source3/locking/brlock.c

   1 /*
   2    Unix SMB/CIFS implementation.
   3    byte range locking code
   4    Updated to handle range splits/merges.
   5
   6    Copyright (C) Andrew Tridgell 1992-2000
   7    Copyright (C) Jeremy Allison 1992-2000
   8
   9    This program is free software; you can redistribute it and/or modify
  10    it under the terms of the GNU General Public License as published by
  11    the Free Software Foundation; either version 3 of the License, or
  12    (at your option) any later version.
  13
  14    This program is distributed in the hope that it will be useful,
  15    but WITHOUT ANY WARRANTY; without even the implied warranty of
  16    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17    GNU General Public License for more details.
  18
  19    You should have received a copy of the GNU General Public License
  20    along with this program.  If not, see <http://www.gnu.org/licenses/>.
  21 */
  22
  23 /* This module implements a tdb based byte range locking service,
  24    replacing the fcntl() based byte range locking previously
  25    used. This allows us to provide the same semantics as NT */
  26
  27 #include "includes.h"
  28 #include "system/filesys.h"
  29 #include "locking/proto.h"
  30 #include "smbd/globals.h"
  31 #include "dbwrap/dbwrap.h"
  32 #include "dbwrap/dbwrap_open.h"
  33 #include "serverid.h"
  34 #include "messages.h"
  35 #include "util_tdb.h"
  36
  37 #undef DBGC_CLASS
  38 #define DBGC_CLASS DBGC_LOCKING
  39
  40 #define ZERO_ZERO 0
  41
  42 /* The open brlock.tdb database. */
  43
  44 static struct db_context *brlock_db;
  45
  46 struct byte_range_lock {
  47         struct files_struct *fsp;
  48         unsigned int num_locks;
  49         bool modified;
  50         bool have_read_oplocks;
  51         struct lock_struct *lock_data;
  52         struct db_record *record;
  53 };
  54
  55 /****************************************************************************
  56  Debug info at level 10 for lock struct.
  57 ****************************************************************************/
  58
  59 static void print_lock_struct(unsigned int i, const struct lock_struct *pls)
  60 {
  61         DEBUG(10,("[%u]: smblctx = %llu, tid = %u, pid = %s, ",
  62                         i,
  63                         (unsigned long long)pls->context.smblctx,
  64                         (unsigned int)pls->context.tid,
  65                         server_id_str(talloc_tos(), &pls->context.pid) ));
  66
  67         DEBUG(10,("start = %.0f, size = %.0f, fnum = %llu, %s %s\n",
  68                 (double)pls->start,
  69                 (double)pls->size,
  70                 (unsigned long long)pls->fnum,
  71                 lock_type_name(pls->lock_type),
  72                 lock_flav_name(pls->lock_flav) ));
  73 }
  74
  75 unsigned int brl_num_locks(const struct byte_range_lock *brl)
  76 {
  77         return brl->num_locks;
  78 }
  79
  80 struct files_struct *brl_fsp(struct byte_range_lock *brl)
  81 {
  82         return brl->fsp;
  83 }
  84
  85 bool brl_have_read_oplocks(const struct byte_range_lock *brl)
  86 {
  87         return brl->have_read_oplocks;
  88 }
  89
  90 void brl_set_have_read_oplocks(struct byte_range_lock *brl,
  91                                bool have_read_oplocks)
  92 {
  93         DEBUG(10, ("Setting have_read_oplocks to %s\n",
  94                    have_read_oplocks ? "true" : "false"));
  95         SMB_ASSERT(brl->record != NULL); /* otherwise we're readonly */
  96         brl->have_read_oplocks = have_read_oplocks;
  97         brl->modified = true;
  98 }
  99
 100 /****************************************************************************
 101  See if two locking contexts are equal.
 102 ****************************************************************************/
 103
 104 static bool brl_same_context(const struct lock_context *ctx1,
 105                              const struct lock_context *ctx2)
 106 {
 107         return (serverid_equal(&ctx1->pid, &ctx2->pid) &&
 108                 (ctx1->smblctx == ctx2->smblctx) &&
 109                 (ctx1->tid == ctx2->tid));
 110 }
 111
 112 /****************************************************************************
 113  See if lck1 and lck2 overlap.
 114 ****************************************************************************/
 115
 116 static bool brl_overlap(const struct lock_struct *lck1,
 117                         const struct lock_struct *lck2)
 118 {
 119         /* XXX Remove for Win7 compatibility. */
 120         /* this extra check is not redundant - it copes with locks
 121            that go beyond the end of 64 bit file space */
 122         if (lck1->size != 0 &&
 123             lck1->start == lck2->start &&
 124             lck1->size == lck2->size) {
 125                 return True;
 126         }
 127
 128         if (lck1->start >= (lck2->start+lck2->size) ||
 129             lck2->start >= (lck1->start+lck1->size)) {
 130                 return False;
 131         }
 132         return True;
 133 }
 134
 135 /****************************************************************************
 136  See if lock2 can be added when lock1 is in place.
 137 ****************************************************************************/
 138
 139 static bool brl_conflict(const struct lock_struct *lck1,
 140                          const struct lock_struct *lck2)
 141 {
 142         /* Ignore PENDING locks. */
 143         if (IS_PENDING_LOCK(lck1->lock_type) || IS_PENDING_LOCK(lck2->lock_type))
 144                 return False;
 145
 146         /* Read locks never conflict. */
 147         if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK) {
 148                 return False;
 149         }
 150
 151         /* A READ lock can stack on top of a WRITE lock if they have the same
 152          * context & fnum. */
 153         if (lck1->lock_type == WRITE_LOCK && lck2->lock_type == READ_LOCK &&
 154             brl_same_context(&lck1->context, &lck2->context) &&
 155             lck1->fnum == lck2->fnum) {
 156                 return False;
 157         }
 158
 159         return brl_overlap(lck1, lck2);
 160 }
 161
 162 /****************************************************************************
 163  See if lock2 can be added when lock1 is in place - when both locks are POSIX
 164  flavour. POSIX locks ignore fnum - they only care about dev/ino which we
 165  know already match.
 166 ****************************************************************************/
 167
 168 static bool brl_conflict_posix(const struct lock_struct *lck1,
 169                                 const struct lock_struct *lck2)
 170 {
 171 #if defined(DEVELOPER)
 172         SMB_ASSERT(lck1->lock_flav == POSIX_LOCK);
 173         SMB_ASSERT(lck2->lock_flav == POSIX_LOCK);
 174 #endif
 175
 176         /* Ignore PENDING locks. */
 177         if (IS_PENDING_LOCK(lck1->lock_type) || IS_PENDING_LOCK(lck2->lock_type))
 178                 return False;
 179
 180         /* Read locks never conflict. */
 181         if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK) {
 182                 return False;
 183         }
 184
 185         /* Locks on the same context don't conflict. Ignore fnum. */
 186         if (brl_same_context(&lck1->context, &lck2->context)) {
 187                 return False;
 188         }
 189
 190         /* One is read, the other write, or the context is different,
 191            do they overlap ? */
 192         return brl_overlap(lck1, lck2);
 193 }
 194
 195 #if ZERO_ZERO
 196 static bool brl_conflict1(const struct lock_struct *lck1,
 197                          const struct lock_struct *lck2)
 198 {
 199         if (IS_PENDING_LOCK(lck1->lock_type) || IS_PENDING_LOCK(lck2->lock_type))
 200                 return False;
 201
 202         if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK) {
 203                 return False;
 204         }
 205
 206         if (brl_same_context(&lck1->context, &lck2->context) &&
 207             lck2->lock_type == READ_LOCK && lck1->fnum == lck2->fnum) {
 208                 return False;
 209         }
 210
 211         if (lck2->start == 0 && lck2->size == 0 && lck1->size != 0) {
 212                 return True;
 213         }
 214
 215         if (lck1->start >= (lck2->start + lck2->size) ||
 216             lck2->start >= (lck1->start + lck1->size)) {
 217                 return False;
 218         }
 219
 220         return True;
 221 }
 222 #endif
 223
 224 /****************************************************************************
 225  Check to see if this lock conflicts, but ignore our own locks on the
 226  same fnum only. This is the read/write lock check code path.
 227  This is never used in the POSIX lock case.
 228 ****************************************************************************/
 229
 230 static bool brl_conflict_other(const struct lock_struct *lck1, const struct lock_struct *lck2)
 231 {
 232         if (IS_PENDING_LOCK(lck1->lock_type) || IS_PENDING_LOCK(lck2->lock_type))
 233                 return False;
 234
 235         if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK)
 236                 return False;
 237
 238         /* POSIX flavour locks never conflict here - this is only called
 239            in the read/write path. */
 240
 241         if (lck1->lock_flav == POSIX_LOCK && lck2->lock_flav == POSIX_LOCK)
 242                 return False;
 243
 244         /*
 245          * Incoming WRITE locks conflict with existing READ locks even
 246          * if the context is the same. JRA. See LOCKTEST7 in smbtorture.
 247          */
 248
 249         if (!(lck2->lock_type == WRITE_LOCK && lck1->lock_type == READ_LOCK)) {
 250                 if (brl_same_context(&lck1->context, &lck2->context) &&
 251                                         lck1->fnum == lck2->fnum)
 252                         return False;
 253         }
 254
 255         return brl_overlap(lck1, lck2);
 256 }
 257
 258 /****************************************************************************
 259  Check if an unlock overlaps a pending lock.
 260 ****************************************************************************/
 261
 262 static bool brl_pending_overlap(const struct lock_struct *lock, const struct lock_struct *pend_lock)
 263 {
 264         if ((lock->start <= pend_lock->start) && (lock->start + lock->size > pend_lock->start))
 265                 return True;
 266         if ((lock->start >= pend_lock->start) && (lock->start < pend_lock->start + pend_lock->size))
 267                 return True;
 268         return False;
 269 }
 270
 271 /****************************************************************************
 272  Amazingly enough, w2k3 "remembers" whether the last lock failure on a fnum
 273  is the same as this one and changes its error code. I wonder if any
 274  app depends on this ?
 275 ****************************************************************************/
 276
 277 static NTSTATUS brl_lock_failed(files_struct *fsp,
 278                                 const struct lock_struct *lock,
 279                                 bool blocking_lock)
 280 {
 281         if (lock->start >= 0xEF000000 && (lock->start >> 63) == 0) {
 282                 /* amazing the little things you learn with a test
 283                    suite. Locks beyond this offset (as a 64 bit
 284                    number!) always generate the conflict error code,
 285                    unless the top bit is set */
 286                 if (!blocking_lock) {
 287                         fsp->last_lock_failure = *lock;
 288                 }
 289                 return NT_STATUS_FILE_LOCK_CONFLICT;
 290         }
 291
 292         if (serverid_equal(&lock->context.pid, &fsp->last_lock_failure.context.pid) &&
 293                         lock->context.tid == fsp->last_lock_failure.context.tid &&
 294                         lock->fnum == fsp->last_lock_failure.fnum &&
 295                         lock->start == fsp->last_lock_failure.start) {
 296                 return NT_STATUS_FILE_LOCK_CONFLICT;
 297         }
 298
 299         if (!blocking_lock) {
 300                 fsp->last_lock_failure = *lock;
 301         }
 302         return NT_STATUS_LOCK_NOT_GRANTED;
 303 }
 304
 305 /****************************************************************************
 306  Open up the brlock.tdb database.
 307 ****************************************************************************/
 308
 309 void brl_init(bool read_only)
 310 {
 311         int tdb_flags;
 312
 313         if (brlock_db) {
 314                 return;
 315         }
 316
 317         tdb_flags = TDB_DEFAULT|TDB_VOLATILE|TDB_CLEAR_IF_FIRST|TDB_INCOMPATIBLE_HASH;
 318
 319         if (!lp_clustering()) {
 320                 /*
 321                  * We can't use the SEQNUM trick to cache brlock
 322                  * entries in the clustering case because ctdb seqnum
 323                  * propagation has a delay.
 324                  */
 325                 tdb_flags |= TDB_SEQNUM;
 326         }
 327
 328         brlock_db = db_open(NULL, lock_path("brlock.tdb"),
 329                             SMB_OPEN_DATABASE_TDB_HASH_SIZE, tdb_flags,
 330                             read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644,
 331                             DBWRAP_LOCK_ORDER_2, DBWRAP_FLAG_NONE);
 332         if (!brlock_db) {
 333                 DEBUG(0,("Failed to open byte range locking database %s\n",
 334                         lock_path("brlock.tdb")));
 335                 return;
 336         }
 337 }
 338
 339 /****************************************************************************
 340  Close down the brlock.tdb database.
 341 ****************************************************************************/
 342
 343 void brl_shutdown(void)
 344 {
 345         TALLOC_FREE(brlock_db);
 346 }
 347
 348 #if ZERO_ZERO
 349 /****************************************************************************
 350  Compare two locks for sorting.
 351 ****************************************************************************/
 352
 353 static int lock_compare(const struct lock_struct *lck1,
 354                          const struct lock_struct *lck2)
 355 {
 356         if (lck1->start != lck2->start) {
 357                 return (lck1->start - lck2->start);
 358         }
 359         if (lck2->size != lck1->size) {
 360                 return ((int)lck1->size - (int)lck2->size);
 361         }
 362         return 0;
 363 }
 364 #endif
 365
 366 /****************************************************************************
 367  Lock a range of bytes - Windows lock semantics.
 368 ****************************************************************************/
 369
 370 NTSTATUS brl_lock_windows_default(struct byte_range_lock *br_lck,
 371     struct lock_struct *plock, bool blocking_lock)
 372 {
 373         unsigned int i;
 374         files_struct *fsp = br_lck->fsp;
 375         struct lock_struct *locks = br_lck->lock_data;
 376         NTSTATUS status;
 377
 378         SMB_ASSERT(plock->lock_type != UNLOCK_LOCK);
 379
 380         if ((plock->start + plock->size - 1 < plock->start) &&
 381                         plock->size != 0) {
 382                 return NT_STATUS_INVALID_LOCK_RANGE;
 383         }
 384
 385         for (i=0; i < br_lck->num_locks; i++) {
 386                 /* Do any Windows or POSIX locks conflict ? */
 387                 if (brl_conflict(&locks[i], plock)) {
 388                         /* Remember who blocked us. */
 389                         plock->context.smblctx = locks[i].context.smblctx;
 390                         return brl_lock_failed(fsp,plock,blocking_lock);
 391                 }
 392 #if ZERO_ZERO
 393                 if (plock->start == 0 && plock->size == 0 &&
 394                                 locks[i].size == 0) {
 395                         break;
 396                 }
 397 #endif
 398         }
 399
 400         if (!IS_PENDING_LOCK(plock->lock_type)) {
 401                 contend_level2_oplocks_begin(fsp, LEVEL2_CONTEND_WINDOWS_BRL);
 402         }
 403
 404         /* We can get the Windows lock, now see if it needs to
 405            be mapped into a lower level POSIX one, and if so can
 406            we get it ? */
 407
 408         if (!IS_PENDING_LOCK(plock->lock_type) && lp_posix_locking(fsp->conn->params)) {
 409                 int errno_ret;
 410                 if (!set_posix_lock_windows_flavour(fsp,
 411                                 plock->start,
 412                                 plock->size,
 413                                 plock->lock_type,
 414                                 &plock->context,
 415                                 locks,
 416                                 br_lck->num_locks,
 417                                 &errno_ret)) {
 418
 419                         /* We don't know who blocked us. */
 420                         plock->context.smblctx = 0xFFFFFFFFFFFFFFFFLL;
 421
 422                         if (errno_ret == EACCES || errno_ret == EAGAIN) {
 423                                 status = NT_STATUS_FILE_LOCK_CONFLICT;
 424                                 goto fail;
 425                         } else {
 426                                 status = map_nt_error_from_unix(errno);
 427                                 goto fail;
 428                         }
 429                 }
 430         }
 431
 432         /* no conflicts - add it to the list of locks */
 433         locks = talloc_realloc(br_lck, locks, struct lock_struct,
 434                                (br_lck->num_locks + 1));
 435         if (!locks) {
 436                 status = NT_STATUS_NO_MEMORY;
 437                 goto fail;
 438         }
 439
 440         memcpy(&locks[br_lck->num_locks], plock, sizeof(struct lock_struct));
 441         br_lck->num_locks += 1;
 442         br_lck->lock_data = locks;
 443         br_lck->modified = True;
 444
 445         return NT_STATUS_OK;
 446  fail:
 447         if (!IS_PENDING_LOCK(plock->lock_type)) {
 448                 contend_level2_oplocks_end(fsp, LEVEL2_CONTEND_WINDOWS_BRL);
 449         }
 450         return status;
 451 }
 452
 453 /****************************************************************************
 454  Cope with POSIX range splits and merges.
 455 ****************************************************************************/
 456
 457 static unsigned int brlock_posix_split_merge(struct lock_struct *lck_arr,       /* Output array. */
 458                                                 struct lock_struct *ex,         /* existing lock. */
 459                                                 struct lock_struct *plock)      /* proposed lock. */
 460 {
 461         bool lock_types_differ = (ex->lock_type != plock->lock_type);
 462
 463         /* We can't merge non-conflicting locks on different context - ignore fnum. */
 464
 465         if (!brl_same_context(&ex->context, &plock->context)) {
 466                 /* Just copy. */
 467                 memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
 468                 return 1;
 469         }
 470
 471         /* We now know we have the same context. */
 472
 473         /* Did we overlap ? */
 474
 475 /*********************************************
 476                                         +---------+
 477                                         | ex      |
 478                                         +---------+
 479                          +-------+
 480                          | plock |
 481                          +-------+
 482 OR....
 483         +---------+
 484         |  ex     |
 485         +---------+
 486 **********************************************/
 487
 488         if ( (ex->start > (plock->start + plock->size)) ||
 489                 (plock->start > (ex->start + ex->size))) {
 490
 491                 /* No overlap with this lock - copy existing. */
 492
 493                 memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
 494                 return 1;
 495         }
 496
 497 /*********************************************
 498         +---------------------------+
 499         |          ex               |
 500         +---------------------------+
 501         +---------------------------+
 502         |       plock               | -> replace with plock.
 503         +---------------------------+
 504 OR
 505              +---------------+
 506              |       ex      |
 507              +---------------+
 508         +---------------------------+
 509         |       plock               | -> replace with plock.
 510         +---------------------------+
 511
 512 **********************************************/
 513
 514         if ( (ex->start >= plock->start) &&
 515                 (ex->start + ex->size <= plock->start + plock->size) ) {
 516
 517                 /* Replace - discard existing lock. */
 518
 519                 return 0;
 520         }
 521
 522 /*********************************************
 523 Adjacent after.
 524                         +-------+
 525                         |  ex   |
 526                         +-------+
 527         +---------------+
 528         |   plock       |
 529         +---------------+
 530
 531 BECOMES....
 532         +---------------+-------+
 533         |   plock       | ex    | - different lock types.
 534         +---------------+-------+
 535 OR.... (merge)
 536         +-----------------------+
 537         |   plock               | - same lock type.
 538         +-----------------------+
 539 **********************************************/
 540
 541         if (plock->start + plock->size == ex->start) {
 542
 543                 /* If the lock types are the same, we merge, if different, we
 544                    add the remainder of the old lock. */
 545
 546                 if (lock_types_differ) {
 547                         /* Add existing. */
 548                         memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
 549                         return 1;
 550                 } else {
 551                         /* Merge - adjust incoming lock as we may have more
 552                          * merging to come. */
 553                         plock->size += ex->size;
 554                         return 0;
 555                 }
 556         }
 557
 558 /*********************************************
 559 Adjacent before.
 560         +-------+
 561         |  ex   |
 562         +-------+
 563                 +---------------+
 564                 |   plock       |
 565                 +---------------+
 566 BECOMES....
 567         +-------+---------------+
 568         | ex    |   plock       | - different lock types
 569         +-------+---------------+
 570
 571 OR.... (merge)
 572         +-----------------------+
 573         |      plock            | - same lock type.
 574         +-----------------------+
 575
 576 **********************************************/
 577
 578         if (ex->start + ex->size == plock->start) {
 579
 580                 /* If the lock types are the same, we merge, if different, we
 581                    add the existing lock. */
 582
 583                 if (lock_types_differ) {
 584                         memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
 585                         return 1;
 586                 } else {
 587                         /* Merge - adjust incoming lock as we may have more
 588                          * merging to come. */
 589                         plock->start = ex->start;
 590                         plock->size += ex->size;
 591                         return 0;
 592                 }
 593         }
 594
 595 /*********************************************
 596 Overlap after.
 597         +-----------------------+
 598         |          ex           |
 599         +-----------------------+
 600         +---------------+
 601         |   plock       |
 602         +---------------+
 603 OR
 604                +----------------+
 605                |       ex       |
 606                +----------------+
 607         +---------------+
 608         |   plock       |
 609         +---------------+
 610
 611 BECOMES....
 612         +---------------+-------+
 613         |   plock       | ex    | - different lock types.
 614         +---------------+-------+
 615 OR.... (merge)
 616         +-----------------------+
 617         |   plock               | - same lock type.
 618         +-----------------------+
 619 **********************************************/
 620
 621         if ( (ex->start >= plock->start) &&
 622                 (ex->start <= plock->start + plock->size) &&
 623                 (ex->start + ex->size > plock->start + plock->size) ) {
 624
 625                 /* If the lock types are the same, we merge, if different, we
 626                    add the remainder of the old lock. */
 627
 628                 if (lock_types_differ) {
 629                         /* Add remaining existing. */
 630                         memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
 631                         /* Adjust existing start and size. */
 632                         lck_arr[0].start = plock->start + plock->size;
 633                         lck_arr[0].size = (ex->start + ex->size) - (plock->start + plock->size);
 634                         return 1;
 635                 } else {
 636                         /* Merge - adjust incoming lock as we may have more
 637                          * merging to come. */
 638                         plock->size += (ex->start + ex->size) - (plock->start + plock->size);
 639                         return 0;
 640                 }
 641         }
 642
 643 /*********************************************
 644 Overlap before.
 645         +-----------------------+
 646         |  ex                   |
 647         +-----------------------+
 648                 +---------------+
 649                 |   plock       |
 650                 +---------------+
 651 OR
 652         +-------------+
 653         |  ex         |
 654         +-------------+
 655                 +---------------+
 656                 |   plock       |
 657                 +---------------+
 658
 659 BECOMES....
 660         +-------+---------------+
 661         | ex    |   plock       | - different lock types
 662         +-------+---------------+
 663
 664 OR.... (merge)
 665         +-----------------------+
 666         |      plock            | - same lock type.
 667         +-----------------------+
 668
 669 **********************************************/
 670
 671         if ( (ex->start < plock->start) &&
 672                         (ex->start + ex->size >= plock->start) &&
 673                         (ex->start + ex->size <= plock->start + plock->size) ) {
 674
 675                 /* If the lock types are the same, we merge, if different, we
 676                    add the truncated old lock. */
 677
 678                 if (lock_types_differ) {
 679                         memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
 680                         /* Adjust existing size. */
 681                         lck_arr[0].size = plock->start - ex->start;
 682                         return 1;
 683                 } else {
 684                         /* Merge - adjust incoming lock as we may have more
 685                          * merging to come. MUST ADJUST plock SIZE FIRST ! */
 686                         plock->size += (plock->start - ex->start);
 687                         plock->start = ex->start;
 688                         return 0;
 689                 }
 690         }
 691
 692 /*********************************************
 693 Complete overlap.
 694         +---------------------------+
 695         |        ex                 |
 696         +---------------------------+
 697                 +---------+
 698                 |  plock  |
 699                 +---------+
 700 BECOMES.....
 701         +-------+---------+---------+
 702         | ex    |  plock  | ex      | - different lock types.
 703         +-------+---------+---------+
 704 OR
 705         +---------------------------+
 706         |        plock              | - same lock type.
 707         +---------------------------+
 708 **********************************************/
 709
 710         if ( (ex->start < plock->start) && (ex->start + ex->size > plock->start + plock->size) ) {
 711
 712                 if (lock_types_differ) {
 713
 714                         /* We have to split ex into two locks here. */
 715
 716                         memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
 717                         memcpy(&lck_arr[1], ex, sizeof(struct lock_struct));
 718
 719                         /* Adjust first existing size. */
 720                         lck_arr[0].size = plock->start - ex->start;
 721
 722                         /* Adjust second existing start and size. */
 723                         lck_arr[1].start = plock->start + plock->size;
 724                         lck_arr[1].size = (ex->start + ex->size) - (plock->start + plock->size);
 725                         return 2;
 726                 } else {
 727                         /* Just eat the existing locks, merge them into plock. */
 728                         plock->start = ex->start;
 729                         plock->size = ex->size;
 730                         return 0;
 731                 }
 732         }
 733
 734         /* Never get here. */
 735         smb_panic("brlock_posix_split_merge");
 736         /* Notreached. */
 737
 738         /* Keep some compilers happy. */
 739         return 0;
 740 }
 741
 742 /****************************************************************************
 743  Lock a range of bytes - POSIX lock semantics.
 744  We must cope with range splits and merges.
 745 ****************************************************************************/
 746
 747 static NTSTATUS brl_lock_posix(struct messaging_context *msg_ctx,
 748                                struct byte_range_lock *br_lck,
 749                                struct lock_struct *plock)
 750 {
 751         unsigned int i, count, posix_count;
 752         struct lock_struct *locks = br_lck->lock_data;
 753         struct lock_struct *tp;
 754         bool signal_pending_read = False;
 755         bool break_oplocks = false;
 756         NTSTATUS status;
 757
 758         /* No zero-zero locks for POSIX. */
 759         if (plock->start == 0 && plock->size == 0) {
 760                 return NT_STATUS_INVALID_PARAMETER;
 761         }
 762
 763         /* Don't allow 64-bit lock wrap. */
 764         if (plock->start + plock->size - 1 < plock->start) {
 765                 return NT_STATUS_INVALID_PARAMETER;
 766         }
 767
 768         /* The worst case scenario here is we have to split an
 769            existing POSIX lock range into two, and add our lock,
 770            so we need at most 2 more entries. */
 771
 772         tp = talloc_array(br_lck, struct lock_struct, br_lck->num_locks + 2);
 773         if (!tp) {
 774                 return NT_STATUS_NO_MEMORY;
 775         }
 776
 777         count = posix_count = 0;
 778
 779         for (i=0; i < br_lck->num_locks; i++) {
 780                 struct lock_struct *curr_lock = &locks[i];
 781
 782                 /* If we have a pending read lock, a lock downgrade should
 783                    trigger a lock re-evaluation. */
 784                 if (curr_lock->lock_type == PENDING_READ_LOCK &&
 785                                 brl_pending_overlap(plock, curr_lock)) {
 786                         signal_pending_read = True;
 787                 }
 788
 789                 if (curr_lock->lock_flav == WINDOWS_LOCK) {
 790                         /* Do any Windows flavour locks conflict ? */
 791                         if (brl_conflict(curr_lock, plock)) {
 792                                 /* No games with error messages. */
 793                                 TALLOC_FREE(tp);
 794                                 /* Remember who blocked us. */
 795                                 plock->context.smblctx = curr_lock->context.smblctx;
 796                                 return NT_STATUS_FILE_LOCK_CONFLICT;
 797                         }
 798                         /* Just copy the Windows lock into the new array. */
 799                         memcpy(&tp[count], curr_lock, sizeof(struct lock_struct));
 800                         count++;
 801                 } else {
 802                         unsigned int tmp_count = 0;
 803
 804                         /* POSIX conflict semantics are different. */
 805                         if (brl_conflict_posix(curr_lock, plock)) {
 806                                 /* Can't block ourselves with POSIX locks. */
 807                                 /* No games with error messages. */
 808                                 TALLOC_FREE(tp);
 809                                 /* Remember who blocked us. */
 810                                 plock->context.smblctx = curr_lock->context.smblctx;
 811                                 return NT_STATUS_FILE_LOCK_CONFLICT;
 812                         }
 813
 814                         /* Work out overlaps. */
 815                         tmp_count += brlock_posix_split_merge(&tp[count], curr_lock, plock);
 816                         posix_count += tmp_count;
 817                         count += tmp_count;
 818                 }
 819         }
 820
 821         /*
 822          * Break oplocks while we hold a brl. Since lock() and unlock() calls
 823          * are not symetric with POSIX semantics, we cannot guarantee our
 824          * contend_level2_oplocks_begin/end calls will be acquired and
 825          * released one-for-one as with Windows semantics. Therefore we only
 826          * call contend_level2_oplocks_begin if this is the first POSIX brl on
 827          * the file.
 828          */
 829         break_oplocks = (!IS_PENDING_LOCK(plock->lock_type) &&
 830                          posix_count == 0);
 831         if (break_oplocks) {
 832                 contend_level2_oplocks_begin(br_lck->fsp,
 833                                              LEVEL2_CONTEND_POSIX_BRL);
 834         }
 835
 836         /* Try and add the lock in order, sorted by lock start. */
 837         for (i=0; i < count; i++) {
 838                 struct lock_struct *curr_lock = &tp[i];
 839
 840                 if (curr_lock->start <= plock->start) {
 841                         continue;
 842                 }
 843         }
 844
 845         if (i < count) {
 846                 memmove(&tp[i+1], &tp[i],
 847                         (count - i)*sizeof(struct lock_struct));
 848         }
 849         memcpy(&tp[i], plock, sizeof(struct lock_struct));
 850         count++;
 851
 852         /* We can get the POSIX lock, now see if it needs to
 853            be mapped into a lower level POSIX one, and if so can
 854            we get it ? */
 855
 856         if (!IS_PENDING_LOCK(plock->lock_type) && lp_posix_locking(br_lck->fsp->conn->params)) {
 857                 int errno_ret;
 858
 859                 /* The lower layer just needs to attempt to
 860                    get the system POSIX lock. We've weeded out
 861                    any conflicts above. */
 862
 863                 if (!set_posix_lock_posix_flavour(br_lck->fsp,
 864                                 plock->start,
 865                                 plock->size,
 866                                 plock->lock_type,
 867                                 &errno_ret)) {
 868
 869                         /* We don't know who blocked us. */
 870                         plock->context.smblctx = 0xFFFFFFFFFFFFFFFFLL;
 871
 872                         if (errno_ret == EACCES || errno_ret == EAGAIN) {
 873                                 TALLOC_FREE(tp);
 874                                 status = NT_STATUS_FILE_LOCK_CONFLICT;
 875                                 goto fail;
 876                         } else {
 877                                 TALLOC_FREE(tp);
 878                                 status = map_nt_error_from_unix(errno);
 879                                 goto fail;
 880                         }
 881                 }
 882         }
 883
 884         /* If we didn't use all the allocated size,
 885          * Realloc so we don't leak entries per lock call. */
 886         if (count < br_lck->num_locks + 2) {
 887                 tp = talloc_realloc(br_lck, tp, struct lock_struct, count);
 888                 if (!tp) {
 889                         status = NT_STATUS_NO_MEMORY;
 890                         goto fail;
 891                 }
 892         }
 893
 894         br_lck->num_locks = count;
 895         TALLOC_FREE(br_lck->lock_data);
 896         br_lck->lock_data = tp;
 897         locks = tp;
 898         br_lck->modified = True;
 899
 900         /* A successful downgrade from write to read lock can trigger a lock
 901            re-evalutation where waiting readers can now proceed. */
 902
 903         if (signal_pending_read) {
 904                 /* Send unlock messages to any pending read waiters that overlap. */
 905                 for (i=0; i < br_lck->num_locks; i++) {
 906                         struct lock_struct *pend_lock = &locks[i];
 907
 908                         /* Ignore non-pending locks. */
 909                         if (!IS_PENDING_LOCK(pend_lock->lock_type)) {
 910                                 continue;
 911                         }
 912
 913                         if (pend_lock->lock_type == PENDING_READ_LOCK &&
 914                                         brl_pending_overlap(plock, pend_lock)) {
 915                                 DEBUG(10,("brl_lock_posix: sending unlock message to pid %s\n",
 916                                         procid_str_static(&pend_lock->context.pid )));
 917
 918                                 messaging_send(msg_ctx, pend_lock->context.pid,
 919                                                MSG_SMB_UNLOCK, &data_blob_null);
 920                         }
 921                 }
 922         }
 923
 924         return NT_STATUS_OK;
 925  fail:
 926         if (break_oplocks) {
 927                 contend_level2_oplocks_end(br_lck->fsp,
 928                                            LEVEL2_CONTEND_POSIX_BRL);
 929         }
 930         return status;
 931 }
 932
 933 NTSTATUS smb_vfs_call_brl_lock_windows(struct vfs_handle_struct *handle,
 934                                        struct byte_range_lock *br_lck,
 935                                        struct lock_struct *plock,
 936                                        bool blocking_lock)
 937 {
 938         VFS_FIND(brl_lock_windows);
 939         return handle->fns->brl_lock_windows_fn(handle, br_lck, plock,
 940                                                 blocking_lock);
 941 }
 942
 943 /****************************************************************************
 944  Lock a range of bytes.
 945 ****************************************************************************/
 946
 947 NTSTATUS brl_lock(struct messaging_context *msg_ctx,
 948                 struct byte_range_lock *br_lck,
 949                 uint64_t smblctx,
 950                 struct server_id pid,
 951                 br_off start,
 952                 br_off size,
 953                 enum brl_type lock_type,
 954                 enum brl_flavour lock_flav,
 955                 bool blocking_lock,
 956                 uint64_t *psmblctx)
 957 {
 958         NTSTATUS ret;
 959         struct lock_struct lock;
 960
 961 #if !ZERO_ZERO
 962         if (start == 0 && size == 0) {
 963                 DEBUG(0,("client sent 0/0 lock - please report this\n"));
 964         }
 965 #endif
 966
 967 #ifdef DEVELOPER
 968         /* Quieten valgrind on test. */
 969         ZERO_STRUCT(lock);
 970 #endif
 971
 972         lock.context.smblctx = smblctx;
 973         lock.context.pid = pid;
 974         lock.context.tid = br_lck->fsp->conn->cnum;
 975         lock.start = start;
 976         lock.size = size;
 977         lock.fnum = br_lck->fsp->fnum;
 978         lock.lock_type = lock_type;
 979         lock.lock_flav = lock_flav;
 980
 981         if (lock_flav == WINDOWS_LOCK) {
 982                 ret = SMB_VFS_BRL_LOCK_WINDOWS(br_lck->fsp->conn, br_lck,
 983                                                &lock, blocking_lock);
 984         } else {
 985                 ret = brl_lock_posix(msg_ctx, br_lck, &lock);
 986         }
 987
 988 #if ZERO_ZERO
 989         /* sort the lock list */
 990         TYPESAFE_QSORT(br_lck->lock_data, (size_t)br_lck->num_locks, lock_compare);
 991 #endif
 992
 993         /* If we're returning an error, return who blocked us. */
 994         if (!NT_STATUS_IS_OK(ret) && psmblctx) {
 995                 *psmblctx = lock.context.smblctx;
 996         }
 997         return ret;
 998 }
 999
1000 static void brl_delete_lock_struct(struct lock_struct *locks,
1001                                    unsigned num_locks,
1002                                    unsigned del_idx)
1003 {
1004         if (del_idx >= num_locks) {
1005                 return;
1006         }
1007         memmove(&locks[del_idx], &locks[del_idx+1],
1008                 sizeof(*locks) * (num_locks - del_idx - 1));
1009 }
1010
1011 /****************************************************************************
1012  Unlock a range of bytes - Windows semantics.
1013 ****************************************************************************/
1014
1015 bool brl_unlock_windows_default(struct messaging_context *msg_ctx,
1016                                struct byte_range_lock *br_lck,
1017                                const struct lock_struct *plock)
1018 {
1019         unsigned int i, j;
1020         struct lock_struct *locks = br_lck->lock_data;
1021         enum brl_type deleted_lock_type = READ_LOCK; /* shut the compiler up.... */
1022
1023         SMB_ASSERT(plock->lock_type == UNLOCK_LOCK);
1024
1025 #if ZERO_ZERO
1026         /* Delete write locks by preference... The lock list
1027            is sorted in the zero zero case. */
1028
1029         for (i = 0; i < br_lck->num_locks; i++) {
1030                 struct lock_struct *lock = &locks[i];
1031
1032                 if (lock->lock_type == WRITE_LOCK &&
1033                     brl_same_context(&lock->context, &plock->context) &&
1034                     lock->fnum == plock->fnum &&
1035                     lock->lock_flav == WINDOWS_LOCK &&
1036                     lock->start == plock->start &&
1037                     lock->size == plock->size) {
1038
1039                         /* found it - delete it */
1040                         deleted_lock_type = lock->lock_type;
1041                         break;
1042                 }
1043         }
1044
1045         if (i != br_lck->num_locks) {
1046                 /* We found it - don't search again. */
1047                 goto unlock_continue;
1048         }
1049 #endif
1050
1051         for (i = 0; i < br_lck->num_locks; i++) {
1052                 struct lock_struct *lock = &locks[i];
1053
1054                 if (IS_PENDING_LOCK(lock->lock_type)) {
1055                         continue;
1056                 }
1057
1058                 /* Only remove our own locks that match in start, size, and flavour. */
1059                 if (brl_same_context(&lock->context, &plock->context) &&
1060                                         lock->fnum == plock->fnum &&
1061                                         lock->lock_flav == WINDOWS_LOCK &&
1062                                         lock->start == plock->start &&
1063                                         lock->size == plock->size ) {
1064                         deleted_lock_type = lock->lock_type;
1065                         break;
1066                 }
1067         }
1068
1069         if (i == br_lck->num_locks) {
1070                 /* we didn't find it */
1071                 return False;
1072         }
1073
1074 #if ZERO_ZERO
1075   unlock_continue:
1076 #endif
1077
1078         brl_delete_lock_struct(locks, br_lck->num_locks, i);
1079         br_lck->num_locks -= 1;
1080         br_lck->modified = True;
1081
1082         /* Unlock the underlying POSIX regions. */
1083         if(lp_posix_locking(br_lck->fsp->conn->params)) {
1084                 release_posix_lock_windows_flavour(br_lck->fsp,
1085                                 plock->start,
1086                                 plock->size,
1087                                 deleted_lock_type,
1088                                 &plock->context,
1089                                 locks,
1090                                 br_lck->num_locks);
1091         }
1092
1093         /* Send unlock messages to any pending waiters that overlap. */
1094         for (j=0; j < br_lck->num_locks; j++) {
1095                 struct lock_struct *pend_lock = &locks[j];
1096
1097                 /* Ignore non-pending locks. */
1098                 if (!IS_PENDING_LOCK(pend_lock->lock_type)) {
1099                         continue;
1100                 }
1101
1102                 /* We could send specific lock info here... */
1103                 if (brl_pending_overlap(plock, pend_lock)) {
1104                         DEBUG(10,("brl_unlock: sending unlock message to pid %s\n",
1105                                 procid_str_static(&pend_lock->context.pid )));
1106
1107                         messaging_send(msg_ctx, pend_lock->context.pid,
1108                                        MSG_SMB_UNLOCK, &data_blob_null);
1109                 }
1110         }
1111
1112         contend_level2_oplocks_end(br_lck->fsp, LEVEL2_CONTEND_WINDOWS_BRL);
1113         return True;
1114 }
1115
1116 /****************************************************************************
1117  Unlock a range of bytes - POSIX semantics.
1118 ****************************************************************************/
1119
1120 static bool brl_unlock_posix(struct messaging_context *msg_ctx,
1121                              struct byte_range_lock *br_lck,
1122                              struct lock_struct *plock)
1123 {
1124         unsigned int i, j, count;
1125         struct lock_struct *tp;
1126         struct lock_struct *locks = br_lck->lock_data;
1127         bool overlap_found = False;
1128
1129         /* No zero-zero locks for POSIX. */
1130         if (plock->start == 0 && plock->size == 0) {
1131                 return False;
1132         }
1133
1134         /* Don't allow 64-bit lock wrap. */
1135         if (plock->start + plock->size < plock->start ||
1136                         plock->start + plock->size < plock->size) {
1137                 DEBUG(10,("brl_unlock_posix: lock wrap\n"));
1138                 return False;
1139         }
1140
1141         /* The worst case scenario here is we have to split an
1142            existing POSIX lock range into two, so we need at most
1143            1 more entry. */
1144
1145         tp = talloc_array(br_lck, struct lock_struct, br_lck->num_locks + 1);
1146         if (!tp) {
1147                 DEBUG(10,("brl_unlock_posix: malloc fail\n"));
1148                 return False;
1149         }
1150
1151         count = 0;
1152         for (i = 0; i < br_lck->num_locks; i++) {
1153                 struct lock_struct *lock = &locks[i];
1154                 unsigned int tmp_count;
1155
1156                 /* Only remove our own locks - ignore fnum. */
1157                 if (IS_PENDING_LOCK(lock->lock_type) ||
1158                                 !brl_same_context(&lock->context, &plock->context)) {
1159                         memcpy(&tp[count], lock, sizeof(struct lock_struct));
1160                         count++;
1161                         continue;
1162                 }
1163
1164                 if (lock->lock_flav == WINDOWS_LOCK) {
1165                         /* Do any Windows flavour locks conflict ? */
1166                         if (brl_conflict(lock, plock)) {
1167                                 TALLOC_FREE(tp);
1168                                 return false;
1169                         }
1170                         /* Just copy the Windows lock into the new array. */
1171                         memcpy(&tp[count], lock, sizeof(struct lock_struct));
1172                         count++;
1173                         continue;
1174                 }
1175
1176                 /* Work out overlaps. */
1177                 tmp_count = brlock_posix_split_merge(&tp[count], lock, plock);
1178
1179                 if (tmp_count == 0) {
1180                         /* plock overlapped the existing lock completely,
1181                            or replaced it. Don't copy the existing lock. */
1182                         overlap_found = true;
1183                 } else if (tmp_count == 1) {
1184                         /* Either no overlap, (simple copy of existing lock) or
1185                          * an overlap of an existing lock. */
1186                         /* If the lock changed size, we had an overlap. */
1187                         if (tp[count].size != lock->size) {
1188                                 overlap_found = true;
1189                         }
1190                         count += tmp_count;
1191                 } else if (tmp_count == 2) {
1192                         /* We split a lock range in two. */
1193                         overlap_found = true;
1194                         count += tmp_count;
1195
1196                         /* Optimisation... */
1197                         /* We know we're finished here as we can't overlap any
1198                            more POSIX locks. Copy the rest of the lock array. */
1199
1200                         if (i < br_lck->num_locks - 1) {
1201                                 memcpy(&tp[count], &locks[i+1],
1202                                         sizeof(*locks)*((br_lck->num_locks-1) - i));
1203                                 count += ((br_lck->num_locks-1) - i);
1204                         }
1205                         break;
1206                 }
1207
1208         }
1209
1210         if (!overlap_found) {
1211                 /* Just ignore - no change. */
1212                 TALLOC_FREE(tp);
1213                 DEBUG(10,("brl_unlock_posix: No overlap - unlocked.\n"));
1214                 return True;
1215         }
1216
1217         /* Unlock any POSIX regions. */
1218         if(lp_posix_locking(br_lck->fsp->conn->params)) {
1219                 release_posix_lock_posix_flavour(br_lck->fsp,
1220                                                 plock->start,
1221                                                 plock->size,
1222                                                 &plock->context,
1223                                                 tp,
1224                                                 count);
1225         }
1226
1227         /* Realloc so we don't leak entries per unlock call. */
1228         if (count) {
1229                 tp = talloc_realloc(br_lck, tp, struct lock_struct, count);
1230                 if (!tp) {
1231                         DEBUG(10,("brl_unlock_posix: realloc fail\n"));
1232                         return False;
1233                 }
1234         } else {
1235                 /* We deleted the last lock. */
1236                 TALLOC_FREE(tp);
1237                 tp = NULL;
1238         }
1239
1240         contend_level2_oplocks_end(br_lck->fsp,
1241                                    LEVEL2_CONTEND_POSIX_BRL);
1242
1243         br_lck->num_locks = count;
1244         TALLOC_FREE(br_lck->lock_data);
1245         locks = tp;
1246         br_lck->lock_data = tp;
1247         br_lck->modified = True;
1248
1249         /* Send unlock messages to any pending waiters that overlap. */
1250
1251         for (j=0; j < br_lck->num_locks; j++) {
1252                 struct lock_struct *pend_lock = &locks[j];
1253
1254                 /* Ignore non-pending locks. */
1255                 if (!IS_PENDING_LOCK(pend_lock->lock_type)) {
1256                         continue;
1257                 }
1258
1259                 /* We could send specific lock info here... */
1260                 if (brl_pending_overlap(plock, pend_lock)) {
1261                         DEBUG(10,("brl_unlock: sending unlock message to pid %s\n",
1262                                 procid_str_static(&pend_lock->context.pid )));
1263
1264                         messaging_send(msg_ctx, pend_lock->context.pid,
1265                                        MSG_SMB_UNLOCK, &data_blob_null);
1266                 }
1267         }
1268
1269         return True;
1270 }
1271
1272 bool smb_vfs_call_brl_unlock_windows(struct vfs_handle_struct *handle,
1273                                      struct messaging_context *msg_ctx,
1274                                      struct byte_range_lock *br_lck,
1275                                      const struct lock_struct *plock)
1276 {
1277         VFS_FIND(brl_unlock_windows);
1278         return handle->fns->brl_unlock_windows_fn(handle, msg_ctx, br_lck,
1279                                                   plock);
1280 }
1281
1282 /****************************************************************************
1283  Unlock a range of bytes.
1284 ****************************************************************************/
1285
1286 bool brl_unlock(struct messaging_context *msg_ctx,
1287                 struct byte_range_lock *br_lck,
1288                 uint64_t smblctx,
1289                 struct server_id pid,
1290                 br_off start,
1291                 br_off size,
1292                 enum brl_flavour lock_flav)
1293 {
1294         struct lock_struct lock;
1295
1296         lock.context.smblctx = smblctx;
1297         lock.context.pid = pid;
1298         lock.context.tid = br_lck->fsp->conn->cnum;
1299         lock.start = start;
1300         lock.size = size;
1301         lock.fnum = br_lck->fsp->fnum;
1302         lock.lock_type = UNLOCK_LOCK;
1303         lock.lock_flav = lock_flav;
1304
1305         if (lock_flav == WINDOWS_LOCK) {
1306                 return SMB_VFS_BRL_UNLOCK_WINDOWS(br_lck->fsp->conn, msg_ctx,
1307                     br_lck, &lock);
1308         } else {
1309                 return brl_unlock_posix(msg_ctx, br_lck, &lock);
1310         }
1311 }
1312
1313 /****************************************************************************
1314  Test if we could add a lock if we wanted to.
1315  Returns True if the region required is currently unlocked, False if locked.
1316 ****************************************************************************/
1317
1318 bool brl_locktest(struct byte_range_lock *br_lck,
1319                 uint64_t smblctx,
1320                 struct server_id pid,
1321                 br_off start,
1322                 br_off size,
1323                 enum brl_type lock_type,
1324                 enum brl_flavour lock_flav)
1325 {
1326         bool ret = True;
1327         unsigned int i;
1328         struct lock_struct lock;
1329         const struct lock_struct *locks = br_lck->lock_data;
1330         files_struct *fsp = br_lck->fsp;
1331
1332         lock.context.smblctx = smblctx;
1333         lock.context.pid = pid;
1334         lock.context.tid = br_lck->fsp->conn->cnum;
1335         lock.start = start;
1336         lock.size = size;
1337         lock.fnum = fsp->fnum;
1338         lock.lock_type = lock_type;
1339         lock.lock_flav = lock_flav;
1340
1341         /* Make sure existing locks don't conflict */
1342         for (i=0; i < br_lck->num_locks; i++) {
1343                 /*
1344                  * Our own locks don't conflict.
1345                  */
1346                 if (brl_conflict_other(&locks[i], &lock)) {
1347                         return False;
1348                 }
1349         }
1350
1351         /*
1352          * There is no lock held by an SMB daemon, check to
1353          * see if there is a POSIX lock from a UNIX or NFS process.
1354          * This only conflicts with Windows locks, not POSIX locks.
1355          */
1356
1357         if(lp_posix_locking(fsp->conn->params) && (lock_flav == WINDOWS_LOCK)) {
1358                 ret = is_posix_locked(fsp, &start, &size, &lock_type, WINDOWS_LOCK);
1359
1360                 DEBUG(10,("brl_locktest: posix start=%.0f len=%.0f %s for %s file %s\n",
1361                         (double)start, (double)size, ret ? "locked" : "unlocked",
1362                         fsp_fnum_dbg(fsp), fsp_str_dbg(fsp)));
1363
1364                 /* We need to return the inverse of is_posix_locked. */
1365                 ret = !ret;
1366         }
1367
1368         /* no conflicts - we could have added it */
1369         return ret;
1370 }
1371
1372 /****************************************************************************
1373  Query for existing locks.
1374 ****************************************************************************/
1375
1376 NTSTATUS brl_lockquery(struct byte_range_lock *br_lck,
1377                 uint64_t *psmblctx,
1378                 struct server_id pid,
1379                 br_off *pstart,
1380                 br_off *psize,
1381                 enum brl_type *plock_type,
1382                 enum brl_flavour lock_flav)
1383 {
1384         unsigned int i;
1385         struct lock_struct lock;
1386         const struct lock_struct *locks = br_lck->lock_data;
1387         files_struct *fsp = br_lck->fsp;
1388
1389         lock.context.smblctx = *psmblctx;
1390         lock.context.pid = pid;
1391         lock.context.tid = br_lck->fsp->conn->cnum;
1392         lock.start = *pstart;
1393         lock.size = *psize;
1394         lock.fnum = fsp->fnum;
1395         lock.lock_type = *plock_type;
1396         lock.lock_flav = lock_flav;
1397
1398         /* Make sure existing locks don't conflict */
1399         for (i=0; i < br_lck->num_locks; i++) {
1400                 const struct lock_struct *exlock = &locks[i];
1401                 bool conflict = False;
1402
1403                 if (exlock->lock_flav == WINDOWS_LOCK) {
1404                         conflict = brl_conflict(exlock, &lock);
1405                 } else {
1406                         conflict = brl_conflict_posix(exlock, &lock);
1407                 }
1408
1409                 if (conflict) {
1410                         *psmblctx = exlock->context.smblctx;
1411                         *pstart = exlock->start;
1412                         *psize = exlock->size;
1413                         *plock_type = exlock->lock_type;
1414                         return NT_STATUS_LOCK_NOT_GRANTED;
1415                 }
1416         }
1417
1418         /*
1419          * There is no lock held by an SMB daemon, check to
1420          * see if there is a POSIX lock from a UNIX or NFS process.
1421          */
1422
1423         if(lp_posix_locking(fsp->conn->params)) {
1424                 bool ret = is_posix_locked(fsp, pstart, psize, plock_type, POSIX_LOCK);
1425
1426                 DEBUG(10,("brl_lockquery: posix start=%.0f len=%.0f %s for %s file %s\n",
1427                         (double)*pstart, (double)*psize, ret ? "locked" : "unlocked",
1428                         fsp_fnum_dbg(fsp), fsp_str_dbg(fsp)));
1429
1430                 if (ret) {
1431                         /* Hmmm. No clue what to set smblctx to - use -1. */
1432                         *psmblctx = 0xFFFFFFFFFFFFFFFFLL;
1433                         return NT_STATUS_LOCK_NOT_GRANTED;
1434                 }
1435         }
1436
1437         return NT_STATUS_OK;
1438 }
1439
1440
1441 bool smb_vfs_call_brl_cancel_windows(struct vfs_handle_struct *handle,
1442                                      struct byte_range_lock *br_lck,
1443                                      struct lock_struct *plock)
1444 {
1445         VFS_FIND(brl_cancel_windows);
1446         return handle->fns->brl_cancel_windows_fn(handle, br_lck, plock);
1447 }
1448
1449 /****************************************************************************
1450  Remove a particular pending lock.
1451 ****************************************************************************/
1452 bool brl_lock_cancel(struct byte_range_lock *br_lck,
1453                 uint64_t smblctx,
1454                 struct server_id pid,
1455                 br_off start,
1456                 br_off size,
1457                 enum brl_flavour lock_flav)
1458 {
1459         bool ret;
1460         struct lock_struct lock;
1461
1462         lock.context.smblctx = smblctx;
1463         lock.context.pid = pid;
1464         lock.context.tid = br_lck->fsp->conn->cnum;
1465         lock.start = start;
1466         lock.size = size;
1467         lock.fnum = br_lck->fsp->fnum;
1468         lock.lock_flav = lock_flav;
1469         /* lock.lock_type doesn't matter */
1470
1471         if (lock_flav == WINDOWS_LOCK) {
1472                 ret = SMB_VFS_BRL_CANCEL_WINDOWS(br_lck->fsp->conn, br_lck,
1473                                                  &lock);
1474         } else {
1475                 ret = brl_lock_cancel_default(br_lck, &lock);
1476         }
1477
1478         return ret;
1479 }
1480
1481 bool brl_lock_cancel_default(struct byte_range_lock *br_lck,
1482                 struct lock_struct *plock)
1483 {
1484         unsigned int i;
1485         struct lock_struct *locks = br_lck->lock_data;
1486
1487         SMB_ASSERT(plock);
1488
1489         for (i = 0; i < br_lck->num_locks; i++) {
1490                 struct lock_struct *lock = &locks[i];
1491
1492                 /* For pending locks we *always* care about the fnum. */
1493                 if (brl_same_context(&lock->context, &plock->context) &&
1494                                 lock->fnum == plock->fnum &&
1495                                 IS_PENDING_LOCK(lock->lock_type) &&
1496                                 lock->lock_flav == plock->lock_flav &&
1497                                 lock->start == plock->start &&
1498                                 lock->size == plock->size) {
1499                         break;
1500                 }
1501         }
1502
1503         if (i == br_lck->num_locks) {
1504                 /* Didn't find it. */
1505                 return False;
1506         }
1507
1508         brl_delete_lock_struct(locks, br_lck->num_locks, i);
1509         br_lck->num_locks -= 1;
1510         br_lck->modified = True;
1511         return True;
1512 }
1513
1514 /****************************************************************************
1515  Remove any locks associated with a open file.
1516  We return True if this process owns any other Windows locks on this
1517  fd and so we should not immediately close the fd.
1518 ****************************************************************************/
1519
1520 void brl_close_fnum(struct messaging_context *msg_ctx,
1521                     struct byte_range_lock *br_lck)
1522 {
1523         files_struct *fsp = br_lck->fsp;
1524         uint32_t tid = fsp->conn->cnum;
1525         uint64_t fnum = fsp->fnum;
1526         unsigned int i;
1527         struct lock_struct *locks = br_lck->lock_data;
1528         struct server_id pid = messaging_server_id(fsp->conn->sconn->msg_ctx);
1529         struct lock_struct *locks_copy;
1530         unsigned int num_locks_copy;
1531
1532         /* Copy the current lock array. */
1533         if (br_lck->num_locks) {
1534                 locks_copy = (struct lock_struct *)talloc_memdup(br_lck, locks, br_lck->num_locks * sizeof(struct lock_struct));
1535                 if (!locks_copy) {
1536                         smb_panic("brl_close_fnum: talloc failed");
1537                         }
1538         } else {
1539                 locks_copy = NULL;
1540         }
1541
1542         num_locks_copy = br_lck->num_locks;
1543
1544         for (i=0; i < num_locks_copy; i++) {
1545                 struct lock_struct *lock = &locks_copy[i];
1546
1547                 if (lock->context.tid == tid && serverid_equal(&lock->context.pid, &pid) &&
1548                                 (lock->fnum == fnum)) {
1549                         brl_unlock(msg_ctx,
1550                                 br_lck,
1551                                 lock->context.smblctx,
1552                                 pid,
1553                                 lock->start,
1554                                 lock->size,
1555                                 lock->lock_flav);
1556                 }
1557         }
1558 }
1559
1560 bool brl_mark_disconnected(struct files_struct *fsp)
1561 {
1562         uint32_t tid = fsp->conn->cnum;
1563         uint64_t smblctx;
1564         uint64_t fnum = fsp->fnum;
1565         unsigned int i;
1566         struct server_id self = messaging_server_id(fsp->conn->sconn->msg_ctx);
1567         struct byte_range_lock *br_lck = NULL;
1568
1569         if (fsp->op == NULL) {
1570                 return false;
1571         }
1572
1573         smblctx = fsp->op->global->open_persistent_id;
1574
1575         if (!fsp->op->global->durable) {
1576                 return false;
1577         }
1578
1579         if (fsp->current_lock_count == 0) {
1580                 return true;
1581         }
1582
1583         br_lck = brl_get_locks(talloc_tos(), fsp);
1584         if (br_lck == NULL) {
1585                 return false;
1586         }
1587
1588         for (i=0; i < br_lck->num_locks; i++) {
1589                 struct lock_struct *lock = &br_lck->lock_data[i];
1590
1591                 /*
1592                  * as this is a durable handle, we only expect locks
1593                  * of the current file handle!
1594                  */
1595
1596                 if (lock->context.smblctx != smblctx) {
1597                         TALLOC_FREE(br_lck);
1598                         return false;
1599                 }
1600
1601                 if (lock->context.tid != tid) {
1602                         TALLOC_FREE(br_lck);
1603                         return false;
1604                 }
1605
1606                 if (!serverid_equal(&lock->context.pid, &self)) {
1607                         TALLOC_FREE(br_lck);
1608                         return false;
1609                 }
1610
1611                 if (lock->fnum != fnum) {
1612                         TALLOC_FREE(br_lck);
1613                         return false;
1614                 }
1615
1616                 server_id_set_disconnected(&lock->context.pid);
1617                 lock->context.tid = TID_FIELD_INVALID;
1618                 lock->fnum = FNUM_FIELD_INVALID;
1619         }
1620
1621         br_lck->modified = true;
1622         TALLOC_FREE(br_lck);
1623         return true;
1624 }
1625
1626 bool brl_reconnect_disconnected(struct files_struct *fsp)
1627 {
1628         uint32_t tid = fsp->conn->cnum;
1629         uint64_t smblctx;
1630         uint64_t fnum = fsp->fnum;
1631         unsigned int i;
1632         struct server_id self = messaging_server_id(fsp->conn->sconn->msg_ctx);
1633         struct byte_range_lock *br_lck = NULL;
1634
1635         if (fsp->op == NULL) {
1636                 return false;
1637         }
1638
1639         smblctx = fsp->op->global->open_persistent_id;
1640
1641         if (!fsp->op->global->durable) {
1642                 return false;
1643         }
1644
1645         /*
1646          * When reconnecting, we do not want to validate the brlock entries
1647          * and thereby remove our own (disconnected) entries but reactivate
1648          * them instead.
1649          */
1650         fsp->lockdb_clean = true;
1651
1652         br_lck = brl_get_locks(talloc_tos(), fsp);
1653         if (br_lck == NULL) {
1654                 return false;
1655         }
1656
1657         if (br_lck->num_locks == 0) {
1658                 TALLOC_FREE(br_lck);
1659                 return true;
1660         }
1661
1662         for (i=0; i < br_lck->num_locks; i++) {
1663                 struct lock_struct *lock = &br_lck->lock_data[i];
1664
1665                 /*
1666                  * as this is a durable handle we only expect locks
1667                  * of the current file handle!
1668                  */
1669
1670                 if (lock->context.smblctx != smblctx) {
1671                         TALLOC_FREE(br_lck);
1672                         return false;
1673                 }
1674
1675                 if (lock->context.tid != TID_FIELD_INVALID) {
1676                         TALLOC_FREE(br_lck);
1677                         return false;
1678                 }
1679
1680                 if (!server_id_is_disconnected(&lock->context.pid)) {
1681                         TALLOC_FREE(br_lck);
1682                         return false;
1683                 }
1684
1685                 if (lock->fnum != FNUM_FIELD_INVALID) {
1686                         TALLOC_FREE(br_lck);
1687                         return false;
1688                 }
1689
1690                 lock->context.pid = self;
1691                 lock->context.tid = tid;
1692                 lock->fnum = fnum;
1693         }
1694
1695         fsp->current_lock_count = br_lck->num_locks;
1696         br_lck->modified = true;
1697         TALLOC_FREE(br_lck);
1698         return true;
1699 }
1700
1701 /****************************************************************************
1702  Ensure this set of lock entries is valid.
1703 ****************************************************************************/
1704 static bool validate_lock_entries(unsigned int *pnum_entries, struct lock_struct **pplocks,
1705                                   bool keep_disconnected)
1706 {
1707         unsigned int i;
1708         struct lock_struct *locks = *pplocks;
1709         unsigned int num_entries = *pnum_entries;
1710         TALLOC_CTX *frame;
1711         struct server_id *ids;
1712         bool *exists;
1713
1714         if (num_entries == 0) {
1715                 return true;
1716         }
1717
1718         frame = talloc_stackframe();
1719
1720         ids = talloc_array(frame, struct server_id, num_entries);
1721         if (ids == NULL) {
1722                 DEBUG(0, ("validate_lock_entries: "
1723                           "talloc_array(struct server_id, %u) failed\n",
1724                           num_entries));
1725                 talloc_free(frame);
1726                 return false;
1727         }
1728
1729         exists = talloc_array(frame, bool, num_entries);
1730         if (exists == NULL) {
1731                 DEBUG(0, ("validate_lock_entries: "
1732                           "talloc_array(bool, %u) failed\n",
1733                           num_entries));
1734                 talloc_free(frame);
1735                 return false;
1736         }
1737
1738         for (i = 0; i < num_entries; i++) {
1739                 ids[i] = locks[i].context.pid;
1740         }
1741
1742         if (!serverids_exist(ids, num_entries, exists)) {
1743                 DEBUG(3, ("validate_lock_entries: serverids_exists failed\n"));
1744                 talloc_free(frame);
1745                 return false;
1746         }
1747
1748         i = 0;
1749
1750         while (i < num_entries) {
1751                 if (exists[i]) {
1752                         i++;
1753                         continue;
1754                 }
1755
1756                 if (keep_disconnected &&
1757                     server_id_is_disconnected(&ids[i]))
1758                 {
1759                         i++;
1760                         continue;
1761                 }
1762
1763                 /* This process no longer exists */
1764
1765                 brl_delete_lock_struct(locks, num_entries, i);
1766                 num_entries -= 1;
1767         }
1768         TALLOC_FREE(frame);
1769
1770         *pnum_entries = num_entries;
1771
1772         return True;
1773 }
1774
1775 struct brl_forall_cb {
1776         void (*fn)(struct file_id id, struct server_id pid,
1777                    enum brl_type lock_type,
1778                    enum brl_flavour lock_flav,
1779                    br_off start, br_off size,
1780                    void *private_data);
1781         void *private_data;
1782 };
1783
1784 /****************************************************************************
1785  Traverse the whole database with this function, calling traverse_callback
1786  on each lock.
1787 ****************************************************************************/
1788
1789 static int brl_traverse_fn(struct db_record *rec, void *state)
1790 {
1791         struct brl_forall_cb *cb = (struct brl_forall_cb *)state;
1792         struct lock_struct *locks;
1793         struct file_id *key;
1794         unsigned int i;
1795         unsigned int num_locks = 0;
1796         unsigned int orig_num_locks = 0;
1797         TDB_DATA dbkey;
1798         TDB_DATA value;
1799
1800         dbkey = dbwrap_record_get_key(rec);
1801         value = dbwrap_record_get_value(rec);
1802
1803         /* In a traverse function we must make a copy of
1804            dbuf before modifying it. */
1805
1806         locks = (struct lock_struct *)talloc_memdup(
1807                 talloc_tos(), value.dptr, value.dsize);
1808         if (!locks) {
1809                 return -1; /* Terminate traversal. */
1810         }
1811
1812         key = (struct file_id *)dbkey.dptr;
1813         orig_num_locks = num_locks = value.dsize/sizeof(*locks);
1814
1815         /* Ensure the lock db is clean of entries from invalid processes. */
1816
1817         if (!validate_lock_entries(&num_locks, &locks, true)) {
1818                 TALLOC_FREE(locks);
1819                 return -1; /* Terminate traversal */
1820         }
1821
1822         if (orig_num_locks != num_locks) {
1823                 if (num_locks) {
1824                         TDB_DATA data;
1825                         data.dptr = (uint8_t *)locks;
1826                         data.dsize = num_locks*sizeof(struct lock_struct);
1827                         dbwrap_record_store(rec, data, TDB_REPLACE);
1828                 } else {
1829                         dbwrap_record_delete(rec);
1830                 }
1831         }
1832
1833         if (cb->fn) {
1834                 for ( i=0; i<num_locks; i++) {
1835                         cb->fn(*key,
1836                                 locks[i].context.pid,
1837                                 locks[i].lock_type,
1838                                 locks[i].lock_flav,
1839                                 locks[i].start,
1840                                 locks[i].size,
1841                                 cb->private_data);
1842                 }
1843         }
1844
1845         TALLOC_FREE(locks);
1846         return 0;
1847 }
1848
1849 /*******************************************************************
1850  Call the specified function on each lock in the database.
1851 ********************************************************************/
1852
1853 int brl_forall(void (*fn)(struct file_id id, struct server_id pid,
1854                           enum brl_type lock_type,
1855                           enum brl_flavour lock_flav,
1856                           br_off start, br_off size,
1857                           void *private_data),
1858                void *private_data)
1859 {
1860         struct brl_forall_cb cb;
1861         NTSTATUS status;
1862         int count = 0;
1863
1864         if (!brlock_db) {
1865                 return 0;
1866         }
1867         cb.fn = fn;
1868         cb.private_data = private_data;
1869         status = dbwrap_traverse(brlock_db, brl_traverse_fn, &cb, &count);
1870
1871         if (!NT_STATUS_IS_OK(status)) {
1872                 return -1;
1873         } else {
1874                 return count;
1875         }
1876 }
1877
1878 /*******************************************************************
1879  Store a potentially modified set of byte range lock data back into
1880  the database.
1881  Unlock the record.
1882 ********************************************************************/
1883
1884 static void byte_range_lock_flush(struct byte_range_lock *br_lck)
1885 {
1886         size_t data_len;
1887         if (!br_lck->modified) {
1888                 DEBUG(10, ("br_lck not modified\n"));
1889                 goto done;
1890         }
1891
1892         data_len = br_lck->num_locks * sizeof(struct lock_struct);
1893
1894         if (br_lck->have_read_oplocks) {
1895                 data_len += 1;
1896         }
1897
1898         DEBUG(10, ("data_len=%d\n", (int)data_len));
1899
1900         if (data_len == 0) {
1901                 /* No locks - delete this entry. */
1902                 NTSTATUS status = dbwrap_record_delete(br_lck->record);
1903                 if (!NT_STATUS_IS_OK(status)) {
1904                         DEBUG(0, ("delete_rec returned %s\n",
1905                                   nt_errstr(status)));
1906                         smb_panic("Could not delete byte range lock entry");
1907                 }
1908         } else {
1909                 TDB_DATA data;
1910                 NTSTATUS status;
1911
1912                 data.dsize = data_len;
1913                 data.dptr = talloc_array(talloc_tos(), uint8_t, data_len);
1914                 SMB_ASSERT(data.dptr != NULL);
1915
1916                 memcpy(data.dptr, br_lck->lock_data,
1917                        br_lck->num_locks * sizeof(struct lock_struct));
1918
1919                 if (br_lck->have_read_oplocks) {
1920                         data.dptr[data_len-1] = 1;
1921                 }
1922
1923                 status = dbwrap_record_store(br_lck->record, data, TDB_REPLACE);
1924                 TALLOC_FREE(data.dptr);
1925                 if (!NT_STATUS_IS_OK(status)) {
1926                         DEBUG(0, ("store returned %s\n", nt_errstr(status)));
1927                         smb_panic("Could not store byte range mode entry");
1928                 }
1929         }
1930
1931         DEBUG(10, ("seqnum=%d\n", dbwrap_get_seqnum(brlock_db)));
1932
1933  done:
1934         br_lck->modified = false;
1935         TALLOC_FREE(br_lck->record);
1936 }
1937
1938 static int byte_range_lock_destructor(struct byte_range_lock *br_lck)
1939 {
1940         byte_range_lock_flush(br_lck);
1941         return 0;
1942 }
1943
1944 /*******************************************************************
1945  Fetch a set of byte range lock data from the database.
1946  Leave the record locked.
1947  TALLOC_FREE(brl) will release the lock in the destructor.
1948 ********************************************************************/
1949
1950 struct byte_range_lock *brl_get_locks(TALLOC_CTX *mem_ctx, files_struct *fsp)
1951 {
1952         TDB_DATA key, data;
1953         struct byte_range_lock *br_lck = talloc(mem_ctx, struct byte_range_lock);
1954
1955         if (br_lck == NULL) {
1956                 return NULL;
1957         }
1958
1959         br_lck->fsp = fsp;
1960         br_lck->num_locks = 0;
1961         br_lck->have_read_oplocks = false;
1962         br_lck->modified = False;
1963
1964         key.dptr = (uint8 *)&fsp->file_id;
1965         key.dsize = sizeof(struct file_id);
1966
1967         br_lck->record = dbwrap_fetch_locked(brlock_db, br_lck, key);
1968
1969         if (br_lck->record == NULL) {
1970                 DEBUG(3, ("Could not lock byte range lock entry\n"));
1971                 TALLOC_FREE(br_lck);
1972                 return NULL;
1973         }
1974
1975         data = dbwrap_record_get_value(br_lck->record);
1976
1977         br_lck->lock_data = NULL;
1978
1979         talloc_set_destructor(br_lck, byte_range_lock_destructor);
1980
1981         br_lck->num_locks = data.dsize / sizeof(struct lock_struct);
1982
1983         if (br_lck->num_locks != 0) {
1984                 br_lck->lock_data = talloc_array(
1985                         br_lck, struct lock_struct, br_lck->num_locks);
1986                 if (br_lck->lock_data == NULL) {
1987                         DEBUG(0, ("malloc failed\n"));
1988                         TALLOC_FREE(br_lck);
1989                         return NULL;
1990                 }
1991
1992                 memcpy(br_lck->lock_data, data.dptr,
1993                        talloc_get_size(br_lck->lock_data));
1994         }
1995
1996         DEBUG(10, ("data.dsize=%d\n", (int)data.dsize));
1997
1998         if ((data.dsize % sizeof(struct lock_struct)) == 1) {
1999                 br_lck->have_read_oplocks = (data.dptr[data.dsize-1] == 1);
2000         }
2001
2002         if (!fsp->lockdb_clean) {
2003                 int orig_num_locks = br_lck->num_locks;
2004
2005                 /*
2006                  * This is the first time we access the byte range lock
2007                  * record with this fsp. Go through and ensure all entries
2008                  * are valid - remove any that don't.
2009                  * This makes the lockdb self cleaning at low cost.
2010                  *
2011                  * Note: Disconnected entries belong to disconnected
2012                  * durable handles. So at this point, we have a new
2013                  * handle on the file and the disconnected durable has
2014                  * already been closed (we are not a durable reconnect).
2015                  * So we need to clean the disconnected brl entry.
2016                  */
2017
2018                 if (!validate_lock_entries(&br_lck->num_locks,
2019                                            &br_lck->lock_data, false)) {
2020                         TALLOC_FREE(br_lck);
2021                         return NULL;
2022                 }
2023
2024                 /* Ensure invalid locks are cleaned up in the destructor. */
2025                 if (orig_num_locks != br_lck->num_locks) {
2026                         br_lck->modified = True;
2027                 }
2028
2029                 /* Mark the lockdb as "clean" as seen from this open file. */
2030                 fsp->lockdb_clean = True;
2031         }
2032
2033         if (DEBUGLEVEL >= 10) {
2034                 unsigned int i;
2035                 struct lock_struct *locks = br_lck->lock_data;
2036                 DEBUG(10,("brl_get_locks_internal: %u current locks on file_id %s\n",
2037                         br_lck->num_locks,
2038                           file_id_string_tos(&fsp->file_id)));
2039                 for( i = 0; i < br_lck->num_locks; i++) {
2040                         print_lock_struct(i, &locks[i]);
2041                 }
2042         }
2043
2044         return br_lck;
2045 }
2046
2047 struct brl_get_locks_readonly_state {
2048         TALLOC_CTX *mem_ctx;
2049         struct byte_range_lock **br_lock;
2050 };
2051
2052 static void brl_get_locks_readonly_parser(TDB_DATA key, TDB_DATA data,
2053                                           void *private_data)
2054 {
2055         struct brl_get_locks_readonly_state *state =
2056                 (struct brl_get_locks_readonly_state *)private_data;
2057         struct byte_range_lock *br_lock;
2058
2059         br_lock = talloc_pooled_object(
2060                 state->mem_ctx, struct byte_range_lock, 1, data.dsize);
2061         if (br_lock == NULL) {
2062                 *state->br_lock = NULL;
2063                 return;
2064         }
2065         br_lock->lock_data = (struct lock_struct *)talloc_memdup(
2066                 br_lock, data.dptr, data.dsize);
2067         br_lock->num_locks = data.dsize / sizeof(struct lock_struct);
2068
2069         if ((data.dsize % sizeof(struct lock_struct)) == 1) {
2070                 br_lock->have_read_oplocks = (data.dptr[data.dsize-1] == 1);
2071         } else {
2072                 br_lock->have_read_oplocks = false;
2073         }
2074
2075         DEBUG(10, ("Got %d bytes, have_read_oplocks: %s\n", (int)data.dsize,
2076                    br_lock->have_read_oplocks ? "true" : "false"));
2077
2078         *state->br_lock = br_lock;
2079 }
2080
2081 struct byte_range_lock *brl_get_locks_readonly(files_struct *fsp)
2082 {
2083         struct byte_range_lock *br_lock = NULL;
2084         struct byte_range_lock *rw = NULL;
2085
2086         DEBUG(10, ("seqnum=%d, fsp->brlock_seqnum=%d\n",
2087                    dbwrap_get_seqnum(brlock_db), fsp->brlock_seqnum));
2088
2089         if ((fsp->brlock_rec != NULL)
2090             && (dbwrap_get_seqnum(brlock_db) == fsp->brlock_seqnum)) {
2091                 /*
2092                  * We have cached the brlock_rec and the database did not
2093                  * change.
2094                  */
2095                 return fsp->brlock_rec;
2096         }
2097
2098         if (!fsp->lockdb_clean) {
2099                 /*
2100                  * Fetch the record in R/W mode to give validate_lock_entries
2101                  * a chance to kick in once.
2102                  */
2103                 rw = brl_get_locks(talloc_tos(), fsp);
2104                 if (rw == NULL) {
2105                         return NULL;
2106                 }
2107                 fsp->lockdb_clean = true;
2108         }
2109
2110         if (rw != NULL) {
2111                 size_t lock_data_size;
2112
2113                 /*
2114                  * Make a copy of the already retrieved and sanitized rw record
2115                  */
2116                 lock_data_size = rw->num_locks * sizeof(struct lock_struct);
2117                 br_lock = talloc_pooled_object(
2118                         fsp, struct byte_range_lock, 1, lock_data_size);
2119                 if (br_lock == NULL) {
2120                         goto fail;
2121                 }
2122                 br_lock->have_read_oplocks = rw->have_read_oplocks;
2123                 br_lock->num_locks = rw->num_locks;
2124                 br_lock->lock_data = (struct lock_struct *)talloc_memdup(
2125                         br_lock, rw->lock_data, lock_data_size);
2126         } else {
2127                 struct brl_get_locks_readonly_state state;
2128                 NTSTATUS status;
2129
2130                 /*
2131                  * Parse the record fresh from the database
2132                  */
2133
2134                 state.mem_ctx = fsp;
2135                 state.br_lock = &br_lock;
2136
2137                 status = dbwrap_parse_record(
2138                         brlock_db,
2139                         make_tdb_data((uint8_t *)&fsp->file_id,
2140                                       sizeof(fsp->file_id)),
2141                         brl_get_locks_readonly_parser, &state);
2142
2143                 if (NT_STATUS_EQUAL(status,NT_STATUS_NOT_FOUND)) {
2144                         /*
2145                          * No locks on this file. Return an empty br_lock.
2146                          */
2147                         br_lock = talloc(fsp, struct byte_range_lock);
2148                         if (br_lock == NULL) {
2149                                 goto fail;
2150                         }
2151
2152                         br_lock->have_read_oplocks = false;
2153                         br_lock->num_locks = 0;
2154                         br_lock->lock_data = NULL;
2155
2156                 } else if (!NT_STATUS_IS_OK(status)) {
2157                         DEBUG(3, ("Could not parse byte range lock record: "
2158                                   "%s\n", nt_errstr(status)));
2159                         goto fail;
2160                 }
2161                 if (br_lock == NULL) {
2162                         goto fail;
2163                 }
2164         }
2165
2166         br_lock->fsp = fsp;
2167         br_lock->modified = false;
2168         br_lock->record = NULL;
2169
2170         if (lp_clustering()) {
2171                 /*
2172                  * In the cluster case we can't cache the brlock struct
2173                  * because dbwrap_get_seqnum does not work reliably over
2174                  * ctdb. Thus we have to throw away the brlock struct soon.
2175                  */
2176                 talloc_steal(talloc_tos(), br_lock);
2177         } else {
2178                 /*
2179                  * Cache the brlock struct, invalidated when the dbwrap_seqnum
2180                  * changes. See beginning of this routine.
2181                  */
2182                 TALLOC_FREE(fsp->brlock_rec);
2183                 fsp->brlock_rec = br_lock;
2184                 fsp->brlock_seqnum = dbwrap_get_seqnum(brlock_db);
2185         }
2186
2187 fail:
2188         TALLOC_FREE(rw);
2189         return br_lock;
2190 }
2191
2192 struct brl_revalidate_state {
2193         ssize_t array_size;
2194         uint32 num_pids;
2195         struct server_id *pids;
2196 };
2197
2198 /*
2199  * Collect PIDs of all processes with pending entries
2200  */
2201
2202 static void brl_revalidate_collect(struct file_id id, struct server_id pid,
2203                                    enum brl_type lock_type,
2204                                    enum brl_flavour lock_flav,
2205                                    br_off start, br_off size,
2206                                    void *private_data)
2207 {
2208         struct brl_revalidate_state *state =
2209                 (struct brl_revalidate_state *)private_data;
2210
2211         if (!IS_PENDING_LOCK(lock_type)) {
2212                 return;
2213         }
2214
2215         add_to_large_array(state, sizeof(pid), (void *)&pid,
2216                            &state->pids, &state->num_pids,
2217                            &state->array_size);
2218 }
2219
2220 /*
2221  * qsort callback to sort the processes
2222  */
2223
2224 static int compare_procids(const void *p1, const void *p2)
2225 {
2226         const struct server_id *i1 = (const struct server_id *)p1;
2227         const struct server_id *i2 = (const struct server_id *)p2;
2228
2229         if (i1->pid < i2->pid) return -1;
2230         if (i1->pid > i2->pid) return 1;
2231         return 0;
2232 }
2233
2234 /*
2235  * Send a MSG_SMB_UNLOCK message to all processes with pending byte range
2236  * locks so that they retry. Mainly used in the cluster code after a node has
2237  * died.
2238  *
2239  * Done in two steps to avoid double-sends: First we collect all entries in an
2240  * array, then qsort that array and only send to non-dupes.
2241  */
2242
2243 void brl_revalidate(struct messaging_context *msg_ctx,
2244                     void *private_data,
2245                     uint32_t msg_type,
2246                     struct server_id server_id,
2247                     DATA_BLOB *data)
2248 {
2249         struct brl_revalidate_state *state;
2250         uint32 i;
2251         struct server_id last_pid;
2252
2253         if (!(state = talloc_zero(NULL, struct brl_revalidate_state))) {
2254                 DEBUG(0, ("talloc failed\n"));
2255                 return;
2256         }
2257
2258         brl_forall(brl_revalidate_collect, state);
2259
2260         if (state->array_size == -1) {
2261                 DEBUG(0, ("talloc failed\n"));
2262                 goto done;
2263         }
2264
2265         if (state->num_pids == 0) {
2266                 goto done;
2267         }
2268
2269         TYPESAFE_QSORT(state->pids, state->num_pids, compare_procids);
2270
2271         ZERO_STRUCT(last_pid);
2272
2273         for (i=0; i<state->num_pids; i++) {
2274                 if (serverid_equal(&last_pid, &state->pids[i])) {
2275                         /*
2276                          * We've seen that one already
2277                          */
2278                         continue;
2279                 }
2280
2281                 messaging_send(msg_ctx, state->pids[i], MSG_SMB_UNLOCK,
2282                                &data_blob_null);
2283                 last_pid = state->pids[i];
2284         }
2285
2286  done:
2287         TALLOC_FREE(state);
2288         return;
2289 }
2290
2291 bool brl_cleanup_disconnected(struct file_id fid, uint64_t open_persistent_id)
2292 {
2293         bool ret = false;
2294         TALLOC_CTX *frame = talloc_stackframe();
2295         TDB_DATA key, val;
2296         struct db_record *rec;
2297         struct lock_struct *lock;
2298         unsigned n, num;
2299         NTSTATUS status;
2300
2301         key = make_tdb_data((void*)&fid, sizeof(fid));
2302
2303         rec = dbwrap_fetch_locked(brlock_db, frame, key);
2304         if (rec == NULL) {
2305                 DEBUG(5, ("brl_cleanup_disconnected: failed to fetch record "
2306                           "for file %s\n", file_id_string(frame, &fid)));
2307                 goto done;
2308         }
2309
2310         val = dbwrap_record_get_value(rec);
2311         lock = (struct lock_struct*)val.dptr;
2312         num = val.dsize / sizeof(struct lock_struct);
2313         if (lock == NULL) {
2314                 DEBUG(10, ("brl_cleanup_disconnected: no byte range locks for "
2315                            "file %s\n", file_id_string(frame, &fid)));
2316                 ret = true;
2317                 goto done;
2318         }
2319
2320         for (n=0; n<num; n++) {
2321                 struct lock_context *ctx = &lock[n].context;
2322
2323                 if (!server_id_is_disconnected(&ctx->pid)) {
2324                         DEBUG(5, ("brl_cleanup_disconnected: byte range lock "
2325                                   "%s used by server %s, do not cleanup\n",
2326                                   file_id_string(frame, &fid),
2327                                   server_id_str(frame, &ctx->pid)));
2328                         goto done;
2329                 }
2330
2331                 if (ctx->smblctx != open_persistent_id) {
2332                         DEBUG(5, ("brl_cleanup_disconnected: byte range lock "
2333                                   "%s expected smblctx %llu but found %llu"
2334                                   ", do not cleanup\n",
2335                                   file_id_string(frame, &fid),
2336                                   (unsigned long long)open_persistent_id,
2337                                   (unsigned long long)ctx->smblctx));
2338                         goto done;
2339                 }
2340         }
2341
2342         status = dbwrap_record_delete(rec);
2343         if (!NT_STATUS_IS_OK(status)) {
2344                 DEBUG(5, ("brl_cleanup_disconnected: failed to delete record "
2345                           "for file %s from %s, open %llu: %s\n",
2346                           file_id_string(frame, &fid), dbwrap_name(brlock_db),
2347                           (unsigned long long)open_persistent_id,
2348                           nt_errstr(status)));
2349                 goto done;
2350         }
2351
2352         DEBUG(10, ("brl_cleanup_disconnected: "
2353                    "file %s cleaned up %u entries from open %llu\n",
2354                    file_id_string(frame, &fid), num,
2355                    (unsigned long long)open_persistent_id));
2356
2357         ret = true;
2358 done:
2359         talloc_free(frame);
2360         return ret;
2361 }