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
  91 /* Locking: */
  92
  93 /*
  94  * Currently used for the del_recall_lru and file hash table.  In an
  95  * effort to decrease the scope of the client_mutex, this spinlock may
  96  * eventually cover more:
  97  */
  98 static DEFINE_SPINLOCK(state_lock);
  99
 100 enum nfsd4_st_mutex_lock_subclass {
 101         OPEN_STATEID_MUTEX = 0,
 102         LOCK_STATEID_MUTEX = 1,
 103 };
 104
 105 /*
 106  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
 107  * the refcount on the open stateid to drop.
 108  */
 109 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
 110
 111 /*
 112  * A waitqueue where a writer to clients/#/ctl destroying a client can
 113  * wait for cl_rpc_users to drop to 0 and then for the client to be
 114  * unhashed.
 115  */
 116 static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
 117
 118 static struct kmem_cache *client_slab;
 119 static struct kmem_cache *openowner_slab;
 120 static struct kmem_cache *lockowner_slab;
 121 static struct kmem_cache *file_slab;
 122 static struct kmem_cache *stateid_slab;
 123 static struct kmem_cache *deleg_slab;
 124 static struct kmem_cache *odstate_slab;
 125
 126 static void free_session(struct nfsd4_session *);
 127
 128 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 129 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
 130
 131 static struct workqueue_struct *laundry_wq;
 132
 133 int nfsd4_create_laundry_wq(void)
 134 {
 135         int rc = 0;
 136
 137         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
 138         if (laundry_wq == NULL)
 139                 rc = -ENOMEM;
 140         return rc;
 141 }
 142
 143 void nfsd4_destroy_laundry_wq(void)
 144 {
 145         destroy_workqueue(laundry_wq);
 146 }
 147
 148 static bool is_session_dead(struct nfsd4_session *ses)
 149 {
 150         return ses->se_flags & NFS4_SESSION_DEAD;
 151 }
 152
 153 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 154 {
 155         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 156                 return nfserr_jukebox;
 157         ses->se_flags |= NFS4_SESSION_DEAD;
 158         return nfs_ok;
 159 }
 160
 161 static bool is_client_expired(struct nfs4_client *clp)
 162 {
 163         return clp->cl_time == 0;
 164 }
 165
 166 static void nfsd4_dec_courtesy_client_count(struct nfsd_net *nn,
 167                                         struct nfs4_client *clp)
 168 {
 169         if (clp->cl_state != NFSD4_ACTIVE)
 170                 atomic_add_unless(&nn->nfsd_courtesy_clients, -1, 0);
 171 }
 172
 173 static __be32 get_client_locked(struct nfs4_client *clp)
 174 {
 175         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 176
 177         lockdep_assert_held(&nn->client_lock);
 178
 179         if (is_client_expired(clp))
 180                 return nfserr_expired;
 181         atomic_inc(&clp->cl_rpc_users);
 182         nfsd4_dec_courtesy_client_count(nn, clp);
 183         clp->cl_state = NFSD4_ACTIVE;
 184         return nfs_ok;
 185 }
 186
 187 /* must be called under the client_lock */
 188 static inline void
 189 renew_client_locked(struct nfs4_client *clp)
 190 {
 191         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 192
 193         if (is_client_expired(clp)) {
 194                 WARN_ON(1);
 195                 printk("%s: client (clientid %08x/%08x) already expired\n",
 196                         __func__,
 197                         clp->cl_clientid.cl_boot,
 198                         clp->cl_clientid.cl_id);
 199                 return;
 200         }
 201
 202         list_move_tail(&clp->cl_lru, &nn->client_lru);
 203         clp->cl_time = ktime_get_boottime_seconds();
 204         nfsd4_dec_courtesy_client_count(nn, clp);
 205         clp->cl_state = NFSD4_ACTIVE;
 206 }
 207
 208 static void put_client_renew_locked(struct nfs4_client *clp)
 209 {
 210         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 211
 212         lockdep_assert_held(&nn->client_lock);
 213
 214         if (!atomic_dec_and_test(&clp->cl_rpc_users))
 215                 return;
 216         if (!is_client_expired(clp))
 217                 renew_client_locked(clp);
 218         else
 219                 wake_up_all(&expiry_wq);
 220 }
 221
 222 static void put_client_renew(struct nfs4_client *clp)
 223 {
 224         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 225
 226         if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
 227                 return;
 228         if (!is_client_expired(clp))
 229                 renew_client_locked(clp);
 230         else
 231                 wake_up_all(&expiry_wq);
 232         spin_unlock(&nn->client_lock);
 233 }
 234
 235 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 236 {
 237         __be32 status;
 238
 239         if (is_session_dead(ses))
 240                 return nfserr_badsession;
 241         status = get_client_locked(ses->se_client);
 242         if (status)
 243                 return status;
 244         atomic_inc(&ses->se_ref);
 245         return nfs_ok;
 246 }
 247
 248 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 249 {
 250         struct nfs4_client *clp = ses->se_client;
 251         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 252
 253         lockdep_assert_held(&nn->client_lock);
 254
 255         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 256                 free_session(ses);
 257         put_client_renew_locked(clp);
 258 }
 259
 260 static void nfsd4_put_session(struct nfsd4_session *ses)
 261 {
 262         struct nfs4_client *clp = ses->se_client;
 263         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 264
 265         spin_lock(&nn->client_lock);
 266         nfsd4_put_session_locked(ses);
 267         spin_unlock(&nn->client_lock);
 268 }
 269
 270 static struct nfsd4_blocked_lock *
 271 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 272                         struct nfsd_net *nn)
 273 {
 274         struct nfsd4_blocked_lock *cur, *found = NULL;
 275
 276         spin_lock(&nn->blocked_locks_lock);
 277         list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
 278                 if (fh_match(fh, &cur->nbl_fh)) {
 279                         list_del_init(&cur->nbl_list);
 280                         WARN_ON(list_empty(&cur->nbl_lru));
 281                         list_del_init(&cur->nbl_lru);
 282                         found = cur;
 283                         break;
 284                 }
 285         }
 286         spin_unlock(&nn->blocked_locks_lock);
 287         if (found)
 288                 locks_delete_block(&found->nbl_lock);
 289         return found;
 290 }
 291
 292 static struct nfsd4_blocked_lock *
 293 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 294                         struct nfsd_net *nn)
 295 {
 296         struct nfsd4_blocked_lock *nbl;
 297
 298         nbl = find_blocked_lock(lo, fh, nn);
 299         if (!nbl) {
 300                 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
 301                 if (nbl) {
 302                         INIT_LIST_HEAD(&nbl->nbl_list);
 303                         INIT_LIST_HEAD(&nbl->nbl_lru);
 304                         fh_copy_shallow(&nbl->nbl_fh, fh);
 305                         locks_init_lock(&nbl->nbl_lock);
 306                         kref_init(&nbl->nbl_kref);
 307                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
 308                                         &nfsd4_cb_notify_lock_ops,
 309                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
 310                 }
 311         }
 312         return nbl;
 313 }
 314
 315 static void
 316 free_nbl(struct kref *kref)
 317 {
 318         struct nfsd4_blocked_lock *nbl;
 319
 320         nbl = container_of(kref, struct nfsd4_blocked_lock, nbl_kref);
 321         kfree(nbl);
 322 }
 323
 324 static void
 325 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
 326 {
 327         locks_delete_block(&nbl->nbl_lock);
 328         locks_release_private(&nbl->nbl_lock);
 329         kref_put(&nbl->nbl_kref, free_nbl);
 330 }
 331
 332 static void
 333 remove_blocked_locks(struct nfs4_lockowner *lo)
 334 {
 335         struct nfs4_client *clp = lo->lo_owner.so_client;
 336         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 337         struct nfsd4_blocked_lock *nbl;
 338         LIST_HEAD(reaplist);
 339
 340         /* Dequeue all blocked locks */
 341         spin_lock(&nn->blocked_locks_lock);
 342         while (!list_empty(&lo->lo_blocked)) {
 343                 nbl = list_first_entry(&lo->lo_blocked,
 344                                         struct nfsd4_blocked_lock,
 345                                         nbl_list);
 346                 list_del_init(&nbl->nbl_list);
 347                 WARN_ON(list_empty(&nbl->nbl_lru));
 348                 list_move(&nbl->nbl_lru, &reaplist);
 349         }
 350         spin_unlock(&nn->blocked_locks_lock);
 351
 352         /* Now free them */
 353         while (!list_empty(&reaplist)) {
 354                 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
 355                                         nbl_lru);
 356                 list_del_init(&nbl->nbl_lru);
 357                 free_blocked_lock(nbl);
 358         }
 359 }
 360
 361 static void
 362 nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
 363 {
 364         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 365                                                 struct nfsd4_blocked_lock, nbl_cb);
 366         locks_delete_block(&nbl->nbl_lock);
 367 }
 368
 369 static int
 370 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
 371 {
 372         trace_nfsd_cb_notify_lock_done(&zero_stateid, task);
 373
 374         /*
 375          * Since this is just an optimization, we don't try very hard if it
 376          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
 377          * just quit trying on anything else.
 378          */
 379         switch (task->tk_status) {
 380         case -NFS4ERR_DELAY:
 381                 rpc_delay(task, 1 * HZ);
 382                 return 0;
 383         default:
 384                 return 1;
 385         }
 386 }
 387
 388 static void
 389 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
 390 {
 391         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 392                                                 struct nfsd4_blocked_lock, nbl_cb);
 393
 394         free_blocked_lock(nbl);
 395 }
 396
 397 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
 398         .prepare        = nfsd4_cb_notify_lock_prepare,
 399         .done           = nfsd4_cb_notify_lock_done,
 400         .release        = nfsd4_cb_notify_lock_release,
 401 };
 402
 403 /*
 404  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 405  * st_{access,deny}_bmap field of the stateid, in order to track not
 406  * only what share bits are currently in force, but also what
 407  * combinations of share bits previous opens have used.  This allows us
 408  * to enforce the recommendation in
 409  * https://datatracker.ietf.org/doc/html/rfc7530#section-16.19.4 that
 410  * the server return an error if the client attempt to downgrade to a
 411  * combination of share bits not explicable by closing some of its
 412  * previous opens.
 413  *
 414  * This enforcement is arguably incomplete, since we don't keep
 415  * track of access/deny bit combinations; so, e.g., we allow:
 416  *
 417  *      OPEN allow read, deny write
 418  *      OPEN allow both, deny none
 419  *      DOWNGRADE allow read, deny none
 420  *
 421  * which we should reject.
 422  *
 423  * But you could also argue that our current code is already overkill,
 424  * since it only exists to return NFS4ERR_INVAL on incorrect client
 425  * behavior.
 426  */
 427 static unsigned int
 428 bmap_to_share_mode(unsigned long bmap)
 429 {
 430         int i;
 431         unsigned int access = 0;
 432
 433         for (i = 1; i < 4; i++) {
 434                 if (test_bit(i, &bmap))
 435                         access |= i;
 436         }
 437         return access;
 438 }
 439
 440 /* set share access for a given stateid */
 441 static inline void
 442 set_access(u32 access, struct nfs4_ol_stateid *stp)
 443 {
 444         unsigned char mask = 1 << access;
 445
 446         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 447         stp->st_access_bmap |= mask;
 448 }
 449
 450 /* clear share access for a given stateid */
 451 static inline void
 452 clear_access(u32 access, struct nfs4_ol_stateid *stp)
 453 {
 454         unsigned char mask = 1 << access;
 455
 456         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 457         stp->st_access_bmap &= ~mask;
 458 }
 459
 460 /* test whether a given stateid has access */
 461 static inline bool
 462 test_access(u32 access, struct nfs4_ol_stateid *stp)
 463 {
 464         unsigned char mask = 1 << access;
 465
 466         return (bool)(stp->st_access_bmap & mask);
 467 }
 468
 469 /* set share deny for a given stateid */
 470 static inline void
 471 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
 472 {
 473         unsigned char mask = 1 << deny;
 474
 475         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 476         stp->st_deny_bmap |= mask;
 477 }
 478
 479 /* clear share deny for a given stateid */
 480 static inline void
 481 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
 482 {
 483         unsigned char mask = 1 << deny;
 484
 485         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 486         stp->st_deny_bmap &= ~mask;
 487 }
 488
 489 /* test whether a given stateid is denying specific access */
 490 static inline bool
 491 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
 492 {
 493         unsigned char mask = 1 << deny;
 494
 495         return (bool)(stp->st_deny_bmap & mask);
 496 }
 497
 498 static int nfs4_access_to_omode(u32 access)
 499 {
 500         switch (access & NFS4_SHARE_ACCESS_BOTH) {
 501         case NFS4_SHARE_ACCESS_READ:
 502                 return O_RDONLY;
 503         case NFS4_SHARE_ACCESS_WRITE:
 504                 return O_WRONLY;
 505         case NFS4_SHARE_ACCESS_BOTH:
 506                 return O_RDWR;
 507         }
 508         WARN_ON_ONCE(1);
 509         return O_RDONLY;
 510 }
 511
 512 static inline int
 513 access_permit_read(struct nfs4_ol_stateid *stp)
 514 {
 515         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
 516                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
 517                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
 518 }
 519
 520 static inline int
 521 access_permit_write(struct nfs4_ol_stateid *stp)
 522 {
 523         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
 524                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
 525 }
 526
 527 static inline struct nfs4_stateowner *
 528 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 529 {
 530         atomic_inc(&sop->so_count);
 531         return sop;
 532 }
 533
 534 static int
 535 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 536 {
 537         return (sop->so_owner.len == owner->len) &&
 538                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 539 }
 540
 541 static struct nfs4_openowner *
 542 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 543                         struct nfs4_client *clp)
 544 {
 545         struct nfs4_stateowner *so;
 546
 547         lockdep_assert_held(&clp->cl_lock);
 548
 549         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 550                             so_strhash) {
 551                 if (!so->so_is_open_owner)
 552                         continue;
 553                 if (same_owner_str(so, &open->op_owner))
 554                         return openowner(nfs4_get_stateowner(so));
 555         }
 556         return NULL;
 557 }
 558
 559 static struct nfs4_openowner *
 560 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 561                         struct nfs4_client *clp)
 562 {
 563         struct nfs4_openowner *oo;
 564
 565         spin_lock(&clp->cl_lock);
 566         oo = find_openstateowner_str_locked(hashval, open, clp);
 567         spin_unlock(&clp->cl_lock);
 568         return oo;
 569 }
 570
 571 static inline u32
 572 opaque_hashval(const void *ptr, int nbytes)
 573 {
 574         unsigned char *cptr = (unsigned char *) ptr;
 575
 576         u32 x = 0;
 577         while (nbytes--) {
 578                 x *= 37;
 579                 x += *cptr++;
 580         }
 581         return x;
 582 }
 583
 584 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 585 {
 586         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 587
 588         kmem_cache_free(file_slab, fp);
 589 }
 590
 591 void
 592 put_nfs4_file(struct nfs4_file *fi)
 593 {
 594         if (refcount_dec_and_test(&fi->fi_ref)) {
 595                 nfsd4_file_hash_remove(fi);
 596                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 597                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 598                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 599         }
 600 }
 601
 602 static struct nfsd_file *
 603 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 604 {
 605         if (f->fi_fds[oflag])
 606                 return nfsd_file_get(f->fi_fds[oflag]);
 607         return NULL;
 608 }
 609
 610 static struct nfsd_file *
 611 find_writeable_file_locked(struct nfs4_file *f)
 612 {
 613         struct nfsd_file *ret;
 614
 615         lockdep_assert_held(&f->fi_lock);
 616
 617         ret = __nfs4_get_fd(f, O_WRONLY);
 618         if (!ret)
 619                 ret = __nfs4_get_fd(f, O_RDWR);
 620         return ret;
 621 }
 622
 623 static struct nfsd_file *
 624 find_writeable_file(struct nfs4_file *f)
 625 {
 626         struct nfsd_file *ret;
 627
 628         spin_lock(&f->fi_lock);
 629         ret = find_writeable_file_locked(f);
 630         spin_unlock(&f->fi_lock);
 631
 632         return ret;
 633 }
 634
 635 static struct nfsd_file *
 636 find_readable_file_locked(struct nfs4_file *f)
 637 {
 638         struct nfsd_file *ret;
 639
 640         lockdep_assert_held(&f->fi_lock);
 641
 642         ret = __nfs4_get_fd(f, O_RDONLY);
 643         if (!ret)
 644                 ret = __nfs4_get_fd(f, O_RDWR);
 645         return ret;
 646 }
 647
 648 static struct nfsd_file *
 649 find_readable_file(struct nfs4_file *f)
 650 {
 651         struct nfsd_file *ret;
 652
 653         spin_lock(&f->fi_lock);
 654         ret = find_readable_file_locked(f);
 655         spin_unlock(&f->fi_lock);
 656
 657         return ret;
 658 }
 659
 660 struct nfsd_file *
 661 find_any_file(struct nfs4_file *f)
 662 {
 663         struct nfsd_file *ret;
 664
 665         if (!f)
 666                 return NULL;
 667         spin_lock(&f->fi_lock);
 668         ret = __nfs4_get_fd(f, O_RDWR);
 669         if (!ret) {
 670                 ret = __nfs4_get_fd(f, O_WRONLY);
 671                 if (!ret)
 672                         ret = __nfs4_get_fd(f, O_RDONLY);
 673         }
 674         spin_unlock(&f->fi_lock);
 675         return ret;
 676 }
 677
 678 static struct nfsd_file *find_any_file_locked(struct nfs4_file *f)
 679 {
 680         lockdep_assert_held(&f->fi_lock);
 681
 682         if (f->fi_fds[O_RDWR])
 683                 return f->fi_fds[O_RDWR];
 684         if (f->fi_fds[O_WRONLY])
 685                 return f->fi_fds[O_WRONLY];
 686         if (f->fi_fds[O_RDONLY])
 687                 return f->fi_fds[O_RDONLY];
 688         return NULL;
 689 }
 690
 691 static struct nfsd_file *find_deleg_file_locked(struct nfs4_file *f)
 692 {
 693         lockdep_assert_held(&f->fi_lock);
 694
 695         if (f->fi_deleg_file)
 696                 return f->fi_deleg_file;
 697         return NULL;
 698 }
 699
 700 static atomic_long_t num_delegations;
 701 unsigned long max_delegations;
 702
 703 /*
 704  * Open owner state (share locks)
 705  */
 706
 707 /* hash tables for lock and open owners */
 708 #define OWNER_HASH_BITS              8
 709 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 710 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 711
 712 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 713 {
 714         unsigned int ret;
 715
 716         ret = opaque_hashval(ownername->data, ownername->len);
 717         return ret & OWNER_HASH_MASK;
 718 }
 719
 720 static struct rhltable nfs4_file_rhltable ____cacheline_aligned_in_smp;
 721
 722 static const struct rhashtable_params nfs4_file_rhash_params = {
 723         .key_len                = sizeof_field(struct nfs4_file, fi_inode),
 724         .key_offset             = offsetof(struct nfs4_file, fi_inode),
 725         .head_offset            = offsetof(struct nfs4_file, fi_rlist),
 726
 727         /*
 728          * Start with a single page hash table to reduce resizing churn
 729          * on light workloads.
 730          */
 731         .min_size               = 256,
 732         .automatic_shrinking    = true,
 733 };
 734
 735 /*
 736  * Check if courtesy clients have conflicting access and resolve it if possible
 737  *
 738  * access:  is op_share_access if share_access is true.
 739  *          Check if access mode, op_share_access, would conflict with
 740  *          the current deny mode of the file 'fp'.
 741  * access:  is op_share_deny if share_access is false.
 742  *          Check if the deny mode, op_share_deny, would conflict with
 743  *          current access of the file 'fp'.
 744  * stp:     skip checking this entry.
 745  * new_stp: normal open, not open upgrade.
 746  *
 747  * Function returns:
 748  *      false - access/deny mode conflict with normal client.
 749  *      true  - no conflict or conflict with courtesy client(s) is resolved.
 750  */
 751 static bool
 752 nfs4_resolve_deny_conflicts_locked(struct nfs4_file *fp, bool new_stp,
 753                 struct nfs4_ol_stateid *stp, u32 access, bool share_access)
 754 {
 755         struct nfs4_ol_stateid *st;
 756         bool resolvable = true;
 757         unsigned char bmap;
 758         struct nfsd_net *nn;
 759         struct nfs4_client *clp;
 760
 761         lockdep_assert_held(&fp->fi_lock);
 762         list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
 763                 /* ignore lock stateid */
 764                 if (st->st_openstp)
 765                         continue;
 766                 if (st == stp && new_stp)
 767                         continue;
 768                 /* check file access against deny mode or vice versa */
 769                 bmap = share_access ? st->st_deny_bmap : st->st_access_bmap;
 770                 if (!(access & bmap_to_share_mode(bmap)))
 771                         continue;
 772                 clp = st->st_stid.sc_client;
 773                 if (try_to_expire_client(clp))
 774                         continue;
 775                 resolvable = false;
 776                 break;
 777         }
 778         if (resolvable) {
 779                 clp = stp->st_stid.sc_client;
 780                 nn = net_generic(clp->net, nfsd_net_id);
 781                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
 782         }
 783         return resolvable;
 784 }
 785
 786 static void
 787 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 788 {
 789         lockdep_assert_held(&fp->fi_lock);
 790
 791         if (access & NFS4_SHARE_ACCESS_WRITE)
 792                 atomic_inc(&fp->fi_access[O_WRONLY]);
 793         if (access & NFS4_SHARE_ACCESS_READ)
 794                 atomic_inc(&fp->fi_access[O_RDONLY]);
 795 }
 796
 797 static __be32
 798 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 799 {
 800         lockdep_assert_held(&fp->fi_lock);
 801
 802         /* Does this access mode make sense? */
 803         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 804                 return nfserr_inval;
 805
 806         /* Does it conflict with a deny mode already set? */
 807         if ((access & fp->fi_share_deny) != 0)
 808                 return nfserr_share_denied;
 809
 810         __nfs4_file_get_access(fp, access);
 811         return nfs_ok;
 812 }
 813
 814 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 815 {
 816         /* Common case is that there is no deny mode. */
 817         if (deny) {
 818                 /* Does this deny mode make sense? */
 819                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 820                         return nfserr_inval;
 821
 822                 if ((deny & NFS4_SHARE_DENY_READ) &&
 823                     atomic_read(&fp->fi_access[O_RDONLY]))
 824                         return nfserr_share_denied;
 825
 826                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 827                     atomic_read(&fp->fi_access[O_WRONLY]))
 828                         return nfserr_share_denied;
 829         }
 830         return nfs_ok;
 831 }
 832
 833 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 834 {
 835         might_lock(&fp->fi_lock);
 836
 837         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 838                 struct nfsd_file *f1 = NULL;
 839                 struct nfsd_file *f2 = NULL;
 840
 841                 swap(f1, fp->fi_fds[oflag]);
 842                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 843                         swap(f2, fp->fi_fds[O_RDWR]);
 844                 spin_unlock(&fp->fi_lock);
 845                 if (f1)
 846                         nfsd_file_put(f1);
 847                 if (f2)
 848                         nfsd_file_put(f2);
 849         }
 850 }
 851
 852 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 853 {
 854         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 855
 856         if (access & NFS4_SHARE_ACCESS_WRITE)
 857                 __nfs4_file_put_access(fp, O_WRONLY);
 858         if (access & NFS4_SHARE_ACCESS_READ)
 859                 __nfs4_file_put_access(fp, O_RDONLY);
 860 }
 861
 862 /*
 863  * Allocate a new open/delegation state counter. This is needed for
 864  * pNFS for proper return on close semantics.
 865  *
 866  * Note that we only allocate it for pNFS-enabled exports, otherwise
 867  * all pointers to struct nfs4_clnt_odstate are always NULL.
 868  */
 869 static struct nfs4_clnt_odstate *
 870 alloc_clnt_odstate(struct nfs4_client *clp)
 871 {
 872         struct nfs4_clnt_odstate *co;
 873
 874         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 875         if (co) {
 876                 co->co_client = clp;
 877                 refcount_set(&co->co_odcount, 1);
 878         }
 879         return co;
 880 }
 881
 882 static void
 883 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 884 {
 885         struct nfs4_file *fp = co->co_file;
 886
 887         lockdep_assert_held(&fp->fi_lock);
 888         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 889 }
 890
 891 static inline void
 892 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 893 {
 894         if (co)
 895                 refcount_inc(&co->co_odcount);
 896 }
 897
 898 static void
 899 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 900 {
 901         struct nfs4_file *fp;
 902
 903         if (!co)
 904                 return;
 905
 906         fp = co->co_file;
 907         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 908                 list_del(&co->co_perfile);
 909                 spin_unlock(&fp->fi_lock);
 910
 911                 nfsd4_return_all_file_layouts(co->co_client, fp);
 912                 kmem_cache_free(odstate_slab, co);
 913         }
 914 }
 915
 916 static struct nfs4_clnt_odstate *
 917 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 918 {
 919         struct nfs4_clnt_odstate *co;
 920         struct nfs4_client *cl;
 921
 922         if (!new)
 923                 return NULL;
 924
 925         cl = new->co_client;
 926
 927         spin_lock(&fp->fi_lock);
 928         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 929                 if (co->co_client == cl) {
 930                         get_clnt_odstate(co);
 931                         goto out;
 932                 }
 933         }
 934         co = new;
 935         co->co_file = fp;
 936         hash_clnt_odstate_locked(new);
 937 out:
 938         spin_unlock(&fp->fi_lock);
 939         return co;
 940 }
 941
 942 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 943                                   void (*sc_free)(struct nfs4_stid *))
 944 {
 945         struct nfs4_stid *stid;
 946         int new_id;
 947
 948         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 949         if (!stid)
 950                 return NULL;
 951
 952         idr_preload(GFP_KERNEL);
 953         spin_lock(&cl->cl_lock);
 954         /* Reserving 0 for start of file in nfsdfs "states" file: */
 955         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
 956         spin_unlock(&cl->cl_lock);
 957         idr_preload_end();
 958         if (new_id < 0)
 959                 goto out_free;
 960
 961         stid->sc_free = sc_free;
 962         stid->sc_client = cl;
 963         stid->sc_stateid.si_opaque.so_id = new_id;
 964         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 965         /* Will be incremented before return to client: */
 966         refcount_set(&stid->sc_count, 1);
 967         spin_lock_init(&stid->sc_lock);
 968         INIT_LIST_HEAD(&stid->sc_cp_list);
 969
 970         /*
 971          * It shouldn't be a problem to reuse an opaque stateid value.
 972          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 973          * example, a stray write retransmission could be accepted by
 974          * the server when it should have been rejected.  Therefore,
 975          * adopt a trick from the sctp code to attempt to maximize the
 976          * amount of time until an id is reused, by ensuring they always
 977          * "increase" (mod INT_MAX):
 978          */
 979         return stid;
 980 out_free:
 981         kmem_cache_free(slab, stid);
 982         return NULL;
 983 }
 984
 985 /*
 986  * Create a unique stateid_t to represent each COPY.
 987  */
 988 static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
 989                               unsigned char cs_type)
 990 {
 991         int new_id;
 992
 993         stid->cs_stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
 994         stid->cs_stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
 995         stid->cs_type = cs_type;
 996
 997         idr_preload(GFP_KERNEL);
 998         spin_lock(&nn->s2s_cp_lock);
 999         new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
1000         stid->cs_stid.si_opaque.so_id = new_id;
1001         stid->cs_stid.si_generation = 1;
1002         spin_unlock(&nn->s2s_cp_lock);
1003         idr_preload_end();
1004         if (new_id < 0)
1005                 return 0;
1006         return 1;
1007 }
1008
1009 int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
1010 {
1011         return nfs4_init_cp_state(nn, &copy->cp_stateid, NFS4_COPY_STID);
1012 }
1013
1014 struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
1015                                                      struct nfs4_stid *p_stid)
1016 {
1017         struct nfs4_cpntf_state *cps;
1018
1019         cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
1020         if (!cps)
1021                 return NULL;
1022         cps->cpntf_time = ktime_get_boottime_seconds();
1023         refcount_set(&cps->cp_stateid.cs_count, 1);
1024         if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
1025                 goto out_free;
1026         spin_lock(&nn->s2s_cp_lock);
1027         list_add(&cps->cp_list, &p_stid->sc_cp_list);
1028         spin_unlock(&nn->s2s_cp_lock);
1029         return cps;
1030 out_free:
1031         kfree(cps);
1032         return NULL;
1033 }
1034
1035 void nfs4_free_copy_state(struct nfsd4_copy *copy)
1036 {
1037         struct nfsd_net *nn;
1038
1039         WARN_ON_ONCE(copy->cp_stateid.cs_type != NFS4_COPY_STID);
1040         nn = net_generic(copy->cp_clp->net, nfsd_net_id);
1041         spin_lock(&nn->s2s_cp_lock);
1042         idr_remove(&nn->s2s_cp_stateids,
1043                    copy->cp_stateid.cs_stid.si_opaque.so_id);
1044         spin_unlock(&nn->s2s_cp_lock);
1045 }
1046
1047 static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
1048 {
1049         struct nfs4_cpntf_state *cps;
1050         struct nfsd_net *nn;
1051
1052         nn = net_generic(net, nfsd_net_id);
1053         spin_lock(&nn->s2s_cp_lock);
1054         while (!list_empty(&stid->sc_cp_list)) {
1055                 cps = list_first_entry(&stid->sc_cp_list,
1056                                        struct nfs4_cpntf_state, cp_list);
1057                 _free_cpntf_state_locked(nn, cps);
1058         }
1059         spin_unlock(&nn->s2s_cp_lock);
1060 }
1061
1062 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
1063 {
1064         struct nfs4_stid *stid;
1065
1066         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
1067         if (!stid)
1068                 return NULL;
1069
1070         return openlockstateid(stid);
1071 }
1072
1073 static void nfs4_free_deleg(struct nfs4_stid *stid)
1074 {
1075         struct nfs4_delegation *dp = delegstateid(stid);
1076
1077         WARN_ON_ONCE(!list_empty(&stid->sc_cp_list));
1078         WARN_ON_ONCE(!list_empty(&dp->dl_perfile));
1079         WARN_ON_ONCE(!list_empty(&dp->dl_perclnt));
1080         WARN_ON_ONCE(!list_empty(&dp->dl_recall_lru));
1081         kmem_cache_free(deleg_slab, stid);
1082         atomic_long_dec(&num_delegations);
1083 }
1084
1085 /*
1086  * When we recall a delegation, we should be careful not to hand it
1087  * out again straight away.
1088  * To ensure this we keep a pair of bloom filters ('new' and 'old')
1089  * in which the filehandles of recalled delegations are "stored".
1090  * If a filehandle appear in either filter, a delegation is blocked.
1091  * When a delegation is recalled, the filehandle is stored in the "new"
1092  * filter.
1093  * Every 30 seconds we swap the filters and clear the "new" one,
1094  * unless both are empty of course.
1095  *
1096  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
1097  * low 3 bytes as hash-table indices.
1098  *
1099  * 'blocked_delegations_lock', which is always taken in block_delegations(),
1100  * is used to manage concurrent access.  Testing does not need the lock
1101  * except when swapping the two filters.
1102  */
1103 static DEFINE_SPINLOCK(blocked_delegations_lock);
1104 static struct bloom_pair {
1105         int     entries, old_entries;
1106         time64_t swap_time;
1107         int     new; /* index into 'set' */
1108         DECLARE_BITMAP(set[2], 256);
1109 } blocked_delegations;
1110
1111 static int delegation_blocked(struct knfsd_fh *fh)
1112 {
1113         u32 hash;
1114         struct bloom_pair *bd = &blocked_delegations;
1115
1116         if (bd->entries == 0)
1117                 return 0;
1118         if (ktime_get_seconds() - bd->swap_time > 30) {
1119                 spin_lock(&blocked_delegations_lock);
1120                 if (ktime_get_seconds() - bd->swap_time > 30) {
1121                         bd->entries -= bd->old_entries;
1122                         bd->old_entries = bd->entries;
1123                         memset(bd->set[bd->new], 0,
1124                                sizeof(bd->set[0]));
1125                         bd->new = 1-bd->new;
1126                         bd->swap_time = ktime_get_seconds();
1127                 }
1128                 spin_unlock(&blocked_delegations_lock);
1129         }
1130         hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1131         if (test_bit(hash&255, bd->set[0]) &&
1132             test_bit((hash>>8)&255, bd->set[0]) &&
1133             test_bit((hash>>16)&255, bd->set[0]))
1134                 return 1;
1135
1136         if (test_bit(hash&255, bd->set[1]) &&
1137             test_bit((hash>>8)&255, bd->set[1]) &&
1138             test_bit((hash>>16)&255, bd->set[1]))
1139                 return 1;
1140
1141         return 0;
1142 }
1143
1144 static void block_delegations(struct knfsd_fh *fh)
1145 {
1146         u32 hash;
1147         struct bloom_pair *bd = &blocked_delegations;
1148
1149         hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1150
1151         spin_lock(&blocked_delegations_lock);
1152         __set_bit(hash&255, bd->set[bd->new]);
1153         __set_bit((hash>>8)&255, bd->set[bd->new]);
1154         __set_bit((hash>>16)&255, bd->set[bd->new]);
1155         if (bd->entries == 0)
1156                 bd->swap_time = ktime_get_seconds();
1157         bd->entries += 1;
1158         spin_unlock(&blocked_delegations_lock);
1159 }
1160
1161 static struct nfs4_delegation *
1162 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
1163                  struct nfs4_clnt_odstate *odstate)
1164 {
1165         struct nfs4_delegation *dp;
1166         long n;
1167
1168         dprintk("NFSD alloc_init_deleg\n");
1169         n = atomic_long_inc_return(&num_delegations);
1170         if (n < 0 || n > max_delegations)
1171                 goto out_dec;
1172         if (delegation_blocked(&fp->fi_fhandle))
1173                 goto out_dec;
1174         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
1175         if (dp == NULL)
1176                 goto out_dec;
1177
1178         /*
1179          * delegation seqid's are never incremented.  The 4.1 special
1180          * meaning of seqid 0 isn't meaningful, really, but let's avoid
1181          * 0 anyway just for consistency and use 1:
1182          */
1183         dp->dl_stid.sc_stateid.si_generation = 1;
1184         INIT_LIST_HEAD(&dp->dl_perfile);
1185         INIT_LIST_HEAD(&dp->dl_perclnt);
1186         INIT_LIST_HEAD(&dp->dl_recall_lru);
1187         dp->dl_clnt_odstate = odstate;
1188         get_clnt_odstate(odstate);
1189         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
1190         dp->dl_retries = 1;
1191         dp->dl_recalled = false;
1192         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
1193                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
1194         get_nfs4_file(fp);
1195         dp->dl_stid.sc_file = fp;
1196         return dp;
1197 out_dec:
1198         atomic_long_dec(&num_delegations);
1199         return NULL;
1200 }
1201
1202 void
1203 nfs4_put_stid(struct nfs4_stid *s)
1204 {
1205         struct nfs4_file *fp = s->sc_file;
1206         struct nfs4_client *clp = s->sc_client;
1207
1208         might_lock(&clp->cl_lock);
1209
1210         if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
1211                 wake_up_all(&close_wq);
1212                 return;
1213         }
1214         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1215         nfs4_free_cpntf_statelist(clp->net, s);
1216         spin_unlock(&clp->cl_lock);
1217         s->sc_free(s);
1218         if (fp)
1219                 put_nfs4_file(fp);
1220 }
1221
1222 void
1223 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
1224 {
1225         stateid_t *src = &stid->sc_stateid;
1226
1227         spin_lock(&stid->sc_lock);
1228         if (unlikely(++src->si_generation == 0))
1229                 src->si_generation = 1;
1230         memcpy(dst, src, sizeof(*dst));
1231         spin_unlock(&stid->sc_lock);
1232 }
1233
1234 static void put_deleg_file(struct nfs4_file *fp)
1235 {
1236         struct nfsd_file *nf = NULL;
1237
1238         spin_lock(&fp->fi_lock);
1239         if (--fp->fi_delegees == 0)
1240                 swap(nf, fp->fi_deleg_file);
1241         spin_unlock(&fp->fi_lock);
1242
1243         if (nf)
1244                 nfsd_file_put(nf);
1245 }
1246
1247 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
1248 {
1249         struct nfs4_file *fp = dp->dl_stid.sc_file;
1250         struct nfsd_file *nf = fp->fi_deleg_file;
1251
1252         WARN_ON_ONCE(!fp->fi_delegees);
1253
1254         vfs_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
1255         put_deleg_file(fp);
1256 }
1257
1258 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
1259 {
1260         put_clnt_odstate(dp->dl_clnt_odstate);
1261         nfs4_unlock_deleg_lease(dp);
1262         nfs4_put_stid(&dp->dl_stid);
1263 }
1264
1265 void nfs4_unhash_stid(struct nfs4_stid *s)
1266 {
1267         s->sc_type = 0;
1268 }
1269
1270 /**
1271  * nfs4_delegation_exists - Discover if this delegation already exists
1272  * @clp:     a pointer to the nfs4_client we're granting a delegation to
1273  * @fp:      a pointer to the nfs4_file we're granting a delegation on
1274  *
1275  * Return:
1276  *      On success: true iff an existing delegation is found
1277  */
1278
1279 static bool
1280 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
1281 {
1282         struct nfs4_delegation *searchdp = NULL;
1283         struct nfs4_client *searchclp = NULL;
1284
1285         lockdep_assert_held(&state_lock);
1286         lockdep_assert_held(&fp->fi_lock);
1287
1288         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
1289                 searchclp = searchdp->dl_stid.sc_client;
1290                 if (clp == searchclp) {
1291                         return true;
1292                 }
1293         }
1294         return false;
1295 }
1296
1297 /**
1298  * hash_delegation_locked - Add a delegation to the appropriate lists
1299  * @dp:     a pointer to the nfs4_delegation we are adding.
1300  * @fp:     a pointer to the nfs4_file we're granting a delegation on
1301  *
1302  * Return:
1303  *      On success: NULL if the delegation was successfully hashed.
1304  *
1305  *      On error: -EAGAIN if one was previously granted to this
1306  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
1307  *
1308  */
1309
1310 static int
1311 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1312 {
1313         struct nfs4_client *clp = dp->dl_stid.sc_client;
1314
1315         lockdep_assert_held(&state_lock);
1316         lockdep_assert_held(&fp->fi_lock);
1317
1318         if (nfs4_delegation_exists(clp, fp))
1319                 return -EAGAIN;
1320         refcount_inc(&dp->dl_stid.sc_count);
1321         dp->dl_stid.sc_type = NFS4_DELEG_STID;
1322         list_add(&dp->dl_perfile, &fp->fi_delegations);
1323         list_add(&dp->dl_perclnt, &clp->cl_delegations);
1324         return 0;
1325 }
1326
1327 static bool delegation_hashed(struct nfs4_delegation *dp)
1328 {
1329         return !(list_empty(&dp->dl_perfile));
1330 }
1331
1332 static bool
1333 unhash_delegation_locked(struct nfs4_delegation *dp)
1334 {
1335         struct nfs4_file *fp = dp->dl_stid.sc_file;
1336
1337         lockdep_assert_held(&state_lock);
1338
1339         if (!delegation_hashed(dp))
1340                 return false;
1341
1342         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
1343         /* Ensure that deleg break won't try to requeue it */
1344         ++dp->dl_time;
1345         spin_lock(&fp->fi_lock);
1346         list_del_init(&dp->dl_perclnt);
1347         list_del_init(&dp->dl_recall_lru);
1348         list_del_init(&dp->dl_perfile);
1349         spin_unlock(&fp->fi_lock);
1350         return true;
1351 }
1352
1353 static void destroy_delegation(struct nfs4_delegation *dp)
1354 {
1355         bool unhashed;
1356
1357         spin_lock(&state_lock);
1358         unhashed = unhash_delegation_locked(dp);
1359         spin_unlock(&state_lock);
1360         if (unhashed)
1361                 destroy_unhashed_deleg(dp);
1362 }
1363
1364 static void revoke_delegation(struct nfs4_delegation *dp)
1365 {
1366         struct nfs4_client *clp = dp->dl_stid.sc_client;
1367
1368         WARN_ON(!list_empty(&dp->dl_recall_lru));
1369
1370         trace_nfsd_stid_revoke(&dp->dl_stid);
1371
1372         if (clp->cl_minorversion) {
1373                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1374                 refcount_inc(&dp->dl_stid.sc_count);
1375                 spin_lock(&clp->cl_lock);
1376                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1377                 spin_unlock(&clp->cl_lock);
1378         }
1379         destroy_unhashed_deleg(dp);
1380 }
1381
1382 /*
1383  * SETCLIENTID state
1384  */
1385
1386 static unsigned int clientid_hashval(u32 id)
1387 {
1388         return id & CLIENT_HASH_MASK;
1389 }
1390
1391 static unsigned int clientstr_hashval(struct xdr_netobj name)
1392 {
1393         return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1394 }
1395
1396 /*
1397  * A stateid that had a deny mode associated with it is being released
1398  * or downgraded. Recalculate the deny mode on the file.
1399  */
1400 static void
1401 recalculate_deny_mode(struct nfs4_file *fp)
1402 {
1403         struct nfs4_ol_stateid *stp;
1404
1405         spin_lock(&fp->fi_lock);
1406         fp->fi_share_deny = 0;
1407         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1408                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1409         spin_unlock(&fp->fi_lock);
1410 }
1411
1412 static void
1413 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1414 {
1415         int i;
1416         bool change = false;
1417
1418         for (i = 1; i < 4; i++) {
1419                 if ((i & deny) != i) {
1420                         change = true;
1421                         clear_deny(i, stp);
1422                 }
1423         }
1424
1425         /* Recalculate per-file deny mode if there was a change */
1426         if (change)
1427                 recalculate_deny_mode(stp->st_stid.sc_file);
1428 }
1429
1430 /* release all access and file references for a given stateid */
1431 static void
1432 release_all_access(struct nfs4_ol_stateid *stp)
1433 {
1434         int i;
1435         struct nfs4_file *fp = stp->st_stid.sc_file;
1436
1437         if (fp && stp->st_deny_bmap != 0)
1438                 recalculate_deny_mode(fp);
1439
1440         for (i = 1; i < 4; i++) {
1441                 if (test_access(i, stp))
1442                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1443                 clear_access(i, stp);
1444         }
1445 }
1446
1447 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1448 {
1449         kfree(sop->so_owner.data);
1450         sop->so_ops->so_free(sop);
1451 }
1452
1453 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1454 {
1455         struct nfs4_client *clp = sop->so_client;
1456
1457         might_lock(&clp->cl_lock);
1458
1459         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1460                 return;
1461         sop->so_ops->so_unhash(sop);
1462         spin_unlock(&clp->cl_lock);
1463         nfs4_free_stateowner(sop);
1464 }
1465
1466 static bool
1467 nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1468 {
1469         return list_empty(&stp->st_perfile);
1470 }
1471
1472 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1473 {
1474         struct nfs4_file *fp = stp->st_stid.sc_file;
1475
1476         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1477
1478         if (list_empty(&stp->st_perfile))
1479                 return false;
1480
1481         spin_lock(&fp->fi_lock);
1482         list_del_init(&stp->st_perfile);
1483         spin_unlock(&fp->fi_lock);
1484         list_del(&stp->st_perstateowner);
1485         return true;
1486 }
1487
1488 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1489 {
1490         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1491
1492         put_clnt_odstate(stp->st_clnt_odstate);
1493         release_all_access(stp);
1494         if (stp->st_stateowner)
1495                 nfs4_put_stateowner(stp->st_stateowner);
1496         WARN_ON(!list_empty(&stid->sc_cp_list));
1497         kmem_cache_free(stateid_slab, stid);
1498 }
1499
1500 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1501 {
1502         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1503         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1504         struct nfsd_file *nf;
1505
1506         nf = find_any_file(stp->st_stid.sc_file);
1507         if (nf) {
1508                 get_file(nf->nf_file);
1509                 filp_close(nf->nf_file, (fl_owner_t)lo);
1510                 nfsd_file_put(nf);
1511         }
1512         nfs4_free_ol_stateid(stid);
1513 }
1514
1515 /*
1516  * Put the persistent reference to an already unhashed generic stateid, while
1517  * holding the cl_lock. If it's the last reference, then put it onto the
1518  * reaplist for later destruction.
1519  */
1520 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1521                                        struct list_head *reaplist)
1522 {
1523         struct nfs4_stid *s = &stp->st_stid;
1524         struct nfs4_client *clp = s->sc_client;
1525
1526         lockdep_assert_held(&clp->cl_lock);
1527
1528         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1529
1530         if (!refcount_dec_and_test(&s->sc_count)) {
1531                 wake_up_all(&close_wq);
1532                 return;
1533         }
1534
1535         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1536         list_add(&stp->st_locks, reaplist);
1537 }
1538
1539 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1540 {
1541         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1542
1543         if (!unhash_ol_stateid(stp))
1544                 return false;
1545         list_del_init(&stp->st_locks);
1546         nfs4_unhash_stid(&stp->st_stid);
1547         return true;
1548 }
1549
1550 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1551 {
1552         struct nfs4_client *clp = stp->st_stid.sc_client;
1553         bool unhashed;
1554
1555         spin_lock(&clp->cl_lock);
1556         unhashed = unhash_lock_stateid(stp);
1557         spin_unlock(&clp->cl_lock);
1558         if (unhashed)
1559                 nfs4_put_stid(&stp->st_stid);
1560 }
1561
1562 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1563 {
1564         struct nfs4_client *clp = lo->lo_owner.so_client;
1565
1566         lockdep_assert_held(&clp->cl_lock);
1567
1568         list_del_init(&lo->lo_owner.so_strhash);
1569 }
1570
1571 /*
1572  * Free a list of generic stateids that were collected earlier after being
1573  * fully unhashed.
1574  */
1575 static void
1576 free_ol_stateid_reaplist(struct list_head *reaplist)
1577 {
1578         struct nfs4_ol_stateid *stp;
1579         struct nfs4_file *fp;
1580
1581         might_sleep();
1582
1583         while (!list_empty(reaplist)) {
1584                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1585                                        st_locks);
1586                 list_del(&stp->st_locks);
1587                 fp = stp->st_stid.sc_file;
1588                 stp->st_stid.sc_free(&stp->st_stid);
1589                 if (fp)
1590                         put_nfs4_file(fp);
1591         }
1592 }
1593
1594 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1595                                        struct list_head *reaplist)
1596 {
1597         struct nfs4_ol_stateid *stp;
1598
1599         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1600
1601         while (!list_empty(&open_stp->st_locks)) {
1602                 stp = list_entry(open_stp->st_locks.next,
1603                                 struct nfs4_ol_stateid, st_locks);
1604                 WARN_ON(!unhash_lock_stateid(stp));
1605                 put_ol_stateid_locked(stp, reaplist);
1606         }
1607 }
1608
1609 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1610                                 struct list_head *reaplist)
1611 {
1612         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1613
1614         if (!unhash_ol_stateid(stp))
1615                 return false;
1616         release_open_stateid_locks(stp, reaplist);
1617         return true;
1618 }
1619
1620 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1621 {
1622         LIST_HEAD(reaplist);
1623
1624         spin_lock(&stp->st_stid.sc_client->cl_lock);
1625         if (unhash_open_stateid(stp, &reaplist))
1626                 put_ol_stateid_locked(stp, &reaplist);
1627         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1628         free_ol_stateid_reaplist(&reaplist);
1629 }
1630
1631 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1632 {
1633         struct nfs4_client *clp = oo->oo_owner.so_client;
1634
1635         lockdep_assert_held(&clp->cl_lock);
1636
1637         list_del_init(&oo->oo_owner.so_strhash);
1638         list_del_init(&oo->oo_perclient);
1639 }
1640
1641 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1642 {
1643         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1644                                           nfsd_net_id);
1645         struct nfs4_ol_stateid *s;
1646
1647         spin_lock(&nn->client_lock);
1648         s = oo->oo_last_closed_stid;
1649         if (s) {
1650                 list_del_init(&oo->oo_close_lru);
1651                 oo->oo_last_closed_stid = NULL;
1652         }
1653         spin_unlock(&nn->client_lock);
1654         if (s)
1655                 nfs4_put_stid(&s->st_stid);
1656 }
1657
1658 static void release_openowner(struct nfs4_openowner *oo)
1659 {
1660         struct nfs4_ol_stateid *stp;
1661         struct nfs4_client *clp = oo->oo_owner.so_client;
1662         struct list_head reaplist;
1663
1664         INIT_LIST_HEAD(&reaplist);
1665
1666         spin_lock(&clp->cl_lock);
1667         unhash_openowner_locked(oo);
1668         while (!list_empty(&oo->oo_owner.so_stateids)) {
1669                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1670                                 struct nfs4_ol_stateid, st_perstateowner);
1671                 if (unhash_open_stateid(stp, &reaplist))
1672                         put_ol_stateid_locked(stp, &reaplist);
1673         }
1674         spin_unlock(&clp->cl_lock);
1675         free_ol_stateid_reaplist(&reaplist);
1676         release_last_closed_stateid(oo);
1677         nfs4_put_stateowner(&oo->oo_owner);
1678 }
1679
1680 static inline int
1681 hash_sessionid(struct nfs4_sessionid *sessionid)
1682 {
1683         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1684
1685         return sid->sequence % SESSION_HASH_SIZE;
1686 }
1687
1688 #ifdef CONFIG_SUNRPC_DEBUG
1689 static inline void
1690 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1691 {
1692         u32 *ptr = (u32 *)(&sessionid->data[0]);
1693         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1694 }
1695 #else
1696 static inline void
1697 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1698 {
1699 }
1700 #endif
1701
1702 /*
1703  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1704  * won't be used for replay.
1705  */
1706 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1707 {
1708         struct nfs4_stateowner *so = cstate->replay_owner;
1709
1710         if (nfserr == nfserr_replay_me)
1711                 return;
1712
1713         if (!seqid_mutating_err(ntohl(nfserr))) {
1714                 nfsd4_cstate_clear_replay(cstate);
1715                 return;
1716         }
1717         if (!so)
1718                 return;
1719         if (so->so_is_open_owner)
1720                 release_last_closed_stateid(openowner(so));
1721         so->so_seqid++;
1722         return;
1723 }
1724
1725 static void
1726 gen_sessionid(struct nfsd4_session *ses)
1727 {
1728         struct nfs4_client *clp = ses->se_client;
1729         struct nfsd4_sessionid *sid;
1730
1731         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1732         sid->clientid = clp->cl_clientid;
1733         sid->sequence = current_sessionid++;
1734         sid->reserved = 0;
1735 }
1736
1737 /*
1738  * The protocol defines ca_maxresponssize_cached to include the size of
1739  * the rpc header, but all we need to cache is the data starting after
1740  * the end of the initial SEQUENCE operation--the rest we regenerate
1741  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1742  * value that is the number of bytes in our cache plus a few additional
1743  * bytes.  In order to stay on the safe side, and not promise more than
1744  * we can cache, those additional bytes must be the minimum possible: 24
1745  * bytes of rpc header (xid through accept state, with AUTH_NULL
1746  * verifier), 12 for the compound header (with zero-length tag), and 44
1747  * for the SEQUENCE op response:
1748  */
1749 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1750
1751 static void
1752 free_session_slots(struct nfsd4_session *ses)
1753 {
1754         int i;
1755
1756         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1757                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1758                 kfree(ses->se_slots[i]);
1759         }
1760 }
1761
1762 /*
1763  * We don't actually need to cache the rpc and session headers, so we
1764  * can allocate a little less for each slot:
1765  */
1766 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1767 {
1768         u32 size;
1769
1770         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1771                 size = 0;
1772         else
1773                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1774         return size + sizeof(struct nfsd4_slot);
1775 }
1776
1777 /*
1778  * XXX: If we run out of reserved DRC memory we could (up to a point)
1779  * re-negotiate active sessions and reduce their slot usage to make
1780  * room for new connections. For now we just fail the create session.
1781  */
1782 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1783 {
1784         u32 slotsize = slot_bytes(ca);
1785         u32 num = ca->maxreqs;
1786         unsigned long avail, total_avail;
1787         unsigned int scale_factor;
1788
1789         spin_lock(&nfsd_drc_lock);
1790         if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1791                 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1792         else
1793                 /* We have handed out more space than we chose in
1794                  * set_max_drc() to allow.  That isn't really a
1795                  * problem as long as that doesn't make us think we
1796                  * have lots more due to integer overflow.
1797                  */
1798                 total_avail = 0;
1799         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1800         /*
1801          * Never use more than a fraction of the remaining memory,
1802          * unless it's the only way to give this client a slot.
1803          * The chosen fraction is either 1/8 or 1/number of threads,
1804          * whichever is smaller.  This ensures there are adequate
1805          * slots to support multiple clients per thread.
1806          * Give the client one slot even if that would require
1807          * over-allocation--it is better than failure.
1808          */
1809         scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1810
1811         avail = clamp_t(unsigned long, avail, slotsize,
1812                         total_avail/scale_factor);
1813         num = min_t(int, num, avail / slotsize);
1814         num = max_t(int, num, 1);
1815         nfsd_drc_mem_used += num * slotsize;
1816         spin_unlock(&nfsd_drc_lock);
1817
1818         return num;
1819 }
1820
1821 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1822 {
1823         int slotsize = slot_bytes(ca);
1824
1825         spin_lock(&nfsd_drc_lock);
1826         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1827         spin_unlock(&nfsd_drc_lock);
1828 }
1829
1830 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1831                                            struct nfsd4_channel_attrs *battrs)
1832 {
1833         int numslots = fattrs->maxreqs;
1834         int slotsize = slot_bytes(fattrs);
1835         struct nfsd4_session *new;
1836         int i;
1837
1838         BUILD_BUG_ON(struct_size(new, se_slots, NFSD_MAX_SLOTS_PER_SESSION)
1839                      > PAGE_SIZE);
1840
1841         new = kzalloc(struct_size(new, se_slots, numslots), GFP_KERNEL);
1842         if (!new)
1843                 return NULL;
1844         /* allocate each struct nfsd4_slot and data cache in one piece */
1845         for (i = 0; i < numslots; i++) {
1846                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1847                 if (!new->se_slots[i])
1848                         goto out_free;
1849         }
1850
1851         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1852         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1853
1854         return new;
1855 out_free:
1856         while (i--)
1857                 kfree(new->se_slots[i]);
1858         kfree(new);
1859         return NULL;
1860 }
1861
1862 static void free_conn(struct nfsd4_conn *c)
1863 {
1864         svc_xprt_put(c->cn_xprt);
1865         kfree(c);
1866 }
1867
1868 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1869 {
1870         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1871         struct nfs4_client *clp = c->cn_session->se_client;
1872
1873         trace_nfsd_cb_lost(clp);
1874
1875         spin_lock(&clp->cl_lock);
1876         if (!list_empty(&c->cn_persession)) {
1877                 list_del(&c->cn_persession);
1878                 free_conn(c);
1879         }
1880         nfsd4_probe_callback(clp);
1881         spin_unlock(&clp->cl_lock);
1882 }
1883
1884 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1885 {
1886         struct nfsd4_conn *conn;
1887
1888         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1889         if (!conn)
1890                 return NULL;
1891         svc_xprt_get(rqstp->rq_xprt);
1892         conn->cn_xprt = rqstp->rq_xprt;
1893         conn->cn_flags = flags;
1894         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1895         return conn;
1896 }
1897
1898 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1899 {
1900         conn->cn_session = ses;
1901         list_add(&conn->cn_persession, &ses->se_conns);
1902 }
1903
1904 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1905 {
1906         struct nfs4_client *clp = ses->se_client;
1907
1908         spin_lock(&clp->cl_lock);
1909         __nfsd4_hash_conn(conn, ses);
1910         spin_unlock(&clp->cl_lock);
1911 }
1912
1913 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1914 {
1915         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1916         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1917 }
1918
1919 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1920 {
1921         int ret;
1922
1923         nfsd4_hash_conn(conn, ses);
1924         ret = nfsd4_register_conn(conn);
1925         if (ret)
1926                 /* oops; xprt is already down: */
1927                 nfsd4_conn_lost(&conn->cn_xpt_user);
1928         /* We may have gained or lost a callback channel: */
1929         nfsd4_probe_callback_sync(ses->se_client);
1930 }
1931
1932 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1933 {
1934         u32 dir = NFS4_CDFC4_FORE;
1935
1936         if (cses->flags & SESSION4_BACK_CHAN)
1937                 dir |= NFS4_CDFC4_BACK;
1938         return alloc_conn(rqstp, dir);
1939 }
1940
1941 /* must be called under client_lock */
1942 static void nfsd4_del_conns(struct nfsd4_session *s)
1943 {
1944         struct nfs4_client *clp = s->se_client;
1945         struct nfsd4_conn *c;
1946
1947         spin_lock(&clp->cl_lock);
1948         while (!list_empty(&s->se_conns)) {
1949                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1950                 list_del_init(&c->cn_persession);
1951                 spin_unlock(&clp->cl_lock);
1952
1953                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1954                 free_conn(c);
1955
1956                 spin_lock(&clp->cl_lock);
1957         }
1958         spin_unlock(&clp->cl_lock);
1959 }
1960
1961 static void __free_session(struct nfsd4_session *ses)
1962 {
1963         free_session_slots(ses);
1964         kfree(ses);
1965 }
1966
1967 static void free_session(struct nfsd4_session *ses)
1968 {
1969         nfsd4_del_conns(ses);
1970         nfsd4_put_drc_mem(&ses->se_fchannel);
1971         __free_session(ses);
1972 }
1973
1974 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1975 {
1976         int idx;
1977         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1978
1979         new->se_client = clp;
1980         gen_sessionid(new);
1981
1982         INIT_LIST_HEAD(&new->se_conns);
1983
1984         new->se_cb_seq_nr = 1;
1985         new->se_flags = cses->flags;
1986         new->se_cb_prog = cses->callback_prog;
1987         new->se_cb_sec = cses->cb_sec;
1988         atomic_set(&new->se_ref, 0);
1989         idx = hash_sessionid(&new->se_sessionid);
1990         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1991         spin_lock(&clp->cl_lock);
1992         list_add(&new->se_perclnt, &clp->cl_sessions);
1993         spin_unlock(&clp->cl_lock);
1994
1995         {
1996                 struct sockaddr *sa = svc_addr(rqstp);
1997                 /*
1998                  * This is a little silly; with sessions there's no real
1999                  * use for the callback address.  Use the peer address
2000                  * as a reasonable default for now, but consider fixing
2001                  * the rpc client not to require an address in the
2002                  * future:
2003                  */
2004                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
2005                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
2006         }
2007 }
2008
2009 /* caller must hold client_lock */
2010 static struct nfsd4_session *
2011 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
2012 {
2013         struct nfsd4_session *elem;
2014         int idx;
2015         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2016
2017         lockdep_assert_held(&nn->client_lock);
2018
2019         dump_sessionid(__func__, sessionid);
2020         idx = hash_sessionid(sessionid);
2021         /* Search in the appropriate list */
2022         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
2023                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
2024                             NFS4_MAX_SESSIONID_LEN)) {
2025                         return elem;
2026                 }
2027         }
2028
2029         dprintk("%s: session not found\n", __func__);
2030         return NULL;
2031 }
2032
2033 static struct nfsd4_session *
2034 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
2035                 __be32 *ret)
2036 {
2037         struct nfsd4_session *session;
2038         __be32 status = nfserr_badsession;
2039
2040         session = __find_in_sessionid_hashtbl(sessionid, net);
2041         if (!session)
2042                 goto out;
2043         status = nfsd4_get_session_locked(session);
2044         if (status)
2045                 session = NULL;
2046 out:
2047         *ret = status;
2048         return session;
2049 }
2050
2051 /* caller must hold client_lock */
2052 static void
2053 unhash_session(struct nfsd4_session *ses)
2054 {
2055         struct nfs4_client *clp = ses->se_client;
2056         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2057
2058         lockdep_assert_held(&nn->client_lock);
2059
2060         list_del(&ses->se_hash);
2061         spin_lock(&ses->se_client->cl_lock);
2062         list_del(&ses->se_perclnt);
2063         spin_unlock(&ses->se_client->cl_lock);
2064 }
2065
2066 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
2067 static int
2068 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
2069 {
2070         /*
2071          * We're assuming the clid was not given out from a boot
2072          * precisely 2^32 (about 136 years) before this one.  That seems
2073          * a safe assumption:
2074          */
2075         if (clid->cl_boot == (u32)nn->boot_time)
2076                 return 0;
2077         trace_nfsd_clid_stale(clid);
2078         return 1;
2079 }
2080
2081 /*
2082  * XXX Should we use a slab cache ?
2083  * This type of memory management is somewhat inefficient, but we use it
2084  * anyway since SETCLIENTID is not a common operation.
2085  */
2086 static struct nfs4_client *alloc_client(struct xdr_netobj name,
2087                                 struct nfsd_net *nn)
2088 {
2089         struct nfs4_client *clp;
2090         int i;
2091
2092         if (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) {
2093                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
2094                 return NULL;
2095         }
2096         clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
2097         if (clp == NULL)
2098                 return NULL;
2099         xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
2100         if (clp->cl_name.data == NULL)
2101                 goto err_no_name;
2102         clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
2103                                                  sizeof(struct list_head),
2104                                                  GFP_KERNEL);
2105         if (!clp->cl_ownerstr_hashtbl)
2106                 goto err_no_hashtbl;
2107         for (i = 0; i < OWNER_HASH_SIZE; i++)
2108                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
2109         INIT_LIST_HEAD(&clp->cl_sessions);
2110         idr_init(&clp->cl_stateids);
2111         atomic_set(&clp->cl_rpc_users, 0);
2112         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
2113         clp->cl_state = NFSD4_ACTIVE;
2114         atomic_inc(&nn->nfs4_client_count);
2115         atomic_set(&clp->cl_delegs_in_recall, 0);
2116         INIT_LIST_HEAD(&clp->cl_idhash);
2117         INIT_LIST_HEAD(&clp->cl_openowners);
2118         INIT_LIST_HEAD(&clp->cl_delegations);
2119         INIT_LIST_HEAD(&clp->cl_lru);
2120         INIT_LIST_HEAD(&clp->cl_revoked);
2121 #ifdef CONFIG_NFSD_PNFS
2122         INIT_LIST_HEAD(&clp->cl_lo_states);
2123 #endif
2124         INIT_LIST_HEAD(&clp->async_copies);
2125         spin_lock_init(&clp->async_lock);
2126         spin_lock_init(&clp->cl_lock);
2127         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
2128         return clp;
2129 err_no_hashtbl:
2130         kfree(clp->cl_name.data);
2131 err_no_name:
2132         kmem_cache_free(client_slab, clp);
2133         return NULL;
2134 }
2135
2136 static void __free_client(struct kref *k)
2137 {
2138         struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
2139         struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
2140
2141         free_svc_cred(&clp->cl_cred);
2142         kfree(clp->cl_ownerstr_hashtbl);
2143         kfree(clp->cl_name.data);
2144         kfree(clp->cl_nii_domain.data);
2145         kfree(clp->cl_nii_name.data);
2146         idr_destroy(&clp->cl_stateids);
2147         kfree(clp->cl_ra);
2148         kmem_cache_free(client_slab, clp);
2149 }
2150
2151 static void drop_client(struct nfs4_client *clp)
2152 {
2153         kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
2154 }
2155
2156 static void
2157 free_client(struct nfs4_client *clp)
2158 {
2159         while (!list_empty(&clp->cl_sessions)) {
2160                 struct nfsd4_session *ses;
2161                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
2162                                 se_perclnt);
2163                 list_del(&ses->se_perclnt);
2164                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
2165                 free_session(ses);
2166         }
2167         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
2168         if (clp->cl_nfsd_dentry) {
2169                 nfsd_client_rmdir(clp->cl_nfsd_dentry);
2170                 clp->cl_nfsd_dentry = NULL;
2171                 wake_up_all(&expiry_wq);
2172         }
2173         drop_client(clp);
2174 }
2175
2176 /* must be called under the client_lock */
2177 static void
2178 unhash_client_locked(struct nfs4_client *clp)
2179 {
2180         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2181         struct nfsd4_session *ses;
2182
2183         lockdep_assert_held(&nn->client_lock);
2184
2185         /* Mark the client as expired! */
2186         clp->cl_time = 0;
2187         /* Make it invisible */
2188         if (!list_empty(&clp->cl_idhash)) {
2189                 list_del_init(&clp->cl_idhash);
2190                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2191                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
2192                 else
2193                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2194         }
2195         list_del_init(&clp->cl_lru);
2196         spin_lock(&clp->cl_lock);
2197         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2198                 list_del_init(&ses->se_hash);
2199         spin_unlock(&clp->cl_lock);
2200 }
2201
2202 static void
2203 unhash_client(struct nfs4_client *clp)
2204 {
2205         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2206
2207         spin_lock(&nn->client_lock);
2208         unhash_client_locked(clp);
2209         spin_unlock(&nn->client_lock);
2210 }
2211
2212 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
2213 {
2214         if (atomic_read(&clp->cl_rpc_users))
2215                 return nfserr_jukebox;
2216         unhash_client_locked(clp);
2217         return nfs_ok;
2218 }
2219
2220 static void
2221 __destroy_client(struct nfs4_client *clp)
2222 {
2223         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2224         int i;
2225         struct nfs4_openowner *oo;
2226         struct nfs4_delegation *dp;
2227         struct list_head reaplist;
2228
2229         INIT_LIST_HEAD(&reaplist);
2230         spin_lock(&state_lock);
2231         while (!list_empty(&clp->cl_delegations)) {
2232                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2233                 WARN_ON(!unhash_delegation_locked(dp));
2234                 list_add(&dp->dl_recall_lru, &reaplist);
2235         }
2236         spin_unlock(&state_lock);
2237         while (!list_empty(&reaplist)) {
2238                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2239                 list_del_init(&dp->dl_recall_lru);
2240                 destroy_unhashed_deleg(dp);
2241         }
2242         while (!list_empty(&clp->cl_revoked)) {
2243                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2244                 list_del_init(&dp->dl_recall_lru);
2245                 nfs4_put_stid(&dp->dl_stid);
2246         }
2247         while (!list_empty(&clp->cl_openowners)) {
2248                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2249                 nfs4_get_stateowner(&oo->oo_owner);
2250                 release_openowner(oo);
2251         }
2252         for (i = 0; i < OWNER_HASH_SIZE; i++) {
2253                 struct nfs4_stateowner *so, *tmp;
2254
2255                 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2256                                          so_strhash) {
2257                         /* Should be no openowners at this point */
2258                         WARN_ON_ONCE(so->so_is_open_owner);
2259                         remove_blocked_locks(lockowner(so));
2260                 }
2261         }
2262         nfsd4_return_all_client_layouts(clp);
2263         nfsd4_shutdown_copy(clp);
2264         nfsd4_shutdown_callback(clp);
2265         if (clp->cl_cb_conn.cb_xprt)
2266                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2267         atomic_add_unless(&nn->nfs4_client_count, -1, 0);
2268         nfsd4_dec_courtesy_client_count(nn, clp);
2269         free_client(clp);
2270         wake_up_all(&expiry_wq);
2271 }
2272
2273 static void
2274 destroy_client(struct nfs4_client *clp)
2275 {
2276         unhash_client(clp);
2277         __destroy_client(clp);
2278 }
2279
2280 static void inc_reclaim_complete(struct nfs4_client *clp)
2281 {
2282         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2283
2284         if (!nn->track_reclaim_completes)
2285                 return;
2286         if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2287                 return;
2288         if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2289                         nn->reclaim_str_hashtbl_size) {
2290                 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2291                                 clp->net->ns.inum);
2292                 nfsd4_end_grace(nn);
2293         }
2294 }
2295
2296 static void expire_client(struct nfs4_client *clp)
2297 {
2298         unhash_client(clp);
2299         nfsd4_client_record_remove(clp);
2300         __destroy_client(clp);
2301 }
2302
2303 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2304 {
2305         memcpy(target->cl_verifier.data, source->data,
2306                         sizeof(target->cl_verifier.data));
2307 }
2308
2309 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2310 {
2311         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2312         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2313 }
2314
2315 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2316 {
2317         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2318         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2319                                                                 GFP_KERNEL);
2320         target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2321         if ((source->cr_principal && !target->cr_principal) ||
2322             (source->cr_raw_principal && !target->cr_raw_principal) ||
2323             (source->cr_targ_princ && !target->cr_targ_princ))
2324                 return -ENOMEM;
2325
2326         target->cr_flavor = source->cr_flavor;
2327         target->cr_uid = source->cr_uid;
2328         target->cr_gid = source->cr_gid;
2329         target->cr_group_info = source->cr_group_info;
2330         get_group_info(target->cr_group_info);
2331         target->cr_gss_mech = source->cr_gss_mech;
2332         if (source->cr_gss_mech)
2333                 gss_mech_get(source->cr_gss_mech);
2334         return 0;
2335 }
2336
2337 static int
2338 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2339 {
2340         if (o1->len < o2->len)
2341                 return -1;
2342         if (o1->len > o2->len)
2343                 return 1;
2344         return memcmp(o1->data, o2->data, o1->len);
2345 }
2346
2347 static int
2348 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2349 {
2350         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2351 }
2352
2353 static int
2354 same_clid(clientid_t *cl1, clientid_t *cl2)
2355 {
2356         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2357 }
2358
2359 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2360 {
2361         int i;
2362
2363         if (g1->ngroups != g2->ngroups)
2364                 return false;
2365         for (i=0; i<g1->ngroups; i++)
2366                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2367                         return false;
2368         return true;
2369 }
2370
2371 /*
2372  * RFC 3530 language requires clid_inuse be returned when the
2373  * "principal" associated with a requests differs from that previously
2374  * used.  We use uid, gid's, and gss principal string as our best
2375  * approximation.  We also don't want to allow non-gss use of a client
2376  * established using gss: in theory cr_principal should catch that
2377  * change, but in practice cr_principal can be null even in the gss case
2378  * since gssd doesn't always pass down a principal string.
2379  */
2380 static bool is_gss_cred(struct svc_cred *cr)
2381 {
2382         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2383         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2384 }
2385
2386
2387 static bool
2388 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2389 {
2390         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2391                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2392                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2393                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2394                 return false;
2395         /* XXX: check that cr_targ_princ fields match ? */
2396         if (cr1->cr_principal == cr2->cr_principal)
2397                 return true;
2398         if (!cr1->cr_principal || !cr2->cr_principal)
2399                 return false;
2400         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2401 }
2402
2403 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2404 {
2405         struct svc_cred *cr = &rqstp->rq_cred;
2406         u32 service;
2407
2408         if (!cr->cr_gss_mech)
2409                 return false;
2410         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2411         return service == RPC_GSS_SVC_INTEGRITY ||
2412                service == RPC_GSS_SVC_PRIVACY;
2413 }
2414
2415 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2416 {
2417         struct svc_cred *cr = &rqstp->rq_cred;
2418
2419         if (!cl->cl_mach_cred)
2420                 return true;
2421         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2422                 return false;
2423         if (!svc_rqst_integrity_protected(rqstp))
2424                 return false;
2425         if (cl->cl_cred.cr_raw_principal)
2426                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2427                                                 cr->cr_raw_principal);
2428         if (!cr->cr_principal)
2429                 return false;
2430         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2431 }
2432
2433 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2434 {
2435         __be32 verf[2];
2436
2437         /*
2438          * This is opaque to client, so no need to byte-swap. Use
2439          * __force to keep sparse happy
2440          */
2441         verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2442         verf[1] = (__force __be32)nn->clverifier_counter++;
2443         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2444 }
2445
2446 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2447 {
2448         clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2449         clp->cl_clientid.cl_id = nn->clientid_counter++;
2450         gen_confirm(clp, nn);
2451 }
2452
2453 static struct nfs4_stid *
2454 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2455 {
2456         struct nfs4_stid *ret;
2457
2458         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2459         if (!ret || !ret->sc_type)
2460                 return NULL;
2461         return ret;
2462 }
2463
2464 static struct nfs4_stid *
2465 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2466 {
2467         struct nfs4_stid *s;
2468
2469         spin_lock(&cl->cl_lock);
2470         s = find_stateid_locked(cl, t);
2471         if (s != NULL) {
2472                 if (typemask & s->sc_type)
2473                         refcount_inc(&s->sc_count);
2474                 else
2475                         s = NULL;
2476         }
2477         spin_unlock(&cl->cl_lock);
2478         return s;
2479 }
2480
2481 static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2482 {
2483         struct nfsdfs_client *nc;
2484         nc = get_nfsdfs_client(inode);
2485         if (!nc)
2486                 return NULL;
2487         return container_of(nc, struct nfs4_client, cl_nfsdfs);
2488 }
2489
2490 static void seq_quote_mem(struct seq_file *m, char *data, int len)
2491 {
2492         seq_printf(m, "\"");
2493         seq_escape_mem(m, data, len, ESCAPE_HEX | ESCAPE_NAP | ESCAPE_APPEND, "\"\\");
2494         seq_printf(m, "\"");
2495 }
2496
2497 static const char *cb_state2str(int state)
2498 {
2499         switch (state) {
2500         case NFSD4_CB_UP:
2501                 return "UP";
2502         case NFSD4_CB_UNKNOWN:
2503                 return "UNKNOWN";
2504         case NFSD4_CB_DOWN:
2505                 return "DOWN";
2506         case NFSD4_CB_FAULT:
2507                 return "FAULT";
2508         }
2509         return "UNDEFINED";
2510 }
2511
2512 static int client_info_show(struct seq_file *m, void *v)
2513 {
2514         struct inode *inode = file_inode(m->file);
2515         struct nfs4_client *clp;
2516         u64 clid;
2517
2518         clp = get_nfsdfs_clp(inode);
2519         if (!clp)
2520                 return -ENXIO;
2521         memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2522         seq_printf(m, "clientid: 0x%llx\n", clid);
2523         seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2524
2525         if (clp->cl_state == NFSD4_COURTESY)
2526                 seq_puts(m, "status: courtesy\n");
2527         else if (clp->cl_state == NFSD4_EXPIRABLE)
2528                 seq_puts(m, "status: expirable\n");
2529         else if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2530                 seq_puts(m, "status: confirmed\n");
2531         else
2532                 seq_puts(m, "status: unconfirmed\n");
2533         seq_printf(m, "seconds from last renew: %lld\n",
2534                 ktime_get_boottime_seconds() - clp->cl_time);
2535         seq_printf(m, "name: ");
2536         seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2537         seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2538         if (clp->cl_nii_domain.data) {
2539                 seq_printf(m, "Implementation domain: ");
2540                 seq_quote_mem(m, clp->cl_nii_domain.data,
2541                                         clp->cl_nii_domain.len);
2542                 seq_printf(m, "\nImplementation name: ");
2543                 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2544                 seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2545                         clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2546         }
2547         seq_printf(m, "callback state: %s\n", cb_state2str(clp->cl_cb_state));
2548         seq_printf(m, "callback address: %pISpc\n", &clp->cl_cb_conn.cb_addr);
2549         drop_client(clp);
2550
2551         return 0;
2552 }
2553
2554 DEFINE_SHOW_ATTRIBUTE(client_info);
2555
2556 static void *states_start(struct seq_file *s, loff_t *pos)
2557         __acquires(&clp->cl_lock)
2558 {
2559         struct nfs4_client *clp = s->private;
2560         unsigned long id = *pos;
2561         void *ret;
2562
2563         spin_lock(&clp->cl_lock);
2564         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2565         *pos = id;
2566         return ret;
2567 }
2568
2569 static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2570 {
2571         struct nfs4_client *clp = s->private;
2572         unsigned long id = *pos;
2573         void *ret;
2574
2575         id = *pos;
2576         id++;
2577         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2578         *pos = id;
2579         return ret;
2580 }
2581
2582 static void states_stop(struct seq_file *s, void *v)
2583         __releases(&clp->cl_lock)
2584 {
2585         struct nfs4_client *clp = s->private;
2586
2587         spin_unlock(&clp->cl_lock);
2588 }
2589
2590 static void nfs4_show_fname(struct seq_file *s, struct nfsd_file *f)
2591 {
2592          seq_printf(s, "filename: \"%pD2\"", f->nf_file);
2593 }
2594
2595 static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2596 {
2597         struct inode *inode = file_inode(f->nf_file);
2598
2599         seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2600                                         MAJOR(inode->i_sb->s_dev),
2601                                          MINOR(inode->i_sb->s_dev),
2602                                          inode->i_ino);
2603 }
2604
2605 static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2606 {
2607         seq_printf(s, "owner: ");
2608         seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2609 }
2610
2611 static void nfs4_show_stateid(struct seq_file *s, stateid_t *stid)
2612 {
2613         seq_printf(s, "0x%.8x", stid->si_generation);
2614         seq_printf(s, "%12phN", &stid->si_opaque);
2615 }
2616
2617 static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2618 {
2619         struct nfs4_ol_stateid *ols;
2620         struct nfs4_file *nf;
2621         struct nfsd_file *file;
2622         struct nfs4_stateowner *oo;
2623         unsigned int access, deny;
2624
2625         if (st->sc_type != NFS4_OPEN_STID && st->sc_type != NFS4_LOCK_STID)
2626                 return 0; /* XXX: or SEQ_SKIP? */
2627         ols = openlockstateid(st);
2628         oo = ols->st_stateowner;
2629         nf = st->sc_file;
2630
2631         spin_lock(&nf->fi_lock);
2632         file = find_any_file_locked(nf);
2633         if (!file)
2634                 goto out;
2635
2636         seq_printf(s, "- ");
2637         nfs4_show_stateid(s, &st->sc_stateid);
2638         seq_printf(s, ": { type: open, ");
2639
2640         access = bmap_to_share_mode(ols->st_access_bmap);
2641         deny   = bmap_to_share_mode(ols->st_deny_bmap);
2642
2643         seq_printf(s, "access: %s%s, ",
2644                 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2645                 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2646         seq_printf(s, "deny: %s%s, ",
2647                 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2648                 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2649
2650         nfs4_show_superblock(s, file);
2651         seq_printf(s, ", ");
2652         nfs4_show_fname(s, file);
2653         seq_printf(s, ", ");
2654         nfs4_show_owner(s, oo);
2655         seq_printf(s, " }\n");
2656 out:
2657         spin_unlock(&nf->fi_lock);
2658         return 0;
2659 }
2660
2661 static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2662 {
2663         struct nfs4_ol_stateid *ols;
2664         struct nfs4_file *nf;
2665         struct nfsd_file *file;
2666         struct nfs4_stateowner *oo;
2667
2668         ols = openlockstateid(st);
2669         oo = ols->st_stateowner;
2670         nf = st->sc_file;
2671         spin_lock(&nf->fi_lock);
2672         file = find_any_file_locked(nf);
2673         if (!file)
2674                 goto out;
2675
2676         seq_printf(s, "- ");
2677         nfs4_show_stateid(s, &st->sc_stateid);
2678         seq_printf(s, ": { type: lock, ");
2679
2680         /*
2681          * Note: a lock stateid isn't really the same thing as a lock,
2682          * it's the locking state held by one owner on a file, and there
2683          * may be multiple (or no) lock ranges associated with it.
2684          * (Same for the matter is true of open stateids.)
2685          */
2686
2687         nfs4_show_superblock(s, file);
2688         /* XXX: open stateid? */
2689         seq_printf(s, ", ");
2690         nfs4_show_fname(s, file);
2691         seq_printf(s, ", ");
2692         nfs4_show_owner(s, oo);
2693         seq_printf(s, " }\n");
2694 out:
2695         spin_unlock(&nf->fi_lock);
2696         return 0;
2697 }
2698
2699 static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2700 {
2701         struct nfs4_delegation *ds;
2702         struct nfs4_file *nf;
2703         struct nfsd_file *file;
2704
2705         ds = delegstateid(st);
2706         nf = st->sc_file;
2707         spin_lock(&nf->fi_lock);
2708         file = find_deleg_file_locked(nf);
2709         if (!file)
2710                 goto out;
2711
2712         seq_printf(s, "- ");
2713         nfs4_show_stateid(s, &st->sc_stateid);
2714         seq_printf(s, ": { type: deleg, ");
2715
2716         /* Kinda dead code as long as we only support read delegs: */
2717         seq_printf(s, "access: %s, ",
2718                 ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2719
2720         /* XXX: lease time, whether it's being recalled. */
2721
2722         nfs4_show_superblock(s, file);
2723         seq_printf(s, ", ");
2724         nfs4_show_fname(s, file);
2725         seq_printf(s, " }\n");
2726 out:
2727         spin_unlock(&nf->fi_lock);
2728         return 0;
2729 }
2730
2731 static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2732 {
2733         struct nfs4_layout_stateid *ls;
2734         struct nfsd_file *file;
2735
2736         ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2737         file = ls->ls_file;
2738
2739         seq_printf(s, "- ");
2740         nfs4_show_stateid(s, &st->sc_stateid);
2741         seq_printf(s, ": { type: layout, ");
2742
2743         /* XXX: What else would be useful? */
2744
2745         nfs4_show_superblock(s, file);
2746         seq_printf(s, ", ");
2747         nfs4_show_fname(s, file);
2748         seq_printf(s, " }\n");
2749
2750         return 0;
2751 }
2752
2753 static int states_show(struct seq_file *s, void *v)
2754 {
2755         struct nfs4_stid *st = v;
2756
2757         switch (st->sc_type) {
2758         case NFS4_OPEN_STID:
2759                 return nfs4_show_open(s, st);
2760         case NFS4_LOCK_STID:
2761                 return nfs4_show_lock(s, st);
2762         case NFS4_DELEG_STID:
2763                 return nfs4_show_deleg(s, st);
2764         case NFS4_LAYOUT_STID:
2765                 return nfs4_show_layout(s, st);
2766         default:
2767                 return 0; /* XXX: or SEQ_SKIP? */
2768         }
2769         /* XXX: copy stateids? */
2770 }
2771
2772 static struct seq_operations states_seq_ops = {
2773         .start = states_start,
2774         .next = states_next,
2775         .stop = states_stop,
2776         .show = states_show
2777 };
2778
2779 static int client_states_open(struct inode *inode, struct file *file)
2780 {
2781         struct seq_file *s;
2782         struct nfs4_client *clp;
2783         int ret;
2784
2785         clp = get_nfsdfs_clp(inode);
2786         if (!clp)
2787                 return -ENXIO;
2788
2789         ret = seq_open(file, &states_seq_ops);
2790         if (ret)
2791                 return ret;
2792         s = file->private_data;
2793         s->private = clp;
2794         return 0;
2795 }
2796
2797 static int client_opens_release(struct inode *inode, struct file *file)
2798 {
2799         struct seq_file *m = file->private_data;
2800         struct nfs4_client *clp = m->private;
2801
2802         /* XXX: alternatively, we could get/drop in seq start/stop */
2803         drop_client(clp);
2804         return 0;
2805 }
2806
2807 static const struct file_operations client_states_fops = {
2808         .open           = client_states_open,
2809         .read           = seq_read,
2810         .llseek         = seq_lseek,
2811         .release        = client_opens_release,
2812 };
2813
2814 /*
2815  * Normally we refuse to destroy clients that are in use, but here the
2816  * administrator is telling us to just do it.  We also want to wait
2817  * so the caller has a guarantee that the client's locks are gone by
2818  * the time the write returns:
2819  */
2820 static void force_expire_client(struct nfs4_client *clp)
2821 {
2822         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2823         bool already_expired;
2824
2825         trace_nfsd_clid_admin_expired(&clp->cl_clientid);
2826
2827         spin_lock(&nn->client_lock);
2828         clp->cl_time = 0;
2829         spin_unlock(&nn->client_lock);
2830
2831         wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2832         spin_lock(&nn->client_lock);
2833         already_expired = list_empty(&clp->cl_lru);
2834         if (!already_expired)
2835                 unhash_client_locked(clp);
2836         spin_unlock(&nn->client_lock);
2837
2838         if (!already_expired)
2839                 expire_client(clp);
2840         else
2841                 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2842 }
2843
2844 static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2845                                    size_t size, loff_t *pos)
2846 {
2847         char *data;
2848         struct nfs4_client *clp;
2849
2850         data = simple_transaction_get(file, buf, size);
2851         if (IS_ERR(data))
2852                 return PTR_ERR(data);
2853         if (size != 7 || 0 != memcmp(data, "expire\n", 7))
2854                 return -EINVAL;
2855         clp = get_nfsdfs_clp(file_inode(file));
2856         if (!clp)
2857                 return -ENXIO;
2858         force_expire_client(clp);
2859         drop_client(clp);
2860         return 7;
2861 }
2862
2863 static const struct file_operations client_ctl_fops = {
2864         .write          = client_ctl_write,
2865         .release        = simple_transaction_release,
2866 };
2867
2868 static const struct tree_descr client_files[] = {
2869         [0] = {"info", &client_info_fops, S_IRUSR},
2870         [1] = {"states", &client_states_fops, S_IRUSR},
2871         [2] = {"ctl", &client_ctl_fops, S_IWUSR},
2872         [3] = {""},
2873 };
2874
2875 static int
2876 nfsd4_cb_recall_any_done(struct nfsd4_callback *cb,
2877                                 struct rpc_task *task)
2878 {
2879         trace_nfsd_cb_recall_any_done(cb, task);
2880         switch (task->tk_status) {
2881         case -NFS4ERR_DELAY:
2882                 rpc_delay(task, 2 * HZ);
2883                 return 0;
2884         default:
2885                 return 1;
2886         }
2887 }
2888
2889 static void
2890 nfsd4_cb_recall_any_release(struct nfsd4_callback *cb)
2891 {
2892         struct nfs4_client *clp = cb->cb_clp;
2893         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2894
2895         spin_lock(&nn->client_lock);
2896         clear_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
2897         put_client_renew_locked(clp);
2898         spin_unlock(&nn->client_lock);
2899 }
2900
2901 static const struct nfsd4_callback_ops nfsd4_cb_recall_any_ops = {
2902         .done           = nfsd4_cb_recall_any_done,
2903         .release        = nfsd4_cb_recall_any_release,
2904 };
2905
2906 static struct nfs4_client *create_client(struct xdr_netobj name,
2907                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2908 {
2909         struct nfs4_client *clp;
2910         struct sockaddr *sa = svc_addr(rqstp);
2911         int ret;
2912         struct net *net = SVC_NET(rqstp);
2913         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2914         struct dentry *dentries[ARRAY_SIZE(client_files)];
2915
2916         clp = alloc_client(name, nn);
2917         if (clp == NULL)
2918                 return NULL;
2919
2920         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2921         if (ret) {
2922                 free_client(clp);
2923                 return NULL;
2924         }
2925         gen_clid(clp, nn);
2926         kref_init(&clp->cl_nfsdfs.cl_ref);
2927         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2928         clp->cl_time = ktime_get_boottime_seconds();
2929         clear_bit(0, &clp->cl_cb_slot_busy);
2930         copy_verf(clp, verf);
2931         memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
2932         clp->cl_cb_session = NULL;
2933         clp->net = net;
2934         clp->cl_nfsd_dentry = nfsd_client_mkdir(
2935                 nn, &clp->cl_nfsdfs,
2936                 clp->cl_clientid.cl_id - nn->clientid_base,
2937                 client_files, dentries);
2938         clp->cl_nfsd_info_dentry = dentries[0];
2939         if (!clp->cl_nfsd_dentry) {
2940                 free_client(clp);
2941                 return NULL;
2942         }
2943         clp->cl_ra = kzalloc(sizeof(*clp->cl_ra), GFP_KERNEL);
2944         if (!clp->cl_ra) {
2945                 free_client(clp);
2946                 return NULL;
2947         }
2948         clp->cl_ra_time = 0;
2949         nfsd4_init_cb(&clp->cl_ra->ra_cb, clp, &nfsd4_cb_recall_any_ops,
2950                         NFSPROC4_CLNT_CB_RECALL_ANY);
2951         return clp;
2952 }
2953
2954 static void
2955 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2956 {
2957         struct rb_node **new = &(root->rb_node), *parent = NULL;
2958         struct nfs4_client *clp;
2959
2960         while (*new) {
2961                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2962                 parent = *new;
2963
2964                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2965                         new = &((*new)->rb_left);
2966                 else
2967                         new = &((*new)->rb_right);
2968         }
2969
2970         rb_link_node(&new_clp->cl_namenode, parent, new);
2971         rb_insert_color(&new_clp->cl_namenode, root);
2972 }
2973
2974 static struct nfs4_client *
2975 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2976 {
2977         int cmp;
2978         struct rb_node *node = root->rb_node;
2979         struct nfs4_client *clp;
2980
2981         while (node) {
2982                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2983                 cmp = compare_blob(&clp->cl_name, name);
2984                 if (cmp > 0)
2985                         node = node->rb_left;
2986                 else if (cmp < 0)
2987                         node = node->rb_right;
2988                 else
2989                         return clp;
2990         }
2991         return NULL;
2992 }
2993
2994 static void
2995 add_to_unconfirmed(struct nfs4_client *clp)
2996 {
2997         unsigned int idhashval;
2998         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2999
3000         lockdep_assert_held(&nn->client_lock);
3001
3002         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3003         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
3004         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3005         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
3006         renew_client_locked(clp);
3007 }
3008
3009 static void
3010 move_to_confirmed(struct nfs4_client *clp)
3011 {
3012         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3013         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3014
3015         lockdep_assert_held(&nn->client_lock);
3016
3017         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
3018         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
3019         add_clp_to_name_tree(clp, &nn->conf_name_tree);
3020         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3021         trace_nfsd_clid_confirmed(&clp->cl_clientid);
3022         renew_client_locked(clp);
3023 }
3024
3025 static struct nfs4_client *
3026 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
3027 {
3028         struct nfs4_client *clp;
3029         unsigned int idhashval = clientid_hashval(clid->cl_id);
3030
3031         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
3032                 if (same_clid(&clp->cl_clientid, clid)) {
3033                         if ((bool)clp->cl_minorversion != sessions)
3034                                 return NULL;
3035                         renew_client_locked(clp);
3036                         return clp;
3037                 }
3038         }
3039         return NULL;
3040 }
3041
3042 static struct nfs4_client *
3043 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3044 {
3045         struct list_head *tbl = nn->conf_id_hashtbl;
3046
3047         lockdep_assert_held(&nn->client_lock);
3048         return find_client_in_id_table(tbl, clid, sessions);
3049 }
3050
3051 static struct nfs4_client *
3052 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3053 {
3054         struct list_head *tbl = nn->unconf_id_hashtbl;
3055
3056         lockdep_assert_held(&nn->client_lock);
3057         return find_client_in_id_table(tbl, clid, sessions);
3058 }
3059
3060 static bool clp_used_exchangeid(struct nfs4_client *clp)
3061 {
3062         return clp->cl_exchange_flags != 0;
3063 }
3064
3065 static struct nfs4_client *
3066 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3067 {
3068         lockdep_assert_held(&nn->client_lock);
3069         return find_clp_in_name_tree(name, &nn->conf_name_tree);
3070 }
3071
3072 static struct nfs4_client *
3073 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3074 {
3075         lockdep_assert_held(&nn->client_lock);
3076         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
3077 }
3078
3079 static void
3080 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
3081 {
3082         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
3083         struct sockaddr *sa = svc_addr(rqstp);
3084         u32 scopeid = rpc_get_scope_id(sa);
3085         unsigned short expected_family;
3086
3087         /* Currently, we only support tcp and tcp6 for the callback channel */
3088         if (se->se_callback_netid_len == 3 &&
3089             !memcmp(se->se_callback_netid_val, "tcp", 3))
3090                 expected_family = AF_INET;
3091         else if (se->se_callback_netid_len == 4 &&
3092                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
3093                 expected_family = AF_INET6;
3094         else
3095                 goto out_err;
3096
3097         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
3098                                             se->se_callback_addr_len,
3099                                             (struct sockaddr *)&conn->cb_addr,
3100                                             sizeof(conn->cb_addr));
3101
3102         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
3103                 goto out_err;
3104
3105         if (conn->cb_addr.ss_family == AF_INET6)
3106                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
3107
3108         conn->cb_prog = se->se_callback_prog;
3109         conn->cb_ident = se->se_callback_ident;
3110         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
3111         trace_nfsd_cb_args(clp, conn);
3112         return;
3113 out_err:
3114         conn->cb_addr.ss_family = AF_UNSPEC;
3115         conn->cb_addrlen = 0;
3116         trace_nfsd_cb_nodelegs(clp);
3117         return;
3118 }
3119
3120 /*
3121  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
3122  */
3123 static void
3124 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
3125 {
3126         struct xdr_buf *buf = resp->xdr->buf;
3127         struct nfsd4_slot *slot = resp->cstate.slot;
3128         unsigned int base;
3129
3130         dprintk("--> %s slot %p\n", __func__, slot);
3131
3132         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
3133         slot->sl_opcnt = resp->opcnt;
3134         slot->sl_status = resp->cstate.status;
3135         free_svc_cred(&slot->sl_cred);
3136         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
3137
3138         if (!nfsd4_cache_this(resp)) {
3139                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
3140                 return;
3141         }
3142         slot->sl_flags |= NFSD4_SLOT_CACHED;
3143
3144         base = resp->cstate.data_offset;
3145         slot->sl_datalen = buf->len - base;
3146         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
3147                 WARN(1, "%s: sessions DRC could not cache compound\n",
3148                      __func__);
3149         return;
3150 }
3151
3152 /*
3153  * Encode the replay sequence operation from the slot values.
3154  * If cachethis is FALSE encode the uncached rep error on the next
3155  * operation which sets resp->p and increments resp->opcnt for
3156  * nfs4svc_encode_compoundres.
3157  *
3158  */
3159 static __be32
3160 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
3161                           struct nfsd4_compoundres *resp)
3162 {
3163         struct nfsd4_op *op;
3164         struct nfsd4_slot *slot = resp->cstate.slot;
3165
3166         /* Encode the replayed sequence operation */
3167         op = &args->ops[resp->opcnt - 1];
3168         nfsd4_encode_operation(resp, op);
3169
3170         if (slot->sl_flags & NFSD4_SLOT_CACHED)
3171                 return op->status;
3172         if (args->opcnt == 1) {
3173                 /*
3174                  * The original operation wasn't a solo sequence--we
3175                  * always cache those--so this retry must not match the
3176                  * original:
3177                  */
3178                 op->status = nfserr_seq_false_retry;
3179         } else {
3180                 op = &args->ops[resp->opcnt++];
3181                 op->status = nfserr_retry_uncached_rep;
3182                 nfsd4_encode_operation(resp, op);
3183         }
3184         return op->status;
3185 }
3186
3187 /*
3188  * The sequence operation is not cached because we can use the slot and
3189  * session values.
3190  */
3191 static __be32
3192 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
3193                          struct nfsd4_sequence *seq)
3194 {
3195         struct nfsd4_slot *slot = resp->cstate.slot;
3196         struct xdr_stream *xdr = resp->xdr;
3197         __be32 *p;
3198         __be32 status;
3199
3200         dprintk("--> %s slot %p\n", __func__, slot);
3201
3202         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
3203         if (status)
3204                 return status;
3205
3206         p = xdr_reserve_space(xdr, slot->sl_datalen);
3207         if (!p) {
3208                 WARN_ON_ONCE(1);
3209                 return nfserr_serverfault;
3210         }
3211         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
3212         xdr_commit_encode(xdr);
3213
3214         resp->opcnt = slot->sl_opcnt;
3215         return slot->sl_status;
3216 }
3217
3218 /*
3219  * Set the exchange_id flags returned by the server.
3220  */
3221 static void
3222 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
3223 {
3224 #ifdef CONFIG_NFSD_PNFS
3225         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
3226 #else
3227         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
3228 #endif
3229
3230         /* Referrals are supported, Migration is not. */
3231         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
3232
3233         /* set the wire flags to return to client. */
3234         clid->flags = new->cl_exchange_flags;
3235 }
3236
3237 static bool client_has_openowners(struct nfs4_client *clp)
3238 {
3239         struct nfs4_openowner *oo;
3240
3241         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
3242                 if (!list_empty(&oo->oo_owner.so_stateids))
3243                         return true;
3244         }
3245         return false;
3246 }
3247
3248 static bool client_has_state(struct nfs4_client *clp)
3249 {
3250         return client_has_openowners(clp)
3251 #ifdef CONFIG_NFSD_PNFS
3252                 || !list_empty(&clp->cl_lo_states)
3253 #endif
3254                 || !list_empty(&clp->cl_delegations)
3255                 || !list_empty(&clp->cl_sessions)
3256                 || !list_empty(&clp->async_copies);
3257 }
3258
3259 static __be32 copy_impl_id(struct nfs4_client *clp,
3260                                 struct nfsd4_exchange_id *exid)
3261 {
3262         if (!exid->nii_domain.data)
3263                 return 0;
3264         xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
3265         if (!clp->cl_nii_domain.data)
3266                 return nfserr_jukebox;
3267         xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
3268         if (!clp->cl_nii_name.data)
3269                 return nfserr_jukebox;
3270         clp->cl_nii_time = exid->nii_time;
3271         return 0;
3272 }
3273
3274 __be32
3275 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3276                 union nfsd4_op_u *u)
3277 {
3278         struct nfsd4_exchange_id *exid = &u->exchange_id;
3279         struct nfs4_client *conf, *new;
3280         struct nfs4_client *unconf = NULL;
3281         __be32 status;
3282         char                    addr_str[INET6_ADDRSTRLEN];
3283         nfs4_verifier           verf = exid->verifier;
3284         struct sockaddr         *sa = svc_addr(rqstp);
3285         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3286         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3287
3288         rpc_ntop(sa, addr_str, sizeof(addr_str));
3289         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3290                 "ip_addr=%s flags %x, spa_how %u\n",
3291                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
3292                 addr_str, exid->flags, exid->spa_how);
3293
3294         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3295                 return nfserr_inval;
3296
3297         new = create_client(exid->clname, rqstp, &verf);
3298         if (new == NULL)
3299                 return nfserr_jukebox;
3300         status = copy_impl_id(new, exid);
3301         if (status)
3302                 goto out_nolock;
3303
3304         switch (exid->spa_how) {
3305         case SP4_MACH_CRED:
3306                 exid->spo_must_enforce[0] = 0;
3307                 exid->spo_must_enforce[1] = (
3308                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
3309                         1 << (OP_EXCHANGE_ID - 32) |
3310                         1 << (OP_CREATE_SESSION - 32) |
3311                         1 << (OP_DESTROY_SESSION - 32) |
3312                         1 << (OP_DESTROY_CLIENTID - 32));
3313
3314                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3315                                         1 << (OP_OPEN_DOWNGRADE) |
3316                                         1 << (OP_LOCKU) |
3317                                         1 << (OP_DELEGRETURN));
3318
3319                 exid->spo_must_allow[1] &= (
3320                                         1 << (OP_TEST_STATEID - 32) |
3321                                         1 << (OP_FREE_STATEID - 32));
3322                 if (!svc_rqst_integrity_protected(rqstp)) {
3323                         status = nfserr_inval;
3324                         goto out_nolock;
3325                 }
3326                 /*
3327                  * Sometimes userspace doesn't give us a principal.
3328                  * Which is a bug, really.  Anyway, we can't enforce
3329                  * MACH_CRED in that case, better to give up now:
3330                  */
3331                 if (!new->cl_cred.cr_principal &&
3332                                         !new->cl_cred.cr_raw_principal) {
3333                         status = nfserr_serverfault;
3334                         goto out_nolock;
3335                 }
3336                 new->cl_mach_cred = true;
3337                 break;
3338         case SP4_NONE:
3339                 break;
3340         default:                                /* checked by xdr code */
3341                 WARN_ON_ONCE(1);
3342                 fallthrough;
3343         case SP4_SSV:
3344                 status = nfserr_encr_alg_unsupp;
3345                 goto out_nolock;
3346         }
3347
3348         /* Cases below refer to rfc 5661 section 18.35.4: */
3349         spin_lock(&nn->client_lock);
3350         conf = find_confirmed_client_by_name(&exid->clname, nn);
3351         if (conf) {
3352                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3353                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3354
3355                 if (update) {
3356                         if (!clp_used_exchangeid(conf)) { /* buggy client */
3357                                 status = nfserr_inval;
3358                                 goto out;
3359                         }
3360                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
3361                                 status = nfserr_wrong_cred;
3362                                 goto out;
3363                         }
3364                         if (!creds_match) { /* case 9 */
3365                                 status = nfserr_perm;
3366                                 goto out;
3367                         }
3368                         if (!verfs_match) { /* case 8 */
3369                                 status = nfserr_not_same;
3370                                 goto out;
3371                         }
3372                         /* case 6 */
3373                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3374                         trace_nfsd_clid_confirmed_r(conf);
3375                         goto out_copy;
3376                 }
3377                 if (!creds_match) { /* case 3 */
3378                         if (client_has_state(conf)) {
3379                                 status = nfserr_clid_inuse;
3380                                 trace_nfsd_clid_cred_mismatch(conf, rqstp);
3381                                 goto out;
3382                         }
3383                         goto out_new;
3384                 }
3385                 if (verfs_match) { /* case 2 */
3386                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3387                         trace_nfsd_clid_confirmed_r(conf);
3388                         goto out_copy;
3389                 }
3390                 /* case 5, client reboot */
3391                 trace_nfsd_clid_verf_mismatch(conf, rqstp, &verf);
3392                 conf = NULL;
3393                 goto out_new;
3394         }
3395
3396         if (update) { /* case 7 */
3397                 status = nfserr_noent;
3398                 goto out;
3399         }
3400
3401         unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
3402         if (unconf) /* case 4, possible retry or client restart */
3403                 unhash_client_locked(unconf);
3404
3405         /* case 1, new owner ID */
3406         trace_nfsd_clid_fresh(new);
3407
3408 out_new:
3409         if (conf) {
3410                 status = mark_client_expired_locked(conf);
3411                 if (status)
3412                         goto out;
3413                 trace_nfsd_clid_replaced(&conf->cl_clientid);
3414         }
3415         new->cl_minorversion = cstate->minorversion;
3416         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3417         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3418
3419         add_to_unconfirmed(new);
3420         swap(new, conf);
3421 out_copy:
3422         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3423         exid->clientid.cl_id = conf->cl_clientid.cl_id;
3424
3425         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3426         nfsd4_set_ex_flags(conf, exid);
3427
3428         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3429                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3430         status = nfs_ok;
3431
3432 out:
3433         spin_unlock(&nn->client_lock);
3434 out_nolock:
3435         if (new)
3436                 expire_client(new);
3437         if (unconf) {
3438                 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
3439                 expire_client(unconf);
3440         }
3441         return status;
3442 }
3443
3444 static __be32
3445 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3446 {
3447         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3448                 slot_seqid);
3449
3450         /* The slot is in use, and no response has been sent. */
3451         if (slot_inuse) {
3452                 if (seqid == slot_seqid)
3453                         return nfserr_jukebox;
3454                 else
3455                         return nfserr_seq_misordered;
3456         }
3457         /* Note unsigned 32-bit arithmetic handles wraparound: */
3458         if (likely(seqid == slot_seqid + 1))
3459                 return nfs_ok;
3460         if (seqid == slot_seqid)
3461                 return nfserr_replay_cache;
3462         return nfserr_seq_misordered;
3463 }
3464
3465 /*
3466  * Cache the create session result into the create session single DRC
3467  * slot cache by saving the xdr structure. sl_seqid has been set.
3468  * Do this for solo or embedded create session operations.
3469  */
3470 static void
3471 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3472                            struct nfsd4_clid_slot *slot, __be32 nfserr)
3473 {
3474         slot->sl_status = nfserr;
3475         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3476 }
3477
3478 static __be32
3479 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3480                             struct nfsd4_clid_slot *slot)
3481 {
3482         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3483         return slot->sl_status;
3484 }
3485
3486 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3487                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3488                         1 +     /* MIN tag is length with zero, only length */ \
3489                         3 +     /* version, opcount, opcode */ \
3490                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3491                                 /* seqid, slotID, slotID, cache */ \
3492                         4 ) * sizeof(__be32))
3493
3494 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3495                         2 +     /* verifier: AUTH_NULL, length 0 */\
3496                         1 +     /* status */ \
3497                         1 +     /* MIN tag is length with zero, only length */ \
3498                         3 +     /* opcount, opcode, opstatus*/ \
3499                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3500                                 /* seqid, slotID, slotID, slotID, status */ \
3501                         5 ) * sizeof(__be32))
3502
3503 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3504 {
3505         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3506
3507         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3508                 return nfserr_toosmall;
3509         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3510                 return nfserr_toosmall;
3511         ca->headerpadsz = 0;
3512         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3513         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3514         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3515         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3516                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3517         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3518         /*
3519          * Note decreasing slot size below client's request may make it
3520          * difficult for client to function correctly, whereas
3521          * decreasing the number of slots will (just?) affect
3522          * performance.  When short on memory we therefore prefer to
3523          * decrease number of slots instead of their size.  Clients that
3524          * request larger slots than they need will get poor results:
3525          * Note that we always allow at least one slot, because our
3526          * accounting is soft and provides no guarantees either way.
3527          */
3528         ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3529
3530         return nfs_ok;
3531 }
3532
3533 /*
3534  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3535  * These are based on similar macros in linux/sunrpc/msg_prot.h .
3536  */
3537 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3538         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3539
3540 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3541         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3542
3543 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
3544                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3545 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
3546                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3547                                  sizeof(__be32))
3548
3549 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3550 {
3551         ca->headerpadsz = 0;
3552
3553         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3554                 return nfserr_toosmall;
3555         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3556                 return nfserr_toosmall;
3557         ca->maxresp_cached = 0;
3558         if (ca->maxops < 2)
3559                 return nfserr_toosmall;
3560
3561         return nfs_ok;
3562 }
3563
3564 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3565 {
3566         switch (cbs->flavor) {
3567         case RPC_AUTH_NULL:
3568         case RPC_AUTH_UNIX:
3569                 return nfs_ok;
3570         default:
3571                 /*
3572                  * GSS case: the spec doesn't allow us to return this
3573                  * error.  But it also doesn't allow us not to support
3574                  * GSS.
3575                  * I'd rather this fail hard than return some error the
3576                  * client might think it can already handle:
3577                  */
3578                 return nfserr_encr_alg_unsupp;
3579         }
3580 }
3581
3582 __be32
3583 nfsd4_create_session(struct svc_rqst *rqstp,
3584                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3585 {
3586         struct nfsd4_create_session *cr_ses = &u->create_session;
3587         struct sockaddr *sa = svc_addr(rqstp);
3588         struct nfs4_client *conf, *unconf;
3589         struct nfs4_client *old = NULL;
3590         struct nfsd4_session *new;
3591         struct nfsd4_conn *conn;
3592         struct nfsd4_clid_slot *cs_slot = NULL;
3593         __be32 status = 0;
3594         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3595
3596         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3597                 return nfserr_inval;
3598         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3599         if (status)
3600                 return status;
3601         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3602         if (status)
3603                 return status;
3604         status = check_backchannel_attrs(&cr_ses->back_channel);
3605         if (status)
3606                 goto out_release_drc_mem;
3607         status = nfserr_jukebox;
3608         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3609         if (!new)
3610                 goto out_release_drc_mem;
3611         conn = alloc_conn_from_crses(rqstp, cr_ses);
3612         if (!conn)
3613                 goto out_free_session;
3614
3615         spin_lock(&nn->client_lock);
3616         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3617         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3618         WARN_ON_ONCE(conf && unconf);
3619
3620         if (conf) {
3621                 status = nfserr_wrong_cred;
3622                 if (!nfsd4_mach_creds_match(conf, rqstp))
3623                         goto out_free_conn;
3624                 cs_slot = &conf->cl_cs_slot;
3625                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3626                 if (status) {
3627                         if (status == nfserr_replay_cache)
3628                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3629                         goto out_free_conn;
3630                 }
3631         } else if (unconf) {
3632                 status = nfserr_clid_inuse;
3633                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3634                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3635                         trace_nfsd_clid_cred_mismatch(unconf, rqstp);
3636                         goto out_free_conn;
3637                 }
3638                 status = nfserr_wrong_cred;
3639                 if (!nfsd4_mach_creds_match(unconf, rqstp))
3640                         goto out_free_conn;
3641                 cs_slot = &unconf->cl_cs_slot;
3642                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3643                 if (status) {
3644                         /* an unconfirmed replay returns misordered */
3645                         status = nfserr_seq_misordered;
3646                         goto out_free_conn;
3647                 }
3648                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3649                 if (old) {
3650                         status = mark_client_expired_locked(old);
3651                         if (status) {
3652                                 old = NULL;
3653                                 goto out_free_conn;
3654                         }
3655                         trace_nfsd_clid_replaced(&old->cl_clientid);
3656                 }
3657                 move_to_confirmed(unconf);
3658                 conf = unconf;
3659         } else {
3660                 status = nfserr_stale_clientid;
3661                 goto out_free_conn;
3662         }
3663         status = nfs_ok;
3664         /* Persistent sessions are not supported */
3665         cr_ses->flags &= ~SESSION4_PERSIST;
3666         /* Upshifting from TCP to RDMA is not supported */
3667         cr_ses->flags &= ~SESSION4_RDMA;
3668
3669         init_session(rqstp, new, conf, cr_ses);
3670         nfsd4_get_session_locked(new);
3671
3672         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3673                NFS4_MAX_SESSIONID_LEN);
3674         cs_slot->sl_seqid++;
3675         cr_ses->seqid = cs_slot->sl_seqid;
3676
3677         /* cache solo and embedded create sessions under the client_lock */
3678         nfsd4_cache_create_session(cr_ses, cs_slot, status);
3679         spin_unlock(&nn->client_lock);
3680         if (conf == unconf)
3681                 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
3682         /* init connection and backchannel */
3683         nfsd4_init_conn(rqstp, conn, new);
3684         nfsd4_put_session(new);
3685         if (old)
3686                 expire_client(old);
3687         return status;
3688 out_free_conn:
3689         spin_unlock(&nn->client_lock);
3690         free_conn(conn);
3691         if (old)
3692                 expire_client(old);
3693 out_free_session:
3694         __free_session(new);
3695 out_release_drc_mem:
3696         nfsd4_put_drc_mem(&cr_ses->fore_channel);
3697         return status;
3698 }
3699
3700 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3701 {
3702         switch (*dir) {
3703         case NFS4_CDFC4_FORE:
3704         case NFS4_CDFC4_BACK:
3705                 return nfs_ok;
3706         case NFS4_CDFC4_FORE_OR_BOTH:
3707         case NFS4_CDFC4_BACK_OR_BOTH:
3708                 *dir = NFS4_CDFC4_BOTH;
3709                 return nfs_ok;
3710         }
3711         return nfserr_inval;
3712 }
3713
3714 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3715                 struct nfsd4_compound_state *cstate,
3716                 union nfsd4_op_u *u)
3717 {
3718         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3719         struct nfsd4_session *session = cstate->session;
3720         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3721         __be32 status;
3722
3723         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3724         if (status)
3725                 return status;
3726         spin_lock(&nn->client_lock);
3727         session->se_cb_prog = bc->bc_cb_program;
3728         session->se_cb_sec = bc->bc_cb_sec;
3729         spin_unlock(&nn->client_lock);
3730
3731         nfsd4_probe_callback(session->se_client);
3732
3733         return nfs_ok;
3734 }
3735
3736 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3737 {
3738         struct nfsd4_conn *c;
3739
3740         list_for_each_entry(c, &s->se_conns, cn_persession) {
3741                 if (c->cn_xprt == xpt) {
3742                         return c;
3743                 }
3744         }
3745         return NULL;
3746 }
3747
3748 static __be32 nfsd4_match_existing_connection(struct svc_rqst *rqst,
3749                 struct nfsd4_session *session, u32 req, struct nfsd4_conn **conn)
3750 {
3751         struct nfs4_client *clp = session->se_client;
3752         struct svc_xprt *xpt = rqst->rq_xprt;
3753         struct nfsd4_conn *c;
3754         __be32 status;
3755
3756         /* Following the last paragraph of RFC 5661 Section 18.34.3: */
3757         spin_lock(&clp->cl_lock);
3758         c = __nfsd4_find_conn(xpt, session);
3759         if (!c)
3760                 status = nfserr_noent;
3761         else if (req == c->cn_flags)
3762                 status = nfs_ok;
3763         else if (req == NFS4_CDFC4_FORE_OR_BOTH &&
3764                                 c->cn_flags != NFS4_CDFC4_BACK)
3765                 status = nfs_ok;
3766         else if (req == NFS4_CDFC4_BACK_OR_BOTH &&
3767                                 c->cn_flags != NFS4_CDFC4_FORE)
3768                 status = nfs_ok;
3769         else
3770                 status = nfserr_inval;
3771         spin_unlock(&clp->cl_lock);
3772         if (status == nfs_ok && conn)
3773                 *conn = c;
3774         return status;
3775 }
3776
3777 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
3778                      struct nfsd4_compound_state *cstate,
3779                      union nfsd4_op_u *u)
3780 {
3781         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
3782         __be32 status;
3783         struct nfsd4_conn *conn;
3784         struct nfsd4_session *session;
3785         struct net *net = SVC_NET(rqstp);
3786         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3787
3788         if (!nfsd4_last_compound_op(rqstp))
3789                 return nfserr_not_only_op;
3790         spin_lock(&nn->client_lock);
3791         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
3792         spin_unlock(&nn->client_lock);
3793         if (!session)
3794                 goto out_no_session;
3795         status = nfserr_wrong_cred;
3796         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
3797                 goto out;
3798         status = nfsd4_match_existing_connection(rqstp, session,
3799                         bcts->dir, &conn);
3800         if (status == nfs_ok) {
3801                 if (bcts->dir == NFS4_CDFC4_FORE_OR_BOTH ||
3802                                 bcts->dir == NFS4_CDFC4_BACK)
3803                         conn->cn_flags |= NFS4_CDFC4_BACK;
3804                 nfsd4_probe_callback(session->se_client);
3805                 goto out;
3806         }
3807         if (status == nfserr_inval)
3808                 goto out;
3809         status = nfsd4_map_bcts_dir(&bcts->dir);
3810         if (status)
3811                 goto out;
3812         conn = alloc_conn(rqstp, bcts->dir);
3813         status = nfserr_jukebox;
3814         if (!conn)
3815                 goto out;
3816         nfsd4_init_conn(rqstp, conn, session);
3817         status = nfs_ok;
3818 out:
3819         nfsd4_put_session(session);
3820 out_no_session:
3821         return status;
3822 }
3823
3824 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3825 {
3826         if (!cstate->session)
3827                 return false;
3828         return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3829 }
3830
3831 __be32
3832 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3833                 union nfsd4_op_u *u)
3834 {
3835         struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3836         struct nfsd4_session *ses;
3837         __be32 status;
3838         int ref_held_by_me = 0;
3839         struct net *net = SVC_NET(r);
3840         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3841
3842         status = nfserr_not_only_op;
3843         if (nfsd4_compound_in_session(cstate, sessionid)) {
3844                 if (!nfsd4_last_compound_op(r))
3845                         goto out;
3846                 ref_held_by_me++;
3847         }
3848         dump_sessionid(__func__, sessionid);
3849         spin_lock(&nn->client_lock);
3850         ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3851         if (!ses)
3852                 goto out_client_lock;
3853         status = nfserr_wrong_cred;
3854         if (!nfsd4_mach_creds_match(ses->se_client, r))
3855                 goto out_put_session;
3856         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3857         if (status)
3858                 goto out_put_session;
3859         unhash_session(ses);
3860         spin_unlock(&nn->client_lock);
3861
3862         nfsd4_probe_callback_sync(ses->se_client);
3863
3864         spin_lock(&nn->client_lock);
3865         status = nfs_ok;
3866 out_put_session:
3867         nfsd4_put_session_locked(ses);
3868 out_client_lock:
3869         spin_unlock(&nn->client_lock);
3870 out:
3871         return status;
3872 }
3873
3874 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3875 {
3876         struct nfs4_client *clp = ses->se_client;
3877         struct nfsd4_conn *c;
3878         __be32 status = nfs_ok;
3879         int ret;
3880
3881         spin_lock(&clp->cl_lock);
3882         c = __nfsd4_find_conn(new->cn_xprt, ses);
3883         if (c)
3884                 goto out_free;
3885         status = nfserr_conn_not_bound_to_session;
3886         if (clp->cl_mach_cred)
3887                 goto out_free;
3888         __nfsd4_hash_conn(new, ses);
3889         spin_unlock(&clp->cl_lock);
3890         ret = nfsd4_register_conn(new);
3891         if (ret)
3892                 /* oops; xprt is already down: */
3893                 nfsd4_conn_lost(&new->cn_xpt_user);
3894         return nfs_ok;
3895 out_free:
3896         spin_unlock(&clp->cl_lock);
3897         free_conn(new);
3898         return status;
3899 }
3900
3901 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3902 {
3903         struct nfsd4_compoundargs *args = rqstp->rq_argp;
3904
3905         return args->opcnt > session->se_fchannel.maxops;
3906 }
3907
3908 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3909                                   struct nfsd4_session *session)
3910 {
3911         struct xdr_buf *xb = &rqstp->rq_arg;
3912
3913         return xb->len > session->se_fchannel.maxreq_sz;
3914 }
3915
3916 static bool replay_matches_cache(struct svc_rqst *rqstp,
3917                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3918 {
3919         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3920
3921         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3922             (bool)seq->cachethis)
3923                 return false;
3924         /*
3925          * If there's an error then the reply can have fewer ops than
3926          * the call.
3927          */
3928         if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
3929                 return false;
3930         /*
3931          * But if we cached a reply with *more* ops than the call you're
3932          * sending us now, then this new call is clearly not really a
3933          * replay of the old one:
3934          */
3935         if (slot->sl_opcnt > argp->opcnt)
3936                 return false;
3937         /* This is the only check explicitly called by spec: */
3938         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3939                 return false;
3940         /*
3941          * There may be more comparisons we could actually do, but the
3942          * spec doesn't require us to catch every case where the calls
3943          * don't match (that would require caching the call as well as
3944          * the reply), so we don't bother.
3945          */
3946         return true;
3947 }
3948
3949 __be32
3950 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3951                 union nfsd4_op_u *u)
3952 {
3953         struct nfsd4_sequence *seq = &u->sequence;
3954         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3955         struct xdr_stream *xdr = resp->xdr;
3956         struct nfsd4_session *session;
3957         struct nfs4_client *clp;
3958         struct nfsd4_slot *slot;
3959         struct nfsd4_conn *conn;
3960         __be32 status;
3961         int buflen;
3962         struct net *net = SVC_NET(rqstp);
3963         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3964
3965         if (resp->opcnt != 1)
3966                 return nfserr_sequence_pos;
3967
3968         /*
3969          * Will be either used or freed by nfsd4_sequence_check_conn
3970          * below.
3971          */
3972         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3973         if (!conn)
3974                 return nfserr_jukebox;
3975
3976         spin_lock(&nn->client_lock);
3977         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3978         if (!session)
3979                 goto out_no_session;
3980         clp = session->se_client;
3981
3982         status = nfserr_too_many_ops;
3983         if (nfsd4_session_too_many_ops(rqstp, session))
3984                 goto out_put_session;
3985
3986         status = nfserr_req_too_big;
3987         if (nfsd4_request_too_big(rqstp, session))
3988                 goto out_put_session;
3989
3990         status = nfserr_badslot;
3991         if (seq->slotid >= session->se_fchannel.maxreqs)
3992                 goto out_put_session;
3993
3994         slot = session->se_slots[seq->slotid];
3995         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3996
3997         /* We do not negotiate the number of slots yet, so set the
3998          * maxslots to the session maxreqs which is used to encode
3999          * sr_highest_slotid and the sr_target_slot id to maxslots */
4000         seq->maxslots = session->se_fchannel.maxreqs;
4001
4002         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
4003                                         slot->sl_flags & NFSD4_SLOT_INUSE);
4004         if (status == nfserr_replay_cache) {
4005                 status = nfserr_seq_misordered;
4006                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
4007                         goto out_put_session;
4008                 status = nfserr_seq_false_retry;
4009                 if (!replay_matches_cache(rqstp, seq, slot))
4010                         goto out_put_session;
4011                 cstate->slot = slot;
4012                 cstate->session = session;
4013                 cstate->clp = clp;
4014                 /* Return the cached reply status and set cstate->status
4015                  * for nfsd4_proc_compound processing */
4016                 status = nfsd4_replay_cache_entry(resp, seq);
4017                 cstate->status = nfserr_replay_cache;
4018                 goto out;
4019         }
4020         if (status)
4021                 goto out_put_session;
4022
4023         status = nfsd4_sequence_check_conn(conn, session);
4024         conn = NULL;
4025         if (status)
4026                 goto out_put_session;
4027
4028         buflen = (seq->cachethis) ?
4029                         session->se_fchannel.maxresp_cached :
4030                         session->se_fchannel.maxresp_sz;
4031         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
4032                                     nfserr_rep_too_big;
4033         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
4034                 goto out_put_session;
4035         svc_reserve(rqstp, buflen);
4036
4037         status = nfs_ok;
4038         /* Success! bump slot seqid */
4039         slot->sl_seqid = seq->seqid;
4040         slot->sl_flags |= NFSD4_SLOT_INUSE;
4041         if (seq->cachethis)
4042                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
4043         else
4044                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
4045
4046         cstate->slot = slot;
4047         cstate->session = session;
4048         cstate->clp = clp;
4049
4050 out:
4051         switch (clp->cl_cb_state) {
4052         case NFSD4_CB_DOWN:
4053                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
4054                 break;
4055         case NFSD4_CB_FAULT:
4056                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
4057                 break;
4058         default:
4059                 seq->status_flags = 0;
4060         }
4061         if (!list_empty(&clp->cl_revoked))
4062                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
4063 out_no_session:
4064         if (conn)
4065                 free_conn(conn);
4066         spin_unlock(&nn->client_lock);
4067         return status;
4068 out_put_session:
4069         nfsd4_put_session_locked(session);
4070         goto out_no_session;
4071 }
4072
4073 void
4074 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
4075 {
4076         struct nfsd4_compound_state *cs = &resp->cstate;
4077
4078         if (nfsd4_has_session(cs)) {
4079                 if (cs->status != nfserr_replay_cache) {
4080                         nfsd4_store_cache_entry(resp);
4081                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
4082                 }
4083                 /* Drop session reference that was taken in nfsd4_sequence() */
4084                 nfsd4_put_session(cs->session);
4085         } else if (cs->clp)
4086                 put_client_renew(cs->clp);
4087 }
4088
4089 __be32
4090 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
4091                 struct nfsd4_compound_state *cstate,
4092                 union nfsd4_op_u *u)
4093 {
4094         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
4095         struct nfs4_client *conf, *unconf;
4096         struct nfs4_client *clp = NULL;
4097         __be32 status = 0;
4098         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4099
4100         spin_lock(&nn->client_lock);
4101         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
4102         conf = find_confirmed_client(&dc->clientid, true, nn);
4103         WARN_ON_ONCE(conf && unconf);
4104
4105         if (conf) {
4106                 if (client_has_state(conf)) {
4107                         status = nfserr_clientid_busy;
4108                         goto out;
4109                 }
4110                 status = mark_client_expired_locked(conf);
4111                 if (status)
4112                         goto out;
4113                 clp = conf;
4114         } else if (unconf)
4115                 clp = unconf;
4116         else {
4117                 status = nfserr_stale_clientid;
4118                 goto out;
4119         }
4120         if (!nfsd4_mach_creds_match(clp, rqstp)) {
4121                 clp = NULL;
4122                 status = nfserr_wrong_cred;
4123                 goto out;
4124         }
4125         trace_nfsd_clid_destroyed(&clp->cl_clientid);
4126         unhash_client_locked(clp);
4127 out:
4128         spin_unlock(&nn->client_lock);
4129         if (clp)
4130                 expire_client(clp);
4131         return status;
4132 }
4133
4134 __be32
4135 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
4136                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
4137 {
4138         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
4139         struct nfs4_client *clp = cstate->clp;
4140         __be32 status = 0;
4141
4142         if (rc->rca_one_fs) {
4143                 if (!cstate->current_fh.fh_dentry)
4144                         return nfserr_nofilehandle;
4145                 /*
4146                  * We don't take advantage of the rca_one_fs case.
4147                  * That's OK, it's optional, we can safely ignore it.
4148                  */
4149                 return nfs_ok;
4150         }
4151
4152         status = nfserr_complete_already;
4153         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
4154                 goto out;
4155
4156         status = nfserr_stale_clientid;
4157         if (is_client_expired(clp))
4158                 /*
4159                  * The following error isn't really legal.
4160                  * But we only get here if the client just explicitly
4161                  * destroyed the client.  Surely it no longer cares what
4162                  * error it gets back on an operation for the dead
4163                  * client.
4164                  */
4165                 goto out;
4166
4167         status = nfs_ok;
4168         trace_nfsd_clid_reclaim_complete(&clp->cl_clientid);
4169         nfsd4_client_record_create(clp);
4170         inc_reclaim_complete(clp);
4171 out:
4172         return status;
4173 }
4174
4175 __be32
4176 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4177                   union nfsd4_op_u *u)
4178 {
4179         struct nfsd4_setclientid *setclid = &u->setclientid;
4180         struct xdr_netobj       clname = setclid->se_name;
4181         nfs4_verifier           clverifier = setclid->se_verf;
4182         struct nfs4_client      *conf, *new;
4183         struct nfs4_client      *unconf = NULL;
4184         __be32                  status;
4185         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4186
4187         new = create_client(clname, rqstp, &clverifier);
4188         if (new == NULL)
4189                 return nfserr_jukebox;
4190         spin_lock(&nn->client_lock);
4191         conf = find_confirmed_client_by_name(&clname, nn);
4192         if (conf && client_has_state(conf)) {
4193                 status = nfserr_clid_inuse;
4194                 if (clp_used_exchangeid(conf))
4195                         goto out;
4196                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4197                         trace_nfsd_clid_cred_mismatch(conf, rqstp);
4198                         goto out;
4199                 }
4200         }
4201         unconf = find_unconfirmed_client_by_name(&clname, nn);
4202         if (unconf)
4203                 unhash_client_locked(unconf);
4204         if (conf) {
4205                 if (same_verf(&conf->cl_verifier, &clverifier)) {
4206                         copy_clid(new, conf);
4207                         gen_confirm(new, nn);
4208                 } else
4209                         trace_nfsd_clid_verf_mismatch(conf, rqstp,
4210                                                       &clverifier);
4211         } else
4212                 trace_nfsd_clid_fresh(new);
4213         new->cl_minorversion = 0;
4214         gen_callback(new, setclid, rqstp);
4215         add_to_unconfirmed(new);
4216         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
4217         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
4218         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
4219         new = NULL;
4220         status = nfs_ok;
4221 out:
4222         spin_unlock(&nn->client_lock);
4223         if (new)
4224                 free_client(new);
4225         if (unconf) {
4226                 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
4227                 expire_client(unconf);
4228         }
4229         return status;
4230 }
4231
4232 __be32
4233 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
4234                         struct nfsd4_compound_state *cstate,
4235                         union nfsd4_op_u *u)
4236 {
4237         struct nfsd4_setclientid_confirm *setclientid_confirm =
4238                         &u->setclientid_confirm;
4239         struct nfs4_client *conf, *unconf;
4240         struct nfs4_client *old = NULL;
4241         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
4242         clientid_t * clid = &setclientid_confirm->sc_clientid;
4243         __be32 status;
4244         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4245
4246         if (STALE_CLIENTID(clid, nn))
4247                 return nfserr_stale_clientid;
4248
4249         spin_lock(&nn->client_lock);
4250         conf = find_confirmed_client(clid, false, nn);
4251         unconf = find_unconfirmed_client(clid, false, nn);
4252         /*
4253          * We try hard to give out unique clientid's, so if we get an
4254          * attempt to confirm the same clientid with a different cred,
4255          * the client may be buggy; this should never happen.
4256          *
4257          * Nevertheless, RFC 7530 recommends INUSE for this case:
4258          */
4259         status = nfserr_clid_inuse;
4260         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
4261                 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
4262                 goto out;
4263         }
4264         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4265                 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4266                 goto out;
4267         }
4268         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
4269                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
4270                         status = nfs_ok;
4271                 } else
4272                         status = nfserr_stale_clientid;
4273                 goto out;
4274         }
4275         status = nfs_ok;
4276         if (conf) {
4277                 old = unconf;
4278                 unhash_client_locked(old);
4279                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
4280         } else {
4281                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
4282                 if (old) {
4283                         status = nfserr_clid_inuse;
4284                         if (client_has_state(old)
4285                                         && !same_creds(&unconf->cl_cred,
4286                                                         &old->cl_cred)) {
4287                                 old = NULL;
4288                                 goto out;
4289                         }
4290                         status = mark_client_expired_locked(old);
4291                         if (status) {
4292                                 old = NULL;
4293                                 goto out;
4294                         }
4295                         trace_nfsd_clid_replaced(&old->cl_clientid);
4296                 }
4297                 move_to_confirmed(unconf);
4298                 conf = unconf;
4299         }
4300         get_client_locked(conf);
4301         spin_unlock(&nn->client_lock);
4302         if (conf == unconf)
4303                 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
4304         nfsd4_probe_callback(conf);
4305         spin_lock(&nn->client_lock);
4306         put_client_renew_locked(conf);
4307 out:
4308         spin_unlock(&nn->client_lock);
4309         if (old)
4310                 expire_client(old);
4311         return status;
4312 }
4313
4314 static struct nfs4_file *nfsd4_alloc_file(void)
4315 {
4316         return kmem_cache_alloc(file_slab, GFP_KERNEL);
4317 }
4318
4319 /* OPEN Share state helper functions */
4320
4321 static void nfsd4_file_init(const struct svc_fh *fh, struct nfs4_file *fp)
4322 {
4323         refcount_set(&fp->fi_ref, 1);
4324         spin_lock_init(&fp->fi_lock);
4325         INIT_LIST_HEAD(&fp->fi_stateids);
4326         INIT_LIST_HEAD(&fp->fi_delegations);
4327         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4328         fh_copy_shallow(&fp->fi_fhandle, &fh->fh_handle);
4329         fp->fi_deleg_file = NULL;
4330         fp->fi_had_conflict = false;
4331         fp->fi_share_deny = 0;
4332         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4333         memset(fp->fi_access, 0, sizeof(fp->fi_access));
4334         fp->fi_aliased = false;
4335         fp->fi_inode = d_inode(fh->fh_dentry);
4336 #ifdef CONFIG_NFSD_PNFS
4337         INIT_LIST_HEAD(&fp->fi_lo_states);
4338         atomic_set(&fp->fi_lo_recalls, 0);
4339 #endif
4340 }
4341
4342 void
4343 nfsd4_free_slabs(void)
4344 {
4345         kmem_cache_destroy(client_slab);
4346         kmem_cache_destroy(openowner_slab);
4347         kmem_cache_destroy(lockowner_slab);
4348         kmem_cache_destroy(file_slab);
4349         kmem_cache_destroy(stateid_slab);
4350         kmem_cache_destroy(deleg_slab);
4351         kmem_cache_destroy(odstate_slab);
4352 }
4353
4354 int
4355 nfsd4_init_slabs(void)
4356 {
4357         client_slab = kmem_cache_create("nfsd4_clients",
4358                         sizeof(struct nfs4_client), 0, 0, NULL);
4359         if (client_slab == NULL)
4360                 goto out;
4361         openowner_slab = kmem_cache_create("nfsd4_openowners",
4362                         sizeof(struct nfs4_openowner), 0, 0, NULL);
4363         if (openowner_slab == NULL)
4364                 goto out_free_client_slab;
4365         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
4366                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
4367         if (lockowner_slab == NULL)
4368                 goto out_free_openowner_slab;
4369         file_slab = kmem_cache_create("nfsd4_files",
4370                         sizeof(struct nfs4_file), 0, 0, NULL);
4371         if (file_slab == NULL)
4372                 goto out_free_lockowner_slab;
4373         stateid_slab = kmem_cache_create("nfsd4_stateids",
4374                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
4375         if (stateid_slab == NULL)
4376                 goto out_free_file_slab;
4377         deleg_slab = kmem_cache_create("nfsd4_delegations",
4378                         sizeof(struct nfs4_delegation), 0, 0, NULL);
4379         if (deleg_slab == NULL)
4380                 goto out_free_stateid_slab;
4381         odstate_slab = kmem_cache_create("nfsd4_odstate",
4382                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
4383         if (odstate_slab == NULL)
4384                 goto out_free_deleg_slab;
4385         return 0;
4386
4387 out_free_deleg_slab:
4388         kmem_cache_destroy(deleg_slab);
4389 out_free_stateid_slab:
4390         kmem_cache_destroy(stateid_slab);
4391 out_free_file_slab:
4392         kmem_cache_destroy(file_slab);
4393 out_free_lockowner_slab:
4394         kmem_cache_destroy(lockowner_slab);
4395 out_free_openowner_slab:
4396         kmem_cache_destroy(openowner_slab);
4397 out_free_client_slab:
4398         kmem_cache_destroy(client_slab);
4399 out:
4400         return -ENOMEM;
4401 }
4402
4403 static unsigned long
4404 nfsd4_state_shrinker_count(struct shrinker *shrink, struct shrink_control *sc)
4405 {
4406         int count;
4407         struct nfsd_net *nn = container_of(shrink,
4408                         struct nfsd_net, nfsd_client_shrinker);
4409
4410         count = atomic_read(&nn->nfsd_courtesy_clients);
4411         if (!count)
4412                 count = atomic_long_read(&num_delegations);
4413         if (count)
4414                 mod_delayed_work(laundry_wq, &nn->nfsd_shrinker_work, 0);
4415         return (unsigned long)count;
4416 }
4417
4418 static unsigned long
4419 nfsd4_state_shrinker_scan(struct shrinker *shrink, struct shrink_control *sc)
4420 {
4421         return SHRINK_STOP;
4422 }
4423
4424 int
4425 nfsd4_init_leases_net(struct nfsd_net *nn)
4426 {
4427         struct sysinfo si;
4428         u64 max_clients;
4429
4430         nn->nfsd4_lease = 90;   /* default lease time */
4431         nn->nfsd4_grace = 90;
4432         nn->somebody_reclaimed = false;
4433         nn->track_reclaim_completes = false;
4434         nn->clverifier_counter = get_random_u32();
4435         nn->clientid_base = get_random_u32();
4436         nn->clientid_counter = nn->clientid_base + 1;
4437         nn->s2s_cp_cl_id = nn->clientid_counter++;
4438
4439         atomic_set(&nn->nfs4_client_count, 0);
4440         si_meminfo(&si);
4441         max_clients = (u64)si.totalram * si.mem_unit / (1024 * 1024 * 1024);
4442         max_clients *= NFS4_CLIENTS_PER_GB;
4443         nn->nfs4_max_clients = max_t(int, max_clients, NFS4_CLIENTS_PER_GB);
4444
4445         atomic_set(&nn->nfsd_courtesy_clients, 0);
4446         nn->nfsd_client_shrinker.scan_objects = nfsd4_state_shrinker_scan;
4447         nn->nfsd_client_shrinker.count_objects = nfsd4_state_shrinker_count;
4448         nn->nfsd_client_shrinker.seeks = DEFAULT_SEEKS;
4449         return register_shrinker(&nn->nfsd_client_shrinker, "nfsd-client");
4450 }
4451
4452 void
4453 nfsd4_leases_net_shutdown(struct nfsd_net *nn)
4454 {
4455         unregister_shrinker(&nn->nfsd_client_shrinker);
4456 }
4457
4458 static void init_nfs4_replay(struct nfs4_replay *rp)
4459 {
4460         rp->rp_status = nfserr_serverfault;
4461         rp->rp_buflen = 0;
4462         rp->rp_buf = rp->rp_ibuf;
4463         mutex_init(&rp->rp_mutex);
4464 }
4465
4466 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4467                 struct nfs4_stateowner *so)
4468 {
4469         if (!nfsd4_has_session(cstate)) {
4470                 mutex_lock(&so->so_replay.rp_mutex);
4471                 cstate->replay_owner = nfs4_get_stateowner(so);
4472         }
4473 }
4474
4475 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4476 {
4477         struct nfs4_stateowner *so = cstate->replay_owner;
4478
4479         if (so != NULL) {
4480                 cstate->replay_owner = NULL;
4481                 mutex_unlock(&so->so_replay.rp_mutex);
4482                 nfs4_put_stateowner(so);
4483         }
4484 }
4485
4486 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4487 {
4488         struct nfs4_stateowner *sop;
4489
4490         sop = kmem_cache_alloc(slab, GFP_KERNEL);
4491         if (!sop)
4492                 return NULL;
4493
4494         xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4495         if (!sop->so_owner.data) {
4496                 kmem_cache_free(slab, sop);
4497                 return NULL;
4498         }
4499
4500         INIT_LIST_HEAD(&sop->so_stateids);
4501         sop->so_client = clp;
4502         init_nfs4_replay(&sop->so_replay);
4503         atomic_set(&sop->so_count, 1);
4504         return sop;
4505 }
4506
4507 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4508 {
4509         lockdep_assert_held(&clp->cl_lock);
4510
4511         list_add(&oo->oo_owner.so_strhash,
4512                  &clp->cl_ownerstr_hashtbl[strhashval]);
4513         list_add(&oo->oo_perclient, &clp->cl_openowners);
4514 }
4515
4516 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4517 {
4518         unhash_openowner_locked(openowner(so));
4519 }
4520
4521 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4522 {
4523         struct nfs4_openowner *oo = openowner(so);
4524
4525         kmem_cache_free(openowner_slab, oo);
4526 }
4527
4528 static const struct nfs4_stateowner_operations openowner_ops = {
4529         .so_unhash =    nfs4_unhash_openowner,
4530         .so_free =      nfs4_free_openowner,
4531 };
4532
4533 static struct nfs4_ol_stateid *
4534 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4535 {
4536         struct nfs4_ol_stateid *local, *ret = NULL;
4537         struct nfs4_openowner *oo = open->op_openowner;
4538
4539         lockdep_assert_held(&fp->fi_lock);
4540
4541         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4542                 /* ignore lock owners */
4543                 if (local->st_stateowner->so_is_open_owner == 0)
4544                         continue;
4545                 if (local->st_stateowner != &oo->oo_owner)
4546                         continue;
4547                 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
4548                         ret = local;
4549                         refcount_inc(&ret->st_stid.sc_count);
4550                         break;
4551                 }
4552         }
4553         return ret;
4554 }
4555
4556 static __be32
4557 nfsd4_verify_open_stid(struct nfs4_stid *s)
4558 {
4559         __be32 ret = nfs_ok;
4560
4561         switch (s->sc_type) {
4562         default:
4563                 break;
4564         case 0:
4565         case NFS4_CLOSED_STID:
4566         case NFS4_CLOSED_DELEG_STID:
4567                 ret = nfserr_bad_stateid;
4568                 break;
4569         case NFS4_REVOKED_DELEG_STID:
4570                 ret = nfserr_deleg_revoked;
4571         }
4572         return ret;
4573 }
4574
4575 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4576 static __be32
4577 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4578 {
4579         __be32 ret;
4580
4581         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4582         ret = nfsd4_verify_open_stid(&stp->st_stid);
4583         if (ret != nfs_ok)
4584                 mutex_unlock(&stp->st_mutex);
4585         return ret;
4586 }
4587
4588 static struct nfs4_ol_stateid *
4589 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4590 {
4591         struct nfs4_ol_stateid *stp;
4592         for (;;) {
4593                 spin_lock(&fp->fi_lock);
4594                 stp = nfsd4_find_existing_open(fp, open);
4595                 spin_unlock(&fp->fi_lock);
4596                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4597                         break;
4598                 nfs4_put_stid(&stp->st_stid);
4599         }
4600         return stp;
4601 }
4602
4603 static struct nfs4_openowner *
4604 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4605                            struct nfsd4_compound_state *cstate)
4606 {
4607         struct nfs4_client *clp = cstate->clp;
4608         struct nfs4_openowner *oo, *ret;
4609
4610         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4611         if (!oo)
4612                 return NULL;
4613         oo->oo_owner.so_ops = &openowner_ops;
4614         oo->oo_owner.so_is_open_owner = 1;
4615         oo->oo_owner.so_seqid = open->op_seqid;
4616         oo->oo_flags = 0;
4617         if (nfsd4_has_session(cstate))
4618                 oo->oo_flags |= NFS4_OO_CONFIRMED;
4619         oo->oo_time = 0;
4620         oo->oo_last_closed_stid = NULL;
4621         INIT_LIST_HEAD(&oo->oo_close_lru);
4622         spin_lock(&clp->cl_lock);
4623         ret = find_openstateowner_str_locked(strhashval, open, clp);
4624         if (ret == NULL) {
4625                 hash_openowner(oo, clp, strhashval);
4626                 ret = oo;
4627         } else
4628                 nfs4_free_stateowner(&oo->oo_owner);
4629
4630         spin_unlock(&clp->cl_lock);
4631         return ret;
4632 }
4633
4634 static struct nfs4_ol_stateid *
4635 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4636 {
4637
4638         struct nfs4_openowner *oo = open->op_openowner;
4639         struct nfs4_ol_stateid *retstp = NULL;
4640         struct nfs4_ol_stateid *stp;
4641
4642         stp = open->op_stp;
4643         /* We are moving these outside of the spinlocks to avoid the warnings */
4644         mutex_init(&stp->st_mutex);
4645         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4646
4647 retry:
4648         spin_lock(&oo->oo_owner.so_client->cl_lock);
4649         spin_lock(&fp->fi_lock);
4650
4651         retstp = nfsd4_find_existing_open(fp, open);
4652         if (retstp)
4653                 goto out_unlock;
4654
4655         open->op_stp = NULL;
4656         refcount_inc(&stp->st_stid.sc_count);
4657         stp->st_stid.sc_type = NFS4_OPEN_STID;
4658         INIT_LIST_HEAD(&stp->st_locks);
4659         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4660         get_nfs4_file(fp);
4661         stp->st_stid.sc_file = fp;
4662         stp->st_access_bmap = 0;
4663         stp->st_deny_bmap = 0;
4664         stp->st_openstp = NULL;
4665         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4666         list_add(&stp->st_perfile, &fp->fi_stateids);
4667
4668 out_unlock:
4669         spin_unlock(&fp->fi_lock);
4670         spin_unlock(&oo->oo_owner.so_client->cl_lock);
4671         if (retstp) {
4672                 /* Handle races with CLOSE */
4673                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4674                         nfs4_put_stid(&retstp->st_stid);
4675                         goto retry;
4676                 }
4677                 /* To keep mutex tracking happy */
4678                 mutex_unlock(&stp->st_mutex);
4679                 stp = retstp;
4680         }
4681         return stp;
4682 }
4683
4684 /*
4685  * In the 4.0 case we need to keep the owners around a little while to handle
4686  * CLOSE replay. We still do need to release any file access that is held by
4687  * them before returning however.
4688  */
4689 static void
4690 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4691 {
4692         struct nfs4_ol_stateid *last;
4693         struct nfs4_openowner *oo = openowner(s->st_stateowner);
4694         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4695                                                 nfsd_net_id);
4696
4697         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4698
4699         /*
4700          * We know that we hold one reference via nfsd4_close, and another
4701          * "persistent" reference for the client. If the refcount is higher
4702          * than 2, then there are still calls in progress that are using this
4703          * stateid. We can't put the sc_file reference until they are finished.
4704          * Wait for the refcount to drop to 2. Since it has been unhashed,
4705          * there should be no danger of the refcount going back up again at
4706          * this point.
4707          */
4708         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4709
4710         release_all_access(s);
4711         if (s->st_stid.sc_file) {
4712                 put_nfs4_file(s->st_stid.sc_file);
4713                 s->st_stid.sc_file = NULL;
4714         }
4715
4716         spin_lock(&nn->client_lock);
4717         last = oo->oo_last_closed_stid;
4718         oo->oo_last_closed_stid = s;
4719         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4720         oo->oo_time = ktime_get_boottime_seconds();
4721         spin_unlock(&nn->client_lock);
4722         if (last)
4723                 nfs4_put_stid(&last->st_stid);
4724 }
4725
4726 static noinline_for_stack struct nfs4_file *
4727 nfsd4_file_hash_lookup(const struct svc_fh *fhp)
4728 {
4729         struct inode *inode = d_inode(fhp->fh_dentry);
4730         struct rhlist_head *tmp, *list;
4731         struct nfs4_file *fi;
4732
4733         rcu_read_lock();
4734         list = rhltable_lookup(&nfs4_file_rhltable, &inode,
4735                                nfs4_file_rhash_params);
4736         rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
4737                 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
4738                         if (refcount_inc_not_zero(&fi->fi_ref)) {
4739                                 rcu_read_unlock();
4740                                 return fi;
4741                         }
4742                 }
4743         }
4744         rcu_read_unlock();
4745         return NULL;
4746 }
4747
4748 /*
4749  * On hash insertion, identify entries with the same inode but
4750  * distinct filehandles. They will all be on the list returned
4751  * by rhltable_lookup().
4752  *
4753  * inode->i_lock prevents racing insertions from adding an entry
4754  * for the same inode/fhp pair twice.
4755  */
4756 static noinline_for_stack struct nfs4_file *
4757 nfsd4_file_hash_insert(struct nfs4_file *new, const struct svc_fh *fhp)
4758 {
4759         struct inode *inode = d_inode(fhp->fh_dentry);
4760         struct rhlist_head *tmp, *list;
4761         struct nfs4_file *ret = NULL;
4762         bool alias_found = false;
4763         struct nfs4_file *fi;
4764         int err;
4765
4766         rcu_read_lock();
4767         spin_lock(&inode->i_lock);
4768
4769         list = rhltable_lookup(&nfs4_file_rhltable, &inode,
4770                                nfs4_file_rhash_params);
4771         rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
4772                 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
4773                         if (refcount_inc_not_zero(&fi->fi_ref))
4774                                 ret = fi;
4775                 } else
4776                         fi->fi_aliased = alias_found = true;
4777         }
4778         if (ret)
4779                 goto out_unlock;
4780
4781         nfsd4_file_init(fhp, new);
4782         err = rhltable_insert(&nfs4_file_rhltable, &new->fi_rlist,
4783                               nfs4_file_rhash_params);
4784         if (err)
4785                 goto out_unlock;
4786
4787         new->fi_aliased = alias_found;
4788         ret = new;
4789
4790 out_unlock:
4791         spin_unlock(&inode->i_lock);
4792         rcu_read_unlock();
4793         return ret;
4794 }
4795
4796 static noinline_for_stack void nfsd4_file_hash_remove(struct nfs4_file *fi)
4797 {
4798         rhltable_remove(&nfs4_file_rhltable, &fi->fi_rlist,
4799                         nfs4_file_rhash_params);
4800 }
4801
4802 /*
4803  * Called to check deny when READ with all zero stateid or
4804  * WRITE with all zero or all one stateid
4805  */
4806 static __be32
4807 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
4808 {
4809         struct nfs4_file *fp;
4810         __be32 ret = nfs_ok;
4811
4812         fp = nfsd4_file_hash_lookup(current_fh);
4813         if (!fp)
4814                 return ret;
4815
4816         /* Check for conflicting share reservations */
4817         spin_lock(&fp->fi_lock);
4818         if (fp->fi_share_deny & deny_type)
4819                 ret = nfserr_locked;
4820         spin_unlock(&fp->fi_lock);
4821         put_nfs4_file(fp);
4822         return ret;
4823 }
4824
4825 static bool nfsd4_deleg_present(const struct inode *inode)
4826 {
4827         struct file_lock_context *ctx = locks_inode_context(inode);
4828
4829         return ctx && !list_empty_careful(&ctx->flc_lease);
4830 }
4831
4832 /**
4833  * nfsd_wait_for_delegreturn - wait for delegations to be returned
4834  * @rqstp: the RPC transaction being executed
4835  * @inode: in-core inode of the file being waited for
4836  *
4837  * The timeout prevents deadlock if all nfsd threads happen to be
4838  * tied up waiting for returning delegations.
4839  *
4840  * Return values:
4841  *   %true: delegation was returned
4842  *   %false: timed out waiting for delegreturn
4843  */
4844 bool nfsd_wait_for_delegreturn(struct svc_rqst *rqstp, struct inode *inode)
4845 {
4846         long __maybe_unused timeo;
4847
4848         timeo = wait_var_event_timeout(inode, !nfsd4_deleg_present(inode),
4849                                        NFSD_DELEGRETURN_TIMEOUT);
4850         trace_nfsd_delegret_wakeup(rqstp, inode, timeo);
4851         return timeo > 0;
4852 }
4853
4854 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
4855 {
4856         struct nfs4_delegation *dp = cb_to_delegation(cb);
4857         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
4858                                           nfsd_net_id);
4859
4860         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
4861
4862         /*
4863          * We can't do this in nfsd_break_deleg_cb because it is
4864          * already holding inode->i_lock.
4865          *
4866          * If the dl_time != 0, then we know that it has already been
4867          * queued for a lease break. Don't queue it again.
4868          */
4869         spin_lock(&state_lock);
4870         if (delegation_hashed(dp) && dp->dl_time == 0) {
4871                 dp->dl_time = ktime_get_boottime_seconds();
4872                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
4873         }
4874         spin_unlock(&state_lock);
4875 }
4876
4877 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
4878                 struct rpc_task *task)
4879 {
4880         struct nfs4_delegation *dp = cb_to_delegation(cb);
4881
4882         trace_nfsd_cb_recall_done(&dp->dl_stid.sc_stateid, task);
4883
4884         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID ||
4885             dp->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4886                 return 1;
4887
4888         switch (task->tk_status) {
4889         case 0:
4890                 return 1;
4891         case -NFS4ERR_DELAY:
4892                 rpc_delay(task, 2 * HZ);
4893                 return 0;
4894         case -EBADHANDLE:
4895         case -NFS4ERR_BAD_STATEID:
4896                 /*
4897                  * Race: client probably got cb_recall before open reply
4898                  * granting delegation.
4899                  */
4900                 if (dp->dl_retries--) {
4901                         rpc_delay(task, 2 * HZ);
4902                         return 0;
4903                 }
4904                 fallthrough;
4905         default:
4906                 return 1;
4907         }
4908 }
4909
4910 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
4911 {
4912         struct nfs4_delegation *dp = cb_to_delegation(cb);
4913
4914         nfs4_put_stid(&dp->dl_stid);
4915 }
4916
4917 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
4918         .prepare        = nfsd4_cb_recall_prepare,
4919         .done           = nfsd4_cb_recall_done,
4920         .release        = nfsd4_cb_recall_release,
4921 };
4922
4923 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
4924 {
4925         /*
4926          * We're assuming the state code never drops its reference
4927          * without first removing the lease.  Since we're in this lease
4928          * callback (and since the lease code is serialized by the
4929          * flc_lock) we know the server hasn't removed the lease yet, and
4930          * we know it's safe to take a reference.
4931          */
4932         refcount_inc(&dp->dl_stid.sc_count);
4933         WARN_ON_ONCE(!nfsd4_run_cb(&dp->dl_recall));
4934 }
4935
4936 /* Called from break_lease() with flc_lock held. */
4937 static bool
4938 nfsd_break_deleg_cb(struct file_lock *fl)
4939 {
4940         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4941         struct nfs4_file *fp = dp->dl_stid.sc_file;
4942         struct nfs4_client *clp = dp->dl_stid.sc_client;
4943         struct nfsd_net *nn;
4944
4945         trace_nfsd_cb_recall(&dp->dl_stid);
4946
4947         dp->dl_recalled = true;
4948         atomic_inc(&clp->cl_delegs_in_recall);
4949         if (try_to_expire_client(clp)) {
4950                 nn = net_generic(clp->net, nfsd_net_id);
4951                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
4952         }
4953
4954         /*
4955          * We don't want the locks code to timeout the lease for us;
4956          * we'll remove it ourself if a delegation isn't returned
4957          * in time:
4958          */
4959         fl->fl_break_time = 0;
4960
4961         spin_lock(&fp->fi_lock);
4962         fp->fi_had_conflict = true;
4963         nfsd_break_one_deleg(dp);
4964         spin_unlock(&fp->fi_lock);
4965         return false;
4966 }
4967
4968 /**
4969  * nfsd_breaker_owns_lease - Check if lease conflict was resolved
4970  * @fl: Lock state to check
4971  *
4972  * Return values:
4973  *   %true: Lease conflict was resolved
4974  *   %false: Lease conflict was not resolved.
4975  */
4976 static bool nfsd_breaker_owns_lease(struct file_lock *fl)
4977 {
4978         struct nfs4_delegation *dl = fl->fl_owner;
4979         struct svc_rqst *rqst;
4980         struct nfs4_client *clp;
4981
4982         if (!i_am_nfsd())
4983                 return false;
4984         rqst = kthread_data(current);
4985         /* Note rq_prog == NFS_ACL_PROGRAM is also possible: */
4986         if (rqst->rq_prog != NFS_PROGRAM || rqst->rq_vers < 4)
4987                 return false;
4988         clp = *(rqst->rq_lease_breaker);
4989         return dl->dl_stid.sc_client == clp;
4990 }
4991
4992 static int
4993 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4994                      struct list_head *dispose)
4995 {
4996         struct nfs4_delegation *dp = (struct nfs4_delegation *)onlist->fl_owner;
4997         struct nfs4_client *clp = dp->dl_stid.sc_client;
4998
4999         if (arg & F_UNLCK) {
5000                 if (dp->dl_recalled)
5001                         atomic_dec(&clp->cl_delegs_in_recall);
5002                 return lease_modify(onlist, arg, dispose);
5003         } else
5004                 return -EAGAIN;
5005 }
5006
5007 static const struct lock_manager_operations nfsd_lease_mng_ops = {
5008         .lm_breaker_owns_lease = nfsd_breaker_owns_lease,
5009         .lm_break = nfsd_break_deleg_cb,
5010         .lm_change = nfsd_change_deleg_cb,
5011 };
5012
5013 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
5014 {
5015         if (nfsd4_has_session(cstate))
5016                 return nfs_ok;
5017         if (seqid == so->so_seqid - 1)
5018                 return nfserr_replay_me;
5019         if (seqid == so->so_seqid)
5020                 return nfs_ok;
5021         return nfserr_bad_seqid;
5022 }
5023
5024 static struct nfs4_client *lookup_clientid(clientid_t *clid, bool sessions,
5025                                                 struct nfsd_net *nn)
5026 {
5027         struct nfs4_client *found;
5028
5029         spin_lock(&nn->client_lock);
5030         found = find_confirmed_client(clid, sessions, nn);
5031         if (found)
5032                 atomic_inc(&found->cl_rpc_users);
5033         spin_unlock(&nn->client_lock);
5034         return found;
5035 }
5036
5037 static __be32 set_client(clientid_t *clid,
5038                 struct nfsd4_compound_state *cstate,
5039                 struct nfsd_net *nn)
5040 {
5041         if (cstate->clp) {
5042                 if (!same_clid(&cstate->clp->cl_clientid, clid))
5043                         return nfserr_stale_clientid;
5044                 return nfs_ok;
5045         }
5046         if (STALE_CLIENTID(clid, nn))
5047                 return nfserr_stale_clientid;
5048         /*
5049          * We're in the 4.0 case (otherwise the SEQUENCE op would have
5050          * set cstate->clp), so session = false:
5051          */
5052         cstate->clp = lookup_clientid(clid, false, nn);
5053         if (!cstate->clp)
5054                 return nfserr_expired;
5055         return nfs_ok;
5056 }
5057
5058 __be32
5059 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
5060                     struct nfsd4_open *open, struct nfsd_net *nn)
5061 {
5062         clientid_t *clientid = &open->op_clientid;
5063         struct nfs4_client *clp = NULL;
5064         unsigned int strhashval;
5065         struct nfs4_openowner *oo = NULL;
5066         __be32 status;
5067
5068         /*
5069          * In case we need it later, after we've already created the
5070          * file and don't want to risk a further failure:
5071          */
5072         open->op_file = nfsd4_alloc_file();
5073         if (open->op_file == NULL)
5074                 return nfserr_jukebox;
5075
5076         status = set_client(clientid, cstate, nn);
5077         if (status)
5078                 return status;
5079         clp = cstate->clp;
5080
5081         strhashval = ownerstr_hashval(&open->op_owner);
5082         oo = find_openstateowner_str(strhashval, open, clp);
5083         open->op_openowner = oo;
5084         if (!oo) {
5085                 goto new_owner;
5086         }
5087         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5088                 /* Replace unconfirmed owners without checking for replay. */
5089                 release_openowner(oo);
5090                 open->op_openowner = NULL;
5091                 goto new_owner;
5092         }
5093         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
5094         if (status)
5095                 return status;
5096         goto alloc_stateid;
5097 new_owner:
5098         oo = alloc_init_open_stateowner(strhashval, open, cstate);
5099         if (oo == NULL)
5100                 return nfserr_jukebox;
5101         open->op_openowner = oo;
5102 alloc_stateid:
5103         open->op_stp = nfs4_alloc_open_stateid(clp);
5104         if (!open->op_stp)
5105                 return nfserr_jukebox;
5106
5107         if (nfsd4_has_session(cstate) &&
5108             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
5109                 open->op_odstate = alloc_clnt_odstate(clp);
5110                 if (!open->op_odstate)
5111                         return nfserr_jukebox;
5112         }
5113
5114         return nfs_ok;
5115 }
5116
5117 static inline __be32
5118 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
5119 {
5120         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
5121                 return nfserr_openmode;
5122         else
5123                 return nfs_ok;
5124 }
5125
5126 static int share_access_to_flags(u32 share_access)
5127 {
5128         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
5129 }
5130
5131 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
5132 {
5133         struct nfs4_stid *ret;
5134
5135         ret = find_stateid_by_type(cl, s,
5136                                 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
5137         if (!ret)
5138                 return NULL;
5139         return delegstateid(ret);
5140 }
5141
5142 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
5143 {
5144         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
5145                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
5146 }
5147
5148 static __be32
5149 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
5150                 struct nfs4_delegation **dp)
5151 {
5152         int flags;
5153         __be32 status = nfserr_bad_stateid;
5154         struct nfs4_delegation *deleg;
5155
5156         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
5157         if (deleg == NULL)
5158                 goto out;
5159         if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
5160                 nfs4_put_stid(&deleg->dl_stid);
5161                 if (cl->cl_minorversion)
5162                         status = nfserr_deleg_revoked;
5163                 goto out;
5164         }
5165         flags = share_access_to_flags(open->op_share_access);
5166         status = nfs4_check_delegmode(deleg, flags);
5167         if (status) {
5168                 nfs4_put_stid(&deleg->dl_stid);
5169                 goto out;
5170         }
5171         *dp = deleg;
5172 out:
5173         if (!nfsd4_is_deleg_cur(open))
5174                 return nfs_ok;
5175         if (status)
5176                 return status;
5177         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5178         return nfs_ok;
5179 }
5180
5181 static inline int nfs4_access_to_access(u32 nfs4_access)
5182 {
5183         int flags = 0;
5184
5185         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
5186                 flags |= NFSD_MAY_READ;
5187         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
5188                 flags |= NFSD_MAY_WRITE;
5189         return flags;
5190 }
5191
5192 static inline __be32
5193 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
5194                 struct nfsd4_open *open)
5195 {
5196         struct iattr iattr = {
5197                 .ia_valid = ATTR_SIZE,
5198                 .ia_size = 0,
5199         };
5200         struct nfsd_attrs attrs = {
5201                 .na_iattr       = &iattr,
5202         };
5203         if (!open->op_truncate)
5204                 return 0;
5205         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
5206                 return nfserr_inval;
5207         return nfsd_setattr(rqstp, fh, &attrs, 0, (time64_t)0);
5208 }
5209
5210 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
5211                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5212                 struct nfsd4_open *open, bool new_stp)
5213 {
5214         struct nfsd_file *nf = NULL;
5215         __be32 status;
5216         int oflag = nfs4_access_to_omode(open->op_share_access);
5217         int access = nfs4_access_to_access(open->op_share_access);
5218         unsigned char old_access_bmap, old_deny_bmap;
5219
5220         spin_lock(&fp->fi_lock);
5221
5222         /*
5223          * Are we trying to set a deny mode that would conflict with
5224          * current access?
5225          */
5226         status = nfs4_file_check_deny(fp, open->op_share_deny);
5227         if (status != nfs_ok) {
5228                 if (status != nfserr_share_denied) {
5229                         spin_unlock(&fp->fi_lock);
5230                         goto out;
5231                 }
5232                 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5233                                 stp, open->op_share_deny, false))
5234                         status = nfserr_jukebox;
5235                 spin_unlock(&fp->fi_lock);
5236                 goto out;
5237         }
5238
5239         /* set access to the file */
5240         status = nfs4_file_get_access(fp, open->op_share_access);
5241         if (status != nfs_ok) {
5242                 if (status != nfserr_share_denied) {
5243                         spin_unlock(&fp->fi_lock);
5244                         goto out;
5245                 }
5246                 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5247                                 stp, open->op_share_access, true))
5248                         status = nfserr_jukebox;
5249                 spin_unlock(&fp->fi_lock);
5250                 goto out;
5251         }
5252
5253         /* Set access bits in stateid */
5254         old_access_bmap = stp->st_access_bmap;
5255         set_access(open->op_share_access, stp);
5256
5257         /* Set new deny mask */
5258         old_deny_bmap = stp->st_deny_bmap;
5259         set_deny(open->op_share_deny, stp);
5260         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5261
5262         if (!fp->fi_fds[oflag]) {
5263                 spin_unlock(&fp->fi_lock);
5264
5265                 if (!open->op_filp) {
5266                         status = nfsd_file_acquire(rqstp, cur_fh, access, &nf);
5267                         if (status != nfs_ok)
5268                                 goto out_put_access;
5269                 } else {
5270                         status = nfsd_file_create(rqstp, cur_fh, access, &nf);
5271                         if (status != nfs_ok)
5272                                 goto out_put_access;
5273                         nf->nf_file = open->op_filp;
5274                         open->op_filp = NULL;
5275                         trace_nfsd_file_create(rqstp, access, nf);
5276                 }
5277
5278                 spin_lock(&fp->fi_lock);
5279                 if (!fp->fi_fds[oflag]) {
5280                         fp->fi_fds[oflag] = nf;
5281                         nf = NULL;
5282                 }
5283         }
5284         spin_unlock(&fp->fi_lock);
5285         if (nf)
5286                 nfsd_file_put(nf);
5287
5288         status = nfserrno(nfsd_open_break_lease(cur_fh->fh_dentry->d_inode,
5289                                                                 access));
5290         if (status)
5291                 goto out_put_access;
5292
5293         status = nfsd4_truncate(rqstp, cur_fh, open);
5294         if (status)
5295                 goto out_put_access;
5296 out:
5297         return status;
5298 out_put_access:
5299         stp->st_access_bmap = old_access_bmap;
5300         nfs4_file_put_access(fp, open->op_share_access);
5301         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
5302         goto out;
5303 }
5304
5305 static __be32
5306 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp,
5307                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5308                 struct nfsd4_open *open)
5309 {
5310         __be32 status;
5311         unsigned char old_deny_bmap = stp->st_deny_bmap;
5312
5313         if (!test_access(open->op_share_access, stp))
5314                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open, false);
5315
5316         /* test and set deny mode */
5317         spin_lock(&fp->fi_lock);
5318         status = nfs4_file_check_deny(fp, open->op_share_deny);
5319         if (status == nfs_ok) {
5320                 if (status != nfserr_share_denied) {
5321                         set_deny(open->op_share_deny, stp);
5322                         fp->fi_share_deny |=
5323                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5324                 } else {
5325                         if (nfs4_resolve_deny_conflicts_locked(fp, false,
5326                                         stp, open->op_share_deny, false))
5327                                 status = nfserr_jukebox;
5328                 }
5329         }
5330         spin_unlock(&fp->fi_lock);
5331
5332         if (status != nfs_ok)
5333                 return status;
5334
5335         status = nfsd4_truncate(rqstp, cur_fh, open);
5336         if (status != nfs_ok)
5337                 reset_union_bmap_deny(old_deny_bmap, stp);
5338         return status;
5339 }
5340
5341 /* Should we give out recallable state?: */
5342 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
5343 {
5344         if (clp->cl_cb_state == NFSD4_CB_UP)
5345                 return true;
5346         /*
5347          * In the sessions case, since we don't have to establish a
5348          * separate connection for callbacks, we assume it's OK
5349          * until we hear otherwise:
5350          */
5351         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
5352 }
5353
5354 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
5355                                                 int flag)
5356 {
5357         struct file_lock *fl;
5358
5359         fl = locks_alloc_lock();
5360         if (!fl)
5361                 return NULL;
5362         fl->fl_lmops = &nfsd_lease_mng_ops;
5363         fl->fl_flags = FL_DELEG;
5364         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
5365         fl->fl_end = OFFSET_MAX;
5366         fl->fl_owner = (fl_owner_t)dp;
5367         fl->fl_pid = current->tgid;
5368         fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
5369         return fl;
5370 }
5371
5372 static int nfsd4_check_conflicting_opens(struct nfs4_client *clp,
5373                                          struct nfs4_file *fp)
5374 {
5375         struct nfs4_ol_stateid *st;
5376         struct file *f = fp->fi_deleg_file->nf_file;
5377         struct inode *ino = locks_inode(f);
5378         int writes;
5379
5380         writes = atomic_read(&ino->i_writecount);
5381         if (!writes)
5382                 return 0;
5383         /*
5384          * There could be multiple filehandles (hence multiple
5385          * nfs4_files) referencing this file, but that's not too
5386          * common; let's just give up in that case rather than
5387          * trying to go look up all the clients using that other
5388          * nfs4_file as well:
5389          */
5390         if (fp->fi_aliased)
5391                 return -EAGAIN;
5392         /*
5393          * If there's a close in progress, make sure that we see it
5394          * clear any fi_fds[] entries before we see it decrement
5395          * i_writecount:
5396          */
5397         smp_mb__after_atomic();
5398
5399         if (fp->fi_fds[O_WRONLY])
5400                 writes--;
5401         if (fp->fi_fds[O_RDWR])
5402                 writes--;
5403         if (writes > 0)
5404                 return -EAGAIN; /* There may be non-NFSv4 writers */
5405         /*
5406          * It's possible there are non-NFSv4 write opens in progress,
5407          * but if they haven't incremented i_writecount yet then they
5408          * also haven't called break lease yet; so, they'll break this
5409          * lease soon enough.  So, all that's left to check for is NFSv4
5410          * opens:
5411          */
5412         spin_lock(&fp->fi_lock);
5413         list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
5414                 if (st->st_openstp == NULL /* it's an open */ &&
5415                     access_permit_write(st) &&
5416                     st->st_stid.sc_client != clp) {
5417                         spin_unlock(&fp->fi_lock);
5418                         return -EAGAIN;
5419                 }
5420         }
5421         spin_unlock(&fp->fi_lock);
5422         /*
5423          * There's a small chance that we could be racing with another
5424          * NFSv4 open.  However, any open that hasn't added itself to
5425          * the fi_stateids list also hasn't called break_lease yet; so,
5426          * they'll break this lease soon enough.
5427          */
5428         return 0;
5429 }
5430
5431 /*
5432  * It's possible that between opening the dentry and setting the delegation,
5433  * that it has been renamed or unlinked. Redo the lookup to verify that this
5434  * hasn't happened.
5435  */
5436 static int
5437 nfsd4_verify_deleg_dentry(struct nfsd4_open *open, struct nfs4_file *fp,
5438                           struct svc_fh *parent)
5439 {
5440         struct svc_export *exp;
5441         struct dentry *child;
5442         __be32 err;
5443
5444         err = nfsd_lookup_dentry(open->op_rqstp, parent,
5445                                  open->op_fname, open->op_fnamelen,
5446                                  &exp, &child);
5447
5448         if (err)
5449                 return -EAGAIN;
5450
5451         exp_put(exp);
5452         dput(child);
5453         if (child != file_dentry(fp->fi_deleg_file->nf_file))
5454                 return -EAGAIN;
5455
5456         return 0;
5457 }
5458
5459 static struct nfs4_delegation *
5460 nfs4_set_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5461                     struct svc_fh *parent)
5462 {
5463         int status = 0;
5464         struct nfs4_client *clp = stp->st_stid.sc_client;
5465         struct nfs4_file *fp = stp->st_stid.sc_file;
5466         struct nfs4_clnt_odstate *odstate = stp->st_clnt_odstate;
5467         struct nfs4_delegation *dp;
5468         struct nfsd_file *nf;
5469         struct file_lock *fl;
5470
5471         /*
5472          * The fi_had_conflict and nfs_get_existing_delegation checks
5473          * here are just optimizations; we'll need to recheck them at
5474          * the end:
5475          */
5476         if (fp->fi_had_conflict)
5477                 return ERR_PTR(-EAGAIN);
5478
5479         nf = find_readable_file(fp);
5480         if (!nf) {
5481                 /*
5482                  * We probably could attempt another open and get a read
5483                  * delegation, but for now, don't bother until the
5484                  * client actually sends us one.
5485                  */
5486                 return ERR_PTR(-EAGAIN);
5487         }
5488         spin_lock(&state_lock);
5489         spin_lock(&fp->fi_lock);
5490         if (nfs4_delegation_exists(clp, fp))
5491                 status = -EAGAIN;
5492         else if (!fp->fi_deleg_file) {
5493                 fp->fi_deleg_file = nf;
5494                 /* increment early to prevent fi_deleg_file from being
5495                  * cleared */
5496                 fp->fi_delegees = 1;
5497                 nf = NULL;
5498         } else
5499                 fp->fi_delegees++;
5500         spin_unlock(&fp->fi_lock);
5501         spin_unlock(&state_lock);
5502         if (nf)
5503                 nfsd_file_put(nf);
5504         if (status)
5505                 return ERR_PTR(status);
5506
5507         status = -ENOMEM;
5508         dp = alloc_init_deleg(clp, fp, odstate);
5509         if (!dp)
5510                 goto out_delegees;
5511
5512         fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
5513         if (!fl)
5514                 goto out_clnt_odstate;
5515
5516         status = vfs_setlease(fp->fi_deleg_file->nf_file, fl->fl_type, &fl, NULL);
5517         if (fl)
5518                 locks_free_lock(fl);
5519         if (status)
5520                 goto out_clnt_odstate;
5521
5522         if (parent) {
5523                 status = nfsd4_verify_deleg_dentry(open, fp, parent);
5524                 if (status)
5525                         goto out_unlock;
5526         }
5527
5528         status = nfsd4_check_conflicting_opens(clp, fp);
5529         if (status)
5530                 goto out_unlock;
5531
5532         spin_lock(&state_lock);
5533         spin_lock(&fp->fi_lock);
5534         if (fp->fi_had_conflict)
5535                 status = -EAGAIN;
5536         else
5537                 status = hash_delegation_locked(dp, fp);
5538         spin_unlock(&fp->fi_lock);
5539         spin_unlock(&state_lock);
5540
5541         if (status)
5542                 goto out_unlock;
5543
5544         return dp;
5545 out_unlock:
5546         vfs_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
5547 out_clnt_odstate:
5548         put_clnt_odstate(dp->dl_clnt_odstate);
5549         nfs4_put_stid(&dp->dl_stid);
5550 out_delegees:
5551         put_deleg_file(fp);
5552         return ERR_PTR(status);
5553 }
5554
5555 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
5556 {
5557         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5558         if (status == -EAGAIN)
5559                 open->op_why_no_deleg = WND4_CONTENTION;
5560         else {
5561                 open->op_why_no_deleg = WND4_RESOURCE;
5562                 switch (open->op_deleg_want) {
5563                 case NFS4_SHARE_WANT_READ_DELEG:
5564                 case NFS4_SHARE_WANT_WRITE_DELEG:
5565                 case NFS4_SHARE_WANT_ANY_DELEG:
5566                         break;
5567                 case NFS4_SHARE_WANT_CANCEL:
5568                         open->op_why_no_deleg = WND4_CANCELLED;
5569                         break;
5570                 case NFS4_SHARE_WANT_NO_DELEG:
5571                         WARN_ON_ONCE(1);
5572                 }
5573         }
5574 }
5575
5576 /*
5577  * Attempt to hand out a delegation.
5578  *
5579  * Note we don't support write delegations, and won't until the vfs has
5580  * proper support for them.
5581  */
5582 static void
5583 nfs4_open_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5584                      struct svc_fh *currentfh)
5585 {
5586         struct nfs4_delegation *dp;
5587         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
5588         struct nfs4_client *clp = stp->st_stid.sc_client;
5589         struct svc_fh *parent = NULL;
5590         int cb_up;
5591         int status = 0;
5592
5593         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
5594         open->op_recall = 0;
5595         switch (open->op_claim_type) {
5596                 case NFS4_OPEN_CLAIM_PREVIOUS:
5597                         if (!cb_up)
5598                                 open->op_recall = 1;
5599                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
5600                                 goto out_no_deleg;
5601                         break;
5602                 case NFS4_OPEN_CLAIM_NULL:
5603                         parent = currentfh;
5604                         fallthrough;
5605                 case NFS4_OPEN_CLAIM_FH:
5606                         /*
5607                          * Let's not give out any delegations till everyone's
5608                          * had the chance to reclaim theirs, *and* until
5609                          * NLM locks have all been reclaimed:
5610                          */
5611                         if (locks_in_grace(clp->net))
5612                                 goto out_no_deleg;
5613                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
5614                                 goto out_no_deleg;
5615                         break;
5616                 default:
5617                         goto out_no_deleg;
5618         }
5619         dp = nfs4_set_delegation(open, stp, parent);
5620         if (IS_ERR(dp))
5621                 goto out_no_deleg;
5622
5623         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5624
5625         trace_nfsd_deleg_read(&dp->dl_stid.sc_stateid);
5626         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
5627         nfs4_put_stid(&dp->dl_stid);
5628         return;
5629 out_no_deleg:
5630         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
5631         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
5632             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
5633                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
5634                 open->op_recall = 1;
5635         }
5636
5637         /* 4.1 client asking for a delegation? */
5638         if (open->op_deleg_want)
5639                 nfsd4_open_deleg_none_ext(open, status);
5640         return;
5641 }
5642
5643 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
5644                                         struct nfs4_delegation *dp)
5645 {
5646         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
5647             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5648                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5649                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
5650         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
5651                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5652                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5653                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
5654         }
5655         /* Otherwise the client must be confused wanting a delegation
5656          * it already has, therefore we don't return
5657          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
5658          */
5659 }
5660
5661 /**
5662  * nfsd4_process_open2 - finish open processing
5663  * @rqstp: the RPC transaction being executed
5664  * @current_fh: NFSv4 COMPOUND's current filehandle
5665  * @open: OPEN arguments
5666  *
5667  * If successful, (1) truncate the file if open->op_truncate was
5668  * set, (2) set open->op_stateid, (3) set open->op_delegation.
5669  *
5670  * Returns %nfs_ok on success; otherwise an nfs4stat value in
5671  * network byte order is returned.
5672  */
5673 __be32
5674 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
5675 {
5676         struct nfsd4_compoundres *resp = rqstp->rq_resp;
5677         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
5678         struct nfs4_file *fp = NULL;
5679         struct nfs4_ol_stateid *stp = NULL;
5680         struct nfs4_delegation *dp = NULL;
5681         __be32 status;
5682         bool new_stp = false;
5683
5684         /*
5685          * Lookup file; if found, lookup stateid and check open request,
5686          * and check for delegations in the process of being recalled.
5687          * If not found, create the nfs4_file struct
5688          */
5689         fp = nfsd4_file_hash_insert(open->op_file, current_fh);
5690         if (unlikely(!fp))
5691                 return nfserr_jukebox;
5692         if (fp != open->op_file) {
5693                 status = nfs4_check_deleg(cl, open, &dp);
5694                 if (status)
5695                         goto out;
5696                 stp = nfsd4_find_and_lock_existing_open(fp, open);
5697         } else {
5698                 open->op_file = NULL;
5699                 status = nfserr_bad_stateid;
5700                 if (nfsd4_is_deleg_cur(open))
5701                         goto out;
5702         }
5703
5704         if (!stp) {
5705                 stp = init_open_stateid(fp, open);
5706                 if (!open->op_stp)
5707                         new_stp = true;
5708         }
5709
5710         /*
5711          * OPEN the file, or upgrade an existing OPEN.
5712          * If truncate fails, the OPEN fails.
5713          *
5714          * stp is already locked.
5715          */
5716         if (!new_stp) {
5717                 /* Stateid was found, this is an OPEN upgrade */
5718                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
5719                 if (status) {
5720                         mutex_unlock(&stp->st_mutex);
5721                         goto out;
5722                 }
5723         } else {
5724                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open, true);
5725                 if (status) {
5726                         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5727                         release_open_stateid(stp);
5728                         mutex_unlock(&stp->st_mutex);
5729                         goto out;
5730                 }
5731
5732                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
5733                                                         open->op_odstate);
5734                 if (stp->st_clnt_odstate == open->op_odstate)
5735                         open->op_odstate = NULL;
5736         }
5737
5738         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
5739         mutex_unlock(&stp->st_mutex);
5740
5741         if (nfsd4_has_session(&resp->cstate)) {
5742                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
5743                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5744                         open->op_why_no_deleg = WND4_NOT_WANTED;
5745                         goto nodeleg;
5746                 }
5747         }
5748
5749         /*
5750         * Attempt to hand out a delegation. No error return, because the
5751         * OPEN succeeds even if we fail.
5752         */
5753         nfs4_open_delegation(open, stp, &resp->cstate.current_fh);
5754 nodeleg:
5755         status = nfs_ok;
5756         trace_nfsd_open(&stp->st_stid.sc_stateid);
5757 out:
5758         /* 4.1 client trying to upgrade/downgrade delegation? */
5759         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
5760             open->op_deleg_want)
5761                 nfsd4_deleg_xgrade_none_ext(open, dp);
5762
5763         if (fp)
5764                 put_nfs4_file(fp);
5765         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
5766                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5767         /*
5768         * To finish the open response, we just need to set the rflags.
5769         */
5770         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
5771         if (nfsd4_has_session(&resp->cstate))
5772                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
5773         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
5774                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
5775
5776         if (dp)
5777                 nfs4_put_stid(&dp->dl_stid);
5778         if (stp)
5779                 nfs4_put_stid(&stp->st_stid);
5780
5781         return status;
5782 }
5783
5784 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
5785                               struct nfsd4_open *open)
5786 {
5787         if (open->op_openowner) {
5788                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
5789
5790                 nfsd4_cstate_assign_replay(cstate, so);
5791                 nfs4_put_stateowner(so);
5792         }
5793         if (open->op_file)
5794                 kmem_cache_free(file_slab, open->op_file);
5795         if (open->op_stp)
5796                 nfs4_put_stid(&open->op_stp->st_stid);
5797         if (open->op_odstate)
5798                 kmem_cache_free(odstate_slab, open->op_odstate);
5799 }
5800
5801 __be32
5802 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5803             union nfsd4_op_u *u)
5804 {
5805         clientid_t *clid = &u->renew;
5806         struct nfs4_client *clp;
5807         __be32 status;
5808         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5809
5810         trace_nfsd_clid_renew(clid);
5811         status = set_client(clid, cstate, nn);
5812         if (status)
5813                 return status;
5814         clp = cstate->clp;
5815         if (!list_empty(&clp->cl_delegations)
5816                         && clp->cl_cb_state != NFSD4_CB_UP)
5817                 return nfserr_cb_path_down;
5818         return nfs_ok;
5819 }
5820
5821 void
5822 nfsd4_end_grace(struct nfsd_net *nn)
5823 {
5824         /* do nothing if grace period already ended */
5825         if (nn->grace_ended)
5826                 return;
5827
5828         trace_nfsd_grace_complete(nn);
5829         nn->grace_ended = true;
5830         /*
5831          * If the server goes down again right now, an NFSv4
5832          * client will still be allowed to reclaim after it comes back up,
5833          * even if it hasn't yet had a chance to reclaim state this time.
5834          *
5835          */
5836         nfsd4_record_grace_done(nn);
5837         /*
5838          * At this point, NFSv4 clients can still reclaim.  But if the
5839          * server crashes, any that have not yet reclaimed will be out
5840          * of luck on the next boot.
5841          *
5842          * (NFSv4.1+ clients are considered to have reclaimed once they
5843          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
5844          * have reclaimed after their first OPEN.)
5845          */
5846         locks_end_grace(&nn->nfsd4_manager);
5847         /*
5848          * At this point, and once lockd and/or any other containers
5849          * exit their grace period, further reclaims will fail and
5850          * regular locking can resume.
5851          */
5852 }
5853
5854 /*
5855  * If we've waited a lease period but there are still clients trying to
5856  * reclaim, wait a little longer to give them a chance to finish.
5857  */
5858 static bool clients_still_reclaiming(struct nfsd_net *nn)
5859 {
5860         time64_t double_grace_period_end = nn->boot_time +
5861                                            2 * nn->nfsd4_lease;
5862
5863         if (nn->track_reclaim_completes &&
5864                         atomic_read(&nn->nr_reclaim_complete) ==
5865                         nn->reclaim_str_hashtbl_size)
5866                 return false;
5867         if (!nn->somebody_reclaimed)
5868                 return false;
5869         nn->somebody_reclaimed = false;
5870         /*
5871          * If we've given them *two* lease times to reclaim, and they're
5872          * still not done, give up:
5873          */
5874         if (ktime_get_boottime_seconds() > double_grace_period_end)
5875                 return false;
5876         return true;
5877 }
5878
5879 struct laundry_time {
5880         time64_t cutoff;
5881         time64_t new_timeo;
5882 };
5883
5884 static bool state_expired(struct laundry_time *lt, time64_t last_refresh)
5885 {
5886         time64_t time_remaining;
5887
5888         if (last_refresh < lt->cutoff)
5889                 return true;
5890         time_remaining = last_refresh - lt->cutoff;
5891         lt->new_timeo = min(lt->new_timeo, time_remaining);
5892         return false;
5893 }
5894
5895 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
5896 void nfsd4_ssc_init_umount_work(struct nfsd_net *nn)
5897 {
5898         spin_lock_init(&nn->nfsd_ssc_lock);
5899         INIT_LIST_HEAD(&nn->nfsd_ssc_mount_list);
5900         init_waitqueue_head(&nn->nfsd_ssc_waitq);
5901 }
5902 EXPORT_SYMBOL_GPL(nfsd4_ssc_init_umount_work);
5903
5904 /*
5905  * This is called when nfsd is being shutdown, after all inter_ssc
5906  * cleanup were done, to destroy the ssc delayed unmount list.
5907  */
5908 static void nfsd4_ssc_shutdown_umount(struct nfsd_net *nn)
5909 {
5910         struct nfsd4_ssc_umount_item *ni = NULL;
5911         struct nfsd4_ssc_umount_item *tmp;
5912
5913         spin_lock(&nn->nfsd_ssc_lock);
5914         list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
5915                 list_del(&ni->nsui_list);
5916                 spin_unlock(&nn->nfsd_ssc_lock);
5917                 mntput(ni->nsui_vfsmount);
5918                 kfree(ni);
5919                 spin_lock(&nn->nfsd_ssc_lock);
5920         }
5921         spin_unlock(&nn->nfsd_ssc_lock);
5922 }
5923
5924 static void nfsd4_ssc_expire_umount(struct nfsd_net *nn)
5925 {
5926         bool do_wakeup = false;
5927         struct nfsd4_ssc_umount_item *ni = NULL;
5928         struct nfsd4_ssc_umount_item *tmp;
5929
5930         spin_lock(&nn->nfsd_ssc_lock);
5931         list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
5932                 if (time_after(jiffies, ni->nsui_expire)) {
5933                         if (refcount_read(&ni->nsui_refcnt) > 1)
5934                                 continue;
5935
5936                         /* mark being unmount */
5937                         ni->nsui_busy = true;
5938                         spin_unlock(&nn->nfsd_ssc_lock);
5939                         mntput(ni->nsui_vfsmount);
5940                         spin_lock(&nn->nfsd_ssc_lock);
5941
5942                         /* waiters need to start from begin of list */
5943                         list_del(&ni->nsui_list);
5944                         kfree(ni);
5945
5946                         /* wakeup ssc_connect waiters */
5947                         do_wakeup = true;
5948                         continue;
5949                 }
5950                 break;
5951         }
5952         if (do_wakeup)
5953                 wake_up_all(&nn->nfsd_ssc_waitq);
5954         spin_unlock(&nn->nfsd_ssc_lock);
5955 }
5956 #endif
5957
5958 /* Check if any lock belonging to this lockowner has any blockers */
5959 static bool
5960 nfs4_lockowner_has_blockers(struct nfs4_lockowner *lo)
5961 {
5962         struct file_lock_context *ctx;
5963         struct nfs4_ol_stateid *stp;
5964         struct nfs4_file *nf;
5965
5966         list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
5967                 nf = stp->st_stid.sc_file;
5968                 ctx = locks_inode_context(nf->fi_inode);
5969                 if (!ctx)
5970                         continue;
5971                 if (locks_owner_has_blockers(ctx, lo))
5972                         return true;
5973         }
5974         return false;
5975 }
5976
5977 static bool
5978 nfs4_anylock_blockers(struct nfs4_client *clp)
5979 {
5980         int i;
5981         struct nfs4_stateowner *so;
5982         struct nfs4_lockowner *lo;
5983
5984         if (atomic_read(&clp->cl_delegs_in_recall))
5985                 return true;
5986         spin_lock(&clp->cl_lock);
5987         for (i = 0; i < OWNER_HASH_SIZE; i++) {
5988                 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[i],
5989                                 so_strhash) {
5990                         if (so->so_is_open_owner)
5991                                 continue;
5992                         lo = lockowner(so);
5993                         if (nfs4_lockowner_has_blockers(lo)) {
5994                                 spin_unlock(&clp->cl_lock);
5995                                 return true;
5996                         }
5997                 }
5998         }
5999         spin_unlock(&clp->cl_lock);
6000         return false;
6001 }
6002
6003 static void
6004 nfs4_get_client_reaplist(struct nfsd_net *nn, struct list_head *reaplist,
6005                                 struct laundry_time *lt)
6006 {
6007         unsigned int maxreap, reapcnt = 0;
6008         struct list_head *pos, *next;
6009         struct nfs4_client *clp;
6010
6011         maxreap = (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) ?
6012                         NFSD_CLIENT_MAX_TRIM_PER_RUN : 0;
6013         INIT_LIST_HEAD(reaplist);
6014         spin_lock(&nn->client_lock);
6015         list_for_each_safe(pos, next, &nn->client_lru) {
6016                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6017                 if (clp->cl_state == NFSD4_EXPIRABLE)
6018                         goto exp_client;
6019                 if (!state_expired(lt, clp->cl_time))
6020                         break;
6021                 if (!atomic_read(&clp->cl_rpc_users)) {
6022                         if (clp->cl_state == NFSD4_ACTIVE)
6023                                 atomic_inc(&nn->nfsd_courtesy_clients);
6024                         clp->cl_state = NFSD4_COURTESY;
6025                 }
6026                 if (!client_has_state(clp))
6027                         goto exp_client;
6028                 if (!nfs4_anylock_blockers(clp))
6029                         if (reapcnt >= maxreap)
6030                                 continue;
6031 exp_client:
6032                 if (!mark_client_expired_locked(clp)) {
6033                         list_add(&clp->cl_lru, reaplist);
6034                         reapcnt++;
6035                 }
6036         }
6037         spin_unlock(&nn->client_lock);
6038 }
6039
6040 static void
6041 nfs4_get_courtesy_client_reaplist(struct nfsd_net *nn,
6042                                 struct list_head *reaplist)
6043 {
6044         unsigned int maxreap = 0, reapcnt = 0;
6045         struct list_head *pos, *next;
6046         struct nfs4_client *clp;
6047
6048         maxreap = NFSD_CLIENT_MAX_TRIM_PER_RUN;
6049         INIT_LIST_HEAD(reaplist);
6050
6051         spin_lock(&nn->client_lock);
6052         list_for_each_safe(pos, next, &nn->client_lru) {
6053                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6054                 if (clp->cl_state == NFSD4_ACTIVE)
6055                         break;
6056                 if (reapcnt >= maxreap)
6057                         break;
6058                 if (!mark_client_expired_locked(clp)) {
6059                         list_add(&clp->cl_lru, reaplist);
6060                         reapcnt++;
6061                 }
6062         }
6063         spin_unlock(&nn->client_lock);
6064 }
6065
6066 static void
6067 nfs4_process_client_reaplist(struct list_head *reaplist)
6068 {
6069         struct list_head *pos, *next;
6070         struct nfs4_client *clp;
6071
6072         list_for_each_safe(pos, next, reaplist) {
6073                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6074                 trace_nfsd_clid_purged(&clp->cl_clientid);
6075                 list_del_init(&clp->cl_lru);
6076                 expire_client(clp);
6077         }
6078 }
6079
6080 static time64_t
6081 nfs4_laundromat(struct nfsd_net *nn)
6082 {
6083         struct nfs4_openowner *oo;
6084         struct nfs4_delegation *dp;
6085         struct nfs4_ol_stateid *stp;
6086         struct nfsd4_blocked_lock *nbl;
6087         struct list_head *pos, *next, reaplist;
6088         struct laundry_time lt = {
6089                 .cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease,
6090                 .new_timeo = nn->nfsd4_lease
6091         };
6092         struct nfs4_cpntf_state *cps;
6093         copy_stateid_t *cps_t;
6094         int i;
6095
6096         if (clients_still_reclaiming(nn)) {
6097                 lt.new_timeo = 0;
6098                 goto out;
6099         }
6100         nfsd4_end_grace(nn);
6101
6102         spin_lock(&nn->s2s_cp_lock);
6103         idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
6104                 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
6105                 if (cps->cp_stateid.cs_type == NFS4_COPYNOTIFY_STID &&
6106                                 state_expired(&lt, cps->cpntf_time))
6107                         _free_cpntf_state_locked(nn, cps);
6108         }
6109         spin_unlock(&nn->s2s_cp_lock);
6110         nfs4_get_client_reaplist(nn, &reaplist, &lt);
6111         nfs4_process_client_reaplist(&reaplist);
6112
6113         spin_lock(&state_lock);
6114         list_for_each_safe(pos, next, &nn->del_recall_lru) {
6115                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6116                 if (!state_expired(&lt, dp->dl_time))
6117                         break;
6118                 WARN_ON(!unhash_delegation_locked(dp));
6119                 list_add(&dp->dl_recall_lru, &reaplist);
6120         }
6121         spin_unlock(&state_lock);
6122         while (!list_empty(&reaplist)) {
6123                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
6124                                         dl_recall_lru);
6125                 list_del_init(&dp->dl_recall_lru);
6126                 revoke_delegation(dp);
6127         }
6128
6129         spin_lock(&nn->client_lock);
6130         while (!list_empty(&nn->close_lru)) {
6131                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
6132                                         oo_close_lru);
6133                 if (!state_expired(&lt, oo->oo_time))
6134                         break;
6135                 list_del_init(&oo->oo_close_lru);
6136                 stp = oo->oo_last_closed_stid;
6137                 oo->oo_last_closed_stid = NULL;
6138                 spin_unlock(&nn->client_lock);
6139                 nfs4_put_stid(&stp->st_stid);
6140                 spin_lock(&nn->client_lock);
6141         }
6142         spin_unlock(&nn->client_lock);
6143
6144         /*
6145          * It's possible for a client to try and acquire an already held lock
6146          * that is being held for a long time, and then lose interest in it.
6147          * So, we clean out any un-revisited request after a lease period
6148          * under the assumption that the client is no longer interested.
6149          *
6150          * RFC5661, sec. 9.6 states that the client must not rely on getting
6151          * notifications and must continue to poll for locks, even when the
6152          * server supports them. Thus this shouldn't lead to clients blocking
6153          * indefinitely once the lock does become free.
6154          */
6155         BUG_ON(!list_empty(&reaplist));
6156         spin_lock(&nn->blocked_locks_lock);
6157         while (!list_empty(&nn->blocked_locks_lru)) {
6158                 nbl = list_first_entry(&nn->blocked_locks_lru,
6159                                         struct nfsd4_blocked_lock, nbl_lru);
6160                 if (!state_expired(&lt, nbl->nbl_time))
6161                         break;
6162                 list_move(&nbl->nbl_lru, &reaplist);
6163                 list_del_init(&nbl->nbl_list);
6164         }
6165         spin_unlock(&nn->blocked_locks_lock);
6166
6167         while (!list_empty(&reaplist)) {
6168                 nbl = list_first_entry(&reaplist,
6169                                         struct nfsd4_blocked_lock, nbl_lru);
6170                 list_del_init(&nbl->nbl_lru);
6171                 free_blocked_lock(nbl);
6172         }
6173 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
6174         /* service the server-to-server copy delayed unmount list */
6175         nfsd4_ssc_expire_umount(nn);
6176 #endif
6177 out:
6178         return max_t(time64_t, lt.new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
6179 }
6180
6181 static void laundromat_main(struct work_struct *);
6182
6183 static void
6184 laundromat_main(struct work_struct *laundry)
6185 {
6186         time64_t t;
6187         struct delayed_work *dwork = to_delayed_work(laundry);
6188         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
6189                                            laundromat_work);
6190
6191         t = nfs4_laundromat(nn);
6192         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
6193 }
6194
6195 static void
6196 courtesy_client_reaper(struct nfsd_net *nn)
6197 {
6198         struct list_head reaplist;
6199
6200         nfs4_get_courtesy_client_reaplist(nn, &reaplist);
6201         nfs4_process_client_reaplist(&reaplist);
6202 }
6203
6204 static void
6205 deleg_reaper(struct nfsd_net *nn)
6206 {
6207         struct list_head *pos, *next;
6208         struct nfs4_client *clp;
6209         struct list_head cblist;
6210
6211         INIT_LIST_HEAD(&cblist);
6212         spin_lock(&nn->client_lock);
6213         list_for_each_safe(pos, next, &nn->client_lru) {
6214                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6215                 if (clp->cl_state != NFSD4_ACTIVE ||
6216                         list_empty(&clp->cl_delegations) ||
6217                         atomic_read(&clp->cl_delegs_in_recall) ||
6218                         test_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags) ||
6219                         (ktime_get_boottime_seconds() -
6220                                 clp->cl_ra_time < 5)) {
6221                         continue;
6222                 }
6223                 list_add(&clp->cl_ra_cblist, &cblist);
6224
6225                 /* release in nfsd4_cb_recall_any_release */
6226                 atomic_inc(&clp->cl_rpc_users);
6227                 set_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
6228                 clp->cl_ra_time = ktime_get_boottime_seconds();
6229         }
6230         spin_unlock(&nn->client_lock);
6231
6232         while (!list_empty(&cblist)) {
6233                 clp = list_first_entry(&cblist, struct nfs4_client,
6234                                         cl_ra_cblist);
6235                 list_del_init(&clp->cl_ra_cblist);
6236                 clp->cl_ra->ra_keep = 0;
6237                 clp->cl_ra->ra_bmval[0] = BIT(RCA4_TYPE_MASK_RDATA_DLG);
6238                 trace_nfsd_cb_recall_any(clp->cl_ra);
6239                 nfsd4_run_cb(&clp->cl_ra->ra_cb);
6240         }
6241 }
6242
6243 static void
6244 nfsd4_state_shrinker_worker(struct work_struct *work)
6245 {
6246         struct delayed_work *dwork = to_delayed_work(work);
6247         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
6248                                 nfsd_shrinker_work);
6249
6250         courtesy_client_reaper(nn);
6251         deleg_reaper(nn);
6252 }
6253
6254 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
6255 {
6256         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
6257                 return nfserr_bad_stateid;
6258         return nfs_ok;
6259 }
6260
6261 static
6262 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
6263 {
6264         __be32 status = nfserr_openmode;
6265
6266         /* For lock stateid's, we test the parent open, not the lock: */
6267         if (stp->st_openstp)
6268                 stp = stp->st_openstp;
6269         if ((flags & WR_STATE) && !access_permit_write(stp))
6270                 goto out;
6271         if ((flags & RD_STATE) && !access_permit_read(stp))
6272                 goto out;
6273         status = nfs_ok;
6274 out:
6275         return status;
6276 }
6277
6278 static inline __be32
6279 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
6280 {
6281         if (ONE_STATEID(stateid) && (flags & RD_STATE))
6282                 return nfs_ok;
6283         else if (opens_in_grace(net)) {
6284                 /* Answer in remaining cases depends on existence of
6285                  * conflicting state; so we must wait out the grace period. */
6286                 return nfserr_grace;
6287         } else if (flags & WR_STATE)
6288                 return nfs4_share_conflict(current_fh,
6289                                 NFS4_SHARE_DENY_WRITE);
6290         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
6291                 return nfs4_share_conflict(current_fh,
6292                                 NFS4_SHARE_DENY_READ);
6293 }
6294
6295 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
6296 {
6297         /*
6298          * When sessions are used the stateid generation number is ignored
6299          * when it is zero.
6300          */
6301         if (has_session && in->si_generation == 0)
6302                 return nfs_ok;
6303
6304         if (in->si_generation == ref->si_generation)
6305                 return nfs_ok;
6306
6307         /* If the client sends us a stateid from the future, it's buggy: */
6308         if (nfsd4_stateid_generation_after(in, ref))
6309                 return nfserr_bad_stateid;
6310         /*
6311          * However, we could see a stateid from the past, even from a
6312          * non-buggy client.  For example, if the client sends a lock
6313          * while some IO is outstanding, the lock may bump si_generation
6314          * while the IO is still in flight.  The client could avoid that
6315          * situation by waiting for responses on all the IO requests,
6316          * but better performance may result in retrying IO that
6317          * receives an old_stateid error if requests are rarely
6318          * reordered in flight:
6319          */
6320         return nfserr_old_stateid;
6321 }
6322
6323 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
6324 {
6325         __be32 ret;
6326
6327         spin_lock(&s->sc_lock);
6328         ret = nfsd4_verify_open_stid(s);
6329         if (ret == nfs_ok)
6330                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
6331         spin_unlock(&s->sc_lock);
6332         return ret;
6333 }
6334
6335 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
6336 {
6337         if (ols->st_stateowner->so_is_open_owner &&
6338             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
6339                 return nfserr_bad_stateid;
6340         return nfs_ok;
6341 }
6342
6343 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
6344 {
6345         struct nfs4_stid *s;
6346         __be32 status = nfserr_bad_stateid;
6347
6348         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6349                 CLOSE_STATEID(stateid))
6350                 return status;
6351         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid))
6352                 return status;
6353         spin_lock(&cl->cl_lock);
6354         s = find_stateid_locked(cl, stateid);
6355         if (!s)
6356                 goto out_unlock;
6357         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
6358         if (status)
6359                 goto out_unlock;
6360         switch (s->sc_type) {
6361         case NFS4_DELEG_STID:
6362                 status = nfs_ok;
6363                 break;
6364         case NFS4_REVOKED_DELEG_STID:
6365                 status = nfserr_deleg_revoked;
6366                 break;
6367         case NFS4_OPEN_STID:
6368         case NFS4_LOCK_STID:
6369                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
6370                 break;
6371         default:
6372                 printk("unknown stateid type %x\n", s->sc_type);
6373                 fallthrough;
6374         case NFS4_CLOSED_STID:
6375         case NFS4_CLOSED_DELEG_STID:
6376                 status = nfserr_bad_stateid;
6377         }
6378 out_unlock:
6379         spin_unlock(&cl->cl_lock);
6380         return status;
6381 }
6382
6383 __be32
6384 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
6385                      stateid_t *stateid, unsigned char typemask,
6386                      struct nfs4_stid **s, struct nfsd_net *nn)
6387 {
6388         __be32 status;
6389         struct nfs4_stid *stid;
6390         bool return_revoked = false;
6391
6392         /*
6393          *  only return revoked delegations if explicitly asked.
6394          *  otherwise we report revoked or bad_stateid status.
6395          */
6396         if (typemask & NFS4_REVOKED_DELEG_STID)
6397                 return_revoked = true;
6398         else if (typemask & NFS4_DELEG_STID)
6399                 typemask |= NFS4_REVOKED_DELEG_STID;
6400
6401         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6402                 CLOSE_STATEID(stateid))
6403                 return nfserr_bad_stateid;
6404         status = set_client(&stateid->si_opaque.so_clid, cstate, nn);
6405         if (status == nfserr_stale_clientid) {
6406                 if (cstate->session)
6407                         return nfserr_bad_stateid;
6408                 return nfserr_stale_stateid;
6409         }
6410         if (status)
6411                 return status;
6412         stid = find_stateid_by_type(cstate->clp, stateid, typemask);
6413         if (!stid)
6414                 return nfserr_bad_stateid;
6415         if ((stid->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
6416                 nfs4_put_stid(stid);
6417                 if (cstate->minorversion)
6418                         return nfserr_deleg_revoked;
6419                 return nfserr_bad_stateid;
6420         }
6421         *s = stid;
6422         return nfs_ok;
6423 }
6424
6425 static struct nfsd_file *
6426 nfs4_find_file(struct nfs4_stid *s, int flags)
6427 {
6428         if (!s)
6429                 return NULL;
6430
6431         switch (s->sc_type) {
6432         case NFS4_DELEG_STID:
6433                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
6434                         return NULL;
6435                 return nfsd_file_get(s->sc_file->fi_deleg_file);
6436         case NFS4_OPEN_STID:
6437         case NFS4_LOCK_STID:
6438                 if (flags & RD_STATE)
6439                         return find_readable_file(s->sc_file);
6440                 else
6441                         return find_writeable_file(s->sc_file);
6442         }
6443
6444         return NULL;
6445 }
6446
6447 static __be32
6448 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
6449 {
6450         __be32 status;
6451
6452         status = nfsd4_check_openowner_confirmed(ols);
6453         if (status)
6454                 return status;
6455         return nfs4_check_openmode(ols, flags);
6456 }
6457
6458 static __be32
6459 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
6460                 struct nfsd_file **nfp, int flags)
6461 {
6462         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
6463         struct nfsd_file *nf;
6464         __be32 status;
6465
6466         nf = nfs4_find_file(s, flags);
6467         if (nf) {
6468                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
6469                                 acc | NFSD_MAY_OWNER_OVERRIDE);
6470                 if (status) {
6471                         nfsd_file_put(nf);
6472                         goto out;
6473                 }
6474         } else {
6475                 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
6476                 if (status)
6477                         return status;
6478         }
6479         *nfp = nf;
6480 out:
6481         return status;
6482 }
6483 static void
6484 _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6485 {
6486         WARN_ON_ONCE(cps->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID);
6487         if (!refcount_dec_and_test(&cps->cp_stateid.cs_count))
6488                 return;
6489         list_del(&cps->cp_list);
6490         idr_remove(&nn->s2s_cp_stateids,
6491                    cps->cp_stateid.cs_stid.si_opaque.so_id);
6492         kfree(cps);
6493 }
6494 /*
6495  * A READ from an inter server to server COPY will have a
6496  * copy stateid. Look up the copy notify stateid from the
6497  * idr structure and take a reference on it.
6498  */
6499 __be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6500                           struct nfs4_client *clp,
6501                           struct nfs4_cpntf_state **cps)
6502 {
6503         copy_stateid_t *cps_t;
6504         struct nfs4_cpntf_state *state = NULL;
6505
6506         if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
6507                 return nfserr_bad_stateid;
6508         spin_lock(&nn->s2s_cp_lock);
6509         cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
6510         if (cps_t) {
6511                 state = container_of(cps_t, struct nfs4_cpntf_state,
6512                                      cp_stateid);
6513                 if (state->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID) {
6514                         state = NULL;
6515                         goto unlock;
6516                 }
6517                 if (!clp)
6518                         refcount_inc(&state->cp_stateid.cs_count);
6519                 else
6520                         _free_cpntf_state_locked(nn, state);
6521         }
6522 unlock:
6523         spin_unlock(&nn->s2s_cp_lock);
6524         if (!state)
6525                 return nfserr_bad_stateid;
6526         if (!clp && state)
6527                 *cps = state;
6528         return 0;
6529 }
6530
6531 static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6532                                struct nfs4_stid **stid)
6533 {
6534         __be32 status;
6535         struct nfs4_cpntf_state *cps = NULL;
6536         struct nfs4_client *found;
6537
6538         status = manage_cpntf_state(nn, st, NULL, &cps);
6539         if (status)
6540                 return status;
6541
6542         cps->cpntf_time = ktime_get_boottime_seconds();
6543
6544         status = nfserr_expired;
6545         found = lookup_clientid(&cps->cp_p_clid, true, nn);
6546         if (!found)
6547                 goto out;
6548
6549         *stid = find_stateid_by_type(found, &cps->cp_p_stateid,
6550                         NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID);
6551         if (*stid)
6552                 status = nfs_ok;
6553         else
6554                 status = nfserr_bad_stateid;
6555
6556         put_client_renew(found);
6557 out:
6558         nfs4_put_cpntf_state(nn, cps);
6559         return status;
6560 }
6561
6562 void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6563 {
6564         spin_lock(&nn->s2s_cp_lock);
6565         _free_cpntf_state_locked(nn, cps);
6566         spin_unlock(&nn->s2s_cp_lock);
6567 }
6568
6569 /*
6570  * Checks for stateid operations
6571  */
6572 __be32
6573 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
6574                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
6575                 stateid_t *stateid, int flags, struct nfsd_file **nfp,
6576                 struct nfs4_stid **cstid)
6577 {
6578         struct net *net = SVC_NET(rqstp);
6579         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6580         struct nfs4_stid *s = NULL;
6581         __be32 status;
6582
6583         if (nfp)
6584                 *nfp = NULL;
6585
6586         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
6587                 if (cstid)
6588                         status = nfserr_bad_stateid;
6589                 else
6590                         status = check_special_stateids(net, fhp, stateid,
6591                                                                         flags);
6592                 goto done;
6593         }
6594
6595         status = nfsd4_lookup_stateid(cstate, stateid,
6596                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
6597                                 &s, nn);
6598         if (status == nfserr_bad_stateid)
6599                 status = find_cpntf_state(nn, stateid, &s);
6600         if (status)
6601                 return status;
6602         status = nfsd4_stid_check_stateid_generation(stateid, s,
6603                         nfsd4_has_session(cstate));
6604         if (status)
6605                 goto out;
6606
6607         switch (s->sc_type) {
6608         case NFS4_DELEG_STID:
6609                 status = nfs4_check_delegmode(delegstateid(s), flags);
6610                 break;
6611         case NFS4_OPEN_STID:
6612         case NFS4_LOCK_STID:
6613                 status = nfs4_check_olstateid(openlockstateid(s), flags);
6614                 break;
6615         default:
6616                 status = nfserr_bad_stateid;
6617                 break;
6618         }
6619         if (status)
6620                 goto out;
6621         status = nfs4_check_fh(fhp, s);
6622
6623 done:
6624         if (status == nfs_ok && nfp)
6625                 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
6626 out:
6627         if (s) {
6628                 if (!status && cstid)
6629                         *cstid = s;
6630                 else
6631                         nfs4_put_stid(s);
6632         }
6633         return status;
6634 }
6635
6636 /*
6637  * Test if the stateid is valid
6638  */
6639 __be32
6640 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6641                    union nfsd4_op_u *u)
6642 {
6643         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
6644         struct nfsd4_test_stateid_id *stateid;
6645         struct nfs4_client *cl = cstate->clp;
6646
6647         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
6648                 stateid->ts_id_status =
6649                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
6650
6651         return nfs_ok;
6652 }
6653
6654 static __be32
6655 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
6656 {
6657         struct nfs4_ol_stateid *stp = openlockstateid(s);
6658         __be32 ret;
6659
6660         ret = nfsd4_lock_ol_stateid(stp);
6661         if (ret)
6662                 goto out_put_stid;
6663
6664         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
6665         if (ret)
6666                 goto out;
6667
6668         ret = nfserr_locks_held;
6669         if (check_for_locks(stp->st_stid.sc_file,
6670                             lockowner(stp->st_stateowner)))
6671                 goto out;
6672
6673         release_lock_stateid(stp);
6674         ret = nfs_ok;
6675
6676 out:
6677         mutex_unlock(&stp->st_mutex);
6678 out_put_stid:
6679         nfs4_put_stid(s);
6680         return ret;
6681 }
6682
6683 __be32
6684 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6685                    union nfsd4_op_u *u)
6686 {
6687         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
6688         stateid_t *stateid = &free_stateid->fr_stateid;
6689         struct nfs4_stid *s;
6690         struct nfs4_delegation *dp;
6691         struct nfs4_client *cl = cstate->clp;
6692         __be32 ret = nfserr_bad_stateid;
6693
6694         spin_lock(&cl->cl_lock);
6695         s = find_stateid_locked(cl, stateid);
6696         if (!s)
6697                 goto out_unlock;
6698         spin_lock(&s->sc_lock);
6699         switch (s->sc_type) {
6700         case NFS4_DELEG_STID:
6701                 ret = nfserr_locks_held;
6702                 break;
6703         case NFS4_OPEN_STID:
6704                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
6705                 if (ret)
6706                         break;
6707                 ret = nfserr_locks_held;
6708                 break;
6709         case NFS4_LOCK_STID:
6710                 spin_unlock(&s->sc_lock);
6711                 refcount_inc(&s->sc_count);
6712                 spin_unlock(&cl->cl_lock);
6713                 ret = nfsd4_free_lock_stateid(stateid, s);
6714                 goto out;
6715         case NFS4_REVOKED_DELEG_STID:
6716                 spin_unlock(&s->sc_lock);
6717                 dp = delegstateid(s);
6718                 list_del_init(&dp->dl_recall_lru);
6719                 spin_unlock(&cl->cl_lock);
6720                 nfs4_put_stid(s);
6721                 ret = nfs_ok;
6722                 goto out;
6723         /* Default falls through and returns nfserr_bad_stateid */
6724         }
6725         spin_unlock(&s->sc_lock);
6726 out_unlock:
6727         spin_unlock(&cl->cl_lock);
6728 out:
6729         return ret;
6730 }
6731
6732 static inline int
6733 setlkflg (int type)
6734 {
6735         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
6736                 RD_STATE : WR_STATE;
6737 }
6738
6739 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
6740 {
6741         struct svc_fh *current_fh = &cstate->current_fh;
6742         struct nfs4_stateowner *sop = stp->st_stateowner;
6743         __be32 status;
6744
6745         status = nfsd4_check_seqid(cstate, sop, seqid);
6746         if (status)
6747                 return status;
6748         status = nfsd4_lock_ol_stateid(stp);
6749         if (status != nfs_ok)
6750                 return status;
6751         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
6752         if (status == nfs_ok)
6753                 status = nfs4_check_fh(current_fh, &stp->st_stid);
6754         if (status != nfs_ok)
6755                 mutex_unlock(&stp->st_mutex);
6756         return status;
6757 }
6758
6759 /*
6760  * Checks for sequence id mutating operations.
6761  */
6762 static __be32
6763 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
6764                          stateid_t *stateid, char typemask,
6765                          struct nfs4_ol_stateid **stpp,
6766                          struct nfsd_net *nn)
6767 {
6768         __be32 status;
6769         struct nfs4_stid *s;
6770         struct nfs4_ol_stateid *stp = NULL;
6771
6772         trace_nfsd_preprocess(seqid, stateid);
6773
6774         *stpp = NULL;
6775         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
6776         if (status)
6777                 return status;
6778         stp = openlockstateid(s);
6779         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
6780
6781         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
6782         if (!status)
6783                 *stpp = stp;
6784         else
6785                 nfs4_put_stid(&stp->st_stid);
6786         return status;
6787 }
6788
6789 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
6790                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
6791 {
6792         __be32 status;
6793         struct nfs4_openowner *oo;
6794         struct nfs4_ol_stateid *stp;
6795
6796         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
6797                                                 NFS4_OPEN_STID, &stp, nn);
6798         if (status)
6799                 return status;
6800         oo = openowner(stp->st_stateowner);
6801         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
6802                 mutex_unlock(&stp->st_mutex);
6803                 nfs4_put_stid(&stp->st_stid);
6804                 return nfserr_bad_stateid;
6805         }
6806         *stpp = stp;
6807         return nfs_ok;
6808 }
6809
6810 __be32
6811 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6812                    union nfsd4_op_u *u)
6813 {
6814         struct nfsd4_open_confirm *oc = &u->open_confirm;
6815         __be32 status;
6816         struct nfs4_openowner *oo;
6817         struct nfs4_ol_stateid *stp;
6818         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6819
6820         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
6821                         cstate->current_fh.fh_dentry);
6822
6823         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
6824         if (status)
6825                 return status;
6826
6827         status = nfs4_preprocess_seqid_op(cstate,
6828                                         oc->oc_seqid, &oc->oc_req_stateid,
6829                                         NFS4_OPEN_STID, &stp, nn);
6830         if (status)
6831                 goto out;
6832         oo = openowner(stp->st_stateowner);
6833         status = nfserr_bad_stateid;
6834         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
6835                 mutex_unlock(&stp->st_mutex);
6836                 goto put_stateid;
6837         }
6838         oo->oo_flags |= NFS4_OO_CONFIRMED;
6839         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
6840         mutex_unlock(&stp->st_mutex);
6841         trace_nfsd_open_confirm(oc->oc_seqid, &stp->st_stid.sc_stateid);
6842         nfsd4_client_record_create(oo->oo_owner.so_client);
6843         status = nfs_ok;
6844 put_stateid:
6845         nfs4_put_stid(&stp->st_stid);
6846 out:
6847         nfsd4_bump_seqid(cstate, status);
6848         return status;
6849 }
6850
6851 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
6852 {
6853         if (!test_access(access, stp))
6854                 return;
6855         nfs4_file_put_access(stp->st_stid.sc_file, access);
6856         clear_access(access, stp);
6857 }
6858
6859 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
6860 {
6861         switch (to_access) {
6862         case NFS4_SHARE_ACCESS_READ:
6863                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
6864                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6865                 break;
6866         case NFS4_SHARE_ACCESS_WRITE:
6867                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
6868                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6869                 break;
6870         case NFS4_SHARE_ACCESS_BOTH:
6871                 break;
6872         default:
6873                 WARN_ON_ONCE(1);
6874         }
6875 }
6876
6877 __be32
6878 nfsd4_open_downgrade(struct svc_rqst *rqstp,
6879                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
6880 {
6881         struct nfsd4_open_downgrade *od = &u->open_downgrade;
6882         __be32 status;
6883         struct nfs4_ol_stateid *stp;
6884         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6885
6886         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
6887                         cstate->current_fh.fh_dentry);
6888
6889         /* We don't yet support WANT bits: */
6890         if (od->od_deleg_want)
6891                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
6892                         od->od_deleg_want);
6893
6894         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
6895                                         &od->od_stateid, &stp, nn);
6896         if (status)
6897                 goto out;
6898         status = nfserr_inval;
6899         if (!test_access(od->od_share_access, stp)) {
6900                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
6901                         stp->st_access_bmap, od->od_share_access);
6902                 goto put_stateid;
6903         }
6904         if (!test_deny(od->od_share_deny, stp)) {
6905                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
6906                         stp->st_deny_bmap, od->od_share_deny);
6907                 goto put_stateid;
6908         }
6909         nfs4_stateid_downgrade(stp, od->od_share_access);
6910         reset_union_bmap_deny(od->od_share_deny, stp);
6911         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
6912         status = nfs_ok;
6913 put_stateid:
6914         mutex_unlock(&stp->st_mutex);
6915         nfs4_put_stid(&stp->st_stid);
6916 out:
6917         nfsd4_bump_seqid(cstate, status);
6918         return status;
6919 }
6920
6921 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
6922 {
6923         struct nfs4_client *clp = s->st_stid.sc_client;
6924         bool unhashed;
6925         LIST_HEAD(reaplist);
6926         struct nfs4_ol_stateid *stp;
6927
6928         spin_lock(&clp->cl_lock);
6929         unhashed = unhash_open_stateid(s, &reaplist);
6930
6931         if (clp->cl_minorversion) {
6932                 if (unhashed)
6933                         put_ol_stateid_locked(s, &reaplist);
6934                 spin_unlock(&clp->cl_lock);
6935                 list_for_each_entry(stp, &reaplist, st_locks)
6936                         nfs4_free_cpntf_statelist(clp->net, &stp->st_stid);
6937                 free_ol_stateid_reaplist(&reaplist);
6938         } else {
6939                 spin_unlock(&clp->cl_lock);
6940                 free_ol_stateid_reaplist(&reaplist);
6941                 if (unhashed)
6942                         move_to_close_lru(s, clp->net);
6943         }
6944 }
6945
6946 /*
6947  * nfs4_unlock_state() called after encode
6948  */
6949 __be32
6950 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6951                 union nfsd4_op_u *u)
6952 {
6953         struct nfsd4_close *close = &u->close;
6954         __be32 status;
6955         struct nfs4_ol_stateid *stp;
6956         struct net *net = SVC_NET(rqstp);
6957         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6958
6959         dprintk("NFSD: nfsd4_close on file %pd\n",
6960                         cstate->current_fh.fh_dentry);
6961
6962         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
6963                                         &close->cl_stateid,
6964                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
6965                                         &stp, nn);
6966         nfsd4_bump_seqid(cstate, status);
6967         if (status)
6968                 goto out;
6969
6970         stp->st_stid.sc_type = NFS4_CLOSED_STID;
6971
6972         /*
6973          * Technically we don't _really_ have to increment or copy it, since
6974          * it should just be gone after this operation and we clobber the
6975          * copied value below, but we continue to do so here just to ensure
6976          * that racing ops see that there was a state change.
6977          */
6978         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
6979
6980         nfsd4_close_open_stateid(stp);
6981         mutex_unlock(&stp->st_mutex);
6982
6983         /* v4.1+ suggests that we send a special stateid in here, since the
6984          * clients should just ignore this anyway. Since this is not useful
6985          * for v4.0 clients either, we set it to the special close_stateid
6986          * universally.
6987          *
6988          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
6989          */
6990         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
6991
6992         /* put reference from nfs4_preprocess_seqid_op */
6993         nfs4_put_stid(&stp->st_stid);
6994 out:
6995         return status;
6996 }
6997
6998 __be32
6999 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7000                   union nfsd4_op_u *u)
7001 {
7002         struct nfsd4_delegreturn *dr = &u->delegreturn;
7003         struct nfs4_delegation *dp;
7004         stateid_t *stateid = &dr->dr_stateid;
7005         struct nfs4_stid *s;
7006         __be32 status;
7007         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7008
7009         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
7010                 return status;
7011
7012         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
7013         if (status)
7014                 goto out;
7015         dp = delegstateid(s);
7016         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
7017         if (status)
7018                 goto put_stateid;
7019
7020         trace_nfsd_deleg_return(stateid);
7021         wake_up_var(d_inode(cstate->current_fh.fh_dentry));
7022         destroy_delegation(dp);
7023 put_stateid:
7024         nfs4_put_stid(&dp->dl_stid);
7025 out:
7026         return status;
7027 }
7028
7029 /* last octet in a range */
7030 static inline u64
7031 last_byte_offset(u64 start, u64 len)
7032 {
7033         u64 end;
7034
7035         WARN_ON_ONCE(!len);
7036         end = start + len;
7037         return end > start ? end - 1: NFS4_MAX_UINT64;
7038 }
7039
7040 /*
7041  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
7042  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
7043  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
7044  * locking, this prevents us from being completely protocol-compliant.  The
7045  * real solution to this problem is to start using unsigned file offsets in
7046  * the VFS, but this is a very deep change!
7047  */
7048 static inline void
7049 nfs4_transform_lock_offset(struct file_lock *lock)
7050 {
7051         if (lock->fl_start < 0)
7052                 lock->fl_start = OFFSET_MAX;
7053         if (lock->fl_end < 0)
7054                 lock->fl_end = OFFSET_MAX;
7055 }
7056
7057 static fl_owner_t
7058 nfsd4_lm_get_owner(fl_owner_t owner)
7059 {
7060         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7061
7062         nfs4_get_stateowner(&lo->lo_owner);
7063         return owner;
7064 }
7065
7066 static void
7067 nfsd4_lm_put_owner(fl_owner_t owner)
7068 {
7069         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7070
7071         if (lo)
7072                 nfs4_put_stateowner(&lo->lo_owner);
7073 }
7074
7075 /* return pointer to struct nfs4_client if client is expirable */
7076 static bool
7077 nfsd4_lm_lock_expirable(struct file_lock *cfl)
7078 {
7079         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)cfl->fl_owner;
7080         struct nfs4_client *clp = lo->lo_owner.so_client;
7081         struct nfsd_net *nn;
7082
7083         if (try_to_expire_client(clp)) {
7084                 nn = net_generic(clp->net, nfsd_net_id);
7085                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
7086                 return true;
7087         }
7088         return false;
7089 }
7090
7091 /* schedule laundromat to run immediately and wait for it to complete */
7092 static void
7093 nfsd4_lm_expire_lock(void)
7094 {
7095         flush_workqueue(laundry_wq);
7096 }
7097
7098 static void
7099 nfsd4_lm_notify(struct file_lock *fl)
7100 {
7101         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
7102         struct net                      *net = lo->lo_owner.so_client->net;
7103         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
7104         struct nfsd4_blocked_lock       *nbl = container_of(fl,
7105                                                 struct nfsd4_blocked_lock, nbl_lock);
7106         bool queue = false;
7107
7108         /* An empty list means that something else is going to be using it */
7109         spin_lock(&nn->blocked_locks_lock);
7110         if (!list_empty(&nbl->nbl_list)) {
7111                 list_del_init(&nbl->nbl_list);
7112                 list_del_init(&nbl->nbl_lru);
7113                 queue = true;
7114         }
7115         spin_unlock(&nn->blocked_locks_lock);
7116
7117         if (queue) {
7118                 trace_nfsd_cb_notify_lock(lo, nbl);
7119                 nfsd4_run_cb(&nbl->nbl_cb);
7120         }
7121 }
7122
7123 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
7124         .lm_mod_owner = THIS_MODULE,
7125         .lm_notify = nfsd4_lm_notify,
7126         .lm_get_owner = nfsd4_lm_get_owner,
7127         .lm_put_owner = nfsd4_lm_put_owner,
7128         .lm_lock_expirable = nfsd4_lm_lock_expirable,
7129         .lm_expire_lock = nfsd4_lm_expire_lock,
7130 };
7131
7132 static inline void
7133 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
7134 {
7135         struct nfs4_lockowner *lo;
7136
7137         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
7138                 lo = (struct nfs4_lockowner *) fl->fl_owner;
7139                 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
7140                                                 GFP_KERNEL);
7141                 if (!deny->ld_owner.data)
7142                         /* We just don't care that much */
7143                         goto nevermind;
7144                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
7145         } else {
7146 nevermind:
7147                 deny->ld_owner.len = 0;
7148                 deny->ld_owner.data = NULL;
7149                 deny->ld_clientid.cl_boot = 0;
7150                 deny->ld_clientid.cl_id = 0;
7151         }
7152         deny->ld_start = fl->fl_start;
7153         deny->ld_length = NFS4_MAX_UINT64;
7154         if (fl->fl_end != NFS4_MAX_UINT64)
7155                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
7156         deny->ld_type = NFS4_READ_LT;
7157         if (fl->fl_type != F_RDLCK)
7158                 deny->ld_type = NFS4_WRITE_LT;
7159 }
7160
7161 static struct nfs4_lockowner *
7162 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
7163 {
7164         unsigned int strhashval = ownerstr_hashval(owner);
7165         struct nfs4_stateowner *so;
7166
7167         lockdep_assert_held(&clp->cl_lock);
7168
7169         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
7170                             so_strhash) {
7171                 if (so->so_is_open_owner)
7172                         continue;
7173                 if (same_owner_str(so, owner))
7174                         return lockowner(nfs4_get_stateowner(so));
7175         }
7176         return NULL;
7177 }
7178
7179 static struct nfs4_lockowner *
7180 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
7181 {
7182         struct nfs4_lockowner *lo;
7183
7184         spin_lock(&clp->cl_lock);
7185         lo = find_lockowner_str_locked(clp, owner);
7186         spin_unlock(&clp->cl_lock);
7187         return lo;
7188 }
7189
7190 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
7191 {
7192         unhash_lockowner_locked(lockowner(sop));
7193 }
7194
7195 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
7196 {
7197         struct nfs4_lockowner *lo = lockowner(sop);
7198
7199         kmem_cache_free(lockowner_slab, lo);
7200 }
7201
7202 static const struct nfs4_stateowner_operations lockowner_ops = {
7203         .so_unhash =    nfs4_unhash_lockowner,
7204         .so_free =      nfs4_free_lockowner,
7205 };
7206
7207 /*
7208  * Alloc a lock owner structure.
7209  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
7210  * occurred.
7211  *
7212  * strhashval = ownerstr_hashval
7213  */
7214 static struct nfs4_lockowner *
7215 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
7216                            struct nfs4_ol_stateid *open_stp,
7217                            struct nfsd4_lock *lock)
7218 {
7219         struct nfs4_lockowner *lo, *ret;
7220
7221         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
7222         if (!lo)
7223                 return NULL;
7224         INIT_LIST_HEAD(&lo->lo_blocked);
7225         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
7226         lo->lo_owner.so_is_open_owner = 0;
7227         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
7228         lo->lo_owner.so_ops = &lockowner_ops;
7229         spin_lock(&clp->cl_lock);
7230         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
7231         if (ret == NULL) {
7232                 list_add(&lo->lo_owner.so_strhash,
7233                          &clp->cl_ownerstr_hashtbl[strhashval]);
7234                 ret = lo;
7235         } else
7236                 nfs4_free_stateowner(&lo->lo_owner);
7237
7238         spin_unlock(&clp->cl_lock);
7239         return ret;
7240 }
7241
7242 static struct nfs4_ol_stateid *
7243 find_lock_stateid(const struct nfs4_lockowner *lo,
7244                   const struct nfs4_ol_stateid *ost)
7245 {
7246         struct nfs4_ol_stateid *lst;
7247
7248         lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
7249
7250         /* If ost is not hashed, ost->st_locks will not be valid */
7251         if (!nfs4_ol_stateid_unhashed(ost))
7252                 list_for_each_entry(lst, &ost->st_locks, st_locks) {
7253                         if (lst->st_stateowner == &lo->lo_owner) {
7254                                 refcount_inc(&lst->st_stid.sc_count);
7255                                 return lst;
7256                         }
7257                 }
7258         return NULL;
7259 }
7260
7261 static struct nfs4_ol_stateid *
7262 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
7263                   struct nfs4_file *fp, struct inode *inode,
7264                   struct nfs4_ol_stateid *open_stp)
7265 {
7266         struct nfs4_client *clp = lo->lo_owner.so_client;
7267         struct nfs4_ol_stateid *retstp;
7268
7269         mutex_init(&stp->st_mutex);
7270         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
7271 retry:
7272         spin_lock(&clp->cl_lock);
7273         if (nfs4_ol_stateid_unhashed(open_stp))
7274                 goto out_close;
7275         retstp = find_lock_stateid(lo, open_stp);
7276         if (retstp)
7277                 goto out_found;
7278         refcount_inc(&stp->st_stid.sc_count);
7279         stp->st_stid.sc_type = NFS4_LOCK_STID;
7280         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
7281         get_nfs4_file(fp);
7282         stp->st_stid.sc_file = fp;
7283         stp->st_access_bmap = 0;
7284         stp->st_deny_bmap = open_stp->st_deny_bmap;
7285         stp->st_openstp = open_stp;
7286         spin_lock(&fp->fi_lock);
7287         list_add(&stp->st_locks, &open_stp->st_locks);
7288         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
7289         list_add(&stp->st_perfile, &fp->fi_stateids);
7290         spin_unlock(&fp->fi_lock);
7291         spin_unlock(&clp->cl_lock);
7292         return stp;
7293 out_found:
7294         spin_unlock(&clp->cl_lock);
7295         if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
7296                 nfs4_put_stid(&retstp->st_stid);
7297                 goto retry;
7298         }
7299         /* To keep mutex tracking happy */
7300         mutex_unlock(&stp->st_mutex);
7301         return retstp;
7302 out_close:
7303         spin_unlock(&clp->cl_lock);
7304         mutex_unlock(&stp->st_mutex);
7305         return NULL;
7306 }
7307
7308 static struct nfs4_ol_stateid *
7309 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
7310                             struct inode *inode, struct nfs4_ol_stateid *ost,
7311                             bool *new)
7312 {
7313         struct nfs4_stid *ns = NULL;
7314         struct nfs4_ol_stateid *lst;
7315         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7316         struct nfs4_client *clp = oo->oo_owner.so_client;
7317
7318         *new = false;
7319         spin_lock(&clp->cl_lock);
7320         lst = find_lock_stateid(lo, ost);
7321         spin_unlock(&clp->cl_lock);
7322         if (lst != NULL) {
7323                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
7324                         goto out;
7325                 nfs4_put_stid(&lst->st_stid);
7326         }
7327         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
7328         if (ns == NULL)
7329                 return NULL;
7330
7331         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
7332         if (lst == openlockstateid(ns))
7333                 *new = true;
7334         else
7335                 nfs4_put_stid(ns);
7336 out:
7337         return lst;
7338 }
7339
7340 static int
7341 check_lock_length(u64 offset, u64 length)
7342 {
7343         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
7344                 (length > ~offset)));
7345 }
7346
7347 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
7348 {
7349         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
7350
7351         lockdep_assert_held(&fp->fi_lock);
7352
7353         if (test_access(access, lock_stp))
7354                 return;
7355         __nfs4_file_get_access(fp, access);
7356         set_access(access, lock_stp);
7357 }
7358
7359 static __be32
7360 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
7361                             struct nfs4_ol_stateid *ost,
7362                             struct nfsd4_lock *lock,
7363                             struct nfs4_ol_stateid **plst, bool *new)
7364 {
7365         __be32 status;
7366         struct nfs4_file *fi = ost->st_stid.sc_file;
7367         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7368         struct nfs4_client *cl = oo->oo_owner.so_client;
7369         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
7370         struct nfs4_lockowner *lo;
7371         struct nfs4_ol_stateid *lst;
7372         unsigned int strhashval;
7373
7374         lo = find_lockowner_str(cl, &lock->lk_new_owner);
7375         if (!lo) {
7376                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
7377                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
7378                 if (lo == NULL)
7379                         return nfserr_jukebox;
7380         } else {
7381                 /* with an existing lockowner, seqids must be the same */
7382                 status = nfserr_bad_seqid;
7383                 if (!cstate->minorversion &&
7384                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
7385                         goto out;
7386         }
7387
7388         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
7389         if (lst == NULL) {
7390                 status = nfserr_jukebox;
7391                 goto out;
7392         }
7393
7394         status = nfs_ok;
7395         *plst = lst;
7396 out:
7397         nfs4_put_stateowner(&lo->lo_owner);
7398         return status;
7399 }
7400
7401 /*
7402  *  LOCK operation
7403  */
7404 __be32
7405 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7406            union nfsd4_op_u *u)
7407 {
7408         struct nfsd4_lock *lock = &u->lock;
7409         struct nfs4_openowner *open_sop = NULL;
7410         struct nfs4_lockowner *lock_sop = NULL;
7411         struct nfs4_ol_stateid *lock_stp = NULL;
7412         struct nfs4_ol_stateid *open_stp = NULL;
7413         struct nfs4_file *fp;
7414         struct nfsd_file *nf = NULL;
7415         struct nfsd4_blocked_lock *nbl = NULL;
7416         struct file_lock *file_lock = NULL;
7417         struct file_lock *conflock = NULL;
7418         __be32 status = 0;
7419         int lkflg;
7420         int err;
7421         bool new = false;
7422         unsigned char fl_type;
7423         unsigned int fl_flags = FL_POSIX;
7424         struct net *net = SVC_NET(rqstp);
7425         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7426
7427         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
7428                 (long long) lock->lk_offset,
7429                 (long long) lock->lk_length);
7430
7431         if (check_lock_length(lock->lk_offset, lock->lk_length))
7432                  return nfserr_inval;
7433
7434         if ((status = fh_verify(rqstp, &cstate->current_fh,
7435                                 S_IFREG, NFSD_MAY_LOCK))) {
7436                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
7437                 return status;
7438         }
7439
7440         if (lock->lk_is_new) {
7441                 if (nfsd4_has_session(cstate))
7442                         /* See rfc 5661 18.10.3: given clientid is ignored: */
7443                         memcpy(&lock->lk_new_clientid,
7444                                 &cstate->clp->cl_clientid,
7445                                 sizeof(clientid_t));
7446
7447                 /* validate and update open stateid and open seqid */
7448                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
7449                                         lock->lk_new_open_seqid,
7450                                         &lock->lk_new_open_stateid,
7451                                         &open_stp, nn);
7452                 if (status)
7453                         goto out;
7454                 mutex_unlock(&open_stp->st_mutex);
7455                 open_sop = openowner(open_stp->st_stateowner);
7456                 status = nfserr_bad_stateid;
7457                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
7458                                                 &lock->lk_new_clientid))
7459                         goto out;
7460                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
7461                                                         &lock_stp, &new);
7462         } else {
7463                 status = nfs4_preprocess_seqid_op(cstate,
7464                                        lock->lk_old_lock_seqid,
7465                                        &lock->lk_old_lock_stateid,
7466                                        NFS4_LOCK_STID, &lock_stp, nn);
7467         }
7468         if (status)
7469                 goto out;
7470         lock_sop = lockowner(lock_stp->st_stateowner);
7471
7472         lkflg = setlkflg(lock->lk_type);
7473         status = nfs4_check_openmode(lock_stp, lkflg);
7474         if (status)
7475                 goto out;
7476
7477         status = nfserr_grace;
7478         if (locks_in_grace(net) && !lock->lk_reclaim)
7479                 goto out;
7480         status = nfserr_no_grace;
7481         if (!locks_in_grace(net) && lock->lk_reclaim)
7482                 goto out;
7483
7484         if (lock->lk_reclaim)
7485                 fl_flags |= FL_RECLAIM;
7486
7487         fp = lock_stp->st_stid.sc_file;
7488         switch (lock->lk_type) {
7489                 case NFS4_READW_LT:
7490                         if (nfsd4_has_session(cstate))
7491                                 fl_flags |= FL_SLEEP;
7492                         fallthrough;
7493                 case NFS4_READ_LT:
7494                         spin_lock(&fp->fi_lock);
7495                         nf = find_readable_file_locked(fp);
7496                         if (nf)
7497                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
7498                         spin_unlock(&fp->fi_lock);
7499                         fl_type = F_RDLCK;
7500                         break;
7501                 case NFS4_WRITEW_LT:
7502                         if (nfsd4_has_session(cstate))
7503                                 fl_flags |= FL_SLEEP;
7504                         fallthrough;
7505                 case NFS4_WRITE_LT:
7506                         spin_lock(&fp->fi_lock);
7507                         nf = find_writeable_file_locked(fp);
7508                         if (nf)
7509                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
7510                         spin_unlock(&fp->fi_lock);
7511                         fl_type = F_WRLCK;
7512                         break;
7513                 default:
7514                         status = nfserr_inval;
7515                 goto out;
7516         }
7517
7518         if (!nf) {
7519                 status = nfserr_openmode;
7520                 goto out;
7521         }
7522
7523         /*
7524          * Most filesystems with their own ->lock operations will block
7525          * the nfsd thread waiting to acquire the lock.  That leads to
7526          * deadlocks (we don't want every nfsd thread tied up waiting
7527          * for file locks), so don't attempt blocking lock notifications
7528          * on those filesystems:
7529          */
7530         if (nf->nf_file->f_op->lock)
7531                 fl_flags &= ~FL_SLEEP;
7532
7533         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
7534         if (!nbl) {
7535                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
7536                 status = nfserr_jukebox;
7537                 goto out;
7538         }
7539
7540         file_lock = &nbl->nbl_lock;
7541         file_lock->fl_type = fl_type;
7542         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
7543         file_lock->fl_pid = current->tgid;
7544         file_lock->fl_file = nf->nf_file;
7545         file_lock->fl_flags = fl_flags;
7546         file_lock->fl_lmops = &nfsd_posix_mng_ops;
7547         file_lock->fl_start = lock->lk_offset;
7548         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
7549         nfs4_transform_lock_offset(file_lock);
7550
7551         conflock = locks_alloc_lock();
7552         if (!conflock) {
7553                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7554                 status = nfserr_jukebox;
7555                 goto out;
7556         }
7557
7558         if (fl_flags & FL_SLEEP) {
7559                 nbl->nbl_time = ktime_get_boottime_seconds();
7560                 spin_lock(&nn->blocked_locks_lock);
7561                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
7562                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
7563                 kref_get(&nbl->nbl_kref);
7564                 spin_unlock(&nn->blocked_locks_lock);
7565         }
7566
7567         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
7568         switch (err) {
7569         case 0: /* success! */
7570                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
7571                 status = 0;
7572                 if (lock->lk_reclaim)
7573                         nn->somebody_reclaimed = true;
7574                 break;
7575         case FILE_LOCK_DEFERRED:
7576                 kref_put(&nbl->nbl_kref, free_nbl);
7577                 nbl = NULL;
7578                 fallthrough;
7579         case -EAGAIN:           /* conflock holds conflicting lock */
7580                 status = nfserr_denied;
7581                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
7582                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
7583                 break;
7584         case -EDEADLK:
7585                 status = nfserr_deadlock;
7586                 break;
7587         default:
7588                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
7589                 status = nfserrno(err);
7590                 break;
7591         }
7592 out:
7593         if (nbl) {
7594                 /* dequeue it if we queued it before */
7595                 if (fl_flags & FL_SLEEP) {
7596                         spin_lock(&nn->blocked_locks_lock);
7597                         if (!list_empty(&nbl->nbl_list) &&
7598                             !list_empty(&nbl->nbl_lru)) {
7599                                 list_del_init(&nbl->nbl_list);
7600                                 list_del_init(&nbl->nbl_lru);
7601                                 kref_put(&nbl->nbl_kref, free_nbl);
7602                         }
7603                         /* nbl can use one of lists to be linked to reaplist */
7604                         spin_unlock(&nn->blocked_locks_lock);
7605                 }
7606                 free_blocked_lock(nbl);
7607         }
7608         if (nf)
7609                 nfsd_file_put(nf);
7610         if (lock_stp) {
7611                 /* Bump seqid manually if the 4.0 replay owner is openowner */
7612                 if (cstate->replay_owner &&
7613                     cstate->replay_owner != &lock_sop->lo_owner &&
7614                     seqid_mutating_err(ntohl(status)))
7615                         lock_sop->lo_owner.so_seqid++;
7616
7617                 /*
7618                  * If this is a new, never-before-used stateid, and we are
7619                  * returning an error, then just go ahead and release it.
7620                  */
7621                 if (status && new)
7622                         release_lock_stateid(lock_stp);
7623
7624                 mutex_unlock(&lock_stp->st_mutex);
7625
7626                 nfs4_put_stid(&lock_stp->st_stid);
7627         }
7628         if (open_stp)
7629                 nfs4_put_stid(&open_stp->st_stid);
7630         nfsd4_bump_seqid(cstate, status);
7631         if (conflock)
7632                 locks_free_lock(conflock);
7633         return status;
7634 }
7635
7636 /*
7637  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
7638  * so we do a temporary open here just to get an open file to pass to
7639  * vfs_test_lock.
7640  */
7641 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
7642 {
7643         struct nfsd_file *nf;
7644         struct inode *inode;
7645         __be32 err;
7646
7647         err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
7648         if (err)
7649                 return err;
7650         inode = fhp->fh_dentry->d_inode;
7651         inode_lock(inode); /* to block new leases till after test_lock: */
7652         err = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
7653         if (err)
7654                 goto out;
7655         lock->fl_file = nf->nf_file;
7656         err = nfserrno(vfs_test_lock(nf->nf_file, lock));
7657         lock->fl_file = NULL;
7658 out:
7659         inode_unlock(inode);
7660         nfsd_file_put(nf);
7661         return err;
7662 }
7663
7664 /*
7665  * LOCKT operation
7666  */
7667 __be32
7668 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7669             union nfsd4_op_u *u)
7670 {
7671         struct nfsd4_lockt *lockt = &u->lockt;
7672         struct file_lock *file_lock = NULL;
7673         struct nfs4_lockowner *lo = NULL;
7674         __be32 status;
7675         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7676
7677         if (locks_in_grace(SVC_NET(rqstp)))
7678                 return nfserr_grace;
7679
7680         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
7681                  return nfserr_inval;
7682
7683         if (!nfsd4_has_session(cstate)) {
7684                 status = set_client(&lockt->lt_clientid, cstate, nn);
7685                 if (status)
7686                         goto out;
7687         }
7688
7689         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
7690                 goto out;
7691
7692         file_lock = locks_alloc_lock();
7693         if (!file_lock) {
7694                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7695                 status = nfserr_jukebox;
7696                 goto out;
7697         }
7698
7699         switch (lockt->lt_type) {
7700                 case NFS4_READ_LT:
7701                 case NFS4_READW_LT:
7702                         file_lock->fl_type = F_RDLCK;
7703                         break;
7704                 case NFS4_WRITE_LT:
7705                 case NFS4_WRITEW_LT:
7706                         file_lock->fl_type = F_WRLCK;
7707                         break;
7708                 default:
7709                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
7710                         status = nfserr_inval;
7711                         goto out;
7712         }
7713
7714         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
7715         if (lo)
7716                 file_lock->fl_owner = (fl_owner_t)lo;
7717         file_lock->fl_pid = current->tgid;
7718         file_lock->fl_flags = FL_POSIX;
7719
7720         file_lock->fl_start = lockt->lt_offset;
7721         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
7722
7723         nfs4_transform_lock_offset(file_lock);
7724
7725         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
7726         if (status)
7727                 goto out;
7728
7729         if (file_lock->fl_type != F_UNLCK) {
7730                 status = nfserr_denied;
7731                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
7732         }
7733 out:
7734         if (lo)
7735                 nfs4_put_stateowner(&lo->lo_owner);
7736         if (file_lock)
7737                 locks_free_lock(file_lock);
7738         return status;
7739 }
7740
7741 __be32
7742 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7743             union nfsd4_op_u *u)
7744 {
7745         struct nfsd4_locku *locku = &u->locku;
7746         struct nfs4_ol_stateid *stp;
7747         struct nfsd_file *nf = NULL;
7748         struct file_lock *file_lock = NULL;
7749         __be32 status;
7750         int err;
7751         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7752
7753         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
7754                 (long long) locku->lu_offset,
7755                 (long long) locku->lu_length);
7756
7757         if (check_lock_length(locku->lu_offset, locku->lu_length))
7758                  return nfserr_inval;
7759
7760         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
7761                                         &locku->lu_stateid, NFS4_LOCK_STID,
7762                                         &stp, nn);
7763         if (status)
7764                 goto out;
7765         nf = find_any_file(stp->st_stid.sc_file);
7766         if (!nf) {
7767                 status = nfserr_lock_range;
7768                 goto put_stateid;
7769         }
7770         file_lock = locks_alloc_lock();
7771         if (!file_lock) {
7772                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7773                 status = nfserr_jukebox;
7774                 goto put_file;
7775         }
7776
7777         file_lock->fl_type = F_UNLCK;
7778         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
7779         file_lock->fl_pid = current->tgid;
7780         file_lock->fl_file = nf->nf_file;
7781         file_lock->fl_flags = FL_POSIX;
7782         file_lock->fl_lmops = &nfsd_posix_mng_ops;
7783         file_lock->fl_start = locku->lu_offset;
7784
7785         file_lock->fl_end = last_byte_offset(locku->lu_offset,
7786                                                 locku->lu_length);
7787         nfs4_transform_lock_offset(file_lock);
7788
7789         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
7790         if (err) {
7791                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
7792                 goto out_nfserr;
7793         }
7794         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
7795 put_file:
7796         nfsd_file_put(nf);
7797 put_stateid:
7798         mutex_unlock(&stp->st_mutex);
7799         nfs4_put_stid(&stp->st_stid);
7800 out:
7801         nfsd4_bump_seqid(cstate, status);
7802         if (file_lock)
7803                 locks_free_lock(file_lock);
7804         return status;
7805
7806 out_nfserr:
7807         status = nfserrno(err);
7808         goto put_file;
7809 }
7810
7811 /*
7812  * returns
7813  *      true:  locks held by lockowner
7814  *      false: no locks held by lockowner
7815  */
7816 static bool
7817 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
7818 {
7819         struct file_lock *fl;
7820         int status = false;
7821         struct nfsd_file *nf = find_any_file(fp);
7822         struct inode *inode;
7823         struct file_lock_context *flctx;
7824
7825         if (!nf) {
7826                 /* Any valid lock stateid should have some sort of access */
7827                 WARN_ON_ONCE(1);
7828                 return status;
7829         }
7830
7831         inode = locks_inode(nf->nf_file);
7832         flctx = locks_inode_context(inode);
7833
7834         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
7835                 spin_lock(&flctx->flc_lock);
7836                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
7837                         if (fl->fl_owner == (fl_owner_t)lowner) {
7838                                 status = true;
7839                                 break;
7840                         }
7841                 }
7842                 spin_unlock(&flctx->flc_lock);
7843         }
7844         nfsd_file_put(nf);
7845         return status;
7846 }
7847
7848 /**
7849  * nfsd4_release_lockowner - process NFSv4.0 RELEASE_LOCKOWNER operations
7850  * @rqstp: RPC transaction
7851  * @cstate: NFSv4 COMPOUND state
7852  * @u: RELEASE_LOCKOWNER arguments
7853  *
7854  * The lockowner's so_count is bumped when a lock record is added
7855  * or when copying a conflicting lock. The latter case is brief,
7856  * but can lead to fleeting false positives when looking for
7857  * locks-in-use.
7858  *
7859  * Return values:
7860  *   %nfs_ok: lockowner released or not found
7861  *   %nfserr_locks_held: lockowner still in use
7862  *   %nfserr_stale_clientid: clientid no longer active
7863  *   %nfserr_expired: clientid not recognized
7864  */
7865 __be32
7866 nfsd4_release_lockowner(struct svc_rqst *rqstp,
7867                         struct nfsd4_compound_state *cstate,
7868                         union nfsd4_op_u *u)
7869 {
7870         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
7871         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7872         clientid_t *clid = &rlockowner->rl_clientid;
7873         struct nfs4_ol_stateid *stp;
7874         struct nfs4_lockowner *lo;
7875         struct nfs4_client *clp;
7876         LIST_HEAD(reaplist);
7877         __be32 status;
7878
7879         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
7880                 clid->cl_boot, clid->cl_id);
7881
7882         status = set_client(clid, cstate, nn);
7883         if (status)
7884                 return status;
7885         clp = cstate->clp;
7886
7887         spin_lock(&clp->cl_lock);
7888         lo = find_lockowner_str_locked(clp, &rlockowner->rl_owner);
7889         if (!lo) {
7890                 spin_unlock(&clp->cl_lock);
7891                 return nfs_ok;
7892         }
7893         if (atomic_read(&lo->lo_owner.so_count) != 2) {
7894                 spin_unlock(&clp->cl_lock);
7895                 nfs4_put_stateowner(&lo->lo_owner);
7896                 return nfserr_locks_held;
7897         }
7898         unhash_lockowner_locked(lo);
7899         while (!list_empty(&lo->lo_owner.so_stateids)) {
7900                 stp = list_first_entry(&lo->lo_owner.so_stateids,
7901                                        struct nfs4_ol_stateid,
7902                                        st_perstateowner);
7903                 WARN_ON(!unhash_lock_stateid(stp));
7904                 put_ol_stateid_locked(stp, &reaplist);
7905         }
7906         spin_unlock(&clp->cl_lock);
7907
7908         free_ol_stateid_reaplist(&reaplist);
7909         remove_blocked_locks(lo);
7910         nfs4_put_stateowner(&lo->lo_owner);
7911         return nfs_ok;
7912 }
7913
7914 static inline struct nfs4_client_reclaim *
7915 alloc_reclaim(void)
7916 {
7917         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
7918 }
7919
7920 bool
7921 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
7922 {
7923         struct nfs4_client_reclaim *crp;
7924
7925         crp = nfsd4_find_reclaim_client(name, nn);
7926         return (crp && crp->cr_clp);
7927 }
7928
7929 /*
7930  * failure => all reset bets are off, nfserr_no_grace...
7931  *
7932  * The caller is responsible for freeing name.data if NULL is returned (it
7933  * will be freed in nfs4_remove_reclaim_record in the normal case).
7934  */
7935 struct nfs4_client_reclaim *
7936 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
7937                 struct nfsd_net *nn)
7938 {
7939         unsigned int strhashval;
7940         struct nfs4_client_reclaim *crp;
7941
7942         crp = alloc_reclaim();
7943         if (crp) {
7944                 strhashval = clientstr_hashval(name);
7945                 INIT_LIST_HEAD(&crp->cr_strhash);
7946                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
7947                 crp->cr_name.data = name.data;
7948                 crp->cr_name.len = name.len;
7949                 crp->cr_princhash.data = princhash.data;
7950                 crp->cr_princhash.len = princhash.len;
7951                 crp->cr_clp = NULL;
7952                 nn->reclaim_str_hashtbl_size++;
7953         }
7954         return crp;
7955 }
7956
7957 void
7958 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
7959 {
7960         list_del(&crp->cr_strhash);
7961         kfree(crp->cr_name.data);
7962         kfree(crp->cr_princhash.data);
7963         kfree(crp);
7964         nn->reclaim_str_hashtbl_size--;
7965 }
7966
7967 void
7968 nfs4_release_reclaim(struct nfsd_net *nn)
7969 {
7970         struct nfs4_client_reclaim *crp = NULL;
7971         int i;
7972
7973         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7974                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
7975                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
7976                                         struct nfs4_client_reclaim, cr_strhash);
7977                         nfs4_remove_reclaim_record(crp, nn);
7978                 }
7979         }
7980         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
7981 }
7982
7983 /*
7984  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
7985 struct nfs4_client_reclaim *
7986 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
7987 {
7988         unsigned int strhashval;
7989         struct nfs4_client_reclaim *crp = NULL;
7990
7991         strhashval = clientstr_hashval(name);
7992         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
7993                 if (compare_blob(&crp->cr_name, &name) == 0) {
7994                         return crp;
7995                 }
7996         }
7997         return NULL;
7998 }
7999
8000 __be32
8001 nfs4_check_open_reclaim(struct nfs4_client *clp)
8002 {
8003         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
8004                 return nfserr_no_grace;
8005
8006         if (nfsd4_client_record_check(clp))
8007                 return nfserr_reclaim_bad;
8008
8009         return nfs_ok;
8010 }
8011
8012 /*
8013  * Since the lifetime of a delegation isn't limited to that of an open, a
8014  * client may quite reasonably hang on to a delegation as long as it has
8015  * the inode cached.  This becomes an obvious problem the first time a
8016  * client's inode cache approaches the size of the server's total memory.
8017  *
8018  * For now we avoid this problem by imposing a hard limit on the number
8019  * of delegations, which varies according to the server's memory size.
8020  */
8021 static void
8022 set_max_delegations(void)
8023 {
8024         /*
8025          * Allow at most 4 delegations per megabyte of RAM.  Quick
8026          * estimates suggest that in the worst case (where every delegation
8027          * is for a different inode), a delegation could take about 1.5K,
8028          * giving a worst case usage of about 6% of memory.
8029          */
8030         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
8031 }
8032
8033 static int nfs4_state_create_net(struct net *net)
8034 {
8035         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8036         int i;
8037
8038         nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8039                                             sizeof(struct list_head),
8040                                             GFP_KERNEL);
8041         if (!nn->conf_id_hashtbl)
8042                 goto err;
8043         nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8044                                               sizeof(struct list_head),
8045                                               GFP_KERNEL);
8046         if (!nn->unconf_id_hashtbl)
8047                 goto err_unconf_id;
8048         nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
8049                                               sizeof(struct list_head),
8050                                               GFP_KERNEL);
8051         if (!nn->sessionid_hashtbl)
8052                 goto err_sessionid;
8053
8054         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8055                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
8056                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
8057         }
8058         for (i = 0; i < SESSION_HASH_SIZE; i++)
8059                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
8060         nn->conf_name_tree = RB_ROOT;
8061         nn->unconf_name_tree = RB_ROOT;
8062         nn->boot_time = ktime_get_real_seconds();
8063         nn->grace_ended = false;
8064         nn->nfsd4_manager.block_opens = true;
8065         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
8066         INIT_LIST_HEAD(&nn->client_lru);
8067         INIT_LIST_HEAD(&nn->close_lru);
8068         INIT_LIST_HEAD(&nn->del_recall_lru);
8069         spin_lock_init(&nn->client_lock);
8070         spin_lock_init(&nn->s2s_cp_lock);
8071         idr_init(&nn->s2s_cp_stateids);
8072
8073         spin_lock_init(&nn->blocked_locks_lock);
8074         INIT_LIST_HEAD(&nn->blocked_locks_lru);
8075
8076         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
8077         INIT_DELAYED_WORK(&nn->nfsd_shrinker_work, nfsd4_state_shrinker_worker);
8078         get_net(net);
8079
8080         return 0;
8081
8082 err_sessionid:
8083         kfree(nn->unconf_id_hashtbl);
8084 err_unconf_id:
8085         kfree(nn->conf_id_hashtbl);
8086 err:
8087         return -ENOMEM;
8088 }
8089
8090 static void
8091 nfs4_state_destroy_net(struct net *net)
8092 {
8093         int i;
8094         struct nfs4_client *clp = NULL;
8095         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8096
8097         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8098                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
8099                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8100                         destroy_client(clp);
8101                 }
8102         }
8103
8104         WARN_ON(!list_empty(&nn->blocked_locks_lru));
8105
8106         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8107                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
8108                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8109                         destroy_client(clp);
8110                 }
8111         }
8112
8113         kfree(nn->sessionid_hashtbl);
8114         kfree(nn->unconf_id_hashtbl);
8115         kfree(nn->conf_id_hashtbl);
8116         put_net(net);
8117 }
8118
8119 int
8120 nfs4_state_start_net(struct net *net)
8121 {
8122         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8123         int ret;
8124
8125         ret = nfs4_state_create_net(net);
8126         if (ret)
8127                 return ret;
8128         locks_start_grace(net, &nn->nfsd4_manager);
8129         nfsd4_client_tracking_init(net);
8130         if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
8131                 goto skip_grace;
8132         printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
8133                nn->nfsd4_grace, net->ns.inum);
8134         trace_nfsd_grace_start(nn);
8135         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
8136         return 0;
8137
8138 skip_grace:
8139         printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
8140                         net->ns.inum);
8141         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
8142         nfsd4_end_grace(nn);
8143         return 0;
8144 }
8145
8146 /* initialization to perform when the nfsd service is started: */
8147
8148 int
8149 nfs4_state_start(void)
8150 {
8151         int ret;
8152
8153         ret = rhltable_init(&nfs4_file_rhltable, &nfs4_file_rhash_params);
8154         if (ret)
8155                 return ret;
8156
8157         ret = nfsd4_create_callback_queue();
8158         if (ret) {
8159                 rhltable_destroy(&nfs4_file_rhltable);
8160                 return ret;
8161         }
8162
8163         set_max_delegations();
8164         return 0;
8165 }
8166
8167 void
8168 nfs4_state_shutdown_net(struct net *net)
8169 {
8170         struct nfs4_delegation *dp = NULL;
8171         struct list_head *pos, *next, reaplist;
8172         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8173
8174         cancel_delayed_work_sync(&nn->laundromat_work);
8175         locks_end_grace(&nn->nfsd4_manager);
8176
8177         INIT_LIST_HEAD(&reaplist);
8178         spin_lock(&state_lock);
8179         list_for_each_safe(pos, next, &nn->del_recall_lru) {
8180                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8181                 WARN_ON(!unhash_delegation_locked(dp));
8182                 list_add(&dp->dl_recall_lru, &reaplist);
8183         }
8184         spin_unlock(&state_lock);
8185         list_for_each_safe(pos, next, &reaplist) {
8186                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8187                 list_del_init(&dp->dl_recall_lru);
8188                 destroy_unhashed_deleg(dp);
8189         }
8190
8191         nfsd4_client_tracking_exit(net);
8192         nfs4_state_destroy_net(net);
8193         rhltable_destroy(&nfs4_file_rhltable);
8194 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
8195         nfsd4_ssc_shutdown_umount(nn);
8196 #endif
8197 }
8198
8199 void
8200 nfs4_state_shutdown(void)
8201 {
8202         nfsd4_destroy_callback_queue();
8203 }
8204
8205 static void
8206 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8207 {
8208         if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
8209             CURRENT_STATEID(stateid))
8210                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
8211 }
8212
8213 static void
8214 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8215 {
8216         if (cstate->minorversion) {
8217                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
8218                 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8219         }
8220 }
8221
8222 void
8223 clear_current_stateid(struct nfsd4_compound_state *cstate)
8224 {
8225         CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8226 }
8227
8228 /*
8229  * functions to set current state id
8230  */
8231 void
8232 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
8233                 union nfsd4_op_u *u)
8234 {
8235         put_stateid(cstate, &u->open_downgrade.od_stateid);
8236 }
8237
8238 void
8239 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
8240                 union nfsd4_op_u *u)
8241 {
8242         put_stateid(cstate, &u->open.op_stateid);
8243 }
8244
8245 void
8246 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
8247                 union nfsd4_op_u *u)
8248 {
8249         put_stateid(cstate, &u->close.cl_stateid);
8250 }
8251
8252 void
8253 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
8254                 union nfsd4_op_u *u)
8255 {
8256         put_stateid(cstate, &u->lock.lk_resp_stateid);
8257 }
8258
8259 /*
8260  * functions to consume current state id
8261  */
8262
8263 void
8264 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
8265                 union nfsd4_op_u *u)
8266 {
8267         get_stateid(cstate, &u->open_downgrade.od_stateid);
8268 }
8269
8270 void
8271 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
8272                 union nfsd4_op_u *u)
8273 {
8274         get_stateid(cstate, &u->delegreturn.dr_stateid);
8275 }
8276
8277 void
8278 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
8279                 union nfsd4_op_u *u)
8280 {
8281         get_stateid(cstate, &u->free_stateid.fr_stateid);
8282 }
8283
8284 void
8285 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
8286                 union nfsd4_op_u *u)
8287 {
8288         get_stateid(cstate, &u->setattr.sa_stateid);
8289 }
8290
8291 void
8292 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
8293                 union nfsd4_op_u *u)
8294 {
8295         get_stateid(cstate, &u->close.cl_stateid);
8296 }
8297
8298 void
8299 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
8300                 union nfsd4_op_u *u)
8301 {
8302         get_stateid(cstate, &u->locku.lu_stateid);
8303 }
8304
8305 void
8306 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
8307                 union nfsd4_op_u *u)
8308 {
8309         get_stateid(cstate, &u->read.rd_stateid);
8310 }
8311
8312 void
8313 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
8314                 union nfsd4_op_u *u)
8315 {
8316         get_stateid(cstate, &u->write.wr_stateid);
8317 }