source3/smbd/oplock.c

   1 /*
   2    Unix SMB/CIFS implementation.
   3    oplock processing
   4    Copyright (C) Andrew Tridgell 1992-1998
   5    Copyright (C) Jeremy Allison 1998 - 2001
   6    Copyright (C) Volker Lendecke 2005
   7
   8    This program is free software; you can redistribute it and/or modify
   9    it under the terms of the GNU General Public License as published by
  10    the Free Software Foundation; either version 3 of the License, or
  11    (at your option) any later version.
  12
  13    This program is distributed in the hope that it will be useful,
  14    but WITHOUT ANY WARRANTY; without even the implied warranty of
  15    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16    GNU General Public License for more details.
  17
  18    You should have received a copy of the GNU General Public License
  19    along with this program.  If not, see <http://www.gnu.org/licenses/>.
  20 */
  21
  22 #define DBGC_CLASS DBGC_LOCKING
  23 #include "includes.h"
  24 #include "smbd/globals.h"
  25 #include "librpc/gen_ndr/messaging.h"
  26
  27 /****************************************************************************
  28  Get the number of current exclusive oplocks.
  29 ****************************************************************************/
  30
  31 int32 get_number_of_exclusive_open_oplocks(void)
  32 {
  33   return exclusive_oplocks_open;
  34 }
  35
  36 /*
  37  * helper function used by the kernel oplock backends to post the break message
  38  */
  39 void break_kernel_oplock(struct messaging_context *msg_ctx, files_struct *fsp)
  40 {
  41         uint8_t msg[MSG_SMB_KERNEL_BREAK_SIZE];
  42
  43         /* Put the kernel break info into the message. */
  44         push_file_id_24((char *)msg, &fsp->file_id);
  45         SIVAL(msg,24,fsp->fh->gen_id);
  46
  47         /* Don't need to be root here as we're only ever
  48            sending to ourselves. */
  49
  50         messaging_send_buf(msg_ctx, messaging_server_id(msg_ctx),
  51                            MSG_SMB_KERNEL_BREAK,
  52                            msg, MSG_SMB_KERNEL_BREAK_SIZE);
  53 }
  54
  55 /****************************************************************************
  56  Attempt to set an oplock on a file. Succeeds if kernel oplocks are
  57  disabled (just sets flags) and no byte-range locks in the file. Returns True
  58  if oplock set.
  59 ****************************************************************************/
  60
  61 bool set_file_oplock(files_struct *fsp, int oplock_type)
  62 {
  63         if (fsp->oplock_type == LEVEL_II_OPLOCK) {
  64                 if (koplocks &&
  65                     !(koplocks->flags & KOPLOCKS_LEVEL2_SUPPORTED)) {
  66                         DEBUG(10, ("Refusing level2 oplock, kernel oplocks "
  67                                    "don't support them\n"));
  68                         return false;
  69                 }
  70         }
  71
  72         if ((fsp->oplock_type != NO_OPLOCK) &&
  73             (fsp->oplock_type != FAKE_LEVEL_II_OPLOCK) &&
  74             koplocks &&
  75             !koplocks->ops->set_oplock(koplocks, fsp, oplock_type)) {
  76                 return False;
  77         }
  78
  79         fsp->oplock_type = oplock_type;
  80         fsp->sent_oplock_break = NO_BREAK_SENT;
  81         if (oplock_type == LEVEL_II_OPLOCK) {
  82                 level_II_oplocks_open++;
  83         } else if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
  84                 exclusive_oplocks_open++;
  85         }
  86
  87         DEBUG(5,("set_file_oplock: granted oplock on file %s, %s/%lu, "
  88                     "tv_sec = %x, tv_usec = %x\n",
  89                  fsp_str_dbg(fsp), file_id_string_tos(&fsp->file_id),
  90                  fsp->fh->gen_id, (int)fsp->open_time.tv_sec,
  91                  (int)fsp->open_time.tv_usec ));
  92
  93         return True;
  94 }
  95
  96 /****************************************************************************
  97  Attempt to release an oplock on a file. Decrements oplock count.
  98 ****************************************************************************/
  99
 100 void release_file_oplock(files_struct *fsp)
 101 {
 102         if ((fsp->oplock_type != NO_OPLOCK) &&
 103             (fsp->oplock_type != FAKE_LEVEL_II_OPLOCK) &&
 104             koplocks) {
 105                 koplocks->ops->release_oplock(koplocks, fsp, NO_OPLOCK);
 106         }
 107
 108         if (fsp->oplock_type == LEVEL_II_OPLOCK) {
 109                 level_II_oplocks_open--;
 110         } else if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
 111                 exclusive_oplocks_open--;
 112         }
 113
 114         SMB_ASSERT(exclusive_oplocks_open>=0);
 115         SMB_ASSERT(level_II_oplocks_open>=0);
 116
 117         if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
 118                 /* This doesn't matter for close. */
 119                 fsp->oplock_type = FAKE_LEVEL_II_OPLOCK;
 120         } else {
 121                 fsp->oplock_type = NO_OPLOCK;
 122         }
 123         fsp->sent_oplock_break = NO_BREAK_SENT;
 124
 125         flush_write_cache(fsp, OPLOCK_RELEASE_FLUSH);
 126         delete_write_cache(fsp);
 127
 128         TALLOC_FREE(fsp->oplock_timeout);
 129 }
 130
 131 /****************************************************************************
 132  Attempt to downgrade an oplock on a file. Doesn't decrement oplock count.
 133 ****************************************************************************/
 134
 135 static void downgrade_file_oplock(files_struct *fsp)
 136 {
 137         if (!EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
 138                 DEBUG(0, ("trying to downgrade an already-downgraded oplock!\n"));
 139                 return;
 140         }
 141
 142         if (koplocks) {
 143                 koplocks->ops->release_oplock(koplocks, fsp, LEVEL_II_OPLOCK);
 144         }
 145         fsp->oplock_type = LEVEL_II_OPLOCK;
 146         exclusive_oplocks_open--;
 147         level_II_oplocks_open++;
 148         fsp->sent_oplock_break = NO_BREAK_SENT;
 149 }
 150
 151 /****************************************************************************
 152  Remove a file oplock. Copes with level II and exclusive.
 153  Locks then unlocks the share mode lock. Client can decide to go directly
 154  to none even if a "break-to-level II" was sent.
 155 ****************************************************************************/
 156
 157 bool remove_oplock(files_struct *fsp)
 158 {
 159         bool ret;
 160         struct share_mode_lock *lck;
 161
 162         /* Remove the oplock flag from the sharemode. */
 163         lck = get_share_mode_lock(talloc_tos(), fsp->file_id, NULL, NULL,
 164                                   NULL);
 165         if (lck == NULL) {
 166                 DEBUG(0,("remove_oplock: failed to lock share entry for "
 167                          "file %s\n", fsp_str_dbg(fsp)));
 168                 return False;
 169         }
 170         ret = remove_share_oplock(lck, fsp);
 171         if (!ret) {
 172                 DEBUG(0,("remove_oplock: failed to remove share oplock for "
 173                          "file %s fnum %d, %s\n",
 174                          fsp_str_dbg(fsp), fsp->fnum,
 175                          file_id_string_tos(&fsp->file_id)));
 176         }
 177         release_file_oplock(fsp);
 178         TALLOC_FREE(lck);
 179         return ret;
 180 }
 181
 182 /*
 183  * Deal with a reply when a break-to-level II was sent.
 184  */
 185 bool downgrade_oplock(files_struct *fsp)
 186 {
 187         bool ret;
 188         struct share_mode_lock *lck;
 189
 190         lck = get_share_mode_lock(talloc_tos(), fsp->file_id, NULL, NULL,
 191                                   NULL);
 192         if (lck == NULL) {
 193                 DEBUG(0,("downgrade_oplock: failed to lock share entry for "
 194                          "file %s\n", fsp_str_dbg(fsp)));
 195                 return False;
 196         }
 197         ret = downgrade_share_oplock(lck, fsp);
 198         if (!ret) {
 199                 DEBUG(0,("downgrade_oplock: failed to downgrade share oplock "
 200                          "for file %s fnum %d, file_id %s\n",
 201                          fsp_str_dbg(fsp), fsp->fnum,
 202                          file_id_string_tos(&fsp->file_id)));
 203         }
 204
 205         downgrade_file_oplock(fsp);
 206         TALLOC_FREE(lck);
 207         return ret;
 208 }
 209
 210 /*
 211  * Some kernel oplock implementations handle the notification themselves.
 212  */
 213 bool should_notify_deferred_opens()
 214 {
 215         return !(koplocks &&
 216                 (koplocks->flags & KOPLOCKS_DEFERRED_OPEN_NOTIFICATION));
 217 }
 218
 219 /****************************************************************************
 220  Set up an oplock break message.
 221 ****************************************************************************/
 222
 223 static char *new_break_message_smb1(TALLOC_CTX *mem_ctx,
 224                                    files_struct *fsp, int cmd)
 225 {
 226         char *result = TALLOC_ARRAY(mem_ctx, char, smb_size + 8*2 + 0);
 227
 228         if (result == NULL) {
 229                 DEBUG(0, ("talloc failed\n"));
 230                 return NULL;
 231         }
 232
 233         memset(result,'\0',smb_size);
 234         srv_set_message(result,8,0,true);
 235         SCVAL(result,smb_com,SMBlockingX);
 236         SSVAL(result,smb_tid,fsp->conn->cnum);
 237         SSVAL(result,smb_pid,0xFFFF);
 238         SSVAL(result,smb_uid,0);
 239         SSVAL(result,smb_mid,0xFFFF);
 240         SCVAL(result,smb_vwv0,0xFF);
 241         SSVAL(result,smb_vwv2,fsp->fnum);
 242         SCVAL(result,smb_vwv3,LOCKING_ANDX_OPLOCK_RELEASE);
 243         SCVAL(result,smb_vwv3+1,cmd);
 244         return result;
 245 }
 246
 247 /****************************************************************************
 248  Function to do the waiting before sending a local break.
 249 ****************************************************************************/
 250
 251 static void wait_before_sending_break(void)
 252 {
 253         long wait_time = (long)lp_oplock_break_wait_time();
 254
 255         if (wait_time) {
 256                 smb_msleep(wait_time);
 257         }
 258 }
 259
 260 /****************************************************************************
 261  Ensure that we have a valid oplock.
 262 ****************************************************************************/
 263
 264 static files_struct *initial_break_processing(
 265         struct smbd_server_connection *sconn, struct file_id id,
 266         unsigned long file_id)
 267 {
 268         files_struct *fsp = NULL;
 269
 270         if( DEBUGLVL( 3 ) ) {
 271                 dbgtext( "initial_break_processing: called for %s/%u\n",
 272                          file_id_string_tos(&id), (int)file_id);
 273                 dbgtext( "Current oplocks_open (exclusive = %d, levelII = %d)\n",
 274                         exclusive_oplocks_open, level_II_oplocks_open );
 275         }
 276
 277         /*
 278          * We need to search the file open table for the
 279          * entry containing this dev and inode, and ensure
 280          * we have an oplock on it.
 281          */
 282
 283         fsp = file_find_dif(sconn, id, file_id);
 284
 285         if(fsp == NULL) {
 286                 /* The file could have been closed in the meantime - return success. */
 287                 if( DEBUGLVL( 3 ) ) {
 288                         dbgtext( "initial_break_processing: cannot find open file with " );
 289                         dbgtext( "file_id %s gen_id = %lu", file_id_string_tos(&id), file_id);
 290                         dbgtext( "allowing break to succeed.\n" );
 291                 }
 292                 return NULL;
 293         }
 294
 295         /* Ensure we have an oplock on the file */
 296
 297         /*
 298          * There is a potential race condition in that an oplock could
 299          * have been broken due to another udp request, and yet there are
 300          * still oplock break messages being sent in the udp message
 301          * queue for this file. So return true if we don't have an oplock,
 302          * as we may have just freed it.
 303          */
 304
 305         if(fsp->oplock_type == NO_OPLOCK) {
 306                 if( DEBUGLVL( 3 ) ) {
 307                         dbgtext( "initial_break_processing: file %s ",
 308                                  fsp_str_dbg(fsp));
 309                         dbgtext( "(file_id = %s gen_id = %lu) has no oplock.\n",
 310                                  file_id_string_tos(&id), fsp->fh->gen_id );
 311                         dbgtext( "Allowing break to succeed regardless.\n" );
 312                 }
 313                 return NULL;
 314         }
 315
 316         return fsp;
 317 }
 318
 319 static void oplock_timeout_handler(struct event_context *ctx,
 320                                    struct timed_event *te,
 321                                    struct timeval now,
 322                                    void *private_data)
 323 {
 324         files_struct *fsp = (files_struct *)private_data;
 325
 326         /* Remove the timed event handler. */
 327         TALLOC_FREE(fsp->oplock_timeout);
 328         DEBUG(0, ("Oplock break failed for file %s -- replying anyway\n",
 329                   fsp_str_dbg(fsp)));
 330         remove_oplock(fsp);
 331         reply_to_oplock_break_requests(fsp);
 332 }
 333
 334 /*******************************************************************
 335  Add a timeout handler waiting for the client reply.
 336 *******************************************************************/
 337
 338 static void add_oplock_timeout_handler(files_struct *fsp)
 339 {
 340         /*
 341          * If kernel oplocks already notifies smbds when an oplock break times
 342          * out, just return.
 343          */
 344         if (koplocks &&
 345             (koplocks->flags & KOPLOCKS_TIMEOUT_NOTIFICATION)) {
 346                 return;
 347         }
 348
 349         if (fsp->oplock_timeout != NULL) {
 350                 DEBUG(0, ("Logic problem -- have an oplock event hanging "
 351                           "around\n"));
 352         }
 353
 354         fsp->oplock_timeout =
 355                 event_add_timed(smbd_event_context(), fsp,
 356                                 timeval_current_ofs(OPLOCK_BREAK_TIMEOUT, 0),
 357                                 oplock_timeout_handler, fsp);
 358
 359         if (fsp->oplock_timeout == NULL) {
 360                 DEBUG(0, ("Could not add oplock timeout handler\n"));
 361         }
 362 }
 363
 364 static void send_break_message_smb1(files_struct *fsp, int level)
 365 {
 366         char *break_msg = new_break_message_smb1(talloc_tos(),
 367                                         fsp,
 368                                         level);
 369         if (break_msg == NULL) {
 370                 exit_server("Could not talloc break_msg\n");
 371         }
 372
 373         show_msg(break_msg);
 374         if (!srv_send_smb(fsp->conn->sconn,
 375                         break_msg, false, 0,
 376                         IS_CONN_ENCRYPTED(fsp->conn),
 377                         NULL)) {
 378                 exit_server_cleanly("send_break_message_smb1: "
 379                         "srv_send_smb failed.");
 380         }
 381
 382         TALLOC_FREE(break_msg);
 383 }
 384
 385 void break_level2_to_none_async(files_struct *fsp)
 386 {
 387         struct smbd_server_connection *sconn = fsp->conn->sconn;
 388
 389         if (fsp->oplock_type == NO_OPLOCK) {
 390                 /* We already got a "break to none" message and we've handled
 391                  * it.  just ignore. */
 392                 DEBUG(3, ("process_oplock_async_level2_break_message: already "
 393                           "broken to none, ignoring.\n"));
 394                 return;
 395         }
 396
 397         if (fsp->oplock_type == FAKE_LEVEL_II_OPLOCK) {
 398                 /* Don't tell the client, just downgrade. */
 399                 DEBUG(3, ("process_oplock_async_level2_break_message: "
 400                           "downgrading fake level 2 oplock.\n"));
 401                 remove_oplock(fsp);
 402                 return;
 403         }
 404
 405         /* Ensure we're really at level2 state. */
 406         SMB_ASSERT(fsp->oplock_type == LEVEL_II_OPLOCK);
 407
 408         DEBUG(10,("process_oplock_async_level2_break_message: sending break "
 409                   "to none message for fid %d, file %s\n", fsp->fnum,
 410                   fsp_str_dbg(fsp)));
 411
 412         /* Now send a break to none message to our client. */
 413         if (sconn->using_smb2) {
 414                 send_break_message_smb2(fsp, OPLOCKLEVEL_NONE);
 415         } else {
 416                 send_break_message_smb1(fsp, OPLOCKLEVEL_NONE);
 417         }
 418
 419         /* Async level2 request, don't send a reply, just remove the oplock. */
 420         remove_oplock(fsp);
 421 }
 422
 423 /*******************************************************************
 424  This handles the case of a write triggering a break to none
 425  message on a level2 oplock.
 426  When we get this message we may be in any of three states :
 427  NO_OPLOCK, LEVEL_II, FAKE_LEVEL2. We only send a message to
 428  the client for LEVEL2.
 429 *******************************************************************/
 430
 431 void process_oplock_async_level2_break_message(struct messaging_context *msg_ctx,
 432                                                       void *private_data,
 433                                                       uint32_t msg_type,
 434                                                       struct server_id src,
 435                                                       DATA_BLOB *data)
 436 {
 437         struct smbd_server_connection *sconn;
 438         struct share_mode_entry msg;
 439         files_struct *fsp;
 440
 441         if (data->data == NULL) {
 442                 DEBUG(0, ("Got NULL buffer\n"));
 443                 return;
 444         }
 445
 446         sconn = msg_ctx_to_sconn(msg_ctx);
 447         if (sconn == NULL) {
 448                 DEBUG(1, ("could not find sconn\n"));
 449                 return;
 450         }
 451
 452         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
 453                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
 454                 return;
 455         }
 456
 457         /* De-linearize incoming message. */
 458         message_to_share_mode_entry(&msg, (char *)data->data);
 459
 460         DEBUG(10, ("Got oplock async level 2 break message from pid %s: "
 461                    "%s/%lu\n", procid_str(talloc_tos(), &src),
 462                    file_id_string_tos(&msg.id), msg.share_file_id));
 463
 464         fsp = initial_break_processing(sconn, msg.id, msg.share_file_id);
 465
 466         if (fsp == NULL) {
 467                 /* We hit a race here. Break messages are sent, and before we
 468                  * get to process this message, we have closed the file.
 469                  * No need to reply as this is an async message. */
 470                 DEBUG(3, ("process_oplock_async_level2_break_message: Did not find fsp, ignoring\n"));
 471                 return;
 472         }
 473
 474         break_level2_to_none_async(fsp);
 475 }
 476
 477 /*******************************************************************
 478  This handles the generic oplock break message from another smbd.
 479 *******************************************************************/
 480
 481 static void process_oplock_break_message(struct messaging_context *msg_ctx,
 482                                          void *private_data,
 483                                          uint32_t msg_type,
 484                                          struct server_id src,
 485                                          DATA_BLOB *data)
 486 {
 487         struct smbd_server_connection *sconn;
 488         struct share_mode_entry msg;
 489         files_struct *fsp;
 490         bool break_to_level2 = False;
 491
 492         if (data->data == NULL) {
 493                 DEBUG(0, ("Got NULL buffer\n"));
 494                 return;
 495         }
 496
 497         sconn = msg_ctx_to_sconn(msg_ctx);
 498         if (sconn == NULL) {
 499                 DEBUG(1, ("could not find sconn\n"));
 500                 return;
 501         }
 502
 503         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
 504                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
 505                 return;
 506         }
 507
 508         /* De-linearize incoming message. */
 509         message_to_share_mode_entry(&msg, (char *)data->data);
 510
 511         DEBUG(10, ("Got oplock break message from pid %s: %s/%lu\n",
 512                    procid_str(talloc_tos(), &src), file_id_string_tos(&msg.id),
 513                    msg.share_file_id));
 514
 515         fsp = initial_break_processing(sconn, msg.id, msg.share_file_id);
 516
 517         if (fsp == NULL) {
 518                 /* We hit a race here. Break messages are sent, and before we
 519                  * get to process this message, we have closed the file. Reply
 520                  * with 'ok, oplock broken' */
 521                 DEBUG(3, ("Did not find fsp\n"));
 522
 523                 /* We just send the same message back. */
 524                 messaging_send_buf(msg_ctx, src, MSG_SMB_BREAK_RESPONSE,
 525                                    (uint8 *)data->data,
 526                                    MSG_SMB_SHARE_MODE_ENTRY_SIZE);
 527                 return;
 528         }
 529
 530         if (fsp->sent_oplock_break != NO_BREAK_SENT) {
 531                 /* Remember we have to inform the requesting PID when the
 532                  * client replies */
 533                 msg.pid = src;
 534                 ADD_TO_ARRAY(NULL, struct share_mode_entry, msg,
 535                              &fsp->pending_break_messages,
 536                              &fsp->num_pending_break_messages);
 537                 return;
 538         }
 539
 540         if (EXCLUSIVE_OPLOCK_TYPE(msg.op_type) &&
 541             !EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
 542                 DEBUG(3, ("Already downgraded oplock on %s: %s\n",
 543                           file_id_string_tos(&fsp->file_id),
 544                           fsp_str_dbg(fsp)));
 545                 /* We just send the same message back. */
 546                 messaging_send_buf(msg_ctx, src, MSG_SMB_BREAK_RESPONSE,
 547                                    (uint8 *)data->data,
 548                                    MSG_SMB_SHARE_MODE_ENTRY_SIZE);
 549                 return;
 550         }
 551
 552         if ((global_client_caps & CAP_LEVEL_II_OPLOCKS) &&
 553             !(msg.op_type & FORCE_OPLOCK_BREAK_TO_NONE) &&
 554             !(koplocks && !(koplocks->flags & KOPLOCKS_LEVEL2_SUPPORTED)) &&
 555             lp_level2_oplocks(SNUM(fsp->conn))) {
 556                 break_to_level2 = True;
 557         }
 558
 559         /* Need to wait before sending a break
 560            message if we sent ourselves this message. */
 561         if (procid_is_me(&src)) {
 562                 wait_before_sending_break();
 563         }
 564
 565         if (sconn->using_smb2) {
 566                 send_break_message_smb2(fsp, break_to_level2 ?
 567                         OPLOCKLEVEL_II : OPLOCKLEVEL_NONE);
 568         } else {
 569                 send_break_message_smb1(fsp, break_to_level2 ?
 570                         OPLOCKLEVEL_II : OPLOCKLEVEL_NONE);
 571         }
 572
 573         fsp->sent_oplock_break = break_to_level2 ? LEVEL_II_BREAK_SENT:BREAK_TO_NONE_SENT;
 574
 575         msg.pid = src;
 576         ADD_TO_ARRAY(NULL, struct share_mode_entry, msg,
 577                      &fsp->pending_break_messages,
 578                      &fsp->num_pending_break_messages);
 579
 580         add_oplock_timeout_handler(fsp);
 581 }
 582
 583 /*******************************************************************
 584  This handles the kernel oplock break message.
 585 *******************************************************************/
 586
 587 static void process_kernel_oplock_break(struct messaging_context *msg_ctx,
 588                                         void *private_data,
 589                                         uint32_t msg_type,
 590                                         struct server_id src,
 591                                         DATA_BLOB *data)
 592 {
 593         struct smbd_server_connection *sconn;
 594         struct file_id id;
 595         unsigned long file_id;
 596         files_struct *fsp;
 597
 598         if (data->data == NULL) {
 599                 DEBUG(0, ("Got NULL buffer\n"));
 600                 return;
 601         }
 602
 603         if (data->length != MSG_SMB_KERNEL_BREAK_SIZE) {
 604                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
 605                 return;
 606         }
 607
 608         sconn = msg_ctx_to_sconn(msg_ctx);
 609         if (sconn == NULL) {
 610                 DEBUG(1, ("could not find sconn\n"));
 611                 return;
 612         }
 613
 614         /* Pull the data from the message. */
 615         pull_file_id_24((char *)data->data, &id);
 616         file_id = (unsigned long)IVAL(data->data, 24);
 617
 618         DEBUG(10, ("Got kernel oplock break message from pid %s: %s/%u\n",
 619                    procid_str(talloc_tos(), &src), file_id_string_tos(&id),
 620                    (unsigned int)file_id));
 621
 622         fsp = initial_break_processing(sconn, id, file_id);
 623
 624         if (fsp == NULL) {
 625                 DEBUG(3, ("Got a kernel oplock break message for a file "
 626                           "I don't know about\n"));
 627                 return;
 628         }
 629
 630         if (fsp->sent_oplock_break != NO_BREAK_SENT) {
 631                 /* This is ok, kernel oplocks come in completely async */
 632                 DEBUG(3, ("Got a kernel oplock request while waiting for a "
 633                           "break reply\n"));
 634                 return;
 635         }
 636
 637         if (sconn->using_smb2) {
 638                 send_break_message_smb2(fsp, OPLOCKLEVEL_NONE);
 639         } else {
 640                 send_break_message_smb1(fsp, OPLOCKLEVEL_NONE);
 641         }
 642
 643         fsp->sent_oplock_break = BREAK_TO_NONE_SENT;
 644
 645         add_oplock_timeout_handler(fsp);
 646 }
 647
 648 void reply_to_oplock_break_requests(files_struct *fsp)
 649 {
 650         int i;
 651
 652         /*
 653          * If kernel oplocks already notifies smbds when oplocks are
 654          * broken/removed, just return.
 655          */
 656         if (koplocks &&
 657             (koplocks->flags & KOPLOCKS_OPLOCK_BROKEN_NOTIFICATION)) {
 658                 return;
 659         }
 660
 661         for (i=0; i<fsp->num_pending_break_messages; i++) {
 662                 struct share_mode_entry *e = &fsp->pending_break_messages[i];
 663                 char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
 664
 665                 share_mode_entry_to_message(msg, e);
 666
 667                 messaging_send_buf(fsp->conn->sconn->msg_ctx, e->pid,
 668                                    MSG_SMB_BREAK_RESPONSE,
 669                                    (uint8 *)msg,
 670                                    MSG_SMB_SHARE_MODE_ENTRY_SIZE);
 671         }
 672
 673         SAFE_FREE(fsp->pending_break_messages);
 674         fsp->num_pending_break_messages = 0;
 675         if (fsp->oplock_timeout != NULL) {
 676                 /* Remove the timed event handler. */
 677                 TALLOC_FREE(fsp->oplock_timeout);
 678                 fsp->oplock_timeout = NULL;
 679         }
 680         return;
 681 }
 682
 683 static void process_oplock_break_response(struct messaging_context *msg_ctx,
 684                                           void *private_data,
 685                                           uint32_t msg_type,
 686                                           struct server_id src,
 687                                           DATA_BLOB *data)
 688 {
 689         struct share_mode_entry msg;
 690
 691         if (data->data == NULL) {
 692                 DEBUG(0, ("Got NULL buffer\n"));
 693                 return;
 694         }
 695
 696         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
 697                 DEBUG(0, ("Got invalid msg len %u\n",
 698                           (unsigned int)data->length));
 699                 return;
 700         }
 701
 702         /* De-linearize incoming message. */
 703         message_to_share_mode_entry(&msg, (char *)data->data);
 704
 705         DEBUG(10, ("Got oplock break response from pid %s: %s/%lu mid %llu\n",
 706                    procid_str(talloc_tos(), &src), file_id_string_tos(&msg.id),
 707                    msg.share_file_id, (unsigned long long)msg.op_mid));
 708
 709         schedule_deferred_open_message_smb(msg.op_mid);
 710 }
 711
 712 static void process_open_retry_message(struct messaging_context *msg_ctx,
 713                                        void *private_data,
 714                                        uint32_t msg_type,
 715                                        struct server_id src,
 716                                        DATA_BLOB *data)
 717 {
 718         struct share_mode_entry msg;
 719
 720         if (data->data == NULL) {
 721                 DEBUG(0, ("Got NULL buffer\n"));
 722                 return;
 723         }
 724
 725         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
 726                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
 727                 return;
 728         }
 729
 730         /* De-linearize incoming message. */
 731         message_to_share_mode_entry(&msg, (char *)data->data);
 732
 733         DEBUG(10, ("Got open retry msg from pid %s: %s mid %llu\n",
 734                    procid_str(talloc_tos(), &src), file_id_string_tos(&msg.id),
 735                    (unsigned long long)msg.op_mid));
 736
 737         schedule_deferred_open_message_smb(msg.op_mid);
 738 }
 739
 740 /****************************************************************************
 741  This function is called on any file modification or lock request. If a file
 742  is level 2 oplocked then it must tell all other level 2 holders to break to
 743  none.
 744 ****************************************************************************/
 745
 746 static void contend_level2_oplocks_begin_default(files_struct *fsp,
 747                                               enum level2_contention_type type)
 748 {
 749         int i;
 750         struct share_mode_lock *lck;
 751
 752         /*
 753          * If this file is level II oplocked then we need
 754          * to grab the shared memory lock and inform all
 755          * other files with a level II lock that they need
 756          * to flush their read caches. We keep the lock over
 757          * the shared memory area whilst doing this.
 758          */
 759
 760         if (!LEVEL_II_OPLOCK_TYPE(fsp->oplock_type))
 761                 return;
 762
 763         lck = get_share_mode_lock(talloc_tos(), fsp->file_id, NULL, NULL,
 764                                   NULL);
 765         if (lck == NULL) {
 766                 DEBUG(0,("release_level_2_oplocks_on_change: failed to lock "
 767                          "share mode entry for file %s.\n", fsp_str_dbg(fsp)));
 768                 return;
 769         }
 770
 771         DEBUG(10,("release_level_2_oplocks_on_change: num_share_modes = %d\n",
 772                   lck->num_share_modes ));
 773
 774         for(i = 0; i < lck->num_share_modes; i++) {
 775                 struct share_mode_entry *share_entry = &lck->share_modes[i];
 776                 char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
 777
 778                 if (!is_valid_share_mode_entry(share_entry)) {
 779                         continue;
 780                 }
 781
 782                 /*
 783                  * As there could have been multiple writes waiting at the
 784                  * lock_share_entry gate we may not be the first to
 785                  * enter. Hence the state of the op_types in the share mode
 786                  * entries may be partly NO_OPLOCK and partly LEVEL_II or FAKE_LEVEL_II
 787                  * oplock. It will do no harm to re-send break messages to
 788                  * those smbd's that are still waiting their turn to remove
 789                  * their LEVEL_II state, and also no harm to ignore existing
 790                  * NO_OPLOCK states. JRA.
 791                  */
 792
 793                 DEBUG(10,("release_level_2_oplocks_on_change: "
 794                           "share_entry[%i]->op_type == %d\n",
 795                           i, share_entry->op_type ));
 796
 797                 if (share_entry->op_type == NO_OPLOCK) {
 798                         continue;
 799                 }
 800
 801                 /* Paranoia .... */
 802                 if (EXCLUSIVE_OPLOCK_TYPE(share_entry->op_type)) {
 803                         DEBUG(0,("release_level_2_oplocks_on_change: PANIC. "
 804                                  "share mode entry %d is an exlusive "
 805                                  "oplock !\n", i ));
 806                         TALLOC_FREE(lck);
 807                         abort();
 808                 }
 809
 810                 share_mode_entry_to_message(msg, share_entry);
 811
 812                 /*
 813                  * Deal with a race condition when breaking level2
 814                  * oplocks. Don't send all the messages and release
 815                  * the lock, this allows someone else to come in and
 816                  * get a level2 lock before any of the messages are
 817                  * processed, and thus miss getting a break message.
 818                  * Ensure at least one entry (the one we're breaking)
 819                  * is processed immediately under the lock and becomes
 820                  * set as NO_OPLOCK to stop any waiter getting a level2.
 821                  * Bugid #5980.
 822                  */
 823
 824                 if (procid_is_me(&share_entry->pid)) {
 825                         struct files_struct *cur_fsp =
 826                                 initial_break_processing(fsp->conn->sconn,
 827                                         share_entry->id,
 828                                         share_entry->share_file_id);
 829                         wait_before_sending_break();
 830                         if (cur_fsp != NULL) {
 831                                 break_level2_to_none_async(cur_fsp);
 832                         } else {
 833                                 DEBUG(3, ("release_level_2_oplocks_on_change: "
 834                                 "Did not find fsp, ignoring\n"));
 835                         }
 836                 } else {
 837                         messaging_send_buf(fsp->conn->sconn->msg_ctx,
 838                                         share_entry->pid,
 839                                         MSG_SMB_ASYNC_LEVEL2_BREAK,
 840                                         (uint8 *)msg,
 841                                         MSG_SMB_SHARE_MODE_ENTRY_SIZE);
 842                 }
 843         }
 844
 845         /* We let the message receivers handle removing the oplock state
 846            in the share mode lock db. */
 847
 848         TALLOC_FREE(lck);
 849 }
 850
 851 void contend_level2_oplocks_begin(files_struct *fsp,
 852                                   enum level2_contention_type type)
 853 {
 854         if (koplocks && koplocks->ops->contend_level2_oplocks_begin) {
 855                 koplocks->ops->contend_level2_oplocks_begin(fsp, type);
 856                 return;
 857         }
 858
 859         contend_level2_oplocks_begin_default(fsp, type);
 860 }
 861
 862 void contend_level2_oplocks_end(files_struct *fsp,
 863                                 enum level2_contention_type type)
 864 {
 865         /* Only kernel oplocks implement this so far */
 866         if (koplocks && koplocks->ops->contend_level2_oplocks_end) {
 867                 koplocks->ops->contend_level2_oplocks_end(fsp, type);
 868         }
 869 }
 870
 871 /****************************************************************************
 872  Linearize a share mode entry struct to an internal oplock break message.
 873 ****************************************************************************/
 874
 875 void share_mode_entry_to_message(char *msg, const struct share_mode_entry *e)
 876 {
 877         SIVAL(msg,OP_BREAK_MSG_PID_OFFSET,(uint32)e->pid.pid);
 878         SBVAL(msg,OP_BREAK_MSG_MID_OFFSET,e->op_mid);
 879         SSVAL(msg,OP_BREAK_MSG_OP_TYPE_OFFSET,e->op_type);
 880         SIVAL(msg,OP_BREAK_MSG_ACCESS_MASK_OFFSET,e->access_mask);
 881         SIVAL(msg,OP_BREAK_MSG_SHARE_ACCESS_OFFSET,e->share_access);
 882         SIVAL(msg,OP_BREAK_MSG_PRIV_OFFSET,e->private_options);
 883         SIVAL(msg,OP_BREAK_MSG_TIME_SEC_OFFSET,(uint32_t)e->time.tv_sec);
 884         SIVAL(msg,OP_BREAK_MSG_TIME_USEC_OFFSET,(uint32_t)e->time.tv_usec);
 885         push_file_id_24(msg+OP_BREAK_MSG_DEV_OFFSET, &e->id);
 886         SIVAL(msg,OP_BREAK_MSG_FILE_ID_OFFSET,e->share_file_id);
 887         SIVAL(msg,OP_BREAK_MSG_UID_OFFSET,e->uid);
 888         SSVAL(msg,OP_BREAK_MSG_FLAGS_OFFSET,e->flags);
 889         SIVAL(msg,OP_BREAK_MSG_NAME_HASH_OFFSET,e->name_hash);
 890 #ifdef CLUSTER_SUPPORT
 891         SIVAL(msg,OP_BREAK_MSG_VNN_OFFSET,e->pid.vnn);
 892 #endif
 893 }
 894
 895 /****************************************************************************
 896  De-linearize an internal oplock break message to a share mode entry struct.
 897 ****************************************************************************/
 898
 899 void message_to_share_mode_entry(struct share_mode_entry *e, char *msg)
 900 {
 901         e->pid.pid = (pid_t)IVAL(msg,OP_BREAK_MSG_PID_OFFSET);
 902         e->op_mid = BVAL(msg,OP_BREAK_MSG_MID_OFFSET);
 903         e->op_type = SVAL(msg,OP_BREAK_MSG_OP_TYPE_OFFSET);
 904         e->access_mask = IVAL(msg,OP_BREAK_MSG_ACCESS_MASK_OFFSET);
 905         e->share_access = IVAL(msg,OP_BREAK_MSG_SHARE_ACCESS_OFFSET);
 906         e->private_options = IVAL(msg,OP_BREAK_MSG_PRIV_OFFSET);
 907         e->time.tv_sec = (time_t)IVAL(msg,OP_BREAK_MSG_TIME_SEC_OFFSET);
 908         e->time.tv_usec = (int)IVAL(msg,OP_BREAK_MSG_TIME_USEC_OFFSET);
 909         pull_file_id_24(msg+OP_BREAK_MSG_DEV_OFFSET, &e->id);
 910         e->share_file_id = (unsigned long)IVAL(msg,OP_BREAK_MSG_FILE_ID_OFFSET);
 911         e->uid = (uint32)IVAL(msg,OP_BREAK_MSG_UID_OFFSET);
 912         e->flags = (uint16)SVAL(msg,OP_BREAK_MSG_FLAGS_OFFSET);
 913         e->name_hash = IVAL(msg,OP_BREAK_MSG_NAME_HASH_OFFSET);
 914 #ifdef CLUSTER_SUPPORT
 915         e->pid.vnn = IVAL(msg,OP_BREAK_MSG_VNN_OFFSET);
 916 #endif
 917 }
 918
 919 /****************************************************************************
 920  Setup oplocks for this process.
 921 ****************************************************************************/
 922
 923 bool init_oplocks(struct messaging_context *msg_ctx)
 924 {
 925         DEBUG(3,("init_oplocks: initializing messages.\n"));
 926
 927         messaging_register(msg_ctx, NULL, MSG_SMB_BREAK_REQUEST,
 928                            process_oplock_break_message);
 929         messaging_register(msg_ctx, NULL, MSG_SMB_ASYNC_LEVEL2_BREAK,
 930                            process_oplock_async_level2_break_message);
 931         messaging_register(msg_ctx, NULL, MSG_SMB_BREAK_RESPONSE,
 932                            process_oplock_break_response);
 933         messaging_register(msg_ctx, NULL, MSG_SMB_KERNEL_BREAK,
 934                            process_kernel_oplock_break);
 935         messaging_register(msg_ctx, NULL, MSG_SMB_OPEN_RETRY,
 936                            process_open_retry_message);
 937
 938         if (lp_kernel_oplocks()) {
 939 #if HAVE_KERNEL_OPLOCKS_IRIX
 940                 koplocks = irix_init_kernel_oplocks(NULL);
 941 #elif HAVE_KERNEL_OPLOCKS_LINUX
 942                 koplocks = linux_init_kernel_oplocks(NULL);
 943 #elif HAVE_ONEFS
 944 #error Isilon, please check if the NULL context is okay here. Thanks!
 945                 koplocks = onefs_init_kernel_oplocks(NULL);
 946 #endif
 947         }
 948
 949         return True;
 950 }