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/clnt.h>
  44 #include "xdr4.h"
  45 #include "vfs.h"
  46 #include "current_stateid.h"
  47 #include "fault_inject.h"
  48
  49 #include "netns.h"
  50
  51 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  52
  53 /* Globals */
  54 time_t nfsd4_lease = 90;     /* default lease time */
  55 time_t nfsd4_grace = 90;
  56
  57 #define all_ones {{~0,~0},~0}
  58 static const stateid_t one_stateid = {
  59         .si_generation = ~0,
  60         .si_opaque = all_ones,
  61 };
  62 static const stateid_t zero_stateid = {
  63         /* all fields zero */
  64 };
  65 static const stateid_t currentstateid = {
  66         .si_generation = 1,
  67 };
  68
  69 static u64 current_sessionid = 1;
  70
  71 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  72 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  73 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  74
  75 /* forward declarations */
  76 static int check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner);
  77
  78 /* Locking: */
  79
  80 /* Currently used for almost all code touching nfsv4 state: */
  81 static DEFINE_MUTEX(client_mutex);
  82
  83 /*
  84  * Currently used for the del_recall_lru and file hash table.  In an
  85  * effort to decrease the scope of the client_mutex, this spinlock may
  86  * eventually cover more:
  87  */
  88 static DEFINE_SPINLOCK(recall_lock);
  89
  90 static struct kmem_cache *openowner_slab = NULL;
  91 static struct kmem_cache *lockowner_slab = NULL;
  92 static struct kmem_cache *file_slab = NULL;
  93 static struct kmem_cache *stateid_slab = NULL;
  94 static struct kmem_cache *deleg_slab = NULL;
  95
  96 void
  97 nfs4_lock_state(void)
  98 {
  99         mutex_lock(&client_mutex);
 100 }
 101
 102 static void free_session(struct kref *);
 103
 104 /* Must be called under the client_lock */
 105 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 106 {
 107         kref_put(&ses->se_ref, free_session);
 108 }
 109
 110 static void nfsd4_get_session(struct nfsd4_session *ses)
 111 {
 112         kref_get(&ses->se_ref);
 113 }
 114
 115 void
 116 nfs4_unlock_state(void)
 117 {
 118         mutex_unlock(&client_mutex);
 119 }
 120
 121 static inline u32
 122 opaque_hashval(const void *ptr, int nbytes)
 123 {
 124         unsigned char *cptr = (unsigned char *) ptr;
 125
 126         u32 x = 0;
 127         while (nbytes--) {
 128                 x *= 37;
 129                 x += *cptr++;
 130         }
 131         return x;
 132 }
 133
 134 static struct list_head del_recall_lru;
 135
 136 static void nfsd4_free_file(struct nfs4_file *f)
 137 {
 138         kmem_cache_free(file_slab, f);
 139 }
 140
 141 static inline void
 142 put_nfs4_file(struct nfs4_file *fi)
 143 {
 144         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
 145                 list_del(&fi->fi_hash);
 146                 spin_unlock(&recall_lock);
 147                 iput(fi->fi_inode);
 148                 nfsd4_free_file(fi);
 149         }
 150 }
 151
 152 static inline void
 153 get_nfs4_file(struct nfs4_file *fi)
 154 {
 155         atomic_inc(&fi->fi_ref);
 156 }
 157
 158 static int num_delegations;
 159 unsigned int max_delegations;
 160
 161 /*
 162  * Open owner state (share locks)
 163  */
 164
 165 /* hash tables for lock and open owners */
 166 #define OWNER_HASH_BITS              8
 167 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 168 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 169
 170 static unsigned int ownerstr_hashval(u32 clientid, struct xdr_netobj *ownername)
 171 {
 172         unsigned int ret;
 173
 174         ret = opaque_hashval(ownername->data, ownername->len);
 175         ret += clientid;
 176         return ret & OWNER_HASH_MASK;
 177 }
 178
 179 /* hash table for nfs4_file */
 180 #define FILE_HASH_BITS                   8
 181 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 182
 183 static unsigned int file_hashval(struct inode *ino)
 184 {
 185         /* XXX: why are we hashing on inode pointer, anyway? */
 186         return hash_ptr(ino, FILE_HASH_BITS);
 187 }
 188
 189 static struct list_head file_hashtbl[FILE_HASH_SIZE];
 190
 191 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
 192 {
 193         WARN_ON_ONCE(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
 194         atomic_inc(&fp->fi_access[oflag]);
 195 }
 196
 197 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
 198 {
 199         if (oflag == O_RDWR) {
 200                 __nfs4_file_get_access(fp, O_RDONLY);
 201                 __nfs4_file_get_access(fp, O_WRONLY);
 202         } else
 203                 __nfs4_file_get_access(fp, oflag);
 204 }
 205
 206 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
 207 {
 208         if (fp->fi_fds[oflag]) {
 209                 fput(fp->fi_fds[oflag]);
 210                 fp->fi_fds[oflag] = NULL;
 211         }
 212 }
 213
 214 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 215 {
 216         if (atomic_dec_and_test(&fp->fi_access[oflag])) {
 217                 nfs4_file_put_fd(fp, oflag);
 218                 /*
 219                  * It's also safe to get rid of the RDWR open *if*
 220                  * we no longer have need of the other kind of access
 221                  * or if we already have the other kind of open:
 222                  */
 223                 if (fp->fi_fds[1-oflag]
 224                         || atomic_read(&fp->fi_access[1 - oflag]) == 0)
 225                         nfs4_file_put_fd(fp, O_RDWR);
 226         }
 227 }
 228
 229 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 230 {
 231         if (oflag == O_RDWR) {
 232                 __nfs4_file_put_access(fp, O_RDONLY);
 233                 __nfs4_file_put_access(fp, O_WRONLY);
 234         } else
 235                 __nfs4_file_put_access(fp, oflag);
 236 }
 237
 238 static inline int get_new_stid(struct nfs4_stid *stid)
 239 {
 240         static int min_stateid = 0;
 241         struct idr *stateids = &stid->sc_client->cl_stateids;
 242         int new_stid;
 243         int error;
 244
 245         error = idr_get_new_above(stateids, stid, min_stateid, &new_stid);
 246         /*
 247          * Note: the necessary preallocation was done in
 248          * nfs4_alloc_stateid().  The idr code caps the number of
 249          * preallocations that can exist at a time, but the state lock
 250          * prevents anyone from using ours before we get here:
 251          */
 252         WARN_ON_ONCE(error);
 253         /*
 254          * It shouldn't be a problem to reuse an opaque stateid value.
 255          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 256          * example, a stray write retransmission could be accepted by
 257          * the server when it should have been rejected.  Therefore,
 258          * adopt a trick from the sctp code to attempt to maximize the
 259          * amount of time until an id is reused, by ensuring they always
 260          * "increase" (mod INT_MAX):
 261          */
 262
 263         min_stateid = new_stid+1;
 264         if (min_stateid == INT_MAX)
 265                 min_stateid = 0;
 266         return new_stid;
 267 }
 268
 269 static void init_stid(struct nfs4_stid *stid, struct nfs4_client *cl, unsigned char type)
 270 {
 271         stateid_t *s = &stid->sc_stateid;
 272         int new_id;
 273
 274         stid->sc_type = type;
 275         stid->sc_client = cl;
 276         s->si_opaque.so_clid = cl->cl_clientid;
 277         new_id = get_new_stid(stid);
 278         s->si_opaque.so_id = (u32)new_id;
 279         /* Will be incremented before return to client: */
 280         s->si_generation = 0;
 281 }
 282
 283 static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab)
 284 {
 285         struct idr *stateids = &cl->cl_stateids;
 286
 287         if (!idr_pre_get(stateids, GFP_KERNEL))
 288                 return NULL;
 289         /*
 290          * Note: if we fail here (or any time between now and the time
 291          * we actually get the new idr), we won't need to undo the idr
 292          * preallocation, since the idr code caps the number of
 293          * preallocated entries.
 294          */
 295         return kmem_cache_alloc(slab, GFP_KERNEL);
 296 }
 297
 298 static struct nfs4_ol_stateid * nfs4_alloc_stateid(struct nfs4_client *clp)
 299 {
 300         return openlockstateid(nfs4_alloc_stid(clp, stateid_slab));
 301 }
 302
 303 static struct nfs4_delegation *
 304 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_ol_stateid *stp, struct svc_fh *current_fh, u32 type)
 305 {
 306         struct nfs4_delegation *dp;
 307         struct nfs4_file *fp = stp->st_file;
 308
 309         dprintk("NFSD alloc_init_deleg\n");
 310         /*
 311          * Major work on the lease subsystem (for example, to support
 312          * calbacks on stat) will be required before we can support
 313          * write delegations properly.
 314          */
 315         if (type != NFS4_OPEN_DELEGATE_READ)
 316                 return NULL;
 317         if (fp->fi_had_conflict)
 318                 return NULL;
 319         if (num_delegations > max_delegations)
 320                 return NULL;
 321         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
 322         if (dp == NULL)
 323                 return dp;
 324         init_stid(&dp->dl_stid, clp, NFS4_DELEG_STID);
 325         /*
 326          * delegation seqid's are never incremented.  The 4.1 special
 327          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 328          * 0 anyway just for consistency and use 1:
 329          */
 330         dp->dl_stid.sc_stateid.si_generation = 1;
 331         num_delegations++;
 332         INIT_LIST_HEAD(&dp->dl_perfile);
 333         INIT_LIST_HEAD(&dp->dl_perclnt);
 334         INIT_LIST_HEAD(&dp->dl_recall_lru);
 335         get_nfs4_file(fp);
 336         dp->dl_file = fp;
 337         dp->dl_type = type;
 338         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
 339         dp->dl_time = 0;
 340         atomic_set(&dp->dl_count, 1);
 341         nfsd4_init_callback(&dp->dl_recall);
 342         return dp;
 343 }
 344
 345 void
 346 nfs4_put_delegation(struct nfs4_delegation *dp)
 347 {
 348         if (atomic_dec_and_test(&dp->dl_count)) {
 349                 dprintk("NFSD: freeing dp %p\n",dp);
 350                 put_nfs4_file(dp->dl_file);
 351                 kmem_cache_free(deleg_slab, dp);
 352                 num_delegations--;
 353         }
 354 }
 355
 356 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 357 {
 358         if (atomic_dec_and_test(&fp->fi_delegees)) {
 359                 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
 360                 fp->fi_lease = NULL;
 361                 fput(fp->fi_deleg_file);
 362                 fp->fi_deleg_file = NULL;
 363         }
 364 }
 365
 366 static void unhash_stid(struct nfs4_stid *s)
 367 {
 368         struct idr *stateids = &s->sc_client->cl_stateids;
 369
 370         idr_remove(stateids, s->sc_stateid.si_opaque.so_id);
 371 }
 372
 373 /* Called under the state lock. */
 374 static void
 375 unhash_delegation(struct nfs4_delegation *dp)
 376 {
 377         unhash_stid(&dp->dl_stid);
 378         list_del_init(&dp->dl_perclnt);
 379         spin_lock(&recall_lock);
 380         list_del_init(&dp->dl_perfile);
 381         list_del_init(&dp->dl_recall_lru);
 382         spin_unlock(&recall_lock);
 383         nfs4_put_deleg_lease(dp->dl_file);
 384         nfs4_put_delegation(dp);
 385 }
 386
 387 /*
 388  * SETCLIENTID state
 389  */
 390
 391 /* client_lock protects the client lru list and session hash table */
 392 static DEFINE_SPINLOCK(client_lock);
 393
 394 static unsigned int clientid_hashval(u32 id)
 395 {
 396         return id & CLIENT_HASH_MASK;
 397 }
 398
 399 static unsigned int clientstr_hashval(const char *name)
 400 {
 401         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
 402 }
 403
 404 /*
 405  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 406  * st_{access,deny}_bmap field of the stateid, in order to track not
 407  * only what share bits are currently in force, but also what
 408  * combinations of share bits previous opens have used.  This allows us
 409  * to enforce the recommendation of rfc 3530 14.2.19 that the server
 410  * return an error if the client attempt to downgrade to a combination
 411  * of share bits not explicable by closing some of its previous opens.
 412  *
 413  * XXX: This enforcement is actually incomplete, since we don't keep
 414  * track of access/deny bit combinations; so, e.g., we allow:
 415  *
 416  *      OPEN allow read, deny write
 417  *      OPEN allow both, deny none
 418  *      DOWNGRADE allow read, deny none
 419  *
 420  * which we should reject.
 421  */
 422 static unsigned int
 423 bmap_to_share_mode(unsigned long bmap) {
 424         int i;
 425         unsigned int access = 0;
 426
 427         for (i = 1; i < 4; i++) {
 428                 if (test_bit(i, &bmap))
 429                         access |= i;
 430         }
 431         return access;
 432 }
 433
 434 static bool
 435 test_share(struct nfs4_ol_stateid *stp, struct nfsd4_open *open) {
 436         unsigned int access, deny;
 437
 438         access = bmap_to_share_mode(stp->st_access_bmap);
 439         deny = bmap_to_share_mode(stp->st_deny_bmap);
 440         if ((access & open->op_share_deny) || (deny & open->op_share_access))
 441                 return false;
 442         return true;
 443 }
 444
 445 /* set share access for a given stateid */
 446 static inline void
 447 set_access(u32 access, struct nfs4_ol_stateid *stp)
 448 {
 449         __set_bit(access, &stp->st_access_bmap);
 450 }
 451
 452 /* clear share access for a given stateid */
 453 static inline void
 454 clear_access(u32 access, struct nfs4_ol_stateid *stp)
 455 {
 456         __clear_bit(access, &stp->st_access_bmap);
 457 }
 458
 459 /* test whether a given stateid has access */
 460 static inline bool
 461 test_access(u32 access, struct nfs4_ol_stateid *stp)
 462 {
 463         return test_bit(access, &stp->st_access_bmap);
 464 }
 465
 466 /* set share deny for a given stateid */
 467 static inline void
 468 set_deny(u32 access, struct nfs4_ol_stateid *stp)
 469 {
 470         __set_bit(access, &stp->st_deny_bmap);
 471 }
 472
 473 /* clear share deny for a given stateid */
 474 static inline void
 475 clear_deny(u32 access, struct nfs4_ol_stateid *stp)
 476 {
 477         __clear_bit(access, &stp->st_deny_bmap);
 478 }
 479
 480 /* test whether a given stateid is denying specific access */
 481 static inline bool
 482 test_deny(u32 access, struct nfs4_ol_stateid *stp)
 483 {
 484         return test_bit(access, &stp->st_deny_bmap);
 485 }
 486
 487 static int nfs4_access_to_omode(u32 access)
 488 {
 489         switch (access & NFS4_SHARE_ACCESS_BOTH) {
 490         case NFS4_SHARE_ACCESS_READ:
 491                 return O_RDONLY;
 492         case NFS4_SHARE_ACCESS_WRITE:
 493                 return O_WRONLY;
 494         case NFS4_SHARE_ACCESS_BOTH:
 495                 return O_RDWR;
 496         }
 497         WARN_ON_ONCE(1);
 498         return O_RDONLY;
 499 }
 500
 501 /* release all access and file references for a given stateid */
 502 static void
 503 release_all_access(struct nfs4_ol_stateid *stp)
 504 {
 505         int i;
 506
 507         for (i = 1; i < 4; i++) {
 508                 if (test_access(i, stp))
 509                         nfs4_file_put_access(stp->st_file,
 510                                              nfs4_access_to_omode(i));
 511                 clear_access(i, stp);
 512         }
 513 }
 514
 515 static void unhash_generic_stateid(struct nfs4_ol_stateid *stp)
 516 {
 517         list_del(&stp->st_perfile);
 518         list_del(&stp->st_perstateowner);
 519 }
 520
 521 static void close_generic_stateid(struct nfs4_ol_stateid *stp)
 522 {
 523         release_all_access(stp);
 524         put_nfs4_file(stp->st_file);
 525         stp->st_file = NULL;
 526 }
 527
 528 static void free_generic_stateid(struct nfs4_ol_stateid *stp)
 529 {
 530         kmem_cache_free(stateid_slab, stp);
 531 }
 532
 533 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
 534 {
 535         struct file *file;
 536
 537         unhash_generic_stateid(stp);
 538         unhash_stid(&stp->st_stid);
 539         file = find_any_file(stp->st_file);
 540         if (file)
 541                 locks_remove_posix(file, (fl_owner_t)lockowner(stp->st_stateowner));
 542         close_generic_stateid(stp);
 543         free_generic_stateid(stp);
 544 }
 545
 546 static void unhash_lockowner(struct nfs4_lockowner *lo)
 547 {
 548         struct nfs4_ol_stateid *stp;
 549
 550         list_del(&lo->lo_owner.so_strhash);
 551         list_del(&lo->lo_perstateid);
 552         list_del(&lo->lo_owner_ino_hash);
 553         while (!list_empty(&lo->lo_owner.so_stateids)) {
 554                 stp = list_first_entry(&lo->lo_owner.so_stateids,
 555                                 struct nfs4_ol_stateid, st_perstateowner);
 556                 release_lock_stateid(stp);
 557         }
 558 }
 559
 560 static void release_lockowner(struct nfs4_lockowner *lo)
 561 {
 562         unhash_lockowner(lo);
 563         nfs4_free_lockowner(lo);
 564 }
 565
 566 static void
 567 release_stateid_lockowners(struct nfs4_ol_stateid *open_stp)
 568 {
 569         struct nfs4_lockowner *lo;
 570
 571         while (!list_empty(&open_stp->st_lockowners)) {
 572                 lo = list_entry(open_stp->st_lockowners.next,
 573                                 struct nfs4_lockowner, lo_perstateid);
 574                 release_lockowner(lo);
 575         }
 576 }
 577
 578 static void unhash_open_stateid(struct nfs4_ol_stateid *stp)
 579 {
 580         unhash_generic_stateid(stp);
 581         release_stateid_lockowners(stp);
 582         close_generic_stateid(stp);
 583 }
 584
 585 static void release_open_stateid(struct nfs4_ol_stateid *stp)
 586 {
 587         unhash_open_stateid(stp);
 588         unhash_stid(&stp->st_stid);
 589         free_generic_stateid(stp);
 590 }
 591
 592 static void unhash_openowner(struct nfs4_openowner *oo)
 593 {
 594         struct nfs4_ol_stateid *stp;
 595
 596         list_del(&oo->oo_owner.so_strhash);
 597         list_del(&oo->oo_perclient);
 598         while (!list_empty(&oo->oo_owner.so_stateids)) {
 599                 stp = list_first_entry(&oo->oo_owner.so_stateids,
 600                                 struct nfs4_ol_stateid, st_perstateowner);
 601                 release_open_stateid(stp);
 602         }
 603 }
 604
 605 static void release_last_closed_stateid(struct nfs4_openowner *oo)
 606 {
 607         struct nfs4_ol_stateid *s = oo->oo_last_closed_stid;
 608
 609         if (s) {
 610                 unhash_stid(&s->st_stid);
 611                 free_generic_stateid(s);
 612                 oo->oo_last_closed_stid = NULL;
 613         }
 614 }
 615
 616 static void release_openowner(struct nfs4_openowner *oo)
 617 {
 618         unhash_openowner(oo);
 619         list_del(&oo->oo_close_lru);
 620         release_last_closed_stateid(oo);
 621         nfs4_free_openowner(oo);
 622 }
 623
 624 static inline int
 625 hash_sessionid(struct nfs4_sessionid *sessionid)
 626 {
 627         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
 628
 629         return sid->sequence % SESSION_HASH_SIZE;
 630 }
 631
 632 #ifdef NFSD_DEBUG
 633 static inline void
 634 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 635 {
 636         u32 *ptr = (u32 *)(&sessionid->data[0]);
 637         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
 638 }
 639 #else
 640 static inline void
 641 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 642 {
 643 }
 644 #endif
 645
 646
 647 static void
 648 gen_sessionid(struct nfsd4_session *ses)
 649 {
 650         struct nfs4_client *clp = ses->se_client;
 651         struct nfsd4_sessionid *sid;
 652
 653         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
 654         sid->clientid = clp->cl_clientid;
 655         sid->sequence = current_sessionid++;
 656         sid->reserved = 0;
 657 }
 658
 659 /*
 660  * The protocol defines ca_maxresponssize_cached to include the size of
 661  * the rpc header, but all we need to cache is the data starting after
 662  * the end of the initial SEQUENCE operation--the rest we regenerate
 663  * each time.  Therefore we can advertise a ca_maxresponssize_cached
 664  * value that is the number of bytes in our cache plus a few additional
 665  * bytes.  In order to stay on the safe side, and not promise more than
 666  * we can cache, those additional bytes must be the minimum possible: 24
 667  * bytes of rpc header (xid through accept state, with AUTH_NULL
 668  * verifier), 12 for the compound header (with zero-length tag), and 44
 669  * for the SEQUENCE op response:
 670  */
 671 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
 672
 673 static void
 674 free_session_slots(struct nfsd4_session *ses)
 675 {
 676         int i;
 677
 678         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
 679                 kfree(ses->se_slots[i]);
 680 }
 681
 682 /*
 683  * We don't actually need to cache the rpc and session headers, so we
 684  * can allocate a little less for each slot:
 685  */
 686 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
 687 {
 688         return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
 689 }
 690
 691 static int nfsd4_sanitize_slot_size(u32 size)
 692 {
 693         size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
 694         size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
 695
 696         return size;
 697 }
 698
 699 /*
 700  * XXX: If we run out of reserved DRC memory we could (up to a point)
 701  * re-negotiate active sessions and reduce their slot usage to make
 702  * room for new connections. For now we just fail the create session.
 703  */
 704 static int nfsd4_get_drc_mem(int slotsize, u32 num)
 705 {
 706         int avail;
 707
 708         num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
 709
 710         spin_lock(&nfsd_drc_lock);
 711         avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
 712                         nfsd_drc_max_mem - nfsd_drc_mem_used);
 713         num = min_t(int, num, avail / slotsize);
 714         nfsd_drc_mem_used += num * slotsize;
 715         spin_unlock(&nfsd_drc_lock);
 716
 717         return num;
 718 }
 719
 720 static void nfsd4_put_drc_mem(int slotsize, int num)
 721 {
 722         spin_lock(&nfsd_drc_lock);
 723         nfsd_drc_mem_used -= slotsize * num;
 724         spin_unlock(&nfsd_drc_lock);
 725 }
 726
 727 static struct nfsd4_session *__alloc_session(int slotsize, int numslots)
 728 {
 729         struct nfsd4_session *new;
 730         int mem, i;
 731
 732         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
 733                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
 734         mem = numslots * sizeof(struct nfsd4_slot *);
 735
 736         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
 737         if (!new)
 738                 return NULL;
 739         /* allocate each struct nfsd4_slot and data cache in one piece */
 740         for (i = 0; i < numslots; i++) {
 741                 mem = sizeof(struct nfsd4_slot) + slotsize;
 742                 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
 743                 if (!new->se_slots[i])
 744                         goto out_free;
 745         }
 746         return new;
 747 out_free:
 748         while (i--)
 749                 kfree(new->se_slots[i]);
 750         kfree(new);
 751         return NULL;
 752 }
 753
 754 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
 755 {
 756         u32 maxrpc = nfsd_serv->sv_max_mesg;
 757
 758         new->maxreqs = numslots;
 759         new->maxresp_cached = min_t(u32, req->maxresp_cached,
 760                                         slotsize + NFSD_MIN_HDR_SEQ_SZ);
 761         new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
 762         new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
 763         new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
 764 }
 765
 766 static void free_conn(struct nfsd4_conn *c)
 767 {
 768         svc_xprt_put(c->cn_xprt);
 769         kfree(c);
 770 }
 771
 772 static void nfsd4_conn_lost(struct svc_xpt_user *u)
 773 {
 774         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
 775         struct nfs4_client *clp = c->cn_session->se_client;
 776
 777         spin_lock(&clp->cl_lock);
 778         if (!list_empty(&c->cn_persession)) {
 779                 list_del(&c->cn_persession);
 780                 free_conn(c);
 781         }
 782         spin_unlock(&clp->cl_lock);
 783         nfsd4_probe_callback(clp);
 784 }
 785
 786 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
 787 {
 788         struct nfsd4_conn *conn;
 789
 790         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
 791         if (!conn)
 792                 return NULL;
 793         svc_xprt_get(rqstp->rq_xprt);
 794         conn->cn_xprt = rqstp->rq_xprt;
 795         conn->cn_flags = flags;
 796         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
 797         return conn;
 798 }
 799
 800 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 801 {
 802         conn->cn_session = ses;
 803         list_add(&conn->cn_persession, &ses->se_conns);
 804 }
 805
 806 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 807 {
 808         struct nfs4_client *clp = ses->se_client;
 809
 810         spin_lock(&clp->cl_lock);
 811         __nfsd4_hash_conn(conn, ses);
 812         spin_unlock(&clp->cl_lock);
 813 }
 814
 815 static int nfsd4_register_conn(struct nfsd4_conn *conn)
 816 {
 817         conn->cn_xpt_user.callback = nfsd4_conn_lost;
 818         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
 819 }
 820
 821 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
 822 {
 823         int ret;
 824
 825         nfsd4_hash_conn(conn, ses);
 826         ret = nfsd4_register_conn(conn);
 827         if (ret)
 828                 /* oops; xprt is already down: */
 829                 nfsd4_conn_lost(&conn->cn_xpt_user);
 830         if (conn->cn_flags & NFS4_CDFC4_BACK) {
 831                 /* callback channel may be back up */
 832                 nfsd4_probe_callback(ses->se_client);
 833         }
 834 }
 835
 836 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
 837 {
 838         u32 dir = NFS4_CDFC4_FORE;
 839
 840         if (cses->flags & SESSION4_BACK_CHAN)
 841                 dir |= NFS4_CDFC4_BACK;
 842         return alloc_conn(rqstp, dir);
 843 }
 844
 845 /* must be called under client_lock */
 846 static void nfsd4_del_conns(struct nfsd4_session *s)
 847 {
 848         struct nfs4_client *clp = s->se_client;
 849         struct nfsd4_conn *c;
 850
 851         spin_lock(&clp->cl_lock);
 852         while (!list_empty(&s->se_conns)) {
 853                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
 854                 list_del_init(&c->cn_persession);
 855                 spin_unlock(&clp->cl_lock);
 856
 857                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
 858                 free_conn(c);
 859
 860                 spin_lock(&clp->cl_lock);
 861         }
 862         spin_unlock(&clp->cl_lock);
 863 }
 864
 865 static void __free_session(struct nfsd4_session *ses)
 866 {
 867         nfsd4_put_drc_mem(slot_bytes(&ses->se_fchannel), ses->se_fchannel.maxreqs);
 868         free_session_slots(ses);
 869         kfree(ses);
 870 }
 871
 872 static void free_session(struct kref *kref)
 873 {
 874         struct nfsd4_session *ses;
 875
 876         lockdep_assert_held(&client_lock);
 877         ses = container_of(kref, struct nfsd4_session, se_ref);
 878         nfsd4_del_conns(ses);
 879         __free_session(ses);
 880 }
 881
 882 void nfsd4_put_session(struct nfsd4_session *ses)
 883 {
 884         spin_lock(&client_lock);
 885         nfsd4_put_session_locked(ses);
 886         spin_unlock(&client_lock);
 887 }
 888
 889 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fchan)
 890 {
 891         struct nfsd4_session *new;
 892         int numslots, slotsize;
 893         /*
 894          * Note decreasing slot size below client's request may
 895          * make it difficult for client to function correctly, whereas
 896          * decreasing the number of slots will (just?) affect
 897          * performance.  When short on memory we therefore prefer to
 898          * decrease number of slots instead of their size.
 899          */
 900         slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
 901         numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
 902         if (numslots < 1)
 903                 return NULL;
 904
 905         new = __alloc_session(slotsize, numslots);
 906         if (!new) {
 907                 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
 908                 return NULL;
 909         }
 910         init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
 911         return new;
 912 }
 913
 914 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
 915 {
 916         int idx;
 917         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
 918
 919         new->se_client = clp;
 920         gen_sessionid(new);
 921
 922         INIT_LIST_HEAD(&new->se_conns);
 923
 924         new->se_cb_seq_nr = 1;
 925         new->se_flags = cses->flags;
 926         new->se_cb_prog = cses->callback_prog;
 927         new->se_cb_sec = cses->cb_sec;
 928         kref_init(&new->se_ref);
 929         idx = hash_sessionid(&new->se_sessionid);
 930         spin_lock(&client_lock);
 931         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
 932         spin_lock(&clp->cl_lock);
 933         list_add(&new->se_perclnt, &clp->cl_sessions);
 934         spin_unlock(&clp->cl_lock);
 935         spin_unlock(&client_lock);
 936
 937         if (cses->flags & SESSION4_BACK_CHAN) {
 938                 struct sockaddr *sa = svc_addr(rqstp);
 939                 /*
 940                  * This is a little silly; with sessions there's no real
 941                  * use for the callback address.  Use the peer address
 942                  * as a reasonable default for now, but consider fixing
 943                  * the rpc client not to require an address in the
 944                  * future:
 945                  */
 946                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
 947                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
 948         }
 949 }
 950
 951 /* caller must hold client_lock */
 952 static struct nfsd4_session *
 953 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
 954 {
 955         struct nfsd4_session *elem;
 956         int idx;
 957         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 958
 959         dump_sessionid(__func__, sessionid);
 960         idx = hash_sessionid(sessionid);
 961         /* Search in the appropriate list */
 962         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
 963                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
 964                             NFS4_MAX_SESSIONID_LEN)) {
 965                         return elem;
 966                 }
 967         }
 968
 969         dprintk("%s: session not found\n", __func__);
 970         return NULL;
 971 }
 972
 973 /* caller must hold client_lock */
 974 static void
 975 unhash_session(struct nfsd4_session *ses)
 976 {
 977         list_del(&ses->se_hash);
 978         spin_lock(&ses->se_client->cl_lock);
 979         list_del(&ses->se_perclnt);
 980         spin_unlock(&ses->se_client->cl_lock);
 981 }
 982
 983 /* must be called under the client_lock */
 984 static inline void
 985 renew_client_locked(struct nfs4_client *clp)
 986 {
 987         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 988
 989         if (is_client_expired(clp)) {
 990                 WARN_ON(1);
 991                 printk("%s: client (clientid %08x/%08x) already expired\n",
 992                         __func__,
 993                         clp->cl_clientid.cl_boot,
 994                         clp->cl_clientid.cl_id);
 995                 return;
 996         }
 997
 998         dprintk("renewing client (clientid %08x/%08x)\n",
 999                         clp->cl_clientid.cl_boot,
1000                         clp->cl_clientid.cl_id);
1001         list_move_tail(&clp->cl_lru, &nn->client_lru);
1002         clp->cl_time = get_seconds();
1003 }
1004
1005 static inline void
1006 renew_client(struct nfs4_client *clp)
1007 {
1008         spin_lock(&client_lock);
1009         renew_client_locked(clp);
1010         spin_unlock(&client_lock);
1011 }
1012
1013 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1014 static int
1015 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1016 {
1017         if (clid->cl_boot == nn->boot_time)
1018                 return 0;
1019         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1020                 clid->cl_boot, clid->cl_id, nn->boot_time);
1021         return 1;
1022 }
1023
1024 /*
1025  * XXX Should we use a slab cache ?
1026  * This type of memory management is somewhat inefficient, but we use it
1027  * anyway since SETCLIENTID is not a common operation.
1028  */
1029 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1030 {
1031         struct nfs4_client *clp;
1032
1033         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1034         if (clp == NULL)
1035                 return NULL;
1036         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1037         if (clp->cl_name.data == NULL) {
1038                 kfree(clp);
1039                 return NULL;
1040         }
1041         clp->cl_name.len = name.len;
1042         return clp;
1043 }
1044
1045 static inline void
1046 free_client(struct nfs4_client *clp)
1047 {
1048         lockdep_assert_held(&client_lock);
1049         while (!list_empty(&clp->cl_sessions)) {
1050                 struct nfsd4_session *ses;
1051                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1052                                 se_perclnt);
1053                 list_del(&ses->se_perclnt);
1054                 nfsd4_put_session_locked(ses);
1055         }
1056         free_svc_cred(&clp->cl_cred);
1057         kfree(clp->cl_name.data);
1058         kfree(clp);
1059 }
1060
1061 void
1062 release_session_client(struct nfsd4_session *session)
1063 {
1064         struct nfs4_client *clp = session->se_client;
1065
1066         if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
1067                 return;
1068         if (is_client_expired(clp)) {
1069                 free_client(clp);
1070                 session->se_client = NULL;
1071         } else
1072                 renew_client_locked(clp);
1073         spin_unlock(&client_lock);
1074 }
1075
1076 /* must be called under the client_lock */
1077 static inline void
1078 unhash_client_locked(struct nfs4_client *clp)
1079 {
1080         struct nfsd4_session *ses;
1081
1082         mark_client_expired(clp);
1083         list_del(&clp->cl_lru);
1084         spin_lock(&clp->cl_lock);
1085         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1086                 list_del_init(&ses->se_hash);
1087         spin_unlock(&clp->cl_lock);
1088 }
1089
1090 static void
1091 destroy_client(struct nfs4_client *clp)
1092 {
1093         struct nfs4_openowner *oo;
1094         struct nfs4_delegation *dp;
1095         struct list_head reaplist;
1096         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1097
1098         INIT_LIST_HEAD(&reaplist);
1099         spin_lock(&recall_lock);
1100         while (!list_empty(&clp->cl_delegations)) {
1101                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1102                 list_del_init(&dp->dl_perclnt);
1103                 list_move(&dp->dl_recall_lru, &reaplist);
1104         }
1105         spin_unlock(&recall_lock);
1106         while (!list_empty(&reaplist)) {
1107                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1108                 unhash_delegation(dp);
1109         }
1110         while (!list_empty(&clp->cl_openowners)) {
1111                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1112                 release_openowner(oo);
1113         }
1114         nfsd4_shutdown_callback(clp);
1115         if (clp->cl_cb_conn.cb_xprt)
1116                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1117         list_del(&clp->cl_idhash);
1118         if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1119                 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1120         else
1121                 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1122         spin_lock(&client_lock);
1123         unhash_client_locked(clp);
1124         if (atomic_read(&clp->cl_refcount) == 0)
1125                 free_client(clp);
1126         spin_unlock(&client_lock);
1127 }
1128
1129 static void expire_client(struct nfs4_client *clp)
1130 {
1131         nfsd4_client_record_remove(clp);
1132         destroy_client(clp);
1133 }
1134
1135 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1136 {
1137         memcpy(target->cl_verifier.data, source->data,
1138                         sizeof(target->cl_verifier.data));
1139 }
1140
1141 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1142 {
1143         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1144         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1145 }
1146
1147 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1148 {
1149         if (source->cr_principal) {
1150                 target->cr_principal =
1151                                 kstrdup(source->cr_principal, GFP_KERNEL);
1152                 if (target->cr_principal == NULL)
1153                         return -ENOMEM;
1154         } else
1155                 target->cr_principal = NULL;
1156         target->cr_flavor = source->cr_flavor;
1157         target->cr_uid = source->cr_uid;
1158         target->cr_gid = source->cr_gid;
1159         target->cr_group_info = source->cr_group_info;
1160         get_group_info(target->cr_group_info);
1161         return 0;
1162 }
1163
1164 static long long
1165 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1166 {
1167         long long res;
1168
1169         res = o1->len - o2->len;
1170         if (res)
1171                 return res;
1172         return (long long)memcmp(o1->data, o2->data, o1->len);
1173 }
1174
1175 static int same_name(const char *n1, const char *n2)
1176 {
1177         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1178 }
1179
1180 static int
1181 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1182 {
1183         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1184 }
1185
1186 static int
1187 same_clid(clientid_t *cl1, clientid_t *cl2)
1188 {
1189         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1190 }
1191
1192 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1193 {
1194         int i;
1195
1196         if (g1->ngroups != g2->ngroups)
1197                 return false;
1198         for (i=0; i<g1->ngroups; i++)
1199                 if (GROUP_AT(g1, i) != GROUP_AT(g2, i))
1200                         return false;
1201         return true;
1202 }
1203
1204 /*
1205  * RFC 3530 language requires clid_inuse be returned when the
1206  * "principal" associated with a requests differs from that previously
1207  * used.  We use uid, gid's, and gss principal string as our best
1208  * approximation.  We also don't want to allow non-gss use of a client
1209  * established using gss: in theory cr_principal should catch that
1210  * change, but in practice cr_principal can be null even in the gss case
1211  * since gssd doesn't always pass down a principal string.
1212  */
1213 static bool is_gss_cred(struct svc_cred *cr)
1214 {
1215         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1216         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1217 }
1218
1219
1220 static bool
1221 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1222 {
1223         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1224                 || (cr1->cr_uid != cr2->cr_uid)
1225                 || (cr1->cr_gid != cr2->cr_gid)
1226                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1227                 return false;
1228         if (cr1->cr_principal == cr2->cr_principal)
1229                 return true;
1230         if (!cr1->cr_principal || !cr2->cr_principal)
1231                 return false;
1232         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1233 }
1234
1235 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
1236 {
1237         static u32 current_clientid = 1;
1238
1239         clp->cl_clientid.cl_boot = nn->boot_time;
1240         clp->cl_clientid.cl_id = current_clientid++;
1241 }
1242
1243 static void gen_confirm(struct nfs4_client *clp)
1244 {
1245         __be32 verf[2];
1246         static u32 i;
1247
1248         verf[0] = (__be32)get_seconds();
1249         verf[1] = (__be32)i++;
1250         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1251 }
1252
1253 static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1254 {
1255         return idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1256 }
1257
1258 static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1259 {
1260         struct nfs4_stid *s;
1261
1262         s = find_stateid(cl, t);
1263         if (!s)
1264                 return NULL;
1265         if (typemask & s->sc_type)
1266                 return s;
1267         return NULL;
1268 }
1269
1270 static struct nfs4_client *create_client(struct xdr_netobj name,
1271                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1272 {
1273         struct nfs4_client *clp;
1274         struct sockaddr *sa = svc_addr(rqstp);
1275         int ret;
1276         struct net *net = SVC_NET(rqstp);
1277
1278         clp = alloc_client(name);
1279         if (clp == NULL)
1280                 return NULL;
1281
1282         INIT_LIST_HEAD(&clp->cl_sessions);
1283         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1284         if (ret) {
1285                 spin_lock(&client_lock);
1286                 free_client(clp);
1287                 spin_unlock(&client_lock);
1288                 return NULL;
1289         }
1290         idr_init(&clp->cl_stateids);
1291         atomic_set(&clp->cl_refcount, 0);
1292         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1293         INIT_LIST_HEAD(&clp->cl_idhash);
1294         INIT_LIST_HEAD(&clp->cl_openowners);
1295         INIT_LIST_HEAD(&clp->cl_delegations);
1296         INIT_LIST_HEAD(&clp->cl_lru);
1297         INIT_LIST_HEAD(&clp->cl_callbacks);
1298         spin_lock_init(&clp->cl_lock);
1299         nfsd4_init_callback(&clp->cl_cb_null);
1300         clp->cl_time = get_seconds();
1301         clear_bit(0, &clp->cl_cb_slot_busy);
1302         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1303         copy_verf(clp, verf);
1304         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1305         gen_confirm(clp);
1306         clp->cl_cb_session = NULL;
1307         clp->net = net;
1308         return clp;
1309 }
1310
1311 static void
1312 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
1313 {
1314         struct rb_node **new = &(root->rb_node), *parent = NULL;
1315         struct nfs4_client *clp;
1316
1317         while (*new) {
1318                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
1319                 parent = *new;
1320
1321                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
1322                         new = &((*new)->rb_left);
1323                 else
1324                         new = &((*new)->rb_right);
1325         }
1326
1327         rb_link_node(&new_clp->cl_namenode, parent, new);
1328         rb_insert_color(&new_clp->cl_namenode, root);
1329 }
1330
1331 static struct nfs4_client *
1332 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
1333 {
1334         long long cmp;
1335         struct rb_node *node = root->rb_node;
1336         struct nfs4_client *clp;
1337
1338         while (node) {
1339                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
1340                 cmp = compare_blob(&clp->cl_name, name);
1341                 if (cmp > 0)
1342                         node = node->rb_left;
1343                 else if (cmp < 0)
1344                         node = node->rb_right;
1345                 else
1346                         return clp;
1347         }
1348         return NULL;
1349 }
1350
1351 static void
1352 add_to_unconfirmed(struct nfs4_client *clp)
1353 {
1354         unsigned int idhashval;
1355         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1356
1357         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1358         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
1359         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1360         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
1361         renew_client(clp);
1362 }
1363
1364 static void
1365 move_to_confirmed(struct nfs4_client *clp)
1366 {
1367         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1368         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1369
1370         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1371         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
1372         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1373         add_clp_to_name_tree(clp, &nn->conf_name_tree);
1374         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1375         renew_client(clp);
1376 }
1377
1378 static struct nfs4_client *
1379 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1380 {
1381         struct nfs4_client *clp;
1382         unsigned int idhashval = clientid_hashval(clid->cl_id);
1383
1384         list_for_each_entry(clp, &nn->conf_id_hashtbl[idhashval], cl_idhash) {
1385                 if (same_clid(&clp->cl_clientid, clid)) {
1386                         if ((bool)clp->cl_minorversion != sessions)
1387                                 return NULL;
1388                         renew_client(clp);
1389                         return clp;
1390                 }
1391         }
1392         return NULL;
1393 }
1394
1395 static struct nfs4_client *
1396 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1397 {
1398         struct nfs4_client *clp;
1399         unsigned int idhashval = clientid_hashval(clid->cl_id);
1400
1401         list_for_each_entry(clp, &nn->unconf_id_hashtbl[idhashval], cl_idhash) {
1402                 if (same_clid(&clp->cl_clientid, clid)) {
1403                         if ((bool)clp->cl_minorversion != sessions)
1404                                 return NULL;
1405                         return clp;
1406                 }
1407         }
1408         return NULL;
1409 }
1410
1411 static bool clp_used_exchangeid(struct nfs4_client *clp)
1412 {
1413         return clp->cl_exchange_flags != 0;
1414 }
1415
1416 static struct nfs4_client *
1417 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1418 {
1419         return find_clp_in_name_tree(name, &nn->conf_name_tree);
1420 }
1421
1422 static struct nfs4_client *
1423 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1424 {
1425         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
1426 }
1427
1428 static void
1429 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1430 {
1431         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1432         struct sockaddr *sa = svc_addr(rqstp);
1433         u32 scopeid = rpc_get_scope_id(sa);
1434         unsigned short expected_family;
1435
1436         /* Currently, we only support tcp and tcp6 for the callback channel */
1437         if (se->se_callback_netid_len == 3 &&
1438             !memcmp(se->se_callback_netid_val, "tcp", 3))
1439                 expected_family = AF_INET;
1440         else if (se->se_callback_netid_len == 4 &&
1441                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1442                 expected_family = AF_INET6;
1443         else
1444                 goto out_err;
1445
1446         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
1447                                             se->se_callback_addr_len,
1448                                             (struct sockaddr *)&conn->cb_addr,
1449                                             sizeof(conn->cb_addr));
1450
1451         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1452                 goto out_err;
1453
1454         if (conn->cb_addr.ss_family == AF_INET6)
1455                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1456
1457         conn->cb_prog = se->se_callback_prog;
1458         conn->cb_ident = se->se_callback_ident;
1459         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1460         return;
1461 out_err:
1462         conn->cb_addr.ss_family = AF_UNSPEC;
1463         conn->cb_addrlen = 0;
1464         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1465                 "will not receive delegations\n",
1466                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1467
1468         return;
1469 }
1470
1471 /*
1472  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1473  */
1474 void
1475 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1476 {
1477         struct nfsd4_slot *slot = resp->cstate.slot;
1478         unsigned int base;
1479
1480         dprintk("--> %s slot %p\n", __func__, slot);
1481
1482         slot->sl_opcnt = resp->opcnt;
1483         slot->sl_status = resp->cstate.status;
1484
1485         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
1486         if (nfsd4_not_cached(resp)) {
1487                 slot->sl_datalen = 0;
1488                 return;
1489         }
1490         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1491         base = (char *)resp->cstate.datap -
1492                                         (char *)resp->xbuf->head[0].iov_base;
1493         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1494                                     slot->sl_datalen))
1495                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1496         return;
1497 }
1498
1499 /*
1500  * Encode the replay sequence operation from the slot values.
1501  * If cachethis is FALSE encode the uncached rep error on the next
1502  * operation which sets resp->p and increments resp->opcnt for
1503  * nfs4svc_encode_compoundres.
1504  *
1505  */
1506 static __be32
1507 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1508                           struct nfsd4_compoundres *resp)
1509 {
1510         struct nfsd4_op *op;
1511         struct nfsd4_slot *slot = resp->cstate.slot;
1512
1513         /* Encode the replayed sequence operation */
1514         op = &args->ops[resp->opcnt - 1];
1515         nfsd4_encode_operation(resp, op);
1516
1517         /* Return nfserr_retry_uncached_rep in next operation. */
1518         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
1519                 op = &args->ops[resp->opcnt++];
1520                 op->status = nfserr_retry_uncached_rep;
1521                 nfsd4_encode_operation(resp, op);
1522         }
1523         return op->status;
1524 }
1525
1526 /*
1527  * The sequence operation is not cached because we can use the slot and
1528  * session values.
1529  */
1530 __be32
1531 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1532                          struct nfsd4_sequence *seq)
1533 {
1534         struct nfsd4_slot *slot = resp->cstate.slot;
1535         __be32 status;
1536
1537         dprintk("--> %s slot %p\n", __func__, slot);
1538
1539         /* Either returns 0 or nfserr_retry_uncached */
1540         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1541         if (status == nfserr_retry_uncached_rep)
1542                 return status;
1543
1544         /* The sequence operation has been encoded, cstate->datap set. */
1545         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1546
1547         resp->opcnt = slot->sl_opcnt;
1548         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1549         status = slot->sl_status;
1550
1551         return status;
1552 }
1553
1554 /*
1555  * Set the exchange_id flags returned by the server.
1556  */
1557 static void
1558 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1559 {
1560         /* pNFS is not supported */
1561         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1562
1563         /* Referrals are supported, Migration is not. */
1564         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1565
1566         /* set the wire flags to return to client. */
1567         clid->flags = new->cl_exchange_flags;
1568 }
1569
1570 static bool client_has_state(struct nfs4_client *clp)
1571 {
1572         /*
1573          * Note clp->cl_openowners check isn't quite right: there's no
1574          * need to count owners without stateid's.
1575          *
1576          * Also note we should probably be using this in 4.0 case too.
1577          */
1578         return !list_empty(&clp->cl_openowners)
1579                 || !list_empty(&clp->cl_delegations)
1580                 || !list_empty(&clp->cl_sessions);
1581 }
1582
1583 __be32
1584 nfsd4_exchange_id(struct svc_rqst *rqstp,
1585                   struct nfsd4_compound_state *cstate,
1586                   struct nfsd4_exchange_id *exid)
1587 {
1588         struct nfs4_client *unconf, *conf, *new;
1589         __be32 status;
1590         char                    addr_str[INET6_ADDRSTRLEN];
1591         nfs4_verifier           verf = exid->verifier;
1592         struct sockaddr         *sa = svc_addr(rqstp);
1593         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
1594         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1595
1596         rpc_ntop(sa, addr_str, sizeof(addr_str));
1597         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1598                 "ip_addr=%s flags %x, spa_how %d\n",
1599                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1600                 addr_str, exid->flags, exid->spa_how);
1601
1602         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1603                 return nfserr_inval;
1604
1605         /* Currently only support SP4_NONE */
1606         switch (exid->spa_how) {
1607         case SP4_NONE:
1608                 break;
1609         default:                                /* checked by xdr code */
1610                 WARN_ON_ONCE(1);
1611         case SP4_SSV:
1612         case SP4_MACH_CRED:
1613                 return nfserr_serverfault;      /* no excuse :-/ */
1614         }
1615
1616         /* Cases below refer to rfc 5661 section 18.35.4: */
1617         nfs4_lock_state();
1618         conf = find_confirmed_client_by_name(&exid->clname, nn);
1619         if (conf) {
1620                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
1621                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
1622
1623                 if (update) {
1624                         if (!clp_used_exchangeid(conf)) { /* buggy client */
1625                                 status = nfserr_inval;
1626                                 goto out;
1627                         }
1628                         if (!creds_match) { /* case 9 */
1629                                 status = nfserr_perm;
1630                                 goto out;
1631                         }
1632                         if (!verfs_match) { /* case 8 */
1633                                 status = nfserr_not_same;
1634                                 goto out;
1635                         }
1636                         /* case 6 */
1637                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1638                         new = conf;
1639                         goto out_copy;
1640                 }
1641                 if (!creds_match) { /* case 3 */
1642                         if (client_has_state(conf)) {
1643                                 status = nfserr_clid_inuse;
1644                                 goto out;
1645                         }
1646                         expire_client(conf);
1647                         goto out_new;
1648                 }
1649                 if (verfs_match) { /* case 2 */
1650                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
1651                         new = conf;
1652                         goto out_copy;
1653                 }
1654                 /* case 5, client reboot */
1655                 goto out_new;
1656         }
1657
1658         if (update) { /* case 7 */
1659                 status = nfserr_noent;
1660                 goto out;
1661         }
1662
1663         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
1664         if (unconf) /* case 4, possible retry or client restart */
1665                 expire_client(unconf);
1666
1667         /* case 1 (normal case) */
1668 out_new:
1669         new = create_client(exid->clname, rqstp, &verf);
1670         if (new == NULL) {
1671                 status = nfserr_jukebox;
1672                 goto out;
1673         }
1674         new->cl_minorversion = 1;
1675
1676         gen_clid(new, nn);
1677         add_to_unconfirmed(new);
1678 out_copy:
1679         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1680         exid->clientid.cl_id = new->cl_clientid.cl_id;
1681
1682         exid->seqid = new->cl_cs_slot.sl_seqid + 1;
1683         nfsd4_set_ex_flags(new, exid);
1684
1685         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1686                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1687         status = nfs_ok;
1688
1689 out:
1690         nfs4_unlock_state();
1691         return status;
1692 }
1693
1694 static __be32
1695 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1696 {
1697         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1698                 slot_seqid);
1699
1700         /* The slot is in use, and no response has been sent. */
1701         if (slot_inuse) {
1702                 if (seqid == slot_seqid)
1703                         return nfserr_jukebox;
1704                 else
1705                         return nfserr_seq_misordered;
1706         }
1707         /* Note unsigned 32-bit arithmetic handles wraparound: */
1708         if (likely(seqid == slot_seqid + 1))
1709                 return nfs_ok;
1710         if (seqid == slot_seqid)
1711                 return nfserr_replay_cache;
1712         return nfserr_seq_misordered;
1713 }
1714
1715 /*
1716  * Cache the create session result into the create session single DRC
1717  * slot cache by saving the xdr structure. sl_seqid has been set.
1718  * Do this for solo or embedded create session operations.
1719  */
1720 static void
1721 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1722                            struct nfsd4_clid_slot *slot, __be32 nfserr)
1723 {
1724         slot->sl_status = nfserr;
1725         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1726 }
1727
1728 static __be32
1729 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1730                             struct nfsd4_clid_slot *slot)
1731 {
1732         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1733         return slot->sl_status;
1734 }
1735
1736 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1737                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1738                         1 +     /* MIN tag is length with zero, only length */ \
1739                         3 +     /* version, opcount, opcode */ \
1740                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1741                                 /* seqid, slotID, slotID, cache */ \
1742                         4 ) * sizeof(__be32))
1743
1744 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1745                         2 +     /* verifier: AUTH_NULL, length 0 */\
1746                         1 +     /* status */ \
1747                         1 +     /* MIN tag is length with zero, only length */ \
1748                         3 +     /* opcount, opcode, opstatus*/ \
1749                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1750                                 /* seqid, slotID, slotID, slotID, status */ \
1751                         5 ) * sizeof(__be32))
1752
1753 static bool check_forechannel_attrs(struct nfsd4_channel_attrs fchannel)
1754 {
1755         return fchannel.maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ
1756                 || fchannel.maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ;
1757 }
1758
1759 __be32
1760 nfsd4_create_session(struct svc_rqst *rqstp,
1761                      struct nfsd4_compound_state *cstate,
1762                      struct nfsd4_create_session *cr_ses)
1763 {
1764         struct sockaddr *sa = svc_addr(rqstp);
1765         struct nfs4_client *conf, *unconf;
1766         struct nfsd4_session *new;
1767         struct nfsd4_conn *conn;
1768         struct nfsd4_clid_slot *cs_slot = NULL;
1769         __be32 status = 0;
1770         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1771
1772         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1773                 return nfserr_inval;
1774         if (check_forechannel_attrs(cr_ses->fore_channel))
1775                 return nfserr_toosmall;
1776         new = alloc_session(&cr_ses->fore_channel);
1777         if (!new)
1778                 return nfserr_jukebox;
1779         status = nfserr_jukebox;
1780         conn = alloc_conn_from_crses(rqstp, cr_ses);
1781         if (!conn)
1782                 goto out_free_session;
1783
1784         nfs4_lock_state();
1785         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
1786         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
1787
1788         if (conf) {
1789                 cs_slot = &conf->cl_cs_slot;
1790                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1791                 if (status == nfserr_replay_cache) {
1792                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1793                         goto out_free_conn;
1794                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1795                         status = nfserr_seq_misordered;
1796                         goto out_free_conn;
1797                 }
1798         } else if (unconf) {
1799                 struct nfs4_client *old;
1800                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1801                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1802                         status = nfserr_clid_inuse;
1803                         goto out_free_conn;
1804                 }
1805                 cs_slot = &unconf->cl_cs_slot;
1806                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1807                 if (status) {
1808                         /* an unconfirmed replay returns misordered */
1809                         status = nfserr_seq_misordered;
1810                         goto out_free_conn;
1811                 }
1812                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
1813                 if (old)
1814                         expire_client(old);
1815                 move_to_confirmed(unconf);
1816                 conf = unconf;
1817         } else {
1818                 status = nfserr_stale_clientid;
1819                 goto out_free_conn;
1820         }
1821         status = nfs_ok;
1822         /*
1823          * We do not support RDMA or persistent sessions
1824          */
1825         cr_ses->flags &= ~SESSION4_PERSIST;
1826         cr_ses->flags &= ~SESSION4_RDMA;
1827
1828         init_session(rqstp, new, conf, cr_ses);
1829         nfsd4_init_conn(rqstp, conn, new);
1830
1831         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1832                NFS4_MAX_SESSIONID_LEN);
1833         memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1834                 sizeof(struct nfsd4_channel_attrs));
1835         cs_slot->sl_seqid++;
1836         cr_ses->seqid = cs_slot->sl_seqid;
1837
1838         /* cache solo and embedded create sessions under the state lock */
1839         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1840 out:
1841         nfs4_unlock_state();
1842         dprintk("%s returns %d\n", __func__, ntohl(status));
1843         return status;
1844 out_free_conn:
1845         free_conn(conn);
1846 out_free_session:
1847         __free_session(new);
1848         goto out;
1849 }
1850
1851 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1852 {
1853         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1854         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1855
1856         return argp->opcnt == resp->opcnt;
1857 }
1858
1859 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1860 {
1861         switch (*dir) {
1862         case NFS4_CDFC4_FORE:
1863         case NFS4_CDFC4_BACK:
1864                 return nfs_ok;
1865         case NFS4_CDFC4_FORE_OR_BOTH:
1866         case NFS4_CDFC4_BACK_OR_BOTH:
1867                 *dir = NFS4_CDFC4_BOTH;
1868                 return nfs_ok;
1869         };
1870         return nfserr_inval;
1871 }
1872
1873 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
1874 {
1875         struct nfsd4_session *session = cstate->session;
1876
1877         spin_lock(&client_lock);
1878         session->se_cb_prog = bc->bc_cb_program;
1879         session->se_cb_sec = bc->bc_cb_sec;
1880         spin_unlock(&client_lock);
1881
1882         nfsd4_probe_callback(session->se_client);
1883
1884         return nfs_ok;
1885 }
1886
1887 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1888                      struct nfsd4_compound_state *cstate,
1889                      struct nfsd4_bind_conn_to_session *bcts)
1890 {
1891         __be32 status;
1892         struct nfsd4_conn *conn;
1893
1894         if (!nfsd4_last_compound_op(rqstp))
1895                 return nfserr_not_only_op;
1896         spin_lock(&client_lock);
1897         cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid, SVC_NET(rqstp));
1898         /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1899          * client_lock iself: */
1900         if (cstate->session) {
1901                 nfsd4_get_session(cstate->session);
1902                 atomic_inc(&cstate->session->se_client->cl_refcount);
1903         }
1904         spin_unlock(&client_lock);
1905         if (!cstate->session)
1906                 return nfserr_badsession;
1907
1908         status = nfsd4_map_bcts_dir(&bcts->dir);
1909         if (status)
1910                 return status;
1911         conn = alloc_conn(rqstp, bcts->dir);
1912         if (!conn)
1913                 return nfserr_jukebox;
1914         nfsd4_init_conn(rqstp, conn, cstate->session);
1915         return nfs_ok;
1916 }
1917
1918 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1919 {
1920         if (!session)
1921                 return 0;
1922         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1923 }
1924
1925 __be32
1926 nfsd4_destroy_session(struct svc_rqst *r,
1927                       struct nfsd4_compound_state *cstate,
1928                       struct nfsd4_destroy_session *sessionid)
1929 {
1930         struct nfsd4_session *ses;
1931         __be32 status = nfserr_badsession;
1932
1933         /* Notes:
1934          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1935          * - Should we return nfserr_back_chan_busy if waiting for
1936          *   callbacks on to-be-destroyed session?
1937          * - Do we need to clear any callback info from previous session?
1938          */
1939
1940         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1941                 if (!nfsd4_last_compound_op(r))
1942                         return nfserr_not_only_op;
1943         }
1944         dump_sessionid(__func__, &sessionid->sessionid);
1945         spin_lock(&client_lock);
1946         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, SVC_NET(r));
1947         if (!ses) {
1948                 spin_unlock(&client_lock);
1949                 goto out;
1950         }
1951
1952         unhash_session(ses);
1953         spin_unlock(&client_lock);
1954
1955         nfs4_lock_state();
1956         nfsd4_probe_callback_sync(ses->se_client);
1957         nfs4_unlock_state();
1958
1959         spin_lock(&client_lock);
1960         nfsd4_del_conns(ses);
1961         nfsd4_put_session_locked(ses);
1962         spin_unlock(&client_lock);
1963         status = nfs_ok;
1964 out:
1965         dprintk("%s returns %d\n", __func__, ntohl(status));
1966         return status;
1967 }
1968
1969 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1970 {
1971         struct nfsd4_conn *c;
1972
1973         list_for_each_entry(c, &s->se_conns, cn_persession) {
1974                 if (c->cn_xprt == xpt) {
1975                         return c;
1976                 }
1977         }
1978         return NULL;
1979 }
1980
1981 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1982 {
1983         struct nfs4_client *clp = ses->se_client;
1984         struct nfsd4_conn *c;
1985         int ret;
1986
1987         spin_lock(&clp->cl_lock);
1988         c = __nfsd4_find_conn(new->cn_xprt, ses);
1989         if (c) {
1990                 spin_unlock(&clp->cl_lock);
1991                 free_conn(new);
1992                 return;
1993         }
1994         __nfsd4_hash_conn(new, ses);
1995         spin_unlock(&clp->cl_lock);
1996         ret = nfsd4_register_conn(new);
1997         if (ret)
1998                 /* oops; xprt is already down: */
1999                 nfsd4_conn_lost(&new->cn_xpt_user);
2000         return;
2001 }
2002
2003 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2004 {
2005         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2006
2007         return args->opcnt > session->se_fchannel.maxops;
2008 }
2009
2010 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2011                                   struct nfsd4_session *session)
2012 {
2013         struct xdr_buf *xb = &rqstp->rq_arg;
2014
2015         return xb->len > session->se_fchannel.maxreq_sz;
2016 }
2017
2018 __be32
2019 nfsd4_sequence(struct svc_rqst *rqstp,
2020                struct nfsd4_compound_state *cstate,
2021                struct nfsd4_sequence *seq)
2022 {
2023         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2024         struct nfsd4_session *session;
2025         struct nfsd4_slot *slot;
2026         struct nfsd4_conn *conn;
2027         __be32 status;
2028
2029         if (resp->opcnt != 1)
2030                 return nfserr_sequence_pos;
2031
2032         /*
2033          * Will be either used or freed by nfsd4_sequence_check_conn
2034          * below.
2035          */
2036         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2037         if (!conn)
2038                 return nfserr_jukebox;
2039
2040         spin_lock(&client_lock);
2041         status = nfserr_badsession;
2042         session = find_in_sessionid_hashtbl(&seq->sessionid, SVC_NET(rqstp));
2043         if (!session)
2044                 goto out;
2045
2046         status = nfserr_too_many_ops;
2047         if (nfsd4_session_too_many_ops(rqstp, session))
2048                 goto out;
2049
2050         status = nfserr_req_too_big;
2051         if (nfsd4_request_too_big(rqstp, session))
2052                 goto out;
2053
2054         status = nfserr_badslot;
2055         if (seq->slotid >= session->se_fchannel.maxreqs)
2056                 goto out;
2057
2058         slot = session->se_slots[seq->slotid];
2059         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2060
2061         /* We do not negotiate the number of slots yet, so set the
2062          * maxslots to the session maxreqs which is used to encode
2063          * sr_highest_slotid and the sr_target_slot id to maxslots */
2064         seq->maxslots = session->se_fchannel.maxreqs;
2065
2066         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2067                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2068         if (status == nfserr_replay_cache) {
2069                 status = nfserr_seq_misordered;
2070                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2071                         goto out;
2072                 cstate->slot = slot;
2073                 cstate->session = session;
2074                 /* Return the cached reply status and set cstate->status
2075                  * for nfsd4_proc_compound processing */
2076                 status = nfsd4_replay_cache_entry(resp, seq);
2077                 cstate->status = nfserr_replay_cache;
2078                 goto out;
2079         }
2080         if (status)
2081                 goto out;
2082
2083         nfsd4_sequence_check_conn(conn, session);
2084         conn = NULL;
2085
2086         /* Success! bump slot seqid */
2087         slot->sl_seqid = seq->seqid;
2088         slot->sl_flags |= NFSD4_SLOT_INUSE;
2089         if (seq->cachethis)
2090                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2091         else
2092                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2093
2094         cstate->slot = slot;
2095         cstate->session = session;
2096
2097 out:
2098         /* Hold a session reference until done processing the compound. */
2099         if (cstate->session) {
2100                 struct nfs4_client *clp = session->se_client;
2101
2102                 nfsd4_get_session(cstate->session);
2103                 atomic_inc(&clp->cl_refcount);
2104                 switch (clp->cl_cb_state) {
2105                 case NFSD4_CB_DOWN:
2106                         seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2107                         break;
2108                 case NFSD4_CB_FAULT:
2109                         seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2110                         break;
2111                 default:
2112                         seq->status_flags = 0;
2113                 }
2114         }
2115         kfree(conn);
2116         spin_unlock(&client_lock);
2117         dprintk("%s: return %d\n", __func__, ntohl(status));
2118         return status;
2119 }
2120
2121 __be32
2122 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2123 {
2124         struct nfs4_client *conf, *unconf, *clp;
2125         __be32 status = 0;
2126         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2127
2128         nfs4_lock_state();
2129         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
2130         conf = find_confirmed_client(&dc->clientid, true, nn);
2131
2132         if (conf) {
2133                 clp = conf;
2134
2135                 if (!is_client_expired(conf) && client_has_state(conf)) {
2136                         status = nfserr_clientid_busy;
2137                         goto out;
2138                 }
2139
2140                 /* rfc5661 18.50.3 */
2141                 if (cstate->session && conf == cstate->session->se_client) {
2142                         status = nfserr_clientid_busy;
2143                         goto out;
2144                 }
2145         } else if (unconf)
2146                 clp = unconf;
2147         else {
2148                 status = nfserr_stale_clientid;
2149                 goto out;
2150         }
2151
2152         expire_client(clp);
2153 out:
2154         nfs4_unlock_state();
2155         dprintk("%s return %d\n", __func__, ntohl(status));
2156         return status;
2157 }
2158
2159 __be32
2160 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2161 {
2162         __be32 status = 0;
2163
2164         if (rc->rca_one_fs) {
2165                 if (!cstate->current_fh.fh_dentry)
2166                         return nfserr_nofilehandle;
2167                 /*
2168                  * We don't take advantage of the rca_one_fs case.
2169                  * That's OK, it's optional, we can safely ignore it.
2170                  */
2171                  return nfs_ok;
2172         }
2173
2174         nfs4_lock_state();
2175         status = nfserr_complete_already;
2176         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2177                              &cstate->session->se_client->cl_flags))
2178                 goto out;
2179
2180         status = nfserr_stale_clientid;
2181         if (is_client_expired(cstate->session->se_client))
2182                 /*
2183                  * The following error isn't really legal.
2184                  * But we only get here if the client just explicitly
2185                  * destroyed the client.  Surely it no longer cares what
2186                  * error it gets back on an operation for the dead
2187                  * client.
2188                  */
2189                 goto out;
2190
2191         status = nfs_ok;
2192         nfsd4_client_record_create(cstate->session->se_client);
2193 out:
2194         nfs4_unlock_state();
2195         return status;
2196 }
2197
2198 __be32
2199 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2200                   struct nfsd4_setclientid *setclid)
2201 {
2202         struct xdr_netobj       clname = setclid->se_name;
2203         nfs4_verifier           clverifier = setclid->se_verf;
2204         struct nfs4_client      *conf, *unconf, *new;
2205         __be32                  status;
2206         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2207
2208         /* Cases below refer to rfc 3530 section 14.2.33: */
2209         nfs4_lock_state();
2210         conf = find_confirmed_client_by_name(&clname, nn);
2211         if (conf) {
2212                 /* case 0: */
2213                 status = nfserr_clid_inuse;
2214                 if (clp_used_exchangeid(conf))
2215                         goto out;
2216                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2217                         char addr_str[INET6_ADDRSTRLEN];
2218                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2219                                  sizeof(addr_str));
2220                         dprintk("NFSD: setclientid: string in use by client "
2221                                 "at %s\n", addr_str);
2222                         goto out;
2223                 }
2224         }
2225         unconf = find_unconfirmed_client_by_name(&clname, nn);
2226         if (unconf)
2227                 expire_client(unconf);
2228         status = nfserr_jukebox;
2229         new = create_client(clname, rqstp, &clverifier);
2230         if (new == NULL)
2231                 goto out;
2232         if (conf && same_verf(&conf->cl_verifier, &clverifier))
2233                 /* case 1: probable callback update */
2234                 copy_clid(new, conf);
2235         else /* case 4 (new client) or cases 2, 3 (client reboot): */
2236                 gen_clid(new, nn);
2237         new->cl_minorversion = 0;
2238         gen_callback(new, setclid, rqstp);
2239         add_to_unconfirmed(new);
2240         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2241         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2242         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2243         status = nfs_ok;
2244 out:
2245         nfs4_unlock_state();
2246         return status;
2247 }
2248
2249
2250 __be32
2251 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2252                          struct nfsd4_compound_state *cstate,
2253                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2254 {
2255         struct nfs4_client *conf, *unconf;
2256         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2257         clientid_t * clid = &setclientid_confirm->sc_clientid;
2258         __be32 status;
2259         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2260
2261         if (STALE_CLIENTID(clid, nn))
2262                 return nfserr_stale_clientid;
2263         nfs4_lock_state();
2264
2265         conf = find_confirmed_client(clid, false, nn);
2266         unconf = find_unconfirmed_client(clid, false, nn);
2267         /*
2268          * We try hard to give out unique clientid's, so if we get an
2269          * attempt to confirm the same clientid with a different cred,
2270          * there's a bug somewhere.  Let's charitably assume it's our
2271          * bug.
2272          */
2273         status = nfserr_serverfault;
2274         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
2275                 goto out;
2276         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
2277                 goto out;
2278         /* cases below refer to rfc 3530 section 14.2.34: */
2279         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
2280                 if (conf && !unconf) /* case 2: probable retransmit */
2281                         status = nfs_ok;
2282                 else /* case 4: client hasn't noticed we rebooted yet? */
2283                         status = nfserr_stale_clientid;
2284                 goto out;
2285         }
2286         status = nfs_ok;
2287         if (conf) { /* case 1: callback update */
2288                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2289                 nfsd4_probe_callback(conf);
2290                 expire_client(unconf);
2291         } else { /* case 3: normal case; new or rebooted client */
2292                 conf = find_confirmed_client_by_name(&unconf->cl_name, nn);
2293                 if (conf)
2294                         expire_client(conf);
2295                 move_to_confirmed(unconf);
2296                 nfsd4_probe_callback(unconf);
2297         }
2298 out:
2299         nfs4_unlock_state();
2300         return status;
2301 }
2302
2303 static struct nfs4_file *nfsd4_alloc_file(void)
2304 {
2305         return kmem_cache_alloc(file_slab, GFP_KERNEL);
2306 }
2307
2308 /* OPEN Share state helper functions */
2309 static void nfsd4_init_file(struct nfs4_file *fp, struct inode *ino)
2310 {
2311         unsigned int hashval = file_hashval(ino);
2312
2313         atomic_set(&fp->fi_ref, 1);
2314         INIT_LIST_HEAD(&fp->fi_hash);
2315         INIT_LIST_HEAD(&fp->fi_stateids);
2316         INIT_LIST_HEAD(&fp->fi_delegations);
2317         fp->fi_inode = igrab(ino);
2318         fp->fi_had_conflict = false;
2319         fp->fi_lease = NULL;
2320         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2321         memset(fp->fi_access, 0, sizeof(fp->fi_access));
2322         spin_lock(&recall_lock);
2323         list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2324         spin_unlock(&recall_lock);
2325 }
2326
2327 static void
2328 nfsd4_free_slab(struct kmem_cache **slab)
2329 {
2330         if (*slab == NULL)
2331                 return;
2332         kmem_cache_destroy(*slab);
2333         *slab = NULL;
2334 }
2335
2336 void
2337 nfsd4_free_slabs(void)
2338 {
2339         nfsd4_free_slab(&openowner_slab);
2340         nfsd4_free_slab(&lockowner_slab);
2341         nfsd4_free_slab(&file_slab);
2342         nfsd4_free_slab(&stateid_slab);
2343         nfsd4_free_slab(&deleg_slab);
2344 }
2345
2346 int
2347 nfsd4_init_slabs(void)
2348 {
2349         openowner_slab = kmem_cache_create("nfsd4_openowners",
2350                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2351         if (openowner_slab == NULL)
2352                 goto out_nomem;
2353         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2354                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
2355         if (lockowner_slab == NULL)
2356                 goto out_nomem;
2357         file_slab = kmem_cache_create("nfsd4_files",
2358                         sizeof(struct nfs4_file), 0, 0, NULL);
2359         if (file_slab == NULL)
2360                 goto out_nomem;
2361         stateid_slab = kmem_cache_create("nfsd4_stateids",
2362                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2363         if (stateid_slab == NULL)
2364                 goto out_nomem;
2365         deleg_slab = kmem_cache_create("nfsd4_delegations",
2366                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2367         if (deleg_slab == NULL)
2368                 goto out_nomem;
2369         return 0;
2370 out_nomem:
2371         nfsd4_free_slabs();
2372         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2373         return -ENOMEM;
2374 }
2375
2376 void nfs4_free_openowner(struct nfs4_openowner *oo)
2377 {
2378         kfree(oo->oo_owner.so_owner.data);
2379         kmem_cache_free(openowner_slab, oo);
2380 }
2381
2382 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2383 {
2384         kfree(lo->lo_owner.so_owner.data);
2385         kmem_cache_free(lockowner_slab, lo);
2386 }
2387
2388 static void init_nfs4_replay(struct nfs4_replay *rp)
2389 {
2390         rp->rp_status = nfserr_serverfault;
2391         rp->rp_buflen = 0;
2392         rp->rp_buf = rp->rp_ibuf;
2393 }
2394
2395 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2396 {
2397         struct nfs4_stateowner *sop;
2398
2399         sop = kmem_cache_alloc(slab, GFP_KERNEL);
2400         if (!sop)
2401                 return NULL;
2402
2403         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2404         if (!sop->so_owner.data) {
2405                 kmem_cache_free(slab, sop);
2406                 return NULL;
2407         }
2408         sop->so_owner.len = owner->len;
2409
2410         INIT_LIST_HEAD(&sop->so_stateids);
2411         sop->so_client = clp;
2412         init_nfs4_replay(&sop->so_replay);
2413         return sop;
2414 }
2415
2416 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2417 {
2418         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2419
2420         list_add(&oo->oo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
2421         list_add(&oo->oo_perclient, &clp->cl_openowners);
2422 }
2423
2424 static struct nfs4_openowner *
2425 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2426         struct nfs4_openowner *oo;
2427
2428         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2429         if (!oo)
2430                 return NULL;
2431         oo->oo_owner.so_is_open_owner = 1;
2432         oo->oo_owner.so_seqid = open->op_seqid;
2433         oo->oo_flags = NFS4_OO_NEW;
2434         oo->oo_time = 0;
2435         oo->oo_last_closed_stid = NULL;
2436         INIT_LIST_HEAD(&oo->oo_close_lru);
2437         hash_openowner(oo, clp, strhashval);
2438         return oo;
2439 }
2440
2441 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2442         struct nfs4_openowner *oo = open->op_openowner;
2443         struct nfs4_client *clp = oo->oo_owner.so_client;
2444
2445         init_stid(&stp->st_stid, clp, NFS4_OPEN_STID);
2446         INIT_LIST_HEAD(&stp->st_lockowners);
2447         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2448         list_add(&stp->st_perfile, &fp->fi_stateids);
2449         stp->st_stateowner = &oo->oo_owner;
2450         get_nfs4_file(fp);
2451         stp->st_file = fp;
2452         stp->st_access_bmap = 0;
2453         stp->st_deny_bmap = 0;
2454         set_access(open->op_share_access, stp);
2455         set_deny(open->op_share_deny, stp);
2456         stp->st_openstp = NULL;
2457 }
2458
2459 static void
2460 move_to_close_lru(struct nfs4_openowner *oo, struct net *net)
2461 {
2462         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2463
2464         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2465
2466         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
2467         oo->oo_time = get_seconds();
2468 }
2469
2470 static int
2471 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2472                                                         clientid_t *clid)
2473 {
2474         return (sop->so_owner.len == owner->len) &&
2475                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2476                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2477 }
2478
2479 static struct nfs4_openowner *
2480 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
2481                         bool sessions, struct nfsd_net *nn)
2482 {
2483         struct nfs4_stateowner *so;
2484         struct nfs4_openowner *oo;
2485         struct nfs4_client *clp;
2486
2487         list_for_each_entry(so, &nn->ownerstr_hashtbl[hashval], so_strhash) {
2488                 if (!so->so_is_open_owner)
2489                         continue;
2490                 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2491                         oo = openowner(so);
2492                         clp = oo->oo_owner.so_client;
2493                         if ((bool)clp->cl_minorversion != sessions)
2494                                 return NULL;
2495                         renew_client(oo->oo_owner.so_client);
2496                         return oo;
2497                 }
2498         }
2499         return NULL;
2500 }
2501
2502 /* search file_hashtbl[] for file */
2503 static struct nfs4_file *
2504 find_file(struct inode *ino)
2505 {
2506         unsigned int hashval = file_hashval(ino);
2507         struct nfs4_file *fp;
2508
2509         spin_lock(&recall_lock);
2510         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2511                 if (fp->fi_inode == ino) {
2512                         get_nfs4_file(fp);
2513                         spin_unlock(&recall_lock);
2514                         return fp;
2515                 }
2516         }
2517         spin_unlock(&recall_lock);
2518         return NULL;
2519 }
2520
2521 /*
2522  * Called to check deny when READ with all zero stateid or
2523  * WRITE with all zero or all one stateid
2524  */
2525 static __be32
2526 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2527 {
2528         struct inode *ino = current_fh->fh_dentry->d_inode;
2529         struct nfs4_file *fp;
2530         struct nfs4_ol_stateid *stp;
2531         __be32 ret;
2532
2533         dprintk("NFSD: nfs4_share_conflict\n");
2534
2535         fp = find_file(ino);
2536         if (!fp)
2537                 return nfs_ok;
2538         ret = nfserr_locked;
2539         /* Search for conflicting share reservations */
2540         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2541                 if (test_deny(deny_type, stp) ||
2542                     test_deny(NFS4_SHARE_DENY_BOTH, stp))
2543                         goto out;
2544         }
2545         ret = nfs_ok;
2546 out:
2547         put_nfs4_file(fp);
2548         return ret;
2549 }
2550
2551 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2552 {
2553         /* We're assuming the state code never drops its reference
2554          * without first removing the lease.  Since we're in this lease
2555          * callback (and since the lease code is serialized by the kernel
2556          * lock) we know the server hasn't removed the lease yet, we know
2557          * it's safe to take a reference: */
2558         atomic_inc(&dp->dl_count);
2559
2560         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2561
2562         /* only place dl_time is set. protected by lock_flocks*/
2563         dp->dl_time = get_seconds();
2564
2565         nfsd4_cb_recall(dp);
2566 }
2567
2568 /* Called from break_lease() with lock_flocks() held. */
2569 static void nfsd_break_deleg_cb(struct file_lock *fl)
2570 {
2571         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2572         struct nfs4_delegation *dp;
2573
2574         if (!fp) {
2575                 WARN(1, "(%p)->fl_owner NULL\n", fl);
2576                 return;
2577         }
2578         if (fp->fi_had_conflict) {
2579                 WARN(1, "duplicate break on %p\n", fp);
2580                 return;
2581         }
2582         /*
2583          * We don't want the locks code to timeout the lease for us;
2584          * we'll remove it ourself if a delegation isn't returned
2585          * in time:
2586          */
2587         fl->fl_break_time = 0;
2588
2589         spin_lock(&recall_lock);
2590         fp->fi_had_conflict = true;
2591         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2592                 nfsd_break_one_deleg(dp);
2593         spin_unlock(&recall_lock);
2594 }
2595
2596 static
2597 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2598 {
2599         if (arg & F_UNLCK)
2600                 return lease_modify(onlist, arg);
2601         else
2602                 return -EAGAIN;
2603 }
2604
2605 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2606         .lm_break = nfsd_break_deleg_cb,
2607         .lm_change = nfsd_change_deleg_cb,
2608 };
2609
2610 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2611 {
2612         if (nfsd4_has_session(cstate))
2613                 return nfs_ok;
2614         if (seqid == so->so_seqid - 1)
2615                 return nfserr_replay_me;
2616         if (seqid == so->so_seqid)
2617                 return nfs_ok;
2618         return nfserr_bad_seqid;
2619 }
2620
2621 __be32
2622 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2623                     struct nfsd4_open *open, struct nfsd_net *nn)
2624 {
2625         clientid_t *clientid = &open->op_clientid;
2626         struct nfs4_client *clp = NULL;
2627         unsigned int strhashval;
2628         struct nfs4_openowner *oo = NULL;
2629         __be32 status;
2630
2631         if (STALE_CLIENTID(&open->op_clientid, nn))
2632                 return nfserr_stale_clientid;
2633         /*
2634          * In case we need it later, after we've already created the
2635          * file and don't want to risk a further failure:
2636          */
2637         open->op_file = nfsd4_alloc_file();
2638         if (open->op_file == NULL)
2639                 return nfserr_jukebox;
2640
2641         strhashval = ownerstr_hashval(clientid->cl_id, &open->op_owner);
2642         oo = find_openstateowner_str(strhashval, open, cstate->minorversion, nn);
2643         open->op_openowner = oo;
2644         if (!oo) {
2645                 clp = find_confirmed_client(clientid, cstate->minorversion,
2646                                             nn);
2647                 if (clp == NULL)
2648                         return nfserr_expired;
2649                 goto new_owner;
2650         }
2651         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2652                 /* Replace unconfirmed owners without checking for replay. */
2653                 clp = oo->oo_owner.so_client;
2654                 release_openowner(oo);
2655                 open->op_openowner = NULL;
2656                 goto new_owner;
2657         }
2658         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2659         if (status)
2660                 return status;
2661         clp = oo->oo_owner.so_client;
2662         goto alloc_stateid;
2663 new_owner:
2664         oo = alloc_init_open_stateowner(strhashval, clp, open);
2665         if (oo == NULL)
2666                 return nfserr_jukebox;
2667         open->op_openowner = oo;
2668 alloc_stateid:
2669         open->op_stp = nfs4_alloc_stateid(clp);
2670         if (!open->op_stp)
2671                 return nfserr_jukebox;
2672         return nfs_ok;
2673 }
2674
2675 static inline __be32
2676 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2677 {
2678         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2679                 return nfserr_openmode;
2680         else
2681                 return nfs_ok;
2682 }
2683
2684 static int share_access_to_flags(u32 share_access)
2685 {
2686         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2687 }
2688
2689 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2690 {
2691         struct nfs4_stid *ret;
2692
2693         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2694         if (!ret)
2695                 return NULL;
2696         return delegstateid(ret);
2697 }
2698
2699 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
2700 {
2701         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
2702                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
2703 }
2704
2705 static __be32
2706 nfs4_check_deleg(struct nfs4_client *cl, struct nfs4_file *fp, struct nfsd4_open *open,
2707                 struct nfs4_delegation **dp)
2708 {
2709         int flags;
2710         __be32 status = nfserr_bad_stateid;
2711
2712         *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2713         if (*dp == NULL)
2714                 goto out;
2715         flags = share_access_to_flags(open->op_share_access);
2716         status = nfs4_check_delegmode(*dp, flags);
2717         if (status)
2718                 *dp = NULL;
2719 out:
2720         if (!nfsd4_is_deleg_cur(open))
2721                 return nfs_ok;
2722         if (status)
2723                 return status;
2724         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2725         return nfs_ok;
2726 }
2727
2728 static __be32
2729 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2730 {
2731         struct nfs4_ol_stateid *local;
2732         struct nfs4_openowner *oo = open->op_openowner;
2733
2734         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2735                 /* ignore lock owners */
2736                 if (local->st_stateowner->so_is_open_owner == 0)
2737                         continue;
2738                 /* remember if we have seen this open owner */
2739                 if (local->st_stateowner == &oo->oo_owner)
2740                         *stpp = local;
2741                 /* check for conflicting share reservations */
2742                 if (!test_share(local, open))
2743                         return nfserr_share_denied;
2744         }
2745         return nfs_ok;
2746 }
2747
2748 static inline int nfs4_access_to_access(u32 nfs4_access)
2749 {
2750         int flags = 0;
2751
2752         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2753                 flags |= NFSD_MAY_READ;
2754         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2755                 flags |= NFSD_MAY_WRITE;
2756         return flags;
2757 }
2758
2759 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2760                 struct svc_fh *cur_fh, struct nfsd4_open *open)
2761 {
2762         __be32 status;
2763         int oflag = nfs4_access_to_omode(open->op_share_access);
2764         int access = nfs4_access_to_access(open->op_share_access);
2765
2766         if (!fp->fi_fds[oflag]) {
2767                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2768                         &fp->fi_fds[oflag]);
2769                 if (status)
2770                         return status;
2771         }
2772         nfs4_file_get_access(fp, oflag);
2773
2774         return nfs_ok;
2775 }
2776
2777 static inline __be32
2778 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2779                 struct nfsd4_open *open)
2780 {
2781         struct iattr iattr = {
2782                 .ia_valid = ATTR_SIZE,
2783                 .ia_size = 0,
2784         };
2785         if (!open->op_truncate)
2786                 return 0;
2787         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2788                 return nfserr_inval;
2789         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2790 }
2791
2792 static __be32
2793 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)
2794 {
2795         u32 op_share_access = open->op_share_access;
2796         bool new_access;
2797         __be32 status;
2798
2799         new_access = !test_access(op_share_access, stp);
2800         if (new_access) {
2801                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2802                 if (status)
2803                         return status;
2804         }
2805         status = nfsd4_truncate(rqstp, cur_fh, open);
2806         if (status) {
2807                 if (new_access) {
2808                         int oflag = nfs4_access_to_omode(op_share_access);
2809                         nfs4_file_put_access(fp, oflag);
2810                 }
2811                 return status;
2812         }
2813         /* remember the open */
2814         set_access(op_share_access, stp);
2815         set_deny(open->op_share_deny, stp);
2816
2817         return nfs_ok;
2818 }
2819
2820
2821 static void
2822 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
2823 {
2824         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2825 }
2826
2827 /* Should we give out recallable state?: */
2828 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2829 {
2830         if (clp->cl_cb_state == NFSD4_CB_UP)
2831                 return true;
2832         /*
2833          * In the sessions case, since we don't have to establish a
2834          * separate connection for callbacks, we assume it's OK
2835          * until we hear otherwise:
2836          */
2837         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2838 }
2839
2840 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2841 {
2842         struct file_lock *fl;
2843
2844         fl = locks_alloc_lock();
2845         if (!fl)
2846                 return NULL;
2847         locks_init_lock(fl);
2848         fl->fl_lmops = &nfsd_lease_mng_ops;
2849         fl->fl_flags = FL_LEASE;
2850         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2851         fl->fl_end = OFFSET_MAX;
2852         fl->fl_owner = (fl_owner_t)(dp->dl_file);
2853         fl->fl_pid = current->tgid;
2854         return fl;
2855 }
2856
2857 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2858 {
2859         struct nfs4_file *fp = dp->dl_file;
2860         struct file_lock *fl;
2861         int status;
2862
2863         fl = nfs4_alloc_init_lease(dp, flag);
2864         if (!fl)
2865                 return -ENOMEM;
2866         fl->fl_file = find_readable_file(fp);
2867         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2868         status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2869         if (status) {
2870                 list_del_init(&dp->dl_perclnt);
2871                 locks_free_lock(fl);
2872                 return -ENOMEM;
2873         }
2874         fp->fi_lease = fl;
2875         fp->fi_deleg_file = get_file(fl->fl_file);
2876         atomic_set(&fp->fi_delegees, 1);
2877         list_add(&dp->dl_perfile, &fp->fi_delegations);
2878         return 0;
2879 }
2880
2881 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2882 {
2883         struct nfs4_file *fp = dp->dl_file;
2884
2885         if (!fp->fi_lease)
2886                 return nfs4_setlease(dp, flag);
2887         spin_lock(&recall_lock);
2888         if (fp->fi_had_conflict) {
2889                 spin_unlock(&recall_lock);
2890                 return -EAGAIN;
2891         }
2892         atomic_inc(&fp->fi_delegees);
2893         list_add(&dp->dl_perfile, &fp->fi_delegations);
2894         spin_unlock(&recall_lock);
2895         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2896         return 0;
2897 }
2898
2899 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
2900 {
2901         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2902         if (status == -EAGAIN)
2903                 open->op_why_no_deleg = WND4_CONTENTION;
2904         else {
2905                 open->op_why_no_deleg = WND4_RESOURCE;
2906                 switch (open->op_deleg_want) {
2907                 case NFS4_SHARE_WANT_READ_DELEG:
2908                 case NFS4_SHARE_WANT_WRITE_DELEG:
2909                 case NFS4_SHARE_WANT_ANY_DELEG:
2910                         break;
2911                 case NFS4_SHARE_WANT_CANCEL:
2912                         open->op_why_no_deleg = WND4_CANCELLED;
2913                         break;
2914                 case NFS4_SHARE_WANT_NO_DELEG:
2915                         WARN_ON_ONCE(1);
2916                 }
2917         }
2918 }
2919
2920 /*
2921  * Attempt to hand out a delegation.
2922  */
2923 static void
2924 nfs4_open_delegation(struct net *net, struct svc_fh *fh,
2925                      struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
2926 {
2927         struct nfs4_delegation *dp;
2928         struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
2929         int cb_up;
2930         int status = 0, flag = 0;
2931
2932         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
2933         flag = NFS4_OPEN_DELEGATE_NONE;
2934         open->op_recall = 0;
2935         switch (open->op_claim_type) {
2936                 case NFS4_OPEN_CLAIM_PREVIOUS:
2937                         if (!cb_up)
2938                                 open->op_recall = 1;
2939                         flag = open->op_delegate_type;
2940                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2941                                 goto out;
2942                         break;
2943                 case NFS4_OPEN_CLAIM_NULL:
2944                         /* Let's not give out any delegations till everyone's
2945                          * had the chance to reclaim theirs.... */
2946                         if (locks_in_grace(net))
2947                                 goto out;
2948                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
2949                                 goto out;
2950                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2951                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2952                         else
2953                                 flag = NFS4_OPEN_DELEGATE_READ;
2954                         break;
2955                 default:
2956                         goto out;
2957         }
2958
2959         dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh, flag);
2960         if (dp == NULL)
2961                 goto out_no_deleg;
2962         status = nfs4_set_delegation(dp, flag);
2963         if (status)
2964                 goto out_free;
2965
2966         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
2967
2968         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2969                 STATEID_VAL(&dp->dl_stid.sc_stateid));
2970 out:
2971         open->op_delegate_type = flag;
2972         if (flag == NFS4_OPEN_DELEGATE_NONE) {
2973                 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
2974                     open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2975                         dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2976
2977                 /* 4.1 client asking for a delegation? */
2978                 if (open->op_deleg_want)
2979                         nfsd4_open_deleg_none_ext(open, status);
2980         }
2981         return;
2982 out_free:
2983         nfs4_put_delegation(dp);
2984 out_no_deleg:
2985         flag = NFS4_OPEN_DELEGATE_NONE;
2986         goto out;
2987 }
2988
2989 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
2990                                         struct nfs4_delegation *dp)
2991 {
2992         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
2993             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
2994                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2995                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
2996         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
2997                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
2998                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2999                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
3000         }
3001         /* Otherwise the client must be confused wanting a delegation
3002          * it already has, therefore we don't return
3003          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
3004          */
3005 }
3006
3007 /*
3008  * called with nfs4_lock_state() held.
3009  */
3010 __be32
3011 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
3012 {
3013         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3014         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
3015         struct nfs4_file *fp = NULL;
3016         struct inode *ino = current_fh->fh_dentry->d_inode;
3017         struct nfs4_ol_stateid *stp = NULL;
3018         struct nfs4_delegation *dp = NULL;
3019         __be32 status;
3020
3021         /*
3022          * Lookup file; if found, lookup stateid and check open request,
3023          * and check for delegations in the process of being recalled.
3024          * If not found, create the nfs4_file struct
3025          */
3026         fp = find_file(ino);
3027         if (fp) {
3028                 if ((status = nfs4_check_open(fp, open, &stp)))
3029                         goto out;
3030                 status = nfs4_check_deleg(cl, fp, open, &dp);
3031                 if (status)
3032                         goto out;
3033         } else {
3034                 status = nfserr_bad_stateid;
3035                 if (nfsd4_is_deleg_cur(open))
3036                         goto out;
3037                 status = nfserr_jukebox;
3038                 fp = open->op_file;
3039                 open->op_file = NULL;
3040                 nfsd4_init_file(fp, ino);
3041         }
3042
3043         /*
3044          * OPEN the file, or upgrade an existing OPEN.
3045          * If truncate fails, the OPEN fails.
3046          */
3047         if (stp) {
3048                 /* Stateid was found, this is an OPEN upgrade */
3049                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
3050                 if (status)
3051                         goto out;
3052         } else {
3053                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, open);
3054                 if (status)
3055                         goto out;
3056                 status = nfsd4_truncate(rqstp, current_fh, open);
3057                 if (status)
3058                         goto out;
3059                 stp = open->op_stp;
3060                 open->op_stp = NULL;
3061                 init_open_stateid(stp, fp, open);
3062         }
3063         update_stateid(&stp->st_stid.sc_stateid);
3064         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3065
3066         if (nfsd4_has_session(&resp->cstate)) {
3067                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3068
3069                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
3070                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3071                         open->op_why_no_deleg = WND4_NOT_WANTED;
3072                         goto nodeleg;
3073                 }
3074         }
3075
3076         /*
3077         * Attempt to hand out a delegation. No error return, because the
3078         * OPEN succeeds even if we fail.
3079         */
3080         nfs4_open_delegation(SVC_NET(rqstp), current_fh, open, stp);
3081 nodeleg:
3082         status = nfs_ok;
3083
3084         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
3085                 STATEID_VAL(&stp->st_stid.sc_stateid));
3086 out:
3087         /* 4.1 client trying to upgrade/downgrade delegation? */
3088         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
3089             open->op_deleg_want)
3090                 nfsd4_deleg_xgrade_none_ext(open, dp);
3091
3092         if (fp)
3093                 put_nfs4_file(fp);
3094         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
3095                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
3096         /*
3097         * To finish the open response, we just need to set the rflags.
3098         */
3099         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
3100         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
3101             !nfsd4_has_session(&resp->cstate))
3102                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
3103
3104         return status;
3105 }
3106
3107 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
3108 {
3109         if (open->op_openowner) {
3110                 struct nfs4_openowner *oo = open->op_openowner;
3111
3112                 if (!list_empty(&oo->oo_owner.so_stateids))
3113                         list_del_init(&oo->oo_close_lru);
3114                 if (oo->oo_flags & NFS4_OO_NEW) {
3115                         if (status) {
3116                                 release_openowner(oo);
3117                                 open->op_openowner = NULL;
3118                         } else
3119                                 oo->oo_flags &= ~NFS4_OO_NEW;
3120                 }
3121         }
3122         if (open->op_file)
3123                 nfsd4_free_file(open->op_file);
3124         if (open->op_stp)
3125                 free_generic_stateid(open->op_stp);
3126 }
3127
3128 __be32
3129 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3130             clientid_t *clid)
3131 {
3132         struct nfs4_client *clp;
3133         __be32 status;
3134         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3135
3136         nfs4_lock_state();
3137         dprintk("process_renew(%08x/%08x): starting\n",
3138                         clid->cl_boot, clid->cl_id);
3139         status = nfserr_stale_clientid;
3140         if (STALE_CLIENTID(clid, nn))
3141                 goto out;
3142         clp = find_confirmed_client(clid, cstate->minorversion, nn);
3143         status = nfserr_expired;
3144         if (clp == NULL) {
3145                 /* We assume the client took too long to RENEW. */
3146                 dprintk("nfsd4_renew: clientid not found!\n");
3147                 goto out;
3148         }
3149         status = nfserr_cb_path_down;
3150         if (!list_empty(&clp->cl_delegations)
3151                         && clp->cl_cb_state != NFSD4_CB_UP)
3152                 goto out;
3153         status = nfs_ok;
3154 out:
3155         nfs4_unlock_state();
3156         return status;
3157 }
3158
3159 static void
3160 nfsd4_end_grace(struct nfsd_net *nn)
3161 {
3162         /* do nothing if grace period already ended */
3163         if (nn->grace_ended)
3164                 return;
3165
3166         dprintk("NFSD: end of grace period\n");
3167         nn->grace_ended = true;
3168         nfsd4_record_grace_done(nn, nn->boot_time);
3169         locks_end_grace(&nn->nfsd4_manager);
3170         /*
3171          * Now that every NFSv4 client has had the chance to recover and
3172          * to see the (possibly new, possibly shorter) lease time, we
3173          * can safely set the next grace time to the current lease time:
3174          */
3175         nfsd4_grace = nfsd4_lease;
3176 }
3177
3178 static time_t
3179 nfs4_laundromat(struct nfsd_net *nn)
3180 {
3181         struct nfs4_client *clp;
3182         struct nfs4_openowner *oo;
3183         struct nfs4_delegation *dp;
3184         struct list_head *pos, *next, reaplist;
3185         time_t cutoff = get_seconds() - nfsd4_lease;
3186         time_t t, clientid_val = nfsd4_lease;
3187         time_t u, test_val = nfsd4_lease;
3188
3189         nfs4_lock_state();
3190
3191         dprintk("NFSD: laundromat service - starting\n");
3192         nfsd4_end_grace(nn);
3193         INIT_LIST_HEAD(&reaplist);
3194         spin_lock(&client_lock);
3195         list_for_each_safe(pos, next, &nn->client_lru) {
3196                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3197                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3198                         t = clp->cl_time - cutoff;
3199                         if (clientid_val > t)
3200                                 clientid_val = t;
3201                         break;
3202                 }
3203                 if (atomic_read(&clp->cl_refcount)) {
3204                         dprintk("NFSD: client in use (clientid %08x)\n",
3205                                 clp->cl_clientid.cl_id);
3206                         continue;
3207                 }
3208                 unhash_client_locked(clp);
3209                 list_add(&clp->cl_lru, &reaplist);
3210         }
3211         spin_unlock(&client_lock);
3212         list_for_each_safe(pos, next, &reaplist) {
3213                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3214                 dprintk("NFSD: purging unused client (clientid %08x)\n",
3215                         clp->cl_clientid.cl_id);
3216                 expire_client(clp);
3217         }
3218         spin_lock(&recall_lock);
3219         list_for_each_safe(pos, next, &del_recall_lru) {
3220                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3221                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3222                         u = dp->dl_time - cutoff;
3223                         if (test_val > u)
3224                                 test_val = u;
3225                         break;
3226                 }
3227                 list_move(&dp->dl_recall_lru, &reaplist);
3228         }
3229         spin_unlock(&recall_lock);
3230         list_for_each_safe(pos, next, &reaplist) {
3231                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3232                 unhash_delegation(dp);
3233         }
3234         test_val = nfsd4_lease;
3235         list_for_each_safe(pos, next, &nn->close_lru) {
3236                 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3237                 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3238                         u = oo->oo_time - cutoff;
3239                         if (test_val > u)
3240                                 test_val = u;
3241                         break;
3242                 }
3243                 release_openowner(oo);
3244         }
3245         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3246                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3247         nfs4_unlock_state();
3248         return clientid_val;
3249 }
3250
3251 static struct workqueue_struct *laundry_wq;
3252 static void laundromat_main(struct work_struct *);
3253
3254 static void
3255 laundromat_main(struct work_struct *laundry)
3256 {
3257         time_t t;
3258         struct delayed_work *dwork = container_of(laundry, struct delayed_work,
3259                                                   work);
3260         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
3261                                            laundromat_work);
3262
3263         t = nfs4_laundromat(nn);
3264         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3265         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
3266 }
3267
3268 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3269 {
3270         if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3271                 return nfserr_bad_stateid;
3272         return nfs_ok;
3273 }
3274
3275 static int
3276 STALE_STATEID(stateid_t *stateid, struct nfsd_net *nn)
3277 {
3278         if (stateid->si_opaque.so_clid.cl_boot == nn->boot_time)
3279                 return 0;
3280         dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
3281                 STATEID_VAL(stateid));
3282         return 1;
3283 }
3284
3285 static inline int
3286 access_permit_read(struct nfs4_ol_stateid *stp)
3287 {
3288         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
3289                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
3290                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
3291 }
3292
3293 static inline int
3294 access_permit_write(struct nfs4_ol_stateid *stp)
3295 {
3296         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
3297                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
3298 }
3299
3300 static
3301 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3302 {
3303         __be32 status = nfserr_openmode;
3304
3305         /* For lock stateid's, we test the parent open, not the lock: */
3306         if (stp->st_openstp)
3307                 stp = stp->st_openstp;
3308         if ((flags & WR_STATE) && !access_permit_write(stp))
3309                 goto out;
3310         if ((flags & RD_STATE) && !access_permit_read(stp))
3311                 goto out;
3312         status = nfs_ok;
3313 out:
3314         return status;
3315 }
3316
3317 static inline __be32
3318 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
3319 {
3320         if (ONE_STATEID(stateid) && (flags & RD_STATE))
3321                 return nfs_ok;
3322         else if (locks_in_grace(net)) {
3323                 /* Answer in remaining cases depends on existence of
3324                  * conflicting state; so we must wait out the grace period. */
3325                 return nfserr_grace;
3326         } else if (flags & WR_STATE)
3327                 return nfs4_share_conflict(current_fh,
3328                                 NFS4_SHARE_DENY_WRITE);
3329         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3330                 return nfs4_share_conflict(current_fh,
3331                                 NFS4_SHARE_DENY_READ);
3332 }
3333
3334 /*
3335  * Allow READ/WRITE during grace period on recovered state only for files
3336  * that are not able to provide mandatory locking.
3337  */
3338 static inline int
3339 grace_disallows_io(struct net *net, struct inode *inode)
3340 {
3341         return locks_in_grace(net) && mandatory_lock(inode);
3342 }
3343
3344 /* Returns true iff a is later than b: */
3345 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3346 {