type;
8007         file_lock->c.flc_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
8008         file_lock->c.flc_pid = current->tgid;
8009         file_lock->c.flc_file = nf->nf_file;
8010         file_lock->c.flc_flags = flags;
8011         file_lock->fl_lmops = &nfsd_posix_mng_ops;
8012         file_lock->fl_start = lock->lk_offset;
8013         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
8014         nfs4_transform_lock_offset(file_lock);
8015
8016         conflock = locks_alloc_lock();
8017         if (!conflock) {
8018                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8019                 status = nfserr_jukebox;
8020                 goto out;
8021         }
8022
8023         if (flags & FL_SLEEP) {
8024                 nbl->nbl_time = ktime_get_boottime_seconds();
8025                 spin_lock(&nn->blocked_locks_lock);
8026                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
8027                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
8028                 kref_get(&nbl->nbl_kref);
8029                 spin_unlock(&nn->blocked_locks_lock);
8030         }
8031
8032         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
8033         switch (err) {
8034         case 0: /* success! */
8035                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
8036                 status = 0;
8037                 if (lock->lk_reclaim)
8038                         nn->somebody_reclaimed = true;
8039                 break;
8040         case FILE_LOCK_DEFERRED:
8041                 kref_put(&nbl->nbl_kref, free_nbl);
8042                 nbl = NULL;
8043                 fallthrough;
8044         case -EAGAIN:           /* conflock holds conflicting lock */
8045                 status = nfserr_denied;
8046                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
8047                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
8048                 break;
8049         case -EDEADLK:
8050                 status = nfserr_deadlock;
8051                 break;
8052         default:
8053                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
8054                 status = nfserrno(err);
8055                 break;
8056         }
8057 out:
8058         if (nbl) {
8059                 /* dequeue it if we queued it before */
8060                 if (flags & FL_SLEEP) {
8061                         spin_lock(&nn->blocked_locks_lock);
8062                         if (!list_empty(&nbl->nbl_list) &&
8063                             !list_empty(&nbl->nbl_lru)) {
8064                                 list_del_init(&nbl->nbl_list);
8065                                 list_del_init(&nbl->nbl_lru);
8066                                 kref_put(&nbl->nbl_kref, free_nbl);
8067                         }
8068                         /* nbl can use one of lists to be linked to reaplist */
8069                         spin_unlock(&nn->blocked_locks_lock);
8070                 }
8071                 free_blocked_lock(nbl);
8072         }
8073         if (nf)
8074                 nfsd_file_put(nf);
8075         if (lock_stp) {
8076                 /* Bump seqid manually if the 4.0 replay owner is openowner */
8077                 if (cstate->replay_owner &&
8078                     cstate->replay_owner != &lock_sop->lo_owner &&
8079                     seqid_mutating_err(ntohl(status)))
8080                         lock_sop->lo_owner.so_seqid++;
8081
8082                 /*
8083                  * If this is a new, never-before-used stateid, and we are
8084                  * returning an error, then just go ahead and release it.
8085                  */
8086                 if (status && new)
8087                         release_lock_stateid(lock_stp);
8088
8089                 mutex_unlock(&lock_stp->st_mutex);
8090
8091                 nfs4_put_stid(&lock_stp->st_stid);
8092         }
8093         if (open_stp)
8094                 nfs4_put_stid(&open_stp->st_stid);
8095         nfsd4_bump_seqid(cstate, status);
8096         if (conflock)
8097                 locks_free_lock(conflock);
8098         return status;
8099 }
8100
8101 void nfsd4_lock_release(union nfsd4_op_u *u)
8102 {
8103         struct nfsd4_lock *lock = &u->lock;
8104         struct nfsd4_lock_denied *deny = &lock->lk_denied;
8105
8106         kfree(deny->ld_owner.data);
8107 }
8108
8109 /*
8110  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
8111  * so we do a temporary open here just to get an open file to pass to
8112  * vfs_test_lock.
8113  */
8114 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
8115 {
8116         struct nfsd_file *nf;
8117         struct inode *inode;
8118         __be32 err;
8119
8120         err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
8121         if (err)
8122                 return err;
8123         inode = fhp->fh_dentry->d_inode;
8124         inode_lock(inode); /* to block new leases till after test_lock: */
8125         err = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
8126         if (err)
8127                 goto out;
8128         lock->c.flc_file = nf->nf_file;
8129         err = nfserrno(vfs_test_lock(nf->nf_file, lock));
8130         lock->c.flc_file = NULL;
8131 out:
8132         inode_unlock(inode);
8133         nfsd_file_put(nf);
8134         return err;
8135 }
8136
8137 /*
8138  * LOCKT operation
8139  */
8140 __be32
8141 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
8142             union nfsd4_op_u *u)
8143 {
8144         struct nfsd4_lockt *lockt = &u->lockt;
8145         struct file_lock *file_lock = NULL;
8146         struct nfs4_lockowner *lo = NULL;
8147         __be32 status;
8148         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8149
8150         if (locks_in_grace(SVC_NET(rqstp)))
8151                 return nfserr_grace;
8152
8153         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
8154                  return nfserr_inval;
8155
8156         if (!nfsd4_has_session(cstate)) {
8157                 status = set_client(&lockt->lt_clientid, cstate, nn);
8158                 if (status)
8159                         goto out;
8160         }
8161
8162         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
8163                 goto out;
8164
8165         file_lock = locks_alloc_lock();
8166         if (!file_lock) {
8167                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8168                 status = nfserr_jukebox;
8169                 goto out;
8170         }
8171
8172         switch (lockt->lt_type) {
8173                 case NFS4_READ_LT:
8174                 case NFS4_READW_LT:
8175                         file_lock->c.flc_type = F_RDLCK;
8176                         break;
8177                 case NFS4_WRITE_LT:
8178                 case NFS4_WRITEW_LT:
8179                         file_lock->c.flc_type = F_WRLCK;
8180                         break;
8181                 default:
8182                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
8183                         status = nfserr_inval;
8184                         goto out;
8185         }
8186
8187         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
8188         if (lo)
8189                 file_lock->c.flc_owner = (fl_owner_t)lo;
8190         file_lock->c.flc_pid = current->tgid;
8191         file_lock->c.flc_flags = FL_POSIX;
8192
8193         file_lock->fl_start = lockt->lt_offset;
8194         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
8195
8196         nfs4_transform_lock_offset(file_lock);
8197
8198         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
8199         if (status)
8200                 goto out;
8201
8202         if (file_lock->c.flc_type != F_UNLCK) {
8203                 status = nfserr_denied;
8204                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
8205         }
8206 out:
8207         if (lo)
8208                 nfs4_put_stateowner(&lo->lo_owner);
8209         if (file_lock)
8210                 locks_free_lock(file_lock);
8211         return status;
8212 }
8213
8214 void nfsd4_lockt_release(union nfsd4_op_u *u)
8215 {
8216         struct nfsd4_lockt *lockt = &u->lockt;
8217         struct nfsd4_lock_denied *deny = &lockt->lt_denied;
8218
8219         kfree(deny->ld_owner.data);
8220 }
8221
8222 __be32
8223 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
8224             union nfsd4_op_u *u)
8225 {
8226         struct nfsd4_locku *locku = &u->locku;
8227         struct nfs4_ol_stateid *stp;
8228         struct nfsd_file *nf = NULL;
8229         struct file_lock *file_lock = NULL;
8230         __be32 status;
8231         int err;
8232         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8233
8234         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
8235                 (long long) locku->lu_offset,
8236                 (long long) locku->lu_length);
8237
8238         if (check_lock_length(locku->lu_offset, locku->lu_length))
8239                  return nfserr_inval;
8240
8241         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
8242                                           &locku->lu_stateid, SC_TYPE_LOCK, 0,
8243                                           &stp, nn);
8244         if (status)
8245                 goto out;
8246         nf = find_any_file(stp->st_stid.sc_file);
8247         if (!nf) {
8248                 status = nfserr_lock_range;
8249                 goto put_stateid;
8250         }
8251         file_lock = locks_alloc_lock();
8252         if (!file_lock) {
8253                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8254                 status = nfserr_jukebox;
8255                 goto put_file;
8256         }
8257
8258         file_lock->c.flc_type = F_UNLCK;
8259         file_lock->c.flc_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
8260         file_lock->c.flc_pid = current->tgid;
8261         file_lock->c.flc_file = nf->nf_file;
8262         file_lock->c.flc_flags = FL_POSIX;
8263         file_lock->fl_lmops = &nfsd_posix_mng_ops;
8264         file_lock->fl_start = locku->lu_offset;
8265
8266         file_lock->fl_end = last_byte_offset(locku->lu_offset,
8267                                                 locku->lu_length);
8268         nfs4_transform_lock_offset(file_lock);
8269
8270         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
8271         if (err) {
8272                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
8273                 goto out_nfserr;
8274         }
8275         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
8276 put_file:
8277         nfsd_file_put(nf);
8278 put_stateid:
8279         mutex_unlock(&stp->st_mutex);
8280         nfs4_put_stid(&stp->st_stid);
8281 out:
8282         nfsd4_bump_seqid(cstate, status);
8283         if (file_lock)
8284                 locks_free_lock(file_lock);
8285         return status;
8286
8287 out_nfserr:
8288         status = nfserrno(err);
8289         goto put_file;
8290 }
8291
8292 /*
8293  * returns
8294  *      true:  locks held by lockowner
8295  *      false: no locks held by lockowner
8296  */
8297 static bool
8298 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
8299 {
8300         struct file_lock *fl;
8301         int status = false;
8302         struct nfsd_file *nf;
8303         struct inode *inode;
8304         struct file_lock_context *flctx;
8305
8306         spin_lock(&fp->fi_lock);
8307         nf = find_any_file_locked(fp);
8308         if (!nf) {
8309                 /* Any valid lock stateid should have some sort of access */
8310                 WARN_ON_ONCE(1);
8311                 goto out;
8312         }
8313
8314         inode = file_inode(nf->nf_file);
8315         flctx = locks_inode_context(inode);
8316
8317         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
8318                 spin_lock(&flctx->flc_lock);
8319                 for_each_file_lock(fl, &flctx->flc_posix) {
8320                         if (fl->c.flc_owner == (fl_owner_t)lowner) {
8321                                 status = true;
8322                                 break;
8323                         }
8324                 }
8325                 spin_unlock(&flctx->flc_lock);
8326         }
8327 out:
8328         spin_unlock(&fp->fi_lock);
8329         return status;
8330 }
8331
8332 /**
8333  * nfsd4_release_lockowner - process NFSv4.0 RELEASE_LOCKOWNER operations
8334  * @rqstp: RPC transaction
8335  * @cstate: NFSv4 COMPOUND state
8336  * @u: RELEASE_LOCKOWNER arguments
8337  *
8338  * Check if theree are any locks still held and if not - free the lockowner
8339  * and any lock state that is owned.
8340  *
8341  * Return values:
8342  *   %nfs_ok: lockowner released or not found
8343  *   %nfserr_locks_held: lockowner still in use
8344  *   %nfserr_stale_clientid: clientid no longer active
8345  *   %nfserr_expired: clientid not recognized
8346  */
8347 __be32
8348 nfsd4_release_lockowner(struct svc_rqst *rqstp,
8349                         struct nfsd4_compound_state *cstate,
8350                         union nfsd4_op_u *u)
8351 {
8352         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
8353         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8354         clientid_t *clid = &rlockowner->rl_clientid;
8355         struct nfs4_ol_stateid *stp;
8356         struct nfs4_lockowner *lo;
8357         struct nfs4_client *clp;
8358         LIST_HEAD(reaplist);
8359         __be32 status;
8360
8361         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
8362                 clid->cl_boot, clid->cl_id);
8363
8364         status = set_client(clid, cstate, nn);
8365         if (status)
8366                 return status;
8367         clp = cstate->clp;
8368
8369         spin_lock(&clp->cl_lock);
8370         lo = find_lockowner_str_locked(clp, &rlockowner->rl_owner);
8371         if (!lo) {
8372                 spin_unlock(&clp->cl_lock);
8373                 return nfs_ok;
8374         }
8375
8376         list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
8377                 if (check_for_locks(stp->st_stid.sc_file, lo)) {
8378                         spin_unlock(&clp->cl_lock);
8379                         nfs4_put_stateowner(&lo->lo_owner);
8380                         return nfserr_locks_held;
8381                 }
8382         }
8383         unhash_lockowner_locked(lo);
8384         while (!list_empty(&lo->lo_owner.so_stateids)) {
8385                 stp = list_first_entry(&lo->lo_owner.so_stateids,
8386                                        struct nfs4_ol_stateid,
8387                                        st_perstateowner);
8388                 unhash_lock_stateid(stp);
8389                 put_ol_stateid_locked(stp, &reaplist);
8390         }
8391         spin_unlock(&clp->cl_lock);
8392
8393         free_ol_stateid_reaplist(&reaplist);
8394         remove_blocked_locks(lo);
8395         nfs4_put_stateowner(&lo->lo_owner);
8396         return nfs_ok;
8397 }
8398
8399 static inline struct nfs4_client_reclaim *
8400 alloc_reclaim(void)
8401 {
8402         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
8403 }
8404
8405 bool
8406 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
8407 {
8408         struct nfs4_client_reclaim *crp;
8409
8410         crp = nfsd4_find_reclaim_client(name, nn);
8411         return (crp && crp->cr_clp);
8412 }
8413
8414 /*
8415  * failure => all reset bets are off, nfserr_no_grace...
8416  *
8417  * The caller is responsible for freeing name.data if NULL is returned (it
8418  * will be freed in nfs4_remove_reclaim_record in the normal case).
8419  */
8420 struct nfs4_client_reclaim *
8421 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
8422                 struct nfsd_net *nn)
8423 {
8424         unsigned int strhashval;
8425         struct nfs4_client_reclaim *crp;
8426
8427         crp = alloc_reclaim();
8428         if (crp) {
8429                 strhashval = clientstr_hashval(name);
8430                 INIT_LIST_HEAD(&crp->cr_strhash);
8431                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
8432                 crp->cr_name.data = name.data;
8433                 crp->cr_name.len = name.len;
8434                 crp->cr_princhash.data = princhash.data;
8435                 crp->cr_princhash.len = princhash.len;
8436                 crp->cr_clp = NULL;
8437                 nn->reclaim_str_hashtbl_size++;
8438         }
8439         return crp;
8440 }
8441
8442 void
8443 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
8444 {
8445         list_del(&crp->cr_strhash);
8446         kfree(crp->cr_name.data);
8447         kfree(crp->cr_princhash.data);
8448         kfree(crp);
8449         nn->reclaim_str_hashtbl_size--;
8450 }
8451
8452 void
8453 nfs4_release_reclaim(struct nfsd_net *nn)
8454 {
8455         struct nfs4_client_reclaim *crp = NULL;
8456         int i;
8457
8458         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8459                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
8460                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
8461                                         struct nfs4_client_reclaim, cr_strhash);
8462                         nfs4_remove_reclaim_record(crp, nn);
8463                 }
8464         }
8465         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
8466 }
8467
8468 /*
8469  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
8470 struct nfs4_client_reclaim *
8471 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
8472 {
8473         unsigned int strhashval;
8474         struct nfs4_client_reclaim *crp = NULL;
8475
8476         strhashval = clientstr_hashval(name);
8477         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
8478                 if (compare_blob(&crp->cr_name, &name) == 0) {
8479                         return crp;
8480                 }
8481         }
8482         return NULL;
8483 }
8484
8485 __be32
8486 nfs4_check_open_reclaim(struct nfs4_client *clp)
8487 {
8488         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
8489                 return nfserr_no_grace;
8490
8491         if (nfsd4_client_record_check(clp))
8492                 return nfserr_reclaim_bad;
8493
8494         return nfs_ok;
8495 }
8496
8497 /*
8498  * Since the lifetime of a delegation isn't limited to that of an open, a
8499  * client may quite reasonably hang on to a delegation as long as it has
8500  * the inode cached.  This becomes an obvious problem the first time a
8501  * client's inode cache approaches the size of the server's total memory.
8502  *
8503  * For now we avoid this problem by imposing a hard limit on the number
8504  * of delegations, which varies according to the server's memory size.
8505  */
8506 static void
8507 set_max_delegations(void)
8508 {
8509         /*
8510          * Allow at most 4 delegations per megabyte of RAM.  Quick
8511          * estimates suggest that in the worst case (where every delegation
8512          * is for a different inode), a delegation could take about 1.5K,
8513          * giving a worst case usage of about 6% of memory.
8514          */
8515         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
8516 }
8517
8518 static int nfs4_state_create_net(struct net *net)
8519 {
8520         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8521         int i;
8522
8523         nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8524                                             sizeof(struct list_head),
8525                                             GFP_KERNEL);
8526         if (!nn->conf_id_hashtbl)
8527                 goto err;
8528         nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8529                                               sizeof(struct list_head),
8530                                               GFP_KERNEL);
8531         if (!nn->unconf_id_hashtbl)
8532                 goto err_unconf_id;
8533         nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
8534                                               sizeof(struct list_head),
8535                                               GFP_KERNEL);
8536         if (!nn->sessionid_hashtbl)
8537                 goto err_sessionid;
8538
8539         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8540                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
8541                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
8542         }
8543         for (i = 0; i < SESSION_HASH_SIZE; i++)
8544                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
8545         nn->conf_name_tree = RB_ROOT;
8546         nn->unconf_name_tree = RB_ROOT;
8547         nn->boot_time = ktime_get_real_seconds();
8548         nn->grace_ended = false;
8549         nn->nfsd4_manager.block_opens = true;
8550         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
8551         INIT_LIST_HEAD(&nn->client_lru);
8552         INIT_LIST_HEAD(&nn->close_lru);
8553         INIT_LIST_HEAD(&nn->del_recall_lru);
8554         spin_lock_init(&nn->client_lock);
8555         spin_lock_init(&nn->s2s_cp_lock);
8556         idr_init(&nn->s2s_cp_stateids);
8557
8558         spin_lock_init(&nn->blocked_locks_lock);
8559         INIT_LIST_HEAD(&nn->blocked_locks_lru);
8560
8561         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
8562         INIT_WORK(&nn->nfsd_shrinker_work, nfsd4_state_shrinker_worker);
8563         get_net(net);
8564
8565         nn->nfsd_client_shrinker = shrinker_alloc(0, "nfsd-client");
8566         if (!nn->nfsd_client_shrinker)
8567                 goto err_shrinker;
8568
8569         nn->nfsd_client_shrinker->scan_objects = nfsd4_state_shrinker_scan;
8570         nn->nfsd_client_shrinker->count_objects = nfsd4_state_shrinker_count;
8571         nn->nfsd_client_shrinker->private_data = nn;
8572
8573         shrinker_register(nn->nfsd_client_shrinker);
8574
8575         return 0;
8576
8577 err_shrinker:
8578         put_net(net);
8579         kfree(nn->sessionid_hashtbl);
8580 err_sessionid:
8581         kfree(nn->unconf_id_hashtbl);
8582 err_unconf_id:
8583         kfree(nn->conf_id_hashtbl);
8584 err:
8585         return -ENOMEM;
8586 }
8587
8588 static void
8589 nfs4_state_destroy_net(struct net *net)
8590 {
8591         int i;
8592         struct nfs4_client *clp = NULL;
8593         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8594
8595         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8596                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
8597                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8598                         destroy_client(clp);
8599                 }
8600         }
8601
8602         WARN_ON(!list_empty(&nn->blocked_locks_lru));
8603
8604         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8605                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
8606                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8607                         destroy_client(clp);
8608                 }
8609         }
8610
8611         kfree(nn->sessionid_hashtbl);
8612         kfree(nn->unconf_id_hashtbl);
8613         kfree(nn->conf_id_hashtbl);
8614         put_net(net);
8615 }
8616
8617 int
8618 nfs4_state_start_net(struct net *net)
8619 {
8620         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8621         int ret;
8622
8623         ret = nfs4_state_create_net(net);
8624         if (ret)
8625                 return ret;
8626         locks_start_grace(net, &nn->nfsd4_manager);
8627         nfsd4_client_tracking_init(net);
8628         if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
8629                 goto skip_grace;
8630         printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
8631                nn->nfsd4_grace, net->ns.inum);
8632         trace_nfsd_grace_start(nn);
8633         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
8634         return 0;
8635
8636 skip_grace:
8637         printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
8638                         net->ns.inum);
8639         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
8640         nfsd4_end_grace(nn);
8641         return 0;
8642 }
8643
8644 /* initialization to perform when the nfsd service is started: */
8645
8646 int
8647 nfs4_state_start(void)
8648 {
8649         int ret;
8650
8651         ret = rhltable_init(&nfs4_file_rhltable, &nfs4_file_rhash_params);
8652         if (ret)
8653                 return ret;
8654
8655         set_max_delegations();
8656         return 0;
8657 }
8658
8659 void
8660 nfs4_state_shutdown_net(struct net *net)
8661 {
8662         struct nfs4_delegation *dp = NULL;
8663         struct list_head *pos, *next, reaplist;
8664         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8665
8666         shrinker_free(nn->nfsd_client_shrinker);
8667         cancel_work(&nn->nfsd_shrinker_work);
8668         cancel_delayed_work_sync(&nn->laundromat_work);
8669         locks_end_grace(&nn->nfsd4_manager);
8670
8671         INIT_LIST_HEAD(&reaplist);
8672         spin_lock(&state_lock);
8673         list_for_each_safe(pos, next, &nn->del_recall_lru) {
8674                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8675                 unhash_delegation_locked(dp, SC_STATUS_CLOSED);
8676                 list_add(&dp->dl_recall_lru, &reaplist);
8677         }
8678         spin_unlock(&state_lock);
8679         list_for_each_safe(pos, next, &reaplist) {
8680                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8681                 list_del_init(&dp->dl_recall_lru);
8682                 destroy_unhashed_deleg(dp);
8683         }
8684
8685         nfsd4_client_tracking_exit(net);
8686         nfs4_state_destroy_net(net);
8687 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
8688         nfsd4_ssc_shutdown_umount(nn);
8689 #endif
8690 }
8691
8692 void
8693 nfs4_state_shutdown(void)
8694 {
8695         rhltable_destroy(&nfs4_file_rhltable);
8696 }
8697
8698 static void
8699 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8700 {
8701         if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
8702             CURRENT_STATEID(stateid))
8703                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
8704 }
8705
8706 static void
8707 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8708 {
8709         if (cstate->minorversion) {
8710                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
8711                 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8712         }
8713 }
8714
8715 void
8716 clear_current_stateid(struct nfsd4_compound_state *cstate)
8717 {
8718         CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8719 }
8720
8721 /*
8722  * functions to set current state id
8723  */
8724 void
8725 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
8726                 union nfsd4_op_u *u)
8727 {
8728         put_stateid(cstate, &u->open_downgrade.od_stateid);
8729 }
8730
8731 void
8732 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
8733                 union nfsd4_op_u *u)
8734 {
8735         put_stateid(cstate, &u->open.op_stateid);
8736 }
8737
8738 void
8739 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
8740                 union nfsd4_op_u *u)
8741 {
8742         put_stateid(cstate, &u->close.cl_stateid);
8743 }
8744
8745 void
8746 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
8747                 union nfsd4_op_u *u)
8748 {
8749         put_stateid(cstate, &u->lock.lk_resp_stateid);
8750 }
8751
8752 /*
8753  * functions to consume current state id
8754  */
8755
8756 void
8757 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
8758                 union nfsd4_op_u *u)
8759 {
8760         get_stateid(cstate, &u->open_downgrade.od_stateid);
8761 }
8762
8763 void
8764 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
8765                 union nfsd4_op_u *u)
8766 {
8767         get_stateid(cstate, &u->delegreturn.dr_stateid);
8768 }
8769
8770 void
8771 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
8772                 union nfsd4_op_u *u)
8773 {
8774         get_stateid(cstate, &u->free_stateid.fr_stateid);
8775 }
8776
8777 void
8778 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
8779                 union nfsd4_op_u *u)
8780 {
8781         get_stateid(cstate, &u->setattr.sa_stateid);
8782 }
8783
8784 void
8785 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
8786                 union nfsd4_op_u *u)
8787 {
8788         get_stateid(cstate, &u->close.cl_stateid);
8789 }
8790
8791 void
8792 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
8793                 union nfsd4_op_u *u)
8794 {
8795         get_stateid(cstate, &u->locku.lu_stateid);
8796 }
8797
8798 void
8799 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
8800                 union nfsd4_op_u *u)
8801 {
8802         get_stateid(cstate, &u->read.rd_stateid);
8803 }
8804
8805 void
8806 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
8807                 union nfsd4_op_u *u)
8808 {
8809         get_stateid(cstate, &u->write.wr_stateid);
8810 }
8811
8812 /**
8813  * nfsd4_deleg_getattr_conflict - Recall if GETATTR causes conflict
8814  * @rqstp: RPC transaction context
8815  * @inode: file to be checked for a conflict
8816  * @modified: return true if file was modified
8817  * @size: new size of file if modified is true
8818  *
8819  * This function is called when there is a conflict between a write
8820  * delegation and a change/size GETATTR from another client. The server
8821  * must either use the CB_GETATTR to get the current values of the
8822  * attributes from the client that holds the delegation or recall the
8823  * delegation before replying to the GETATTR. See RFC 8881 section
8824  * 18.7.4.
8825  *
8826  * Returns 0 if there is no conflict; otherwise an nfs_stat
8827  * code is returned.
8828  */
8829 __be32
8830 nfsd4_deleg_getattr_conflict(struct svc_rqst *rqstp, struct inode *inode,
8831                                 bool *modified, u64 *size)
8832 {
8833         __be32 status;
8834         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8835         struct file_lock_context *ctx;
8836         struct file_lease *fl;
8837         struct nfs4_delegation *dp;
8838         struct iattr attrs;
8839         struct nfs4_cb_fattr *ncf;
8840
8841         *modified = false;
8842         ctx = locks_inode_context(inode);
8843         if (!ctx)
8844                 return 0;
8845         spin_lock(&ctx->flc_lock);
8846         for_each_file_lock(fl, &ctx->flc_lease) {
8847                 unsigned char type = fl->c.flc_type;
8848
8849                 if (fl->c.flc_flags == FL_LAYOUT)
8850                         continue;
8851                 if (fl->fl_lmops != &nfsd_lease_mng_ops) {
8852                         /*
8853                          * non-nfs lease, if it's a lease with F_RDLCK then
8854                          * we are done; there isn't any write delegation
8855                          * on this inode
8856                          */
8857                         if (type == F_RDLCK)
8858                                 break;
8859                         goto break_lease;
8860                 }
8861                 if (type == F_WRLCK) {
8862                         dp = fl->c.flc_owner;
8863                         if (dp->dl_recall.cb_clp == *(rqstp->rq_lease_breaker)) {
8864                                 spin_unlock(&ctx->flc_lock);
8865                                 return 0;
8866                         }
8867 break_lease:
8868                         nfsd_stats_wdeleg_getattr_inc(nn);
8869                         dp = fl->c.flc_owner;
8870                         ncf = &dp->dl_cb_fattr;
8871                         nfs4_cb_getattr(&dp->dl_cb_fattr);
8872                         spin_unlock(&ctx->flc_lock);
8873                         wait_on_bit_timeout(&ncf->ncf_cb_flags, CB_GETATTR_BUSY,
8874                                         TASK_INTERRUPTIBLE, NFSD_CB_GETATTR_TIMEOUT);
8875                         if (ncf->ncf_cb_status) {
8876                                 /* Recall delegation only if client didn't respond */
8877                                 status = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
8878                                 if (status != nfserr_jukebox ||
8879                                                 !nfsd_wait_for_delegreturn(rqstp, inode))
8880                                         return status;
8881                         }
8882                         if (!ncf->ncf_file_modified &&
8883                                         (ncf->ncf_initial_cinfo != ncf->ncf_cb_change ||
8884                                         ncf->ncf_cur_fsize != ncf->ncf_cb_fsize))
8885                                 ncf->ncf_file_modified = true;
8886                         if (ncf->ncf_file_modified) {
8887                                 /*
8888                                  * Per section 10.4.3 of RFC 8881, the server would
8889                                  * not update the file's metadata with the client's
8890                                  * modified size
8891                                  */
8892                                 attrs.ia_mtime = attrs.ia_ctime = current_time(inode);
8893                                 attrs.ia_valid = ATTR_MTIME | ATTR_CTIME;
8894                                 setattr_copy(&nop_mnt_idmap, inode, &attrs);
8895                                 mark_inode_dirty(inode);
8896                                 ncf->ncf_cur_fsize = ncf->ncf_cb_fsize;
8897                                 *size = ncf->ncf_cur_fsize;
8898                                 *modified = true;
8899                         }
8900                         return 0;
8901                 }
8902                 break;
8903         }
8904         spin_unlock(&ctx->flc_lock);
8905         return 0;
8906 }