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 "xdr4.h"
  46 #include "xdr4cb.h"
  47 #include "vfs.h"
  48 #include "current_stateid.h"
  49
  50 #include "netns.h"
  51 #include "pnfs.h"
  52
  53 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  54
  55 #define all_ones {{~0,~0},~0}
  56 static const stateid_t one_stateid = {
  57         .si_generation = ~0,
  58         .si_opaque = all_ones,
  59 };
  60 static const stateid_t zero_stateid = {
  61         /* all fields zero */
  62 };
  63 static const stateid_t currentstateid = {
  64         .si_generation = 1,
  65 };
  66
  67 static u64 current_sessionid = 1;
  68
  69 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  70 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  71 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  72
  73 /* forward declarations */
  74 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  75 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  76
  77 /* Locking: */
  78
  79 /*
  80  * Currently used for the del_recall_lru and file hash table.  In an
  81  * effort to decrease the scope of the client_mutex, this spinlock may
  82  * eventually cover more:
  83  */
  84 static DEFINE_SPINLOCK(state_lock);
  85
  86 /*
  87  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
  88  * the refcount on the open stateid to drop.
  89  */
  90 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
  91
  92 static struct kmem_cache *openowner_slab;
  93 static struct kmem_cache *lockowner_slab;
  94 static struct kmem_cache *file_slab;
  95 static struct kmem_cache *stateid_slab;
  96 static struct kmem_cache *deleg_slab;
  97 static struct kmem_cache *odstate_slab;
  98
  99 static void free_session(struct nfsd4_session *);
 100
 101 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 102 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
 103
 104 static bool is_session_dead(struct nfsd4_session *ses)
 105 {
 106         return ses->se_flags & NFS4_SESSION_DEAD;
 107 }
 108
 109 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 110 {
 111         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 112                 return nfserr_jukebox;
 113         ses->se_flags |= NFS4_SESSION_DEAD;
 114         return nfs_ok;
 115 }
 116
 117 static bool is_client_expired(struct nfs4_client *clp)
 118 {
 119         return clp->cl_time == 0;
 120 }
 121
 122 static __be32 get_client_locked(struct nfs4_client *clp)
 123 {
 124         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 125
 126         lockdep_assert_held(&nn->client_lock);
 127
 128         if (is_client_expired(clp))
 129                 return nfserr_expired;
 130         atomic_inc(&clp->cl_refcount);
 131         return nfs_ok;
 132 }
 133
 134 /* must be called under the client_lock */
 135 static inline void
 136 renew_client_locked(struct nfs4_client *clp)
 137 {
 138         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 139
 140         if (is_client_expired(clp)) {
 141                 WARN_ON(1);
 142                 printk("%s: client (clientid %08x/%08x) already expired\n",
 143                         __func__,
 144                         clp->cl_clientid.cl_boot,
 145                         clp->cl_clientid.cl_id);
 146                 return;
 147         }
 148
 149         dprintk("renewing client (clientid %08x/%08x)\n",
 150                         clp->cl_clientid.cl_boot,
 151                         clp->cl_clientid.cl_id);
 152         list_move_tail(&clp->cl_lru, &nn->client_lru);
 153         clp->cl_time = get_seconds();
 154 }
 155
 156 static void put_client_renew_locked(struct nfs4_client *clp)
 157 {
 158         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 159
 160         lockdep_assert_held(&nn->client_lock);
 161
 162         if (!atomic_dec_and_test(&clp->cl_refcount))
 163                 return;
 164         if (!is_client_expired(clp))
 165                 renew_client_locked(clp);
 166 }
 167
 168 static void put_client_renew(struct nfs4_client *clp)
 169 {
 170         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 171
 172         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
 173                 return;
 174         if (!is_client_expired(clp))
 175                 renew_client_locked(clp);
 176         spin_unlock(&nn->client_lock);
 177 }
 178
 179 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 180 {
 181         __be32 status;
 182
 183         if (is_session_dead(ses))
 184                 return nfserr_badsession;
 185         status = get_client_locked(ses->se_client);
 186         if (status)
 187                 return status;
 188         atomic_inc(&ses->se_ref);
 189         return nfs_ok;
 190 }
 191
 192 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 193 {
 194         struct nfs4_client *clp = ses->se_client;
 195         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 196
 197         lockdep_assert_held(&nn->client_lock);
 198
 199         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 200                 free_session(ses);
 201         put_client_renew_locked(clp);
 202 }
 203
 204 static void nfsd4_put_session(struct nfsd4_session *ses)
 205 {
 206         struct nfs4_client *clp = ses->se_client;
 207         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 208
 209         spin_lock(&nn->client_lock);
 210         nfsd4_put_session_locked(ses);
 211         spin_unlock(&nn->client_lock);
 212 }
 213
 214 static struct nfsd4_blocked_lock *
 215 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 216                         struct nfsd_net *nn)
 217 {
 218         struct nfsd4_blocked_lock *cur, *found = NULL;
 219
 220         spin_lock(&nn->client_lock);
 221         list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
 222                 if (fh_match(fh, &cur->nbl_fh)) {
 223                         list_del_init(&cur->nbl_list);
 224                         list_del_init(&cur->nbl_lru);
 225                         found = cur;
 226                         break;
 227                 }
 228         }
 229         spin_unlock(&nn->client_lock);
 230         if (found)
 231                 posix_unblock_lock(&found->nbl_lock);
 232         return found;
 233 }
 234
 235 static struct nfsd4_blocked_lock *
 236 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 237                         struct nfsd_net *nn)
 238 {
 239         struct nfsd4_blocked_lock *nbl;
 240
 241         nbl = find_blocked_lock(lo, fh, nn);
 242         if (!nbl) {
 243                 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
 244                 if (nbl) {
 245                         fh_copy_shallow(&nbl->nbl_fh, fh);
 246                         locks_init_lock(&nbl->nbl_lock);
 247                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
 248                                         &nfsd4_cb_notify_lock_ops,
 249                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
 250                 }
 251         }
 252         return nbl;
 253 }
 254
 255 static void
 256 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
 257 {
 258         locks_release_private(&nbl->nbl_lock);
 259         kfree(nbl);
 260 }
 261
 262 static int
 263 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
 264 {
 265         /*
 266          * Since this is just an optimization, we don't try very hard if it
 267          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
 268          * just quit trying on anything else.
 269          */
 270         switch (task->tk_status) {
 271         case -NFS4ERR_DELAY:
 272                 rpc_delay(task, 1 * HZ);
 273                 return 0;
 274         default:
 275                 return 1;
 276         }
 277 }
 278
 279 static void
 280 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
 281 {
 282         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 283                                                 struct nfsd4_blocked_lock, nbl_cb);
 284
 285         free_blocked_lock(nbl);
 286 }
 287
 288 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
 289         .done           = nfsd4_cb_notify_lock_done,
 290         .release        = nfsd4_cb_notify_lock_release,
 291 };
 292
 293 static inline struct nfs4_stateowner *
 294 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 295 {
 296         atomic_inc(&sop->so_count);
 297         return sop;
 298 }
 299
 300 static int
 301 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 302 {
 303         return (sop->so_owner.len == owner->len) &&
 304                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 305 }
 306
 307 static struct nfs4_openowner *
 308 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 309                         struct nfs4_client *clp)
 310 {
 311         struct nfs4_stateowner *so;
 312
 313         lockdep_assert_held(&clp->cl_lock);
 314
 315         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 316                             so_strhash) {
 317                 if (!so->so_is_open_owner)
 318                         continue;
 319                 if (same_owner_str(so, &open->op_owner))
 320                         return openowner(nfs4_get_stateowner(so));
 321         }
 322         return NULL;
 323 }
 324
 325 static struct nfs4_openowner *
 326 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 327                         struct nfs4_client *clp)
 328 {
 329         struct nfs4_openowner *oo;
 330
 331         spin_lock(&clp->cl_lock);
 332         oo = find_openstateowner_str_locked(hashval, open, clp);
 333         spin_unlock(&clp->cl_lock);
 334         return oo;
 335 }
 336
 337 static inline u32
 338 opaque_hashval(const void *ptr, int nbytes)
 339 {
 340         unsigned char *cptr = (unsigned char *) ptr;
 341
 342         u32 x = 0;
 343         while (nbytes--) {
 344                 x *= 37;
 345                 x += *cptr++;
 346         }
 347         return x;
 348 }
 349
 350 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 351 {
 352         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 353
 354         kmem_cache_free(file_slab, fp);
 355 }
 356
 357 void
 358 put_nfs4_file(struct nfs4_file *fi)
 359 {
 360         might_lock(&state_lock);
 361
 362         if (atomic_dec_and_lock(&fi->fi_ref, &state_lock)) {
 363                 hlist_del_rcu(&fi->fi_hash);
 364                 spin_unlock(&state_lock);
 365                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 366                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 367                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 368         }
 369 }
 370
 371 static struct file *
 372 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 373 {
 374         if (f->fi_fds[oflag])
 375                 return get_file(f->fi_fds[oflag]);
 376         return NULL;
 377 }
 378
 379 static struct file *
 380 find_writeable_file_locked(struct nfs4_file *f)
 381 {
 382         struct file *ret;
 383
 384         lockdep_assert_held(&f->fi_lock);
 385
 386         ret = __nfs4_get_fd(f, O_WRONLY);
 387         if (!ret)
 388                 ret = __nfs4_get_fd(f, O_RDWR);
 389         return ret;
 390 }
 391
 392 static struct file *
 393 find_writeable_file(struct nfs4_file *f)
 394 {
 395         struct file *ret;
 396
 397         spin_lock(&f->fi_lock);
 398         ret = find_writeable_file_locked(f);
 399         spin_unlock(&f->fi_lock);
 400
 401         return ret;
 402 }
 403
 404 static struct file *find_readable_file_locked(struct nfs4_file *f)
 405 {
 406         struct file *ret;
 407
 408         lockdep_assert_held(&f->fi_lock);
 409
 410         ret = __nfs4_get_fd(f, O_RDONLY);
 411         if (!ret)
 412                 ret = __nfs4_get_fd(f, O_RDWR);
 413         return ret;
 414 }
 415
 416 static struct file *
 417 find_readable_file(struct nfs4_file *f)
 418 {
 419         struct file *ret;
 420
 421         spin_lock(&f->fi_lock);
 422         ret = find_readable_file_locked(f);
 423         spin_unlock(&f->fi_lock);
 424
 425         return ret;
 426 }
 427
 428 struct file *
 429 find_any_file(struct nfs4_file *f)
 430 {
 431         struct file *ret;
 432
 433         spin_lock(&f->fi_lock);
 434         ret = __nfs4_get_fd(f, O_RDWR);
 435         if (!ret) {
 436                 ret = __nfs4_get_fd(f, O_WRONLY);
 437                 if (!ret)
 438                         ret = __nfs4_get_fd(f, O_RDONLY);
 439         }
 440         spin_unlock(&f->fi_lock);
 441         return ret;
 442 }
 443
 444 static atomic_long_t num_delegations;
 445 unsigned long max_delegations;
 446
 447 /*
 448  * Open owner state (share locks)
 449  */
 450
 451 /* hash tables for lock and open owners */
 452 #define OWNER_HASH_BITS              8
 453 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 454 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 455
 456 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 457 {
 458         unsigned int ret;
 459
 460         ret = opaque_hashval(ownername->data, ownername->len);
 461         return ret & OWNER_HASH_MASK;
 462 }
 463
 464 /* hash table for nfs4_file */
 465 #define FILE_HASH_BITS                   8
 466 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 467
 468 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 469 {
 470         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 471 }
 472
 473 static unsigned int file_hashval(struct knfsd_fh *fh)
 474 {
 475         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 476 }
 477
 478 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 479
 480 static void
 481 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 482 {
 483         lockdep_assert_held(&fp->fi_lock);
 484
 485         if (access & NFS4_SHARE_ACCESS_WRITE)
 486                 atomic_inc(&fp->fi_access[O_WRONLY]);
 487         if (access & NFS4_SHARE_ACCESS_READ)
 488                 atomic_inc(&fp->fi_access[O_RDONLY]);
 489 }
 490
 491 static __be32
 492 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 493 {
 494         lockdep_assert_held(&fp->fi_lock);
 495
 496         /* Does this access mode make sense? */
 497         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 498                 return nfserr_inval;
 499
 500         /* Does it conflict with a deny mode already set? */
 501         if ((access & fp->fi_share_deny) != 0)
 502                 return nfserr_share_denied;
 503
 504         __nfs4_file_get_access(fp, access);
 505         return nfs_ok;
 506 }
 507
 508 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 509 {
 510         /* Common case is that there is no deny mode. */
 511         if (deny) {
 512                 /* Does this deny mode make sense? */
 513                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 514                         return nfserr_inval;
 515
 516                 if ((deny & NFS4_SHARE_DENY_READ) &&
 517                     atomic_read(&fp->fi_access[O_RDONLY]))
 518                         return nfserr_share_denied;
 519
 520                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 521                     atomic_read(&fp->fi_access[O_WRONLY]))
 522                         return nfserr_share_denied;
 523         }
 524         return nfs_ok;
 525 }
 526
 527 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 528 {
 529         might_lock(&fp->fi_lock);
 530
 531         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 532                 struct file *f1 = NULL;
 533                 struct file *f2 = NULL;
 534
 535                 swap(f1, fp->fi_fds[oflag]);
 536                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 537                         swap(f2, fp->fi_fds[O_RDWR]);
 538                 spin_unlock(&fp->fi_lock);
 539                 if (f1)
 540                         fput(f1);
 541                 if (f2)
 542                         fput(f2);
 543         }
 544 }
 545
 546 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 547 {
 548         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 549
 550         if (access & NFS4_SHARE_ACCESS_WRITE)
 551                 __nfs4_file_put_access(fp, O_WRONLY);
 552         if (access & NFS4_SHARE_ACCESS_READ)
 553                 __nfs4_file_put_access(fp, O_RDONLY);
 554 }
 555
 556 /*
 557  * Allocate a new open/delegation state counter. This is needed for
 558  * pNFS for proper return on close semantics.
 559  *
 560  * Note that we only allocate it for pNFS-enabled exports, otherwise
 561  * all pointers to struct nfs4_clnt_odstate are always NULL.
 562  */
 563 static struct nfs4_clnt_odstate *
 564 alloc_clnt_odstate(struct nfs4_client *clp)
 565 {
 566         struct nfs4_clnt_odstate *co;
 567
 568         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 569         if (co) {
 570                 co->co_client = clp;
 571                 atomic_set(&co->co_odcount, 1);
 572         }
 573         return co;
 574 }
 575
 576 static void
 577 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 578 {
 579         struct nfs4_file *fp = co->co_file;
 580
 581         lockdep_assert_held(&fp->fi_lock);
 582         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 583 }
 584
 585 static inline void
 586 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 587 {
 588         if (co)
 589                 atomic_inc(&co->co_odcount);
 590 }
 591
 592 static void
 593 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 594 {
 595         struct nfs4_file *fp;
 596
 597         if (!co)
 598                 return;
 599
 600         fp = co->co_file;
 601         if (atomic_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 602                 list_del(&co->co_perfile);
 603                 spin_unlock(&fp->fi_lock);
 604
 605                 nfsd4_return_all_file_layouts(co->co_client, fp);
 606                 kmem_cache_free(odstate_slab, co);
 607         }
 608 }
 609
 610 static struct nfs4_clnt_odstate *
 611 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 612 {
 613         struct nfs4_clnt_odstate *co;
 614         struct nfs4_client *cl;
 615
 616         if (!new)
 617                 return NULL;
 618
 619         cl = new->co_client;
 620
 621         spin_lock(&fp->fi_lock);
 622         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 623                 if (co->co_client == cl) {
 624                         get_clnt_odstate(co);
 625                         goto out;
 626                 }
 627         }
 628         co = new;
 629         co->co_file = fp;
 630         hash_clnt_odstate_locked(new);
 631 out:
 632         spin_unlock(&fp->fi_lock);
 633         return co;
 634 }
 635
 636 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl,
 637                                          struct kmem_cache *slab)
 638 {
 639         struct nfs4_stid *stid;
 640         int new_id;
 641
 642         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 643         if (!stid)
 644                 return NULL;
 645
 646         idr_preload(GFP_KERNEL);
 647         spin_lock(&cl->cl_lock);
 648         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
 649         spin_unlock(&cl->cl_lock);
 650         idr_preload_end();
 651         if (new_id < 0)
 652                 goto out_free;
 653         stid->sc_client = cl;
 654         stid->sc_stateid.si_opaque.so_id = new_id;
 655         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 656         /* Will be incremented before return to client: */
 657         atomic_set(&stid->sc_count, 1);
 658         spin_lock_init(&stid->sc_lock);
 659
 660         /*
 661          * It shouldn't be a problem to reuse an opaque stateid value.
 662          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 663          * example, a stray write retransmission could be accepted by
 664          * the server when it should have been rejected.  Therefore,
 665          * adopt a trick from the sctp code to attempt to maximize the
 666          * amount of time until an id is reused, by ensuring they always
 667          * "increase" (mod INT_MAX):
 668          */
 669         return stid;
 670 out_free:
 671         kmem_cache_free(slab, stid);
 672         return NULL;
 673 }
 674
 675 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 676 {
 677         struct nfs4_stid *stid;
 678         struct nfs4_ol_stateid *stp;
 679
 680         stid = nfs4_alloc_stid(clp, stateid_slab);
 681         if (!stid)
 682                 return NULL;
 683
 684         stp = openlockstateid(stid);
 685         stp->st_stid.sc_free = nfs4_free_ol_stateid;
 686         return stp;
 687 }
 688
 689 static void nfs4_free_deleg(struct nfs4_stid *stid)
 690 {
 691         kmem_cache_free(deleg_slab, stid);
 692         atomic_long_dec(&num_delegations);
 693 }
 694
 695 /*
 696  * When we recall a delegation, we should be careful not to hand it
 697  * out again straight away.
 698  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 699  * in which the filehandles of recalled delegations are "stored".
 700  * If a filehandle appear in either filter, a delegation is blocked.
 701  * When a delegation is recalled, the filehandle is stored in the "new"
 702  * filter.
 703  * Every 30 seconds we swap the filters and clear the "new" one,
 704  * unless both are empty of course.
 705  *
 706  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 707  * low 3 bytes as hash-table indices.
 708  *
 709  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 710  * is used to manage concurrent access.  Testing does not need the lock
 711  * except when swapping the two filters.
 712  */
 713 static DEFINE_SPINLOCK(blocked_delegations_lock);
 714 static struct bloom_pair {
 715         int     entries, old_entries;
 716         time_t  swap_time;
 717         int     new; /* index into 'set' */
 718         DECLARE_BITMAP(set[2], 256);
 719 } blocked_delegations;
 720
 721 static int delegation_blocked(struct knfsd_fh *fh)
 722 {
 723         u32 hash;
 724         struct bloom_pair *bd = &blocked_delegations;
 725
 726         if (bd->entries == 0)
 727                 return 0;
 728         if (seconds_since_boot() - bd->swap_time > 30) {
 729                 spin_lock(&blocked_delegations_lock);
 730                 if (seconds_since_boot() - bd->swap_time > 30) {
 731                         bd->entries -= bd->old_entries;
 732                         bd->old_entries = bd->entries;
 733                         memset(bd->set[bd->new], 0,
 734                                sizeof(bd->set[0]));
 735                         bd->new = 1-bd->new;
 736                         bd->swap_time = seconds_since_boot();
 737                 }
 738                 spin_unlock(&blocked_delegations_lock);
 739         }
 740         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 741         if (test_bit(hash&255, bd->set[0]) &&
 742             test_bit((hash>>8)&255, bd->set[0]) &&
 743             test_bit((hash>>16)&255, bd->set[0]))
 744                 return 1;
 745
 746         if (test_bit(hash&255, bd->set[1]) &&
 747             test_bit((hash>>8)&255, bd->set[1]) &&
 748             test_bit((hash>>16)&255, bd->set[1]))
 749                 return 1;
 750
 751         return 0;
 752 }
 753
 754 static void block_delegations(struct knfsd_fh *fh)
 755 {
 756         u32 hash;
 757         struct bloom_pair *bd = &blocked_delegations;
 758
 759         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 760
 761         spin_lock(&blocked_delegations_lock);
 762         __set_bit(hash&255, bd->set[bd->new]);
 763         __set_bit((hash>>8)&255, bd->set[bd->new]);
 764         __set_bit((hash>>16)&255, bd->set[bd->new]);
 765         if (bd->entries == 0)
 766                 bd->swap_time = seconds_since_boot();
 767         bd->entries += 1;
 768         spin_unlock(&blocked_delegations_lock);
 769 }
 770
 771 static struct nfs4_delegation *
 772 alloc_init_deleg(struct nfs4_client *clp, struct svc_fh *current_fh,
 773                  struct nfs4_clnt_odstate *odstate)
 774 {
 775         struct nfs4_delegation *dp;
 776         long n;
 777
 778         dprintk("NFSD alloc_init_deleg\n");
 779         n = atomic_long_inc_return(&num_delegations);
 780         if (n < 0 || n > max_delegations)
 781                 goto out_dec;
 782         if (delegation_blocked(&current_fh->fh_handle))
 783                 goto out_dec;
 784         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
 785         if (dp == NULL)
 786                 goto out_dec;
 787
 788         dp->dl_stid.sc_free = nfs4_free_deleg;
 789         /*
 790          * delegation seqid's are never incremented.  The 4.1 special
 791          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 792          * 0 anyway just for consistency and use 1:
 793          */
 794         dp->dl_stid.sc_stateid.si_generation = 1;
 795         INIT_LIST_HEAD(&dp->dl_perfile);
 796         INIT_LIST_HEAD(&dp->dl_perclnt);
 797         INIT_LIST_HEAD(&dp->dl_recall_lru);
 798         dp->dl_clnt_odstate = odstate;
 799         get_clnt_odstate(odstate);
 800         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 801         dp->dl_retries = 1;
 802         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 803                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 804         return dp;
 805 out_dec:
 806         atomic_long_dec(&num_delegations);
 807         return NULL;
 808 }
 809
 810 void
 811 nfs4_put_stid(struct nfs4_stid *s)
 812 {
 813         struct nfs4_file *fp = s->sc_file;
 814         struct nfs4_client *clp = s->sc_client;
 815
 816         might_lock(&clp->cl_lock);
 817
 818         if (!atomic_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 819                 wake_up_all(&close_wq);
 820                 return;
 821         }
 822         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 823         spin_unlock(&clp->cl_lock);
 824         s->sc_free(s);
 825         if (fp)
 826                 put_nfs4_file(fp);
 827 }
 828
 829 void
 830 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
 831 {
 832         stateid_t *src = &stid->sc_stateid;
 833
 834         spin_lock(&stid->sc_lock);
 835         if (unlikely(++src->si_generation == 0))
 836                 src->si_generation = 1;
 837         memcpy(dst, src, sizeof(*dst));
 838         spin_unlock(&stid->sc_lock);
 839 }
 840
 841 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 842 {
 843         struct file *filp = NULL;
 844
 845         spin_lock(&fp->fi_lock);
 846         if (fp->fi_deleg_file && --fp->fi_delegees == 0)
 847                 swap(filp, fp->fi_deleg_file);
 848         spin_unlock(&fp->fi_lock);
 849
 850         if (filp) {
 851                 vfs_setlease(filp, F_UNLCK, NULL, (void **)&fp);
 852                 fput(filp);
 853         }
 854 }
 855
 856 void nfs4_unhash_stid(struct nfs4_stid *s)
 857 {
 858         s->sc_type = 0;
 859 }
 860
 861 /**
 862  * nfs4_get_existing_delegation - Discover if this delegation already exists
 863  * @clp:     a pointer to the nfs4_client we're granting a delegation to
 864  * @fp:      a pointer to the nfs4_file we're granting a delegation on
 865  *
 866  * Return:
 867  *      On success: NULL if an existing delegation was not found.
 868  *
 869  *      On error: -EAGAIN if one was previously granted to this nfs4_client
 870  *                 for this nfs4_file.
 871  *
 872  */
 873
 874 static int
 875 nfs4_get_existing_delegation(struct nfs4_client *clp, struct nfs4_file *fp)
 876 {
 877         struct nfs4_delegation *searchdp = NULL;
 878         struct nfs4_client *searchclp = NULL;
 879
 880         lockdep_assert_held(&state_lock);
 881         lockdep_assert_held(&fp->fi_lock);
 882
 883         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
 884                 searchclp = searchdp->dl_stid.sc_client;
 885                 if (clp == searchclp) {
 886                         return -EAGAIN;
 887                 }
 888         }
 889         return 0;
 890 }
 891
 892 /**
 893  * hash_delegation_locked - Add a delegation to the appropriate lists
 894  * @dp:     a pointer to the nfs4_delegation we are adding.
 895  * @fp:     a pointer to the nfs4_file we're granting a delegation on
 896  *
 897  * Return:
 898  *      On success: NULL if the delegation was successfully hashed.
 899  *
 900  *      On error: -EAGAIN if one was previously granted to this
 901  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
 902  *
 903  */
 904
 905 static int
 906 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
 907 {
 908         int status;
 909         struct nfs4_client *clp = dp->dl_stid.sc_client;
 910
 911         lockdep_assert_held(&state_lock);
 912         lockdep_assert_held(&fp->fi_lock);
 913
 914         status = nfs4_get_existing_delegation(clp, fp);
 915         if (status)
 916                 return status;
 917         ++fp->fi_delegees;
 918         atomic_inc(&dp->dl_stid.sc_count);
 919         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 920         list_add(&dp->dl_perfile, &fp->fi_delegations);
 921         list_add(&dp->dl_perclnt, &clp->cl_delegations);
 922         return 0;
 923 }
 924
 925 static bool
 926 unhash_delegation_locked(struct nfs4_delegation *dp)
 927 {
 928         struct nfs4_file *fp = dp->dl_stid.sc_file;
 929
 930         lockdep_assert_held(&state_lock);
 931
 932         if (list_empty(&dp->dl_perfile))
 933                 return false;
 934
 935         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
 936         /* Ensure that deleg break won't try to requeue it */
 937         ++dp->dl_time;
 938         spin_lock(&fp->fi_lock);
 939         list_del_init(&dp->dl_perclnt);
 940         list_del_init(&dp->dl_recall_lru);
 941         list_del_init(&dp->dl_perfile);
 942         spin_unlock(&fp->fi_lock);
 943         return true;
 944 }
 945
 946 static void destroy_delegation(struct nfs4_delegation *dp)
 947 {
 948         bool unhashed;
 949
 950         spin_lock(&state_lock);
 951         unhashed = unhash_delegation_locked(dp);
 952         spin_unlock(&state_lock);
 953         if (unhashed) {
 954                 put_clnt_odstate(dp->dl_clnt_odstate);
 955                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
 956                 nfs4_put_stid(&dp->dl_stid);
 957         }
 958 }
 959
 960 static void revoke_delegation(struct nfs4_delegation *dp)
 961 {
 962         struct nfs4_client *clp = dp->dl_stid.sc_client;
 963
 964         WARN_ON(!list_empty(&dp->dl_recall_lru));
 965
 966         put_clnt_odstate(dp->dl_clnt_odstate);
 967         nfs4_put_deleg_lease(dp->dl_stid.sc_file);
 968
 969         if (clp->cl_minorversion == 0)
 970                 nfs4_put_stid(&dp->dl_stid);
 971         else {
 972                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
 973                 spin_lock(&clp->cl_lock);
 974                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
 975                 spin_unlock(&clp->cl_lock);
 976         }
 977 }
 978
 979 /*
 980  * SETCLIENTID state
 981  */
 982
 983 static unsigned int clientid_hashval(u32 id)
 984 {
 985         return id & CLIENT_HASH_MASK;
 986 }
 987
 988 static unsigned int clientstr_hashval(const char *name)
 989 {
 990         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
 991 }
 992
 993 /*
 994  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 995  * st_{access,deny}_bmap field of the stateid, in order to track not
 996  * only what share bits are currently in force, but also what
 997  * combinations of share bits previous opens have used.  This allows us
 998  * to enforce the recommendation of rfc 3530 14.2.19 that the server
 999  * return an error if the client attempt to downgrade to a combination
1000  * of share bits not explicable by closing some of its previous opens.
1001  *
1002  * XXX: This enforcement is actually incomplete, since we don't keep
1003  * track of access/deny bit combinations; so, e.g., we allow:
1004  *
1005  *      OPEN allow read, deny write
1006  *      OPEN allow both, deny none
1007  *      DOWNGRADE allow read, deny none
1008  *
1009  * which we should reject.
1010  */
1011 static unsigned int
1012 bmap_to_share_mode(unsigned long bmap) {
1013         int i;
1014         unsigned int access = 0;
1015
1016         for (i = 1; i < 4; i++) {
1017                 if (test_bit(i, &bmap))
1018                         access |= i;
1019         }
1020         return access;
1021 }
1022
1023 /* set share access for a given stateid */
1024 static inline void
1025 set_access(u32 access, struct nfs4_ol_stateid *stp)
1026 {
1027         unsigned char mask = 1 << access;
1028
1029         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1030         stp->st_access_bmap |= mask;
1031 }
1032
1033 /* clear share access for a given stateid */
1034 static inline void
1035 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1036 {
1037         unsigned char mask = 1 << access;
1038
1039         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1040         stp->st_access_bmap &= ~mask;
1041 }
1042
1043 /* test whether a given stateid has access */
1044 static inline bool
1045 test_access(u32 access, struct nfs4_ol_stateid *stp)
1046 {
1047         unsigned char mask = 1 << access;
1048
1049         return (bool)(stp->st_access_bmap & mask);
1050 }
1051
1052 /* set share deny for a given stateid */
1053 static inline void
1054 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1055 {
1056         unsigned char mask = 1 << deny;
1057
1058         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1059         stp->st_deny_bmap |= mask;
1060 }
1061
1062 /* clear share deny for a given stateid */
1063 static inline void
1064 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1065 {
1066         unsigned char mask = 1 << deny;
1067
1068         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1069         stp->st_deny_bmap &= ~mask;
1070 }
1071
1072 /* test whether a given stateid is denying specific access */
1073 static inline bool
1074 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1075 {
1076         unsigned char mask = 1 << deny;
1077
1078         return (bool)(stp->st_deny_bmap & mask);
1079 }
1080
1081 static int nfs4_access_to_omode(u32 access)
1082 {
1083         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1084         case NFS4_SHARE_ACCESS_READ:
1085                 return O_RDONLY;
1086         case NFS4_SHARE_ACCESS_WRITE:
1087                 return O_WRONLY;
1088         case NFS4_SHARE_ACCESS_BOTH:
1089                 return O_RDWR;
1090         }
1091         WARN_ON_ONCE(1);
1092         return O_RDONLY;
1093 }
1094
1095 /*
1096  * A stateid that had a deny mode associated with it is being released
1097  * or downgraded. Recalculate the deny mode on the file.
1098  */
1099 static void
1100 recalculate_deny_mode(struct nfs4_file *fp)
1101 {
1102         struct nfs4_ol_stateid *stp;
1103
1104         spin_lock(&fp->fi_lock);
1105         fp->fi_share_deny = 0;
1106         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1107                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1108         spin_unlock(&fp->fi_lock);
1109 }
1110
1111 static void
1112 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1113 {
1114         int i;
1115         bool change = false;
1116
1117         for (i = 1; i < 4; i++) {
1118                 if ((i & deny) != i) {
1119                         change = true;
1120                         clear_deny(i, stp);
1121                 }
1122         }
1123
1124         /* Recalculate per-file deny mode if there was a change */
1125         if (change)
1126                 recalculate_deny_mode(stp->st_stid.sc_file);
1127 }
1128
1129 /* release all access and file references for a given stateid */
1130 static void
1131 release_all_access(struct nfs4_ol_stateid *stp)
1132 {
1133         int i;
1134         struct nfs4_file *fp = stp->st_stid.sc_file;
1135
1136         if (fp && stp->st_deny_bmap != 0)
1137                 recalculate_deny_mode(fp);
1138
1139         for (i = 1; i < 4; i++) {
1140                 if (test_access(i, stp))
1141                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1142                 clear_access(i, stp);
1143         }
1144 }
1145
1146 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1147 {
1148         kfree(sop->so_owner.data);
1149         sop->so_ops->so_free(sop);
1150 }
1151
1152 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1153 {
1154         struct nfs4_client *clp = sop->so_client;
1155
1156         might_lock(&clp->cl_lock);
1157
1158         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1159                 return;
1160         sop->so_ops->so_unhash(sop);
1161         spin_unlock(&clp->cl_lock);
1162         nfs4_free_stateowner(sop);
1163 }
1164
1165 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1166 {
1167         struct nfs4_file *fp = stp->st_stid.sc_file;
1168
1169         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1170
1171         if (list_empty(&stp->st_perfile))
1172                 return false;
1173
1174         spin_lock(&fp->fi_lock);
1175         list_del_init(&stp->st_perfile);
1176         spin_unlock(&fp->fi_lock);
1177         list_del(&stp->st_perstateowner);
1178         return true;
1179 }
1180
1181 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1182 {
1183         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1184
1185         put_clnt_odstate(stp->st_clnt_odstate);
1186         release_all_access(stp);
1187         if (stp->st_stateowner)
1188                 nfs4_put_stateowner(stp->st_stateowner);
1189         kmem_cache_free(stateid_slab, stid);
1190 }
1191
1192 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1193 {
1194         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1195         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1196         struct file *file;
1197
1198         file = find_any_file(stp->st_stid.sc_file);
1199         if (file)
1200                 filp_close(file, (fl_owner_t)lo);
1201         nfs4_free_ol_stateid(stid);
1202 }
1203
1204 /*
1205  * Put the persistent reference to an already unhashed generic stateid, while
1206  * holding the cl_lock. If it's the last reference, then put it onto the
1207  * reaplist for later destruction.
1208  */
1209 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1210                                        struct list_head *reaplist)
1211 {
1212         struct nfs4_stid *s = &stp->st_stid;
1213         struct nfs4_client *clp = s->sc_client;
1214
1215         lockdep_assert_held(&clp->cl_lock);
1216
1217         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1218
1219         if (!atomic_dec_and_test(&s->sc_count)) {
1220                 wake_up_all(&close_wq);
1221                 return;
1222         }
1223
1224         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1225         list_add(&stp->st_locks, reaplist);
1226 }
1227
1228 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1229 {
1230         struct nfs4_openowner *oo = openowner(stp->st_openstp->st_stateowner);
1231
1232         lockdep_assert_held(&oo->oo_owner.so_client->cl_lock);
1233
1234         list_del_init(&stp->st_locks);
1235         nfs4_unhash_stid(&stp->st_stid);
1236         return unhash_ol_stateid(stp);
1237 }
1238
1239 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1240 {
1241         struct nfs4_openowner *oo = openowner(stp->st_openstp->st_stateowner);
1242         bool unhashed;
1243
1244         spin_lock(&oo->oo_owner.so_client->cl_lock);
1245         unhashed = unhash_lock_stateid(stp);
1246         spin_unlock(&oo->oo_owner.so_client->cl_lock);
1247         if (unhashed)
1248                 nfs4_put_stid(&stp->st_stid);
1249 }
1250
1251 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1252 {
1253         struct nfs4_client *clp = lo->lo_owner.so_client;
1254
1255         lockdep_assert_held(&clp->cl_lock);
1256
1257         list_del_init(&lo->lo_owner.so_strhash);
1258 }
1259
1260 /*
1261  * Free a list of generic stateids that were collected earlier after being
1262  * fully unhashed.
1263  */
1264 static void
1265 free_ol_stateid_reaplist(struct list_head *reaplist)
1266 {
1267         struct nfs4_ol_stateid *stp;
1268         struct nfs4_file *fp;
1269
1270         might_sleep();
1271
1272         while (!list_empty(reaplist)) {
1273                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1274                                        st_locks);
1275                 list_del(&stp->st_locks);
1276                 fp = stp->st_stid.sc_file;
1277                 stp->st_stid.sc_free(&stp->st_stid);
1278                 if (fp)
1279                         put_nfs4_file(fp);
1280         }
1281 }
1282
1283 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1284                                        struct list_head *reaplist)
1285 {
1286         struct nfs4_ol_stateid *stp;
1287
1288         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1289
1290         while (!list_empty(&open_stp->st_locks)) {
1291                 stp = list_entry(open_stp->st_locks.next,
1292                                 struct nfs4_ol_stateid, st_locks);
1293                 WARN_ON(!unhash_lock_stateid(stp));
1294                 put_ol_stateid_locked(stp, reaplist);
1295         }
1296 }
1297
1298 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1299                                 struct list_head *reaplist)
1300 {
1301         bool unhashed;
1302
1303         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1304
1305         unhashed = unhash_ol_stateid(stp);
1306         release_open_stateid_locks(stp, reaplist);
1307         return unhashed;
1308 }
1309
1310 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1311 {
1312         LIST_HEAD(reaplist);
1313
1314         spin_lock(&stp->st_stid.sc_client->cl_lock);
1315         if (unhash_open_stateid(stp, &reaplist))
1316                 put_ol_stateid_locked(stp, &reaplist);
1317         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1318         free_ol_stateid_reaplist(&reaplist);
1319 }
1320
1321 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1322 {
1323         struct nfs4_client *clp = oo->oo_owner.so_client;
1324
1325         lockdep_assert_held(&clp->cl_lock);
1326
1327         list_del_init(&oo->oo_owner.so_strhash);
1328         list_del_init(&oo->oo_perclient);
1329 }
1330
1331 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1332 {
1333         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1334                                           nfsd_net_id);
1335         struct nfs4_ol_stateid *s;
1336
1337         spin_lock(&nn->client_lock);
1338         s = oo->oo_last_closed_stid;
1339         if (s) {
1340                 list_del_init(&oo->oo_close_lru);
1341                 oo->oo_last_closed_stid = NULL;
1342         }
1343         spin_unlock(&nn->client_lock);
1344         if (s)
1345                 nfs4_put_stid(&s->st_stid);
1346 }
1347
1348 static void release_openowner(struct nfs4_openowner *oo)
1349 {
1350         struct nfs4_ol_stateid *stp;
1351         struct nfs4_client *clp = oo->oo_owner.so_client;
1352         struct list_head reaplist;
1353
1354         INIT_LIST_HEAD(&reaplist);
1355
1356         spin_lock(&clp->cl_lock);
1357         unhash_openowner_locked(oo);
1358         while (!list_empty(&oo->oo_owner.so_stateids)) {
1359                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1360                                 struct nfs4_ol_stateid, st_perstateowner);
1361                 if (unhash_open_stateid(stp, &reaplist))
1362                         put_ol_stateid_locked(stp, &reaplist);
1363         }
1364         spin_unlock(&clp->cl_lock);
1365         free_ol_stateid_reaplist(&reaplist);
1366         release_last_closed_stateid(oo);
1367         nfs4_put_stateowner(&oo->oo_owner);
1368 }
1369
1370 static inline int
1371 hash_sessionid(struct nfs4_sessionid *sessionid)
1372 {
1373         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1374
1375         return sid->sequence % SESSION_HASH_SIZE;
1376 }
1377
1378 #ifdef CONFIG_SUNRPC_DEBUG
1379 static inline void
1380 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1381 {
1382         u32 *ptr = (u32 *)(&sessionid->data[0]);
1383         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1384 }
1385 #else
1386 static inline void
1387 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1388 {
1389 }
1390 #endif
1391
1392 /*
1393  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1394  * won't be used for replay.
1395  */
1396 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1397 {
1398         struct nfs4_stateowner *so = cstate->replay_owner;
1399
1400         if (nfserr == nfserr_replay_me)
1401                 return;
1402
1403         if (!seqid_mutating_err(ntohl(nfserr))) {
1404                 nfsd4_cstate_clear_replay(cstate);
1405                 return;
1406         }
1407         if (!so)
1408                 return;
1409         if (so->so_is_open_owner)
1410                 release_last_closed_stateid(openowner(so));
1411         so->so_seqid++;
1412         return;
1413 }
1414
1415 static void
1416 gen_sessionid(struct nfsd4_session *ses)
1417 {
1418         struct nfs4_client *clp = ses->se_client;
1419         struct nfsd4_sessionid *sid;
1420
1421         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1422         sid->clientid = clp->cl_clientid;
1423         sid->sequence = current_sessionid++;
1424         sid->reserved = 0;
1425 }
1426
1427 /*
1428  * The protocol defines ca_maxresponssize_cached to include the size of
1429  * the rpc header, but all we need to cache is the data starting after
1430  * the end of the initial SEQUENCE operation--the rest we regenerate
1431  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1432  * value that is the number of bytes in our cache plus a few additional
1433  * bytes.  In order to stay on the safe side, and not promise more than
1434  * we can cache, those additional bytes must be the minimum possible: 24
1435  * bytes of rpc header (xid through accept state, with AUTH_NULL
1436  * verifier), 12 for the compound header (with zero-length tag), and 44
1437  * for the SEQUENCE op response:
1438  */
1439 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1440
1441 static void
1442 free_session_slots(struct nfsd4_session *ses)
1443 {
1444         int i;
1445
1446         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
1447                 kfree(ses->se_slots[i]);
1448 }
1449
1450 /*
1451  * We don't actually need to cache the rpc and session headers, so we
1452  * can allocate a little less for each slot:
1453  */
1454 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1455 {
1456         u32 size;
1457
1458         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1459                 size = 0;
1460         else
1461                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1462         return size + sizeof(struct nfsd4_slot);
1463 }
1464
1465 /*
1466  * XXX: If we run out of reserved DRC memory we could (up to a point)
1467  * re-negotiate active sessions and reduce their slot usage to make
1468  * room for new connections. For now we just fail the create session.
1469  */
1470 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1471 {
1472         u32 slotsize = slot_bytes(ca);
1473         u32 num = ca->maxreqs;
1474         int avail;
1475
1476         spin_lock(&nfsd_drc_lock);
1477         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1478                     nfsd_drc_max_mem - nfsd_drc_mem_used);
1479         num = min_t(int, num, avail / slotsize);
1480         nfsd_drc_mem_used += num * slotsize;
1481         spin_unlock(&nfsd_drc_lock);
1482
1483         return num;
1484 }
1485
1486 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1487 {
1488         int slotsize = slot_bytes(ca);
1489
1490         spin_lock(&nfsd_drc_lock);
1491         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1492         spin_unlock(&nfsd_drc_lock);
1493 }
1494
1495 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1496                                            struct nfsd4_channel_attrs *battrs)
1497 {
1498         int numslots = fattrs->maxreqs;
1499         int slotsize = slot_bytes(fattrs);
1500         struct nfsd4_session *new;
1501         int mem, i;
1502
1503         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1504                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1505         mem = numslots * sizeof(struct nfsd4_slot *);
1506
1507         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1508         if (!new)
1509                 return NULL;
1510         /* allocate each struct nfsd4_slot and data cache in one piece */
1511         for (i = 0; i < numslots; i++) {
1512                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1513                 if (!new->se_slots[i])
1514                         goto out_free;
1515         }
1516
1517         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1518         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1519
1520         return new;
1521 out_free:
1522         while (i--)
1523                 kfree(new->se_slots[i]);
1524         kfree(new);
1525         return NULL;
1526 }
1527
1528 static void free_conn(struct nfsd4_conn *c)
1529 {
1530         svc_xprt_put(c->cn_xprt);
1531         kfree(c);
1532 }
1533
1534 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1535 {
1536         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1537         struct nfs4_client *clp = c->cn_session->se_client;
1538
1539         spin_lock(&clp->cl_lock);
1540         if (!list_empty(&c->cn_persession)) {
1541                 list_del(&c->cn_persession);
1542                 free_conn(c);
1543         }
1544         nfsd4_probe_callback(clp);
1545         spin_unlock(&clp->cl_lock);
1546 }
1547
1548 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1549 {
1550         struct nfsd4_conn *conn;
1551
1552         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1553         if (!conn)
1554                 return NULL;
1555         svc_xprt_get(rqstp->rq_xprt);
1556         conn->cn_xprt = rqstp->rq_xprt;
1557         conn->cn_flags = flags;
1558         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1559         return conn;
1560 }
1561
1562 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1563 {
1564         conn->cn_session = ses;
1565         list_add(&conn->cn_persession, &ses->se_conns);
1566 }
1567
1568 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1569 {
1570         struct nfs4_client *clp = ses->se_client;
1571
1572         spin_lock(&clp->cl_lock);
1573         __nfsd4_hash_conn(conn, ses);
1574         spin_unlock(&clp->cl_lock);
1575 }
1576
1577 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1578 {
1579         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1580         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1581 }
1582
1583 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1584 {
1585         int ret;
1586
1587         nfsd4_hash_conn(conn, ses);
1588         ret = nfsd4_register_conn(conn);
1589         if (ret)
1590                 /* oops; xprt is already down: */
1591                 nfsd4_conn_lost(&conn->cn_xpt_user);
1592         /* We may have gained or lost a callback channel: */
1593         nfsd4_probe_callback_sync(ses->se_client);
1594 }
1595
1596 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1597 {
1598         u32 dir = NFS4_CDFC4_FORE;
1599
1600         if (cses->flags & SESSION4_BACK_CHAN)
1601                 dir |= NFS4_CDFC4_BACK;
1602         return alloc_conn(rqstp, dir);
1603 }
1604
1605 /* must be called under client_lock */
1606 static void nfsd4_del_conns(struct nfsd4_session *s)
1607 {
1608         struct nfs4_client *clp = s->se_client;
1609         struct nfsd4_conn *c;
1610
1611         spin_lock(&clp->cl_lock);
1612         while (!list_empty(&s->se_conns)) {
1613                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1614                 list_del_init(&c->cn_persession);
1615                 spin_unlock(&clp->cl_lock);
1616
1617                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1618                 free_conn(c);
1619
1620                 spin_lock(&clp->cl_lock);
1621         }
1622         spin_unlock(&clp->cl_lock);
1623 }
1624
1625 static void __free_session(struct nfsd4_session *ses)
1626 {
1627         free_session_slots(ses);
1628         kfree(ses);
1629 }
1630
1631 static void free_session(struct nfsd4_session *ses)
1632 {
1633         nfsd4_del_conns(ses);
1634         nfsd4_put_drc_mem(&ses->se_fchannel);
1635         __free_session(ses);
1636 }
1637
1638 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1639 {
1640         int idx;
1641         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1642
1643         new->se_client = clp;
1644         gen_sessionid(new);
1645
1646         INIT_LIST_HEAD(&new->se_conns);
1647
1648         new->se_cb_seq_nr = 1;
1649         new->se_flags = cses->flags;
1650         new->se_cb_prog = cses->callback_prog;
1651         new->se_cb_sec = cses->cb_sec;
1652         atomic_set(&new->se_ref, 0);
1653         idx = hash_sessionid(&new->se_sessionid);
1654         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1655         spin_lock(&clp->cl_lock);
1656         list_add(&new->se_perclnt, &clp->cl_sessions);
1657         spin_unlock(&clp->cl_lock);
1658
1659         {
1660                 struct sockaddr *sa = svc_addr(rqstp);
1661                 /*
1662                  * This is a little silly; with sessions there's no real
1663                  * use for the callback address.  Use the peer address
1664                  * as a reasonable default for now, but consider fixing
1665                  * the rpc client not to require an address in the
1666                  * future:
1667                  */
1668                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1669                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1670         }
1671 }
1672
1673 /* caller must hold client_lock */
1674 static struct nfsd4_session *
1675 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1676 {
1677         struct nfsd4_session *elem;
1678         int idx;
1679         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1680
1681         lockdep_assert_held(&nn->client_lock);
1682
1683         dump_sessionid(__func__, sessionid);
1684         idx = hash_sessionid(sessionid);
1685         /* Search in the appropriate list */
1686         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1687                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1688                             NFS4_MAX_SESSIONID_LEN)) {
1689                         return elem;
1690                 }
1691         }
1692
1693         dprintk("%s: session not found\n", __func__);
1694         return NULL;
1695 }
1696
1697 static struct nfsd4_session *
1698 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1699                 __be32 *ret)
1700 {
1701         struct nfsd4_session *session;
1702         __be32 status = nfserr_badsession;
1703
1704         session = __find_in_sessionid_hashtbl(sessionid, net);
1705         if (!session)
1706                 goto out;
1707         status = nfsd4_get_session_locked(session);
1708         if (status)
1709                 session = NULL;
1710 out:
1711         *ret = status;
1712         return session;
1713 }
1714
1715 /* caller must hold client_lock */
1716 static void
1717 unhash_session(struct nfsd4_session *ses)
1718 {
1719         struct nfs4_client *clp = ses->se_client;
1720         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1721
1722         lockdep_assert_held(&nn->client_lock);
1723
1724         list_del(&ses->se_hash);
1725         spin_lock(&ses->se_client->cl_lock);
1726         list_del(&ses->se_perclnt);
1727         spin_unlock(&ses->se_client->cl_lock);
1728 }
1729
1730 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1731 static int
1732 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1733 {
1734         /*
1735          * We're assuming the clid was not given out from a boot
1736          * precisely 2^32 (about 136 years) before this one.  That seems
1737          * a safe assumption:
1738          */
1739         if (clid->cl_boot == (u32)nn->boot_time)
1740                 return 0;
1741         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1742                 clid->cl_boot, clid->cl_id, nn->boot_time);
1743         return 1;
1744 }
1745
1746 /*
1747  * XXX Should we use a slab cache ?
1748  * This type of memory management is somewhat inefficient, but we use it
1749  * anyway since SETCLIENTID is not a common operation.
1750  */
1751 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1752 {
1753         struct nfs4_client *clp;
1754         int i;
1755
1756         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1757         if (clp == NULL)
1758                 return NULL;
1759         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1760         if (clp->cl_name.data == NULL)
1761                 goto err_no_name;
1762         clp->cl_ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
1763                         OWNER_HASH_SIZE, GFP_KERNEL);
1764         if (!clp->cl_ownerstr_hashtbl)
1765                 goto err_no_hashtbl;
1766         for (i = 0; i < OWNER_HASH_SIZE; i++)
1767                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1768         clp->cl_name.len = name.len;
1769         INIT_LIST_HEAD(&clp->cl_sessions);
1770         idr_init(&clp->cl_stateids);
1771         atomic_set(&clp->cl_refcount, 0);
1772         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1773         INIT_LIST_HEAD(&clp->cl_idhash);
1774         INIT_LIST_HEAD(&clp->cl_openowners);
1775         INIT_LIST_HEAD(&clp->cl_delegations);
1776         INIT_LIST_HEAD(&clp->cl_lru);
1777         INIT_LIST_HEAD(&clp->cl_revoked);
1778 #ifdef CONFIG_NFSD_PNFS
1779         INIT_LIST_HEAD(&clp->cl_lo_states);
1780 #endif
1781         spin_lock_init(&clp->cl_lock);
1782         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1783         return clp;
1784 err_no_hashtbl:
1785         kfree(clp->cl_name.data);
1786 err_no_name:
1787         kfree(clp);
1788         return NULL;
1789 }
1790
1791 static void
1792 free_client(struct nfs4_client *clp)
1793 {
1794         while (!list_empty(&clp->cl_sessions)) {
1795                 struct nfsd4_session *ses;
1796                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1797                                 se_perclnt);
1798                 list_del(&ses->se_perclnt);
1799                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1800                 free_session(ses);
1801         }
1802         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1803         free_svc_cred(&clp->cl_cred);
1804         kfree(clp->cl_ownerstr_hashtbl);
1805         kfree(clp->cl_name.data);
1806         idr_destroy(&clp->cl_stateids);
1807         kfree(clp);
1808 }
1809
1810 /* must be called under the client_lock */
1811 static void
1812 unhash_client_locked(struct nfs4_client *clp)
1813 {
1814         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1815         struct nfsd4_session *ses;
1816
1817         lockdep_assert_held(&nn->client_lock);
1818
1819         /* Mark the client as expired! */
1820         clp->cl_time = 0;
1821         /* Make it invisible */
1822         if (!list_empty(&clp->cl_idhash)) {
1823                 list_del_init(&clp->cl_idhash);
1824                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1825                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1826                 else
1827                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1828         }
1829         list_del_init(&clp->cl_lru);
1830         spin_lock(&clp->cl_lock);
1831         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1832                 list_del_init(&ses->se_hash);
1833         spin_unlock(&clp->cl_lock);
1834 }
1835
1836 static void
1837 unhash_client(struct nfs4_client *clp)
1838 {
1839         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1840
1841         spin_lock(&nn->client_lock);
1842         unhash_client_locked(clp);
1843         spin_unlock(&nn->client_lock);
1844 }
1845
1846 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1847 {
1848         if (atomic_read(&clp->cl_refcount))
1849                 return nfserr_jukebox;
1850         unhash_client_locked(clp);
1851         return nfs_ok;
1852 }
1853
1854 static void
1855 __destroy_client(struct nfs4_client *clp)
1856 {
1857         struct nfs4_openowner *oo;
1858         struct nfs4_delegation *dp;
1859         struct list_head reaplist;
1860
1861         INIT_LIST_HEAD(&reaplist);
1862         spin_lock(&state_lock);
1863         while (!list_empty(&clp->cl_delegations)) {
1864                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1865                 WARN_ON(!unhash_delegation_locked(dp));
1866                 list_add(&dp->dl_recall_lru, &reaplist);
1867         }
1868         spin_unlock(&state_lock);
1869         while (!list_empty(&reaplist)) {
1870                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1871                 list_del_init(&dp->dl_recall_lru);
1872                 put_clnt_odstate(dp->dl_clnt_odstate);
1873                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1874                 nfs4_put_stid(&dp->dl_stid);
1875         }
1876         while (!list_empty(&clp->cl_revoked)) {
1877                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1878                 list_del_init(&dp->dl_recall_lru);
1879                 nfs4_put_stid(&dp->dl_stid);
1880         }
1881         while (!list_empty(&clp->cl_openowners)) {
1882                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1883                 nfs4_get_stateowner(&oo->oo_owner);
1884                 release_openowner(oo);
1885         }
1886         nfsd4_return_all_client_layouts(clp);
1887         nfsd4_shutdown_callback(clp);
1888         if (clp->cl_cb_conn.cb_xprt)
1889                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1890         free_client(clp);
1891 }
1892
1893 static void
1894 destroy_client(struct nfs4_client *clp)
1895 {
1896         unhash_client(clp);
1897         __destroy_client(clp);
1898 }
1899
1900 static void expire_client(struct nfs4_client *clp)
1901 {
1902         unhash_client(clp);
1903         nfsd4_client_record_remove(clp);
1904         __destroy_client(clp);
1905 }
1906
1907 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1908 {
1909         memcpy(target->cl_verifier.data, source->data,
1910                         sizeof(target->cl_verifier.data));
1911 }
1912
1913 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1914 {
1915         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1916         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1917 }
1918
1919 int strdup_if_nonnull(char **target, char *source)
1920 {
1921         if (source) {
1922                 *target = kstrdup(source, GFP_KERNEL);
1923                 if (!*target)
1924                         return -ENOMEM;
1925         } else
1926                 *target = NULL;
1927         return 0;
1928 }
1929
1930 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1931 {
1932         int ret;
1933
1934         ret = strdup_if_nonnull(&target->cr_principal, source->cr_principal);
1935         if (ret)
1936                 return ret;
1937         ret = strdup_if_nonnull(&target->cr_raw_principal,
1938                                         source->cr_raw_principal);
1939         if (ret)
1940                 return ret;
1941         target->cr_flavor = source->cr_flavor;
1942         target->cr_uid = source->cr_uid;
1943         target->cr_gid = source->cr_gid;
1944         target->cr_group_info = source->cr_group_info;
1945         get_group_info(target->cr_group_info);
1946         target->cr_gss_mech = source->cr_gss_mech;
1947         if (source->cr_gss_mech)
1948                 gss_mech_get(source->cr_gss_mech);
1949         return 0;
1950 }
1951
1952 static int
1953 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1954 {
1955         if (o1->len < o2->len)
1956                 return -1;
1957         if (o1->len > o2->len)
1958                 return 1;
1959         return memcmp(o1->data, o2->data, o1->len);
1960 }
1961
1962 static int same_name(const char *n1, const char *n2)
1963 {
1964         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1965 }
1966
1967 static int
1968 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1969 {
1970         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1971 }
1972
1973 static int
1974 same_clid(clientid_t *cl1, clientid_t *cl2)
1975 {
1976         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1977 }
1978
1979 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1980 {
1981         int i;
1982
1983         if (g1->ngroups != g2->ngroups)
1984                 return false;
1985         for (i=0; i<g1->ngroups; i++)
1986                 if (!gid_eq(g1->gid[i], g2->gid[i]))
1987                         return false;
1988         return true;
1989 }
1990
1991 /*
1992  * RFC 3530 language requires clid_inuse be returned when the
1993  * "principal" associated with a requests differs from that previously
1994  * used.  We use uid, gid's, and gss principal string as our best
1995  * approximation.  We also don't want to allow non-gss use of a client
1996  * established using gss: in theory cr_principal should catch that
1997  * change, but in practice cr_principal can be null even in the gss case
1998  * since gssd doesn't always pass down a principal string.
1999  */
2000 static bool is_gss_cred(struct svc_cred *cr)
2001 {
2002         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2003         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2004 }
2005
2006
2007 static bool
2008 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2009 {
2010         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2011                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2012                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2013                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2014                 return false;
2015         if (cr1->cr_principal == cr2->cr_principal)
2016                 return true;
2017         if (!cr1->cr_principal || !cr2->cr_principal)
2018                 return false;
2019         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2020 }
2021
2022 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2023 {
2024         struct svc_cred *cr = &rqstp->rq_cred;
2025         u32 service;
2026
2027         if (!cr->cr_gss_mech)
2028                 return false;
2029         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2030         return service == RPC_GSS_SVC_INTEGRITY ||
2031                service == RPC_GSS_SVC_PRIVACY;
2032 }
2033
2034 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2035 {
2036         struct svc_cred *cr = &rqstp->rq_cred;
2037
2038         if (!cl->cl_mach_cred)
2039                 return true;
2040         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2041                 return false;
2042         if (!svc_rqst_integrity_protected(rqstp))
2043                 return false;
2044         if (cl->cl_cred.cr_raw_principal)
2045                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2046                                                 cr->cr_raw_principal);
2047         if (!cr->cr_principal)
2048                 return false;
2049         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2050 }
2051
2052 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2053 {
2054         __be32 verf[2];
2055
2056         /*
2057          * This is opaque to client, so no need to byte-swap. Use
2058          * __force to keep sparse happy
2059          */
2060         verf[0] = (__force __be32)get_seconds();
2061         verf[1] = (__force __be32)nn->clverifier_counter++;
2062         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2063 }
2064
2065 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2066 {
2067         clp->cl_clientid.cl_boot = nn->boot_time;
2068         clp->cl_clientid.cl_id = nn->clientid_counter++;
2069         gen_confirm(clp, nn);
2070 }
2071
2072 static struct nfs4_stid *
2073 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2074 {
2075         struct nfs4_stid *ret;
2076
2077         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2078         if (!ret || !ret->sc_type)
2079                 return NULL;
2080         return ret;
2081 }
2082
2083 static struct nfs4_stid *
2084 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2085 {
2086         struct nfs4_stid *s;
2087
2088         spin_lock(&cl->cl_lock);
2089         s = find_stateid_locked(cl, t);
2090         if (s != NULL) {
2091                 if (typemask & s->sc_type)
2092                         atomic_inc(&s->sc_count);
2093                 else
2094                         s = NULL;
2095         }
2096         spin_unlock(&cl->cl_lock);
2097         return s;
2098 }
2099
2100 static struct nfs4_client *create_client(struct xdr_netobj name,
2101                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2102 {
2103         struct nfs4_client *clp;
2104         struct sockaddr *sa = svc_addr(rqstp);
2105         int ret;
2106         struct net *net = SVC_NET(rqstp);
2107
2108         clp = alloc_client(name);
2109         if (clp == NULL)
2110                 return NULL;
2111
2112         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2113         if (ret) {
2114                 free_client(clp);
2115                 return NULL;
2116         }
2117         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2118         clp->cl_time = get_seconds();
2119         clear_bit(0, &clp->cl_cb_slot_busy);
2120         copy_verf(clp, verf);
2121         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2122         clp->cl_cb_session = NULL;
2123         clp->net = net;
2124         return clp;
2125 }
2126
2127 static void
2128 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2129 {
2130         struct rb_node **new = &(root->rb_node), *parent = NULL;
2131         struct nfs4_client *clp;
2132
2133         while (*new) {
2134                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2135                 parent = *new;
2136
2137                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2138                         new = &((*new)->rb_left);
2139                 else
2140                         new = &((*new)->rb_right);
2141         }
2142
2143         rb_link_node(&new_clp->cl_namenode, parent, new);
2144         rb_insert_color(&new_clp->cl_namenode, root);
2145 }
2146
2147 static struct nfs4_client *
2148 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2149 {
2150         int cmp;
2151         struct rb_node *node = root->rb_node;
2152         struct nfs4_client *clp;
2153
2154         while (node) {
2155                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2156                 cmp = compare_blob(&clp->cl_name, name);
2157                 if (cmp > 0)
2158                         node = node->rb_left;
2159                 else if (cmp < 0)
2160                         node = node->rb_right;
2161                 else
2162                         return clp;
2163         }
2164         return NULL;
2165 }
2166
2167 static void
2168 add_to_unconfirmed(struct nfs4_client *clp)
2169 {
2170         unsigned int idhashval;
2171         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2172
2173         lockdep_assert_held(&nn->client_lock);
2174
2175         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2176         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2177         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2178         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2179         renew_client_locked(clp);
2180 }
2181
2182 static void
2183 move_to_confirmed(struct nfs4_client *clp)
2184 {
2185         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2186         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2187
2188         lockdep_assert_held(&nn->client_lock);
2189
2190         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2191         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2192         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2193         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2194         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2195         renew_client_locked(clp);
2196 }
2197
2198 static struct nfs4_client *
2199 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2200 {
2201         struct nfs4_client *clp;
2202         unsigned int idhashval = clientid_hashval(clid->cl_id);
2203
2204         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2205                 if (same_clid(&clp->cl_clientid, clid)) {
2206                         if ((bool)clp->cl_minorversion != sessions)
2207                                 return NULL;
2208                         renew_client_locked(clp);
2209                         return clp;
2210                 }
2211         }
2212         return NULL;
2213 }
2214
2215 static struct nfs4_client *
2216 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2217 {
2218         struct list_head *tbl = nn->conf_id_hashtbl;
2219
2220         lockdep_assert_held(&nn->client_lock);
2221         return find_client_in_id_table(tbl, clid, sessions);
2222 }
2223
2224 static struct nfs4_client *
2225 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2226 {
2227         struct list_head *tbl = nn->unconf_id_hashtbl;
2228
2229         lockdep_assert_held(&nn->client_lock);
2230         return find_client_in_id_table(tbl, clid, sessions);
2231 }
2232
2233 static bool clp_used_exchangeid(struct nfs4_client *clp)
2234 {
2235         return clp->cl_exchange_flags != 0;
2236 }
2237
2238 static struct nfs4_client *
2239 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2240 {
2241         lockdep_assert_held(&nn->client_lock);
2242         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2243 }
2244
2245 static struct nfs4_client *
2246 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2247 {
2248         lockdep_assert_held(&nn->client_lock);
2249         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2250 }
2251
2252 static void
2253 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2254 {
2255         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2256         struct sockaddr *sa = svc_addr(rqstp);
2257         u32 scopeid = rpc_get_scope_id(sa);
2258         unsigned short expected_family;
2259
2260         /* Currently, we only support tcp and tcp6 for the callback channel */
2261         if (se->se_callback_netid_len == 3 &&
2262             !memcmp(se->se_callback_netid_val, "tcp", 3))
2263                 expected_family = AF_INET;
2264         else if (se->se_callback_netid_len == 4 &&
2265                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2266                 expected_family = AF_INET6;
2267         else
2268                 goto out_err;
2269
2270         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2271                                             se->se_callback_addr_len,
2272                                             (struct sockaddr *)&conn->cb_addr,
2273                                             sizeof(conn->cb_addr));
2274
2275         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2276                 goto out_err;
2277
2278         if (conn->cb_addr.ss_family == AF_INET6)
2279                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2280
2281         conn->cb_prog = se->se_callback_prog;
2282         conn->cb_ident = se->se_callback_ident;
2283         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2284         return;
2285 out_err:
2286         conn->cb_addr.ss_family = AF_UNSPEC;
2287         conn->cb_addrlen = 0;
2288         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
2289                 "will not receive delegations\n",
2290                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2291
2292         return;
2293 }
2294
2295 /*
2296  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2297  */
2298 static void
2299 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2300 {
2301         struct xdr_buf *buf = resp->xdr.buf;
2302         struct nfsd4_slot *slot = resp->cstate.slot;
2303         unsigned int base;
2304
2305         dprintk("--> %s slot %p\n", __func__, slot);
2306
2307         slot->sl_opcnt = resp->opcnt;
2308         slot->sl_status = resp->cstate.status;
2309
2310         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2311         if (nfsd4_not_cached(resp)) {
2312                 slot->sl_datalen = 0;
2313                 return;
2314         }
2315         base = resp->cstate.data_offset;
2316         slot->sl_datalen = buf->len - base;
2317         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2318                 WARN(1, "%s: sessions DRC could not cache compound\n",
2319                      __func__);
2320         return;
2321 }
2322
2323 /*
2324  * Encode the replay sequence operation from the slot values.
2325  * If cachethis is FALSE encode the uncached rep error on the next
2326  * operation which sets resp->p and increments resp->opcnt for
2327  * nfs4svc_encode_compoundres.
2328  *
2329  */
2330 static __be32
2331 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2332                           struct nfsd4_compoundres *resp)
2333 {
2334         struct nfsd4_op *op;
2335         struct nfsd4_slot *slot = resp->cstate.slot;
2336
2337         /* Encode the replayed sequence operation */
2338         op = &args->ops[resp->opcnt - 1];
2339         nfsd4_encode_operation(resp, op);
2340
2341         /* Return nfserr_retry_uncached_rep in next operation. */
2342         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
2343                 op = &args->ops[resp->opcnt++];
2344                 op->status = nfserr_retry_uncached_rep;
2345                 nfsd4_encode_operation(resp, op);
2346         }
2347         return op->status;
2348 }
2349
2350 /*
2351  * The sequence operation is not cached because we can use the slot and
2352  * session values.
2353  */
2354 static __be32
2355 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2356                          struct nfsd4_sequence *seq)
2357 {
2358         struct nfsd4_slot *slot = resp->cstate.slot;
2359         struct xdr_stream *xdr = &resp->xdr;
2360         __be32 *p;
2361         __be32 status;
2362
2363         dprintk("--> %s slot %p\n", __func__, slot);
2364
2365         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2366         if (status)
2367                 return status;
2368
2369         p = xdr_reserve_space(xdr, slot->sl_datalen);
2370         if (!p) {
2371                 WARN_ON_ONCE(1);
2372                 return nfserr_serverfault;
2373         }
2374         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2375         xdr_commit_encode(xdr);
2376
2377         resp->opcnt = slot->sl_opcnt;
2378         return slot->sl_status;
2379 }
2380
2381 /*
2382  * Set the exchange_id flags returned by the server.
2383  */
2384 static void
2385 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2386 {
2387 #ifdef CONFIG_NFSD_PNFS
2388         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2389 #else
2390         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2391 #endif
2392
2393         /* Referrals are supported, Migration is not. */
2394         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2395
2396         /* set the wire flags to return to client. */
2397         clid->flags = new->cl_exchange_flags;
2398 }
2399
2400 static bool client_has_openowners(struct nfs4_client *clp)
2401 {
2402         struct nfs4_openowner *oo;
2403
2404         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2405                 if (!list_empty(&oo->oo_owner.so_stateids))
2406                         return true;
2407         }
2408         return false;
2409 }
2410
2411 static bool client_has_state(struct nfs4_client *clp)
2412 {
2413         return client_has_openowners(clp)
2414 #ifdef CONFIG_NFSD_PNFS
2415                 || !list_empty(&clp->cl_lo_states)
2416 #endif
2417                 || !list_empty(&clp->cl_delegations)
2418                 || !list_empty(&clp->cl_sessions);
2419 }
2420
2421 __be32
2422 nfsd4_exchange_id(struct svc_rqst *rqstp,
2423                   struct nfsd4_compound_state *cstate,
2424                   struct nfsd4_exchange_id *exid)
2425 {
2426         struct nfs4_client *conf, *new;
2427         struct nfs4_client *unconf = NULL;
2428         __be32 status;
2429         char                    addr_str[INET6_ADDRSTRLEN];
2430         nfs4_verifier           verf = exid->verifier;
2431         struct sockaddr         *sa = svc_addr(rqstp);
2432         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2433         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2434
2435         rpc_ntop(sa, addr_str, sizeof(addr_str));
2436         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2437                 "ip_addr=%s flags %x, spa_how %d\n",
2438                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2439                 addr_str, exid->flags, exid->spa_how);
2440
2441         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2442                 return nfserr_inval;
2443
2444         new = create_client(exid->clname, rqstp, &verf);
2445         if (new == NULL)
2446                 return nfserr_jukebox;
2447
2448         switch (exid->spa_how) {
2449         case SP4_MACH_CRED:
2450                 exid->spo_must_enforce[0] = 0;
2451                 exid->spo_must_enforce[1] = (
2452                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
2453                         1 << (OP_EXCHANGE_ID - 32) |
2454                         1 << (OP_CREATE_SESSION - 32) |
2455                         1 << (OP_DESTROY_SESSION - 32) |
2456                         1 << (OP_DESTROY_CLIENTID - 32));
2457
2458                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2459                                         1 << (OP_OPEN_DOWNGRADE) |
2460                                         1 << (OP_LOCKU) |
2461                                         1 << (OP_DELEGRETURN));
2462
2463                 exid->spo_must_allow[1] &= (
2464                                         1 << (OP_TEST_STATEID - 32) |
2465                                         1 << (OP_FREE_STATEID - 32));
2466                 if (!svc_rqst_integrity_protected(rqstp)) {
2467                         status = nfserr_inval;
2468                         goto out_nolock;
2469                 }
2470                 /*
2471                  * Sometimes userspace doesn't give us a principal.
2472                  * Which is a bug, really.  Anyway, we can't enforce
2473                  * MACH_CRED in that case, better to give up now:
2474                  */
2475                 if (!new->cl_cred.cr_principal &&
2476                                         !new->cl_cred.cr_raw_principal) {
2477                         status = nfserr_serverfault;
2478                         goto out_nolock;
2479                 }
2480                 new->cl_mach_cred = true;
2481         case SP4_NONE:
2482                 break;
2483         default:                                /* checked by xdr code */
2484                 WARN_ON_ONCE(1);
2485         case SP4_SSV:
2486                 status = nfserr_encr_alg_unsupp;
2487                 goto out_nolock;
2488         }
2489
2490         /* Cases below refer to rfc 5661 section 18.35.4: */
2491         spin_lock(&nn->client_lock);
2492         conf = find_confirmed_client_by_name(&exid->clname, nn);
2493         if (conf) {
2494                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2495                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2496
2497                 if (update) {
2498                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2499                                 status = nfserr_inval;
2500                                 goto out;
2501                         }
2502                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
2503                                 status = nfserr_wrong_cred;
2504                                 goto out;
2505                         }
2506                         if (!creds_match) { /* case 9 */
2507                                 status = nfserr_perm;
2508                                 goto out;
2509                         }
2510                         if (!verfs_match) { /* case 8 */
2511                                 status = nfserr_not_same;
2512                                 goto out;
2513                         }
2514                         /* case 6 */
2515                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2516                         goto out_copy;
2517                 }
2518                 if (!creds_match) { /* case 3 */
2519                         if (client_has_state(conf)) {
2520                                 status = nfserr_clid_inuse;
2521                                 goto out;
2522                         }
2523                         goto out_new;
2524                 }
2525                 if (verfs_match) { /* case 2 */
2526                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2527                         goto out_copy;
2528                 }
2529                 /* case 5, client reboot */
2530                 conf = NULL;
2531                 goto out_new;
2532         }
2533
2534         if (update) { /* case 7 */
2535                 status = nfserr_noent;
2536                 goto out;
2537         }
2538
2539         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2540         if (unconf) /* case 4, possible retry or client restart */
2541                 unhash_client_locked(unconf);
2542
2543         /* case 1 (normal case) */
2544 out_new:
2545         if (conf) {
2546                 status = mark_client_expired_locked(conf);
2547                 if (status)
2548                         goto out;
2549         }
2550         new->cl_minorversion = cstate->minorversion;
2551         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2552         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2553
2554         gen_clid(new, nn);
2555         add_to_unconfirmed(new);
2556         swap(new, conf);
2557 out_copy:
2558         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2559         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2560
2561         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2562         nfsd4_set_ex_flags(conf, exid);
2563
2564         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2565                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2566         status = nfs_ok;
2567
2568 out:
2569         spin_unlock(&nn->client_lock);
2570 out_nolock:
2571         if (new)
2572                 expire_client(new);
2573         if (unconf)
2574                 expire_client(unconf);
2575         return status;
2576 }
2577
2578 static __be32
2579 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2580 {
2581         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2582                 slot_seqid);
2583
2584         /* The slot is in use, and no response has been sent. */
2585         if (slot_inuse) {
2586                 if (seqid == slot_seqid)
2587                         return nfserr_jukebox;
2588                 else
2589                         return nfserr_seq_misordered;
2590         }
2591         /* Note unsigned 32-bit arithmetic handles wraparound: */
2592         if (likely(seqid == slot_seqid + 1))
2593                 return nfs_ok;
2594         if (seqid == slot_seqid)
2595                 return nfserr_replay_cache;
2596         return nfserr_seq_misordered;
2597 }
2598
2599 /*
2600  * Cache the create session result into the create session single DRC
2601  * slot cache by saving the xdr structure. sl_seqid has been set.
2602  * Do this for solo or embedded create session operations.
2603  */
2604 static void
2605 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2606                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2607 {
2608         slot->sl_status = nfserr;
2609         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2610 }
2611
2612 static __be32
2613 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2614                             struct nfsd4_clid_slot *slot)
2615 {
2616         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2617         return slot->sl_status;
2618 }
2619
2620 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2621                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2622                         1 +     /* MIN tag is length with zero, only length */ \
2623                         3 +     /* version, opcount, opcode */ \
2624                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2625                                 /* seqid, slotID, slotID, cache */ \
2626                         4 ) * sizeof(__be32))
2627
2628 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2629                         2 +     /* verifier: AUTH_NULL, length 0 */\
2630                         1 +     /* status */ \
2631                         1 +     /* MIN tag is length with zero, only length */ \
2632                         3 +     /* opcount, opcode, opstatus*/ \
2633                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2634                                 /* seqid, slotID, slotID, slotID, status */ \
2635                         5 ) * sizeof(__be32))
2636
2637 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2638 {
2639         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2640
2641         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2642                 return nfserr_toosmall;
2643         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2644                 return nfserr_toosmall;
2645         ca->headerpadsz = 0;
2646         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2647         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2648         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2649         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2650                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2651         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2652         /*
2653          * Note decreasing slot size below client's request may make it
2654          * difficult for client to function correctly, whereas
2655          * decreasing the number of slots will (just?) affect
2656          * performance.  When short on memory we therefore prefer to
2657          * decrease number of slots instead of their size.  Clients that
2658          * request larger slots than they need will get poor results:
2659          */
2660         ca->maxreqs = nfsd4_get_drc_mem(ca);
2661         if (!ca->maxreqs)
2662                 return nfserr_jukebox;
2663
2664         return nfs_ok;
2665 }
2666
2667 /*
2668  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2669  * These are based on similar macros in linux/sunrpc/msg_prot.h .
2670  */
2671 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2672         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2673
2674 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2675         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2676
2677 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2678                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2679 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2680                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2681                                  sizeof(__be32))
2682
2683 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2684 {
2685         ca->headerpadsz = 0;
2686
2687         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2688                 return nfserr_toosmall;
2689         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2690                 return nfserr_toosmall;
2691         ca->maxresp_cached = 0;
2692         if (ca->maxops < 2)
2693                 return nfserr_toosmall;
2694
2695         return nfs_ok;
2696 }
2697
2698 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2699 {
2700         switch (cbs->flavor) {
2701         case RPC_AUTH_NULL:
2702         case RPC_AUTH_UNIX:
2703                 return nfs_ok;
2704         default:
2705                 /*
2706                  * GSS case: the spec doesn't allow us to return this
2707                  * error.  But it also doesn't allow us not to support
2708                  * GSS.
2709                  * I'd rather this fail hard than return some error the
2710                  * client might think it can already handle:
2711                  */
2712                 return nfserr_encr_alg_unsupp;
2713         }
2714 }
2715
2716 __be32
2717 nfsd4_create_session(struct svc_rqst *rqstp,
2718                      struct nfsd4_compound_state *cstate,
2719                      struct nfsd4_create_session *cr_ses)
2720 {
2721         struct sockaddr *sa = svc_addr(rqstp);
2722         struct nfs4_client *conf, *unconf;
2723         struct nfs4_client *old = NULL;
2724         struct nfsd4_session *new;
2725         struct nfsd4_conn *conn;
2726         struct nfsd4_clid_slot *cs_slot = NULL;
2727         __be32 status = 0;
2728         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2729
2730         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2731                 return nfserr_inval;
2732         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2733         if (status)
2734                 return status;
2735         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2736         if (status)
2737                 return status;
2738         status = check_backchannel_attrs(&cr_ses->back_channel);
2739         if (status)
2740                 goto out_release_drc_mem;
2741         status = nfserr_jukebox;
2742         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2743         if (!new)
2744                 goto out_release_drc_mem;
2745         conn = alloc_conn_from_crses(rqstp, cr_ses);
2746         if (!conn)
2747                 goto out_free_session;
2748
2749         spin_lock(&nn->client_lock);
2750         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2751         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2752         WARN_ON_ONCE(conf && unconf);
2753
2754         if (conf) {
2755                 status = nfserr_wrong_cred;
2756                 if (!nfsd4_mach_creds_match(conf, rqstp))
2757                         goto out_free_conn;
2758                 cs_slot = &conf->cl_cs_slot;
2759                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2760                 if (status) {
2761                         if (status == nfserr_replay_cache)
2762                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2763                         goto out_free_conn;
2764                 }
2765         } else if (unconf) {
2766                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2767                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2768                         status = nfserr_clid_inuse;
2769                         goto out_free_conn;
2770                 }
2771                 status = nfserr_wrong_cred;
2772                 if (!nfsd4_mach_creds_match(unconf, rqstp))
2773                         goto out_free_conn;
2774                 cs_slot = &unconf->cl_cs_slot;
2775                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2776                 if (status) {
2777                         /* an unconfirmed replay returns misordered */
2778                         status = nfserr_seq_misordered;
2779                         goto out_free_conn;
2780                 }
2781                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2782                 if (old) {
2783                         status = mark_client_expired_locked(old);
2784                         if (status) {
2785                                 old = NULL;
2786                                 goto out_free_conn;
2787                         }
2788                 }
2789                 move_to_confirmed(unconf);
2790                 conf = unconf;
2791         } else {
2792                 status = nfserr_stale_clientid;
2793                 goto out_free_conn;
2794         }
2795         status = nfs_ok;
2796         /* Persistent sessions are not supported */
2797         cr_ses->flags &= ~SESSION4_PERSIST;
2798         /* Upshifting from TCP to RDMA is not supported */
2799         cr_ses->flags &= ~SESSION4_RDMA;
2800
2801         init_session(rqstp, new, conf, cr_ses);
2802         nfsd4_get_session_locked(new);
2803
2804         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2805                NFS4_MAX_SESSIONID_LEN);
2806         cs_slot->sl_seqid++;
2807         cr_ses->seqid = cs_slot->sl_seqid;
2808
2809         /* cache solo and embedded create sessions under the client_lock */
2810         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2811         spin_unlock(&nn->client_lock);
2812         /* init connection and backchannel */
2813         nfsd4_init_conn(rqstp, conn, new);
2814         nfsd4_put_session(new);
2815         if (old)
2816                 expire_client(old);
2817         return status;
2818 out_free_conn:
2819         spin_unlock(&nn->client_lock);
2820         free_conn(conn);
2821         if (old)
2822                 expire_client(old);
2823 out_free_session:
2824         __free_session(new);
2825 out_release_drc_mem:
2826         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2827         return status;
2828 }
2829
2830 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2831 {
2832         switch (*dir) {
2833         case NFS4_CDFC4_FORE:
2834         case NFS4_CDFC4_BACK:
2835                 return nfs_ok;
2836         case NFS4_CDFC4_FORE_OR_BOTH:
2837         case NFS4_CDFC4_BACK_OR_BOTH:
2838                 *dir = NFS4_CDFC4_BOTH;
2839                 return nfs_ok;
2840         };
2841         return nfserr_inval;
2842 }
2843
2844 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
2845 {
2846         struct nfsd4_session *session = cstate->session;
2847         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2848         __be32 status;
2849
2850         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2851         if (status)
2852                 return status;
2853         spin_lock(&nn->client_lock);
2854         session->se_cb_prog = bc->bc_cb_program;
2855         session->se_cb_sec = bc->bc_cb_sec;
2856         spin_unlock(&nn->client_lock);
2857
2858         nfsd4_probe_callback(session->se_client);
2859
2860         return nfs_ok;
2861 }
2862
2863 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2864                      struct nfsd4_compound_state *cstate,
2865                      struct nfsd4_bind_conn_to_session *bcts)
2866 {
2867         __be32 status;
2868         struct nfsd4_conn *conn;
2869         struct nfsd4_session *session;
2870         struct net *net = SVC_NET(rqstp);
2871         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2872
2873         if (!nfsd4_last_compound_op(rqstp))
2874                 return nfserr_not_only_op;
2875         spin_lock(&nn->client_lock);
2876         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2877         spin_unlock(&nn->client_lock);
2878         if (!session)
2879                 goto out_no_session;
2880         status = nfserr_wrong_cred;
2881         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2882                 goto out;
2883         status = nfsd4_map_bcts_dir(&bcts->dir);
2884         if (status)
2885                 goto out;
2886         conn = alloc_conn(rqstp, bcts->dir);
2887         status = nfserr_jukebox;
2888         if (!conn)
2889                 goto out;
2890         nfsd4_init_conn(rqstp, conn, session);
2891         status = nfs_ok;
2892 out:
2893         nfsd4_put_session(session);
2894 out_no_session:
2895         return status;
2896 }
2897
2898 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2899 {
2900         if (!session)
2901                 return 0;
2902         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2903 }
2904
2905 __be32
2906 nfsd4_destroy_session(struct svc_rqst *r,
2907                       struct nfsd4_compound_state *cstate,
2908                       struct nfsd4_destroy_session *sessionid)
2909 {
2910         struct nfsd4_session *ses;
2911         __be32 status;
2912         int ref_held_by_me = 0;
2913         struct net *net = SVC_NET(r);
2914         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2915
2916         status = nfserr_not_only_op;
2917         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2918                 if (!nfsd4_last_compound_op(r))
2919                         goto out;
2920                 ref_held_by_me++;
2921         }
2922         dump_sessionid(__func__, &sessionid->sessionid);
2923         spin_lock(&nn->client_lock);
2924         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2925         if (!ses)
2926                 goto out_client_lock;
2927         status = nfserr_wrong_cred;
2928         if (!nfsd4_mach_creds_match(ses->se_client, r))
2929                 goto out_put_session;
2930         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2931         if (status)
2932                 goto out_put_session;
2933         unhash_session(ses);
2934         spin_unlock(&nn->client_lock);
2935
2936         nfsd4_probe_callback_sync(ses->se_client);
2937
2938         spin_lock(&nn->client_lock);
2939         status = nfs_ok;
2940 out_put_session:
2941         nfsd4_put_session_locked(ses);
2942 out_client_lock:
2943         spin_unlock(&nn->client_lock);
2944 out:
2945         return status;
2946 }
2947
2948 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2949 {
2950         struct nfsd4_conn *c;
2951
2952         list_for_each_entry(c, &s->se_conns, cn_persession) {
2953                 if (c->cn_xprt == xpt) {
2954                         return c;
2955                 }
2956         }
2957         return NULL;
2958 }
2959
2960 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2961 {
2962         struct nfs4_client *clp = ses->se_client;
2963         struct nfsd4_conn *c;
2964         __be32 status = nfs_ok;
2965         int ret;
2966
2967         spin_lock(&clp->cl_lock);
2968         c = __nfsd4_find_conn(new->cn_xprt, ses);
2969         if (c)
2970                 goto out_free;
2971         status = nfserr_conn_not_bound_to_session;
2972         if (clp->cl_mach_cred)
2973                 goto out_free;
2974         __nfsd4_hash_conn(new, ses);
2975         spin_unlock(&clp->cl_lock);
2976         ret = nfsd4_register_conn(new);
2977         if (ret)
2978                 /* oops; xprt is already down: */
2979                 nfsd4_conn_lost(&new->cn_xpt_user);
2980         return nfs_ok;
2981 out_free:
2982         spin_unlock(&clp->cl_lock);
2983         free_conn(new);
2984         return status;
2985 }
2986
2987 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2988 {
2989         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2990
2991         return args->opcnt > session->se_fchannel.maxops;
2992 }
2993
2994 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2995                                   struct nfsd4_session *session)
2996 {
2997         struct xdr_buf *xb = &rqstp->rq_arg;
2998
2999         return xb->len > session->se_fchannel.maxreq_sz;
3000 }
3001
3002 __be32
3003 nfsd4_sequence(struct svc_rqst *rqstp,
3004                struct nfsd4_compound_state *cstate,
3005                struct nfsd4_sequence *seq)
3006 {
3007         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3008         struct xdr_stream *xdr = &resp->xdr;
3009         struct nfsd4_session *session;
3010         struct nfs4_client *clp;
3011         struct nfsd4_slot *slot;
3012         struct nfsd4_conn *conn;
3013         __be32 status;
3014         int buflen;
3015         struct net *net = SVC_NET(rqstp);
3016         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3017
3018         if (resp->opcnt != 1)
3019                 return nfserr_sequence_pos;
3020
3021         /*
3022          * Will be either used or freed by nfsd4_sequence_check_conn
3023          * below.
3024          */
3025         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3026         if (!conn)
3027                 return nfserr_jukebox;
3028
3029         spin_lock(&nn->client_lock);
3030         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3031         if (!session)
3032                 goto out_no_session;
3033         clp = session->se_client;
3034
3035         status = nfserr_too_many_ops;
3036         if (nfsd4_session_too_many_ops(rqstp, session))
3037                 goto out_put_session;
3038
3039         status = nfserr_req_too_big;
3040         if (nfsd4_request_too_big(rqstp, session))
3041                 goto out_put_session;
3042
3043         status = nfserr_badslot;
3044         if (seq->slotid >= session->se_fchannel.maxreqs)
3045                 goto out_put_session;
3046
3047         slot = session->se_slots[seq->slotid];
3048         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3049
3050         /* We do not negotiate the number of slots yet, so set the
3051          * maxslots to the session maxreqs which is used to encode
3052          * sr_highest_slotid and the sr_target_slot id to maxslots */
3053         seq->maxslots = session->se_fchannel.maxreqs;
3054
3055         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3056                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3057         if (status == nfserr_replay_cache) {
3058                 status = nfserr_seq_misordered;
3059                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3060                         goto out_put_session;
3061                 cstate->slot = slot;
3062                 cstate->session = session;
3063                 cstate->clp = clp;
3064                 /* Return the cached reply status and set cstate->status
3065                  * for nfsd4_proc_compound processing */
3066                 status = nfsd4_replay_cache_entry(resp, seq);
3067                 cstate->status = nfserr_replay_cache;
3068                 goto out;
3069         }
3070         if (status)
3071                 goto out_put_session;
3072
3073         status = nfsd4_sequence_check_conn(conn, session);
3074         conn = NULL;
3075         if (status)
3076                 goto out_put_session;
3077
3078         buflen = (seq->cachethis) ?
3079                         session->se_fchannel.maxresp_cached :
3080                         session->se_fchannel.maxresp_sz;
3081         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3082                                     nfserr_rep_too_big;
3083         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3084                 goto out_put_session;
3085         svc_reserve(rqstp, buflen);
3086
3087         status = nfs_ok;
3088         /* Success! bump slot seqid */
3089         slot->sl_seqid = seq->seqid;
3090         slot->sl_flags |= NFSD4_SLOT_INUSE;
3091         if (seq->cachethis)
3092                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3093         else
3094                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3095
3096         cstate->slot = slot;
3097         cstate->session = session;
3098         cstate->clp = clp;
3099
3100 out:
3101         switch (clp->cl_cb_state) {
3102         case NFSD4_CB_DOWN:
3103                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3104                 break;
3105         case NFSD4_CB_FAULT:
3106                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3107                 break;
3108         default:
3109                 seq->status_flags = 0;
3110         }
3111         if (!list_empty(&clp->cl_revoked))
3112                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3113 out_no_session:
3114         if (conn)
3115                 free_conn(conn);
3116         spin_unlock(&nn->client_lock);
3117         return status;
3118 out_put_session:
3119         nfsd4_put_session_locked(session);
3120         goto out_no_session;
3121 }
3122
3123 void
3124 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3125 {
3126         struct nfsd4_compound_state *cs = &resp->cstate;
3127
3128         if (nfsd4_has_session(cs)) {
3129                 if (cs->status != nfserr_replay_cache) {
3130                         nfsd4_store_cache_entry(resp);
3131                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3132                 }
3133                 /* Drop session reference that was taken in nfsd4_sequence() */
3134                 nfsd4_put_session(cs->session);
3135         } else if (cs->clp)
3136                 put_client_renew(cs->clp);
3137 }
3138
3139 __be32
3140 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
3141 {
3142         struct nfs4_client *conf, *unconf;
3143         struct nfs4_client *clp = NULL;
3144         __be32 status = 0;
3145         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3146
3147         spin_lock(&nn->client_lock);
3148         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3149         conf = find_confirmed_client(&dc->clientid, true, nn);
3150         WARN_ON_ONCE(conf && unconf);
3151
3152         if (conf) {
3153                 if (client_has_state(conf)) {
3154                         status = nfserr_clientid_busy;
3155                         goto out;
3156                 }
3157                 status = mark_client_expired_locked(conf);
3158                 if (status)
3159                         goto out;
3160                 clp = conf;
3161         } else if (unconf)
3162                 clp = unconf;
3163         else {
3164                 status = nfserr_stale_clientid;
3165                 goto out;
3166         }
3167         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3168                 clp = NULL;
3169                 status = nfserr_wrong_cred;
3170                 goto out;
3171         }
3172         unhash_client_locked(clp);
3173 out:
3174         spin_unlock(&nn->client_lock);
3175         if (clp)
3176                 expire_client(clp);
3177         return status;
3178 }
3179
3180 __be32
3181 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
3182 {
3183         __be32 status = 0;
3184
3185         if (rc->rca_one_fs) {
3186                 if (!cstate->current_fh.fh_dentry)
3187                         return nfserr_nofilehandle;
3188                 /*
3189                  * We don't take advantage of the rca_one_fs case.
3190                  * That's OK, it's optional, we can safely ignore it.
3191                  */
3192                 return nfs_ok;
3193         }
3194
3195         status = nfserr_complete_already;
3196         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3197                              &cstate->session->se_client->cl_flags))
3198                 goto out;
3199
3200         status = nfserr_stale_clientid;
3201         if (is_client_expired(cstate->session->se_client))
3202                 /*
3203                  * The following error isn't really legal.
3204                  * But we only get here if the client just explicitly
3205                  * destroyed the client.  Surely it no longer cares what
3206                  * error it gets back on an operation for the dead
3207                  * client.
3208                  */
3209                 goto out;
3210
3211         status = nfs_ok;
3212         nfsd4_client_record_create(cstate->session->se_client);
3213 out:
3214         return status;
3215 }
3216
3217 __be32
3218 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3219                   struct nfsd4_setclientid *setclid)
3220 {
3221         struct xdr_netobj       clname = setclid->se_name;
3222         nfs4_verifier           clverifier = setclid->se_verf;
3223         struct nfs4_client      *conf, *new;
3224         struct nfs4_client      *unconf = NULL;
3225         __be32                  status;
3226         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3227
3228         new = create_client(clname, rqstp, &clverifier);
3229         if (new == NULL)
3230                 return nfserr_jukebox;
3231         /* Cases below refer to rfc 3530 section 14.2.33: */
3232         spin_lock(&nn->client_lock);
3233         conf = find_confirmed_client_by_name(&clname, nn);
3234         if (conf && client_has_state(conf)) {
3235                 /* case 0: */
3236                 status = nfserr_clid_inuse;
3237                 if (clp_used_exchangeid(conf))
3238                         goto out;
3239                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3240                         char addr_str[INET6_ADDRSTRLEN];
3241                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3242                                  sizeof(addr_str));
3243                         dprintk("NFSD: setclientid: string in use by client "
3244                                 "at %s\n", addr_str);
3245                         goto out;
3246                 }
3247         }
3248         unconf = find_unconfirmed_client_by_name(&clname, nn);
3249         if (unconf)
3250                 unhash_client_locked(unconf);
3251         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3252                 /* case 1: probable callback update */
3253                 copy_clid(new, conf);
3254                 gen_confirm(new, nn);
3255         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3256                 gen_clid(new, nn);
3257         new->cl_minorversion = 0;
3258         gen_callback(new, setclid, rqstp);
3259         add_to_unconfirmed(new);
3260         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3261         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3262         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3263         new = NULL;
3264         status = nfs_ok;
3265 out:
3266         spin_unlock(&nn->client_lock);
3267         if (new)
3268                 free_client(new);
3269         if (unconf)
3270                 expire_client(unconf);
3271         return status;
3272 }
3273
3274
3275 __be32
3276 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3277                          struct nfsd4_compound_state *cstate,
3278                          struct nfsd4_setclientid_confirm *setclientid_confirm)
3279 {
3280         struct nfs4_client *conf, *unconf;
3281         struct nfs4_client *old = NULL;
3282         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3283         clientid_t * clid = &setclientid_confirm->sc_clientid;
3284         __be32 status;
3285         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3286
3287         if (STALE_CLIENTID(clid, nn))
3288                 return nfserr_stale_clientid;
3289
3290         spin_lock(&nn->client_lock);
3291         conf = find_confirmed_client(clid, false, nn);
3292         unconf = find_unconfirmed_client(clid, false, nn);
3293         /*
3294          * We try hard to give out unique clientid's, so if we get an
3295          * attempt to confirm the same clientid with a different cred,
3296          * the client may be buggy; this should never happen.
3297          *
3298          * Nevertheless, RFC 7530 recommends INUSE for this case:
3299          */
3300         status = nfserr_clid_inuse;
3301         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3302                 goto out;
3303         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3304                 goto out;
3305         /* cases below refer to rfc 3530 section 14.2.34: */
3306         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3307                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3308                         /* case 2: probable retransmit */
3309                         status = nfs_ok;
3310                 } else /* case 4: client hasn't noticed we rebooted yet? */
3311                         status = nfserr_stale_clientid;
3312                 goto out;
3313         }
3314         status = nfs_ok;
3315         if (conf) { /* case 1: callback update */
3316                 old = unconf;
3317                 unhash_client_locked(old);
3318                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3319         } else { /* case 3: normal case; new or rebooted client */
3320                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3321                 if (old) {
3322                         status = nfserr_clid_inuse;
3323                         if (client_has_state(old)
3324                                         && !same_creds(&unconf->cl_cred,
3325                                                         &old->cl_cred))
3326                                 goto out;
3327                         status = mark_client_expired_locked(old);
3328                         if (status) {
3329                                 old = NULL;
3330                                 goto out;
3331                         }
3332                 }
3333                 move_to_confirmed(unconf);
3334                 conf = unconf;
3335         }
3336         get_client_locked(conf);
3337         spin_unlock(&nn->client_lock);
3338         nfsd4_probe_callback(conf);
3339         spin_lock(&nn->client_lock);
3340         put_client_renew_locked(conf);
3341 out:
3342         spin_unlock(&nn->client_lock);
3343         if (old)
3344                 expire_client(old);
3345         return status;
3346 }
3347
3348 static struct nfs4_file *nfsd4_alloc_file(void)
3349 {
3350         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3351 }
3352
3353 /* OPEN Share state helper functions */
3354 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3355                                 struct nfs4_file *fp)
3356 {
3357         lockdep_assert_held(&state_lock);
3358
3359         atomic_set(&fp->fi_ref, 1);
3360         spin_lock_init(&fp->fi_lock);
3361         INIT_LIST_HEAD(&fp->fi_stateids);
3362         INIT_LIST_HEAD(&fp->fi_delegations);
3363         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3364         fh_copy_shallow(&fp->fi_fhandle, fh);
3365         fp->fi_deleg_file = NULL;
3366         fp->fi_had_conflict = false;
3367         fp->fi_share_deny = 0;
3368         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3369         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3370 #ifdef CONFIG_NFSD_PNFS
3371         INIT_LIST_HEAD(&fp->fi_lo_states);
3372         atomic_set(&fp->fi_lo_recalls, 0);
3373 #endif
3374         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3375 }
3376
3377 void
3378 nfsd4_free_slabs(void)
3379 {
3380         kmem_cache_destroy(odstate_slab);
3381         kmem_cache_destroy(openowner_slab);
3382         kmem_cache_destroy(lockowner_slab);
3383         kmem_cache_destroy(file_slab);
3384         kmem_cache_destroy(stateid_slab);
3385         kmem_cache_destroy(deleg_slab);
3386 }
3387
3388 int
3389 nfsd4_init_slabs(void)
3390 {
3391         openowner_slab = kmem_cache_create("nfsd4_openowners",
3392                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3393         if (openowner_slab == NULL)
3394                 goto out;
3395         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3396                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3397         if (lockowner_slab == NULL)
3398                 goto out_free_openowner_slab;
3399         file_slab = kmem_cache_create("nfsd4_files",
3400                         sizeof(struct nfs4_file), 0, 0, NULL);
3401         if (file_slab == NULL)
3402                 goto out_free_lockowner_slab;
3403         stateid_slab = kmem_cache_create("nfsd4_stateids",
3404                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3405         if (stateid_slab == NULL)
3406                 goto out_free_file_slab;
3407         deleg_slab = kmem_cache_create("nfsd4_delegations",
3408                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3409         if (deleg_slab == NULL)
3410                 goto out_free_stateid_slab;
3411         odstate_slab = kmem_cache_create("nfsd4_odstate",
3412                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3413         if (odstate_slab == NULL)
3414                 goto out_free_deleg_slab;
3415         return 0;
3416
3417 out_free_deleg_slab:
3418         kmem_cache_destroy(deleg_slab);
3419 out_free_stateid_slab:
3420         kmem_cache_destroy(stateid_slab);
3421 out_free_file_slab:
3422         kmem_cache_destroy(file_slab);
3423 out_free_lockowner_slab:
3424         kmem_cache_destroy(lockowner_slab);
3425 out_free_openowner_slab:
3426         kmem_cache_destroy(openowner_slab);
3427 out:
3428         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3429         return -ENOMEM;
3430 }
3431
3432 static void init_nfs4_replay(struct nfs4_replay *rp)
3433 {
3434         rp->rp_status = nfserr_serverfault;
3435         rp->rp_buflen = 0;
3436         rp->rp_buf = rp->rp_ibuf;
3437         mutex_init(&rp->rp_mutex);
3438 }
3439
3440 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3441                 struct nfs4_stateowner *so)
3442 {
3443         if (!nfsd4_has_session(cstate)) {
3444                 mutex_lock(&so->so_replay.rp_mutex);
3445                 cstate->replay_owner = nfs4_get_stateowner(so);
3446         }
3447 }
3448
3449 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3450 {
3451         struct nfs4_stateowner *so = cstate->replay_owner;
3452
3453         if (so != NULL) {
3454                 cstate->replay_owner = NULL;
3455                 mutex_unlock(&so->so_replay.rp_mutex);
3456                 nfs4_put_stateowner(so);
3457         }
3458 }
3459
3460 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3461 {
3462         struct nfs4_stateowner *sop;
3463
3464         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3465         if (!sop)
3466                 return NULL;
3467
3468         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3469         if (!sop->so_owner.data) {
3470                 kmem_cache_free(slab, sop);
3471                 return NULL;
3472         }
3473         sop->so_owner.len = owner->len;
3474
3475         INIT_LIST_HEAD(&sop->so_stateids);
3476         sop->so_client = clp;
3477         init_nfs4_replay(&sop->so_replay);
3478         atomic_set(&sop->so_count, 1);
3479         return sop;
3480 }
3481
3482 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3483 {
3484         lockdep_assert_held(&clp->cl_lock);
3485
3486         list_add(&oo->oo_owner.so_strhash,
3487                  &clp->cl_ownerstr_hashtbl[strhashval]);
3488         list_add(&oo->oo_perclient, &clp->cl_openowners);
3489 }
3490
3491 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3492 {
3493         unhash_openowner_locked(openowner(so));
3494 }
3495
3496 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3497 {
3498         struct nfs4_openowner *oo = openowner(so);
3499
3500         kmem_cache_free(openowner_slab, oo);
3501 }
3502
3503 static const struct nfs4_stateowner_operations openowner_ops = {
3504         .so_unhash =    nfs4_unhash_openowner,
3505         .so_free =      nfs4_free_openowner,
3506 };
3507
3508 static struct nfs4_ol_stateid *
3509 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3510 {
3511         struct nfs4_ol_stateid *local, *ret = NULL;
3512         struct nfs4_openowner *oo = open->op_openowner;
3513
3514         lockdep_assert_held(&fp->fi_lock);
3515
3516         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3517                 /* ignore lock owners */
3518                 if (local->st_stateowner->so_is_open_owner == 0)
3519                         continue;
3520                 if (local->st_stateowner == &oo->oo_owner) {
3521                         ret = local;
3522                         atomic_inc(&ret->st_stid.sc_count);
3523                         break;
3524                 }
3525         }
3526         return ret;
3527 }
3528
3529 static struct nfs4_openowner *
3530 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3531                            struct nfsd4_compound_state *cstate)
3532 {
3533         struct nfs4_client *clp = cstate->clp;
3534         struct nfs4_openowner *oo, *ret;
3535
3536         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3537         if (!oo)
3538                 return NULL;
3539         oo->oo_owner.so_ops = &openowner_ops;
3540         oo->oo_owner.so_is_open_owner = 1;
3541         oo->oo_owner.so_seqid = open->op_seqid;
3542         oo->oo_flags = 0;
3543         if (nfsd4_has_session(cstate))
3544                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3545         oo->oo_time = 0;
3546         oo->oo_last_closed_stid = NULL;
3547         INIT_LIST_HEAD(&oo->oo_close_lru);
3548         spin_lock(&clp->cl_lock);
3549         ret = find_openstateowner_str_locked(strhashval, open, clp);
3550         if (ret == NULL) {
3551                 hash_openowner(oo, clp, strhashval);
3552                 ret = oo;
3553         } else
3554                 nfs4_free_stateowner(&oo->oo_owner);
3555
3556         spin_unlock(&clp->cl_lock);
3557         return ret;
3558 }
3559
3560 static struct nfs4_ol_stateid *
3561 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3562 {
3563
3564         struct nfs4_openowner *oo = open->op_openowner;
3565         struct nfs4_ol_stateid *retstp = NULL;
3566         struct nfs4_ol_stateid *stp;
3567
3568         stp = open->op_stp;
3569         /* We are moving these outside of the spinlocks to avoid the warnings */
3570         mutex_init(&stp->st_mutex);
3571         mutex_lock(&stp->st_mutex);
3572
3573         spin_lock(&oo->oo_owner.so_client->cl_lock);
3574         spin_lock(&fp->fi_lock);
3575
3576         retstp = nfsd4_find_existing_open(fp, open);
3577         if (retstp)
3578                 goto out_unlock;
3579
3580         open->op_stp = NULL;
3581         atomic_inc(&stp->st_stid.sc_count);
3582         stp->st_stid.sc_type = NFS4_OPEN_STID;
3583         INIT_LIST_HEAD(&stp->st_locks);
3584         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3585         get_nfs4_file(fp);
3586         stp->st_stid.sc_file = fp;
3587         stp->st_access_bmap = 0;
3588         stp->st_deny_bmap = 0;
3589         stp->st_openstp = NULL;
3590         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3591         list_add(&stp->st_perfile, &fp->fi_stateids);
3592
3593 out_unlock:
3594         spin_unlock(&fp->fi_lock);
3595         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3596         if (retstp) {
3597                 mutex_lock(&retstp->st_mutex);
3598                 /* To keep mutex tracking happy */
3599                 mutex_unlock(&stp->st_mutex);
3600                 stp = retstp;
3601         }
3602         return stp;
3603 }
3604
3605 /*
3606  * In the 4.0 case we need to keep the owners around a little while to handle
3607  * CLOSE replay. We still do need to release any file access that is held by
3608  * them before returning however.
3609  */
3610 static void
3611 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3612 {
3613         struct nfs4_ol_stateid *last;
3614         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3615         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3616                                                 nfsd_net_id);
3617
3618         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3619
3620         /*
3621          * We know that we hold one reference via nfsd4_close, and another
3622          * "persistent" reference for the client. If the refcount is higher
3623          * than 2, then there are still calls in progress that are using this
3624          * stateid. We can't put the sc_file reference until they are finished.
3625          * Wait for the refcount to drop to 2. Since it has been unhashed,
3626          * there should be no danger of the refcount going back up again at
3627          * this point.
3628          */
3629         wait_event(close_wq, atomic_read(&s->st_stid.sc_count) == 2);
3630
3631         release_all_access(s);
3632         if (s->st_stid.sc_file) {
3633                 put_nfs4_file(s->st_stid.sc_file);
3634                 s->st_stid.sc_file = NULL;
3635         }
3636
3637         spin_lock(&nn->client_lock);
3638         last = oo->oo_last_closed_stid;
3639         oo->oo_last_closed_stid = s;
3640         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3641         oo->oo_time = get_seconds();
3642         spin_unlock(&nn->client_lock);
3643         if (last)
3644                 nfs4_put_stid(&last->st_stid);
3645 }
3646
3647 /* search file_hashtbl[] for file */
3648 static struct nfs4_file *
3649 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3650 {
3651         struct nfs4_file *fp;
3652
3653         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3654                 if (fh_match(&fp->fi_fhandle, fh)) {
3655                         if (atomic_inc_not_zero(&fp->fi_ref))
3656                                 return fp;
3657                 }
3658         }
3659         return NULL;
3660 }
3661
3662 struct nfs4_file *
3663 find_file(struct knfsd_fh *fh)
3664 {
3665         struct nfs4_file *fp;
3666         unsigned int hashval = file_hashval(fh);
3667
3668         rcu_read_lock();
3669         fp = find_file_locked(fh, hashval);
3670         rcu_read_unlock();
3671         return fp;
3672 }
3673
3674 static struct nfs4_file *
3675 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3676 {
3677         struct nfs4_file *fp;
3678         unsigned int hashval = file_hashval(fh);
3679
3680         rcu_read_lock();
3681         fp = find_file_locked(fh, hashval);
3682         rcu_read_unlock();
3683         if (fp)
3684                 return fp;
3685
3686         spin_lock(&state_lock);
3687         fp = find_file_locked(fh, hashval);
3688         if (likely(fp == NULL)) {
3689                 nfsd4_init_file(fh, hashval, new);
3690                 fp = new;
3691         }
3692         spin_unlock(&state_lock);
3693
3694         return fp;
3695 }
3696
3697 /*
3698  * Called to check deny when READ with all zero stateid or
3699  * WRITE with all zero or all one stateid
3700  */
3701 static __be32
3702 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3703 {
3704         struct nfs4_file *fp;
3705         __be32 ret = nfs_ok;
3706
3707         fp = find_file(&current_fh->fh_handle);
3708         if (!fp)
3709                 return ret;
3710         /* Check for conflicting share reservations */
3711         spin_lock(&fp->fi_lock);
3712         if (fp->fi_share_deny & deny_type)
3713                 ret = nfserr_locked;
3714         spin_unlock(&fp->fi_lock);
3715         put_nfs4_file(fp);
3716         return ret;
3717 }
3718
3719 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3720 {
3721         struct nfs4_delegation *dp = cb_to_delegation(cb);
3722         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3723                                           nfsd_net_id);
3724
3725         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3726
3727         /*
3728          * We can't do this in nfsd_break_deleg_cb because it is
3729          * already holding inode->i_lock.
3730          *
3731          * If the dl_time != 0, then we know that it has already been
3732          * queued for a lease break. Don't queue it again.
3733          */
3734         spin_lock(&state_lock);
3735         if (dp->dl_time == 0) {
3736                 dp->dl_time = get_seconds();
3737                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3738         }
3739         spin_unlock(&state_lock);
3740 }
3741
3742 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3743                 struct rpc_task *task)
3744 {
3745         struct nfs4_delegation *dp = cb_to_delegation(cb);
3746
3747         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3748                 return 1;
3749
3750         switch (task->tk_status) {
3751         case 0:
3752                 return 1;
3753         case -EBADHANDLE:
3754         case -NFS4ERR_BAD_STATEID:
3755                 /*
3756                  * Race: client probably got cb_recall before open reply
3757                  * granting delegation.
3758                  */
3759                 if (dp->dl_retries--) {
3760                         rpc_delay(task, 2 * HZ);
3761                         return 0;
3762                 }
3763                 /*FALLTHRU*/
3764         default:
3765                 return -1;
3766         }
3767 }
3768
3769 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3770 {
3771         struct nfs4_delegation *dp = cb_to_delegation(cb);
3772
3773         nfs4_put_stid(&dp->dl_stid);
3774 }
3775
3776 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3777         .prepare        = nfsd4_cb_recall_prepare,
3778         .done           = nfsd4_cb_recall_done,
3779         .release        = nfsd4_cb_recall_release,
3780 };
3781
3782 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3783 {
3784         /*
3785          * We're assuming the state code never drops its reference
3786          * without first removing the lease.  Since we're in this lease
3787          * callback (and since the lease code is serialized by the kernel
3788          * lock) we know the server hasn't removed the lease yet, we know
3789          * it's safe to take a reference.
3790          */
3791         atomic_inc(&dp->dl_stid.sc_count);
3792         nfsd4_run_cb(&dp->dl_recall);
3793 }
3794
3795 /* Called from break_lease() with i_lock held. */
3796 static bool
3797 nfsd_break_deleg_cb(struct file_lock *fl)
3798 {
3799         bool ret = false;
3800         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3801         struct nfs4_delegation *dp;
3802
3803         if (!fp) {
3804                 WARN(1, "(%p)->fl_owner NULL\n", fl);
3805                 return ret;
3806         }
3807         if (fp->fi_had_conflict) {
3808                 WARN(1, "duplicate break on %p\n", fp);
3809                 return ret;
3810         }
3811         /*
3812          * We don't want the locks code to timeout the lease for us;
3813          * we'll remove it ourself if a delegation isn't returned
3814          * in time:
3815          */
3816         fl->fl_break_time = 0;
3817
3818         spin_lock(&fp->fi_lock);
3819         fp->fi_had_conflict = true;
3820         /*
3821          * If there are no delegations on the list, then return true
3822          * so that the lease code will go ahead and delete it.
3823          */
3824         if (list_empty(&fp->fi_delegations))
3825                 ret = true;
3826         else
3827                 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3828                         nfsd_break_one_deleg(dp);
3829         spin_unlock(&fp->fi_lock);
3830         return ret;
3831 }
3832
3833 static int
3834 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3835                      struct list_head *dispose)
3836 {
3837         if (arg & F_UNLCK)
3838                 return lease_modify(onlist, arg, dispose);
3839         else
3840                 return -EAGAIN;
3841 }
3842
3843 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3844         .lm_break = nfsd_break_deleg_cb,
3845         .lm_change = nfsd_change_deleg_cb,
3846 };
3847
3848 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3849 {
3850         if (nfsd4_has_session(cstate))
3851                 return nfs_ok;
3852         if (seqid == so->so_seqid - 1)
3853                 return nfserr_replay_me;
3854         if (seqid == so->so_seqid)
3855                 return nfs_ok;
3856         return nfserr_bad_seqid;
3857 }
3858
3859 static __be32 lookup_clientid(clientid_t *clid,
3860                 struct nfsd4_compound_state *cstate,
3861                 struct nfsd_net *nn)
3862 {
3863         struct nfs4_client *found;
3864
3865         if (cstate->clp) {
3866                 found = cstate->clp;
3867                 if (!same_clid(&found->cl_clientid, clid))
3868                         return nfserr_stale_clientid;
3869                 return nfs_ok;
3870         }
3871
3872         if (STALE_CLIENTID(clid, nn))
3873                 return nfserr_stale_clientid;
3874
3875         /*
3876          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
3877          * cached already then we know this is for is for v4.0 and "sessions"
3878          * will be false.
3879          */
3880         WARN_ON_ONCE(cstate->session);
3881         spin_lock(&nn->client_lock);
3882         found = find_confirmed_client(clid, false, nn);
3883         if (!found) {
3884                 spin_unlock(&nn->client_lock);
3885                 return nfserr_expired;
3886         }
3887         atomic_inc(&found->cl_refcount);
3888         spin_unlock(&nn->client_lock);
3889
3890         /* Cache the nfs4_client in cstate! */
3891         cstate->clp = found;
3892         return nfs_ok;
3893 }
3894
3895 __be32
3896 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
3897                     struct nfsd4_open *open, struct nfsd_net *nn)
3898 {
3899         clientid_t *clientid = &open->op_clientid;
3900         struct nfs4_client *clp = NULL;
3901         unsigned int strhashval;
3902         struct nfs4_openowner *oo = NULL;
3903         __be32 status;
3904
3905         if (STALE_CLIENTID(&open->op_clientid, nn))
3906                 return nfserr_stale_clientid;
3907         /*
3908          * In case we need it later, after we've already created the
3909          * file and don't want to risk a further failure:
3910          */
3911         open->op_file = nfsd4_alloc_file();
3912         if (open->op_file == NULL)
3913                 return nfserr_jukebox;
3914
3915         status = lookup_clientid(clientid, cstate, nn);
3916         if (status)
3917                 return status;
3918         clp = cstate->clp;
3919
3920         strhashval = ownerstr_hashval(&open->op_owner);
3921         oo = find_openstateowner_str(strhashval, open, clp);
3922         open->op_openowner = oo;
3923         if (!oo) {
3924                 goto new_owner;
3925         }
3926         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
3927                 /* Replace unconfirmed owners without checking for replay. */
3928                 release_openowner(oo);
3929                 open->op_openowner = NULL;
3930                 goto new_owner;
3931         }
3932         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
3933         if (status)
3934                 return status;
3935         goto alloc_stateid;
3936 new_owner:
3937         oo = alloc_init_open_stateowner(strhashval, open, cstate);
3938         if (oo == NULL)
3939                 return nfserr_jukebox;
3940         open->op_openowner = oo;
3941 alloc_stateid:
3942         open->op_stp = nfs4_alloc_open_stateid(clp);
3943         if (!open->op_stp)
3944                 return nfserr_jukebox;
3945
3946         if (nfsd4_has_session(cstate) &&
3947             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
3948                 open->op_odstate = alloc_clnt_odstate(clp);
3949                 if (!open->op_odstate)
3950                         return nfserr_jukebox;
3951         }
3952
3953         return nfs_ok;
3954 }
3955
3956 static inline __be32
3957 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
3958 {
3959         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
3960                 return nfserr_openmode;
3961         else
3962                 return nfs_ok;
3963 }
3964
3965 static int share_access_to_flags(u32 share_access)
3966 {
3967         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
3968 }
3969
3970 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
3971 {
3972         struct nfs4_stid *ret;
3973
3974         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
3975         if (!ret)
3976                 return NULL;
3977         return delegstateid(ret);
3978 }
3979
3980 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
3981 {
3982         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
3983                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
3984 }
3985
3986 static __be32
3987 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
3988                 struct nfs4_delegation **dp)
3989 {
3990         int flags;
3991         __be32 status = nfserr_bad_stateid;
3992         struct nfs4_delegation *deleg;
3993
3994         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
3995         if (deleg == NULL)
3996                 goto out;
3997         flags = share_access_to_flags(open->op_share_access);
3998         status = nfs4_check_delegmode(deleg, flags);
3999         if (status) {
4000                 nfs4_put_stid(&deleg->dl_stid);
4001                 goto out;
4002         }
4003         *dp = deleg;
4004 out:
4005         if (!nfsd4_is_deleg_cur(open))
4006                 return nfs_ok;
4007         if (status)
4008                 return status;
4009         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4010         return nfs_ok;
4011 }
4012
4013 static inline int nfs4_access_to_access(u32 nfs4_access)
4014 {
4015         int flags = 0;
4016
4017         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4018                 flags |= NFSD_MAY_READ;
4019         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4020                 flags |= NFSD_MAY_WRITE;
4021         return flags;
4022 }
4023
4024 static inline __be32
4025 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4026                 struct nfsd4_open *open)
4027 {
4028         struct iattr iattr = {
4029                 .ia_valid = ATTR_SIZE,
4030                 .ia_size = 0,
4031         };
4032         if (!open->op_truncate)
4033                 return 0;
4034         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4035                 return nfserr_inval;
4036         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4037 }
4038
4039 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4040                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4041                 struct nfsd4_open *open)
4042 {
4043         struct file *filp = NULL;
4044         __be32 status;
4045         int oflag = nfs4_access_to_omode(open->op_share_access);
4046         int access = nfs4_access_to_access(open->op_share_access);
4047         unsigned char old_access_bmap, old_deny_bmap;
4048
4049         spin_lock(&fp->fi_lock);
4050
4051         /*
4052          * Are we trying to set a deny mode that would conflict with
4053          * current access?
4054          */
4055         status = nfs4_file_check_deny(fp, open->op_share_deny);
4056         if (status != nfs_ok) {
4057                 spin_unlock(&fp->fi_lock);
4058                 goto out;
4059         }
4060
4061         /* set access to the file */
4062         status = nfs4_file_get_access(fp, open->op_share_access);
4063         if (status != nfs_ok) {
4064                 spin_unlock(&fp->fi_lock);
4065                 goto out;
4066         }
4067
4068         /* Set access bits in stateid */
4069         old_access_bmap = stp->st_access_bmap;
4070         set_access(open->op_share_access, stp);
4071
4072         /* Set new deny mask */
4073         old_deny_bmap = stp->st_deny_bmap;
4074         set_deny(open->op_share_deny, stp);
4075         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4076
4077         if (!fp->fi_fds[oflag]) {
4078                 spin_unlock(&fp->fi_lock);
4079                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4080                 if (status)
4081                         goto out_put_access;
4082                 spin_lock(&fp->fi_lock);
4083                 if (!fp->fi_fds[oflag]) {
4084                         fp->fi_fds[oflag] = filp;
4085                         filp = NULL;
4086                 }
4087         }
4088         spin_unlock(&fp->fi_lock);
4089         if (filp)
4090                 fput(filp);
4091
4092         status = nfsd4_truncate(rqstp, cur_fh, open);
4093         if (status)
4094                 goto out_put_access;
4095 out:
4096         return status;
4097 out_put_access:
4098         stp->st_access_bmap = old_access_bmap;
4099         nfs4_file_put_access(fp, open->op_share_access);
4100         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4101         goto out;
4102 }
4103
4104 static __be32
4105 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp, struct nfsd4_open *open)
4106 {
4107         __be32 status;
4108         unsigned char old_deny_bmap = stp->st_deny_bmap;
4109
4110         if (!test_access(open->op_share_access, stp))
4111                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4112
4113         /* test and set deny mode */
4114         spin_lock(&fp->fi_lock);
4115         status = nfs4_file_check_deny(fp, open->op_share_deny);
4116         if (status == nfs_ok) {
4117                 set_deny(open->op_share_deny, stp);
4118                 fp->fi_share_deny |=
4119                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4120         }
4121         spin_unlock(&fp->fi_lock);
4122
4123         if (status != nfs_ok)
4124                 return status;
4125
4126         status = nfsd4_truncate(rqstp, cur_fh, open);
4127         if (status != nfs_ok)
4128                 reset_union_bmap_deny(old_deny_bmap, stp);
4129         return status;
4130 }
4131
4132 /* Should we give out recallable state?: */
4133 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4134 {
4135         if (clp->cl_cb_state == NFSD4_CB_UP)
4136                 return true;
4137         /*
4138          * In the sessions case, since we don't have to establish a
4139          * separate connection for callbacks, we assume it's OK
4140          * until we hear otherwise:
4141          */
4142         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4143 }
4144
4145 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
4146 {
4147         struct file_lock *fl;
4148
4149         fl = locks_alloc_lock();
4150         if (!fl)
4151                 return NULL;
4152         fl->fl_lmops = &nfsd_lease_mng_ops;
4153         fl->fl_flags = FL_DELEG;
4154         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4155         fl->fl_end = OFFSET_MAX;
4156         fl->fl_owner = (fl_owner_t)fp;
4157         fl->fl_pid = current->tgid;
4158         return fl;
4159 }
4160
4161 /**
4162  * nfs4_setlease - Obtain a delegation by requesting lease from vfs layer
4163  * @dp:   a pointer to the nfs4_delegation we're adding.
4164  *
4165  * Return:
4166  *      On success: Return code will be 0 on success.
4167  *
4168  *      On error: -EAGAIN if there was an existing delegation.
4169  *                 nonzero if there is an error in other cases.
4170  *
4171  */
4172
4173 static int nfs4_setlease(struct nfs4_delegation *dp)
4174 {
4175         struct nfs4_file *fp = dp->dl_stid.sc_file;
4176         struct file_lock *fl;
4177         struct file *filp;
4178         int status = 0;
4179
4180         fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
4181         if (!fl)
4182                 return -ENOMEM;
4183         filp = find_readable_file(fp);
4184         if (!filp) {
4185                 /* We should always have a readable file here */
4186                 WARN_ON_ONCE(1);
4187                 locks_free_lock(fl);
4188                 return -EBADF;
4189         }
4190         fl->fl_file = filp;
4191         status = vfs_setlease(filp, fl->fl_type, &fl, NULL);
4192         if (fl)
4193                 locks_free_lock(fl);
4194         if (status)
4195                 goto out_fput;
4196         spin_lock(&state_lock);
4197         spin_lock(&fp->fi_lock);
4198         /* Did the lease get broken before we took the lock? */
4199         status = -EAGAIN;
4200         if (fp->fi_had_conflict)
4201                 goto out_unlock;
4202         /* Race breaker */
4203         if (fp->fi_deleg_file) {
4204                 status = hash_delegation_locked(dp, fp);
4205                 goto out_unlock;
4206         }
4207         fp->fi_deleg_file = filp;
4208         fp->fi_delegees = 0;
4209         status = hash_delegation_locked(dp, fp);
4210         spin_unlock(&fp->fi_lock);
4211         spin_unlock(&state_lock);
4212         if (status) {
4213                 /* Should never happen, this is a new fi_deleg_file  */
4214                 WARN_ON_ONCE(1);
4215                 goto out_fput;
4216         }
4217         return 0;
4218 out_unlock:
4219         spin_unlock(&fp->fi_lock);
4220         spin_unlock(&state_lock);
4221 out_fput:
4222         fput(filp);
4223         return status;
4224 }
4225
4226 static struct nfs4_delegation *
4227 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4228                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4229 {
4230         int status;
4231         struct nfs4_delegation *dp;
4232
4233         if (fp->fi_had_conflict)
4234                 return ERR_PTR(-EAGAIN);
4235
4236         spin_lock(&state_lock);
4237         spin_lock(&fp->fi_lock);
4238         status = nfs4_get_existing_delegation(clp, fp);
4239         spin_unlock(&fp->fi_lock);
4240         spin_unlock(&state_lock);
4241
4242         if (status)
4243                 return ERR_PTR(status);
4244
4245         dp = alloc_init_deleg(clp, fh, odstate);
4246         if (!dp)
4247                 return ERR_PTR(-ENOMEM);
4248
4249         get_nfs4_file(fp);
4250         spin_lock(&state_lock);
4251         spin_lock(&fp->fi_lock);
4252         dp->dl_stid.sc_file = fp;
4253         if (!fp->fi_deleg_file) {
4254                 spin_unlock(&fp->fi_lock);
4255                 spin_unlock(&state_lock);
4256                 status = nfs4_setlease(dp);
4257                 goto out;
4258         }
4259         if (fp->fi_had_conflict) {
4260                 status = -EAGAIN;
4261                 goto out_unlock;
4262         }
4263         status = hash_delegation_locked(dp, fp);
4264 out_unlock:
4265         spin_unlock(&fp->fi_lock);
4266         spin_unlock(&state_lock);
4267 out:
4268         if (status) {
4269                 put_clnt_odstate(dp->dl_clnt_odstate);
4270                 nfs4_put_stid(&dp->dl_stid);
4271                 return ERR_PTR(status);
4272         }
4273         return dp;
4274 }
4275
4276 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4277 {
4278         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4279         if (status == -EAGAIN)
4280                 open->op_why_no_deleg = WND4_CONTENTION;
4281         else {
4282                 open->op_why_no_deleg = WND4_RESOURCE;
4283                 switch (open->op_deleg_want) {
4284                 case NFS4_SHARE_WANT_READ_DELEG:
4285                 case NFS4_SHARE_WANT_WRITE_DELEG:
4286                 case NFS4_SHARE_WANT_ANY_DELEG:
4287                         break;
4288                 case NFS4_SHARE_WANT_CANCEL:
4289                         open->op_why_no_deleg = WND4_CANCELLED;
4290                         break;
4291                 case NFS4_SHARE_WANT_NO_DELEG:
4292                         WARN_ON_ONCE(1);
4293                 }
4294         }
4295 }
4296
4297 /*
4298  * Attempt to hand out a delegation.
4299  *
4300  * Note we don't support write delegations, and won't until the vfs has
4301  * proper support for them.
4302  */
4303 static void
4304 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4305                         struct nfs4_ol_stateid *stp)
4306 {
4307         struct nfs4_delegation *dp;
4308         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4309         struct nfs4_client *clp = stp->st_stid.sc_client;
4310         int cb_up;
4311         int status = 0;
4312
4313         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4314         open->op_recall = 0;
4315         switch (open->op_claim_type) {
4316                 case NFS4_OPEN_CLAIM_PREVIOUS:
4317                         if (!cb_up)
4318                                 open->op_recall = 1;
4319                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4320                                 goto out_no_deleg;
4321                         break;
4322                 case NFS4_OPEN_CLAIM_NULL:
4323                 case NFS4_OPEN_CLAIM_FH:
4324                         /*
4325                          * Let's not give out any delegations till everyone's
4326                          * had the chance to reclaim theirs, *and* until
4327                          * NLM locks have all been reclaimed:
4328                          */
4329                         if (locks_in_grace(clp->net))
4330                                 goto out_no_deleg;
4331                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4332                                 goto out_no_deleg;
4333                         /*
4334                          * Also, if the file was opened for write or
4335                          * create, there's a good chance the client's
4336                          * about to write to it, resulting in an
4337                          * immediate recall (since we don't support
4338                          * write delegations):
4339                          */
4340                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4341                                 goto out_no_deleg;
4342                         if (open->op_create == NFS4_OPEN_CREATE)
4343                                 goto out_no_deleg;
4344                         break;
4345                 default:
4346                         goto out_no_deleg;
4347         }
4348         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4349         if (IS_ERR(dp))
4350                 goto out_no_deleg;
4351
4352         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4353
4354         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4355                 STATEID_VAL(&dp->dl_stid.sc_stateid));
4356         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4357         nfs4_put_stid(&dp->dl_stid);
4358         return;
4359 out_no_deleg:
4360         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4361         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4362             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4363                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4364                 open->op_recall = 1;
4365         }
4366
4367         /* 4.1 client asking for a delegation? */
4368         if (open->op_deleg_want)
4369                 nfsd4_open_deleg_none_ext(open, status);
4370         return;
4371 }
4372
4373 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4374                                         struct nfs4_delegation *dp)
4375 {
4376         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4377             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4378                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4379                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4380         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4381                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4382                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4383                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4384         }
4385         /* Otherwise the client must be confused wanting a delegation
4386          * it already has, therefore we don't return
4387          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4388          */
4389 }
4390
4391 __be32
4392 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4393 {
4394         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4395         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4396         struct nfs4_file *fp = NULL;
4397         struct nfs4_ol_stateid *stp = NULL;
4398         struct nfs4_delegation *dp = NULL;
4399         __be32 status;
4400
4401         /*
4402          * Lookup file; if found, lookup stateid and check open request,
4403          * and check for delegations in the process of being recalled.
4404          * If not found, create the nfs4_file struct
4405          */
4406         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4407         if (fp != open->op_file) {
4408                 status = nfs4_check_deleg(cl, open, &dp);
4409                 if (status)
4410                         goto out;
4411                 spin_lock(&fp->fi_lock);
4412                 stp = nfsd4_find_existing_open(fp, open);
4413                 spin_unlock(&fp->fi_lock);
4414         } else {
4415                 open->op_file = NULL;
4416                 status = nfserr_bad_stateid;
4417                 if (nfsd4_is_deleg_cur(open))
4418                         goto out;
4419         }
4420
4421         /*
4422          * OPEN the file, or upgrade an existing OPEN.
4423          * If truncate fails, the OPEN fails.
4424          */
4425         if (stp) {
4426                 /* Stateid was found, this is an OPEN upgrade */
4427                 mutex_lock(&stp->st_mutex);
4428                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4429                 if (status) {
4430                         mutex_unlock(&stp->st_mutex);
4431                         goto out;
4432                 }
4433         } else {
4434                 /* stp is returned locked. */
4435                 stp = init_open_stateid(fp, open);
4436                 /* See if we lost the race to some other thread */
4437                 if (stp->st_access_bmap != 0) {
4438                         status = nfs4_upgrade_open(rqstp, fp, current_fh,
4439                                                 stp, open);
4440                         if (status) {
4441                                 mutex_unlock(&stp->st_mutex);
4442                                 goto out;
4443                         }
4444                         goto upgrade_out;
4445                 }
4446                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4447                 if (status) {
4448                         mutex_unlock(&stp->st_mutex);
4449                         release_open_stateid(stp);
4450                         goto out;
4451                 }
4452
4453                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4454                                                         open->op_odstate);
4455                 if (stp->st_clnt_odstate == open->op_odstate)
4456                         open->op_odstate = NULL;
4457         }
4458 upgrade_out:
4459         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4460         mutex_unlock(&stp->st_mutex);
4461
4462         if (nfsd4_has_session(&resp->cstate)) {
4463                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4464                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4465                         open->op_why_no_deleg = WND4_NOT_WANTED;
4466                         goto nodeleg;
4467                 }
4468         }
4469
4470         /*
4471         * Attempt to hand out a delegation. No error return, because the
4472         * OPEN succeeds even if we fail.
4473         */
4474         nfs4_open_delegation(current_fh, open, stp);
4475 nodeleg:
4476         status = nfs_ok;
4477
4478         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4479                 STATEID_VAL(&stp->st_stid.sc_stateid));
4480 out:
4481         /* 4.1 client trying to upgrade/downgrade delegation? */
4482         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4483             open->op_deleg_want)
4484                 nfsd4_deleg_xgrade_none_ext(open, dp);
4485
4486         if (fp)
4487                 put_nfs4_file(fp);
4488         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4489                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4490         /*
4491         * To finish the open response, we just need to set the rflags.
4492         */
4493         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4494         if (nfsd4_has_session(&resp->cstate))
4495                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
4496         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
4497                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4498
4499         if (dp)
4500                 nfs4_put_stid(&dp->dl_stid);
4501         if (stp)
4502                 nfs4_put_stid(&stp->st_stid);
4503
4504         return status;
4505 }
4506
4507 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4508                               struct nfsd4_open *open)
4509 {
4510         if (open->op_openowner) {
4511                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4512
4513                 nfsd4_cstate_assign_replay(cstate, so);
4514                 nfs4_put_stateowner(so);
4515         }
4516         if (open->op_file)
4517                 kmem_cache_free(file_slab, open->op_file);
4518         if (open->op_stp)
4519                 nfs4_put_stid(&open->op_stp->st_stid);
4520         if (open->op_odstate)
4521                 kmem_cache_free(odstate_slab, open->op_odstate);
4522 }
4523
4524 __be32
4525 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4526             clientid_t *clid)
4527 {
4528         struct nfs4_client *clp;
4529         __be32 status;
4530         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4531
4532         dprintk("process_renew(%08x/%08x): starting\n",
4533                         clid->cl_boot, clid->cl_id);
4534         status = lookup_clientid(clid, cstate, nn);
4535         if (status)
4536                 goto out;
4537         clp = cstate->clp;
4538         status = nfserr_cb_path_down;
4539         if (!list_empty(&clp->cl_delegations)
4540                         && clp->cl_cb_state != NFSD4_CB_UP)
4541                 goto out;
4542         status = nfs_ok;
4543 out:
4544         return status;
4545 }
4546
4547 void
4548 nfsd4_end_grace(struct nfsd_net *nn)
4549 {
4550         /* do nothing if grace period already ended */
4551         if (nn->grace_ended)
4552                 return;
4553
4554         dprintk("NFSD: end of grace period\n");
4555         nn->grace_ended = true;
4556         /*
4557          * If the server goes down again right now, an NFSv4
4558          * client will still be allowed to reclaim after it comes back up,
4559          * even if it hasn't yet had a chance to reclaim state this time.
4560          *
4561          */
4562         nfsd4_record_grace_done(nn);
4563         /*
4564          * At this point, NFSv4 clients can still reclaim.  But if the
4565          * server crashes, any that have not yet reclaimed will be out
4566          * of luck on the next boot.
4567          *
4568          * (NFSv4.1+ clients are considered to have reclaimed once they
4569          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
4570          * have reclaimed after their first OPEN.)
4571          */
4572         locks_end_grace(&nn->nfsd4_manager);
4573         /*
4574          * At this point, and once lockd and/or any other containers
4575          * exit their grace period, further reclaims will fail and
4576          * regular locking can resume.
4577          */
4578 }
4579
4580 static time_t
4581 nfs4_laundromat(struct nfsd_net *nn)
4582 {
4583         struct nfs4_client *clp;
4584         struct nfs4_openowner *oo;
4585         struct nfs4_delegation *dp;
4586         struct nfs4_ol_stateid *stp;
4587         struct nfsd4_blocked_lock *nbl;
4588         struct list_head *pos, *next, reaplist;
4589         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4590         time_t t, new_timeo = nn->nfsd4_lease;
4591
4592         dprintk("NFSD: laundromat service - starting\n");
4593         nfsd4_end_grace(nn);
4594         INIT_LIST_HEAD(&reaplist);
4595         spin_lock(&nn->client_lock);
4596         list_for_each_safe(pos, next, &nn->client_lru) {
4597                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4598                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4599                         t = clp->cl_time - cutoff;
4600                         new_timeo = min(new_timeo, t);
4601                         break;
4602                 }
4603                 if (mark_client_expired_locked(clp)) {
4604                         dprintk("NFSD: client in use (clientid %08x)\n",
4605                                 clp->cl_clientid.cl_id);
4606                         continue;
4607                 }
4608                 list_add(&clp->cl_lru, &reaplist);
4609         }
4610         spin_unlock(&nn->client_lock);
4611         list_for_each_safe(pos, next, &reaplist) {
4612                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4613                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4614                         clp->cl_clientid.cl_id);
4615                 list_del_init(&clp->cl_lru);
4616                 expire_client(clp);
4617         }
4618         spin_lock(&state_lock);
4619         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4620                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4621                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4622                         t = dp->dl_time - cutoff;
4623                         new_timeo = min(new_timeo, t);
4624                         break;
4625                 }
4626                 WARN_ON(!unhash_delegation_locked(dp));
4627                 list_add(&dp->dl_recall_lru, &reaplist);
4628         }
4629         spin_unlock(&state_lock);
4630         while (!list_empty(&reaplist)) {
4631                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4632                                         dl_recall_lru);
4633                 list_del_init(&dp->dl_recall_lru);
4634                 revoke_delegation(dp);
4635         }
4636
4637         spin_lock(&nn->client_lock);
4638         while (!list_empty(&nn->close_lru)) {
4639                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4640                                         oo_close_lru);
4641                 if (time_after((unsigned long)oo->oo_time,
4642                                (unsigned long)cutoff)) {
4643                         t = oo->oo_time - cutoff;
4644                         new_timeo = min(new_timeo, t);
4645                         break;
4646                 }
4647                 list_del_init(&oo->oo_close_lru);
4648                 stp = oo->oo_last_closed_stid;
4649                 oo->oo_last_closed_stid = NULL;
4650                 spin_unlock(&nn->client_lock);
4651                 nfs4_put_stid(&stp->st_stid);
4652                 spin_lock(&nn->client_lock);
4653         }
4654         spin_unlock(&nn->client_lock);
4655
4656         /*
4657          * It's possible for a client to try and acquire an already held lock
4658          * that is being held for a long time, and then lose interest in it.
4659          * So, we clean out any un-revisited request after a lease period
4660          * under the assumption that the client is no longer interested.
4661          *
4662          * RFC5661, sec. 9.6 states that the client must not rely on getting
4663          * notifications and must continue to poll for locks, even when the
4664          * server supports them. Thus this shouldn't lead to clients blocking
4665          * indefinitely once the lock does become free.
4666          */
4667         BUG_ON(!list_empty(&reaplist));
4668         spin_lock(&nn->client_lock);
4669         while (!list_empty(&nn->blocked_locks_lru)) {
4670                 nbl = list_first_entry(&nn->blocked_locks_lru,
4671                                         struct nfsd4_blocked_lock, nbl_lru);
4672                 if (time_after((unsigned long)nbl->nbl_time,
4673                                (unsigned long)cutoff)) {
4674                         t = nbl->nbl_time - cutoff;
4675                         new_timeo = min(new_timeo, t);
4676                         break;
4677                 }
4678                 list_move(&nbl->nbl_lru, &reaplist);
4679                 list_del_init(&nbl->nbl_list);
4680         }
4681         spin_unlock(&nn->client_lock);
4682
4683         while (!list_empty(&reaplist)) {
4684                 nbl = list_first_entry(&nn->blocked_locks_lru,
4685                                         struct nfsd4_blocked_lock, nbl_lru);
4686                 list_del_init(&nbl->nbl_lru);
4687                 posix_unblock_lock(&nbl->nbl_lock);
4688                 free_blocked_lock(nbl);
4689         }
4690
4691         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4692         return new_timeo;
4693 }
4694
4695 static struct workqueue_struct *laundry_wq;
4696 static void laundromat_main(struct work_struct *);
4697
4698 static void
4699 laundromat_main(struct work_struct *laundry)
4700 {
4701         time_t t;
4702         struct delayed_work *dwork = to_delayed_work(laundry);
4703         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4704                                            laundromat_work);
4705
4706         t = nfs4_laundromat(nn);
4707         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4708         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4709 }
4710
4711 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4712 {
4713         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4714                 return nfserr_bad_stateid;
4715         return nfs_ok;
4716 }
4717
4718 static inline int
4719 access_permit_read(struct nfs4_ol_stateid *stp)
4720 {
4721         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4722                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4723                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4724 }
4725
4726 static inline int
4727 access_permit_write(struct nfs4_ol_stateid *stp)
4728 {
4729         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4730                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4731 }
4732
4733 static
4734 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4735 {
4736         __be32 status = nfserr_openmode;
4737
4738         /* For lock stateid's, we test the parent open, not the lock: */
4739         if (stp->st_openstp)
4740                 stp = stp->st_openstp;
4741         if ((flags & WR_STATE) && !access_permit_write(stp))
4742                 goto out;
4743         if ((flags & RD_STATE) && !access_permit_read(stp))
4744                 goto out;
4745         status = nfs_ok;
4746 out:
4747         return status;
4748 }
4749
4750 static inline __be32
4751 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4752 {
4753         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4754                 return nfs_ok;
4755         else if (opens_in_grace(net)) {
4756                 /* Answer in remaining cases depends on existence of
4757                  * conflicting state; so we must wait out the grace period. */
4758                 return nfserr_grace;
4759         } else if (flags & WR_STATE)
4760                 return nfs4_share_conflict(current_fh,
4761                                 NFS4_SHARE_DENY_WRITE);
4762         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4763                 return nfs4_share_conflict(current_fh,
4764                                 NFS4_SHARE_DENY_READ);
4765 }
4766
4767 /*
4768  * Allow READ/WRITE during grace period on recovered state only for files
4769  * that are not able to provide mandatory locking.
4770  */
4771 static inline int
4772 grace_disallows_io(struct net *net, struct inode *inode)
4773 {
4774         return opens_in_grace(net) && mandatory_lock(inode);
4775 }
4776
4777 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4778 {
4779         /*
4780          * When sessions are used the stateid generation number is ignored
4781          * when it is zero.
4782          */
4783         if (has_session && in->si_generation == 0)
4784                 return nfs_ok;
4785
4786         if (in->si_generation == ref->si_generation)
4787                 return nfs_ok;
4788
4789         /* If the client sends us a stateid from the future, it's buggy: */
4790         if (nfsd4_stateid_generation_after(in, ref))
4791                 return nfserr_bad_stateid;
4792         /*
4793          * However, we could see a stateid from the past, even from a
4794          * non-buggy client.  For example, if the client sends a lock
4795          * while some IO is outstanding, the lock may bump si_generation
4796          * while the IO is still in flight.  The client could avoid that
4797          * situation by waiting for responses on all the IO requests,
4798          * but better performance may result in retrying IO that
4799          * receives an old_stateid error if requests are rarely
4800          * reordered in flight:
4801          */
4802         return nfserr_old_stateid;
4803 }
4804
4805 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4806 {
4807         if (ols->st_stateowner->so_is_open_owner &&
4808             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4809                 return nfserr_bad_stateid;
4810         return nfs_ok;
4811 }
4812
4813 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4814 {
4815         struct nfs4_stid *s;
4816         __be32 status = nfserr_bad_stateid;
4817
4818         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4819                 return status;
4820         /* Client debugging aid. */
4821         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4822                 char addr_str[INET6_ADDRSTRLEN];
4823                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4824                                  sizeof(addr_str));
4825                 pr_warn_ratelimited("NFSD: client %s testing state ID "
4826                                         "with incorrect client ID\n", addr_str);
4827                 return status;
4828         }
4829         spin_lock(&cl->cl_lock);
4830         s = find_stateid_locked(cl, stateid);
4831         if (!s)
4832                 goto out_unlock;
4833         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
4834         if (status)
4835                 goto out_unlock;
4836         switch (s->sc_type) {
4837         case NFS4_DELEG_STID:
4838                 status = nfs_ok;
4839                 break;
4840         case NFS4_REVOKED_DELEG_STID:
4841                 status = nfserr_deleg_revoked;
4842                 break;
4843         case NFS4_OPEN_STID:
4844         case NFS4_LOCK_STID:
4845                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
4846                 break;
4847         default:
4848                 printk("unknown stateid type %x\n", s->sc_type);
4849                 /* Fallthrough */
4850         case NFS4_CLOSED_STID:
4851         case NFS4_CLOSED_DELEG_STID:
4852                 status = nfserr_bad_stateid;
4853         }
4854 out_unlock:
4855         spin_unlock(&cl->cl_lock);
4856         return status;
4857 }
4858
4859 __be32
4860 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4861                      stateid_t *stateid, unsigned char typemask,
4862                      struct nfs4_stid **s, struct nfsd_net *nn)
4863 {
4864         __be32 status;
4865
4866         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4867                 return nfserr_bad_stateid;
4868         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
4869         if (status == nfserr_stale_clientid) {
4870                 if (cstate->session)
4871                         return nfserr_bad_stateid;
4872                 return nfserr_stale_stateid;
4873         }
4874         if (status)
4875                 return status;
4876         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
4877         if (!*s)
4878                 return nfserr_bad_stateid;
4879         return nfs_ok;
4880 }
4881
4882 static struct file *
4883 nfs4_find_file(struct nfs4_stid *s, int flags)
4884 {
4885         if (!s)
4886                 return NULL;
4887
4888         switch (s->sc_type) {
4889         case NFS4_DELEG_STID:
4890                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
4891                         return NULL;
4892                 return get_file(s->sc_file->fi_deleg_file);
4893         case NFS4_OPEN_STID:
4894         case NFS4_LOCK_STID:
4895                 if (flags & RD_STATE)
4896                         return find_readable_file(s->sc_file);
4897                 else
4898                         return find_writeable_file(s->sc_file);
4899                 break;
4900         }
4901
4902         return NULL;
4903 }
4904
4905 static __be32
4906 nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
4907 {
4908         __be32 status;
4909
4910         status = nfsd4_check_openowner_confirmed(ols);
4911         if (status)
4912                 return status;
4913         return nfs4_check_openmode(ols, flags);
4914 }
4915
4916 static __be32
4917 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
4918                 struct file **filpp, bool *tmp_file, int flags)
4919 {
4920         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
4921         struct file *file;
4922         __be32 status;
4923
4924         file = nfs4_find_file(s, flags);
4925         if (file) {
4926                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
4927                                 acc | NFSD_MAY_OWNER_OVERRIDE);
4928                 if (status) {
4929                         fput(file);
4930                         return status;
4931                 }
4932
4933                 *filpp = file;
4934         } else {
4935                 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
4936                 if (status)
4937                         return status;
4938
4939                 if (tmp_file)
4940                         *tmp_file = true;
4941         }
4942
4943         return 0;
4944 }
4945
4946 /*
4947  * Checks for stateid operations
4948  */
4949 __be32
4950 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
4951                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
4952                 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
4953 {
4954         struct inode *ino = d_inode(fhp->fh_dentry);
4955         struct net *net = SVC_NET(rqstp);
4956         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4957         struct nfs4_stid *s = NULL;
4958         __be32 status;
4959
4960         if (filpp)
4961                 *filpp = NULL;
4962         if (tmp_file)
4963                 *tmp_file = false;
4964
4965         if (grace_disallows_io(net, ino))
4966                 return nfserr_grace;
4967
4968         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
4969                 status = check_special_stateids(net, fhp, stateid, flags);
4970                 goto done;
4971         }
4972
4973         status = nfsd4_lookup_stateid(cstate, stateid,
4974                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
4975                                 &s, nn);
4976         if (status)
4977                 return status;
4978         status = check_stateid_generation(stateid, &s->sc_stateid,
4979                         nfsd4_has_session(cstate));
4980         if (status)
4981                 goto out;
4982
4983         switch (s->sc_type) {
4984         case NFS4_DELEG_STID:
4985                 status = nfs4_check_delegmode(delegstateid(s), flags);
4986                 break;
4987         case NFS4_OPEN_STID:
4988         case NFS4_LOCK_STID:
4989                 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
4990                 break;
4991         default:
4992                 status = nfserr_bad_stateid;
4993                 break;
4994         }
4995         if (status)
4996                 goto out;
4997         status = nfs4_check_fh(fhp, s);
4998
4999 done:
5000         if (!status && filpp)
5001                 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
5002 out:
5003         if (s)
5004                 nfs4_put_stid(s);
5005         return status;
5006 }
5007
5008 /*
5009  * Test if the stateid is valid
5010  */
5011 __be32
5012 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5013                    struct nfsd4_test_stateid *test_stateid)
5014 {
5015         struct nfsd4_test_stateid_id *stateid;
5016         struct nfs4_client *cl = cstate->session->se_client;
5017
5018         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5019                 stateid->ts_id_status =
5020                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5021
5022         return nfs_ok;
5023 }
5024
5025 static __be32
5026 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5027 {
5028         struct nfs4_ol_stateid *stp = openlockstateid(s);
5029         __be32 ret;
5030
5031         mutex_lock(&stp->st_mutex);
5032
5033         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5034         if (ret)
5035                 goto out;
5036
5037         ret = nfserr_locks_held;
5038         if (check_for_locks(stp->st_stid.sc_file,
5039                             lockowner(stp->st_stateowner)))
5040                 goto out;
5041
5042         release_lock_stateid(stp);
5043         ret = nfs_ok;
5044
5045 out:
5046         mutex_unlock(&stp->st_mutex);
5047         nfs4_put_stid(s);
5048         return ret;
5049 }
5050
5051 __be32
5052 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5053                    struct nfsd4_free_stateid *free_stateid)
5054 {
5055         stateid_t *stateid = &free_stateid->fr_stateid;
5056         struct nfs4_stid *s;
5057         struct nfs4_delegation *dp;
5058         struct nfs4_client *cl = cstate->session->se_client;
5059         __be32 ret = nfserr_bad_stateid;
5060
5061         spin_lock(&cl->cl_lock);
5062         s = find_stateid_locked(cl, stateid);
5063         if (!s)
5064                 goto out_unlock;
5065         switch (s->sc_type) {
5066         case NFS4_DELEG_STID:
5067                 ret = nfserr_locks_held;
5068                 break;
5069         case NFS4_OPEN_STID:
5070                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5071                 if (ret)
5072                         break;
5073                 ret = nfserr_locks_held;
5074                 break;
5075         case NFS4_LOCK_STID:
5076                 atomic_inc(&s->sc_count);
5077                 spin_unlock(&cl->cl_lock);
5078                 ret = nfsd4_free_lock_stateid(stateid, s);
5079                 goto out;
5080         case NFS4_REVOKED_DELEG_STID:
5081                 dp = delegstateid(s);
5082                 list_del_init(&dp->dl_recall_lru);
5083                 spin_unlock(&cl->cl_lock);
5084                 nfs4_put_stid(s);
5085                 ret = nfs_ok;
5086                 goto out;
5087         /* Default falls through and returns nfserr_bad_stateid */
5088         }
5089 out_unlock:
5090         spin_unlock(&cl->cl_lock);
5091 out:
5092         return ret;
5093 }
5094
5095 static inline int
5096 setlkflg (int type)
5097 {
5098         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5099                 RD_STATE : WR_STATE;
5100 }
5101
5102 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5103 {
5104         struct svc_fh *current_fh = &cstate->current_fh;
5105         struct nfs4_stateowner *sop = stp->st_stateowner;
5106         __be32 status;
5107
5108         status = nfsd4_check_seqid(cstate, sop, seqid);
5109         if (status)
5110                 return status;
5111         if (stp->st_stid.sc_type == NFS4_CLOSED_STID
5112                 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
5113                 /*
5114                  * "Closed" stateid's exist *only* to return
5115                  * nfserr_replay_me from the previous step, and
5116                  * revoked delegations are kept only for free_stateid.
5117                  */
5118                 return nfserr_bad_stateid;
5119         mutex_lock(&stp->st_mutex);
5120         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5121         if (status == nfs_ok)
5122                 status = nfs4_check_fh(current_fh, &stp->st_stid);
5123         if (status != nfs_ok)
5124                 mutex_unlock(&stp->st_mutex);
5125         return status;
5126 }
5127
5128 /*
5129  * Checks for sequence id mutating operations.
5130  */
5131 static __be32
5132 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5133                          stateid_t *stateid, char typemask,
5134                          struct nfs4_ol_stateid **stpp,
5135                          struct nfsd_net *nn)
5136 {
5137         __be32 status;
5138         struct nfs4_stid *s;
5139         struct nfs4_ol_stateid *stp = NULL;
5140
5141         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5142                 seqid, STATEID_VAL(stateid));
5143
5144         *stpp = NULL;
5145         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5146         if (status)
5147                 return status;
5148         stp = openlockstateid(s);
5149         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5150
5151         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5152         if (!status)
5153                 *stpp = stp;
5154         else
5155                 nfs4_put_stid(&stp->st_stid);
5156         return status;
5157 }
5158
5159 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5160                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5161 {
5162         __be32 status;
5163         struct nfs4_openowner *oo;
5164         struct nfs4_ol_stateid *stp;
5165
5166         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5167                                                 NFS4_OPEN_STID, &stp, nn);
5168         if (status)
5169                 return status;
5170         oo = openowner(stp->st_stateowner);
5171         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5172                 mutex_unlock(&stp->st_mutex);
5173                 nfs4_put_stid(&stp->st_stid);
5174                 return nfserr_bad_stateid;
5175         }
5176         *stpp = stp;
5177         return nfs_ok;
5178 }
5179
5180 __be32
5181 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5182                    struct nfsd4_open_confirm *oc)
5183 {
5184         __be32 status;
5185         struct nfs4_openowner *oo;
5186         struct nfs4_ol_stateid *stp;
5187         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5188
5189         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5190                         cstate->current_fh.fh_dentry);
5191
5192         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5193         if (status)
5194                 return status;
5195
5196         status = nfs4_preprocess_seqid_op(cstate,
5197                                         oc->oc_seqid, &oc->oc_req_stateid,
5198                                         NFS4_OPEN_STID, &stp, nn);
5199         if (status)
5200                 goto out;
5201         oo = openowner(stp->st_stateowner);
5202         status = nfserr_bad_stateid;
5203         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5204                 mutex_unlock(&stp->st_mutex);
5205                 goto put_stateid;
5206         }
5207         oo->oo_flags |= NFS4_OO_CONFIRMED;
5208         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5209         mutex_unlock(&stp->st_mutex);
5210         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5211                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5212
5213         nfsd4_client_record_create(oo->oo_owner.so_client);
5214         status = nfs_ok;
5215 put_stateid:
5216         nfs4_put_stid(&stp->st_stid);
5217 out:
5218         nfsd4_bump_seqid(cstate, status);
5219         return status;
5220 }
5221
5222 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5223 {
5224         if (!test_access(access, stp))
5225                 return;
5226         nfs4_file_put_access(stp->st_stid.sc_file, access);
5227         clear_access(access, stp);
5228 }
5229
5230 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5231 {
5232         switch (to_access) {
5233         case NFS4_SHARE_ACCESS_READ:
5234                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5235                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5236                 break;
5237         case NFS4_SHARE_ACCESS_WRITE:
5238                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5239                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5240                 break;
5241         case NFS4_SHARE_ACCESS_BOTH:
5242                 break;
5243         default:
5244                 WARN_ON_ONCE(1);
5245         }
5246 }
5247
5248 __be32
5249 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5250                      struct nfsd4_compound_state *cstate,
5251                      struct nfsd4_open_downgrade *od)
5252 {
5253         __be32 status;
5254         struct nfs4_ol_stateid *stp;
5255         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5256
5257         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5258                         cstate->current_fh.fh_dentry);
5259
5260         /* We don't yet support WANT bits: */
5261         if (od->od_deleg_want)
5262                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5263                         od->od_deleg_want);
5264
5265         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5266                                         &od->od_stateid, &stp, nn);
5267         if (status)
5268                 goto out;
5269         status = nfserr_inval;
5270         if (!test_access(od->od_share_access, stp)) {
5271                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5272                         stp->st_access_bmap, od->od_share_access);
5273                 goto put_stateid;
5274         }
5275         if (!test_deny(od->od_share_deny, stp)) {
5276                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5277                         stp->st_deny_bmap, od->od_share_deny);
5278                 goto put_stateid;
5279         }
5280         nfs4_stateid_downgrade(stp, od->od_share_access);
5281         reset_union_bmap_deny(od->od_share_deny, stp);
5282         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5283         status = nfs_ok;
5284 put_stateid:
5285         mutex_unlock(&stp->st_mutex);
5286         nfs4_put_stid(&stp->st_stid);
5287 out:
5288         nfsd4_bump_seqid(cstate, status);
5289         return status;
5290 }
5291
5292 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5293 {
5294         struct nfs4_client *clp = s->st_stid.sc_client;
5295         bool unhashed;
5296         LIST_HEAD(reaplist);
5297
5298         s->st_stid.sc_type = NFS4_CLOSED_STID;
5299         spin_lock(&clp->cl_lock);
5300         unhashed = unhash_open_stateid(s, &reaplist);
5301
5302         if (clp->cl_minorversion) {
5303                 if (unhashed)
5304                         put_ol_stateid_locked(s, &reaplist);
5305                 spin_unlock(&clp->cl_lock);
5306                 free_ol_stateid_reaplist(&reaplist);
5307         } else {
5308                 spin_unlock(&clp->cl_lock);
5309                 free_ol_stateid_reaplist(&reaplist);
5310                 if (unhashed)
5311                         move_to_close_lru(s, clp->net);
5312         }
5313 }
5314
5315 /*
5316  * nfs4_unlock_state() called after encode
5317  */
5318 __be32
5319 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5320             struct nfsd4_close *close)
5321 {
5322         __be32 status;
5323         struct nfs4_ol_stateid *stp;
5324         struct net *net = SVC_NET(rqstp);
5325         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5326
5327         dprintk("NFSD: nfsd4_close on file %pd\n",
5328                         cstate->current_fh.fh_dentry);
5329
5330         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5331                                         &close->cl_stateid,
5332                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
5333                                         &stp, nn);
5334         nfsd4_bump_seqid(cstate, status);
5335         if (status)
5336                 goto out;
5337         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5338         mutex_unlock(&stp->st_mutex);
5339
5340         nfsd4_close_open_stateid(stp);
5341
5342         /* put reference from nfs4_preprocess_seqid_op */
5343         nfs4_put_stid(&stp->st_stid);
5344 out:
5345         return status;
5346 }
5347
5348 __be32
5349 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5350                   struct nfsd4_delegreturn *dr)
5351 {
5352         struct nfs4_delegation *dp;
5353         stateid_t *stateid = &dr->dr_stateid;
5354         struct nfs4_stid *s;
5355         __be32 status;
5356         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5357
5358         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5359                 return status;
5360
5361         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5362         if (status)
5363                 goto out;
5364         dp = delegstateid(s);
5365         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
5366         if (status)
5367                 goto put_stateid;
5368
5369         destroy_delegation(dp);
5370 put_stateid:
5371         nfs4_put_stid(&dp->dl_stid);
5372 out:
5373         return status;
5374 }
5375
5376 static inline u64
5377 end_offset(u64 start, u64 len)
5378 {
5379         u64 end;
5380
5381         end = start + len;
5382         return end >= start ? end: NFS4_MAX_UINT64;
5383 }
5384
5385 /* last octet in a range */
5386 static inline u64
5387 last_byte_offset(u64 start, u64 len)
5388 {
5389         u64 end;
5390
5391         WARN_ON_ONCE(!len);
5392         end = start + len;
5393         return end > start ? end - 1: NFS4_MAX_UINT64;
5394 }
5395
5396 /*
5397  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5398  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5399  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
5400  * locking, this prevents us from being completely protocol-compliant.  The
5401  * real solution to this problem is to start using unsigned file offsets in
5402  * the VFS, but this is a very deep change!
5403  */
5404 static inline void
5405 nfs4_transform_lock_offset(struct file_lock *lock)
5406 {
5407         if (lock->fl_start < 0)
5408                 lock->fl_start = OFFSET_MAX;
5409         if (lock->fl_end < 0)
5410                 lock->fl_end = OFFSET_MAX;
5411 }
5412
5413 static fl_owner_t
5414 nfsd4_fl_get_owner(fl_owner_t owner)
5415 {
5416         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5417
5418         nfs4_get_stateowner(&lo->lo_owner);
5419         return owner;
5420 }
5421
5422 static void
5423 nfsd4_fl_put_owner(fl_owner_t owner)
5424 {
5425         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5426
5427         if (lo)
5428                 nfs4_put_stateowner(&lo->lo_owner);
5429 }
5430
5431 static void
5432 nfsd4_lm_notify(struct file_lock *fl)
5433 {
5434         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
5435         struct net                      *net = lo->lo_owner.so_client->net;
5436         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
5437         struct nfsd4_blocked_lock       *nbl = container_of(fl,
5438                                                 struct nfsd4_blocked_lock, nbl_lock);
5439         bool queue = false;
5440
5441         /* An empty list means that something else is going to be using it */
5442         spin_lock(&nn->client_lock);
5443         if (!list_empty(&nbl->nbl_list)) {
5444                 list_del_init(&nbl->nbl_list);
5445                 list_del_init(&nbl->nbl_lru);
5446                 queue = true;
5447         }
5448         spin_unlock(&nn->client_lock);
5449
5450         if (queue)
5451                 nfsd4_run_cb(&nbl->nbl_cb);
5452 }
5453
5454 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
5455         .lm_notify = nfsd4_lm_notify,
5456         .lm_get_owner = nfsd4_fl_get_owner,
5457         .lm_put_owner = nfsd4_fl_put_owner,
5458 };
5459
5460 static inline void
5461 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5462 {
5463         struct nfs4_lockowner *lo;
5464
5465         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5466                 lo = (struct nfs4_lockowner *) fl->fl_owner;
5467                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5468                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
5469                 if (!deny->ld_owner.data)
5470                         /* We just don't care that much */
5471                         goto nevermind;
5472                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5473                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5474         } else {
5475 nevermind:
5476                 deny->ld_owner.len = 0;
5477                 deny->ld_owner.data = NULL;
5478                 deny->ld_clientid.cl_boot = 0;
5479                 deny->ld_clientid.cl_id = 0;
5480         }
5481         deny->ld_start = fl->fl_start;
5482         deny->ld_length = NFS4_MAX_UINT64;
5483         if (fl->fl_end != NFS4_MAX_UINT64)
5484                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5485         deny->ld_type = NFS4_READ_LT;
5486         if (fl->fl_type != F_RDLCK)
5487                 deny->ld_type = NFS4_WRITE_LT;
5488 }
5489
5490 static struct nfs4_lockowner *
5491 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5492 {
5493         unsigned int strhashval = ownerstr_hashval(owner);
5494         struct nfs4_stateowner *so;
5495
5496         lockdep_assert_held(&clp->cl_lock);
5497
5498         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5499                             so_strhash) {
5500                 if (so->so_is_open_owner)
5501                         continue;
5502                 if (same_owner_str(so, owner))
5503                         return lockowner(nfs4_get_stateowner(so));
5504         }
5505         return NULL;
5506 }
5507
5508 static struct nfs4_lockowner *
5509 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5510 {
5511         struct nfs4_lockowner *lo;
5512
5513         spin_lock(&clp->cl_lock);
5514         lo = find_lockowner_str_locked(clp, owner);
5515         spin_unlock(&clp->cl_lock);
5516         return lo;
5517 }
5518
5519 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5520 {
5521         unhash_lockowner_locked(lockowner(sop));
5522 }
5523
5524 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5525 {
5526         struct nfs4_lockowner *lo = lockowner(sop);
5527
5528         kmem_cache_free(lockowner_slab, lo);
5529 }
5530
5531 static const struct nfs4_stateowner_operations lockowner_ops = {
5532         .so_unhash =    nfs4_unhash_lockowner,
5533         .so_free =      nfs4_free_lockowner,
5534 };
5535
5536 /*
5537  * Alloc a lock owner structure.
5538  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5539  * occurred.
5540  *
5541  * strhashval = ownerstr_hashval
5542  */
5543 static struct nfs4_lockowner *
5544 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5545                            struct nfs4_ol_stateid *open_stp,
5546                            struct nfsd4_lock *lock)
5547 {
5548         struct nfs4_lockowner *lo, *ret;
5549
5550         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5551         if (!lo)
5552                 return NULL;
5553         INIT_LIST_HEAD(&lo->lo_blocked);
5554         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5555         lo->lo_owner.so_is_open_owner = 0;
5556         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5557         lo->lo_owner.so_ops = &lockowner_ops;
5558         spin_lock(&clp->cl_lock);
5559         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5560         if (ret == NULL) {
5561                 list_add(&lo->lo_owner.so_strhash,
5562                          &clp->cl_ownerstr_hashtbl[strhashval]);
5563                 ret = lo;
5564         } else
5565                 nfs4_free_stateowner(&lo->lo_owner);
5566
5567         spin_unlock(&clp->cl_lock);
5568         return ret;
5569 }
5570
5571 static void
5572 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5573                   struct nfs4_file *fp, struct inode *inode,
5574                   struct nfs4_ol_stateid *open_stp)
5575 {
5576         struct nfs4_client *clp = lo->lo_owner.so_client;
5577
5578         lockdep_assert_held(&clp->cl_lock);
5579
5580         atomic_inc(&stp->st_stid.sc_count);
5581         stp->st_stid.sc_type = NFS4_LOCK_STID;
5582         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5583         get_nfs4_file(fp);
5584         stp->st_stid.sc_file = fp;
5585         stp->st_stid.sc_free = nfs4_free_lock_stateid;
5586         stp->st_access_bmap = 0;
5587         stp->st_deny_bmap = open_stp->st_deny_bmap;
5588         stp->st_openstp = open_stp;
5589         mutex_init(&stp->st_mutex);
5590         list_add(&stp->st_locks, &open_stp->st_locks);
5591         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5592         spin_lock(&fp->fi_lock);
5593         list_add(&stp->st_perfile, &fp->fi_stateids);
5594         spin_unlock(&fp->fi_lock);
5595 }
5596
5597 static struct nfs4_ol_stateid *
5598 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5599 {
5600         struct nfs4_ol_stateid *lst;
5601         struct nfs4_client *clp = lo->lo_owner.so_client;
5602
5603         lockdep_assert_held(&clp->cl_lock);
5604
5605         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5606                 if (lst->st_stid.sc_file == fp) {
5607                         atomic_inc(&lst->st_stid.sc_count);
5608                         return lst;
5609                 }
5610         }
5611         return NULL;
5612 }
5613
5614 static struct nfs4_ol_stateid *
5615 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5616                             struct inode *inode, struct nfs4_ol_stateid *ost,
5617                             bool *new)
5618 {
5619         struct nfs4_stid *ns = NULL;
5620         struct nfs4_ol_stateid *lst;
5621         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5622         struct nfs4_client *clp = oo->oo_owner.so_client;
5623
5624         spin_lock(&clp->cl_lock);
5625         lst = find_lock_stateid(lo, fi);
5626         if (lst == NULL) {
5627                 spin_unlock(&clp->cl_lock);
5628                 ns = nfs4_alloc_stid(clp, stateid_slab);
5629                 if (ns == NULL)
5630                         return NULL;
5631
5632                 spin_lock(&clp->cl_lock);
5633                 lst = find_lock_stateid(lo, fi);
5634                 if (likely(!lst)) {
5635                         lst = openlockstateid(ns);
5636                         init_lock_stateid(lst, lo, fi, inode, ost);
5637                         ns = NULL;
5638                         *new = true;
5639                 }
5640         }
5641         spin_unlock(&clp->cl_lock);
5642         if (ns)
5643                 nfs4_put_stid(ns);
5644         return lst;
5645 }
5646
5647 static int
5648 check_lock_length(u64 offset, u64 length)
5649 {
5650         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5651                 (length > ~offset)));
5652 }
5653
5654 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5655 {
5656         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5657
5658         lockdep_assert_held(&fp->fi_lock);
5659
5660         if (test_access(access, lock_stp))
5661                 return;
5662         __nfs4_file_get_access(fp, access);
5663         set_access(access, lock_stp);
5664 }
5665
5666 static __be32
5667 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5668                             struct nfs4_ol_stateid *ost,
5669                             struct nfsd4_lock *lock,
5670                             struct nfs4_ol_stateid **plst, bool *new)
5671 {
5672         __be32 status;
5673         struct nfs4_file *fi = ost->st_stid.sc_file;
5674         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5675         struct nfs4_client *cl = oo->oo_owner.so_client;
5676         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5677         struct nfs4_lockowner *lo;
5678         struct nfs4_ol_stateid *lst;
5679         unsigned int strhashval;
5680         bool hashed;
5681
5682         lo = find_lockowner_str(cl, &lock->lk_new_owner);
5683         if (!lo) {
5684                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5685                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5686                 if (lo == NULL)
5687                         return nfserr_jukebox;
5688         } else {
5689                 /* with an existing lockowner, seqids must be the same */
5690                 status = nfserr_bad_seqid;
5691                 if (!cstate->minorversion &&
5692                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5693                         goto out;
5694         }
5695
5696 retry:
5697         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5698         if (lst == NULL) {
5699                 status = nfserr_jukebox;
5700                 goto out;
5701         }
5702
5703         mutex_lock(&lst->st_mutex);
5704
5705         /* See if it's still hashed to avoid race with FREE_STATEID */
5706         spin_lock(&cl->cl_lock);
5707         hashed = !list_empty(&lst->st_perfile);
5708         spin_unlock(&cl->cl_lock);
5709
5710         if (!hashed) {
5711                 mutex_unlock(&lst->st_mutex);
5712                 nfs4_put_stid(&lst->st_stid);
5713                 goto retry;
5714         }
5715         status = nfs_ok;
5716         *plst = lst;
5717 out:
5718         nfs4_put_stateowner(&lo->lo_owner);
5719         return status;
5720 }
5721
5722 /*
5723  *  LOCK operation
5724  */
5725 __be32
5726 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5727            struct nfsd4_lock *lock)
5728 {
5729         struct nfs4_openowner *open_sop = NULL;
5730         struct nfs4_lockowner *lock_sop = NULL;
5731         struct nfs4_ol_stateid *lock_stp = NULL;
5732         struct nfs4_ol_stateid *open_stp = NULL;
5733         struct nfs4_file *fp;
5734         struct file *filp = NULL;
5735         struct nfsd4_blocked_lock *nbl = NULL;
5736         struct file_lock *file_lock = NULL;
5737         struct file_lock *conflock = NULL;
5738         __be32 status = 0;
5739         int lkflg;
5740         int err;
5741         bool new = false;
5742         unsigned char fl_type;
5743         unsigned int fl_flags = FL_POSIX;
5744         struct net *net = SVC_NET(rqstp);
5745         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5746
5747         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5748                 (long long) lock->lk_offset,
5749                 (long long) lock->lk_length);
5750
5751         if (check_lock_length(lock->lk_offset, lock->lk_length))
5752                  return nfserr_inval;
5753
5754         if ((status = fh_verify(rqstp, &cstate->current_fh,
5755                                 S_IFREG, NFSD_MAY_LOCK))) {
5756                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5757                 return status;
5758         }
5759
5760         if (lock->lk_is_new) {
5761                 if (nfsd4_has_session(cstate))
5762                         /* See rfc 5661 18.10.3: given clientid is ignored: */
5763                         memcpy(&lock->lk_new_clientid,
5764                                 &cstate->session->se_client->cl_clientid,
5765                                 sizeof(clientid_t));
5766
5767                 status = nfserr_stale_clientid;
5768                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5769                         goto out;
5770
5771                 /* validate and update open stateid and open seqid */
5772                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5773                                         lock->lk_new_open_seqid,
5774                                         &lock->lk_new_open_stateid,
5775                                         &open_stp, nn);
5776                 if (status)
5777                         goto out;
5778                 mutex_unlock(&open_stp->st_mutex);
5779                 open_sop = openowner(open_stp->st_stateowner);
5780                 status = nfserr_bad_stateid;
5781                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5782                                                 &lock->lk_new_clientid))
5783                         goto out;
5784                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5785                                                         &lock_stp, &new);
5786         } else {
5787                 status = nfs4_preprocess_seqid_op(cstate,
5788                                        lock->lk_old_lock_seqid,
5789                                        &lock->lk_old_lock_stateid,
5790                                        NFS4_LOCK_STID, &lock_stp, nn);
5791         }
5792         if (status)
5793                 goto out;
5794         lock_sop = lockowner(lock_stp->st_stateowner);
5795
5796         lkflg = setlkflg(lock->lk_type);
5797         status = nfs4_check_openmode(lock_stp, lkflg);
5798         if (status)
5799                 goto out;
5800
5801         status = nfserr_grace;
5802         if (locks_in_grace(net) && !lock->lk_reclaim)
5803                 goto out;
5804         status = nfserr_no_grace;
5805         if (!locks_in_grace(net) && lock->lk_reclaim)
5806                 goto out;
5807
5808         fp = lock_stp->st_stid.sc_file;
5809         switch (lock->lk_type) {
5810                 case NFS4_READW_LT:
5811                         if (nfsd4_has_session(cstate))
5812                                 fl_flags |= FL_SLEEP;
5813                         /* Fallthrough */
5814                 case NFS4_READ_LT:
5815                         spin_lock(&fp->fi_lock);
5816                         filp = find_readable_file_locked(fp);
5817                         if (filp)
5818                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5819                         spin_unlock(&fp->fi_lock);
5820                         fl_type = F_RDLCK;
5821                         break;
5822                 case NFS4_WRITEW_LT:
5823                         if (nfsd4_has_session(cstate))
5824                                 fl_flags |= FL_SLEEP;
5825                         /* Fallthrough */
5826                 case NFS4_WRITE_LT:
5827                         spin_lock(&fp->fi_lock);
5828                         filp = find_writeable_file_locked(fp);
5829                         if (filp)
5830                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
5831                         spin_unlock(&fp->fi_lock);
5832                         fl_type = F_WRLCK;
5833                         break;
5834                 default:
5835                         status = nfserr_inval;
5836                 goto out;
5837         }
5838
5839         if (!filp) {
5840                 status = nfserr_openmode;
5841                 goto out;
5842         }
5843
5844         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
5845         if (!nbl) {
5846                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
5847                 status = nfserr_jukebox;
5848                 goto out;
5849         }
5850
5851         file_lock = &nbl->nbl_lock;
5852         file_lock->fl_type = fl_type;
5853         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
5854         file_lock->fl_pid = current->tgid;
5855         file_lock->fl_file = filp;
5856         file_lock->fl_flags = fl_flags;
5857         file_lock->fl_lmops = &nfsd_posix_mng_ops;
5858         file_lock->fl_start = lock->lk_offset;
5859         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
5860         nfs4_transform_lock_offset(file_lock);
5861
5862         conflock = locks_alloc_lock();
5863         if (!conflock) {
5864                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5865                 status = nfserr_jukebox;
5866                 goto out;
5867         }
5868
5869         if (fl_flags & FL_SLEEP) {
5870                 nbl->nbl_time = jiffies;
5871                 spin_lock(&nn->client_lock);
5872                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
5873                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
5874                 spin_unlock(&nn->client_lock);
5875         }
5876
5877         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
5878         switch (err) {
5879         case 0: /* success! */
5880                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
5881                 status = 0;
5882                 break;
5883         case FILE_LOCK_DEFERRED:
5884                 nbl = NULL;
5885                 /* Fallthrough */
5886         case -EAGAIN:           /* conflock holds conflicting lock */
5887                 status = nfserr_denied;
5888                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
5889                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
5890                 break;
5891         case -EDEADLK:
5892                 status = nfserr_deadlock;
5893                 break;
5894         default:
5895                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
5896                 status = nfserrno(err);
5897                 break;
5898         }
5899 out:
5900         if (nbl) {
5901                 /* dequeue it if we queued it before */
5902                 if (fl_flags & FL_SLEEP) {
5903                         spin_lock(&nn->client_lock);
5904                         list_del_init(&nbl->nbl_list);
5905                         list_del_init(&nbl->nbl_lru);
5906                         spin_unlock(&nn->client_lock);
5907                 }
5908                 free_blocked_lock(nbl);
5909         }
5910         if (filp)
5911                 fput(filp);
5912         if (lock_stp) {
5913                 /* Bump seqid manually if the 4.0 replay owner is openowner */
5914                 if (cstate->replay_owner &&
5915                     cstate->replay_owner != &lock_sop->lo_owner &&
5916                     seqid_mutating_err(ntohl(status)))
5917                         lock_sop->lo_owner.so_seqid++;
5918
5919                 mutex_unlock(&lock_stp->st_mutex);
5920
5921                 /*
5922                  * If this is a new, never-before-used stateid, and we are
5923                  * returning an error, then just go ahead and release it.
5924                  */
5925                 if (status && new)
5926                         release_lock_stateid(lock_stp);
5927
5928                 nfs4_put_stid(&lock_stp->st_stid);
5929         }
5930         if (open_stp)
5931                 nfs4_put_stid(&open_stp->st_stid);
5932         nfsd4_bump_seqid(cstate, status);
5933         if (conflock)
5934                 locks_free_lock(conflock);
5935         return status;
5936 }
5937
5938 /*
5939  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
5940  * so we do a temporary open here just to get an open file to pass to
5941  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
5942  * inode operation.)
5943  */
5944 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
5945 {
5946         struct file *file;
5947         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
5948         if (!err) {
5949                 err = nfserrno(vfs_test_lock(file, lock));
5950                 fput(file);
5951         }
5952         return err;
5953 }
5954
5955 /*
5956  * LOCKT operation
5957  */
5958 __be32
5959 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5960             struct nfsd4_lockt *lockt)
5961 {
5962         struct file_lock *file_lock = NULL;
5963         struct nfs4_lockowner *lo = NULL;
5964         __be32 status;
5965         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5966
5967         if (locks_in_grace(SVC_NET(rqstp)))
5968                 return nfserr_grace;
5969
5970         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
5971                  return nfserr_inval;
5972
5973         if (!nfsd4_has_session(cstate)) {
5974                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
5975                 if (status)
5976                         goto out;
5977         }
5978
5979         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5980                 goto out;
5981
5982         file_lock = locks_alloc_lock();
5983         if (!file_lock) {
5984                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5985                 status = nfserr_jukebox;
5986                 goto out;
5987         }
5988
5989         switch (lockt->lt_type) {
5990                 case NFS4_READ_LT:
5991                 case NFS4_READW_LT:
5992                         file_lock->fl_type = F_RDLCK;
5993                 break;
5994                 case NFS4_WRITE_LT:
5995                 case NFS4_WRITEW_LT:
5996                         file_lock->fl_type = F_WRLCK;
5997                 break;
5998                 default:
5999                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6000                         status = nfserr_inval;
6001                 goto out;
6002         }
6003
6004         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6005         if (lo)
6006                 file_lock->fl_owner = (fl_owner_t)lo;
6007         file_lock->fl_pid = current->tgid;
6008         file_lock->fl_flags = FL_POSIX;
6009
6010         file_lock->fl_start = lockt->lt_offset;
6011         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6012
6013         nfs4_transform_lock_offset(file_lock);
6014
6015         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6016         if (status)
6017                 goto out;
6018
6019         if (file_lock->fl_type != F_UNLCK) {
6020                 status = nfserr_denied;
6021                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6022         }
6023 out:
6024         if (lo)
6025                 nfs4_put_stateowner(&lo->lo_owner);
6026         if (file_lock)
6027                 locks_free_lock(file_lock);
6028         return status;
6029 }
6030
6031 __be32
6032 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6033             struct nfsd4_locku *locku)
6034 {
6035         struct nfs4_ol_stateid *stp;
6036         struct file *filp = NULL;
6037         struct file_lock *file_lock = NULL;
6038         __be32 status;
6039         int err;
6040         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6041
6042         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6043                 (long long) locku->lu_offset,
6044                 (long long) locku->lu_length);
6045
6046         if (check_lock_length(locku->lu_offset, locku->lu_length))
6047                  return nfserr_inval;
6048
6049         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6050                                         &locku->lu_stateid, NFS4_LOCK_STID,
6051                                         &stp, nn);
6052         if (status)
6053                 goto out;
6054         filp = find_any_file(stp->st_stid.sc_file);
6055         if (!filp) {
6056                 status = nfserr_lock_range;
6057                 goto put_stateid;
6058         }
6059         file_lock = locks_alloc_lock();
6060         if (!file_lock) {
6061                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6062                 status = nfserr_jukebox;
6063                 goto fput;
6064         }
6065
6066         file_lock->fl_type = F_UNLCK;
6067         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6068         file_lock->fl_pid = current->tgid;
6069         file_lock->fl_file = filp;
6070         file_lock->fl_flags = FL_POSIX;
6071         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6072         file_lock->fl_start = locku->lu_offset;
6073
6074         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6075                                                 locku->lu_length);
6076         nfs4_transform_lock_offset(file_lock);
6077
6078         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6079         if (err) {
6080                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6081                 goto out_nfserr;
6082         }
6083         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6084 fput:
6085         fput(filp);
6086 put_stateid:
6087         mutex_unlock(&stp->st_mutex);
6088         nfs4_put_stid(&stp->st_stid);
6089 out:
6090         nfsd4_bump_seqid(cstate, status);
6091         if (file_lock)
6092                 locks_free_lock(file_lock);
6093         return status;
6094
6095 out_nfserr:
6096         status = nfserrno(err);
6097         goto fput;
6098 }
6099
6100 /*
6101  * returns
6102  *      true:  locks held by lockowner
6103  *      false: no locks held by lockowner
6104  */
6105 static bool
6106 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6107 {
6108         struct file_lock *fl;
6109         int status = false;
6110         struct file *filp = find_any_file(fp);
6111         struct inode *inode;
6112         struct file_lock_context *flctx;
6113
6114         if (!filp) {
6115                 /* Any valid lock stateid should have some sort of access */
6116                 WARN_ON_ONCE(1);
6117                 return status;
6118         }
6119
6120         inode = file_inode(filp);
6121         flctx = inode->i_flctx;
6122
6123         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6124                 spin_lock(&flctx->flc_lock);
6125                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6126                         if (fl->fl_owner == (fl_owner_t)lowner) {
6127                                 status = true;
6128                                 break;
6129                         }
6130                 }
6131                 spin_unlock(&flctx->flc_lock);
6132         }
6133         fput(filp);
6134         return status;
6135 }
6136
6137 __be32
6138 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6139                         struct nfsd4_compound_state *cstate,
6140                         struct nfsd4_release_lockowner *rlockowner)
6141 {
6142         clientid_t *clid = &rlockowner->rl_clientid;
6143         struct nfs4_stateowner *sop;
6144         struct nfs4_lockowner *lo = NULL;
6145         struct nfs4_ol_stateid *stp;
6146         struct xdr_netobj *owner = &rlockowner->rl_owner;
6147         unsigned int hashval = ownerstr_hashval(owner);
6148         __be32 status;
6149         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6150         struct nfs4_client *clp;
6151         LIST_HEAD (reaplist);
6152
6153         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6154                 clid->cl_boot, clid->cl_id);
6155
6156         status = lookup_clientid(clid, cstate, nn);
6157         if (status)
6158                 return status;
6159
6160         clp = cstate->clp;
6161         /* Find the matching lock stateowner */
6162         spin_lock(&clp->cl_lock);
6163         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6164                             so_strhash) {
6165
6166                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6167                         continue;
6168
6169                 /* see if there are still any locks associated with it */
6170                 lo = lockowner(sop);
6171                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6172                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
6173                                 status = nfserr_locks_held;
6174                                 spin_unlock(&clp->cl_lock);
6175                                 return status;
6176                         }
6177                 }
6178
6179                 nfs4_get_stateowner(sop);
6180                 break;
6181         }
6182         if (!lo) {
6183                 spin_unlock(&clp->cl_lock);
6184                 return status;
6185         }
6186
6187         unhash_lockowner_locked(lo);
6188         while (!list_empty(&lo->lo_owner.so_stateids)) {
6189                 stp = list_first_entry(&lo->lo_owner.so_stateids,
6190                                        struct nfs4_ol_stateid,
6191                                        st_perstateowner);
6192                 WARN_ON(!unhash_lock_stateid(stp));
6193                 put_ol_stateid_locked(stp, &reaplist);
6194         }
6195         spin_unlock(&clp->cl_lock);
6196         free_ol_stateid_reaplist(&reaplist);
6197         nfs4_put_stateowner(&lo->lo_owner);
6198
6199         return status;
6200 }
6201
6202 static inline struct nfs4_client_reclaim *
6203 alloc_reclaim(void)
6204 {
6205         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6206 }
6207
6208 bool
6209 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
6210 {
6211         struct nfs4_client_reclaim *crp;
6212
6213         crp = nfsd4_find_reclaim_client(name, nn);
6214         return (crp && crp->cr_clp);
6215 }
6216
6217 /*
6218  * failure => all reset bets are off, nfserr_no_grace...
6219  */
6220 struct nfs4_client_reclaim *
6221 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6222 {
6223         unsigned int strhashval;
6224         struct nfs4_client_reclaim *crp;
6225
6226         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6227         crp = alloc_reclaim();
6228         if (crp) {
6229                 strhashval = clientstr_hashval(name);
6230                 INIT_LIST_HEAD(&crp->cr_strhash);
6231                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6232                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6233                 crp->cr_clp = NULL;
6234                 nn->reclaim_str_hashtbl_size++;
6235         }
6236         return crp;
6237 }
6238
6239 void
6240 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6241 {
6242         list_del(&crp->cr_strhash);
6243         kfree(crp);
6244         nn->reclaim_str_hashtbl_size--;
6245 }
6246
6247 void
6248 nfs4_release_reclaim(struct nfsd_net *nn)
6249 {
6250         struct nfs4_client_reclaim *crp = NULL;
6251         int i;
6252
6253         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6254                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6255                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6256                                         struct nfs4_client_reclaim, cr_strhash);
6257                         nfs4_remove_reclaim_record(crp, nn);
6258                 }
6259         }
6260         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6261 }
6262
6263 /*
6264  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6265 struct nfs4_client_reclaim *
6266 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6267 {
6268         unsigned int strhashval;
6269         struct nfs4_client_reclaim *crp = NULL;
6270
6271         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6272
6273         strhashval = clientstr_hashval(recdir);
6274         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6275                 if (same_name(crp->cr_recdir, recdir)) {
6276                         return crp;
6277                 }
6278         }
6279         return NULL;
6280 }
6281
6282 /*
6283 * Called from OPEN. Look for clientid in reclaim list.
6284 */
6285 __be32
6286 nfs4_check_open_reclaim(clientid_t *clid,
6287                 struct nfsd4_compound_state *cstate,
6288                 struct nfsd_net *nn)
6289 {
6290         __be32 status;
6291
6292         /* find clientid in conf_id_hashtbl */
6293         status = lookup_clientid(clid, cstate, nn);
6294         if (status)
6295                 return nfserr_reclaim_bad;
6296
6297         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6298                 return nfserr_no_grace;
6299
6300         if (nfsd4_client_record_check(cstate->clp))
6301                 return nfserr_reclaim_bad;
6302
6303         return nfs_ok;
6304 }
6305
6306 #ifdef CONFIG_NFSD_FAULT_INJECTION
6307 static inline void
6308 put_client(struct nfs4_client *clp)
6309 {
6310         atomic_dec(&clp->cl_refcount);
6311 }
6312
6313 static struct nfs4_client *
6314 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6315 {
6316         struct nfs4_client *clp;
6317         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6318                                           nfsd_net_id);
6319
6320         if (!nfsd_netns_ready(nn))
6321                 return NULL;
6322
6323         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6324                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6325                         return clp;
6326         }
6327         return NULL;
6328 }
6329
6330 u64
6331 nfsd_inject_print_clients(void)
6332 {
6333         struct nfs4_client *clp;
6334         u64 count = 0;
6335         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6336                                           nfsd_net_id);
6337         char buf[INET6_ADDRSTRLEN];
6338
6339         if (!nfsd_netns_ready(nn))
6340                 return 0;
6341
6342         spin_lock(&nn->client_lock);
6343         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6344                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6345                 pr_info("NFS Client: %s\n", buf);
6346                 ++count;
6347         }
6348         spin_unlock(&nn->client_lock);
6349
6350         return count;
6351 }
6352
6353 u64
6354 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6355 {
6356         u64 count = 0;
6357         struct nfs4_client *clp;
6358         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6359                                           nfsd_net_id);
6360
6361         if (!nfsd_netns_ready(nn))
6362                 return count;
6363
6364         spin_lock(&nn->client_lock);
6365         clp = nfsd_find_client(addr, addr_size);
6366         if (clp) {
6367                 if (mark_client_expired_locked(clp) == nfs_ok)
6368                         ++count;
6369                 else
6370                         clp = NULL;
6371         }
6372         spin_unlock(&nn->client_lock);
6373
6374         if (clp)
6375                 expire_client(clp);
6376
6377         return count;
6378 }
6379
6380 u64
6381 nfsd_inject_forget_clients(u64 max)
6382 {
6383         u64 count = 0;
6384         struct nfs4_client *clp, *next;
6385         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6386                                                 nfsd_net_id);
6387         LIST_HEAD(reaplist);
6388
6389         if (!nfsd_netns_ready(nn))
6390                 return count;
6391
6392         spin_lock(&nn->client_lock);
6393         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6394                 if (mark_client_expired_locked(clp) == nfs_ok) {
6395                         list_add(&clp->cl_lru, &reaplist);
6396                         if (max != 0 && ++count >= max)
6397                                 break;
6398                 }
6399         }
6400         spin_unlock(&nn->client_lock);
6401
6402         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6403                 expire_client(clp);
6404
6405         return count;
6406 }
6407
6408 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6409                              const char *type)
6410 {
6411         char buf[INET6_ADDRSTRLEN];
6412         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6413         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6414 }
6415
6416 static void
6417 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6418                              struct list_head *collect)
6419 {
6420         struct nfs4_client *clp = lst->st_stid.sc_client;
6421         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6422                                           nfsd_net_id);
6423
6424         if (!collect)
6425                 return;
6426
6427         lockdep_assert_held(&nn->client_lock);
6428         atomic_inc(&clp->cl_refcount);
6429         list_add(&lst->st_locks, collect);
6430 }
6431
6432 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6433                                     struct list_head *collect,
6434                                     bool (*func)(struct nfs4_ol_stateid *))
6435 {
6436         struct nfs4_openowner *oop;
6437         struct nfs4_ol_stateid *stp, *st_next;
6438         struct nfs4_ol_stateid *lst, *lst_next;
6439         u64 count = 0;
6440
6441         spin_lock(&clp->cl_lock);
6442         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6443                 list_for_each_entry_safe(stp, st_next,
6444                                 &oop->oo_owner.so_stateids, st_perstateowner) {
6445                         list_for_each_entry_safe(lst, lst_next,
6446                                         &stp->st_locks, st_locks) {
6447                                 if (func) {
6448                                         if (func(lst))
6449                                                 nfsd_inject_add_lock_to_list(lst,
6450                                                                         collect);
6451                                 }
6452                                 ++count;
6453                                 /*
6454                                  * Despite the fact that these functions deal
6455                                  * with 64-bit integers for "count", we must
6456                                  * ensure that it doesn't blow up the
6457                                  * clp->cl_refcount. Throw a warning if we
6458                                  * start to approach INT_MAX here.
6459                                  */
6460                                 WARN_ON_ONCE(count == (INT_MAX / 2));
6461                                 if (count == max)
6462                                         goto out;
6463                         }
6464                 }
6465         }
6466 out:
6467         spin_unlock(&clp->cl_lock);
6468
6469         return count;
6470 }
6471
6472 static u64
6473 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6474                           u64 max)
6475 {
6476         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6477 }
6478
6479 static u64
6480 nfsd_print_client_locks(struct nfs4_client *clp)
6481 {
6482         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6483         nfsd_print_count(clp, count, "locked files");
6484         return count;
6485 }
6486
6487 u64
6488 nfsd_inject_print_locks(void)
6489 {
6490         struct nfs4_client *clp;
6491         u64 count = 0;
6492         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6493                                                 nfsd_net_id);
6494
6495         if (!nfsd_netns_ready(nn))
6496                 return 0;
6497
6498         spin_lock(&nn->client_lock);
6499         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6500                 count += nfsd_print_client_locks(clp);
6501         spin_unlock(&nn->client_lock);
6502
6503         return count;
6504 }
6505
6506 static void
6507 nfsd_reap_locks(struct list_head *reaplist)
6508 {
6509         struct nfs4_client *clp;
6510         struct nfs4_ol_stateid *stp, *next;
6511
6512         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6513                 list_del_init(&stp->st_locks);
6514                 clp = stp->st_stid.sc_client;
6515                 nfs4_put_stid(&stp->st_stid);
6516                 put_client(clp);
6517         }
6518 }
6519
6520 u64
6521 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6522 {
6523         unsigned int count = 0;
6524         struct nfs4_client *clp;
6525         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6526                                                 nfsd_net_id);
6527         LIST_HEAD(reaplist);
6528
6529         if (!nfsd_netns_ready(nn))
6530                 return count;
6531
6532         spin_lock(&nn->client_lock);
6533         clp = nfsd_find_client(addr, addr_size);
6534         if (clp)
6535                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6536         spin_unlock(&nn->client_lock);
6537         nfsd_reap_locks(&reaplist);
6538         return count;
6539 }
6540
6541 u64
6542 nfsd_inject_forget_locks(u64 max)
6543 {
6544         u64 count = 0;
6545         struct nfs4_client *clp;
6546         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6547                                                 nfsd_net_id);
6548         LIST_HEAD(reaplist);
6549
6550         if (!nfsd_netns_ready(nn))
6551                 return count;
6552
6553         spin_lock(&nn->client_lock);
6554         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6555                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6556                 if (max != 0 && count >= max)
6557                         break;
6558         }
6559         spin_unlock(&nn->client_lock);
6560         nfsd_reap_locks(&reaplist);
6561         return count;
6562 }
6563
6564 static u64
6565 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6566                               struct list_head *collect,
6567                               void (*func)(struct nfs4_openowner *))
6568 {
6569         struct nfs4_openowner *oop, *next;
6570         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6571                                                 nfsd_net_id);
6572         u64 count = 0;
6573
6574         lockdep_assert_held(&nn->client_lock);
6575
6576         spin_lock(&clp->cl_lock);
6577         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6578                 if (func) {
6579                         func(oop);
6580                         if (collect) {
6581                                 atomic_inc(&clp->cl_refcount);
6582                                 list_add(&oop->oo_perclient, collect);
6583                         }
6584                 }
6585                 ++count;
6586                 /*
6587                  * Despite the fact that these functions deal with
6588                  * 64-bit integers for "count", we must ensure that
6589                  * it doesn't blow up the clp->cl_refcount. Throw a
6590                  * warning if we start to approach INT_MAX here.
6591                  */
6592                 WARN_ON_ONCE(count == (INT_MAX / 2));
6593                 if (count == max)
6594                         break;
6595         }
6596         spin_unlock(&clp->cl_lock);
6597
6598         return count;
6599 }
6600
6601 static u64
6602 nfsd_print_client_openowners(struct nfs4_client *clp)
6603 {
6604         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6605
6606         nfsd_print_count(clp, count, "openowners");
6607         return count;
6608 }
6609
6610 static u64
6611 nfsd_collect_client_openowners(struct nfs4_client *clp,
6612                                struct list_head *collect, u64 max)
6613 {
6614         return nfsd_foreach_client_openowner(clp, max, collect,
6615                                                 unhash_openowner_locked);
6616 }
6617
6618 u64
6619 nfsd_inject_print_openowners(void)
6620 {
6621         struct nfs4_client *clp;
6622         u64 count = 0;
6623         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6624                                                 nfsd_net_id);
6625
6626         if (!nfsd_netns_ready(nn))
6627                 return 0;
6628
6629         spin_lock(&nn->client_lock);
6630         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6631                 count += nfsd_print_client_openowners(clp);
6632         spin_unlock(&nn->client_lock);
6633
6634         return count;
6635 }
6636
6637 static void
6638 nfsd_reap_openowners(struct list_head *reaplist)
6639 {
6640         struct nfs4_client *clp;
6641         struct nfs4_openowner *oop, *next;
6642
6643         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6644                 list_del_init(&oop->oo_perclient);
6645                 clp = oop->oo_owner.so_client;
6646                 release_openowner(oop);
6647                 put_client(clp);
6648         }
6649 }
6650
6651 u64
6652 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6653                                      size_t addr_size)
6654 {
6655         unsigned int count = 0;
6656         struct nfs4_client *clp;
6657         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6658                                                 nfsd_net_id);
6659         LIST_HEAD(reaplist);
6660
6661         if (!nfsd_netns_ready(nn))
6662                 return count;
6663
6664         spin_lock(&nn->client_lock);
6665         clp = nfsd_find_client(addr, addr_size);
6666         if (clp)
6667                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6668         spin_unlock(&nn->client_lock);
6669         nfsd_reap_openowners(&reaplist);
6670         return count;
6671 }
6672
6673 u64
6674 nfsd_inject_forget_openowners(u64 max)
6675 {
6676         u64 count = 0;
6677         struct nfs4_client *clp;
6678         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6679                                                 nfsd_net_id);
6680         LIST_HEAD(reaplist);
6681
6682         if (!nfsd_netns_ready(nn))
6683                 return count;
6684
6685         spin_lock(&nn->client_lock);
6686         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6687                 count += nfsd_collect_client_openowners(clp, &reaplist,
6688                                                         max - count);
6689                 if (max != 0 && count >= max)
6690                         break;
6691         }
6692         spin_unlock(&nn->client_lock);
6693         nfsd_reap_openowners(&reaplist);
6694         return count;
6695 }
6696
6697 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6698                                      struct list_head *victims)
6699 {
6700         struct nfs4_delegation *dp, *next;
6701         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6702                                                 nfsd_net_id);
6703         u64 count = 0;
6704
6705         lockdep_assert_held(&nn->client_lock);
6706
6707         spin_lock(&state_lock);
6708         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6709                 if (victims) {
6710                         /*
6711                          * It's not safe to mess with delegations that have a
6712                          * non-zero dl_time. They might have already been broken
6713                          * and could be processed by the laundromat outside of
6714                          * the state_lock. Just leave them be.
6715                          */
6716                         if (dp->dl_time != 0)
6717                                 continue;
6718
6719                         atomic_inc(&clp->cl_refcount);
6720                         WARN_ON(!unhash_delegation_locked(dp));
6721                         list_add(&dp->dl_recall_lru, victims);
6722                 }
6723                 ++count;
6724                 /*
6725                  * Despite the fact that these functions deal with
6726                  * 64-bit integers for "count", we must ensure that
6727                  * it doesn't blow up the clp->cl_refcount. Throw a
6728                  * warning if we start to approach INT_MAX here.
6729                  */
6730                 WARN_ON_ONCE(count == (INT_MAX / 2));
6731                 if (count == max)
6732                         break;
6733         }
6734         spin_unlock(&state_lock);
6735         return count;
6736 }
6737
6738 static u64
6739 nfsd_print_client_delegations(struct nfs4_client *clp)
6740 {
6741         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6742
6743         nfsd_print_count(clp, count, "delegations");
6744         return count;
6745 }
6746
6747 u64
6748 nfsd_inject_print_delegations(void)
6749 {
6750         struct nfs4_client *clp;
6751         u64 count = 0;
6752         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6753                                                 nfsd_net_id);
6754
6755         if (!nfsd_netns_ready(nn))
6756                 return 0;
6757
6758         spin_lock(&nn->client_lock);
6759         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6760                 count += nfsd_print_client_delegations(clp);
6761         spin_unlock(&nn->client_lock);
6762
6763         return count;
6764 }
6765
6766 static void
6767 nfsd_forget_delegations(struct list_head *reaplist)
6768 {
6769         struct nfs4_client *clp;
6770         struct nfs4_delegation *dp, *next;
6771
6772         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6773                 list_del_init(&dp->dl_recall_lru);
6774                 clp = dp->dl_stid.sc_client;
6775                 revoke_delegation(dp);
6776                 put_client(clp);
6777         }
6778 }
6779
6780 u64
6781 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6782                                       size_t addr_size)
6783 {
6784         u64 count = 0;
6785         struct nfs4_client *clp;
6786         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6787                                                 nfsd_net_id);
6788         LIST_HEAD(reaplist);
6789
6790         if (!nfsd_netns_ready(nn))
6791                 return count;
6792
6793         spin_lock(&nn->client_lock);
6794         clp = nfsd_find_client(addr, addr_size);
6795         if (clp)
6796                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6797         spin_unlock(&nn->client_lock);
6798
6799         nfsd_forget_delegations(&reaplist);
6800         return count;
6801 }
6802
6803 u64
6804 nfsd_inject_forget_delegations(u64 max)
6805 {
6806         u64 count = 0;
6807         struct nfs4_client *clp;
6808         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6809                                                 nfsd_net_id);
6810         LIST_HEAD(reaplist);
6811
6812         if (!nfsd_netns_ready(nn))
6813                 return count;
6814
6815         spin_lock(&nn->client_lock);
6816         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6817                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6818                 if (max != 0 && count >= max)
6819                         break;
6820         }
6821         spin_unlock(&nn->client_lock);
6822         nfsd_forget_delegations(&reaplist);
6823         return count;
6824 }
6825
6826 static void
6827 nfsd_recall_delegations(struct list_head *reaplist)
6828 {
6829         struct nfs4_client *clp;
6830         struct nfs4_delegation *dp, *next;
6831
6832         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6833                 list_del_init(&dp->dl_recall_lru);
6834                 clp = dp->dl_stid.sc_client;
6835                 /*
6836                  * We skipped all entries that had a zero dl_time before,
6837                  * so we can now reset the dl_time back to 0. If a delegation
6838                  * break comes in now, then it won't make any difference since
6839                  * we're recalling it either way.
6840                  */
6841                 spin_lock(&state_lock);
6842                 dp->dl_time = 0;
6843                 spin_unlock(&state_lock);
6844                 nfsd_break_one_deleg(dp);
6845                 put_client(clp);
6846         }
6847 }
6848
6849 u64
6850 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
6851                                       size_t addr_size)
6852 {
6853         u64 count = 0;
6854         struct nfs4_client *clp;
6855         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6856                                                 nfsd_net_id);
6857         LIST_HEAD(reaplist);
6858
6859         if (!nfsd_netns_ready(nn))
6860                 return count;
6861
6862         spin_lock(&nn->client_lock);
6863         clp = nfsd_find_client(addr, addr_size);
6864         if (clp)
6865                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6866         spin_unlock(&nn->client_lock);
6867
6868         nfsd_recall_delegations(&reaplist);
6869         return count;
6870 }
6871
6872 u64
6873 nfsd_inject_recall_delegations(u64 max)
6874 {
6875         u64 count = 0;
6876         struct nfs4_client *clp, *next;
6877         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6878                                                 nfsd_net_id);
6879         LIST_HEAD(reaplist);
6880
6881         if (!nfsd_netns_ready(nn))
6882                 return count;
6883
6884         spin_lock(&nn->client_lock);
6885         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6886                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6887                 if (max != 0 && ++count >= max)
6888                         break;
6889         }
6890         spin_unlock(&nn->client_lock);
6891         nfsd_recall_delegations(&reaplist);
6892         return count;
6893 }
6894 #endif /* CONFIG_NFSD_FAULT_INJECTION */
6895
6896 /*
6897  * Since the lifetime of a delegation isn't limited to that of an open, a
6898  * client may quite reasonably hang on to a delegation as long as it has
6899  * the inode cached.  This becomes an obvious problem the first time a
6900  * client's inode cache approaches the size of the server's total memory.
6901  *
6902  * For now we avoid this problem by imposing a hard limit on the number
6903  * of delegations, which varies according to the server's memory size.
6904  */
6905 static void
6906 set_max_delegations(void)
6907 {
6908         /*
6909          * Allow at most 4 delegations per megabyte of RAM.  Quick
6910          * estimates suggest that in the worst case (where every delegation
6911          * is for a different inode), a delegation could take about 1.5K,
6912          * giving a worst case usage of about 6% of memory.
6913          */
6914         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
6915 }
6916
6917 static int nfs4_state_create_net(struct net *net)
6918 {
6919         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6920         int i;
6921
6922         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6923                         CLIENT_HASH_SIZE, GFP_KERNEL);
6924         if (!nn->conf_id_hashtbl)
6925                 goto err;
6926         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6927                         CLIENT_HASH_SIZE, GFP_KERNEL);
6928         if (!nn->unconf_id_hashtbl)
6929                 goto err_unconf_id;
6930         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
6931                         SESSION_HASH_SIZE, GFP_KERNEL);
6932         if (!nn->sessionid_hashtbl)
6933                 goto err_sessionid;
6934
6935         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6936                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
6937                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
6938         }
6939         for (i = 0; i < SESSION_HASH_SIZE; i++)
6940                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
6941         nn->conf_name_tree = RB_ROOT;
6942         nn->unconf_name_tree = RB_ROOT;
6943         INIT_LIST_HEAD(&nn->client_lru);
6944         INIT_LIST_HEAD(&nn->close_lru);
6945         INIT_LIST_HEAD(&nn->del_recall_lru);
6946         INIT_LIST_HEAD(&nn->blocked_locks_lru);
6947         spin_lock_init(&nn->client_lock);
6948
6949         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
6950         get_net(net);
6951
6952         return 0;
6953
6954 err_sessionid:
6955         kfree(nn->unconf_id_hashtbl);
6956 err_unconf_id:
6957         kfree(nn->conf_id_hashtbl);
6958 err:
6959         return -ENOMEM;
6960 }
6961
6962 static void
6963 nfs4_state_destroy_net(struct net *net)
6964 {
6965         int i;
6966         struct nfs4_client *clp = NULL;
6967         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6968
6969         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6970                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
6971                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6972                         destroy_client(clp);
6973                 }
6974         }
6975
6976         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6977                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
6978                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6979                         destroy_client(clp);
6980                 }
6981         }
6982
6983         kfree(nn->sessionid_hashtbl);
6984         kfree(nn->unconf_id_hashtbl);
6985         kfree(nn->conf_id_hashtbl);
6986         put_net(net);
6987 }
6988
6989 int
6990 nfs4_state_start_net(struct net *net)
6991 {
6992         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6993         int ret;
6994
6995         ret = nfs4_state_create_net(net);
6996         if (ret)
6997                 return ret;
6998         nn->boot_time = get_seconds();
6999         nn->grace_ended = false;
7000         nn->nfsd4_manager.block_opens = true;
7001         locks_start_grace(net, &nn->nfsd4_manager);
7002         nfsd4_client_tracking_init(net);
7003         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
7004                nn->nfsd4_grace, net);
7005         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7006         return 0;
7007 }
7008
7009 /* initialization to perform when the nfsd service is started: */
7010
7011 int
7012 nfs4_state_start(void)
7013 {
7014         int ret;
7015
7016         ret = set_callback_cred();
7017         if (ret)
7018                 return -ENOMEM;
7019         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7020         if (laundry_wq == NULL) {
7021                 ret = -ENOMEM;
7022                 goto out_recovery;
7023         }
7024         ret = nfsd4_create_callback_queue();
7025         if (ret)
7026                 goto out_free_laundry;
7027
7028         set_max_delegations();
7029
7030         return 0;
7031
7032 out_free_laundry:
7033         destroy_workqueue(laundry_wq);
7034 out_recovery:
7035         return ret;
7036 }
7037
7038 void
7039 nfs4_state_shutdown_net(struct net *net)
7040 {
7041         struct nfs4_delegation *dp = NULL;
7042         struct list_head *pos, *next, reaplist;
7043         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7044         struct nfsd4_blocked_lock *nbl;
7045
7046         cancel_delayed_work_sync(&nn->laundromat_work);
7047         locks_end_grace(&nn->nfsd4_manager);
7048
7049         INIT_LIST_HEAD(&reaplist);
7050         spin_lock(&state_lock);
7051         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7052                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7053                 WARN_ON(!unhash_delegation_locked(dp));
7054                 list_add(&dp->dl_recall_lru, &reaplist);
7055         }
7056         spin_unlock(&state_lock);
7057         list_for_each_safe(pos, next, &reaplist) {
7058                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7059                 list_del_init(&dp->dl_recall_lru);
7060                 put_clnt_odstate(dp->dl_clnt_odstate);
7061                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
7062                 nfs4_put_stid(&dp->dl_stid);
7063         }
7064
7065         BUG_ON(!list_empty(&reaplist));
7066         spin_lock(&nn->client_lock);
7067         while (!list_empty(&nn->blocked_locks_lru)) {
7068                 nbl = list_first_entry(&nn->blocked_locks_lru,
7069                                         struct nfsd4_blocked_lock, nbl_lru);
7070                 list_move(&nbl->nbl_lru, &reaplist);
7071                 list_del_init(&nbl->nbl_list);
7072         }
7073         spin_unlock(&nn->client_lock);
7074
7075         while (!list_empty(&reaplist)) {
7076                 nbl = list_first_entry(&nn->blocked_locks_lru,
7077                                         struct nfsd4_blocked_lock, nbl_lru);
7078                 list_del_init(&nbl->nbl_lru);
7079                 posix_unblock_lock(&nbl->nbl_lock);
7080                 free_blocked_lock(nbl);
7081         }
7082
7083         nfsd4_client_tracking_exit(net);
7084         nfs4_state_destroy_net(net);
7085 }
7086
7087 void
7088 nfs4_state_shutdown(void)
7089 {
7090         destroy_workqueue(laundry_wq);
7091         nfsd4_destroy_callback_queue();
7092 }
7093
7094 static void
7095 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7096 {
7097         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7098                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7099 }
7100
7101 static void
7102 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7103 {
7104         if (cstate->minorversion) {
7105                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7106                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7107         }
7108 }
7109
7110 void
7111 clear_current_stateid(struct nfsd4_compound_state *cstate)
7112 {
7113         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7114 }
7115
7116 /*
7117  * functions to set current state id
7118  */
7119 void
7120 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
7121 {
7122         put_stateid(cstate, &odp->od_stateid);
7123 }
7124
7125 void
7126 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
7127 {
7128         put_stateid(cstate, &open->op_stateid);
7129 }
7130
7131 void
7132 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
7133 {
7134         put_stateid(cstate, &close->cl_stateid);
7135 }
7136
7137 void
7138 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
7139 {
7140         put_stateid(cstate, &lock->lk_resp_stateid);
7141 }
7142
7143 /*
7144  * functions to consume current state id
7145  */
7146
7147 void
7148 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
7149 {
7150         get_stateid(cstate, &odp->od_stateid);
7151 }
7152
7153 void
7154 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
7155 {
7156         get_stateid(cstate, &drp->dr_stateid);
7157 }
7158
7159 void
7160 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
7161 {
7162         get_stateid(cstate, &fsp->fr_stateid);
7163 }
7164
7165 void
7166 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
7167 {
7168         get_stateid(cstate, &setattr->sa_stateid);
7169 }
7170
7171 void
7172 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
7173 {
7174         get_stateid(cstate, &close->cl_stateid);
7175 }
7176
7177 void
7178 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
7179 {
7180         get_stateid(cstate, &locku->lu_stateid);
7181 }
7182
7183 void
7184 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
7185 {
7186         get_stateid(cstate, &read->rd_stateid);
7187 }
7188
7189 void
7190 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
7191 {
7192         get_stateid(cstate, &write->wr_stateid);
7193 }