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/nfs_ssc.h>
  48 #include "xdr4.h"
  49 #include "xdr4cb.h"
  50 #include "vfs.h"
  51 #include "current_stateid.h"
  52
  53 #include "netns.h"
  54 #include "pnfs.h"
  55 #include "filecache.h"
  56 #include "trace.h"
  57
  58 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  59
  60 #define all_ones {{~0,~0},~0}
  61 static const stateid_t one_stateid = {
  62         .si_generation = ~0,
  63         .si_opaque = all_ones,
  64 };
  65 static const stateid_t zero_stateid = {
  66         /* all fields zero */
  67 };
  68 static const stateid_t currentstateid = {
  69         .si_generation = 1,
  70 };
  71 static const stateid_t close_stateid = {
  72         .si_generation = 0xffffffffU,
  73 };
  74
  75 static u64 current_sessionid = 1;
  76
  77 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  78 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  79 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  80 #define CLOSE_STATEID(stateid)  (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
  81
  82 /* forward declarations */
  83 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  84 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  85 void nfsd4_end_grace(struct nfsd_net *nn);
  86 static void _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps);
  87
  88 /* Locking: */
  89
  90 /*
  91  * Currently used for the del_recall_lru and file hash table.  In an
  92  * effort to decrease the scope of the client_mutex, this spinlock may
  93  * eventually cover more:
  94  */
  95 static DEFINE_SPINLOCK(state_lock);
  96
  97 enum nfsd4_st_mutex_lock_subclass {
  98         OPEN_STATEID_MUTEX = 0,
  99         LOCK_STATEID_MUTEX = 1,
 100 };
 101
 102 /*
 103  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
 104  * the refcount on the open stateid to drop.
 105  */
 106 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
 107
 108 /*
 109  * A waitqueue where a writer to clients/#/ctl destroying a client can
 110  * wait for cl_rpc_users to drop to 0 and then for the client to be
 111  * unhashed.
 112  */
 113 static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
 114
 115 static struct kmem_cache *client_slab;
 116 static struct kmem_cache *openowner_slab;
 117 static struct kmem_cache *lockowner_slab;
 118 static struct kmem_cache *file_slab;
 119 static struct kmem_cache *stateid_slab;
 120 static struct kmem_cache *deleg_slab;
 121 static struct kmem_cache *odstate_slab;
 122
 123 static void free_session(struct nfsd4_session *);
 124
 125 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 126 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
 127
 128 static struct workqueue_struct *laundry_wq;
 129
 130 int nfsd4_create_laundry_wq(void)
 131 {
 132         int rc = 0;
 133
 134         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
 135         if (laundry_wq == NULL)
 136                 rc = -ENOMEM;
 137         return rc;
 138 }
 139
 140 void nfsd4_destroy_laundry_wq(void)
 141 {
 142         destroy_workqueue(laundry_wq);
 143 }
 144
 145 static bool is_session_dead(struct nfsd4_session *ses)
 146 {
 147         return ses->se_flags & NFS4_SESSION_DEAD;
 148 }
 149
 150 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 151 {
 152         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 153                 return nfserr_jukebox;
 154         ses->se_flags |= NFS4_SESSION_DEAD;
 155         return nfs_ok;
 156 }
 157
 158 static bool is_client_expired(struct nfs4_client *clp)
 159 {
 160         return clp->cl_time == 0;
 161 }
 162
 163 static void nfsd4_dec_courtesy_client_count(struct nfsd_net *nn,
 164                                         struct nfs4_client *clp)
 165 {
 166         if (clp->cl_state != NFSD4_ACTIVE)
 167                 atomic_add_unless(&nn->nfsd_courtesy_clients, -1, 0);
 168 }
 169
 170 static __be32 get_client_locked(struct nfs4_client *clp)
 171 {
 172         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 173
 174         lockdep_assert_held(&nn->client_lock);
 175
 176         if (is_client_expired(clp))
 177                 return nfserr_expired;
 178         atomic_inc(&clp->cl_rpc_users);
 179         nfsd4_dec_courtesy_client_count(nn, clp);
 180         clp->cl_state = NFSD4_ACTIVE;
 181         return nfs_ok;
 182 }
 183
 184 /* must be called under the client_lock */
 185 static inline void
 186 renew_client_locked(struct nfs4_client *clp)
 187 {
 188         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 189
 190         if (is_client_expired(clp)) {
 191                 WARN_ON(1);
 192                 printk("%s: client (clientid %08x/%08x) already expired\n",
 193                         __func__,
 194                         clp->cl_clientid.cl_boot,
 195                         clp->cl_clientid.cl_id);
 196                 return;
 197         }
 198
 199         list_move_tail(&clp->cl_lru, &nn->client_lru);
 200         clp->cl_time = ktime_get_boottime_seconds();
 201         nfsd4_dec_courtesy_client_count(nn, clp);
 202         clp->cl_state = NFSD4_ACTIVE;
 203 }
 204
 205 static void put_client_renew_locked(struct nfs4_client *clp)
 206 {
 207         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 208
 209         lockdep_assert_held(&nn->client_lock);
 210
 211         if (!atomic_dec_and_test(&clp->cl_rpc_users))
 212                 return;
 213         if (!is_client_expired(clp))
 214                 renew_client_locked(clp);
 215         else
 216                 wake_up_all(&expiry_wq);
 217 }
 218
 219 static void put_client_renew(struct nfs4_client *clp)
 220 {
 221         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 222
 223         if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
 224                 return;
 225         if (!is_client_expired(clp))
 226                 renew_client_locked(clp);
 227         else
 228                 wake_up_all(&expiry_wq);
 229         spin_unlock(&nn->client_lock);
 230 }
 231
 232 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 233 {
 234         __be32 status;
 235
 236         if (is_session_dead(ses))
 237                 return nfserr_badsession;
 238         status = get_client_locked(ses->se_client);
 239         if (status)
 240                 return status;
 241         atomic_inc(&ses->se_ref);
 242         return nfs_ok;
 243 }
 244
 245 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 246 {
 247         struct nfs4_client *clp = ses->se_client;
 248         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 249
 250         lockdep_assert_held(&nn->client_lock);
 251
 252         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 253                 free_session(ses);
 254         put_client_renew_locked(clp);
 255 }
 256
 257 static void nfsd4_put_session(struct nfsd4_session *ses)
 258 {
 259         struct nfs4_client *clp = ses->se_client;
 260         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 261
 262         spin_lock(&nn->client_lock);
 263         nfsd4_put_session_locked(ses);
 264         spin_unlock(&nn->client_lock);
 265 }
 266
 267 static struct nfsd4_blocked_lock *
 268 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 269                         struct nfsd_net *nn)
 270 {
 271         struct nfsd4_blocked_lock *cur, *found = NULL;
 272
 273         spin_lock(&nn->blocked_locks_lock);
 274         list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
 275                 if (fh_match(fh, &cur->nbl_fh)) {
 276                         list_del_init(&cur->nbl_list);
 277                         WARN_ON(list_empty(&cur->nbl_lru));
 278                         list_del_init(&cur->nbl_lru);
 279                         found = cur;
 280                         break;
 281                 }
 282         }
 283         spin_unlock(&nn->blocked_locks_lock);
 284         if (found)
 285                 locks_delete_block(&found->nbl_lock);
 286         return found;
 287 }
 288
 289 static struct nfsd4_blocked_lock *
 290 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 291                         struct nfsd_net *nn)
 292 {
 293         struct nfsd4_blocked_lock *nbl;
 294
 295         nbl = find_blocked_lock(lo, fh, nn);
 296         if (!nbl) {
 297                 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
 298                 if (nbl) {
 299                         INIT_LIST_HEAD(&nbl->nbl_list);
 300                         INIT_LIST_HEAD(&nbl->nbl_lru);
 301                         fh_copy_shallow(&nbl->nbl_fh, fh);
 302                         locks_init_lock(&nbl->nbl_lock);
 303                         kref_init(&nbl->nbl_kref);
 304                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
 305                                         &nfsd4_cb_notify_lock_ops,
 306                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
 307                 }
 308         }
 309         return nbl;
 310 }
 311
 312 static void
 313 free_nbl(struct kref *kref)
 314 {
 315         struct nfsd4_blocked_lock *nbl;
 316
 317         nbl = container_of(kref, struct nfsd4_blocked_lock, nbl_kref);
 318         kfree(nbl);
 319 }
 320
 321 static void
 322 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
 323 {
 324         locks_delete_block(&nbl->nbl_lock);
 325         locks_release_private(&nbl->nbl_lock);
 326         kref_put(&nbl->nbl_kref, free_nbl);
 327 }
 328
 329 static void
 330 remove_blocked_locks(struct nfs4_lockowner *lo)
 331 {
 332         struct nfs4_client *clp = lo->lo_owner.so_client;
 333         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 334         struct nfsd4_blocked_lock *nbl;
 335         LIST_HEAD(reaplist);
 336
 337         /* Dequeue all blocked locks */
 338         spin_lock(&nn->blocked_locks_lock);
 339         while (!list_empty(&lo->lo_blocked)) {
 340                 nbl = list_first_entry(&lo->lo_blocked,
 341                                         struct nfsd4_blocked_lock,
 342                                         nbl_list);
 343                 list_del_init(&nbl->nbl_list);
 344                 WARN_ON(list_empty(&nbl->nbl_lru));
 345                 list_move(&nbl->nbl_lru, &reaplist);
 346         }
 347         spin_unlock(&nn->blocked_locks_lock);
 348
 349         /* Now free them */
 350         while (!list_empty(&reaplist)) {
 351                 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
 352                                         nbl_lru);
 353                 list_del_init(&nbl->nbl_lru);
 354                 free_blocked_lock(nbl);
 355         }
 356 }
 357
 358 static void
 359 nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
 360 {
 361         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 362                                                 struct nfsd4_blocked_lock, nbl_cb);
 363         locks_delete_block(&nbl->nbl_lock);
 364 }
 365
 366 static int
 367 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
 368 {
 369         trace_nfsd_cb_notify_lock_done(&zero_stateid, task);
 370
 371         /*
 372          * Since this is just an optimization, we don't try very hard if it
 373          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
 374          * just quit trying on anything else.
 375          */
 376         switch (task->tk_status) {
 377         case -NFS4ERR_DELAY:
 378                 rpc_delay(task, 1 * HZ);
 379                 return 0;
 380         default:
 381                 return 1;
 382         }
 383 }
 384
 385 static void
 386 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
 387 {
 388         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 389                                                 struct nfsd4_blocked_lock, nbl_cb);
 390
 391         free_blocked_lock(nbl);
 392 }
 393
 394 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
 395         .prepare        = nfsd4_cb_notify_lock_prepare,
 396         .done           = nfsd4_cb_notify_lock_done,
 397         .release        = nfsd4_cb_notify_lock_release,
 398 };
 399
 400 /*
 401  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 402  * st_{access,deny}_bmap field of the stateid, in order to track not
 403  * only what share bits are currently in force, but also what
 404  * combinations of share bits previous opens have used.  This allows us
 405  * to enforce the recommendation in
 406  * https://datatracker.ietf.org/doc/html/rfc7530#section-16.19.4 that
 407  * the server return an error if the client attempt to downgrade to a
 408  * combination of share bits not explicable by closing some of its
 409  * previous opens.
 410  *
 411  * This enforcement is arguably incomplete, since we don't keep
 412  * track of access/deny bit combinations; so, e.g., we allow:
 413  *
 414  *      OPEN allow read, deny write
 415  *      OPEN allow both, deny none
 416  *      DOWNGRADE allow read, deny none
 417  *
 418  * which we should reject.
 419  *
 420  * But you could also argue that our current code is already overkill,
 421  * since it only exists to return NFS4ERR_INVAL on incorrect client
 422  * behavior.
 423  */
 424 static unsigned int
 425 bmap_to_share_mode(unsigned long bmap)
 426 {
 427         int i;
 428         unsigned int access = 0;
 429
 430         for (i = 1; i < 4; i++) {
 431                 if (test_bit(i, &bmap))
 432                         access |= i;
 433         }
 434         return access;
 435 }
 436
 437 /* set share access for a given stateid */
 438 static inline void
 439 set_access(u32 access, struct nfs4_ol_stateid *stp)
 440 {
 441         unsigned char mask = 1 << access;
 442
 443         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 444         stp->st_access_bmap |= mask;
 445 }
 446
 447 /* clear share access for a given stateid */
 448 static inline void
 449 clear_access(u32 access, struct nfs4_ol_stateid *stp)
 450 {
 451         unsigned char mask = 1 << access;
 452
 453         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 454         stp->st_access_bmap &= ~mask;
 455 }
 456
 457 /* test whether a given stateid has access */
 458 static inline bool
 459 test_access(u32 access, struct nfs4_ol_stateid *stp)
 460 {
 461         unsigned char mask = 1 << access;
 462
 463         return (bool)(stp->st_access_bmap & mask);
 464 }
 465
 466 /* set share deny for a given stateid */
 467 static inline void
 468 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
 469 {
 470         unsigned char mask = 1 << deny;
 471
 472         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 473         stp->st_deny_bmap |= mask;
 474 }
 475
 476 /* clear share deny for a given stateid */
 477 static inline void
 478 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
 479 {
 480         unsigned char mask = 1 << deny;
 481
 482         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 483         stp->st_deny_bmap &= ~mask;
 484 }
 485
 486 /* test whether a given stateid is denying specific access */
 487 static inline bool
 488 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
 489 {
 490         unsigned char mask = 1 << deny;
 491
 492         return (bool)(stp->st_deny_bmap & mask);
 493 }
 494
 495 static int nfs4_access_to_omode(u32 access)
 496 {
 497         switch (access & NFS4_SHARE_ACCESS_BOTH) {
 498         case NFS4_SHARE_ACCESS_READ:
 499                 return O_RDONLY;
 500         case NFS4_SHARE_ACCESS_WRITE:
 501                 return O_WRONLY;
 502         case NFS4_SHARE_ACCESS_BOTH:
 503                 return O_RDWR;
 504         }
 505         WARN_ON_ONCE(1);
 506         return O_RDONLY;
 507 }
 508
 509 static inline int
 510 access_permit_read(struct nfs4_ol_stateid *stp)
 511 {
 512         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
 513                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
 514                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
 515 }
 516
 517 static inline int
 518 access_permit_write(struct nfs4_ol_stateid *stp)
 519 {
 520         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
 521                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
 522 }
 523
 524 static inline struct nfs4_stateowner *
 525 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 526 {
 527         atomic_inc(&sop->so_count);
 528         return sop;
 529 }
 530
 531 static int
 532 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 533 {
 534         return (sop->so_owner.len == owner->len) &&
 535                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 536 }
 537
 538 static struct nfs4_openowner *
 539 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 540                         struct nfs4_client *clp)
 541 {
 542         struct nfs4_stateowner *so;
 543
 544         lockdep_assert_held(&clp->cl_lock);
 545
 546         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 547                             so_strhash) {
 548                 if (!so->so_is_open_owner)
 549                         continue;
 550                 if (same_owner_str(so, &open->op_owner))
 551                         return openowner(nfs4_get_stateowner(so));
 552         }
 553         return NULL;
 554 }
 555
 556 static struct nfs4_openowner *
 557 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 558                         struct nfs4_client *clp)
 559 {
 560         struct nfs4_openowner *oo;
 561
 562         spin_lock(&clp->cl_lock);
 563         oo = find_openstateowner_str_locked(hashval, open, clp);
 564         spin_unlock(&clp->cl_lock);
 565         return oo;
 566 }
 567
 568 static inline u32
 569 opaque_hashval(const void *ptr, int nbytes)
 570 {
 571         unsigned char *cptr = (unsigned char *) ptr;
 572
 573         u32 x = 0;
 574         while (nbytes--) {
 575                 x *= 37;
 576                 x += *cptr++;
 577         }
 578         return x;
 579 }
 580
 581 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 582 {
 583         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 584
 585         kmem_cache_free(file_slab, fp);
 586 }
 587
 588 void
 589 put_nfs4_file(struct nfs4_file *fi)
 590 {
 591         might_lock(&state_lock);
 592
 593         if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
 594                 hlist_del_rcu(&fi->fi_hash);
 595                 spin_unlock(&state_lock);
 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 /* hash table for nfs4_file */
 721 #define FILE_HASH_BITS                   8
 722 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 723
 724 static unsigned int file_hashval(struct svc_fh *fh)
 725 {
 726         struct inode *inode = d_inode(fh->fh_dentry);
 727
 728         /* XXX: why not (here & in file cache) use inode? */
 729         return (unsigned int)hash_long(inode->i_ino, FILE_HASH_BITS);
 730 }
 731
 732 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 733
 734 /*
 735  * Check if courtesy clients have conflicting access and resolve it if possible
 736  *
 737  * access:  is op_share_access if share_access is true.
 738  *          Check if access mode, op_share_access, would conflict with
 739  *          the current deny mode of the file 'fp'.
 740  * access:  is op_share_deny if share_access is false.
 741  *          Check if the deny mode, op_share_deny, would conflict with
 742  *          current access of the file 'fp'.
 743  * stp:     skip checking this entry.
 744  * new_stp: normal open, not open upgrade.
 745  *
 746  * Function returns:
 747  *      false - access/deny mode conflict with normal client.
 748  *      true  - no conflict or conflict with courtesy client(s) is resolved.
 749  */
 750 static bool
 751 nfs4_resolve_deny_conflicts_locked(struct nfs4_file *fp, bool new_stp,
 752                 struct nfs4_ol_stateid *stp, u32 access, bool share_access)
 753 {
 754         struct nfs4_ol_stateid *st;
 755         bool resolvable = true;
 756         unsigned char bmap;
 757         struct nfsd_net *nn;
 758         struct nfs4_client *clp;
 759
 760         lockdep_assert_held(&fp->fi_lock);
 761         list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
 762                 /* ignore lock stateid */
 763                 if (st->st_openstp)
 764                         continue;
 765                 if (st == stp && new_stp)
 766                         continue;
 767                 /* check file access against deny mode or vice versa */
 768                 bmap = share_access ? st->st_deny_bmap : st->st_access_bmap;
 769                 if (!(access & bmap_to_share_mode(bmap)))
 770                         continue;
 771                 clp = st->st_stid.sc_client;
 772                 if (try_to_expire_client(clp))
 773                         continue;
 774                 resolvable = false;
 775                 break;
 776         }
 777         if (resolvable) {
 778                 clp = stp->st_stid.sc_client;
 779                 nn = net_generic(clp->net, nfsd_net_id);
 780                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
 781         }
 782         return resolvable;
 783 }
 784
 785 static void
 786 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 787 {
 788         lockdep_assert_held(&fp->fi_lock);
 789
 790         if (access & NFS4_SHARE_ACCESS_WRITE)
 791                 atomic_inc(&fp->fi_access[O_WRONLY]);
 792         if (access & NFS4_SHARE_ACCESS_READ)
 793                 atomic_inc(&fp->fi_access[O_RDONLY]);
 794 }
 795
 796 static __be32
 797 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 798 {
 799         lockdep_assert_held(&fp->fi_lock);
 800
 801         /* Does this access mode make sense? */
 802         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 803                 return nfserr_inval;
 804
 805         /* Does it conflict with a deny mode already set? */
 806         if ((access & fp->fi_share_deny) != 0)
 807                 return nfserr_share_denied;
 808
 809         __nfs4_file_get_access(fp, access);
 810         return nfs_ok;
 811 }
 812
 813 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 814 {
 815         /* Common case is that there is no deny mode. */
 816         if (deny) {
 817                 /* Does this deny mode make sense? */
 818                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 819                         return nfserr_inval;
 820
 821                 if ((deny & NFS4_SHARE_DENY_READ) &&
 822                     atomic_read(&fp->fi_access[O_RDONLY]))
 823                         return nfserr_share_denied;
 824
 825                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 826                     atomic_read(&fp->fi_access[O_WRONLY]))
 827                         return nfserr_share_denied;
 828         }
 829         return nfs_ok;
 830 }
 831
 832 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 833 {
 834         might_lock(&fp->fi_lock);
 835
 836         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 837                 struct nfsd_file *f1 = NULL;
 838                 struct nfsd_file *f2 = NULL;
 839
 840                 swap(f1, fp->fi_fds[oflag]);
 841                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 842                         swap(f2, fp->fi_fds[O_RDWR]);
 843                 spin_unlock(&fp->fi_lock);
 844                 if (f1)
 845                         nfsd_file_put(f1);
 846                 if (f2)
 847                         nfsd_file_put(f2);
 848         }
 849 }
 850
 851 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 852 {
 853         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 854
 855         if (access & NFS4_SHARE_ACCESS_WRITE)
 856                 __nfs4_file_put_access(fp, O_WRONLY);
 857         if (access & NFS4_SHARE_ACCESS_READ)
 858                 __nfs4_file_put_access(fp, O_RDONLY);
 859 }
 860
 861 /*
 862  * Allocate a new open/delegation state counter. This is needed for
 863  * pNFS for proper return on close semantics.
 864  *
 865  * Note that we only allocate it for pNFS-enabled exports, otherwise
 866  * all pointers to struct nfs4_clnt_odstate are always NULL.
 867  */
 868 static struct nfs4_clnt_odstate *
 869 alloc_clnt_odstate(struct nfs4_client *clp)
 870 {
 871         struct nfs4_clnt_odstate *co;
 872
 873         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 874         if (co) {
 875                 co->co_client = clp;
 876                 refcount_set(&co->co_odcount, 1);
 877         }
 878         return co;
 879 }
 880
 881 static void
 882 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 883 {
 884         struct nfs4_file *fp = co->co_file;
 885
 886         lockdep_assert_held(&fp->fi_lock);
 887         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 888 }
 889
 890 static inline void
 891 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 892 {
 893         if (co)
 894                 refcount_inc(&co->co_odcount);
 895 }
 896
 897 static void
 898 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 899 {
 900         struct nfs4_file *fp;
 901
 902         if (!co)
 903                 return;
 904
 905         fp = co->co_file;
 906         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 907                 list_del(&co->co_perfile);
 908                 spin_unlock(&fp->fi_lock);
 909
 910                 nfsd4_return_all_file_layouts(co->co_client, fp);
 911                 kmem_cache_free(odstate_slab, co);
 912         }
 913 }
 914
 915 static struct nfs4_clnt_odstate *
 916 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 917 {
 918         struct nfs4_clnt_odstate *co;
 919         struct nfs4_client *cl;
 920
 921         if (!new)
 922                 return NULL;
 923
 924         cl = new->co_client;
 925
 926         spin_lock(&fp->fi_lock);
 927         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 928                 if (co->co_client == cl) {
 929                         get_clnt_odstate(co);
 930                         goto out;
 931                 }
 932         }
 933         co = new;
 934         co->co_file = fp;
 935         hash_clnt_odstate_locked(new);
 936 out:
 937         spin_unlock(&fp->fi_lock);
 938         return co;
 939 }
 940
 941 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 942                                   void (*sc_free)(struct nfs4_stid *))
 943 {
 944         struct nfs4_stid *stid;
 945         int new_id;
 946
 947         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 948         if (!stid)
 949                 return NULL;
 950
 951         idr_preload(GFP_KERNEL);
 952         spin_lock(&cl->cl_lock);
 953         /* Reserving 0 for start of file in nfsdfs "states" file: */
 954         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
 955         spin_unlock(&cl->cl_lock);
 956         idr_preload_end();
 957         if (new_id < 0)
 958                 goto out_free;
 959
 960         stid->sc_free = sc_free;
 961         stid->sc_client = cl;
 962         stid->sc_stateid.si_opaque.so_id = new_id;
 963         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 964         /* Will be incremented before return to client: */
 965         refcount_set(&stid->sc_count, 1);
 966         spin_lock_init(&stid->sc_lock);
 967         INIT_LIST_HEAD(&stid->sc_cp_list);
 968
 969         /*
 970          * It shouldn't be a problem to reuse an opaque stateid value.
 971          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 972          * example, a stray write retransmission could be accepted by
 973          * the server when it should have been rejected.  Therefore,
 974          * adopt a trick from the sctp code to attempt to maximize the
 975          * amount of time until an id is reused, by ensuring they always
 976          * "increase" (mod INT_MAX):
 977          */
 978         return stid;
 979 out_free:
 980         kmem_cache_free(slab, stid);
 981         return NULL;
 982 }
 983
 984 /*
 985  * Create a unique stateid_t to represent each COPY.
 986  */
 987 static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
 988                               unsigned char cs_type)
 989 {
 990         int new_id;
 991
 992         stid->cs_stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
 993         stid->cs_stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
 994
 995         idr_preload(GFP_KERNEL);
 996         spin_lock(&nn->s2s_cp_lock);
 997         new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
 998         stid->cs_stid.si_opaque.so_id = new_id;
 999         stid->cs_stid.si_generation = 1;
1000         spin_unlock(&nn->s2s_cp_lock);
1001         idr_preload_end();
1002         if (new_id < 0)
1003                 return 0;
1004         stid->cs_type = cs_type;
1005         return 1;
1006 }
1007
1008 int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
1009 {
1010         return nfs4_init_cp_state(nn, &copy->cp_stateid, NFS4_COPY_STID);
1011 }
1012
1013 struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
1014                                                      struct nfs4_stid *p_stid)
1015 {
1016         struct nfs4_cpntf_state *cps;
1017
1018         cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
1019         if (!cps)
1020                 return NULL;
1021         cps->cpntf_time = ktime_get_boottime_seconds();
1022         refcount_set(&cps->cp_stateid.cs_count, 1);
1023         if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
1024                 goto out_free;
1025         spin_lock(&nn->s2s_cp_lock);
1026         list_add(&cps->cp_list, &p_stid->sc_cp_list);
1027         spin_unlock(&nn->s2s_cp_lock);
1028         return cps;
1029 out_free:
1030         kfree(cps);
1031         return NULL;
1032 }
1033
1034 void nfs4_free_copy_state(struct nfsd4_copy *copy)
1035 {
1036         struct nfsd_net *nn;
1037
1038         if (copy->cp_stateid.cs_type != NFS4_COPY_STID)
1039                 return;
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         if (clp->cl_minorversion) {
1371                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1372                 refcount_inc(&dp->dl_stid.sc_count);
1373                 spin_lock(&clp->cl_lock);
1374                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1375                 spin_unlock(&clp->cl_lock);
1376         }
1377         destroy_unhashed_deleg(dp);
1378 }
1379
1380 /*
1381  * SETCLIENTID state
1382  */
1383
1384 static unsigned int clientid_hashval(u32 id)
1385 {
1386         return id & CLIENT_HASH_MASK;
1387 }
1388
1389 static unsigned int clientstr_hashval(struct xdr_netobj name)
1390 {
1391         return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1392 }
1393
1394 /*
1395  * A stateid that had a deny mode associated with it is being released
1396  * or downgraded. Recalculate the deny mode on the file.
1397  */
1398 static void
1399 recalculate_deny_mode(struct nfs4_file *fp)
1400 {
1401         struct nfs4_ol_stateid *stp;
1402
1403         spin_lock(&fp->fi_lock);
1404         fp->fi_share_deny = 0;
1405         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1406                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1407         spin_unlock(&fp->fi_lock);
1408 }
1409
1410 static void
1411 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1412 {
1413         int i;
1414         bool change = false;
1415
1416         for (i = 1; i < 4; i++) {
1417                 if ((i & deny) != i) {
1418                         change = true;
1419                         clear_deny(i, stp);
1420                 }
1421         }
1422
1423         /* Recalculate per-file deny mode if there was a change */
1424         if (change)
1425                 recalculate_deny_mode(stp->st_stid.sc_file);
1426 }
1427
1428 /* release all access and file references for a given stateid */
1429 static void
1430 release_all_access(struct nfs4_ol_stateid *stp)
1431 {
1432         int i;
1433         struct nfs4_file *fp = stp->st_stid.sc_file;
1434
1435         if (fp && stp->st_deny_bmap != 0)
1436                 recalculate_deny_mode(fp);
1437
1438         for (i = 1; i < 4; i++) {
1439                 if (test_access(i, stp))
1440                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1441                 clear_access(i, stp);
1442         }
1443 }
1444
1445 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1446 {
1447         kfree(sop->so_owner.data);
1448         sop->so_ops->so_free(sop);
1449 }
1450
1451 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1452 {
1453         struct nfs4_client *clp = sop->so_client;
1454
1455         might_lock(&clp->cl_lock);
1456
1457         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1458                 return;
1459         sop->so_ops->so_unhash(sop);
1460         spin_unlock(&clp->cl_lock);
1461         nfs4_free_stateowner(sop);
1462 }
1463
1464 static bool
1465 nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1466 {
1467         return list_empty(&stp->st_perfile);
1468 }
1469
1470 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1471 {
1472         struct nfs4_file *fp = stp->st_stid.sc_file;
1473
1474         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1475
1476         if (list_empty(&stp->st_perfile))
1477                 return false;
1478
1479         spin_lock(&fp->fi_lock);
1480         list_del_init(&stp->st_perfile);
1481         spin_unlock(&fp->fi_lock);
1482         list_del(&stp->st_perstateowner);
1483         return true;
1484 }
1485
1486 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1487 {
1488         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1489
1490         put_clnt_odstate(stp->st_clnt_odstate);
1491         release_all_access(stp);
1492         if (stp->st_stateowner)
1493                 nfs4_put_stateowner(stp->st_stateowner);
1494         WARN_ON(!list_empty(&stid->sc_cp_list));
1495         kmem_cache_free(stateid_slab, stid);
1496 }
1497
1498 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1499 {
1500         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1501         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1502         struct nfsd_file *nf;
1503
1504         nf = find_any_file(stp->st_stid.sc_file);
1505         if (nf) {
1506                 get_file(nf->nf_file);
1507                 filp_close(nf->nf_file, (fl_owner_t)lo);
1508                 nfsd_file_put(nf);
1509         }
1510         nfs4_free_ol_stateid(stid);
1511 }
1512
1513 /*
1514  * Put the persistent reference to an already unhashed generic stateid, while
1515  * holding the cl_lock. If it's the last reference, then put it onto the
1516  * reaplist for later destruction.
1517  */
1518 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1519                                        struct list_head *reaplist)
1520 {
1521         struct nfs4_stid *s = &stp->st_stid;
1522         struct nfs4_client *clp = s->sc_client;
1523
1524         lockdep_assert_held(&clp->cl_lock);
1525
1526         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1527
1528         if (!refcount_dec_and_test(&s->sc_count)) {
1529                 wake_up_all(&close_wq);
1530                 return;
1531         }
1532
1533         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1534         list_add(&stp->st_locks, reaplist);
1535 }
1536
1537 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1538 {
1539         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1540
1541         if (!unhash_ol_stateid(stp))
1542                 return false;
1543         list_del_init(&stp->st_locks);
1544         nfs4_unhash_stid(&stp->st_stid);
1545         return true;
1546 }
1547
1548 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1549 {
1550         struct nfs4_client *clp = stp->st_stid.sc_client;
1551         bool unhashed;
1552
1553         spin_lock(&clp->cl_lock);
1554         unhashed = unhash_lock_stateid(stp);
1555         spin_unlock(&clp->cl_lock);
1556         if (unhashed)
1557                 nfs4_put_stid(&stp->st_stid);
1558 }
1559
1560 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1561 {
1562         struct nfs4_client *clp = lo->lo_owner.so_client;
1563
1564         lockdep_assert_held(&clp->cl_lock);
1565
1566         list_del_init(&lo->lo_owner.so_strhash);
1567 }
1568
1569 /*
1570  * Free a list of generic stateids that were collected earlier after being
1571  * fully unhashed.
1572  */
1573 static void
1574 free_ol_stateid_reaplist(struct list_head *reaplist)
1575 {
1576         struct nfs4_ol_stateid *stp;
1577         struct nfs4_file *fp;
1578
1579         might_sleep();
1580
1581         while (!list_empty(reaplist)) {
1582                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1583                                        st_locks);
1584                 list_del(&stp->st_locks);
1585                 fp = stp->st_stid.sc_file;
1586                 stp->st_stid.sc_free(&stp->st_stid);
1587                 if (fp)
1588                         put_nfs4_file(fp);
1589         }
1590 }
1591
1592 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1593                                        struct list_head *reaplist)
1594 {
1595         struct nfs4_ol_stateid *stp;
1596
1597         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1598
1599         while (!list_empty(&open_stp->st_locks)) {
1600                 stp = list_entry(open_stp->st_locks.next,
1601                                 struct nfs4_ol_stateid, st_locks);
1602                 WARN_ON(!unhash_lock_stateid(stp));
1603                 put_ol_stateid_locked(stp, reaplist);
1604         }
1605 }
1606
1607 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1608                                 struct list_head *reaplist)
1609 {
1610         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1611
1612         if (!unhash_ol_stateid(stp))
1613                 return false;
1614         release_open_stateid_locks(stp, reaplist);
1615         return true;
1616 }
1617
1618 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1619 {
1620         LIST_HEAD(reaplist);
1621
1622         spin_lock(&stp->st_stid.sc_client->cl_lock);
1623         if (unhash_open_stateid(stp, &reaplist))
1624                 put_ol_stateid_locked(stp, &reaplist);
1625         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1626         free_ol_stateid_reaplist(&reaplist);
1627 }
1628
1629 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1630 {
1631         struct nfs4_client *clp = oo->oo_owner.so_client;
1632
1633         lockdep_assert_held(&clp->cl_lock);
1634
1635         list_del_init(&oo->oo_owner.so_strhash);
1636         list_del_init(&oo->oo_perclient);
1637 }
1638
1639 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1640 {
1641         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1642                                           nfsd_net_id);
1643         struct nfs4_ol_stateid *s;
1644
1645         spin_lock(&nn->client_lock);
1646         s = oo->oo_last_closed_stid;
1647         if (s) {
1648                 list_del_init(&oo->oo_close_lru);
1649                 oo->oo_last_closed_stid = NULL;
1650         }
1651         spin_unlock(&nn->client_lock);
1652         if (s)
1653                 nfs4_put_stid(&s->st_stid);
1654 }
1655
1656 static void release_openowner(struct nfs4_openowner *oo)
1657 {
1658         struct nfs4_ol_stateid *stp;
1659         struct nfs4_client *clp = oo->oo_owner.so_client;
1660         struct list_head reaplist;
1661
1662         INIT_LIST_HEAD(&reaplist);
1663
1664         spin_lock(&clp->cl_lock);
1665         unhash_openowner_locked(oo);
1666         while (!list_empty(&oo->oo_owner.so_stateids)) {
1667                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1668                                 struct nfs4_ol_stateid, st_perstateowner);
1669                 if (unhash_open_stateid(stp, &reaplist))
1670                         put_ol_stateid_locked(stp, &reaplist);
1671         }
1672         spin_unlock(&clp->cl_lock);
1673         free_ol_stateid_reaplist(&reaplist);
1674         release_last_closed_stateid(oo);
1675         nfs4_put_stateowner(&oo->oo_owner);
1676 }
1677
1678 static inline int
1679 hash_sessionid(struct nfs4_sessionid *sessionid)
1680 {
1681         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1682
1683         return sid->sequence % SESSION_HASH_SIZE;
1684 }
1685
1686 #ifdef CONFIG_SUNRPC_DEBUG
1687 static inline void
1688 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1689 {
1690         u32 *ptr = (u32 *)(&sessionid->data[0]);
1691         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1692 }
1693 #else
1694 static inline void
1695 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1696 {
1697 }
1698 #endif
1699
1700 /*
1701  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1702  * won't be used for replay.
1703  */
1704 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1705 {
1706         struct nfs4_stateowner *so = cstate->replay_owner;
1707
1708         if (nfserr == nfserr_replay_me)
1709                 return;
1710
1711         if (!seqid_mutating_err(ntohl(nfserr))) {
1712                 nfsd4_cstate_clear_replay(cstate);
1713                 return;
1714         }
1715         if (!so)
1716                 return;
1717         if (so->so_is_open_owner)
1718                 release_last_closed_stateid(openowner(so));
1719         so->so_seqid++;
1720         return;
1721 }
1722
1723 static void
1724 gen_sessionid(struct nfsd4_session *ses)
1725 {
1726         struct nfs4_client *clp = ses->se_client;
1727         struct nfsd4_sessionid *sid;
1728
1729         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1730         sid->clientid = clp->cl_clientid;
1731         sid->sequence = current_sessionid++;
1732         sid->reserved = 0;
1733 }
1734
1735 /*
1736  * The protocol defines ca_maxresponssize_cached to include the size of
1737  * the rpc header, but all we need to cache is the data starting after
1738  * the end of the initial SEQUENCE operation--the rest we regenerate
1739  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1740  * value that is the number of bytes in our cache plus a few additional
1741  * bytes.  In order to stay on the safe side, and not promise more than
1742  * we can cache, those additional bytes must be the minimum possible: 24
1743  * bytes of rpc header (xid through accept state, with AUTH_NULL
1744  * verifier), 12 for the compound header (with zero-length tag), and 44
1745  * for the SEQUENCE op response:
1746  */
1747 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1748
1749 static void
1750 free_session_slots(struct nfsd4_session *ses)
1751 {
1752         int i;
1753
1754         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1755                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1756                 kfree(ses->se_slots[i]);
1757         }
1758 }
1759
1760 /*
1761  * We don't actually need to cache the rpc and session headers, so we
1762  * can allocate a little less for each slot:
1763  */
1764 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1765 {
1766         u32 size;
1767
1768         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1769                 size = 0;
1770         else
1771                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1772         return size + sizeof(struct nfsd4_slot);
1773 }
1774
1775 /*
1776  * XXX: If we run out of reserved DRC memory we could (up to a point)
1777  * re-negotiate active sessions and reduce their slot usage to make
1778  * room for new connections. For now we just fail the create session.
1779  */
1780 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1781 {
1782         u32 slotsize = slot_bytes(ca);
1783         u32 num = ca->maxreqs;
1784         unsigned long avail, total_avail;
1785         unsigned int scale_factor;
1786
1787         spin_lock(&nfsd_drc_lock);
1788         if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1789                 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1790         else
1791                 /* We have handed out more space than we chose in
1792                  * set_max_drc() to allow.  That isn't really a
1793                  * problem as long as that doesn't make us think we
1794                  * have lots more due to integer overflow.
1795                  */
1796                 total_avail = 0;
1797         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1798         /*
1799          * Never use more than a fraction of the remaining memory,
1800          * unless it's the only way to give this client a slot.
1801          * The chosen fraction is either 1/8 or 1/number of threads,
1802          * whichever is smaller.  This ensures there are adequate
1803          * slots to support multiple clients per thread.
1804          * Give the client one slot even if that would require
1805          * over-allocation--it is better than failure.
1806          */
1807         scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1808
1809         avail = clamp_t(unsigned long, avail, slotsize,
1810                         total_avail/scale_factor);
1811         num = min_t(int, num, avail / slotsize);
1812         num = max_t(int, num, 1);
1813         nfsd_drc_mem_used += num * slotsize;
1814         spin_unlock(&nfsd_drc_lock);
1815
1816         return num;
1817 }
1818
1819 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1820 {
1821         int slotsize = slot_bytes(ca);
1822
1823         spin_lock(&nfsd_drc_lock);
1824         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1825         spin_unlock(&nfsd_drc_lock);
1826 }
1827
1828 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1829                                            struct nfsd4_channel_attrs *battrs)
1830 {
1831         int numslots = fattrs->maxreqs;
1832         int slotsize = slot_bytes(fattrs);
1833         struct nfsd4_session *new;
1834         int mem, i;
1835
1836         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1837                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1838         mem = numslots * sizeof(struct nfsd4_slot *);
1839
1840         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1841         if (!new)
1842                 return NULL;
1843         /* allocate each struct nfsd4_slot and data cache in one piece */
1844         for (i = 0; i < numslots; i++) {
1845                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1846                 if (!new->se_slots[i])
1847                         goto out_free;
1848         }
1849
1850         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1851         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1852
1853         return new;
1854 out_free:
1855         while (i--)
1856                 kfree(new->se_slots[i]);
1857         kfree(new);
1858         return NULL;
1859 }
1860
1861 static void free_conn(struct nfsd4_conn *c)
1862 {
1863         svc_xprt_put(c->cn_xprt);
1864         kfree(c);
1865 }
1866
1867 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1868 {
1869         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1870         struct nfs4_client *clp = c->cn_session->se_client;
1871
1872         trace_nfsd_cb_lost(clp);
1873
1874         spin_lock(&clp->cl_lock);
1875         if (!list_empty(&c->cn_persession)) {
1876                 list_del(&c->cn_persession);
1877                 free_conn(c);
1878         }
1879         nfsd4_probe_callback(clp);
1880         spin_unlock(&clp->cl_lock);
1881 }
1882
1883 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1884 {
1885         struct nfsd4_conn *conn;
1886
1887         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1888         if (!conn)
1889                 return NULL;
1890         svc_xprt_get(rqstp->rq_xprt);
1891         conn->cn_xprt = rqstp->rq_xprt;
1892         conn->cn_flags = flags;
1893         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1894         return conn;
1895 }
1896
1897 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1898 {
1899         conn->cn_session = ses;
1900         list_add(&conn->cn_persession, &ses->se_conns);
1901 }
1902
1903 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1904 {
1905         struct nfs4_client *clp = ses->se_client;
1906
1907         spin_lock(&clp->cl_lock);
1908         __nfsd4_hash_conn(conn, ses);
1909         spin_unlock(&clp->cl_lock);
1910 }
1911
1912 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1913 {
1914         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1915         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1916 }
1917
1918 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1919 {
1920         int ret;
1921
1922         nfsd4_hash_conn(conn, ses);
1923         ret = nfsd4_register_conn(conn);
1924         if (ret)
1925                 /* oops; xprt is already down: */
1926                 nfsd4_conn_lost(&conn->cn_xpt_user);
1927         /* We may have gained or lost a callback channel: */
1928         nfsd4_probe_callback_sync(ses->se_client);
1929 }
1930
1931 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1932 {
1933         u32 dir = NFS4_CDFC4_FORE;
1934
1935         if (cses->flags & SESSION4_BACK_CHAN)
1936                 dir |= NFS4_CDFC4_BACK;
1937         return alloc_conn(rqstp, dir);
1938 }
1939
1940 /* must be called under client_lock */
1941 static void nfsd4_del_conns(struct nfsd4_session *s)
1942 {
1943         struct nfs4_client *clp = s->se_client;
1944         struct nfsd4_conn *c;
1945
1946         spin_lock(&clp->cl_lock);
1947         while (!list_empty(&s->se_conns)) {
1948                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1949                 list_del_init(&c->cn_persession);
1950                 spin_unlock(&clp->cl_lock);
1951
1952                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1953                 free_conn(c);
1954
1955                 spin_lock(&clp->cl_lock);
1956         }
1957         spin_unlock(&clp->cl_lock);
1958 }
1959
1960 static void __free_session(struct nfsd4_session *ses)
1961 {
1962         free_session_slots(ses);
1963         kfree(ses);
1964 }
1965
1966 static void free_session(struct nfsd4_session *ses)
1967 {
1968         nfsd4_del_conns(ses);
1969         nfsd4_put_drc_mem(&ses->se_fchannel);
1970         __free_session(ses);
1971 }
1972
1973 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1974 {
1975         int idx;
1976         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1977
1978         new->se_client = clp;
1979         gen_sessionid(new);
1980
1981         INIT_LIST_HEAD(&new->se_conns);
1982
1983         new->se_cb_seq_nr = 1;
1984         new->se_flags = cses->flags;
1985         new->se_cb_prog = cses->callback_prog;
1986         new->se_cb_sec = cses->cb_sec;
1987         atomic_set(&new->se_ref, 0);
1988         idx = hash_sessionid(&new->se_sessionid);
1989         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1990         spin_lock(&clp->cl_lock);
1991         list_add(&new->se_perclnt, &clp->cl_sessions);
1992         spin_unlock(&clp->cl_lock);
1993
1994         {
1995                 struct sockaddr *sa = svc_addr(rqstp);
1996                 /*
1997                  * This is a little silly; with sessions there's no real
1998                  * use for the callback address.  Use the peer address
1999                  * as a reasonable default for now, but consider fixing
2000                  * the rpc client not to require an address in the
2001                  * future:
2002                  */
2003                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
2004                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
2005         }
2006 }
2007
2008 /* caller must hold client_lock */
2009 static struct nfsd4_session *
2010 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
2011 {
2012         struct nfsd4_session *elem;
2013         int idx;
2014         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2015
2016         lockdep_assert_held(&nn->client_lock);
2017
2018         dump_sessionid(__func__, sessionid);
2019         idx = hash_sessionid(sessionid);
2020         /* Search in the appropriate list */
2021         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
2022                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
2023                             NFS4_MAX_SESSIONID_LEN)) {
2024                         return elem;
2025                 }
2026         }
2027
2028         dprintk("%s: session not found\n", __func__);
2029         return NULL;
2030 }
2031
2032 static struct nfsd4_session *
2033 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
2034                 __be32 *ret)
2035 {
2036         struct nfsd4_session *session;
2037         __be32 status = nfserr_badsession;
2038
2039         session = __find_in_sessionid_hashtbl(sessionid, net);
2040         if (!session)
2041                 goto out;
2042         status = nfsd4_get_session_locked(session);
2043         if (status)
2044                 session = NULL;
2045 out:
2046         *ret = status;
2047         return session;
2048 }
2049
2050 /* caller must hold client_lock */
2051 static void
2052 unhash_session(struct nfsd4_session *ses)
2053 {
2054         struct nfs4_client *clp = ses->se_client;
2055         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2056
2057         lockdep_assert_held(&nn->client_lock);
2058
2059         list_del(&ses->se_hash);
2060         spin_lock(&ses->se_client->cl_lock);
2061         list_del(&ses->se_perclnt);
2062         spin_unlock(&ses->se_client->cl_lock);
2063 }
2064
2065 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
2066 static int
2067 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
2068 {
2069         /*
2070          * We're assuming the clid was not given out from a boot
2071          * precisely 2^32 (about 136 years) before this one.  That seems
2072          * a safe assumption:
2073          */
2074         if (clid->cl_boot == (u32)nn->boot_time)
2075                 return 0;
2076         trace_nfsd_clid_stale(clid);
2077         return 1;
2078 }
2079
2080 /*
2081  * XXX Should we use a slab cache ?
2082  * This type of memory management is somewhat inefficient, but we use it
2083  * anyway since SETCLIENTID is not a common operation.
2084  */
2085 static struct nfs4_client *alloc_client(struct xdr_netobj name,
2086                                 struct nfsd_net *nn)
2087 {
2088         struct nfs4_client *clp;
2089         int i;
2090
2091         if (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) {
2092                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
2093                 return NULL;
2094         }
2095         clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
2096         if (clp == NULL)
2097                 return NULL;
2098         xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
2099         if (clp->cl_name.data == NULL)
2100                 goto err_no_name;
2101         clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
2102                                                  sizeof(struct list_head),
2103                                                  GFP_KERNEL);
2104         if (!clp->cl_ownerstr_hashtbl)
2105                 goto err_no_hashtbl;
2106         for (i = 0; i < OWNER_HASH_SIZE; i++)
2107                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
2108         INIT_LIST_HEAD(&clp->cl_sessions);
2109         idr_init(&clp->cl_stateids);
2110         atomic_set(&clp->cl_rpc_users, 0);
2111         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
2112         clp->cl_state = NFSD4_ACTIVE;
2113         atomic_inc(&nn->nfs4_client_count);
2114         atomic_set(&clp->cl_delegs_in_recall, 0);
2115         INIT_LIST_HEAD(&clp->cl_idhash);
2116         INIT_LIST_HEAD(&clp->cl_openowners);
2117         INIT_LIST_HEAD(&clp->cl_delegations);
2118         INIT_LIST_HEAD(&clp->cl_lru);
2119         INIT_LIST_HEAD(&clp->cl_revoked);
2120 #ifdef CONFIG_NFSD_PNFS
2121         INIT_LIST_HEAD(&clp->cl_lo_states);
2122 #endif
2123         INIT_LIST_HEAD(&clp->async_copies);
2124         spin_lock_init(&clp->async_lock);
2125         spin_lock_init(&clp->cl_lock);
2126         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
2127         return clp;
2128 err_no_hashtbl:
2129         kfree(clp->cl_name.data);
2130 err_no_name:
2131         kmem_cache_free(client_slab, clp);
2132         return NULL;
2133 }
2134
2135 static void __free_client(struct kref *k)
2136 {
2137         struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
2138         struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
2139
2140         free_svc_cred(&clp->cl_cred);
2141         kfree(clp->cl_ownerstr_hashtbl);
2142         kfree(clp->cl_name.data);
2143         kfree(clp->cl_nii_domain.data);
2144         kfree(clp->cl_nii_name.data);
2145         idr_destroy(&clp->cl_stateids);
2146         kmem_cache_free(client_slab, clp);
2147 }
2148
2149 static void drop_client(struct nfs4_client *clp)
2150 {
2151         kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
2152 }
2153
2154 static void
2155 free_client(struct nfs4_client *clp)
2156 {
2157         while (!list_empty(&clp->cl_sessions)) {
2158                 struct nfsd4_session *ses;
2159                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
2160                                 se_perclnt);
2161                 list_del(&ses->se_perclnt);
2162                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
2163                 free_session(ses);
2164         }
2165         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
2166         if (clp->cl_nfsd_dentry) {
2167                 nfsd_client_rmdir(clp->cl_nfsd_dentry);
2168                 clp->cl_nfsd_dentry = NULL;
2169                 wake_up_all(&expiry_wq);
2170         }
2171         drop_client(clp);
2172 }
2173
2174 /* must be called under the client_lock */
2175 static void
2176 unhash_client_locked(struct nfs4_client *clp)
2177 {
2178         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2179         struct nfsd4_session *ses;
2180
2181         lockdep_assert_held(&nn->client_lock);
2182
2183         /* Mark the client as expired! */
2184         clp->cl_time = 0;
2185         /* Make it invisible */
2186         if (!list_empty(&clp->cl_idhash)) {
2187                 list_del_init(&clp->cl_idhash);
2188                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2189                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
2190                 else
2191                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2192         }
2193         list_del_init(&clp->cl_lru);
2194         spin_lock(&clp->cl_lock);
2195         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2196                 list_del_init(&ses->se_hash);
2197         spin_unlock(&clp->cl_lock);
2198 }
2199
2200 static void
2201 unhash_client(struct nfs4_client *clp)
2202 {
2203         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2204
2205         spin_lock(&nn->client_lock);
2206         unhash_client_locked(clp);
2207         spin_unlock(&nn->client_lock);
2208 }
2209
2210 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
2211 {
2212         if (atomic_read(&clp->cl_rpc_users))
2213                 return nfserr_jukebox;
2214         unhash_client_locked(clp);
2215         return nfs_ok;
2216 }
2217
2218 static void
2219 __destroy_client(struct nfs4_client *clp)
2220 {
2221         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2222         int i;
2223         struct nfs4_openowner *oo;
2224         struct nfs4_delegation *dp;
2225         struct list_head reaplist;
2226
2227         INIT_LIST_HEAD(&reaplist);
2228         spin_lock(&state_lock);
2229         while (!list_empty(&clp->cl_delegations)) {
2230                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2231                 WARN_ON(!unhash_delegation_locked(dp));
2232                 list_add(&dp->dl_recall_lru, &reaplist);
2233         }
2234         spin_unlock(&state_lock);
2235         while (!list_empty(&reaplist)) {
2236                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2237                 list_del_init(&dp->dl_recall_lru);
2238                 destroy_unhashed_deleg(dp);
2239         }
2240         while (!list_empty(&clp->cl_revoked)) {
2241                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2242                 list_del_init(&dp->dl_recall_lru);
2243                 nfs4_put_stid(&dp->dl_stid);
2244         }
2245         while (!list_empty(&clp->cl_openowners)) {
2246                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2247                 nfs4_get_stateowner(&oo->oo_owner);
2248                 release_openowner(oo);
2249         }
2250         for (i = 0; i < OWNER_HASH_SIZE; i++) {
2251                 struct nfs4_stateowner *so, *tmp;
2252
2253                 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2254                                          so_strhash) {
2255                         /* Should be no openowners at this point */
2256                         WARN_ON_ONCE(so->so_is_open_owner);
2257                         remove_blocked_locks(lockowner(so));
2258                 }
2259         }
2260         nfsd4_return_all_client_layouts(clp);
2261         nfsd4_shutdown_copy(clp);
2262         nfsd4_shutdown_callback(clp);
2263         if (clp->cl_cb_conn.cb_xprt)
2264                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2265         atomic_add_unless(&nn->nfs4_client_count, -1, 0);
2266         nfsd4_dec_courtesy_client_count(nn, clp);
2267         free_client(clp);
2268         wake_up_all(&expiry_wq);
2269 }
2270
2271 static void
2272 destroy_client(struct nfs4_client *clp)
2273 {
2274         unhash_client(clp);
2275         __destroy_client(clp);
2276 }
2277
2278 static void inc_reclaim_complete(struct nfs4_client *clp)
2279 {
2280         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2281
2282         if (!nn->track_reclaim_completes)
2283                 return;
2284         if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2285                 return;
2286         if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2287                         nn->reclaim_str_hashtbl_size) {
2288                 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2289                                 clp->net->ns.inum);
2290                 nfsd4_end_grace(nn);
2291         }
2292 }
2293
2294 static void expire_client(struct nfs4_client *clp)
2295 {
2296         unhash_client(clp);
2297         nfsd4_client_record_remove(clp);
2298         __destroy_client(clp);
2299 }
2300
2301 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2302 {
2303         memcpy(target->cl_verifier.data, source->data,
2304                         sizeof(target->cl_verifier.data));
2305 }
2306
2307 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2308 {
2309         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2310         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2311 }
2312
2313 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2314 {
2315         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2316         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2317                                                                 GFP_KERNEL);
2318         target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2319         if ((source->cr_principal && !target->cr_principal) ||
2320             (source->cr_raw_principal && !target->cr_raw_principal) ||
2321             (source->cr_targ_princ && !target->cr_targ_princ))
2322                 return -ENOMEM;
2323
2324         target->cr_flavor = source->cr_flavor;
2325         target->cr_uid = source->cr_uid;
2326         target->cr_gid = source->cr_gid;
2327         target->cr_group_info = source->cr_group_info;
2328         get_group_info(target->cr_group_info);
2329         target->cr_gss_mech = source->cr_gss_mech;
2330         if (source->cr_gss_mech)
2331                 gss_mech_get(source->cr_gss_mech);
2332         return 0;
2333 }
2334
2335 static int
2336 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2337 {
2338         if (o1->len < o2->len)
2339                 return -1;
2340         if (o1->len > o2->len)
2341                 return 1;
2342         return memcmp(o1->data, o2->data, o1->len);
2343 }
2344
2345 static int
2346 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2347 {
2348         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2349 }
2350
2351 static int
2352 same_clid(clientid_t *cl1, clientid_t *cl2)
2353 {
2354         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2355 }
2356
2357 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2358 {
2359         int i;
2360
2361         if (g1->ngroups != g2->ngroups)
2362                 return false;
2363         for (i=0; i<g1->ngroups; i++)
2364                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2365                         return false;
2366         return true;
2367 }
2368
2369 /*
2370  * RFC 3530 language requires clid_inuse be returned when the
2371  * "principal" associated with a requests differs from that previously
2372  * used.  We use uid, gid's, and gss principal string as our best
2373  * approximation.  We also don't want to allow non-gss use of a client
2374  * established using gss: in theory cr_principal should catch that
2375  * change, but in practice cr_principal can be null even in the gss case
2376  * since gssd doesn't always pass down a principal string.
2377  */
2378 static bool is_gss_cred(struct svc_cred *cr)
2379 {
2380         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2381         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2382 }
2383
2384
2385 static bool
2386 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2387 {
2388         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2389                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2390                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2391                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2392                 return false;
2393         /* XXX: check that cr_targ_princ fields match ? */
2394         if (cr1->cr_principal == cr2->cr_principal)
2395                 return true;
2396         if (!cr1->cr_principal || !cr2->cr_principal)
2397                 return false;
2398         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2399 }
2400
2401 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2402 {
2403         struct svc_cred *cr = &rqstp->rq_cred;
2404         u32 service;
2405
2406         if (!cr->cr_gss_mech)
2407                 return false;
2408         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2409         return service == RPC_GSS_SVC_INTEGRITY ||
2410                service == RPC_GSS_SVC_PRIVACY;
2411 }
2412
2413 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2414 {
2415         struct svc_cred *cr = &rqstp->rq_cred;
2416
2417         if (!cl->cl_mach_cred)
2418                 return true;
2419         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2420                 return false;
2421         if (!svc_rqst_integrity_protected(rqstp))
2422                 return false;
2423         if (cl->cl_cred.cr_raw_principal)
2424                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2425                                                 cr->cr_raw_principal);
2426         if (!cr->cr_principal)
2427                 return false;
2428         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2429 }
2430
2431 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2432 {
2433         __be32 verf[2];
2434
2435         /*
2436          * This is opaque to client, so no need to byte-swap. Use
2437          * __force to keep sparse happy
2438          */
2439         verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2440         verf[1] = (__force __be32)nn->clverifier_counter++;
2441         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2442 }
2443
2444 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2445 {
2446         clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2447         clp->cl_clientid.cl_id = nn->clientid_counter++;
2448         gen_confirm(clp, nn);
2449 }
2450
2451 static struct nfs4_stid *
2452 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2453 {
2454         struct nfs4_stid *ret;
2455
2456         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2457         if (!ret || !ret->sc_type)
2458                 return NULL;
2459         return ret;
2460 }
2461
2462 static struct nfs4_stid *
2463 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2464 {
2465         struct nfs4_stid *s;
2466
2467         spin_lock(&cl->cl_lock);
2468         s = find_stateid_locked(cl, t);
2469         if (s != NULL) {
2470                 if (typemask & s->sc_type)
2471                         refcount_inc(&s->sc_count);
2472                 else
2473                         s = NULL;
2474         }
2475         spin_unlock(&cl->cl_lock);
2476         return s;
2477 }
2478
2479 static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2480 {
2481         struct nfsdfs_client *nc;
2482         nc = get_nfsdfs_client(inode);
2483         if (!nc)
2484                 return NULL;
2485         return container_of(nc, struct nfs4_client, cl_nfsdfs);
2486 }
2487
2488 static void seq_quote_mem(struct seq_file *m, char *data, int len)
2489 {
2490         seq_printf(m, "\"");
2491         seq_escape_mem(m, data, len, ESCAPE_HEX | ESCAPE_NAP | ESCAPE_APPEND, "\"\\");
2492         seq_printf(m, "\"");
2493 }
2494
2495 static const char *cb_state2str(int state)
2496 {
2497         switch (state) {
2498         case NFSD4_CB_UP:
2499                 return "UP";
2500         case NFSD4_CB_UNKNOWN:
2501                 return "UNKNOWN";
2502         case NFSD4_CB_DOWN:
2503                 return "DOWN";
2504         case NFSD4_CB_FAULT:
2505                 return "FAULT";
2506         }
2507         return "UNDEFINED";
2508 }
2509
2510 static int client_info_show(struct seq_file *m, void *v)
2511 {
2512         struct inode *inode = file_inode(m->file);
2513         struct nfs4_client *clp;
2514         u64 clid;
2515
2516         clp = get_nfsdfs_clp(inode);
2517         if (!clp)
2518                 return -ENXIO;
2519         memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2520         seq_printf(m, "clientid: 0x%llx\n", clid);
2521         seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2522
2523         if (clp->cl_state == NFSD4_COURTESY)
2524                 seq_puts(m, "status: courtesy\n");
2525         else if (clp->cl_state == NFSD4_EXPIRABLE)
2526                 seq_puts(m, "status: expirable\n");
2527         else if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2528                 seq_puts(m, "status: confirmed\n");
2529         else
2530                 seq_puts(m, "status: unconfirmed\n");
2531         seq_printf(m, "seconds from last renew: %lld\n",
2532                 ktime_get_boottime_seconds() - clp->cl_time);
2533         seq_printf(m, "name: ");
2534         seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2535         seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2536         if (clp->cl_nii_domain.data) {
2537                 seq_printf(m, "Implementation domain: ");
2538                 seq_quote_mem(m, clp->cl_nii_domain.data,
2539                                         clp->cl_nii_domain.len);
2540                 seq_printf(m, "\nImplementation name: ");
2541                 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2542                 seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2543                         clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2544         }
2545         seq_printf(m, "callback state: %s\n", cb_state2str(clp->cl_cb_state));
2546         seq_printf(m, "callback address: %pISpc\n", &clp->cl_cb_conn.cb_addr);
2547         drop_client(clp);
2548
2549         return 0;
2550 }
2551
2552 DEFINE_SHOW_ATTRIBUTE(client_info);
2553
2554 static void *states_start(struct seq_file *s, loff_t *pos)
2555         __acquires(&clp->cl_lock)
2556 {
2557         struct nfs4_client *clp = s->private;
2558         unsigned long id = *pos;
2559         void *ret;
2560
2561         spin_lock(&clp->cl_lock);
2562         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2563         *pos = id;
2564         return ret;
2565 }
2566
2567 static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2568 {
2569         struct nfs4_client *clp = s->private;
2570         unsigned long id = *pos;
2571         void *ret;
2572
2573         id = *pos;
2574         id++;
2575         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2576         *pos = id;
2577         return ret;
2578 }
2579
2580 static void states_stop(struct seq_file *s, void *v)
2581         __releases(&clp->cl_lock)
2582 {
2583         struct nfs4_client *clp = s->private;
2584
2585         spin_unlock(&clp->cl_lock);
2586 }
2587
2588 static void nfs4_show_fname(struct seq_file *s, struct nfsd_file *f)
2589 {
2590          seq_printf(s, "filename: \"%pD2\"", f->nf_file);
2591 }
2592
2593 static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2594 {
2595         struct inode *inode = file_inode(f->nf_file);
2596
2597         seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2598                                         MAJOR(inode->i_sb->s_dev),
2599                                          MINOR(inode->i_sb->s_dev),
2600                                          inode->i_ino);
2601 }
2602
2603 static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2604 {
2605         seq_printf(s, "owner: ");
2606         seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2607 }
2608
2609 static void nfs4_show_stateid(struct seq_file *s, stateid_t *stid)
2610 {
2611         seq_printf(s, "0x%.8x", stid->si_generation);
2612         seq_printf(s, "%12phN", &stid->si_opaque);
2613 }
2614
2615 static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2616 {
2617         struct nfs4_ol_stateid *ols;
2618         struct nfs4_file *nf;
2619         struct nfsd_file *file;
2620         struct nfs4_stateowner *oo;
2621         unsigned int access, deny;
2622
2623         if (st->sc_type != NFS4_OPEN_STID && st->sc_type != NFS4_LOCK_STID)
2624                 return 0; /* XXX: or SEQ_SKIP? */
2625         ols = openlockstateid(st);
2626         oo = ols->st_stateowner;
2627         nf = st->sc_file;
2628
2629         spin_lock(&nf->fi_lock);
2630         file = find_any_file_locked(nf);
2631         if (!file)
2632                 goto out;
2633
2634         seq_printf(s, "- ");
2635         nfs4_show_stateid(s, &st->sc_stateid);
2636         seq_printf(s, ": { type: open, ");
2637
2638         access = bmap_to_share_mode(ols->st_access_bmap);
2639         deny   = bmap_to_share_mode(ols->st_deny_bmap);
2640
2641         seq_printf(s, "access: %s%s, ",
2642                 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2643                 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2644         seq_printf(s, "deny: %s%s, ",
2645                 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2646                 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2647
2648         nfs4_show_superblock(s, file);
2649         seq_printf(s, ", ");
2650         nfs4_show_fname(s, file);
2651         seq_printf(s, ", ");
2652         nfs4_show_owner(s, oo);
2653         seq_printf(s, " }\n");
2654 out:
2655         spin_unlock(&nf->fi_lock);
2656         return 0;
2657 }
2658
2659 static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2660 {
2661         struct nfs4_ol_stateid *ols;
2662         struct nfs4_file *nf;
2663         struct nfsd_file *file;
2664         struct nfs4_stateowner *oo;
2665
2666         ols = openlockstateid(st);
2667         oo = ols->st_stateowner;
2668         nf = st->sc_file;
2669         spin_lock(&nf->fi_lock);
2670         file = find_any_file_locked(nf);
2671         if (!file)
2672                 goto out;
2673
2674         seq_printf(s, "- ");
2675         nfs4_show_stateid(s, &st->sc_stateid);
2676         seq_printf(s, ": { type: lock, ");
2677
2678         /*
2679          * Note: a lock stateid isn't really the same thing as a lock,
2680          * it's the locking state held by one owner on a file, and there
2681          * may be multiple (or no) lock ranges associated with it.
2682          * (Same for the matter is true of open stateids.)
2683          */
2684
2685         nfs4_show_superblock(s, file);
2686         /* XXX: open stateid? */
2687         seq_printf(s, ", ");
2688         nfs4_show_fname(s, file);
2689         seq_printf(s, ", ");
2690         nfs4_show_owner(s, oo);
2691         seq_printf(s, " }\n");
2692 out:
2693         spin_unlock(&nf->fi_lock);
2694         return 0;
2695 }
2696
2697 static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2698 {
2699         struct nfs4_delegation *ds;
2700         struct nfs4_file *nf;
2701         struct nfsd_file *file;
2702
2703         ds = delegstateid(st);
2704         nf = st->sc_file;
2705         spin_lock(&nf->fi_lock);
2706         file = find_deleg_file_locked(nf);
2707         if (!file)
2708                 goto out;
2709
2710         seq_printf(s, "- ");
2711         nfs4_show_stateid(s, &st->sc_stateid);
2712         seq_printf(s, ": { type: deleg, ");
2713
2714         /* Kinda dead code as long as we only support read delegs: */
2715         seq_printf(s, "access: %s, ",
2716                 ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2717
2718         /* XXX: lease time, whether it's being recalled. */
2719
2720         nfs4_show_superblock(s, file);
2721         seq_printf(s, ", ");
2722         nfs4_show_fname(s, file);
2723         seq_printf(s, " }\n");
2724 out:
2725         spin_unlock(&nf->fi_lock);
2726         return 0;
2727 }
2728
2729 static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2730 {
2731         struct nfs4_layout_stateid *ls;
2732         struct nfsd_file *file;
2733
2734         ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2735         file = ls->ls_file;
2736
2737         seq_printf(s, "- ");
2738         nfs4_show_stateid(s, &st->sc_stateid);
2739         seq_printf(s, ": { type: layout, ");
2740
2741         /* XXX: What else would be useful? */
2742
2743         nfs4_show_superblock(s, file);
2744         seq_printf(s, ", ");
2745         nfs4_show_fname(s, file);
2746         seq_printf(s, " }\n");
2747
2748         return 0;
2749 }
2750
2751 static int states_show(struct seq_file *s, void *v)
2752 {
2753         struct nfs4_stid *st = v;
2754
2755         switch (st->sc_type) {
2756         case NFS4_OPEN_STID:
2757                 return nfs4_show_open(s, st);
2758         case NFS4_LOCK_STID:
2759                 return nfs4_show_lock(s, st);
2760         case NFS4_DELEG_STID:
2761                 return nfs4_show_deleg(s, st);
2762         case NFS4_LAYOUT_STID:
2763                 return nfs4_show_layout(s, st);
2764         default:
2765                 return 0; /* XXX: or SEQ_SKIP? */
2766         }
2767         /* XXX: copy stateids? */
2768 }
2769
2770 static struct seq_operations states_seq_ops = {
2771         .start = states_start,
2772         .next = states_next,
2773         .stop = states_stop,
2774         .show = states_show
2775 };
2776
2777 static int client_states_open(struct inode *inode, struct file *file)
2778 {
2779         struct seq_file *s;
2780         struct nfs4_client *clp;
2781         int ret;
2782
2783         clp = get_nfsdfs_clp(inode);
2784         if (!clp)
2785                 return -ENXIO;
2786
2787         ret = seq_open(file, &states_seq_ops);
2788         if (ret)
2789                 return ret;
2790         s = file->private_data;
2791         s->private = clp;
2792         return 0;
2793 }
2794
2795 static int client_opens_release(struct inode *inode, struct file *file)
2796 {
2797         struct seq_file *m = file->private_data;
2798         struct nfs4_client *clp = m->private;
2799
2800         /* XXX: alternatively, we could get/drop in seq start/stop */
2801         drop_client(clp);
2802         return 0;
2803 }
2804
2805 static const struct file_operations client_states_fops = {
2806         .open           = client_states_open,
2807         .read           = seq_read,
2808         .llseek         = seq_lseek,
2809         .release        = client_opens_release,
2810 };
2811
2812 /*
2813  * Normally we refuse to destroy clients that are in use, but here the
2814  * administrator is telling us to just do it.  We also want to wait
2815  * so the caller has a guarantee that the client's locks are gone by
2816  * the time the write returns:
2817  */
2818 static void force_expire_client(struct nfs4_client *clp)
2819 {
2820         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2821         bool already_expired;
2822
2823         trace_nfsd_clid_admin_expired(&clp->cl_clientid);
2824
2825         spin_lock(&nn->client_lock);
2826         clp->cl_time = 0;
2827         spin_unlock(&nn->client_lock);
2828
2829         wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2830         spin_lock(&nn->client_lock);
2831         already_expired = list_empty(&clp->cl_lru);
2832         if (!already_expired)
2833                 unhash_client_locked(clp);
2834         spin_unlock(&nn->client_lock);
2835
2836         if (!already_expired)
2837                 expire_client(clp);
2838         else
2839                 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2840 }
2841
2842 static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2843                                    size_t size, loff_t *pos)
2844 {
2845         char *data;
2846         struct nfs4_client *clp;
2847
2848         data = simple_transaction_get(file, buf, size);
2849         if (IS_ERR(data))
2850                 return PTR_ERR(data);
2851         if (size != 7 || 0 != memcmp(data, "expire\n", 7))
2852                 return -EINVAL;
2853         clp = get_nfsdfs_clp(file_inode(file));
2854         if (!clp)
2855                 return -ENXIO;
2856         force_expire_client(clp);
2857         drop_client(clp);
2858         return 7;
2859 }
2860
2861 static const struct file_operations client_ctl_fops = {
2862         .write          = client_ctl_write,
2863         .release        = simple_transaction_release,
2864 };
2865
2866 static const struct tree_descr client_files[] = {
2867         [0] = {"info", &client_info_fops, S_IRUSR},
2868         [1] = {"states", &client_states_fops, S_IRUSR},
2869         [2] = {"ctl", &client_ctl_fops, S_IWUSR},
2870         [3] = {""},
2871 };
2872
2873 static struct nfs4_client *create_client(struct xdr_netobj name,
2874                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2875 {
2876         struct nfs4_client *clp;
2877         struct sockaddr *sa = svc_addr(rqstp);
2878         int ret;
2879         struct net *net = SVC_NET(rqstp);
2880         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2881         struct dentry *dentries[ARRAY_SIZE(client_files)];
2882
2883         clp = alloc_client(name, nn);
2884         if (clp == NULL)
2885                 return NULL;
2886
2887         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2888         if (ret) {
2889                 free_client(clp);
2890                 return NULL;
2891         }
2892         gen_clid(clp, nn);
2893         kref_init(&clp->cl_nfsdfs.cl_ref);
2894         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2895         clp->cl_time = ktime_get_boottime_seconds();
2896         clear_bit(0, &clp->cl_cb_slot_busy);
2897         copy_verf(clp, verf);
2898         memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
2899         clp->cl_cb_session = NULL;
2900         clp->net = net;
2901         clp->cl_nfsd_dentry = nfsd_client_mkdir(
2902                 nn, &clp->cl_nfsdfs,
2903                 clp->cl_clientid.cl_id - nn->clientid_base,
2904                 client_files, dentries);
2905         clp->cl_nfsd_info_dentry = dentries[0];
2906         if (!clp->cl_nfsd_dentry) {
2907                 free_client(clp);
2908                 return NULL;
2909         }
2910         return clp;
2911 }
2912
2913 static void
2914 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2915 {
2916         struct rb_node **new = &(root->rb_node), *parent = NULL;
2917         struct nfs4_client *clp;
2918
2919         while (*new) {
2920                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2921                 parent = *new;
2922
2923                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2924                         new = &((*new)->rb_left);
2925                 else
2926                         new = &((*new)->rb_right);
2927         }
2928
2929         rb_link_node(&new_clp->cl_namenode, parent, new);
2930         rb_insert_color(&new_clp->cl_namenode, root);
2931 }
2932
2933 static struct nfs4_client *
2934 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2935 {
2936         int cmp;
2937         struct rb_node *node = root->rb_node;
2938         struct nfs4_client *clp;
2939
2940         while (node) {
2941                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2942                 cmp = compare_blob(&clp->cl_name, name);
2943                 if (cmp > 0)
2944                         node = node->rb_left;
2945                 else if (cmp < 0)
2946                         node = node->rb_right;
2947                 else
2948                         return clp;
2949         }
2950         return NULL;
2951 }
2952
2953 static void
2954 add_to_unconfirmed(struct nfs4_client *clp)
2955 {
2956         unsigned int idhashval;
2957         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2958
2959         lockdep_assert_held(&nn->client_lock);
2960
2961         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2962         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2963         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2964         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2965         renew_client_locked(clp);
2966 }
2967
2968 static void
2969 move_to_confirmed(struct nfs4_client *clp)
2970 {
2971         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2972         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2973
2974         lockdep_assert_held(&nn->client_lock);
2975
2976         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2977         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2978         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2979         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2980         trace_nfsd_clid_confirmed(&clp->cl_clientid);
2981         renew_client_locked(clp);
2982 }
2983
2984 static struct nfs4_client *
2985 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2986 {
2987         struct nfs4_client *clp;
2988         unsigned int idhashval = clientid_hashval(clid->cl_id);
2989
2990         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2991                 if (same_clid(&clp->cl_clientid, clid)) {
2992                         if ((bool)clp->cl_minorversion != sessions)
2993                                 return NULL;
2994                         renew_client_locked(clp);
2995                         return clp;
2996                 }
2997         }
2998         return NULL;
2999 }
3000
3001 static struct nfs4_client *
3002 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3003 {
3004         struct list_head *tbl = nn->conf_id_hashtbl;
3005
3006         lockdep_assert_held(&nn->client_lock);
3007         return find_client_in_id_table(tbl, clid, sessions);
3008 }
3009
3010 static struct nfs4_client *
3011 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3012 {
3013         struct list_head *tbl = nn->unconf_id_hashtbl;
3014
3015         lockdep_assert_held(&nn->client_lock);
3016         return find_client_in_id_table(tbl, clid, sessions);
3017 }
3018
3019 static bool clp_used_exchangeid(struct nfs4_client *clp)
3020 {
3021         return clp->cl_exchange_flags != 0;
3022 }
3023
3024 static struct nfs4_client *
3025 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3026 {
3027         lockdep_assert_held(&nn->client_lock);
3028         return find_clp_in_name_tree(name, &nn->conf_name_tree);
3029 }
3030
3031 static struct nfs4_client *
3032 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3033 {
3034         lockdep_assert_held(&nn->client_lock);
3035         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
3036 }
3037
3038 static void
3039 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
3040 {
3041         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
3042         struct sockaddr *sa = svc_addr(rqstp);
3043         u32 scopeid = rpc_get_scope_id(sa);
3044         unsigned short expected_family;
3045
3046         /* Currently, we only support tcp and tcp6 for the callback channel */
3047         if (se->se_callback_netid_len == 3 &&
3048             !memcmp(se->se_callback_netid_val, "tcp", 3))
3049                 expected_family = AF_INET;
3050         else if (se->se_callback_netid_len == 4 &&
3051                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
3052                 expected_family = AF_INET6;
3053         else
3054                 goto out_err;
3055
3056         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
3057                                             se->se_callback_addr_len,
3058                                             (struct sockaddr *)&conn->cb_addr,
3059                                             sizeof(conn->cb_addr));
3060
3061         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
3062                 goto out_err;
3063
3064         if (conn->cb_addr.ss_family == AF_INET6)
3065                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
3066
3067         conn->cb_prog = se->se_callback_prog;
3068         conn->cb_ident = se->se_callback_ident;
3069         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
3070         trace_nfsd_cb_args(clp, conn);
3071         return;
3072 out_err:
3073         conn->cb_addr.ss_family = AF_UNSPEC;
3074         conn->cb_addrlen = 0;
3075         trace_nfsd_cb_nodelegs(clp);
3076         return;
3077 }
3078
3079 /*
3080  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
3081  */
3082 static void
3083 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
3084 {
3085         struct xdr_buf *buf = resp->xdr->buf;
3086         struct nfsd4_slot *slot = resp->cstate.slot;
3087         unsigned int base;
3088
3089         dprintk("--> %s slot %p\n", __func__, slot);
3090
3091         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
3092         slot->sl_opcnt = resp->opcnt;
3093         slot->sl_status = resp->cstate.status;
3094         free_svc_cred(&slot->sl_cred);
3095         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
3096
3097         if (!nfsd4_cache_this(resp)) {
3098                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
3099                 return;
3100         }
3101         slot->sl_flags |= NFSD4_SLOT_CACHED;
3102
3103         base = resp->cstate.data_offset;
3104         slot->sl_datalen = buf->len - base;
3105         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
3106                 WARN(1, "%s: sessions DRC could not cache compound\n",
3107                      __func__);
3108         return;
3109 }
3110
3111 /*
3112  * Encode the replay sequence operation from the slot values.
3113  * If cachethis is FALSE encode the uncached rep error on the next
3114  * operation which sets resp->p and increments resp->opcnt for
3115  * nfs4svc_encode_compoundres.
3116  *
3117  */
3118 static __be32
3119 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
3120                           struct nfsd4_compoundres *resp)
3121 {
3122         struct nfsd4_op *op;
3123         struct nfsd4_slot *slot = resp->cstate.slot;
3124
3125         /* Encode the replayed sequence operation */
3126         op = &args->ops[resp->opcnt - 1];
3127         nfsd4_encode_operation(resp, op);
3128
3129         if (slot->sl_flags & NFSD4_SLOT_CACHED)
3130                 return op->status;
3131         if (args->opcnt == 1) {
3132                 /*
3133                  * The original operation wasn't a solo sequence--we
3134                  * always cache those--so this retry must not match the
3135                  * original:
3136                  */
3137                 op->status = nfserr_seq_false_retry;
3138         } else {
3139                 op = &args->ops[resp->opcnt++];
3140                 op->status = nfserr_retry_uncached_rep;
3141                 nfsd4_encode_operation(resp, op);
3142         }
3143         return op->status;
3144 }
3145
3146 /*
3147  * The sequence operation is not cached because we can use the slot and
3148  * session values.
3149  */
3150 static __be32
3151 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
3152                          struct nfsd4_sequence *seq)
3153 {
3154         struct nfsd4_slot *slot = resp->cstate.slot;
3155         struct xdr_stream *xdr = resp->xdr;
3156         __be32 *p;
3157         __be32 status;
3158
3159         dprintk("--> %s slot %p\n", __func__, slot);
3160
3161         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
3162         if (status)
3163                 return status;
3164
3165         p = xdr_reserve_space(xdr, slot->sl_datalen);
3166         if (!p) {
3167                 WARN_ON_ONCE(1);
3168                 return nfserr_serverfault;
3169         }
3170         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
3171         xdr_commit_encode(xdr);
3172
3173         resp->opcnt = slot->sl_opcnt;
3174         return slot->sl_status;
3175 }
3176
3177 /*
3178  * Set the exchange_id flags returned by the server.
3179  */
3180 static void
3181 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
3182 {
3183 #ifdef CONFIG_NFSD_PNFS
3184         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
3185 #else
3186         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
3187 #endif
3188
3189         /* Referrals are supported, Migration is not. */
3190         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
3191
3192         /* set the wire flags to return to client. */
3193         clid->flags = new->cl_exchange_flags;
3194 }
3195
3196 static bool client_has_openowners(struct nfs4_client *clp)
3197 {
3198         struct nfs4_openowner *oo;
3199
3200         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
3201                 if (!list_empty(&oo->oo_owner.so_stateids))
3202                         return true;
3203         }
3204         return false;
3205 }
3206
3207 static bool client_has_state(struct nfs4_client *clp)
3208 {
3209         return client_has_openowners(clp)
3210 #ifdef CONFIG_NFSD_PNFS
3211                 || !list_empty(&clp->cl_lo_states)
3212 #endif
3213                 || !list_empty(&clp->cl_delegations)
3214                 || !list_empty(&clp->cl_sessions)
3215                 || !list_empty(&clp->async_copies);
3216 }
3217
3218 static __be32 copy_impl_id(struct nfs4_client *clp,
3219                                 struct nfsd4_exchange_id *exid)
3220 {
3221         if (!exid->nii_domain.data)
3222                 return 0;
3223         xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
3224         if (!clp->cl_nii_domain.data)
3225                 return nfserr_jukebox;
3226         xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
3227         if (!clp->cl_nii_name.data)
3228                 return nfserr_jukebox;
3229         clp->cl_nii_time = exid->nii_time;
3230         return 0;
3231 }
3232
3233 __be32
3234 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3235                 union nfsd4_op_u *u)
3236 {
3237         struct nfsd4_exchange_id *exid = &u->exchange_id;
3238         struct nfs4_client *conf, *new;
3239         struct nfs4_client *unconf = NULL;
3240         __be32 status;
3241         char                    addr_str[INET6_ADDRSTRLEN];
3242         nfs4_verifier           verf = exid->verifier;
3243         struct sockaddr         *sa = svc_addr(rqstp);
3244         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3245         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3246
3247         rpc_ntop(sa, addr_str, sizeof(addr_str));
3248         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3249                 "ip_addr=%s flags %x, spa_how %u\n",
3250                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
3251                 addr_str, exid->flags, exid->spa_how);
3252
3253         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3254                 return nfserr_inval;
3255
3256         new = create_client(exid->clname, rqstp, &verf);
3257         if (new == NULL)
3258                 return nfserr_jukebox;
3259         status = copy_impl_id(new, exid);
3260         if (status)
3261                 goto out_nolock;
3262
3263         switch (exid->spa_how) {
3264         case SP4_MACH_CRED:
3265                 exid->spo_must_enforce[0] = 0;
3266                 exid->spo_must_enforce[1] = (
3267                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
3268                         1 << (OP_EXCHANGE_ID - 32) |
3269                         1 << (OP_CREATE_SESSION - 32) |
3270                         1 << (OP_DESTROY_SESSION - 32) |
3271                         1 << (OP_DESTROY_CLIENTID - 32));
3272
3273                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3274                                         1 << (OP_OPEN_DOWNGRADE) |
3275                                         1 << (OP_LOCKU) |
3276                                         1 << (OP_DELEGRETURN));
3277
3278                 exid->spo_must_allow[1] &= (
3279                                         1 << (OP_TEST_STATEID - 32) |
3280                                         1 << (OP_FREE_STATEID - 32));
3281                 if (!svc_rqst_integrity_protected(rqstp)) {
3282                         status = nfserr_inval;
3283                         goto out_nolock;
3284                 }
3285                 /*
3286                  * Sometimes userspace doesn't give us a principal.
3287                  * Which is a bug, really.  Anyway, we can't enforce
3288                  * MACH_CRED in that case, better to give up now:
3289                  */
3290                 if (!new->cl_cred.cr_principal &&
3291                                         !new->cl_cred.cr_raw_principal) {
3292                         status = nfserr_serverfault;
3293                         goto out_nolock;
3294                 }
3295                 new->cl_mach_cred = true;
3296                 break;
3297         case SP4_NONE:
3298                 break;
3299         default:                                /* checked by xdr code */
3300                 WARN_ON_ONCE(1);
3301                 fallthrough;
3302         case SP4_SSV:
3303                 status = nfserr_encr_alg_unsupp;
3304                 goto out_nolock;
3305         }
3306
3307         /* Cases below refer to rfc 5661 section 18.35.4: */
3308         spin_lock(&nn->client_lock);
3309         conf = find_confirmed_client_by_name(&exid->clname, nn);
3310         if (conf) {
3311                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3312                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3313
3314                 if (update) {
3315                         if (!clp_used_exchangeid(conf)) { /* buggy client */
3316                                 status = nfserr_inval;
3317                                 goto out;
3318                         }
3319                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
3320                                 status = nfserr_wrong_cred;
3321                                 goto out;
3322                         }
3323                         if (!creds_match) { /* case 9 */
3324                                 status = nfserr_perm;
3325                                 goto out;
3326                         }
3327                         if (!verfs_match) { /* case 8 */
3328                                 status = nfserr_not_same;
3329                                 goto out;
3330                         }
3331                         /* case 6 */
3332                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3333                         trace_nfsd_clid_confirmed_r(conf);
3334                         goto out_copy;
3335                 }
3336                 if (!creds_match) { /* case 3 */
3337                         if (client_has_state(conf)) {
3338                                 status = nfserr_clid_inuse;
3339                                 trace_nfsd_clid_cred_mismatch(conf, rqstp);
3340                                 goto out;
3341                         }
3342                         goto out_new;
3343                 }
3344                 if (verfs_match) { /* case 2 */
3345                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3346                         trace_nfsd_clid_confirmed_r(conf);
3347                         goto out_copy;
3348                 }
3349                 /* case 5, client reboot */
3350                 trace_nfsd_clid_verf_mismatch(conf, rqstp, &verf);
3351                 conf = NULL;
3352                 goto out_new;
3353         }
3354
3355         if (update) { /* case 7 */
3356                 status = nfserr_noent;
3357                 goto out;
3358         }
3359
3360         unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
3361         if (unconf) /* case 4, possible retry or client restart */
3362                 unhash_client_locked(unconf);
3363
3364         /* case 1, new owner ID */
3365         trace_nfsd_clid_fresh(new);
3366
3367 out_new:
3368         if (conf) {
3369                 status = mark_client_expired_locked(conf);
3370                 if (status)
3371                         goto out;
3372                 trace_nfsd_clid_replaced(&conf->cl_clientid);
3373         }
3374         new->cl_minorversion = cstate->minorversion;
3375         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3376         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3377
3378         add_to_unconfirmed(new);
3379         swap(new, conf);
3380 out_copy:
3381         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3382         exid->clientid.cl_id = conf->cl_clientid.cl_id;
3383
3384         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3385         nfsd4_set_ex_flags(conf, exid);
3386
3387         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3388                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3389         status = nfs_ok;
3390
3391 out:
3392         spin_unlock(&nn->client_lock);
3393 out_nolock:
3394         if (new)
3395                 expire_client(new);
3396         if (unconf) {
3397                 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
3398                 expire_client(unconf);
3399         }
3400         return status;
3401 }
3402
3403 static __be32
3404 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3405 {
3406         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3407                 slot_seqid);
3408
3409         /* The slot is in use, and no response has been sent. */
3410         if (slot_inuse) {
3411                 if (seqid == slot_seqid)
3412                         return nfserr_jukebox;
3413                 else
3414                         return nfserr_seq_misordered;
3415         }
3416         /* Note unsigned 32-bit arithmetic handles wraparound: */
3417         if (likely(seqid == slot_seqid + 1))
3418                 return nfs_ok;
3419         if (seqid == slot_seqid)
3420                 return nfserr_replay_cache;
3421         return nfserr_seq_misordered;
3422 }
3423
3424 /*
3425  * Cache the create session result into the create session single DRC
3426  * slot cache by saving the xdr structure. sl_seqid has been set.
3427  * Do this for solo or embedded create session operations.
3428  */
3429 static void
3430 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3431                            struct nfsd4_clid_slot *slot, __be32 nfserr)
3432 {
3433         slot->sl_status = nfserr;
3434         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3435 }
3436
3437 static __be32
3438 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3439                             struct nfsd4_clid_slot *slot)
3440 {
3441         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3442         return slot->sl_status;
3443 }
3444
3445 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3446                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3447                         1 +     /* MIN tag is length with zero, only length */ \
3448                         3 +     /* version, opcount, opcode */ \
3449                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3450                                 /* seqid, slotID, slotID, cache */ \
3451                         4 ) * sizeof(__be32))
3452
3453 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3454                         2 +     /* verifier: AUTH_NULL, length 0 */\
3455                         1 +     /* status */ \
3456                         1 +     /* MIN tag is length with zero, only length */ \
3457                         3 +     /* opcount, opcode, opstatus*/ \
3458                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3459                                 /* seqid, slotID, slotID, slotID, status */ \
3460                         5 ) * sizeof(__be32))
3461
3462 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3463 {
3464         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3465
3466         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3467                 return nfserr_toosmall;
3468         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3469                 return nfserr_toosmall;
3470         ca->headerpadsz = 0;
3471         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3472         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3473         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3474         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3475                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3476         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3477         /*
3478          * Note decreasing slot size below client's request may make it
3479          * difficult for client to function correctly, whereas
3480          * decreasing the number of slots will (just?) affect
3481          * performance.  When short on memory we therefore prefer to
3482          * decrease number of slots instead of their size.  Clients that
3483          * request larger slots than they need will get poor results:
3484          * Note that we always allow at least one slot, because our
3485          * accounting is soft and provides no guarantees either way.
3486          */
3487         ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3488
3489         return nfs_ok;
3490 }
3491
3492 /*
3493  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3494  * These are based on similar macros in linux/sunrpc/msg_prot.h .
3495  */
3496 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3497         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3498
3499 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3500         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3501
3502 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
3503                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3504 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
3505                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3506                                  sizeof(__be32))
3507
3508 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3509 {
3510         ca->headerpadsz = 0;
3511
3512         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3513                 return nfserr_toosmall;
3514         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3515                 return nfserr_toosmall;
3516         ca->maxresp_cached = 0;
3517         if (ca->maxops < 2)
3518                 return nfserr_toosmall;
3519
3520         return nfs_ok;
3521 }
3522
3523 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3524 {
3525         switch (cbs->flavor) {
3526         case RPC_AUTH_NULL:
3527         case RPC_AUTH_UNIX:
3528                 return nfs_ok;
3529         default:
3530                 /*
3531                  * GSS case: the spec doesn't allow us to return this
3532                  * error.  But it also doesn't allow us not to support
3533                  * GSS.
3534                  * I'd rather this fail hard than return some error the
3535                  * client might think it can already handle:
3536                  */
3537                 return nfserr_encr_alg_unsupp;
3538         }
3539 }
3540
3541 __be32
3542 nfsd4_create_session(struct svc_rqst *rqstp,
3543                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3544 {
3545         struct nfsd4_create_session *cr_ses = &u->create_session;
3546         struct sockaddr *sa = svc_addr(rqstp);
3547         struct nfs4_client *conf, *unconf;
3548         struct nfs4_client *old = NULL;
3549         struct nfsd4_session *new;
3550         struct nfsd4_conn *conn;
3551         struct nfsd4_clid_slot *cs_slot = NULL;
3552         __be32 status = 0;
3553         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3554
3555         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3556                 return nfserr_inval;
3557         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3558         if (status)
3559                 return status;
3560         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3561         if (status)
3562                 return status;
3563         status = check_backchannel_attrs(&cr_ses->back_channel);
3564         if (status)
3565                 goto out_release_drc_mem;
3566         status = nfserr_jukebox;
3567         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3568         if (!new)
3569                 goto out_release_drc_mem;
3570         conn = alloc_conn_from_crses(rqstp, cr_ses);
3571         if (!conn)
3572                 goto out_free_session;
3573
3574         spin_lock(&nn->client_lock);
3575         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3576         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3577         WARN_ON_ONCE(conf && unconf);
3578
3579         if (conf) {
3580                 status = nfserr_wrong_cred;
3581                 if (!nfsd4_mach_creds_match(conf, rqstp))
3582                         goto out_free_conn;
3583                 cs_slot = &conf->cl_cs_slot;
3584                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3585                 if (status) {
3586                         if (status == nfserr_replay_cache)
3587                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3588                         goto out_free_conn;
3589                 }
3590         } else if (unconf) {
3591                 status = nfserr_clid_inuse;
3592                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3593                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3594                         trace_nfsd_clid_cred_mismatch(unconf, rqstp);
3595                         goto out_free_conn;
3596                 }
3597                 status = nfserr_wrong_cred;
3598                 if (!nfsd4_mach_creds_match(unconf, rqstp))
3599                         goto out_free_conn;
3600                 cs_slot = &unconf->cl_cs_slot;
3601                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3602                 if (status) {
3603                         /* an unconfirmed replay returns misordered */
3604                         status = nfserr_seq_misordered;
3605                         goto out_free_conn;
3606                 }
3607                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3608                 if (old) {
3609                         status = mark_client_expired_locked(old);
3610                         if (status) {
3611                                 old = NULL;
3612                                 goto out_free_conn;
3613                         }
3614                         trace_nfsd_clid_replaced(&old->cl_clientid);
3615                 }
3616                 move_to_confirmed(unconf);
3617                 conf = unconf;
3618         } else {
3619                 status = nfserr_stale_clientid;
3620                 goto out_free_conn;
3621         }
3622         status = nfs_ok;
3623         /* Persistent sessions are not supported */
3624         cr_ses->flags &= ~SESSION4_PERSIST;
3625         /* Upshifting from TCP to RDMA is not supported */
3626         cr_ses->flags &= ~SESSION4_RDMA;
3627
3628         init_session(rqstp, new, conf, cr_ses);
3629         nfsd4_get_session_locked(new);
3630
3631         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3632                NFS4_MAX_SESSIONID_LEN);
3633         cs_slot->sl_seqid++;
3634         cr_ses->seqid = cs_slot->sl_seqid;
3635
3636         /* cache solo and embedded create sessions under the client_lock */
3637         nfsd4_cache_create_session(cr_ses, cs_slot, status);
3638         spin_unlock(&nn->client_lock);
3639         if (conf == unconf)
3640                 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
3641         /* init connection and backchannel */
3642         nfsd4_init_conn(rqstp, conn, new);
3643         nfsd4_put_session(new);
3644         if (old)
3645                 expire_client(old);
3646         return status;
3647 out_free_conn:
3648         spin_unlock(&nn->client_lock);
3649         free_conn(conn);
3650         if (old)
3651                 expire_client(old);
3652 out_free_session:
3653         __free_session(new);
3654 out_release_drc_mem:
3655         nfsd4_put_drc_mem(&cr_ses->fore_channel);
3656         return status;
3657 }
3658
3659 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3660 {
3661         switch (*dir) {
3662         case NFS4_CDFC4_FORE:
3663         case NFS4_CDFC4_BACK:
3664                 return nfs_ok;
3665         case NFS4_CDFC4_FORE_OR_BOTH:
3666         case NFS4_CDFC4_BACK_OR_BOTH:
3667                 *dir = NFS4_CDFC4_BOTH;
3668                 return nfs_ok;
3669         }
3670         return nfserr_inval;
3671 }
3672
3673 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3674                 struct nfsd4_compound_state *cstate,
3675                 union nfsd4_op_u *u)
3676 {
3677         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3678         struct nfsd4_session *session = cstate->session;
3679         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3680         __be32 status;
3681
3682         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3683         if (status)
3684                 return status;
3685         spin_lock(&nn->client_lock);
3686         session->se_cb_prog = bc->bc_cb_program;
3687         session->se_cb_sec = bc->bc_cb_sec;
3688         spin_unlock(&nn->client_lock);
3689
3690         nfsd4_probe_callback(session->se_client);
3691
3692         return nfs_ok;
3693 }
3694
3695 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3696 {
3697         struct nfsd4_conn *c;
3698
3699         list_for_each_entry(c, &s->se_conns, cn_persession) {
3700                 if (c->cn_xprt == xpt) {
3701                         return c;
3702                 }
3703         }
3704         return NULL;
3705 }
3706
3707 static __be32 nfsd4_match_existing_connection(struct svc_rqst *rqst,
3708                 struct nfsd4_session *session, u32 req, struct nfsd4_conn **conn)
3709 {
3710         struct nfs4_client *clp = session->se_client;
3711         struct svc_xprt *xpt = rqst->rq_xprt;
3712         struct nfsd4_conn *c;
3713         __be32 status;
3714
3715         /* Following the last paragraph of RFC 5661 Section 18.34.3: */
3716         spin_lock(&clp->cl_lock);
3717         c = __nfsd4_find_conn(xpt, session);
3718         if (!c)
3719                 status = nfserr_noent;
3720         else if (req == c->cn_flags)
3721                 status = nfs_ok;
3722         else if (req == NFS4_CDFC4_FORE_OR_BOTH &&
3723                                 c->cn_flags != NFS4_CDFC4_BACK)
3724                 status = nfs_ok;
3725         else if (req == NFS4_CDFC4_BACK_OR_BOTH &&
3726                                 c->cn_flags != NFS4_CDFC4_FORE)
3727                 status = nfs_ok;
3728         else
3729                 status = nfserr_inval;
3730         spin_unlock(&clp->cl_lock);
3731         if (status == nfs_ok && conn)
3732                 *conn = c;
3733         return status;
3734 }
3735
3736 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
3737                      struct nfsd4_compound_state *cstate,
3738                      union nfsd4_op_u *u)
3739 {
3740         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
3741         __be32 status;
3742         struct nfsd4_conn *conn;
3743         struct nfsd4_session *session;
3744         struct net *net = SVC_NET(rqstp);
3745         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3746
3747         if (!nfsd4_last_compound_op(rqstp))
3748                 return nfserr_not_only_op;
3749         spin_lock(&nn->client_lock);
3750         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
3751         spin_unlock(&nn->client_lock);
3752         if (!session)
3753                 goto out_no_session;
3754         status = nfserr_wrong_cred;
3755         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
3756                 goto out;
3757         status = nfsd4_match_existing_connection(rqstp, session,
3758                         bcts->dir, &conn);
3759         if (status == nfs_ok) {
3760                 if (bcts->dir == NFS4_CDFC4_FORE_OR_BOTH ||
3761                                 bcts->dir == NFS4_CDFC4_BACK)
3762                         conn->cn_flags |= NFS4_CDFC4_BACK;
3763                 nfsd4_probe_callback(session->se_client);
3764                 goto out;
3765         }
3766         if (status == nfserr_inval)
3767                 goto out;
3768         status = nfsd4_map_bcts_dir(&bcts->dir);
3769         if (status)
3770                 goto out;
3771         conn = alloc_conn(rqstp, bcts->dir);
3772         status = nfserr_jukebox;
3773         if (!conn)
3774                 goto out;
3775         nfsd4_init_conn(rqstp, conn, session);
3776         status = nfs_ok;
3777 out:
3778         nfsd4_put_session(session);
3779 out_no_session:
3780         return status;
3781 }
3782
3783 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3784 {
3785         if (!cstate->session)
3786                 return false;
3787         return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3788 }
3789
3790 __be32
3791 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3792                 union nfsd4_op_u *u)
3793 {
3794         struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3795         struct nfsd4_session *ses;
3796         __be32 status;
3797         int ref_held_by_me = 0;
3798         struct net *net = SVC_NET(r);
3799         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3800
3801         status = nfserr_not_only_op;
3802         if (nfsd4_compound_in_session(cstate, sessionid)) {
3803                 if (!nfsd4_last_compound_op(r))
3804                         goto out;
3805                 ref_held_by_me++;
3806         }
3807         dump_sessionid(__func__, sessionid);
3808         spin_lock(&nn->client_lock);
3809         ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3810         if (!ses)
3811                 goto out_client_lock;
3812         status = nfserr_wrong_cred;
3813         if (!nfsd4_mach_creds_match(ses->se_client, r))
3814                 goto out_put_session;
3815         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3816         if (status)
3817                 goto out_put_session;
3818         unhash_session(ses);
3819         spin_unlock(&nn->client_lock);
3820
3821         nfsd4_probe_callback_sync(ses->se_client);
3822
3823         spin_lock(&nn->client_lock);
3824         status = nfs_ok;
3825 out_put_session:
3826         nfsd4_put_session_locked(ses);
3827 out_client_lock:
3828         spin_unlock(&nn->client_lock);
3829 out:
3830         return status;
3831 }
3832
3833 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3834 {
3835         struct nfs4_client *clp = ses->se_client;
3836         struct nfsd4_conn *c;
3837         __be32 status = nfs_ok;
3838         int ret;
3839
3840         spin_lock(&clp->cl_lock);
3841         c = __nfsd4_find_conn(new->cn_xprt, ses);
3842         if (c)
3843                 goto out_free;
3844         status = nfserr_conn_not_bound_to_session;
3845         if (clp->cl_mach_cred)
3846                 goto out_free;
3847         __nfsd4_hash_conn(new, ses);
3848         spin_unlock(&clp->cl_lock);
3849         ret = nfsd4_register_conn(new);
3850         if (ret)
3851                 /* oops; xprt is already down: */
3852                 nfsd4_conn_lost(&new->cn_xpt_user);
3853         return nfs_ok;
3854 out_free:
3855         spin_unlock(&clp->cl_lock);
3856         free_conn(new);
3857         return status;
3858 }
3859
3860 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3861 {
3862         struct nfsd4_compoundargs *args = rqstp->rq_argp;
3863
3864         return args->opcnt > session->se_fchannel.maxops;
3865 }
3866
3867 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3868                                   struct nfsd4_session *session)
3869 {
3870         struct xdr_buf *xb = &rqstp->rq_arg;
3871
3872         return xb->len > session->se_fchannel.maxreq_sz;
3873 }
3874
3875 static bool replay_matches_cache(struct svc_rqst *rqstp,
3876                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3877 {
3878         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3879
3880         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3881             (bool)seq->cachethis)
3882                 return false;
3883         /*
3884          * If there's an error then the reply can have fewer ops than
3885          * the call.
3886          */
3887         if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
3888                 return false;
3889         /*
3890          * But if we cached a reply with *more* ops than the call you're
3891          * sending us now, then this new call is clearly not really a
3892          * replay of the old one:
3893          */
3894         if (slot->sl_opcnt > argp->opcnt)
3895                 return false;
3896         /* This is the only check explicitly called by spec: */
3897         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3898                 return false;
3899         /*
3900          * There may be more comparisons we could actually do, but the
3901          * spec doesn't require us to catch every case where the calls
3902          * don't match (that would require caching the call as well as
3903          * the reply), so we don't bother.
3904          */
3905         return true;
3906 }
3907
3908 __be32
3909 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3910                 union nfsd4_op_u *u)
3911 {
3912         struct nfsd4_sequence *seq = &u->sequence;
3913         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3914         struct xdr_stream *xdr = resp->xdr;
3915         struct nfsd4_session *session;
3916         struct nfs4_client *clp;
3917         struct nfsd4_slot *slot;
3918         struct nfsd4_conn *conn;
3919         __be32 status;
3920         int buflen;
3921         struct net *net = SVC_NET(rqstp);
3922         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3923
3924         if (resp->opcnt != 1)
3925                 return nfserr_sequence_pos;
3926
3927         /*
3928          * Will be either used or freed by nfsd4_sequence_check_conn
3929          * below.
3930          */
3931         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3932         if (!conn)
3933                 return nfserr_jukebox;
3934
3935         spin_lock(&nn->client_lock);
3936         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3937         if (!session)
3938                 goto out_no_session;
3939         clp = session->se_client;
3940
3941         status = nfserr_too_many_ops;
3942         if (nfsd4_session_too_many_ops(rqstp, session))
3943                 goto out_put_session;
3944
3945         status = nfserr_req_too_big;
3946         if (nfsd4_request_too_big(rqstp, session))
3947                 goto out_put_session;
3948
3949         status = nfserr_badslot;
3950         if (seq->slotid >= session->se_fchannel.maxreqs)
3951                 goto out_put_session;
3952
3953         slot = session->se_slots[seq->slotid];
3954         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3955
3956         /* We do not negotiate the number of slots yet, so set the
3957          * maxslots to the session maxreqs which is used to encode
3958          * sr_highest_slotid and the sr_target_slot id to maxslots */
3959         seq->maxslots = session->se_fchannel.maxreqs;
3960
3961         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3962                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3963         if (status == nfserr_replay_cache) {
3964                 status = nfserr_seq_misordered;
3965                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3966                         goto out_put_session;
3967                 status = nfserr_seq_false_retry;
3968                 if (!replay_matches_cache(rqstp, seq, slot))
3969                         goto out_put_session;
3970                 cstate->slot = slot;
3971                 cstate->session = session;
3972                 cstate->clp = clp;
3973                 /* Return the cached reply status and set cstate->status
3974                  * for nfsd4_proc_compound processing */
3975                 status = nfsd4_replay_cache_entry(resp, seq);
3976                 cstate->status = nfserr_replay_cache;
3977                 goto out;
3978         }
3979         if (status)
3980                 goto out_put_session;
3981
3982         status = nfsd4_sequence_check_conn(conn, session);
3983         conn = NULL;
3984         if (status)
3985                 goto out_put_session;
3986
3987         buflen = (seq->cachethis) ?
3988                         session->se_fchannel.maxresp_cached :
3989                         session->se_fchannel.maxresp_sz;
3990         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3991                                     nfserr_rep_too_big;
3992         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3993                 goto out_put_session;
3994         svc_reserve(rqstp, buflen);
3995
3996         status = nfs_ok;
3997         /* Success! bump slot seqid */
3998         slot->sl_seqid = seq->seqid;
3999         slot->sl_flags |= NFSD4_SLOT_INUSE;
4000         if (seq->cachethis)
4001                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
4002         else
4003                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
4004
4005         cstate->slot = slot;
4006         cstate->session = session;
4007         cstate->clp = clp;
4008
4009 out:
4010         switch (clp->cl_cb_state) {
4011         case NFSD4_CB_DOWN:
4012                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
4013                 break;
4014         case NFSD4_CB_FAULT:
4015                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
4016                 break;
4017         default:
4018                 seq->status_flags = 0;
4019         }
4020         if (!list_empty(&clp->cl_revoked))
4021                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
4022 out_no_session:
4023         if (conn)
4024                 free_conn(conn);
4025         spin_unlock(&nn->client_lock);
4026         return status;
4027 out_put_session:
4028         nfsd4_put_session_locked(session);
4029         goto out_no_session;
4030 }
4031
4032 void
4033 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
4034 {
4035         struct nfsd4_compound_state *cs = &resp->cstate;
4036
4037         if (nfsd4_has_session(cs)) {
4038                 if (cs->status != nfserr_replay_cache) {
4039                         nfsd4_store_cache_entry(resp);
4040                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
4041                 }
4042                 /* Drop session reference that was taken in nfsd4_sequence() */
4043                 nfsd4_put_session(cs->session);
4044         } else if (cs->clp)
4045                 put_client_renew(cs->clp);
4046 }
4047
4048 __be32
4049 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
4050                 struct nfsd4_compound_state *cstate,
4051                 union nfsd4_op_u *u)
4052 {
4053         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
4054         struct nfs4_client *conf, *unconf;
4055         struct nfs4_client *clp = NULL;
4056         __be32 status = 0;
4057         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4058
4059         spin_lock(&nn->client_lock);
4060         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
4061         conf = find_confirmed_client(&dc->clientid, true, nn);
4062         WARN_ON_ONCE(conf && unconf);
4063
4064         if (conf) {
4065                 if (client_has_state(conf)) {
4066                         status = nfserr_clientid_busy;
4067                         goto out;
4068                 }
4069                 status = mark_client_expired_locked(conf);
4070                 if (status)
4071                         goto out;
4072                 clp = conf;
4073         } else if (unconf)
4074                 clp = unconf;
4075         else {
4076                 status = nfserr_stale_clientid;
4077                 goto out;
4078         }
4079         if (!nfsd4_mach_creds_match(clp, rqstp)) {
4080                 clp = NULL;
4081                 status = nfserr_wrong_cred;
4082                 goto out;
4083         }
4084         trace_nfsd_clid_destroyed(&clp->cl_clientid);
4085         unhash_client_locked(clp);
4086 out:
4087         spin_unlock(&nn->client_lock);
4088         if (clp)
4089                 expire_client(clp);
4090         return status;
4091 }
4092
4093 __be32
4094 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
4095                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
4096 {
4097         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
4098         struct nfs4_client *clp = cstate->clp;
4099         __be32 status = 0;
4100
4101         if (rc->rca_one_fs) {
4102                 if (!cstate->current_fh.fh_dentry)
4103                         return nfserr_nofilehandle;
4104                 /*
4105                  * We don't take advantage of the rca_one_fs case.
4106                  * That's OK, it's optional, we can safely ignore it.
4107                  */
4108                 return nfs_ok;
4109         }
4110
4111         status = nfserr_complete_already;
4112         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
4113                 goto out;
4114
4115         status = nfserr_stale_clientid;
4116         if (is_client_expired(clp))
4117                 /*
4118                  * The following error isn't really legal.
4119                  * But we only get here if the client just explicitly
4120                  * destroyed the client.  Surely it no longer cares what
4121                  * error it gets back on an operation for the dead
4122                  * client.
4123                  */
4124                 goto out;
4125
4126         status = nfs_ok;
4127         trace_nfsd_clid_reclaim_complete(&clp->cl_clientid);
4128         nfsd4_client_record_create(clp);
4129         inc_reclaim_complete(clp);
4130 out:
4131         return status;
4132 }
4133
4134 __be32
4135 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4136                   union nfsd4_op_u *u)
4137 {
4138         struct nfsd4_setclientid *setclid = &u->setclientid;
4139         struct xdr_netobj       clname = setclid->se_name;
4140         nfs4_verifier           clverifier = setclid->se_verf;
4141         struct nfs4_client      *conf, *new;
4142         struct nfs4_client      *unconf = NULL;
4143         __be32                  status;
4144         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4145
4146         new = create_client(clname, rqstp, &clverifier);
4147         if (new == NULL)
4148                 return nfserr_jukebox;
4149         spin_lock(&nn->client_lock);
4150         conf = find_confirmed_client_by_name(&clname, nn);
4151         if (conf && client_has_state(conf)) {
4152                 status = nfserr_clid_inuse;
4153                 if (clp_used_exchangeid(conf))
4154                         goto out;
4155                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4156                         trace_nfsd_clid_cred_mismatch(conf, rqstp);
4157                         goto out;
4158                 }
4159         }
4160         unconf = find_unconfirmed_client_by_name(&clname, nn);
4161         if (unconf)
4162                 unhash_client_locked(unconf);
4163         if (conf) {
4164                 if (same_verf(&conf->cl_verifier, &clverifier)) {
4165                         copy_clid(new, conf);
4166                         gen_confirm(new, nn);
4167                 } else
4168                         trace_nfsd_clid_verf_mismatch(conf, rqstp,
4169                                                       &clverifier);
4170         } else
4171                 trace_nfsd_clid_fresh(new);
4172         new->cl_minorversion = 0;
4173         gen_callback(new, setclid, rqstp);
4174         add_to_unconfirmed(new);
4175         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
4176         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
4177         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
4178         new = NULL;
4179         status = nfs_ok;
4180 out:
4181         spin_unlock(&nn->client_lock);
4182         if (new)
4183                 free_client(new);
4184         if (unconf) {
4185                 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
4186                 expire_client(unconf);
4187         }
4188         return status;
4189 }
4190
4191 __be32
4192 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
4193                         struct nfsd4_compound_state *cstate,
4194                         union nfsd4_op_u *u)
4195 {
4196         struct nfsd4_setclientid_confirm *setclientid_confirm =
4197                         &u->setclientid_confirm;
4198         struct nfs4_client *conf, *unconf;
4199         struct nfs4_client *old = NULL;
4200         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
4201         clientid_t * clid = &setclientid_confirm->sc_clientid;
4202         __be32 status;
4203         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4204
4205         if (STALE_CLIENTID(clid, nn))
4206                 return nfserr_stale_clientid;
4207
4208         spin_lock(&nn->client_lock);
4209         conf = find_confirmed_client(clid, false, nn);
4210         unconf = find_unconfirmed_client(clid, false, nn);
4211         /*
4212          * We try hard to give out unique clientid's, so if we get an
4213          * attempt to confirm the same clientid with a different cred,
4214          * the client may be buggy; this should never happen.
4215          *
4216          * Nevertheless, RFC 7530 recommends INUSE for this case:
4217          */
4218         status = nfserr_clid_inuse;
4219         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
4220                 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
4221                 goto out;
4222         }
4223         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4224                 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4225                 goto out;
4226         }
4227         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
4228                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
4229                         status = nfs_ok;
4230                 } else
4231                         status = nfserr_stale_clientid;
4232                 goto out;
4233         }
4234         status = nfs_ok;
4235         if (conf) {
4236                 old = unconf;
4237                 unhash_client_locked(old);
4238                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
4239         } else {
4240                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
4241                 if (old) {
4242                         status = nfserr_clid_inuse;
4243                         if (client_has_state(old)
4244                                         && !same_creds(&unconf->cl_cred,
4245                                                         &old->cl_cred)) {
4246                                 old = NULL;
4247                                 goto out;
4248                         }
4249                         status = mark_client_expired_locked(old);
4250                         if (status) {
4251                                 old = NULL;
4252                                 goto out;
4253                         }
4254                         trace_nfsd_clid_replaced(&old->cl_clientid);
4255                 }
4256                 move_to_confirmed(unconf);
4257                 conf = unconf;
4258         }
4259         get_client_locked(conf);
4260         spin_unlock(&nn->client_lock);
4261         if (conf == unconf)
4262                 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
4263         nfsd4_probe_callback(conf);
4264         spin_lock(&nn->client_lock);
4265         put_client_renew_locked(conf);
4266 out:
4267         spin_unlock(&nn->client_lock);
4268         if (old)
4269                 expire_client(old);
4270         return status;
4271 }
4272
4273 static struct nfs4_file *nfsd4_alloc_file(void)
4274 {
4275         return kmem_cache_alloc(file_slab, GFP_KERNEL);
4276 }
4277
4278 /* OPEN Share state helper functions */
4279 static void nfsd4_init_file(struct svc_fh *fh, unsigned int hashval,
4280                                 struct nfs4_file *fp)
4281 {
4282         lockdep_assert_held(&state_lock);
4283
4284         refcount_set(&fp->fi_ref, 1);
4285         spin_lock_init(&fp->fi_lock);
4286         INIT_LIST_HEAD(&fp->fi_stateids);
4287         INIT_LIST_HEAD(&fp->fi_delegations);
4288         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4289         fh_copy_shallow(&fp->fi_fhandle, &fh->fh_handle);
4290         fp->fi_deleg_file = NULL;
4291         fp->fi_had_conflict = false;
4292         fp->fi_share_deny = 0;
4293         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4294         memset(fp->fi_access, 0, sizeof(fp->fi_access));
4295         fp->fi_aliased = false;
4296         fp->fi_inode = d_inode(fh->fh_dentry);
4297 #ifdef CONFIG_NFSD_PNFS
4298         INIT_LIST_HEAD(&fp->fi_lo_states);
4299         atomic_set(&fp->fi_lo_recalls, 0);
4300 #endif
4301         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
4302 }
4303
4304 void
4305 nfsd4_free_slabs(void)
4306 {
4307         kmem_cache_destroy(client_slab);
4308         kmem_cache_destroy(openowner_slab);
4309         kmem_cache_destroy(lockowner_slab);
4310         kmem_cache_destroy(file_slab);
4311         kmem_cache_destroy(stateid_slab);
4312         kmem_cache_destroy(deleg_slab);
4313         kmem_cache_destroy(odstate_slab);
4314 }
4315
4316 int
4317 nfsd4_init_slabs(void)
4318 {
4319         client_slab = kmem_cache_create("nfsd4_clients",
4320                         sizeof(struct nfs4_client), 0, 0, NULL);
4321         if (client_slab == NULL)
4322                 goto out;
4323         openowner_slab = kmem_cache_create("nfsd4_openowners",
4324                         sizeof(struct nfs4_openowner), 0, 0, NULL);
4325         if (openowner_slab == NULL)
4326                 goto out_free_client_slab;
4327         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
4328                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
4329         if (lockowner_slab == NULL)
4330                 goto out_free_openowner_slab;
4331         file_slab = kmem_cache_create("nfsd4_files",
4332                         sizeof(struct nfs4_file), 0, 0, NULL);
4333         if (file_slab == NULL)
4334                 goto out_free_lockowner_slab;
4335         stateid_slab = kmem_cache_create("nfsd4_stateids",
4336                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
4337         if (stateid_slab == NULL)
4338                 goto out_free_file_slab;
4339         deleg_slab = kmem_cache_create("nfsd4_delegations",
4340                         sizeof(struct nfs4_delegation), 0, 0, NULL);
4341         if (deleg_slab == NULL)
4342                 goto out_free_stateid_slab;
4343         odstate_slab = kmem_cache_create("nfsd4_odstate",
4344                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
4345         if (odstate_slab == NULL)
4346                 goto out_free_deleg_slab;
4347         return 0;
4348
4349 out_free_deleg_slab:
4350         kmem_cache_destroy(deleg_slab);
4351 out_free_stateid_slab:
4352         kmem_cache_destroy(stateid_slab);
4353 out_free_file_slab:
4354         kmem_cache_destroy(file_slab);
4355 out_free_lockowner_slab:
4356         kmem_cache_destroy(lockowner_slab);
4357 out_free_openowner_slab:
4358         kmem_cache_destroy(openowner_slab);
4359 out_free_client_slab:
4360         kmem_cache_destroy(client_slab);
4361 out:
4362         return -ENOMEM;
4363 }
4364
4365 static unsigned long
4366 nfsd_courtesy_client_count(struct shrinker *shrink, struct shrink_control *sc)
4367 {
4368         int cnt;
4369         struct nfsd_net *nn = container_of(shrink,
4370                         struct nfsd_net, nfsd_client_shrinker);
4371
4372         cnt = atomic_read(&nn->nfsd_courtesy_clients);
4373         if (cnt > 0)
4374                 mod_delayed_work(laundry_wq, &nn->nfsd_shrinker_work, 0);
4375         return (unsigned long)cnt;
4376 }
4377
4378 static unsigned long
4379 nfsd_courtesy_client_scan(struct shrinker *shrink, struct shrink_control *sc)
4380 {
4381         return SHRINK_STOP;
4382 }
4383
4384 int
4385 nfsd4_init_leases_net(struct nfsd_net *nn)
4386 {
4387         struct sysinfo si;
4388         u64 max_clients;
4389
4390         nn->nfsd4_lease = 90;   /* default lease time */
4391         nn->nfsd4_grace = 90;
4392         nn->somebody_reclaimed = false;
4393         nn->track_reclaim_completes = false;
4394         nn->clverifier_counter = get_random_u32();
4395         nn->clientid_base = get_random_u32();
4396         nn->clientid_counter = nn->clientid_base + 1;
4397         nn->s2s_cp_cl_id = nn->clientid_counter++;
4398
4399         atomic_set(&nn->nfs4_client_count, 0);
4400         si_meminfo(&si);
4401         max_clients = (u64)si.totalram * si.mem_unit / (1024 * 1024 * 1024);
4402         max_clients *= NFS4_CLIENTS_PER_GB;
4403         nn->nfs4_max_clients = max_t(int, max_clients, NFS4_CLIENTS_PER_GB);
4404
4405         atomic_set(&nn->nfsd_courtesy_clients, 0);
4406         nn->nfsd_client_shrinker.scan_objects = nfsd_courtesy_client_scan;
4407         nn->nfsd_client_shrinker.count_objects = nfsd_courtesy_client_count;
4408         nn->nfsd_client_shrinker.seeks = DEFAULT_SEEKS;
4409         return register_shrinker(&nn->nfsd_client_shrinker, "nfsd-client");
4410 }
4411
4412 void
4413 nfsd4_leases_net_shutdown(struct nfsd_net *nn)
4414 {
4415         unregister_shrinker(&nn->nfsd_client_shrinker);
4416 }
4417
4418 static void init_nfs4_replay(struct nfs4_replay *rp)
4419 {
4420         rp->rp_status = nfserr_serverfault;
4421         rp->rp_buflen = 0;
4422         rp->rp_buf = rp->rp_ibuf;
4423         mutex_init(&rp->rp_mutex);
4424 }
4425
4426 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4427                 struct nfs4_stateowner *so)
4428 {
4429         if (!nfsd4_has_session(cstate)) {
4430                 mutex_lock(&so->so_replay.rp_mutex);
4431                 cstate->replay_owner = nfs4_get_stateowner(so);
4432         }
4433 }
4434
4435 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4436 {
4437         struct nfs4_stateowner *so = cstate->replay_owner;
4438
4439         if (so != NULL) {
4440                 cstate->replay_owner = NULL;
4441                 mutex_unlock(&so->so_replay.rp_mutex);
4442                 nfs4_put_stateowner(so);
4443         }
4444 }
4445
4446 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4447 {
4448         struct nfs4_stateowner *sop;
4449
4450         sop = kmem_cache_alloc(slab, GFP_KERNEL);
4451         if (!sop)
4452                 return NULL;
4453
4454         xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4455         if (!sop->so_owner.data) {
4456                 kmem_cache_free(slab, sop);
4457                 return NULL;
4458         }
4459
4460         INIT_LIST_HEAD(&sop->so_stateids);
4461         sop->so_client = clp;
4462         init_nfs4_replay(&sop->so_replay);
4463         atomic_set(&sop->so_count, 1);
4464         return sop;
4465 }
4466
4467 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4468 {
4469         lockdep_assert_held(&clp->cl_lock);
4470
4471         list_add(&oo->oo_owner.so_strhash,
4472                  &clp->cl_ownerstr_hashtbl[strhashval]);
4473         list_add(&oo->oo_perclient, &clp->cl_openowners);
4474 }
4475
4476 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4477 {
4478         unhash_openowner_locked(openowner(so));
4479 }
4480
4481 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4482 {
4483         struct nfs4_openowner *oo = openowner(so);
4484
4485         kmem_cache_free(openowner_slab, oo);
4486 }
4487
4488 static const struct nfs4_stateowner_operations openowner_ops = {
4489         .so_unhash =    nfs4_unhash_openowner,
4490         .so_free =      nfs4_free_openowner,
4491 };
4492
4493 static struct nfs4_ol_stateid *
4494 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4495 {
4496         struct nfs4_ol_stateid *local, *ret = NULL;
4497         struct nfs4_openowner *oo = open->op_openowner;
4498
4499         lockdep_assert_held(&fp->fi_lock);
4500
4501         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4502                 /* ignore lock owners */
4503                 if (local->st_stateowner->so_is_open_owner == 0)
4504                         continue;
4505                 if (local->st_stateowner != &oo->oo_owner)
4506                         continue;
4507                 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
4508                         ret = local;
4509                         refcount_inc(&ret->st_stid.sc_count);
4510                         break;
4511                 }
4512         }
4513         return ret;
4514 }
4515
4516 static __be32
4517 nfsd4_verify_open_stid(struct nfs4_stid *s)
4518 {
4519         __be32 ret = nfs_ok;
4520
4521         switch (s->sc_type) {
4522         default:
4523                 break;
4524         case 0:
4525         case NFS4_CLOSED_STID:
4526         case NFS4_CLOSED_DELEG_STID:
4527                 ret = nfserr_bad_stateid;
4528                 break;
4529         case NFS4_REVOKED_DELEG_STID:
4530                 ret = nfserr_deleg_revoked;
4531         }
4532         return ret;
4533 }
4534
4535 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4536 static __be32
4537 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4538 {
4539         __be32 ret;
4540
4541         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4542         ret = nfsd4_verify_open_stid(&stp->st_stid);
4543         if (ret != nfs_ok)
4544                 mutex_unlock(&stp->st_mutex);
4545         return ret;
4546 }
4547
4548 static struct nfs4_ol_stateid *
4549 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4550 {
4551         struct nfs4_ol_stateid *stp;
4552         for (;;) {
4553                 spin_lock(&fp->fi_lock);
4554                 stp = nfsd4_find_existing_open(fp, open);
4555                 spin_unlock(&fp->fi_lock);
4556                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4557                         break;
4558                 nfs4_put_stid(&stp->st_stid);
4559         }
4560         return stp;
4561 }
4562
4563 static struct nfs4_openowner *
4564 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4565                            struct nfsd4_compound_state *cstate)
4566 {
4567         struct nfs4_client *clp = cstate->clp;
4568         struct nfs4_openowner *oo, *ret;
4569
4570         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4571         if (!oo)
4572                 return NULL;
4573         oo->oo_owner.so_ops = &openowner_ops;
4574         oo->oo_owner.so_is_open_owner = 1;
4575         oo->oo_owner.so_seqid = open->op_seqid;
4576         oo->oo_flags = 0;
4577         if (nfsd4_has_session(cstate))
4578                 oo->oo_flags |= NFS4_OO_CONFIRMED;
4579         oo->oo_time = 0;
4580         oo->oo_last_closed_stid = NULL;
4581         INIT_LIST_HEAD(&oo->oo_close_lru);
4582         spin_lock(&clp->cl_lock);
4583         ret = find_openstateowner_str_locked(strhashval, open, clp);
4584         if (ret == NULL) {
4585                 hash_openowner(oo, clp, strhashval);
4586                 ret = oo;
4587         } else
4588                 nfs4_free_stateowner(&oo->oo_owner);
4589
4590         spin_unlock(&clp->cl_lock);
4591         return ret;
4592 }
4593
4594 static struct nfs4_ol_stateid *
4595 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4596 {
4597
4598         struct nfs4_openowner *oo = open->op_openowner;
4599         struct nfs4_ol_stateid *retstp = NULL;
4600         struct nfs4_ol_stateid *stp;
4601
4602         stp = open->op_stp;
4603         /* We are moving these outside of the spinlocks to avoid the warnings */
4604         mutex_init(&stp->st_mutex);
4605         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4606
4607 retry:
4608         spin_lock(&oo->oo_owner.so_client->cl_lock);
4609         spin_lock(&fp->fi_lock);
4610
4611         retstp = nfsd4_find_existing_open(fp, open);
4612         if (retstp)
4613                 goto out_unlock;
4614
4615         open->op_stp = NULL;
4616         refcount_inc(&stp->st_stid.sc_count);
4617         stp->st_stid.sc_type = NFS4_OPEN_STID;
4618         INIT_LIST_HEAD(&stp->st_locks);
4619         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4620         get_nfs4_file(fp);
4621         stp->st_stid.sc_file = fp;
4622         stp->st_access_bmap = 0;
4623         stp->st_deny_bmap = 0;
4624         stp->st_openstp = NULL;
4625         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4626         list_add(&stp->st_perfile, &fp->fi_stateids);
4627
4628 out_unlock:
4629         spin_unlock(&fp->fi_lock);
4630         spin_unlock(&oo->oo_owner.so_client->cl_lock);
4631         if (retstp) {
4632                 /* Handle races with CLOSE */
4633                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4634                         nfs4_put_stid(&retstp->st_stid);
4635                         goto retry;
4636                 }
4637                 /* To keep mutex tracking happy */
4638                 mutex_unlock(&stp->st_mutex);
4639                 stp = retstp;
4640         }
4641         return stp;
4642 }
4643
4644 /*
4645  * In the 4.0 case we need to keep the owners around a little while to handle
4646  * CLOSE replay. We still do need to release any file access that is held by
4647  * them before returning however.
4648  */
4649 static void
4650 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4651 {
4652         struct nfs4_ol_stateid *last;
4653         struct nfs4_openowner *oo = openowner(s->st_stateowner);
4654         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4655                                                 nfsd_net_id);
4656
4657         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4658
4659         /*
4660          * We know that we hold one reference via nfsd4_close, and another
4661          * "persistent" reference for the client. If the refcount is higher
4662          * than 2, then there are still calls in progress that are using this
4663          * stateid. We can't put the sc_file reference until they are finished.
4664          * Wait for the refcount to drop to 2. Since it has been unhashed,
4665          * there should be no danger of the refcount going back up again at
4666          * this point.
4667          */
4668         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4669
4670         release_all_access(s);
4671         if (s->st_stid.sc_file) {
4672                 put_nfs4_file(s->st_stid.sc_file);
4673                 s->st_stid.sc_file = NULL;
4674         }
4675
4676         spin_lock(&nn->client_lock);
4677         last = oo->oo_last_closed_stid;
4678         oo->oo_last_closed_stid = s;
4679         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4680         oo->oo_time = ktime_get_boottime_seconds();
4681         spin_unlock(&nn->client_lock);
4682         if (last)
4683                 nfs4_put_stid(&last->st_stid);
4684 }
4685
4686 /* search file_hashtbl[] for file */
4687 static struct nfs4_file *
4688 find_file_locked(struct svc_fh *fh, unsigned int hashval)
4689 {
4690         struct nfs4_file *fp;
4691
4692         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash,
4693                                 lockdep_is_held(&state_lock)) {
4694                 if (fh_match(&fp->fi_fhandle, &fh->fh_handle)) {
4695                         if (refcount_inc_not_zero(&fp->fi_ref))
4696                                 return fp;
4697                 }
4698         }
4699         return NULL;
4700 }
4701
4702 static struct nfs4_file *insert_file(struct nfs4_file *new, struct svc_fh *fh,
4703                                      unsigned int hashval)
4704 {
4705         struct nfs4_file *fp;
4706         struct nfs4_file *ret = NULL;
4707         bool alias_found = false;
4708
4709         spin_lock(&state_lock);
4710         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash,
4711                                  lockdep_is_held(&state_lock)) {
4712                 if (fh_match(&fp->fi_fhandle, &fh->fh_handle)) {
4713                         if (refcount_inc_not_zero(&fp->fi_ref))
4714                                 ret = fp;
4715                 } else if (d_inode(fh->fh_dentry) == fp->fi_inode)
4716                         fp->fi_aliased = alias_found = true;
4717         }
4718         if (likely(ret == NULL)) {
4719                 nfsd4_init_file(fh, hashval, new);
4720                 new->fi_aliased = alias_found;
4721                 ret = new;
4722         }
4723         spin_unlock(&state_lock);
4724         return ret;
4725 }
4726
4727 static struct nfs4_file * find_file(struct svc_fh *fh)
4728 {
4729         struct nfs4_file *fp;
4730         unsigned int hashval = file_hashval(fh);
4731
4732         rcu_read_lock();
4733         fp = find_file_locked(fh, hashval);
4734         rcu_read_unlock();
4735         return fp;
4736 }
4737
4738 static struct nfs4_file *
4739 find_or_add_file(struct nfs4_file *new, struct svc_fh *fh)
4740 {
4741         struct nfs4_file *fp;
4742         unsigned int hashval = file_hashval(fh);
4743
4744         rcu_read_lock();
4745         fp = find_file_locked(fh, hashval);
4746         rcu_read_unlock();
4747         if (fp)
4748                 return fp;
4749
4750         return insert_file(new, fh, hashval);
4751 }
4752
4753 /*
4754  * Called to check deny when READ with all zero stateid or
4755  * WRITE with all zero or all one stateid
4756  */
4757 static __be32
4758 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
4759 {
4760         struct nfs4_file *fp;
4761         __be32 ret = nfs_ok;
4762
4763         fp = find_file(current_fh);
4764         if (!fp)
4765                 return ret;
4766         /* Check for conflicting share reservations */
4767         spin_lock(&fp->fi_lock);
4768         if (fp->fi_share_deny & deny_type)
4769                 ret = nfserr_locked;
4770         spin_unlock(&fp->fi_lock);
4771         put_nfs4_file(fp);
4772         return ret;
4773 }
4774
4775 static bool nfsd4_deleg_present(const struct inode *inode)
4776 {
4777         struct file_lock_context *ctx = smp_load_acquire(&inode->i_flctx);
4778
4779         return ctx && !list_empty_careful(&ctx->flc_lease);
4780 }
4781
4782 /**
4783  * nfsd_wait_for_delegreturn - wait for delegations to be returned
4784  * @rqstp: the RPC transaction being executed
4785  * @inode: in-core inode of the file being waited for
4786  *
4787  * The timeout prevents deadlock if all nfsd threads happen to be
4788  * tied up waiting for returning delegations.
4789  *
4790  * Return values:
4791  *   %true: delegation was returned
4792  *   %false: timed out waiting for delegreturn
4793  */
4794 bool nfsd_wait_for_delegreturn(struct svc_rqst *rqstp, struct inode *inode)
4795 {
4796         long __maybe_unused timeo;
4797
4798         timeo = wait_var_event_timeout(inode, !nfsd4_deleg_present(inode),
4799                                        NFSD_DELEGRETURN_TIMEOUT);
4800         trace_nfsd_delegret_wakeup(rqstp, inode, timeo);
4801         return timeo > 0;
4802 }
4803
4804 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
4805 {
4806         struct nfs4_delegation *dp = cb_to_delegation(cb);
4807         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
4808                                           nfsd_net_id);
4809
4810         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
4811
4812         /*
4813          * We can't do this in nfsd_break_deleg_cb because it is
4814          * already holding inode->i_lock.
4815          *
4816          * If the dl_time != 0, then we know that it has already been
4817          * queued for a lease break. Don't queue it again.
4818          */
4819         spin_lock(&state_lock);
4820         if (delegation_hashed(dp) && dp->dl_time == 0) {
4821                 dp->dl_time = ktime_get_boottime_seconds();
4822                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
4823         }
4824         spin_unlock(&state_lock);
4825 }
4826
4827 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
4828                 struct rpc_task *task)
4829 {
4830         struct nfs4_delegation *dp = cb_to_delegation(cb);
4831
4832         trace_nfsd_cb_recall_done(&dp->dl_stid.sc_stateid, task);
4833
4834         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID ||
4835             dp->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4836                 return 1;
4837
4838         switch (task->tk_status) {
4839         case 0:
4840                 return 1;
4841         case -NFS4ERR_DELAY:
4842                 rpc_delay(task, 2 * HZ);
4843                 return 0;
4844         case -EBADHANDLE:
4845         case -NFS4ERR_BAD_STATEID:
4846                 /*
4847                  * Race: client probably got cb_recall before open reply
4848                  * granting delegation.
4849                  */
4850                 if (dp->dl_retries--) {
4851                         rpc_delay(task, 2 * HZ);
4852                         return 0;
4853                 }
4854                 fallthrough;
4855         default:
4856                 return 1;
4857         }
4858 }
4859
4860 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
4861 {
4862         struct nfs4_delegation *dp = cb_to_delegation(cb);
4863
4864         nfs4_put_stid(&dp->dl_stid);
4865 }
4866
4867 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
4868         .prepare        = nfsd4_cb_recall_prepare,
4869         .done           = nfsd4_cb_recall_done,
4870         .release        = nfsd4_cb_recall_release,
4871 };
4872
4873 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
4874 {
4875         /*
4876          * We're assuming the state code never drops its reference
4877          * without first removing the lease.  Since we're in this lease
4878          * callback (and since the lease code is serialized by the
4879          * flc_lock) we know the server hasn't removed the lease yet, and
4880          * we know it's safe to take a reference.
4881          */
4882         refcount_inc(&dp->dl_stid.sc_count);
4883         WARN_ON_ONCE(!nfsd4_run_cb(&dp->dl_recall));
4884 }
4885
4886 /* Called from break_lease() with flc_lock held. */
4887 static bool
4888 nfsd_break_deleg_cb(struct file_lock *fl)
4889 {
4890         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4891         struct nfs4_file *fp = dp->dl_stid.sc_file;
4892         struct nfs4_client *clp = dp->dl_stid.sc_client;
4893         struct nfsd_net *nn;
4894
4895         trace_nfsd_cb_recall(&dp->dl_stid);
4896
4897         dp->dl_recalled = true;
4898         atomic_inc(&clp->cl_delegs_in_recall);
4899         if (try_to_expire_client(clp)) {
4900                 nn = net_generic(clp->net, nfsd_net_id);
4901                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
4902         }
4903
4904         /*
4905          * We don't want the locks code to timeout the lease for us;
4906          * we'll remove it ourself if a delegation isn't returned
4907          * in time:
4908          */
4909         fl->fl_break_time = 0;
4910
4911         spin_lock(&fp->fi_lock);
4912         fp->fi_had_conflict = true;
4913         nfsd_break_one_deleg(dp);
4914         spin_unlock(&fp->fi_lock);
4915         return false;
4916 }
4917
4918 /**
4919  * nfsd_breaker_owns_lease - Check if lease conflict was resolved
4920  * @fl: Lock state to check
4921  *
4922  * Return values:
4923  *   %true: Lease conflict was resolved
4924  *   %false: Lease conflict was not resolved.
4925  */
4926 static bool nfsd_breaker_owns_lease(struct file_lock *fl)
4927 {
4928         struct nfs4_delegation *dl = fl->fl_owner;
4929         struct svc_rqst *rqst;
4930         struct nfs4_client *clp;
4931
4932         if (!i_am_nfsd())
4933                 return false;
4934         rqst = kthread_data(current);
4935         /* Note rq_prog == NFS_ACL_PROGRAM is also possible: */
4936         if (rqst->rq_prog != NFS_PROGRAM || rqst->rq_vers < 4)
4937                 return false;
4938         clp = *(rqst->rq_lease_breaker);
4939         return dl->dl_stid.sc_client == clp;
4940 }
4941
4942 static int
4943 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4944                      struct list_head *dispose)
4945 {
4946         struct nfs4_delegation *dp = (struct nfs4_delegation *)onlist->fl_owner;
4947         struct nfs4_client *clp = dp->dl_stid.sc_client;
4948
4949         if (arg & F_UNLCK) {
4950                 if (dp->dl_recalled)
4951                         atomic_dec(&clp->cl_delegs_in_recall);
4952                 return lease_modify(onlist, arg, dispose);
4953         } else
4954                 return -EAGAIN;
4955 }
4956
4957 static const struct lock_manager_operations nfsd_lease_mng_ops = {
4958         .lm_breaker_owns_lease = nfsd_breaker_owns_lease,
4959         .lm_break = nfsd_break_deleg_cb,
4960         .lm_change = nfsd_change_deleg_cb,
4961 };
4962
4963 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4964 {
4965         if (nfsd4_has_session(cstate))
4966                 return nfs_ok;
4967         if (seqid == so->so_seqid - 1)
4968                 return nfserr_replay_me;
4969         if (seqid == so->so_seqid)
4970                 return nfs_ok;
4971         return nfserr_bad_seqid;
4972 }
4973
4974 static struct nfs4_client *lookup_clientid(clientid_t *clid, bool sessions,
4975                                                 struct nfsd_net *nn)
4976 {
4977         struct nfs4_client *found;
4978
4979         spin_lock(&nn->client_lock);
4980         found = find_confirmed_client(clid, sessions, nn);
4981         if (found)
4982                 atomic_inc(&found->cl_rpc_users);
4983         spin_unlock(&nn->client_lock);
4984         return found;
4985 }
4986
4987 static __be32 set_client(clientid_t *clid,
4988                 struct nfsd4_compound_state *cstate,
4989                 struct nfsd_net *nn)
4990 {
4991         if (cstate->clp) {
4992                 if (!same_clid(&cstate->clp->cl_clientid, clid))
4993                         return nfserr_stale_clientid;
4994                 return nfs_ok;
4995         }
4996         if (STALE_CLIENTID(clid, nn))
4997                 return nfserr_stale_clientid;
4998         /*
4999          * We're in the 4.0 case (otherwise the SEQUENCE op would have
5000          * set cstate->clp), so session = false:
5001          */
5002         cstate->clp = lookup_clientid(clid, false, nn);
5003         if (!cstate->clp)
5004                 return nfserr_expired;
5005         return nfs_ok;
5006 }
5007
5008 __be32
5009 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
5010                     struct nfsd4_open *open, struct nfsd_net *nn)
5011 {
5012         clientid_t *clientid = &open->op_clientid;
5013         struct nfs4_client *clp = NULL;
5014         unsigned int strhashval;
5015         struct nfs4_openowner *oo = NULL;
5016         __be32 status;
5017
5018         /*
5019          * In case we need it later, after we've already created the
5020          * file and don't want to risk a further failure:
5021          */
5022         open->op_file = nfsd4_alloc_file();
5023         if (open->op_file == NULL)
5024                 return nfserr_jukebox;
5025
5026         status = set_client(clientid, cstate, nn);
5027         if (status)
5028                 return status;
5029         clp = cstate->clp;
5030
5031         strhashval = ownerstr_hashval(&open->op_owner);
5032         oo = find_openstateowner_str(strhashval, open, clp);
5033         open->op_openowner = oo;
5034         if (!oo) {
5035                 goto new_owner;
5036         }
5037         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5038                 /* Replace unconfirmed owners without checking for replay. */
5039                 release_openowner(oo);
5040                 open->op_openowner = NULL;
5041                 goto new_owner;
5042         }
5043         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
5044         if (status)
5045                 return status;
5046         goto alloc_stateid;
5047 new_owner:
5048         oo = alloc_init_open_stateowner(strhashval, open, cstate);
5049         if (oo == NULL)
5050                 return nfserr_jukebox;
5051         open->op_openowner = oo;
5052 alloc_stateid:
5053         open->op_stp = nfs4_alloc_open_stateid(clp);
5054         if (!open->op_stp)
5055                 return nfserr_jukebox;
5056
5057         if (nfsd4_has_session(cstate) &&
5058             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
5059                 open->op_odstate = alloc_clnt_odstate(clp);
5060                 if (!open->op_odstate)
5061                         return nfserr_jukebox;
5062         }
5063
5064         return nfs_ok;
5065 }
5066
5067 static inline __be32
5068 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
5069 {
5070         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
5071                 return nfserr_openmode;
5072         else
5073                 return nfs_ok;
5074 }
5075
5076 static int share_access_to_flags(u32 share_access)
5077 {
5078         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
5079 }
5080
5081 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
5082 {
5083         struct nfs4_stid *ret;
5084
5085         ret = find_stateid_by_type(cl, s,
5086                                 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
5087         if (!ret)
5088                 return NULL;
5089         return delegstateid(ret);
5090 }
5091
5092 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
5093 {
5094         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
5095                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
5096 }
5097
5098 static __be32
5099 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
5100                 struct nfs4_delegation **dp)
5101 {
5102         int flags;
5103         __be32 status = nfserr_bad_stateid;
5104         struct nfs4_delegation *deleg;
5105
5106         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
5107         if (deleg == NULL)
5108                 goto out;
5109         if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
5110                 nfs4_put_stid(&deleg->dl_stid);
5111                 if (cl->cl_minorversion)
5112                         status = nfserr_deleg_revoked;
5113                 goto out;
5114         }
5115         flags = share_access_to_flags(open->op_share_access);
5116         status = nfs4_check_delegmode(deleg, flags);
5117         if (status) {
5118                 nfs4_put_stid(&deleg->dl_stid);
5119                 goto out;
5120         }
5121         *dp = deleg;
5122 out:
5123         if (!nfsd4_is_deleg_cur(open))
5124                 return nfs_ok;
5125         if (status)
5126                 return status;
5127         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5128         return nfs_ok;
5129 }
5130
5131 static inline int nfs4_access_to_access(u32 nfs4_access)
5132 {
5133         int flags = 0;
5134
5135         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
5136                 flags |= NFSD_MAY_READ;
5137         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
5138                 flags |= NFSD_MAY_WRITE;
5139         return flags;
5140 }
5141
5142 static inline __be32
5143 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
5144                 struct nfsd4_open *open)
5145 {
5146         struct iattr iattr = {
5147                 .ia_valid = ATTR_SIZE,
5148                 .ia_size = 0,
5149         };
5150         struct nfsd_attrs attrs = {
5151                 .na_iattr       = &iattr,
5152         };
5153         if (!open->op_truncate)
5154                 return 0;
5155         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
5156                 return nfserr_inval;
5157         return nfsd_setattr(rqstp, fh, &attrs, 0, (time64_t)0);
5158 }
5159
5160 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
5161                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5162                 struct nfsd4_open *open, bool new_stp)
5163 {
5164         struct nfsd_file *nf = NULL;
5165         __be32 status;
5166         int oflag = nfs4_access_to_omode(open->op_share_access);
5167         int access = nfs4_access_to_access(open->op_share_access);
5168         unsigned char old_access_bmap, old_deny_bmap;
5169
5170         spin_lock(&fp->fi_lock);
5171
5172         /*
5173          * Are we trying to set a deny mode that would conflict with
5174          * current access?
5175          */
5176         status = nfs4_file_check_deny(fp, open->op_share_deny);
5177         if (status != nfs_ok) {
5178                 if (status != nfserr_share_denied) {
5179                         spin_unlock(&fp->fi_lock);
5180                         goto out;
5181                 }
5182                 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5183                                 stp, open->op_share_deny, false))
5184                         status = nfserr_jukebox;
5185                 spin_unlock(&fp->fi_lock);
5186                 goto out;
5187         }
5188
5189         /* set access to the file */
5190         status = nfs4_file_get_access(fp, open->op_share_access);
5191         if (status != nfs_ok) {
5192                 if (status != nfserr_share_denied) {
5193                         spin_unlock(&fp->fi_lock);
5194                         goto out;
5195                 }
5196                 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5197                                 stp, open->op_share_access, true))
5198                         status = nfserr_jukebox;
5199                 spin_unlock(&fp->fi_lock);
5200                 goto out;
5201         }
5202
5203         /* Set access bits in stateid */
5204         old_access_bmap = stp->st_access_bmap;
5205         set_access(open->op_share_access, stp);
5206
5207         /* Set new deny mask */
5208         old_deny_bmap = stp->st_deny_bmap;
5209         set_deny(open->op_share_deny, stp);
5210         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5211
5212         if (!fp->fi_fds[oflag]) {
5213                 spin_unlock(&fp->fi_lock);
5214
5215                 status = nfsd_file_acquire_opened(rqstp, cur_fh, access,
5216                                                   open->op_filp, &nf);
5217                 if (status != nfs_ok)
5218                         goto out_put_access;
5219
5220                 spin_lock(&fp->fi_lock);
5221                 if (!fp->fi_fds[oflag]) {
5222                         fp->fi_fds[oflag] = nf;
5223                         nf = NULL;
5224                 }
5225         }
5226         spin_unlock(&fp->fi_lock);
5227         if (nf)
5228                 nfsd_file_put(nf);
5229
5230         status = nfserrno(nfsd_open_break_lease(cur_fh->fh_dentry->d_inode,
5231                                                                 access));
5232         if (status)
5233                 goto out_put_access;
5234
5235         status = nfsd4_truncate(rqstp, cur_fh, open);
5236         if (status)
5237                 goto out_put_access;
5238 out:
5239         return status;
5240 out_put_access:
5241         stp->st_access_bmap = old_access_bmap;
5242         nfs4_file_put_access(fp, open->op_share_access);
5243         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
5244         goto out;
5245 }
5246
5247 static __be32
5248 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp,
5249                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5250                 struct nfsd4_open *open)
5251 {
5252         __be32 status;
5253         unsigned char old_deny_bmap = stp->st_deny_bmap;
5254
5255         if (!test_access(open->op_share_access, stp))
5256                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open, false);
5257
5258         /* test and set deny mode */
5259         spin_lock(&fp->fi_lock);
5260         status = nfs4_file_check_deny(fp, open->op_share_deny);
5261         if (status == nfs_ok) {
5262                 if (status != nfserr_share_denied) {
5263                         set_deny(open->op_share_deny, stp);
5264                         fp->fi_share_deny |=
5265                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5266                 } else {
5267                         if (nfs4_resolve_deny_conflicts_locked(fp, false,
5268                                         stp, open->op_share_deny, false))
5269                                 status = nfserr_jukebox;
5270                 }
5271         }
5272         spin_unlock(&fp->fi_lock);
5273
5274         if (status != nfs_ok)
5275                 return status;
5276
5277         status = nfsd4_truncate(rqstp, cur_fh, open);
5278         if (status != nfs_ok)
5279                 reset_union_bmap_deny(old_deny_bmap, stp);
5280         return status;
5281 }
5282
5283 /* Should we give out recallable state?: */
5284 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
5285 {
5286         if (clp->cl_cb_state == NFSD4_CB_UP)
5287                 return true;
5288         /*
5289          * In the sessions case, since we don't have to establish a
5290          * separate connection for callbacks, we assume it's OK
5291          * until we hear otherwise:
5292          */
5293         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
5294 }
5295
5296 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
5297                                                 int flag)
5298 {
5299         struct file_lock *fl;
5300
5301         fl = locks_alloc_lock();
5302         if (!fl)
5303                 return NULL;
5304         fl->fl_lmops = &nfsd_lease_mng_ops;
5305         fl->fl_flags = FL_DELEG;
5306         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
5307         fl->fl_end = OFFSET_MAX;
5308         fl->fl_owner = (fl_owner_t)dp;
5309         fl->fl_pid = current->tgid;
5310         fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
5311         return fl;
5312 }
5313
5314 static int nfsd4_check_conflicting_opens(struct nfs4_client *clp,
5315                                          struct nfs4_file *fp)
5316 {
5317         struct nfs4_ol_stateid *st;
5318         struct file *f = fp->fi_deleg_file->nf_file;
5319         struct inode *ino = locks_inode(f);
5320         int writes;
5321
5322         writes = atomic_read(&ino->i_writecount);
5323         if (!writes)
5324                 return 0;
5325         /*
5326          * There could be multiple filehandles (hence multiple
5327          * nfs4_files) referencing this file, but that's not too
5328          * common; let's just give up in that case rather than
5329          * trying to go look up all the clients using that other
5330          * nfs4_file as well:
5331          */
5332         if (fp->fi_aliased)
5333                 return -EAGAIN;
5334         /*
5335          * If there's a close in progress, make sure that we see it
5336          * clear any fi_fds[] entries before we see it decrement
5337          * i_writecount:
5338          */
5339         smp_mb__after_atomic();
5340
5341         if (fp->fi_fds[O_WRONLY])
5342                 writes--;
5343         if (fp->fi_fds[O_RDWR])
5344                 writes--;
5345         if (writes > 0)
5346                 return -EAGAIN; /* There may be non-NFSv4 writers */
5347         /*
5348          * It's possible there are non-NFSv4 write opens in progress,
5349          * but if they haven't incremented i_writecount yet then they
5350          * also haven't called break lease yet; so, they'll break this
5351          * lease soon enough.  So, all that's left to check for is NFSv4
5352          * opens:
5353          */
5354         spin_lock(&fp->fi_lock);
5355         list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
5356                 if (st->st_openstp == NULL /* it's an open */ &&
5357                     access_permit_write(st) &&
5358                     st->st_stid.sc_client != clp) {
5359                         spin_unlock(&fp->fi_lock);
5360                         return -EAGAIN;
5361                 }
5362         }
5363         spin_unlock(&fp->fi_lock);
5364         /*
5365          * There's a small chance that we could be racing with another
5366          * NFSv4 open.  However, any open that hasn't added itself to
5367          * the fi_stateids list also hasn't called break_lease yet; so,
5368          * they'll break this lease soon enough.
5369          */
5370         return 0;
5371 }
5372
5373 /*
5374  * It's possible that between opening the dentry and setting the delegation,
5375  * that it has been renamed or unlinked. Redo the lookup to verify that this
5376  * hasn't happened.
5377  */
5378 static int
5379 nfsd4_verify_deleg_dentry(struct nfsd4_open *open, struct nfs4_file *fp,
5380                           struct svc_fh *parent)
5381 {
5382         struct svc_export *exp;
5383         struct dentry *child;
5384         __be32 err;
5385
5386         err = nfsd_lookup_dentry(open->op_rqstp, parent,
5387                                  open->op_fname, open->op_fnamelen,
5388                                  &exp, &child);
5389
5390         if (err)
5391                 return -EAGAIN;
5392
5393         exp_put(exp);
5394         dput(child);
5395         if (child != file_dentry(fp->fi_deleg_file->nf_file))
5396                 return -EAGAIN;
5397
5398         return 0;
5399 }
5400
5401 static struct nfs4_delegation *
5402 nfs4_set_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5403                     struct svc_fh *parent)
5404 {
5405         int status = 0;
5406         struct nfs4_client *clp = stp->st_stid.sc_client;
5407         struct nfs4_file *fp = stp->st_stid.sc_file;
5408         struct nfs4_clnt_odstate *odstate = stp->st_clnt_odstate;
5409         struct nfs4_delegation *dp;
5410         struct nfsd_file *nf;
5411         struct file_lock *fl;
5412
5413         /*
5414          * The fi_had_conflict and nfs_get_existing_delegation checks
5415          * here are just optimizations; we'll need to recheck them at
5416          * the end:
5417          */
5418         if (fp->fi_had_conflict)
5419                 return ERR_PTR(-EAGAIN);
5420
5421         nf = find_readable_file(fp);
5422         if (!nf) {
5423                 /*
5424                  * We probably could attempt another open and get a read
5425                  * delegation, but for now, don't bother until the
5426                  * client actually sends us one.
5427                  */
5428                 return ERR_PTR(-EAGAIN);
5429         }
5430         spin_lock(&state_lock);
5431         spin_lock(&fp->fi_lock);
5432         if (nfs4_delegation_exists(clp, fp))
5433                 status = -EAGAIN;
5434         else if (!fp->fi_deleg_file) {
5435                 fp->fi_deleg_file = nf;
5436                 /* increment early to prevent fi_deleg_file from being
5437                  * cleared */
5438                 fp->fi_delegees = 1;
5439                 nf = NULL;
5440         } else
5441                 fp->fi_delegees++;
5442         spin_unlock(&fp->fi_lock);
5443         spin_unlock(&state_lock);
5444         if (nf)
5445                 nfsd_file_put(nf);
5446         if (status)
5447                 return ERR_PTR(status);
5448
5449         status = -ENOMEM;
5450         dp = alloc_init_deleg(clp, fp, odstate);
5451         if (!dp)
5452                 goto out_delegees;
5453
5454         fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
5455         if (!fl)
5456                 goto out_clnt_odstate;
5457
5458         status = vfs_setlease(fp->fi_deleg_file->nf_file, fl->fl_type, &fl, NULL);
5459         if (fl)
5460                 locks_free_lock(fl);
5461         if (status)
5462                 goto out_clnt_odstate;
5463
5464         if (parent) {
5465                 status = nfsd4_verify_deleg_dentry(open, fp, parent);
5466                 if (status)
5467                         goto out_unlock;
5468         }
5469
5470         status = nfsd4_check_conflicting_opens(clp, fp);
5471         if (status)
5472                 goto out_unlock;
5473
5474         spin_lock(&state_lock);
5475         spin_lock(&fp->fi_lock);
5476         if (fp->fi_had_conflict)
5477                 status = -EAGAIN;
5478         else
5479                 status = hash_delegation_locked(dp, fp);
5480         spin_unlock(&fp->fi_lock);
5481         spin_unlock(&state_lock);
5482
5483         if (status)
5484                 goto out_unlock;
5485
5486         return dp;
5487 out_unlock:
5488         vfs_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
5489 out_clnt_odstate:
5490         put_clnt_odstate(dp->dl_clnt_odstate);
5491         nfs4_put_stid(&dp->dl_stid);
5492 out_delegees:
5493         put_deleg_file(fp);
5494         return ERR_PTR(status);
5495 }
5496
5497 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
5498 {
5499         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5500         if (status == -EAGAIN)
5501                 open->op_why_no_deleg = WND4_CONTENTION;
5502         else {
5503                 open->op_why_no_deleg = WND4_RESOURCE;
5504                 switch (open->op_deleg_want) {
5505                 case NFS4_SHARE_WANT_READ_DELEG:
5506                 case NFS4_SHARE_WANT_WRITE_DELEG:
5507                 case NFS4_SHARE_WANT_ANY_DELEG:
5508                         break;
5509                 case NFS4_SHARE_WANT_CANCEL:
5510                         open->op_why_no_deleg = WND4_CANCELLED;
5511                         break;
5512                 case NFS4_SHARE_WANT_NO_DELEG:
5513                         WARN_ON_ONCE(1);
5514                 }
5515         }
5516 }
5517
5518 /*
5519  * Attempt to hand out a delegation.
5520  *
5521  * Note we don't support write delegations, and won't until the vfs has
5522  * proper support for them.
5523  */
5524 static void
5525 nfs4_open_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5526                      struct svc_fh *currentfh)
5527 {
5528         struct nfs4_delegation *dp;
5529         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
5530         struct nfs4_client *clp = stp->st_stid.sc_client;
5531         struct svc_fh *parent = NULL;
5532         int cb_up;
5533         int status = 0;
5534
5535         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
5536         open->op_recall = 0;
5537         switch (open->op_claim_type) {
5538                 case NFS4_OPEN_CLAIM_PREVIOUS:
5539                         if (!cb_up)
5540                                 open->op_recall = 1;
5541                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
5542                                 goto out_no_deleg;
5543                         break;
5544                 case NFS4_OPEN_CLAIM_NULL:
5545                         parent = currentfh;
5546                         fallthrough;
5547                 case NFS4_OPEN_CLAIM_FH:
5548                         /*
5549                          * Let's not give out any delegations till everyone's
5550                          * had the chance to reclaim theirs, *and* until
5551                          * NLM locks have all been reclaimed:
5552                          */
5553                         if (locks_in_grace(clp->net))
5554                                 goto out_no_deleg;
5555                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
5556                                 goto out_no_deleg;
5557                         break;
5558                 default:
5559                         goto out_no_deleg;
5560         }
5561         dp = nfs4_set_delegation(open, stp, parent);
5562         if (IS_ERR(dp))
5563                 goto out_no_deleg;
5564
5565         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5566
5567         trace_nfsd_deleg_read(&dp->dl_stid.sc_stateid);
5568         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
5569         nfs4_put_stid(&dp->dl_stid);
5570         return;
5571 out_no_deleg:
5572         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
5573         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
5574             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
5575                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
5576                 open->op_recall = 1;
5577         }
5578
5579         /* 4.1 client asking for a delegation? */
5580         if (open->op_deleg_want)
5581                 nfsd4_open_deleg_none_ext(open, status);
5582         return;
5583 }
5584
5585 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
5586                                         struct nfs4_delegation *dp)
5587 {
5588         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
5589             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5590                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5591                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
5592         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
5593                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5594                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5595                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
5596         }
5597         /* Otherwise the client must be confused wanting a delegation
5598          * it already has, therefore we don't return
5599          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
5600          */
5601 }
5602
5603 /**
5604  * nfsd4_process_open2 - finish open processing
5605  * @rqstp: the RPC transaction being executed
5606  * @current_fh: NFSv4 COMPOUND's current filehandle
5607  * @open: OPEN arguments
5608  *
5609  * If successful, (1) truncate the file if open->op_truncate was
5610  * set, (2) set open->op_stateid, (3) set open->op_delegation.
5611  *
5612  * Returns %nfs_ok on success; otherwise an nfs4stat value in
5613  * network byte order is returned.
5614  */
5615 __be32
5616 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
5617 {
5618         struct nfsd4_compoundres *resp = rqstp->rq_resp;
5619         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
5620         struct nfs4_file *fp = NULL;
5621         struct nfs4_ol_stateid *stp = NULL;
5622         struct nfs4_delegation *dp = NULL;
5623         __be32 status;
5624         bool new_stp = false;
5625
5626         /*
5627          * Lookup file; if found, lookup stateid and check open request,
5628          * and check for delegations in the process of being recalled.
5629          * If not found, create the nfs4_file struct
5630          */
5631         fp = find_or_add_file(open->op_file, current_fh);
5632         if (fp != open->op_file) {
5633                 status = nfs4_check_deleg(cl, open, &dp);
5634                 if (status)
5635                         goto out;
5636                 stp = nfsd4_find_and_lock_existing_open(fp, open);
5637         } else {
5638                 open->op_file = NULL;
5639                 status = nfserr_bad_stateid;
5640                 if (nfsd4_is_deleg_cur(open))
5641                         goto out;
5642         }
5643
5644         if (!stp) {
5645                 stp = init_open_stateid(fp, open);
5646                 if (!open->op_stp)
5647                         new_stp = true;
5648         }
5649
5650         /*
5651          * OPEN the file, or upgrade an existing OPEN.
5652          * If truncate fails, the OPEN fails.
5653          *
5654          * stp is already locked.
5655          */
5656         if (!new_stp) {
5657                 /* Stateid was found, this is an OPEN upgrade */
5658                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
5659                 if (status) {
5660                         mutex_unlock(&stp->st_mutex);
5661                         goto out;
5662                 }
5663         } else {
5664                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open, true);
5665                 if (status) {
5666                         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5667                         release_open_stateid(stp);
5668                         mutex_unlock(&stp->st_mutex);
5669                         goto out;
5670                 }
5671
5672                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
5673                                                         open->op_odstate);
5674                 if (stp->st_clnt_odstate == open->op_odstate)
5675                         open->op_odstate = NULL;
5676         }
5677
5678         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
5679         mutex_unlock(&stp->st_mutex);
5680
5681         if (nfsd4_has_session(&resp->cstate)) {
5682                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
5683                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5684                         open->op_why_no_deleg = WND4_NOT_WANTED;
5685                         goto nodeleg;
5686                 }
5687         }
5688
5689         /*
5690         * Attempt to hand out a delegation. No error return, because the
5691         * OPEN succeeds even if we fail.
5692         */
5693         nfs4_open_delegation(open, stp, &resp->cstate.current_fh);
5694 nodeleg:
5695         status = nfs_ok;
5696         trace_nfsd_open(&stp->st_stid.sc_stateid);
5697 out:
5698         /* 4.1 client trying to upgrade/downgrade delegation? */
5699         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
5700             open->op_deleg_want)
5701                 nfsd4_deleg_xgrade_none_ext(open, dp);
5702
5703         if (fp)
5704                 put_nfs4_file(fp);
5705         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
5706                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5707         /*
5708         * To finish the open response, we just need to set the rflags.
5709         */
5710         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
5711         if (nfsd4_has_session(&resp->cstate))
5712                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
5713         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
5714                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
5715
5716         if (dp)
5717                 nfs4_put_stid(&dp->dl_stid);
5718         if (stp)
5719                 nfs4_put_stid(&stp->st_stid);
5720
5721         return status;
5722 }
5723
5724 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
5725                               struct nfsd4_open *open)
5726 {
5727         if (open->op_openowner) {
5728                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
5729
5730                 nfsd4_cstate_assign_replay(cstate, so);
5731                 nfs4_put_stateowner(so);
5732         }
5733         if (open->op_file)
5734                 kmem_cache_free(file_slab, open->op_file);
5735         if (open->op_stp)
5736                 nfs4_put_stid(&open->op_stp->st_stid);
5737         if (open->op_odstate)
5738                 kmem_cache_free(odstate_slab, open->op_odstate);
5739 }
5740
5741 __be32
5742 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5743             union nfsd4_op_u *u)
5744 {
5745         clientid_t *clid = &u->renew;
5746         struct nfs4_client *clp;
5747         __be32 status;
5748         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5749
5750         trace_nfsd_clid_renew(clid);
5751         status = set_client(clid, cstate, nn);
5752         if (status)
5753                 return status;
5754         clp = cstate->clp;
5755         if (!list_empty(&clp->cl_delegations)
5756                         && clp->cl_cb_state != NFSD4_CB_UP)
5757                 return nfserr_cb_path_down;
5758         return nfs_ok;
5759 }
5760
5761 void
5762 nfsd4_end_grace(struct nfsd_net *nn)
5763 {
5764         /* do nothing if grace period already ended */
5765         if (nn->grace_ended)
5766                 return;
5767
5768         trace_nfsd_grace_complete(nn);
5769         nn->grace_ended = true;
5770         /*
5771          * If the server goes down again right now, an NFSv4
5772          * client will still be allowed to reclaim after it comes back up,
5773          * even if it hasn't yet had a chance to reclaim state this time.
5774          *
5775          */
5776         nfsd4_record_grace_done(nn);
5777         /*
5778          * At this point, NFSv4 clients can still reclaim.  But if the
5779          * server crashes, any that have not yet reclaimed will be out
5780          * of luck on the next boot.
5781          *
5782          * (NFSv4.1+ clients are considered to have reclaimed once they
5783          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
5784          * have reclaimed after their first OPEN.)
5785          */
5786         locks_end_grace(&nn->nfsd4_manager);
5787         /*
5788          * At this point, and once lockd and/or any other containers
5789          * exit their grace period, further reclaims will fail and
5790          * regular locking can resume.
5791          */
5792 }
5793
5794 /*
5795  * If we've waited a lease period but there are still clients trying to
5796  * reclaim, wait a little longer to give them a chance to finish.
5797  */
5798 static bool clients_still_reclaiming(struct nfsd_net *nn)
5799 {
5800         time64_t double_grace_period_end = nn->boot_time +
5801                                            2 * nn->nfsd4_lease;
5802
5803         if (nn->track_reclaim_completes &&
5804                         atomic_read(&nn->nr_reclaim_complete) ==
5805                         nn->reclaim_str_hashtbl_size)
5806                 return false;
5807         if (!nn->somebody_reclaimed)
5808                 return false;
5809         nn->somebody_reclaimed = false;
5810         /*
5811          * If we've given them *two* lease times to reclaim, and they're
5812          * still not done, give up:
5813          */
5814         if (ktime_get_boottime_seconds() > double_grace_period_end)
5815                 return false;
5816         return true;
5817 }
5818
5819 struct laundry_time {
5820         time64_t cutoff;
5821         time64_t new_timeo;
5822 };
5823
5824 static bool state_expired(struct laundry_time *lt, time64_t last_refresh)
5825 {
5826         time64_t time_remaining;
5827
5828         if (last_refresh < lt->cutoff)
5829                 return true;
5830         time_remaining = last_refresh - lt->cutoff;
5831         lt->new_timeo = min(lt->new_timeo, time_remaining);
5832         return false;
5833 }
5834
5835 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
5836 void nfsd4_ssc_init_umount_work(struct nfsd_net *nn)
5837 {
5838         spin_lock_init(&nn->nfsd_ssc_lock);
5839         INIT_LIST_HEAD(&nn->nfsd_ssc_mount_list);
5840         init_waitqueue_head(&nn->nfsd_ssc_waitq);
5841 }
5842 EXPORT_SYMBOL_GPL(nfsd4_ssc_init_umount_work);
5843
5844 /*
5845  * This is called when nfsd is being shutdown, after all inter_ssc
5846  * cleanup were done, to destroy the ssc delayed unmount list.
5847  */
5848 static void nfsd4_ssc_shutdown_umount(struct nfsd_net *nn)
5849 {
5850         struct nfsd4_ssc_umount_item *ni = NULL;
5851         struct nfsd4_ssc_umount_item *tmp;
5852
5853         spin_lock(&nn->nfsd_ssc_lock);
5854         list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
5855                 list_del(&ni->nsui_list);
5856                 spin_unlock(&nn->nfsd_ssc_lock);
5857                 mntput(ni->nsui_vfsmount);
5858                 kfree(ni);
5859                 spin_lock(&nn->nfsd_ssc_lock);
5860         }
5861         spin_unlock(&nn->nfsd_ssc_lock);
5862 }
5863
5864 static void nfsd4_ssc_expire_umount(struct nfsd_net *nn)
5865 {
5866         bool do_wakeup = false;
5867         struct nfsd4_ssc_umount_item *ni = NULL;
5868         struct nfsd4_ssc_umount_item *tmp;
5869
5870         spin_lock(&nn->nfsd_ssc_lock);
5871         list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
5872                 if (time_after(jiffies, ni->nsui_expire)) {
5873                         if (refcount_read(&ni->nsui_refcnt) > 1)
5874                                 continue;
5875
5876                         /* mark being unmount */
5877                         ni->nsui_busy = true;
5878                         spin_unlock(&nn->nfsd_ssc_lock);
5879                         mntput(ni->nsui_vfsmount);
5880                         spin_lock(&nn->nfsd_ssc_lock);
5881
5882                         /* waiters need to start from begin of list */
5883                         list_del(&ni->nsui_list);
5884                         kfree(ni);
5885
5886                         /* wakeup ssc_connect waiters */
5887                         do_wakeup = true;
5888                         continue;
5889                 }
5890                 break;
5891         }
5892         if (do_wakeup)
5893                 wake_up_all(&nn->nfsd_ssc_waitq);
5894         spin_unlock(&nn->nfsd_ssc_lock);
5895 }
5896 #endif
5897
5898 /* Check if any lock belonging to this lockowner has any blockers */
5899 static bool
5900 nfs4_lockowner_has_blockers(struct nfs4_lockowner *lo)
5901 {
5902         struct file_lock_context *ctx;
5903         struct nfs4_ol_stateid *stp;
5904         struct nfs4_file *nf;
5905
5906         list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
5907                 nf = stp->st_stid.sc_file;
5908                 ctx = nf->fi_inode->i_flctx;
5909                 if (!ctx)
5910                         continue;
5911                 if (locks_owner_has_blockers(ctx, lo))
5912                         return true;
5913         }
5914         return false;
5915 }
5916
5917 static bool
5918 nfs4_anylock_blockers(struct nfs4_client *clp)
5919 {
5920         int i;
5921         struct nfs4_stateowner *so;
5922         struct nfs4_lockowner *lo;
5923
5924         if (atomic_read(&clp->cl_delegs_in_recall))
5925                 return true;
5926         spin_lock(&clp->cl_lock);
5927         for (i = 0; i < OWNER_HASH_SIZE; i++) {
5928                 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[i],
5929                                 so_strhash) {
5930                         if (so->so_is_open_owner)
5931                                 continue;
5932                         lo = lockowner(so);
5933                         if (nfs4_lockowner_has_blockers(lo)) {
5934                                 spin_unlock(&clp->cl_lock);
5935                                 return true;
5936                         }
5937                 }
5938         }
5939         spin_unlock(&clp->cl_lock);
5940         return false;
5941 }
5942
5943 static void
5944 nfs4_get_client_reaplist(struct nfsd_net *nn, struct list_head *reaplist,
5945                                 struct laundry_time *lt)
5946 {
5947         unsigned int maxreap, reapcnt = 0;
5948         struct list_head *pos, *next;
5949         struct nfs4_client *clp;
5950
5951         maxreap = (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) ?
5952                         NFSD_CLIENT_MAX_TRIM_PER_RUN : 0;
5953         INIT_LIST_HEAD(reaplist);
5954         spin_lock(&nn->client_lock);
5955         list_for_each_safe(pos, next, &nn->client_lru) {
5956                 clp = list_entry(pos, struct nfs4_client, cl_lru);
5957                 if (clp->cl_state == NFSD4_EXPIRABLE)
5958                         goto exp_client;
5959                 if (!state_expired(lt, clp->cl_time))
5960                         break;
5961                 if (!atomic_read(&clp->cl_rpc_users)) {
5962                         if (clp->cl_state == NFSD4_ACTIVE)
5963                                 atomic_inc(&nn->nfsd_courtesy_clients);
5964                         clp->cl_state = NFSD4_COURTESY;
5965                 }
5966                 if (!client_has_state(clp))
5967                         goto exp_client;
5968                 if (!nfs4_anylock_blockers(clp))
5969                         if (reapcnt >= maxreap)
5970                                 continue;
5971 exp_client:
5972                 if (!mark_client_expired_locked(clp)) {
5973                         list_add(&clp->cl_lru, reaplist);
5974                         reapcnt++;
5975                 }
5976         }
5977         spin_unlock(&nn->client_lock);
5978 }
5979
5980 static void
5981 nfs4_get_courtesy_client_reaplist(struct nfsd_net *nn,
5982                                 struct list_head *reaplist)
5983 {
5984         unsigned int maxreap = 0, reapcnt = 0;
5985         struct list_head *pos, *next;
5986         struct nfs4_client *clp;
5987
5988         maxreap = NFSD_CLIENT_MAX_TRIM_PER_RUN;
5989         INIT_LIST_HEAD(reaplist);
5990
5991         spin_lock(&nn->client_lock);
5992         list_for_each_safe(pos, next, &nn->client_lru) {
5993                 clp = list_entry(pos, struct nfs4_client, cl_lru);
5994                 if (clp->cl_state == NFSD4_ACTIVE)
5995                         break;
5996                 if (reapcnt >= maxreap)
5997                         break;
5998                 if (!mark_client_expired_locked(clp)) {
5999                         list_add(&clp->cl_lru, reaplist);
6000                         reapcnt++;
6001                 }
6002         }
6003         spin_unlock(&nn->client_lock);
6004 }
6005
6006 static void
6007 nfs4_process_client_reaplist(struct list_head *reaplist)
6008 {
6009         struct list_head *pos, *next;
6010         struct nfs4_client *clp;
6011
6012         list_for_each_safe(pos, next, reaplist) {
6013                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6014                 trace_nfsd_clid_purged(&clp->cl_clientid);
6015                 list_del_init(&clp->cl_lru);
6016                 expire_client(clp);
6017         }
6018 }
6019
6020 static time64_t
6021 nfs4_laundromat(struct nfsd_net *nn)
6022 {
6023         struct nfs4_openowner *oo;
6024         struct nfs4_delegation *dp;
6025         struct nfs4_ol_stateid *stp;
6026         struct nfsd4_blocked_lock *nbl;
6027         struct list_head *pos, *next, reaplist;
6028         struct laundry_time lt = {
6029                 .cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease,
6030                 .new_timeo = nn->nfsd4_lease
6031         };
6032         struct nfs4_cpntf_state *cps;
6033         copy_stateid_t *cps_t;
6034         int i;
6035
6036         if (clients_still_reclaiming(nn)) {
6037                 lt.new_timeo = 0;
6038                 goto out;
6039         }
6040         nfsd4_end_grace(nn);
6041
6042         spin_lock(&nn->s2s_cp_lock);
6043         idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
6044                 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
6045                 if (cps->cp_stateid.cs_type == NFS4_COPYNOTIFY_STID &&
6046                                 state_expired(&lt, cps->cpntf_time))
6047                         _free_cpntf_state_locked(nn, cps);
6048         }
6049         spin_unlock(&nn->s2s_cp_lock);
6050         nfs4_get_client_reaplist(nn, &reaplist, &lt);
6051         nfs4_process_client_reaplist(&reaplist);
6052
6053         spin_lock(&state_lock);
6054         list_for_each_safe(pos, next, &nn->del_recall_lru) {
6055                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6056                 if (!state_expired(&lt, dp->dl_time))
6057                         break;
6058                 WARN_ON(!unhash_delegation_locked(dp));
6059                 list_add(&dp->dl_recall_lru, &reaplist);
6060         }
6061         spin_unlock(&state_lock);
6062         while (!list_empty(&reaplist)) {
6063                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
6064                                         dl_recall_lru);
6065                 list_del_init(&dp->dl_recall_lru);
6066                 revoke_delegation(dp);
6067         }
6068
6069         spin_lock(&nn->client_lock);
6070         while (!list_empty(&nn->close_lru)) {
6071                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
6072                                         oo_close_lru);
6073                 if (!state_expired(&lt, oo->oo_time))
6074                         break;
6075                 list_del_init(&oo->oo_close_lru);
6076                 stp = oo->oo_last_closed_stid;
6077                 oo->oo_last_closed_stid = NULL;
6078                 spin_unlock(&nn->client_lock);
6079                 nfs4_put_stid(&stp->st_stid);
6080                 spin_lock(&nn->client_lock);
6081         }
6082         spin_unlock(&nn->client_lock);
6083
6084         /*
6085          * It's possible for a client to try and acquire an already held lock
6086          * that is being held for a long time, and then lose interest in it.
6087          * So, we clean out any un-revisited request after a lease period
6088          * under the assumption that the client is no longer interested.
6089          *
6090          * RFC5661, sec. 9.6 states that the client must not rely on getting
6091          * notifications and must continue to poll for locks, even when the
6092          * server supports them. Thus this shouldn't lead to clients blocking
6093          * indefinitely once the lock does become free.
6094          */
6095         BUG_ON(!list_empty(&reaplist));
6096         spin_lock(&nn->blocked_locks_lock);
6097         while (!list_empty(&nn->blocked_locks_lru)) {
6098                 nbl = list_first_entry(&nn->blocked_locks_lru,
6099                                         struct nfsd4_blocked_lock, nbl_lru);
6100                 if (!state_expired(&lt, nbl->nbl_time))
6101                         break;
6102                 list_move(&nbl->nbl_lru, &reaplist);
6103                 list_del_init(&nbl->nbl_list);
6104         }
6105         spin_unlock(&nn->blocked_locks_lock);
6106
6107         while (!list_empty(&reaplist)) {
6108                 nbl = list_first_entry(&reaplist,
6109                                         struct nfsd4_blocked_lock, nbl_lru);
6110                 list_del_init(&nbl->nbl_lru);
6111                 free_blocked_lock(nbl);
6112         }
6113 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
6114         /* service the server-to-server copy delayed unmount list */
6115         nfsd4_ssc_expire_umount(nn);
6116 #endif
6117 out:
6118         return max_t(time64_t, lt.new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
6119 }
6120
6121 static void laundromat_main(struct work_struct *);
6122
6123 static void
6124 laundromat_main(struct work_struct *laundry)
6125 {
6126         time64_t t;
6127         struct delayed_work *dwork = to_delayed_work(laundry);
6128         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
6129                                            laundromat_work);
6130
6131         t = nfs4_laundromat(nn);
6132         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
6133 }
6134
6135 static void
6136 courtesy_client_reaper(struct work_struct *reaper)
6137 {
6138         struct list_head reaplist;
6139         struct delayed_work *dwork = to_delayed_work(reaper);
6140         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
6141                                         nfsd_shrinker_work);
6142
6143         nfs4_get_courtesy_client_reaplist(nn, &reaplist);
6144         nfs4_process_client_reaplist(&reaplist);
6145 }
6146
6147 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
6148 {
6149         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
6150                 return nfserr_bad_stateid;
6151         return nfs_ok;
6152 }
6153
6154 static
6155 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
6156 {
6157         __be32 status = nfserr_openmode;
6158
6159         /* For lock stateid's, we test the parent open, not the lock: */
6160         if (stp->st_openstp)
6161                 stp = stp->st_openstp;
6162         if ((flags & WR_STATE) && !access_permit_write(stp))
6163                 goto out;
6164         if ((flags & RD_STATE) && !access_permit_read(stp))
6165                 goto out;
6166         status = nfs_ok;
6167 out:
6168         return status;
6169 }
6170
6171 static inline __be32
6172 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
6173 {
6174         if (ONE_STATEID(stateid) && (flags & RD_STATE))
6175                 return nfs_ok;
6176         else if (opens_in_grace(net)) {
6177                 /* Answer in remaining cases depends on existence of
6178                  * conflicting state; so we must wait out the grace period. */
6179                 return nfserr_grace;
6180         } else if (flags & WR_STATE)
6181                 return nfs4_share_conflict(current_fh,
6182                                 NFS4_SHARE_DENY_WRITE);
6183         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
6184                 return nfs4_share_conflict(current_fh,
6185                                 NFS4_SHARE_DENY_READ);
6186 }
6187
6188 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
6189 {
6190         /*
6191          * When sessions are used the stateid generation number is ignored
6192          * when it is zero.
6193          */
6194         if (has_session && in->si_generation == 0)
6195                 return nfs_ok;
6196
6197         if (in->si_generation == ref->si_generation)
6198                 return nfs_ok;
6199
6200         /* If the client sends us a stateid from the future, it's buggy: */
6201         if (nfsd4_stateid_generation_after(in, ref))
6202                 return nfserr_bad_stateid;
6203         /*
6204          * However, we could see a stateid from the past, even from a
6205          * non-buggy client.  For example, if the client sends a lock
6206          * while some IO is outstanding, the lock may bump si_generation
6207          * while the IO is still in flight.  The client could avoid that
6208          * situation by waiting for responses on all the IO requests,
6209          * but better performance may result in retrying IO that
6210          * receives an old_stateid error if requests are rarely
6211          * reordered in flight:
6212          */
6213         return nfserr_old_stateid;
6214 }
6215
6216 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
6217 {
6218         __be32 ret;
6219
6220         spin_lock(&s->sc_lock);
6221         ret = nfsd4_verify_open_stid(s);
6222         if (ret == nfs_ok)
6223                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
6224         spin_unlock(&s->sc_lock);
6225         return ret;
6226 }
6227
6228 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
6229 {
6230         if (ols->st_stateowner->so_is_open_owner &&
6231             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
6232                 return nfserr_bad_stateid;
6233         return nfs_ok;
6234 }
6235
6236 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
6237 {
6238         struct nfs4_stid *s;
6239         __be32 status = nfserr_bad_stateid;
6240
6241         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6242                 CLOSE_STATEID(stateid))
6243                 return status;
6244         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid))
6245                 return status;
6246         spin_lock(&cl->cl_lock);
6247         s = find_stateid_locked(cl, stateid);
6248         if (!s)
6249                 goto out_unlock;
6250         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
6251         if (status)
6252                 goto out_unlock;
6253         switch (s->sc_type) {
6254         case NFS4_DELEG_STID:
6255                 status = nfs_ok;
6256                 break;
6257         case NFS4_REVOKED_DELEG_STID:
6258                 status = nfserr_deleg_revoked;
6259                 break;
6260         case NFS4_OPEN_STID:
6261         case NFS4_LOCK_STID:
6262                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
6263                 break;
6264         default:
6265                 printk("unknown stateid type %x\n", s->sc_type);
6266                 fallthrough;
6267         case NFS4_CLOSED_STID:
6268         case NFS4_CLOSED_DELEG_STID:
6269                 status = nfserr_bad_stateid;
6270         }
6271 out_unlock:
6272         spin_unlock(&cl->cl_lock);
6273         return status;
6274 }
6275
6276 __be32
6277 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
6278                      stateid_t *stateid, unsigned char typemask,
6279                      struct nfs4_stid **s, struct nfsd_net *nn)
6280 {
6281         __be32 status;
6282         struct nfs4_stid *stid;
6283         bool return_revoked = false;
6284
6285         /*
6286          *  only return revoked delegations if explicitly asked.
6287          *  otherwise we report revoked or bad_stateid status.
6288          */
6289         if (typemask & NFS4_REVOKED_DELEG_STID)
6290                 return_revoked = true;
6291         else if (typemask & NFS4_DELEG_STID)
6292                 typemask |= NFS4_REVOKED_DELEG_STID;
6293
6294         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6295                 CLOSE_STATEID(stateid))
6296                 return nfserr_bad_stateid;
6297         status = set_client(&stateid->si_opaque.so_clid, cstate, nn);
6298         if (status == nfserr_stale_clientid) {
6299                 if (cstate->session)
6300                         return nfserr_bad_stateid;
6301                 return nfserr_stale_stateid;
6302         }
6303         if (status)
6304                 return status;
6305         stid = find_stateid_by_type(cstate->clp, stateid, typemask);
6306         if (!stid)
6307                 return nfserr_bad_stateid;
6308         if ((stid->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
6309                 nfs4_put_stid(stid);
6310                 if (cstate->minorversion)
6311                         return nfserr_deleg_revoked;
6312                 return nfserr_bad_stateid;
6313         }
6314         *s = stid;
6315         return nfs_ok;
6316 }
6317
6318 static struct nfsd_file *
6319 nfs4_find_file(struct nfs4_stid *s, int flags)
6320 {
6321         if (!s)
6322                 return NULL;
6323
6324         switch (s->sc_type) {
6325         case NFS4_DELEG_STID:
6326                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
6327                         return NULL;
6328                 return nfsd_file_get(s->sc_file->fi_deleg_file);
6329         case NFS4_OPEN_STID:
6330         case NFS4_LOCK_STID:
6331                 if (flags & RD_STATE)
6332                         return find_readable_file(s->sc_file);
6333                 else
6334                         return find_writeable_file(s->sc_file);
6335         }
6336
6337         return NULL;
6338 }
6339
6340 static __be32
6341 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
6342 {
6343         __be32 status;
6344
6345         status = nfsd4_check_openowner_confirmed(ols);
6346         if (status)
6347                 return status;
6348         return nfs4_check_openmode(ols, flags);
6349 }
6350
6351 static __be32
6352 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
6353                 struct nfsd_file **nfp, int flags)
6354 {
6355         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
6356         struct nfsd_file *nf;
6357         __be32 status;
6358
6359         nf = nfs4_find_file(s, flags);
6360         if (nf) {
6361                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
6362                                 acc | NFSD_MAY_OWNER_OVERRIDE);
6363                 if (status) {
6364                         nfsd_file_put(nf);
6365                         goto out;
6366                 }
6367         } else {
6368                 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
6369                 if (status)
6370                         return status;
6371         }
6372         *nfp = nf;
6373 out:
6374         return status;
6375 }
6376 static void
6377 _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6378 {
6379         WARN_ON_ONCE(cps->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID);
6380         if (!refcount_dec_and_test(&cps->cp_stateid.cs_count))
6381                 return;
6382         list_del(&cps->cp_list);
6383         idr_remove(&nn->s2s_cp_stateids,
6384                    cps->cp_stateid.cs_stid.si_opaque.so_id);
6385         kfree(cps);
6386 }
6387 /*
6388  * A READ from an inter server to server COPY will have a
6389  * copy stateid. Look up the copy notify stateid from the
6390  * idr structure and take a reference on it.
6391  */
6392 __be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6393                           struct nfs4_client *clp,
6394                           struct nfs4_cpntf_state **cps)
6395 {
6396         copy_stateid_t *cps_t;
6397         struct nfs4_cpntf_state *state = NULL;
6398
6399         if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
6400                 return nfserr_bad_stateid;
6401         spin_lock(&nn->s2s_cp_lock);
6402         cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
6403         if (cps_t) {
6404                 state = container_of(cps_t, struct nfs4_cpntf_state,
6405                                      cp_stateid);
6406                 if (state->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID) {
6407                         state = NULL;
6408                         goto unlock;
6409                 }
6410                 if (!clp)
6411                         refcount_inc(&state->cp_stateid.cs_count);
6412                 else
6413                         _free_cpntf_state_locked(nn, state);
6414         }
6415 unlock:
6416         spin_unlock(&nn->s2s_cp_lock);
6417         if (!state)
6418                 return nfserr_bad_stateid;
6419         if (!clp && state)
6420                 *cps = state;
6421         return 0;
6422 }
6423
6424 static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6425                                struct nfs4_stid **stid)
6426 {
6427         __be32 status;
6428         struct nfs4_cpntf_state *cps = NULL;
6429         struct nfs4_client *found;
6430
6431         status = manage_cpntf_state(nn, st, NULL, &cps);
6432         if (status)
6433                 return status;
6434
6435         cps->cpntf_time = ktime_get_boottime_seconds();
6436
6437         status = nfserr_expired;
6438         found = lookup_clientid(&cps->cp_p_clid, true, nn);
6439         if (!found)
6440                 goto out;
6441
6442         *stid = find_stateid_by_type(found, &cps->cp_p_stateid,
6443                         NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID);
6444         if (*stid)
6445                 status = nfs_ok;
6446         else
6447                 status = nfserr_bad_stateid;
6448
6449         put_client_renew(found);
6450 out:
6451         nfs4_put_cpntf_state(nn, cps);
6452         return status;
6453 }
6454
6455 void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6456 {
6457         spin_lock(&nn->s2s_cp_lock);
6458         _free_cpntf_state_locked(nn, cps);
6459         spin_unlock(&nn->s2s_cp_lock);
6460 }
6461
6462 /*
6463  * Checks for stateid operations
6464  */
6465 __be32
6466 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
6467                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
6468                 stateid_t *stateid, int flags, struct nfsd_file **nfp,
6469                 struct nfs4_stid **cstid)
6470 {
6471         struct net *net = SVC_NET(rqstp);
6472         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6473         struct nfs4_stid *s = NULL;
6474         __be32 status;
6475
6476         if (nfp)
6477                 *nfp = NULL;
6478
6479         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
6480                 if (cstid)
6481                         status = nfserr_bad_stateid;
6482                 else
6483                         status = check_special_stateids(net, fhp, stateid,
6484                                                                         flags);
6485                 goto done;
6486         }
6487
6488         status = nfsd4_lookup_stateid(cstate, stateid,
6489                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
6490                                 &s, nn);
6491         if (status == nfserr_bad_stateid)
6492                 status = find_cpntf_state(nn, stateid, &s);
6493         if (status)
6494                 return status;
6495         status = nfsd4_stid_check_stateid_generation(stateid, s,
6496                         nfsd4_has_session(cstate));
6497         if (status)
6498                 goto out;
6499
6500         switch (s->sc_type) {
6501         case NFS4_DELEG_STID:
6502                 status = nfs4_check_delegmode(delegstateid(s), flags);
6503                 break;
6504         case NFS4_OPEN_STID:
6505         case NFS4_LOCK_STID:
6506                 status = nfs4_check_olstateid(openlockstateid(s), flags);
6507                 break;
6508         default:
6509                 status = nfserr_bad_stateid;
6510                 break;
6511         }
6512         if (status)
6513                 goto out;
6514         status = nfs4_check_fh(fhp, s);
6515
6516 done:
6517         if (status == nfs_ok && nfp)
6518                 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
6519 out:
6520         if (s) {
6521                 if (!status && cstid)
6522                         *cstid = s;
6523                 else
6524                         nfs4_put_stid(s);
6525         }
6526         return status;
6527 }
6528
6529 /*
6530  * Test if the stateid is valid
6531  */
6532 __be32
6533 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6534                    union nfsd4_op_u *u)
6535 {
6536         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
6537         struct nfsd4_test_stateid_id *stateid;
6538         struct nfs4_client *cl = cstate->clp;
6539
6540         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
6541                 stateid->ts_id_status =
6542                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
6543
6544         return nfs_ok;
6545 }
6546
6547 static __be32
6548 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
6549 {
6550         struct nfs4_ol_stateid *stp = openlockstateid(s);
6551         __be32 ret;
6552
6553         ret = nfsd4_lock_ol_stateid(stp);
6554         if (ret)
6555                 goto out_put_stid;
6556
6557         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
6558         if (ret)
6559                 goto out;
6560
6561         ret = nfserr_locks_held;
6562         if (check_for_locks(stp->st_stid.sc_file,
6563                             lockowner(stp->st_stateowner)))
6564                 goto out;
6565
6566         release_lock_stateid(stp);
6567         ret = nfs_ok;
6568
6569 out:
6570         mutex_unlock(&stp->st_mutex);
6571 out_put_stid:
6572         nfs4_put_stid(s);
6573         return ret;
6574 }
6575
6576 __be32
6577 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6578                    union nfsd4_op_u *u)
6579 {
6580         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
6581         stateid_t *stateid = &free_stateid->fr_stateid;
6582         struct nfs4_stid *s;
6583         struct nfs4_delegation *dp;
6584         struct nfs4_client *cl = cstate->clp;
6585         __be32 ret = nfserr_bad_stateid;
6586
6587         spin_lock(&cl->cl_lock);
6588         s = find_stateid_locked(cl, stateid);
6589         if (!s)
6590                 goto out_unlock;
6591         spin_lock(&s->sc_lock);
6592         switch (s->sc_type) {
6593         case NFS4_DELEG_STID:
6594                 ret = nfserr_locks_held;
6595                 break;
6596         case NFS4_OPEN_STID:
6597                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
6598                 if (ret)
6599                         break;
6600                 ret = nfserr_locks_held;
6601                 break;
6602         case NFS4_LOCK_STID:
6603                 spin_unlock(&s->sc_lock);
6604                 refcount_inc(&s->sc_count);
6605                 spin_unlock(&cl->cl_lock);
6606                 ret = nfsd4_free_lock_stateid(stateid, s);
6607                 goto out;
6608         case NFS4_REVOKED_DELEG_STID:
6609                 spin_unlock(&s->sc_lock);
6610                 dp = delegstateid(s);
6611                 list_del_init(&dp->dl_recall_lru);
6612                 spin_unlock(&cl->cl_lock);
6613                 nfs4_put_stid(s);
6614                 ret = nfs_ok;
6615                 goto out;
6616         /* Default falls through and returns nfserr_bad_stateid */
6617         }
6618         spin_unlock(&s->sc_lock);
6619 out_unlock:
6620         spin_unlock(&cl->cl_lock);
6621 out:
6622         return ret;
6623 }
6624
6625 static inline int
6626 setlkflg (int type)
6627 {
6628         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
6629                 RD_STATE : WR_STATE;
6630 }
6631
6632 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
6633 {
6634         struct svc_fh *current_fh = &cstate->current_fh;
6635         struct nfs4_stateowner *sop = stp->st_stateowner;
6636         __be32 status;
6637
6638         status = nfsd4_check_seqid(cstate, sop, seqid);
6639         if (status)
6640                 return status;
6641         status = nfsd4_lock_ol_stateid(stp);
6642         if (status != nfs_ok)
6643                 return status;
6644         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
6645         if (status == nfs_ok)
6646                 status = nfs4_check_fh(current_fh, &stp->st_stid);
6647         if (status != nfs_ok)
6648                 mutex_unlock(&stp->st_mutex);
6649         return status;
6650 }
6651
6652 /*
6653  * Checks for sequence id mutating operations.
6654  */
6655 static __be32
6656 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
6657                          stateid_t *stateid, char typemask,
6658                          struct nfs4_ol_stateid **stpp,
6659                          struct nfsd_net *nn)
6660 {
6661         __be32 status;
6662         struct nfs4_stid *s;
6663         struct nfs4_ol_stateid *stp = NULL;
6664
6665         trace_nfsd_preprocess(seqid, stateid);
6666
6667         *stpp = NULL;
6668         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
6669         if (status)
6670                 return status;
6671         stp = openlockstateid(s);
6672         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
6673
6674         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
6675         if (!status)
6676                 *stpp = stp;
6677         else
6678                 nfs4_put_stid(&stp->st_stid);
6679         return status;
6680 }
6681
6682 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
6683                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
6684 {
6685         __be32 status;
6686         struct nfs4_openowner *oo;
6687         struct nfs4_ol_stateid *stp;
6688
6689         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
6690                                                 NFS4_OPEN_STID, &stp, nn);
6691         if (status)
6692                 return status;
6693         oo = openowner(stp->st_stateowner);
6694         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
6695                 mutex_unlock(&stp->st_mutex);
6696                 nfs4_put_stid(&stp->st_stid);
6697                 return nfserr_bad_stateid;
6698         }
6699         *stpp = stp;
6700         return nfs_ok;
6701 }
6702
6703 __be32
6704 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6705                    union nfsd4_op_u *u)
6706 {
6707         struct nfsd4_open_confirm *oc = &u->open_confirm;
6708         __be32 status;
6709         struct nfs4_openowner *oo;
6710         struct nfs4_ol_stateid *stp;
6711         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6712
6713         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
6714                         cstate->current_fh.fh_dentry);
6715
6716         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
6717         if (status)
6718                 return status;
6719
6720         status = nfs4_preprocess_seqid_op(cstate,
6721                                         oc->oc_seqid, &oc->oc_req_stateid,
6722                                         NFS4_OPEN_STID, &stp, nn);
6723         if (status)
6724                 goto out;
6725         oo = openowner(stp->st_stateowner);
6726         status = nfserr_bad_stateid;
6727         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
6728                 mutex_unlock(&stp->st_mutex);
6729                 goto put_stateid;
6730         }
6731         oo->oo_flags |= NFS4_OO_CONFIRMED;
6732         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
6733         mutex_unlock(&stp->st_mutex);
6734         trace_nfsd_open_confirm(oc->oc_seqid, &stp->st_stid.sc_stateid);
6735         nfsd4_client_record_create(oo->oo_owner.so_client);
6736         status = nfs_ok;
6737 put_stateid:
6738         nfs4_put_stid(&stp->st_stid);
6739 out:
6740         nfsd4_bump_seqid(cstate, status);
6741         return status;
6742 }
6743
6744 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
6745 {
6746         if (!test_access(access, stp))
6747                 return;
6748         nfs4_file_put_access(stp->st_stid.sc_file, access);
6749         clear_access(access, stp);
6750 }
6751
6752 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
6753 {
6754         switch (to_access) {
6755         case NFS4_SHARE_ACCESS_READ:
6756                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
6757                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6758                 break;
6759         case NFS4_SHARE_ACCESS_WRITE:
6760                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
6761                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6762                 break;
6763         case NFS4_SHARE_ACCESS_BOTH:
6764                 break;
6765         default:
6766                 WARN_ON_ONCE(1);
6767         }
6768 }
6769
6770 __be32
6771 nfsd4_open_downgrade(struct svc_rqst *rqstp,
6772                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
6773 {
6774         struct nfsd4_open_downgrade *od = &u->open_downgrade;
6775         __be32 status;
6776         struct nfs4_ol_stateid *stp;
6777         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6778
6779         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
6780                         cstate->current_fh.fh_dentry);
6781
6782         /* We don't yet support WANT bits: */
6783         if (od->od_deleg_want)
6784                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
6785                         od->od_deleg_want);
6786
6787         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
6788                                         &od->od_stateid, &stp, nn);
6789         if (status)
6790                 goto out;
6791         status = nfserr_inval;
6792         if (!test_access(od->od_share_access, stp)) {
6793                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
6794                         stp->st_access_bmap, od->od_share_access);
6795                 goto put_stateid;
6796         }
6797         if (!test_deny(od->od_share_deny, stp)) {
6798                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
6799                         stp->st_deny_bmap, od->od_share_deny);
6800                 goto put_stateid;
6801         }
6802         nfs4_stateid_downgrade(stp, od->od_share_access);
6803         reset_union_bmap_deny(od->od_share_deny, stp);
6804         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
6805         status = nfs_ok;
6806 put_stateid:
6807         mutex_unlock(&stp->st_mutex);
6808         nfs4_put_stid(&stp->st_stid);
6809 out:
6810         nfsd4_bump_seqid(cstate, status);
6811         return status;
6812 }
6813
6814 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
6815 {
6816         struct nfs4_client *clp = s->st_stid.sc_client;
6817         bool unhashed;
6818         LIST_HEAD(reaplist);
6819         struct nfs4_ol_stateid *stp;
6820
6821         spin_lock(&clp->cl_lock);
6822         unhashed = unhash_open_stateid(s, &reaplist);
6823
6824         if (clp->cl_minorversion) {
6825                 if (unhashed)
6826                         put_ol_stateid_locked(s, &reaplist);
6827                 spin_unlock(&clp->cl_lock);
6828                 list_for_each_entry(stp, &reaplist, st_locks)
6829                         nfs4_free_cpntf_statelist(clp->net, &stp->st_stid);
6830                 free_ol_stateid_reaplist(&reaplist);
6831         } else {
6832                 spin_unlock(&clp->cl_lock);
6833                 free_ol_stateid_reaplist(&reaplist);
6834                 if (unhashed)
6835                         move_to_close_lru(s, clp->net);
6836         }
6837 }
6838
6839 /*
6840  * nfs4_unlock_state() called after encode
6841  */
6842 __be32
6843 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6844                 union nfsd4_op_u *u)
6845 {
6846         struct nfsd4_close *close = &u->close;
6847         __be32 status;
6848         struct nfs4_ol_stateid *stp;
6849         struct net *net = SVC_NET(rqstp);
6850         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6851
6852         dprintk("NFSD: nfsd4_close on file %pd\n",
6853                         cstate->current_fh.fh_dentry);
6854
6855         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
6856                                         &close->cl_stateid,
6857                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
6858                                         &stp, nn);
6859         nfsd4_bump_seqid(cstate, status);
6860         if (status)
6861                 goto out;
6862
6863         stp->st_stid.sc_type = NFS4_CLOSED_STID;
6864
6865         /*
6866          * Technically we don't _really_ have to increment or copy it, since
6867          * it should just be gone after this operation and we clobber the
6868          * copied value below, but we continue to do so here just to ensure
6869          * that racing ops see that there was a state change.
6870          */
6871         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
6872
6873         nfsd4_close_open_stateid(stp);
6874         mutex_unlock(&stp->st_mutex);
6875
6876         /* v4.1+ suggests that we send a special stateid in here, since the
6877          * clients should just ignore this anyway. Since this is not useful
6878          * for v4.0 clients either, we set it to the special close_stateid
6879          * universally.
6880          *
6881          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
6882          */
6883         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
6884
6885         /* put reference from nfs4_preprocess_seqid_op */
6886         nfs4_put_stid(&stp->st_stid);
6887 out:
6888         return status;
6889 }
6890
6891 __be32
6892 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6893                   union nfsd4_op_u *u)
6894 {
6895         struct nfsd4_delegreturn *dr = &u->delegreturn;
6896         struct nfs4_delegation *dp;
6897         stateid_t *stateid = &dr->dr_stateid;
6898         struct nfs4_stid *s;
6899         __be32 status;
6900         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6901
6902         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6903                 return status;
6904
6905         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
6906         if (status)
6907                 goto out;
6908         dp = delegstateid(s);
6909         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
6910         if (status)
6911                 goto put_stateid;
6912
6913         wake_up_var(d_inode(cstate->current_fh.fh_dentry));
6914         destroy_delegation(dp);
6915 put_stateid:
6916         nfs4_put_stid(&dp->dl_stid);
6917 out:
6918         return status;
6919 }
6920
6921 /* last octet in a range */
6922 static inline u64
6923 last_byte_offset(u64 start, u64 len)
6924 {
6925         u64 end;
6926
6927         WARN_ON_ONCE(!len);
6928         end = start + len;
6929         return end > start ? end - 1: NFS4_MAX_UINT64;
6930 }
6931
6932 /*
6933  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
6934  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
6935  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
6936  * locking, this prevents us from being completely protocol-compliant.  The
6937  * real solution to this problem is to start using unsigned file offsets in
6938  * the VFS, but this is a very deep change!
6939  */
6940 static inline void
6941 nfs4_transform_lock_offset(struct file_lock *lock)
6942 {
6943         if (lock->fl_start < 0)
6944                 lock->fl_start = OFFSET_MAX;
6945         if (lock->fl_end < 0)
6946                 lock->fl_end = OFFSET_MAX;
6947 }
6948
6949 static fl_owner_t
6950 nfsd4_lm_get_owner(fl_owner_t owner)
6951 {
6952         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6953
6954         nfs4_get_stateowner(&lo->lo_owner);
6955         return owner;
6956 }
6957
6958 static void
6959 nfsd4_lm_put_owner(fl_owner_t owner)
6960 {
6961         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6962
6963         if (lo)
6964                 nfs4_put_stateowner(&lo->lo_owner);
6965 }
6966
6967 /* return pointer to struct nfs4_client if client is expirable */
6968 static bool
6969 nfsd4_lm_lock_expirable(struct file_lock *cfl)
6970 {
6971         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)cfl->fl_owner;
6972         struct nfs4_client *clp = lo->lo_owner.so_client;
6973         struct nfsd_net *nn;
6974
6975         if (try_to_expire_client(clp)) {
6976                 nn = net_generic(clp->net, nfsd_net_id);
6977                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
6978                 return true;
6979         }
6980         return false;
6981 }
6982
6983 /* schedule laundromat to run immediately and wait for it to complete */
6984 static void
6985 nfsd4_lm_expire_lock(void)
6986 {
6987         flush_workqueue(laundry_wq);
6988 }
6989
6990 static void
6991 nfsd4_lm_notify(struct file_lock *fl)
6992 {
6993         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
6994         struct net                      *net = lo->lo_owner.so_client->net;
6995         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
6996         struct nfsd4_blocked_lock       *nbl = container_of(fl,
6997                                                 struct nfsd4_blocked_lock, nbl_lock);
6998         bool queue = false;
6999
7000         /* An empty list means that something else is going to be using it */
7001         spin_lock(&nn->blocked_locks_lock);
7002         if (!list_empty(&nbl->nbl_list)) {
7003                 list_del_init(&nbl->nbl_list);
7004                 list_del_init(&nbl->nbl_lru);
7005                 queue = true;
7006         }
7007         spin_unlock(&nn->blocked_locks_lock);
7008
7009         if (queue) {
7010                 trace_nfsd_cb_notify_lock(lo, nbl);
7011                 nfsd4_run_cb(&nbl->nbl_cb);
7012         }
7013 }
7014
7015 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
7016         .lm_mod_owner = THIS_MODULE,
7017         .lm_notify = nfsd4_lm_notify,
7018         .lm_get_owner = nfsd4_lm_get_owner,
7019         .lm_put_owner = nfsd4_lm_put_owner,
7020         .lm_lock_expirable = nfsd4_lm_lock_expirable,
7021         .lm_expire_lock = nfsd4_lm_expire_lock,
7022 };
7023
7024 static inline void
7025 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
7026 {
7027         struct nfs4_lockowner *lo;
7028
7029         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
7030                 lo = (struct nfs4_lockowner *) fl->fl_owner;
7031                 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
7032                                                 GFP_KERNEL);
7033                 if (!deny->ld_owner.data)
7034                         /* We just don't care that much */
7035                         goto nevermind;
7036                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
7037         } else {
7038 nevermind:
7039                 deny->ld_owner.len = 0;
7040                 deny->ld_owner.data = NULL;
7041                 deny->ld_clientid.cl_boot = 0;
7042                 deny->ld_clientid.cl_id = 0;
7043         }
7044         deny->ld_start = fl->fl_start;
7045         deny->ld_length = NFS4_MAX_UINT64;
7046         if (fl->fl_end != NFS4_MAX_UINT64)
7047                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
7048         deny->ld_type = NFS4_READ_LT;
7049         if (fl->fl_type != F_RDLCK)
7050                 deny->ld_type = NFS4_WRITE_LT;
7051 }
7052
7053 static struct nfs4_lockowner *
7054 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
7055 {
7056         unsigned int strhashval = ownerstr_hashval(owner);
7057         struct nfs4_stateowner *so;
7058
7059         lockdep_assert_held(&clp->cl_lock);
7060
7061         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
7062                             so_strhash) {
7063                 if (so->so_is_open_owner)
7064                         continue;
7065                 if (same_owner_str(so, owner))
7066                         return lockowner(nfs4_get_stateowner(so));
7067         }
7068         return NULL;
7069 }
7070
7071 static struct nfs4_lockowner *
7072 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
7073 {
7074         struct nfs4_lockowner *lo;
7075
7076         spin_lock(&clp->cl_lock);
7077         lo = find_lockowner_str_locked(clp, owner);
7078         spin_unlock(&clp->cl_lock);
7079         return lo;
7080 }
7081
7082 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
7083 {
7084         unhash_lockowner_locked(lockowner(sop));
7085 }
7086
7087 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
7088 {
7089         struct nfs4_lockowner *lo = lockowner(sop);
7090
7091         kmem_cache_free(lockowner_slab, lo);
7092 }
7093
7094 static const struct nfs4_stateowner_operations lockowner_ops = {
7095         .so_unhash =    nfs4_unhash_lockowner,
7096         .so_free =      nfs4_free_lockowner,
7097 };
7098
7099 /*
7100  * Alloc a lock owner structure.
7101  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
7102  * occurred.
7103  *
7104  * strhashval = ownerstr_hashval
7105  */
7106 static struct nfs4_lockowner *
7107 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
7108                            struct nfs4_ol_stateid *open_stp,
7109                            struct nfsd4_lock *lock)
7110 {
7111         struct nfs4_lockowner *lo, *ret;
7112
7113         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
7114         if (!lo)
7115                 return NULL;
7116         INIT_LIST_HEAD(&lo->lo_blocked);
7117         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
7118         lo->lo_owner.so_is_open_owner = 0;
7119         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
7120         lo->lo_owner.so_ops = &lockowner_ops;
7121         spin_lock(&clp->cl_lock);
7122         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
7123         if (ret == NULL) {
7124                 list_add(&lo->lo_owner.so_strhash,
7125                          &clp->cl_ownerstr_hashtbl[strhashval]);
7126                 ret = lo;
7127         } else
7128                 nfs4_free_stateowner(&lo->lo_owner);
7129
7130         spin_unlock(&clp->cl_lock);
7131         return ret;
7132 }
7133
7134 static struct nfs4_ol_stateid *
7135 find_lock_stateid(const struct nfs4_lockowner *lo,
7136                   const struct nfs4_ol_stateid *ost)
7137 {
7138         struct nfs4_ol_stateid *lst;
7139
7140         lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
7141
7142         /* If ost is not hashed, ost->st_locks will not be valid */
7143         if (!nfs4_ol_stateid_unhashed(ost))
7144                 list_for_each_entry(lst, &ost->st_locks, st_locks) {
7145                         if (lst->st_stateowner == &lo->lo_owner) {
7146                                 refcount_inc(&lst->st_stid.sc_count);
7147                                 return lst;
7148                         }
7149                 }
7150         return NULL;
7151 }
7152
7153 static struct nfs4_ol_stateid *
7154 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
7155                   struct nfs4_file *fp, struct inode *inode,
7156                   struct nfs4_ol_stateid *open_stp)
7157 {
7158         struct nfs4_client *clp = lo->lo_owner.so_client;
7159         struct nfs4_ol_stateid *retstp;
7160
7161         mutex_init(&stp->st_mutex);
7162         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
7163 retry:
7164         spin_lock(&clp->cl_lock);
7165         if (nfs4_ol_stateid_unhashed(open_stp))
7166                 goto out_close;
7167         retstp = find_lock_stateid(lo, open_stp);
7168         if (retstp)
7169                 goto out_found;
7170         refcount_inc(&stp->st_stid.sc_count);
7171         stp->st_stid.sc_type = NFS4_LOCK_STID;
7172         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
7173         get_nfs4_file(fp);
7174         stp->st_stid.sc_file = fp;
7175         stp->st_access_bmap = 0;
7176         stp->st_deny_bmap = open_stp->st_deny_bmap;
7177         stp->st_openstp = open_stp;
7178         spin_lock(&fp->fi_lock);
7179         list_add(&stp->st_locks, &open_stp->st_locks);
7180         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
7181         list_add(&stp->st_perfile, &fp->fi_stateids);
7182         spin_unlock(&fp->fi_lock);
7183         spin_unlock(&clp->cl_lock);
7184         return stp;
7185 out_found:
7186         spin_unlock(&clp->cl_lock);
7187         if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
7188                 nfs4_put_stid(&retstp->st_stid);
7189                 goto retry;
7190         }
7191         /* To keep mutex tracking happy */
7192         mutex_unlock(&stp->st_mutex);
7193         return retstp;
7194 out_close:
7195         spin_unlock(&clp->cl_lock);
7196         mutex_unlock(&stp->st_mutex);
7197         return NULL;
7198 }
7199
7200 static struct nfs4_ol_stateid *
7201 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
7202                             struct inode *inode, struct nfs4_ol_stateid *ost,
7203                             bool *new)
7204 {
7205         struct nfs4_stid *ns = NULL;
7206         struct nfs4_ol_stateid *lst;
7207         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7208         struct nfs4_client *clp = oo->oo_owner.so_client;
7209
7210         *new = false;
7211         spin_lock(&clp->cl_lock);
7212         lst = find_lock_stateid(lo, ost);
7213         spin_unlock(&clp->cl_lock);
7214         if (lst != NULL) {
7215                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
7216                         goto out;
7217                 nfs4_put_stid(&lst->st_stid);
7218         }
7219         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
7220         if (ns == NULL)
7221                 return NULL;
7222
7223         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
7224         if (lst == openlockstateid(ns))
7225                 *new = true;
7226         else
7227                 nfs4_put_stid(ns);
7228 out:
7229         return lst;
7230 }
7231
7232 static int
7233 check_lock_length(u64 offset, u64 length)
7234 {
7235         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
7236                 (length > ~offset)));
7237 }
7238
7239 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
7240 {
7241         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
7242
7243         lockdep_assert_held(&fp->fi_lock);
7244
7245         if (test_access(access, lock_stp))
7246                 return;
7247         __nfs4_file_get_access(fp, access);
7248         set_access(access, lock_stp);
7249 }
7250
7251 static __be32
7252 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
7253                             struct nfs4_ol_stateid *ost,
7254                             struct nfsd4_lock *lock,
7255                             struct nfs4_ol_stateid **plst, bool *new)
7256 {
7257         __be32 status;
7258         struct nfs4_file *fi = ost->st_stid.sc_file;
7259         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7260         struct nfs4_client *cl = oo->oo_owner.so_client;
7261         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
7262         struct nfs4_lockowner *lo;
7263         struct nfs4_ol_stateid *lst;
7264         unsigned int strhashval;
7265
7266         lo = find_lockowner_str(cl, &lock->lk_new_owner);
7267         if (!lo) {
7268                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
7269                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
7270                 if (lo == NULL)
7271                         return nfserr_jukebox;
7272         } else {
7273                 /* with an existing lockowner, seqids must be the same */
7274                 status = nfserr_bad_seqid;
7275                 if (!cstate->minorversion &&
7276                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
7277                         goto out;
7278         }
7279
7280         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
7281         if (lst == NULL) {
7282                 status = nfserr_jukebox;
7283                 goto out;
7284         }
7285
7286         status = nfs_ok;
7287         *plst = lst;
7288 out:
7289         nfs4_put_stateowner(&lo->lo_owner);
7290         return status;
7291 }
7292
7293 /*
7294  *  LOCK operation
7295  */
7296 __be32
7297 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7298            union nfsd4_op_u *u)
7299 {
7300         struct nfsd4_lock *lock = &u->lock;
7301         struct nfs4_openowner *open_sop = NULL;
7302         struct nfs4_lockowner *lock_sop = NULL;
7303         struct nfs4_ol_stateid *lock_stp = NULL;
7304         struct nfs4_ol_stateid *open_stp = NULL;
7305         struct nfs4_file *fp;
7306         struct nfsd_file *nf = NULL;
7307         struct nfsd4_blocked_lock *nbl = NULL;
7308         struct file_lock *file_lock = NULL;
7309         struct file_lock *conflock = NULL;
7310         __be32 status = 0;
7311         int lkflg;
7312         int err;
7313         bool new = false;
7314         unsigned char fl_type;
7315         unsigned int fl_flags = FL_POSIX;
7316         struct net *net = SVC_NET(rqstp);
7317         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7318
7319         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
7320                 (long long) lock->lk_offset,
7321                 (long long) lock->lk_length);
7322
7323         if (check_lock_length(lock->lk_offset, lock->lk_length))
7324                  return nfserr_inval;
7325
7326         if ((status = fh_verify(rqstp, &cstate->current_fh,
7327                                 S_IFREG, NFSD_MAY_LOCK))) {
7328                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
7329                 return status;
7330         }
7331
7332         if (lock->lk_is_new) {
7333                 if (nfsd4_has_session(cstate))
7334                         /* See rfc 5661 18.10.3: given clientid is ignored: */
7335                         memcpy(&lock->lk_new_clientid,
7336                                 &cstate->clp->cl_clientid,
7337                                 sizeof(clientid_t));
7338
7339                 /* validate and update open stateid and open seqid */
7340                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
7341                                         lock->lk_new_open_seqid,
7342                                         &lock->lk_new_open_stateid,
7343                                         &open_stp, nn);
7344                 if (status)
7345                         goto out;
7346                 mutex_unlock(&open_stp->st_mutex);
7347                 open_sop = openowner(open_stp->st_stateowner);
7348                 status = nfserr_bad_stateid;
7349                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
7350                                                 &lock->lk_new_clientid))
7351                         goto out;
7352                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
7353                                                         &lock_stp, &new);
7354         } else {
7355                 status = nfs4_preprocess_seqid_op(cstate,
7356                                        lock->lk_old_lock_seqid,
7357                                        &lock->lk_old_lock_stateid,
7358                                        NFS4_LOCK_STID, &lock_stp, nn);
7359         }
7360         if (status)
7361                 goto out;
7362         lock_sop = lockowner(lock_stp->st_stateowner);
7363
7364         lkflg = setlkflg(lock->lk_type);
7365         status = nfs4_check_openmode(lock_stp, lkflg);
7366         if (status)
7367                 goto out;
7368
7369         status = nfserr_grace;
7370         if (locks_in_grace(net) && !lock->lk_reclaim)
7371                 goto out;
7372         status = nfserr_no_grace;
7373         if (!locks_in_grace(net) && lock->lk_reclaim)
7374                 goto out;
7375
7376         if (lock->lk_reclaim)
7377                 fl_flags |= FL_RECLAIM;
7378
7379         fp = lock_stp->st_stid.sc_file;
7380         switch (lock->lk_type) {
7381                 case NFS4_READW_LT:
7382                         if (nfsd4_has_session(cstate))
7383                                 fl_flags |= FL_SLEEP;
7384                         fallthrough;
7385                 case NFS4_READ_LT:
7386                         spin_lock(&fp->fi_lock);
7387                         nf = find_readable_file_locked(fp);
7388                         if (nf)
7389                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
7390                         spin_unlock(&fp->fi_lock);
7391                         fl_type = F_RDLCK;
7392                         break;
7393                 case NFS4_WRITEW_LT:
7394                         if (nfsd4_has_session(cstate))
7395                                 fl_flags |= FL_SLEEP;
7396                         fallthrough;
7397                 case NFS4_WRITE_LT:
7398                         spin_lock(&fp->fi_lock);
7399                         nf = find_writeable_file_locked(fp);
7400                         if (nf)
7401                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
7402                         spin_unlock(&fp->fi_lock);
7403                         fl_type = F_WRLCK;
7404                         break;
7405                 default:
7406                         status = nfserr_inval;
7407                 goto out;
7408         }
7409
7410         if (!nf) {
7411                 status = nfserr_openmode;
7412                 goto out;
7413         }
7414
7415         /*
7416          * Most filesystems with their own ->lock operations will block
7417          * the nfsd thread waiting to acquire the lock.  That leads to
7418          * deadlocks (we don't want every nfsd thread tied up waiting
7419          * for file locks), so don't attempt blocking lock notifications
7420          * on those filesystems:
7421          */
7422         if (nf->nf_file->f_op->lock)
7423                 fl_flags &= ~FL_SLEEP;
7424
7425         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
7426         if (!nbl) {
7427                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
7428                 status = nfserr_jukebox;
7429                 goto out;
7430         }
7431
7432         file_lock = &nbl->nbl_lock;
7433         file_lock->fl_type = fl_type;
7434         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
7435         file_lock->fl_pid = current->tgid;
7436         file_lock->fl_file = nf->nf_file;
7437         file_lock->fl_flags = fl_flags;
7438         file_lock->fl_lmops = &nfsd_posix_mng_ops;
7439         file_lock->fl_start = lock->lk_offset;
7440         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
7441         nfs4_transform_lock_offset(file_lock);
7442
7443         conflock = locks_alloc_lock();
7444         if (!conflock) {
7445                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7446                 status = nfserr_jukebox;
7447                 goto out;
7448         }
7449
7450         if (fl_flags & FL_SLEEP) {
7451                 nbl->nbl_time = ktime_get_boottime_seconds();
7452                 spin_lock(&nn->blocked_locks_lock);
7453                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
7454                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
7455                 kref_get(&nbl->nbl_kref);
7456                 spin_unlock(&nn->blocked_locks_lock);
7457         }
7458
7459         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
7460         switch (err) {
7461         case 0: /* success! */
7462                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
7463                 status = 0;
7464                 if (lock->lk_reclaim)
7465                         nn->somebody_reclaimed = true;
7466                 break;
7467         case FILE_LOCK_DEFERRED:
7468                 kref_put(&nbl->nbl_kref, free_nbl);
7469                 nbl = NULL;
7470                 fallthrough;
7471         case -EAGAIN:           /* conflock holds conflicting lock */
7472                 status = nfserr_denied;
7473                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
7474                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
7475                 break;
7476         case -EDEADLK:
7477                 status = nfserr_deadlock;
7478                 break;
7479         default:
7480                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
7481                 status = nfserrno(err);
7482                 break;
7483         }
7484 out:
7485         if (nbl) {
7486                 /* dequeue it if we queued it before */
7487                 if (fl_flags & FL_SLEEP) {
7488                         spin_lock(&nn->blocked_locks_lock);
7489                         if (!list_empty(&nbl->nbl_list) &&
7490                             !list_empty(&nbl->nbl_lru)) {
7491                                 list_del_init(&nbl->nbl_list);
7492                                 list_del_init(&nbl->nbl_lru);
7493                                 kref_put(&nbl->nbl_kref, free_nbl);
7494                         }
7495                         /* nbl can use one of lists to be linked to reaplist */
7496                         spin_unlock(&nn->blocked_locks_lock);
7497                 }
7498                 free_blocked_lock(nbl);
7499         }
7500         if (nf)
7501                 nfsd_file_put(nf);
7502         if (lock_stp) {
7503                 /* Bump seqid manually if the 4.0 replay owner is openowner */
7504                 if (cstate->replay_owner &&
7505                     cstate->replay_owner != &lock_sop->lo_owner &&
7506                     seqid_mutating_err(ntohl(status)))
7507                         lock_sop->lo_owner.so_seqid++;
7508
7509                 /*
7510                  * If this is a new, never-before-used stateid, and we are
7511                  * returning an error, then just go ahead and release it.
7512                  */
7513                 if (status && new)
7514                         release_lock_stateid(lock_stp);
7515
7516                 mutex_unlock(&lock_stp->st_mutex);
7517
7518                 nfs4_put_stid(&lock_stp->st_stid);
7519         }
7520         if (open_stp)
7521                 nfs4_put_stid(&open_stp->st_stid);
7522         nfsd4_bump_seqid(cstate, status);
7523         if (conflock)
7524                 locks_free_lock(conflock);
7525         return status;
7526 }
7527
7528 /*
7529  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
7530  * so we do a temporary open here just to get an open file to pass to
7531  * vfs_test_lock.
7532  */
7533 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
7534 {
7535         struct nfsd_file *nf;
7536         struct inode *inode;
7537         __be32 err;
7538
7539         err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
7540         if (err)
7541                 return err;
7542         inode = fhp->fh_dentry->d_inode;
7543         inode_lock(inode); /* to block new leases till after test_lock: */
7544         err = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
7545         if (err)
7546                 goto out;
7547         lock->fl_file = nf->nf_file;
7548         err = nfserrno(vfs_test_lock(nf->nf_file, lock));
7549         lock->fl_file = NULL;
7550 out:
7551         inode_unlock(inode);
7552         nfsd_file_put(nf);
7553         return err;
7554 }
7555
7556 /*
7557  * LOCKT operation
7558  */
7559 __be32
7560 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7561             union nfsd4_op_u *u)
7562 {
7563         struct nfsd4_lockt *lockt = &u->lockt;
7564         struct file_lock *file_lock = NULL;
7565         struct nfs4_lockowner *lo = NULL;
7566         __be32 status;
7567         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7568
7569         if (locks_in_grace(SVC_NET(rqstp)))
7570                 return nfserr_grace;
7571
7572         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
7573                  return nfserr_inval;
7574
7575         if (!nfsd4_has_session(cstate)) {
7576                 status = set_client(&lockt->lt_clientid, cstate, nn);
7577                 if (status)
7578                         goto out;
7579         }
7580
7581         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
7582                 goto out;
7583
7584         file_lock = locks_alloc_lock();
7585         if (!file_lock) {
7586                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7587                 status = nfserr_jukebox;
7588                 goto out;
7589         }
7590
7591         switch (lockt->lt_type) {
7592                 case NFS4_READ_LT:
7593                 case NFS4_READW_LT:
7594                         file_lock->fl_type = F_RDLCK;
7595                         break;
7596                 case NFS4_WRITE_LT:
7597                 case NFS4_WRITEW_LT:
7598                         file_lock->fl_type = F_WRLCK;
7599                         break;
7600                 default:
7601                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
7602                         status = nfserr_inval;
7603                         goto out;
7604         }
7605
7606         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
7607         if (lo)
7608                 file_lock->fl_owner = (fl_owner_t)lo;
7609         file_lock->fl_pid = current->tgid;
7610         file_lock->fl_flags = FL_POSIX;
7611
7612         file_lock->fl_start = lockt->lt_offset;
7613         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
7614
7615         nfs4_transform_lock_offset(file_lock);
7616
7617         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
7618         if (status)
7619                 goto out;
7620
7621         if (file_lock->fl_type != F_UNLCK) {
7622                 status = nfserr_denied;
7623                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
7624         }
7625 out:
7626         if (lo)
7627                 nfs4_put_stateowner(&lo->lo_owner);
7628         if (file_lock)
7629                 locks_free_lock(file_lock);
7630         return status;
7631 }
7632
7633 __be32
7634 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7635             union nfsd4_op_u *u)
7636 {
7637         struct nfsd4_locku *locku = &u->locku;
7638         struct nfs4_ol_stateid *stp;
7639         struct nfsd_file *nf = NULL;
7640         struct file_lock *file_lock = NULL;
7641         __be32 status;
7642         int err;
7643         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7644
7645         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
7646                 (long long) locku->lu_offset,
7647                 (long long) locku->lu_length);
7648
7649         if (check_lock_length(locku->lu_offset, locku->lu_length))
7650                  return nfserr_inval;
7651
7652         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
7653                                         &locku->lu_stateid, NFS4_LOCK_STID,
7654                                         &stp, nn);
7655         if (status)
7656                 goto out;
7657         nf = find_any_file(stp->st_stid.sc_file);
7658         if (!nf) {
7659                 status = nfserr_lock_range;
7660                 goto put_stateid;
7661         }
7662         file_lock = locks_alloc_lock();
7663         if (!file_lock) {
7664                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7665                 status = nfserr_jukebox;
7666                 goto put_file;
7667         }
7668
7669         file_lock->fl_type = F_UNLCK;
7670         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
7671         file_lock->fl_pid = current->tgid;
7672         file_lock->fl_file = nf->nf_file;
7673         file_lock->fl_flags = FL_POSIX;
7674         file_lock->fl_lmops = &nfsd_posix_mng_ops;
7675         file_lock->fl_start = locku->lu_offset;
7676
7677         file_lock->fl_end = last_byte_offset(locku->lu_offset,
7678                                                 locku->lu_length);
7679         nfs4_transform_lock_offset(file_lock);
7680
7681         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
7682         if (err) {
7683                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
7684                 goto out_nfserr;
7685         }
7686         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
7687 put_file:
7688         nfsd_file_put(nf);
7689 put_stateid:
7690         mutex_unlock(&stp->st_mutex);
7691         nfs4_put_stid(&stp->st_stid);
7692 out:
7693         nfsd4_bump_seqid(cstate, status);
7694         if (file_lock)
7695                 locks_free_lock(file_lock);
7696         return status;
7697
7698 out_nfserr:
7699         status = nfserrno(err);
7700         goto put_file;
7701 }
7702
7703 /*
7704  * returns
7705  *      true:  locks held by lockowner
7706  *      false: no locks held by lockowner
7707  */
7708 static bool
7709 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
7710 {
7711         struct file_lock *fl;
7712         int status = false;
7713         struct nfsd_file *nf = find_any_file(fp);
7714         struct inode *inode;
7715         struct file_lock_context *flctx;
7716
7717         if (!nf) {
7718                 /* Any valid lock stateid should have some sort of access */
7719                 WARN_ON_ONCE(1);
7720                 return status;
7721         }
7722
7723         inode = locks_inode(nf->nf_file);
7724         flctx = inode->i_flctx;
7725
7726         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
7727                 spin_lock(&flctx->flc_lock);
7728                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
7729                         if (fl->fl_owner == (fl_owner_t)lowner) {
7730                                 status = true;
7731                                 break;
7732                         }
7733                 }
7734                 spin_unlock(&flctx->flc_lock);
7735         }
7736         nfsd_file_put(nf);
7737         return status;
7738 }
7739
7740 /**
7741  * nfsd4_release_lockowner - process NFSv4.0 RELEASE_LOCKOWNER operations
7742  * @rqstp: RPC transaction
7743  * @cstate: NFSv4 COMPOUND state
7744  * @u: RELEASE_LOCKOWNER arguments
7745  *
7746  * The lockowner's so_count is bumped when a lock record is added
7747  * or when copying a conflicting lock. The latter case is brief,
7748  * but can lead to fleeting false positives when looking for
7749  * locks-in-use.
7750  *
7751  * Return values:
7752  *   %nfs_ok: lockowner released or not found
7753  *   %nfserr_locks_held: lockowner still in use
7754  *   %nfserr_stale_clientid: clientid no longer active
7755  *   %nfserr_expired: clientid not recognized
7756  */
7757 __be32
7758 nfsd4_release_lockowner(struct svc_rqst *rqstp,
7759                         struct nfsd4_compound_state *cstate,
7760                         union nfsd4_op_u *u)
7761 {
7762         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
7763         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7764         clientid_t *clid = &rlockowner->rl_clientid;
7765         struct nfs4_ol_stateid *stp;
7766         struct nfs4_lockowner *lo;
7767         struct nfs4_client *clp;
7768         LIST_HEAD(reaplist);
7769         __be32 status;
7770
7771         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
7772                 clid->cl_boot, clid->cl_id);
7773
7774         status = set_client(clid, cstate, nn);
7775         if (status)
7776                 return status;
7777         clp = cstate->clp;
7778
7779         spin_lock(&clp->cl_lock);
7780         lo = find_lockowner_str_locked(clp, &rlockowner->rl_owner);
7781         if (!lo) {
7782                 spin_unlock(&clp->cl_lock);
7783                 return nfs_ok;
7784         }
7785         if (atomic_read(&lo->lo_owner.so_count) != 2) {
7786                 spin_unlock(&clp->cl_lock);
7787                 nfs4_put_stateowner(&lo->lo_owner);
7788                 return nfserr_locks_held;
7789         }
7790         unhash_lockowner_locked(lo);
7791         while (!list_empty(&lo->lo_owner.so_stateids)) {
7792                 stp = list_first_entry(&lo->lo_owner.so_stateids,
7793                                        struct nfs4_ol_stateid,
7794                                        st_perstateowner);
7795                 WARN_ON(!unhash_lock_stateid(stp));
7796                 put_ol_stateid_locked(stp, &reaplist);
7797         }
7798         spin_unlock(&clp->cl_lock);
7799
7800         free_ol_stateid_reaplist(&reaplist);
7801         remove_blocked_locks(lo);
7802         nfs4_put_stateowner(&lo->lo_owner);
7803         return nfs_ok;
7804 }
7805
7806 static inline struct nfs4_client_reclaim *
7807 alloc_reclaim(void)
7808 {
7809         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
7810 }
7811
7812 bool
7813 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
7814 {
7815         struct nfs4_client_reclaim *crp;
7816
7817         crp = nfsd4_find_reclaim_client(name, nn);
7818         return (crp && crp->cr_clp);
7819 }
7820
7821 /*
7822  * failure => all reset bets are off, nfserr_no_grace...
7823  *
7824  * The caller is responsible for freeing name.data if NULL is returned (it
7825  * will be freed in nfs4_remove_reclaim_record in the normal case).
7826  */
7827 struct nfs4_client_reclaim *
7828 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
7829                 struct nfsd_net *nn)
7830 {
7831         unsigned int strhashval;
7832         struct nfs4_client_reclaim *crp;
7833
7834         crp = alloc_reclaim();
7835         if (crp) {
7836                 strhashval = clientstr_hashval(name);
7837                 INIT_LIST_HEAD(&crp->cr_strhash);
7838                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
7839                 crp->cr_name.data = name.data;
7840                 crp->cr_name.len = name.len;
7841                 crp->cr_princhash.data = princhash.data;
7842                 crp->cr_princhash.len = princhash.len;
7843                 crp->cr_clp = NULL;
7844                 nn->reclaim_str_hashtbl_size++;
7845         }
7846         return crp;
7847 }
7848
7849 void
7850 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
7851 {
7852         list_del(&crp->cr_strhash);
7853         kfree(crp->cr_name.data);
7854         kfree(crp->cr_princhash.data);
7855         kfree(crp);
7856         nn->reclaim_str_hashtbl_size--;
7857 }
7858
7859 void
7860 nfs4_release_reclaim(struct nfsd_net *nn)
7861 {
7862         struct nfs4_client_reclaim *crp = NULL;
7863         int i;
7864
7865         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7866                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
7867                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
7868                                         struct nfs4_client_reclaim, cr_strhash);
7869                         nfs4_remove_reclaim_record(crp, nn);
7870                 }
7871         }
7872         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
7873 }
7874
7875 /*
7876  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
7877 struct nfs4_client_reclaim *
7878 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
7879 {
7880         unsigned int strhashval;
7881         struct nfs4_client_reclaim *crp = NULL;
7882
7883         strhashval = clientstr_hashval(name);
7884         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
7885                 if (compare_blob(&crp->cr_name, &name) == 0) {
7886                         return crp;
7887                 }
7888         }
7889         return NULL;
7890 }
7891
7892 __be32
7893 nfs4_check_open_reclaim(struct nfs4_client *clp)
7894 {
7895         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
7896                 return nfserr_no_grace;
7897
7898         if (nfsd4_client_record_check(clp))
7899                 return nfserr_reclaim_bad;
7900
7901         return nfs_ok;
7902 }
7903
7904 /*
7905  * Since the lifetime of a delegation isn't limited to that of an open, a
7906  * client may quite reasonably hang on to a delegation as long as it has
7907  * the inode cached.  This becomes an obvious problem the first time a
7908  * client's inode cache approaches the size of the server's total memory.
7909  *
7910  * For now we avoid this problem by imposing a hard limit on the number
7911  * of delegations, which varies according to the server's memory size.
7912  */
7913 static void
7914 set_max_delegations(void)
7915 {
7916         /*
7917          * Allow at most 4 delegations per megabyte of RAM.  Quick
7918          * estimates suggest that in the worst case (where every delegation
7919          * is for a different inode), a delegation could take about 1.5K,
7920          * giving a worst case usage of about 6% of memory.
7921          */
7922         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7923 }
7924
7925 static int nfs4_state_create_net(struct net *net)
7926 {
7927         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7928         int i;
7929
7930         nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7931                                             sizeof(struct list_head),
7932                                             GFP_KERNEL);
7933         if (!nn->conf_id_hashtbl)
7934                 goto err;
7935         nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7936                                               sizeof(struct list_head),
7937                                               GFP_KERNEL);
7938         if (!nn->unconf_id_hashtbl)
7939                 goto err_unconf_id;
7940         nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7941                                               sizeof(struct list_head),
7942                                               GFP_KERNEL);
7943         if (!nn->sessionid_hashtbl)
7944                 goto err_sessionid;
7945
7946         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7947                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7948                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7949         }
7950         for (i = 0; i < SESSION_HASH_SIZE; i++)
7951                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7952         nn->conf_name_tree = RB_ROOT;
7953         nn->unconf_name_tree = RB_ROOT;
7954         nn->boot_time = ktime_get_real_seconds();
7955         nn->grace_ended = false;
7956         nn->nfsd4_manager.block_opens = true;
7957         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7958         INIT_LIST_HEAD(&nn->client_lru);
7959         INIT_LIST_HEAD(&nn->close_lru);
7960         INIT_LIST_HEAD(&nn->del_recall_lru);
7961         spin_lock_init(&nn->client_lock);
7962         spin_lock_init(&nn->s2s_cp_lock);
7963         idr_init(&nn->s2s_cp_stateids);
7964
7965         spin_lock_init(&nn->blocked_locks_lock);
7966         INIT_LIST_HEAD(&nn->blocked_locks_lru);
7967
7968         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7969         INIT_DELAYED_WORK(&nn->nfsd_shrinker_work, courtesy_client_reaper);
7970         get_net(net);
7971
7972         return 0;
7973
7974 err_sessionid:
7975         kfree(nn->unconf_id_hashtbl);
7976 err_unconf_id:
7977         kfree(nn->conf_id_hashtbl);
7978 err:
7979         return -ENOMEM;
7980 }
7981
7982 static void
7983 nfs4_state_destroy_net(struct net *net)
7984 {
7985         int i;
7986         struct nfs4_client *clp = NULL;
7987         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7988
7989         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7990                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7991                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7992                         destroy_client(clp);
7993                 }
7994         }
7995
7996         WARN_ON(!list_empty(&nn->blocked_locks_lru));
7997
7998         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7999                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
8000                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8001                         destroy_client(clp);
8002                 }
8003         }
8004
8005         kfree(nn->sessionid_hashtbl);
8006         kfree(nn->unconf_id_hashtbl);
8007         kfree(nn->conf_id_hashtbl);
8008         put_net(net);
8009 }
8010
8011 int
8012 nfs4_state_start_net(struct net *net)
8013 {
8014         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8015         int ret;
8016
8017         ret = nfs4_state_create_net(net);
8018         if (ret)
8019                 return ret;
8020         locks_start_grace(net, &nn->nfsd4_manager);
8021         nfsd4_client_tracking_init(net);
8022         if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
8023                 goto skip_grace;
8024         printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
8025                nn->nfsd4_grace, net->ns.inum);
8026         trace_nfsd_grace_start(nn);
8027         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
8028         return 0;
8029
8030 skip_grace:
8031         printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
8032                         net->ns.inum);
8033         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
8034         nfsd4_end_grace(nn);
8035         return 0;
8036 }
8037
8038 /* initialization to perform when the nfsd service is started: */
8039
8040 int
8041 nfs4_state_start(void)
8042 {
8043         int ret;
8044
8045         ret = nfsd4_create_callback_queue();
8046         if (ret)
8047                 return ret;
8048
8049         set_max_delegations();
8050         return 0;
8051 }
8052
8053 void
8054 nfs4_state_shutdown_net(struct net *net)
8055 {
8056         struct nfs4_delegation *dp = NULL;
8057         struct list_head *pos, *next, reaplist;
8058         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8059
8060         cancel_delayed_work_sync(&nn->laundromat_work);
8061         locks_end_grace(&nn->nfsd4_manager);
8062
8063         INIT_LIST_HEAD(&reaplist);
8064         spin_lock(&state_lock);
8065         list_for_each_safe(pos, next, &nn->del_recall_lru) {
8066                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8067                 WARN_ON(!unhash_delegation_locked(dp));
8068                 list_add(&dp->dl_recall_lru, &reaplist);
8069         }
8070         spin_unlock(&state_lock);
8071         list_for_each_safe(pos, next, &reaplist) {
8072                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8073                 list_del_init(&dp->dl_recall_lru);
8074                 destroy_unhashed_deleg(dp);
8075         }
8076
8077         nfsd4_client_tracking_exit(net);
8078         nfs4_state_destroy_net(net);
8079 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
8080         nfsd4_ssc_shutdown_umount(nn);
8081 #endif
8082 }
8083
8084 void
8085 nfs4_state_shutdown(void)
8086 {
8087         nfsd4_destroy_callback_queue();
8088 }
8089
8090 static void
8091 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8092 {
8093         if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
8094             CURRENT_STATEID(stateid))
8095                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
8096 }
8097
8098 static void
8099 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8100 {
8101         if (cstate->minorversion) {
8102                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
8103                 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8104         }
8105 }
8106
8107 void
8108 clear_current_stateid(struct nfsd4_compound_state *cstate)
8109 {
8110         CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8111 }
8112
8113 /*
8114  * functions to set current state id
8115  */
8116 void
8117 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
8118                 union nfsd4_op_u *u)
8119 {
8120         put_stateid(cstate, &u->open_downgrade.od_stateid);
8121 }
8122
8123 void
8124 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
8125                 union nfsd4_op_u *u)
8126 {
8127         put_stateid(cstate, &u->open.op_stateid);
8128 }
8129
8130 void
8131 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
8132                 union nfsd4_op_u *u)
8133 {
8134         put_stateid(cstate, &u->close.cl_stateid);
8135 }
8136
8137 void
8138 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
8139                 union nfsd4_op_u *u)
8140 {
8141         put_stateid(cstate, &u->lock.lk_resp_stateid);
8142 }
8143
8144 /*
8145  * functions to consume current state id
8146  */
8147
8148 void
8149 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
8150                 union nfsd4_op_u *u)
8151 {
8152         get_stateid(cstate, &u->open_downgrade.od_stateid);
8153 }
8154
8155 void
8156 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
8157                 union nfsd4_op_u *u)
8158 {
8159         get_stateid(cstate, &u->delegreturn.dr_stateid);
8160 }
8161
8162 void
8163 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
8164                 union nfsd4_op_u *u)
8165 {
8166         get_stateid(cstate, &u->free_stateid.fr_stateid);
8167 }
8168
8169 void
8170 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
8171                 union nfsd4_op_u *u)
8172 {
8173         get_stateid(cstate, &u->setattr.sa_stateid);
8174 }
8175
8176 void
8177 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
8178                 union nfsd4_op_u *u)
8179 {
8180         get_stateid(cstate, &u->close.cl_stateid);
8181 }
8182
8183 void
8184 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
8185                 union nfsd4_op_u *u)
8186 {
8187         get_stateid(cstate, &u->locku.lu_stateid);
8188 }
8189
8190 void
8191 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
8192                 union nfsd4_op_u *u)
8193 {
8194         get_stateid(cstate, &u->read.rd_stateid);
8195 }
8196
8197 void
8198 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
8199                 union nfsd4_op_u *u)
8200 {
8201         get_stateid(cstate, &u->write.wr_stateid);
8202 }