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/smbd.h"
  25 #include "smbd/globals.h"
  26 #include "messages.h"
  27 #include "../librpc/gen_ndr/open_files.h"
  28
  29 /****************************************************************************
  30  Get the number of current exclusive oplocks.
  31 ****************************************************************************/
  32
  33 int32 get_number_of_exclusive_open_oplocks(void)
  34 {
  35   return exclusive_oplocks_open;
  36 }
  37
  38 /*
  39  * helper function used by the kernel oplock backends to post the break message
  40  */
  41 void break_kernel_oplock(struct messaging_context *msg_ctx, files_struct *fsp)
  42 {
  43         uint8_t msg[MSG_SMB_KERNEL_BREAK_SIZE];
  44
  45         /* Put the kernel break info into the message. */
  46         push_file_id_24((char *)msg, &fsp->file_id);
  47         SIVAL(msg,24,fsp->fh->gen_id);
  48
  49         /* Don't need to be root here as we're only ever
  50            sending to ourselves. */
  51
  52         messaging_send_buf(msg_ctx, messaging_server_id(msg_ctx),
  53                            MSG_SMB_KERNEL_BREAK,
  54                            msg, MSG_SMB_KERNEL_BREAK_SIZE);
  55 }
  56
  57 /****************************************************************************
  58  Attempt to set an oplock on a file. Succeeds if kernel oplocks are
  59  disabled (just sets flags) and no byte-range locks in the file. Returns True
  60  if oplock set.
  61 ****************************************************************************/
  62
  63 bool set_file_oplock(files_struct *fsp, int oplock_type)
  64 {
  65
  66         bool use_kernel = lp_kernel_oplocks(SNUM(fsp->conn)) && koplocks;
  67
  68         if (fsp->oplock_type == LEVEL_II_OPLOCK) {
  69                 if (use_kernel &&
  70                     !(koplocks->flags & KOPLOCKS_LEVEL2_SUPPORTED)) {
  71                         DEBUG(10, ("Refusing level2 oplock, kernel oplocks "
  72                                    "don't support them\n"));
  73                         return false;
  74                 }
  75         }
  76
  77         if ((fsp->oplock_type != NO_OPLOCK) &&
  78             (fsp->oplock_type != FAKE_LEVEL_II_OPLOCK) &&
  79             use_kernel &&
  80             !koplocks->ops->set_oplock(koplocks, fsp, oplock_type)) {
  81                 return False;
  82         }
  83
  84         fsp->oplock_type = oplock_type;
  85         fsp->sent_oplock_break = NO_BREAK_SENT;
  86         if (oplock_type == LEVEL_II_OPLOCK) {
  87                 level_II_oplocks_open++;
  88         } else if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
  89                 exclusive_oplocks_open++;
  90         }
  91
  92         DEBUG(5,("set_file_oplock: granted oplock on file %s, %s/%lu, "
  93                     "tv_sec = %x, tv_usec = %x\n",
  94                  fsp_str_dbg(fsp), file_id_string_tos(&fsp->file_id),
  95                  fsp->fh->gen_id, (int)fsp->open_time.tv_sec,
  96                  (int)fsp->open_time.tv_usec ));
  97
  98         return True;
  99 }
 100
 101 /****************************************************************************
 102  Attempt to release an oplock on a file. Decrements oplock count.
 103 ****************************************************************************/
 104
 105 void release_file_oplock(files_struct *fsp)
 106 {
 107         if ((fsp->oplock_type != NO_OPLOCK) &&
 108             (fsp->oplock_type != FAKE_LEVEL_II_OPLOCK) &&
 109             koplocks) {
 110                 koplocks->ops->release_oplock(koplocks, fsp, NO_OPLOCK);
 111         }
 112
 113         if (fsp->oplock_type == LEVEL_II_OPLOCK) {
 114                 level_II_oplocks_open--;
 115         } else if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
 116                 exclusive_oplocks_open--;
 117         }
 118
 119         SMB_ASSERT(exclusive_oplocks_open>=0);
 120         SMB_ASSERT(level_II_oplocks_open>=0);
 121
 122         if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
 123                 /* This doesn't matter for close. */
 124                 fsp->oplock_type = FAKE_LEVEL_II_OPLOCK;
 125         } else {
 126                 fsp->oplock_type = NO_OPLOCK;
 127         }
 128         fsp->sent_oplock_break = NO_BREAK_SENT;
 129
 130         flush_write_cache(fsp, OPLOCK_RELEASE_FLUSH);
 131         delete_write_cache(fsp);
 132
 133         TALLOC_FREE(fsp->oplock_timeout);
 134 }
 135
 136 /****************************************************************************
 137  Attempt to downgrade an oplock on a file. Doesn't decrement oplock count.
 138 ****************************************************************************/
 139
 140 static void downgrade_file_oplock(files_struct *fsp)
 141 {
 142         if (!EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
 143                 DEBUG(0, ("trying to downgrade an already-downgraded oplock!\n"));
 144                 return;
 145         }
 146
 147         if (koplocks) {
 148                 koplocks->ops->release_oplock(koplocks, fsp, LEVEL_II_OPLOCK);
 149         }
 150         fsp->oplock_type = LEVEL_II_OPLOCK;
 151         exclusive_oplocks_open--;
 152         level_II_oplocks_open++;
 153         fsp->sent_oplock_break = NO_BREAK_SENT;
 154 }
 155
 156 /****************************************************************************
 157  Remove a file oplock. Copes with level II and exclusive.
 158  Locks then unlocks the share mode lock. Client can decide to go directly
 159  to none even if a "break-to-level II" was sent.
 160 ****************************************************************************/
 161
 162 bool remove_oplock(files_struct *fsp)
 163 {
 164         bool ret;
 165         struct share_mode_lock *lck;
 166
 167         /* Remove the oplock flag from the sharemode. */
 168         lck = get_existing_share_mode_lock(talloc_tos(), fsp->file_id);
 169         if (lck == NULL) {
 170                 DEBUG(0,("remove_oplock: failed to lock share entry for "
 171                          "file %s\n", fsp_str_dbg(fsp)));
 172                 return False;
 173         }
 174         ret = remove_share_oplock(lck, fsp);
 175         if (!ret) {
 176                 DEBUG(0,("remove_oplock: failed to remove share oplock for "
 177                          "file %s fnum %d, %s\n",
 178                          fsp_str_dbg(fsp), fsp->fnum,
 179                          file_id_string_tos(&fsp->file_id)));
 180         }
 181         release_file_oplock(fsp);
 182         TALLOC_FREE(lck);
 183         return ret;
 184 }
 185
 186 /*
 187  * Deal with a reply when a break-to-level II was sent.
 188  */
 189 bool downgrade_oplock(files_struct *fsp)
 190 {
 191         bool ret;
 192         struct share_mode_lock *lck;
 193
 194         lck = get_existing_share_mode_lock(talloc_tos(), fsp->file_id);
 195         if (lck == NULL) {
 196                 DEBUG(0,("downgrade_oplock: failed to lock share entry for "
 197                          "file %s\n", fsp_str_dbg(fsp)));
 198                 return False;
 199         }
 200         ret = downgrade_share_oplock(lck, fsp);
 201         if (!ret) {
 202                 DEBUG(0,("downgrade_oplock: failed to downgrade share oplock "
 203                          "for file %s fnum %d, file_id %s\n",
 204                          fsp_str_dbg(fsp), fsp->fnum,
 205                          file_id_string_tos(&fsp->file_id)));
 206         }
 207
 208         downgrade_file_oplock(fsp);
 209         TALLOC_FREE(lck);
 210         return ret;
 211 }
 212
 213 /*
 214  * Some kernel oplock implementations handle the notification themselves.
 215  */
 216 bool should_notify_deferred_opens()
 217 {
 218         return !(koplocks &&
 219                 (koplocks->flags & KOPLOCKS_DEFERRED_OPEN_NOTIFICATION));
 220 }
 221
 222 /****************************************************************************
 223  Set up an oplock break message.
 224 ****************************************************************************/
 225
 226 static char *new_break_message_smb1(TALLOC_CTX *mem_ctx,
 227                                    files_struct *fsp, int cmd)
 228 {
 229         char *result = talloc_array(mem_ctx, char, smb_size + 8*2 + 0);
 230
 231         if (result == NULL) {
 232                 DEBUG(0, ("talloc failed\n"));
 233                 return NULL;
 234         }
 235
 236         memset(result,'\0',smb_size);
 237         srv_set_message(result,8,0,true);
 238         SCVAL(result,smb_com,SMBlockingX);
 239         SSVAL(result,smb_tid,fsp->conn->cnum);
 240         SSVAL(result,smb_pid,0xFFFF);
 241         SSVAL(result,smb_uid,0);
 242         SSVAL(result,smb_mid,0xFFFF);
 243         SCVAL(result,smb_vwv0,0xFF);
 244         SSVAL(result,smb_vwv2,fsp->fnum);
 245         SCVAL(result,smb_vwv3,LOCKING_ANDX_OPLOCK_RELEASE);
 246         SCVAL(result,smb_vwv3+1,cmd);
 247         return result;
 248 }
 249
 250 /****************************************************************************
 251  Function to do the waiting before sending a local break.
 252 ****************************************************************************/
 253
 254 static void wait_before_sending_break(void)
 255 {
 256         long wait_time = (long)lp_oplock_break_wait_time();
 257
 258         if (wait_time) {
 259                 smb_msleep(wait_time);
 260         }
 261 }
 262
 263 /****************************************************************************
 264  Ensure that we have a valid oplock.
 265 ****************************************************************************/
 266
 267 static files_struct *initial_break_processing(
 268         struct smbd_server_connection *sconn, struct file_id id,
 269         unsigned long file_id)
 270 {
 271         files_struct *fsp = NULL;
 272
 273         if( DEBUGLVL( 3 ) ) {
 274                 dbgtext( "initial_break_processing: called for %s/%u\n",
 275                          file_id_string_tos(&id), (int)file_id);
 276                 dbgtext( "Current oplocks_open (exclusive = %d, levelII = %d)\n",
 277                         exclusive_oplocks_open, level_II_oplocks_open );
 278         }
 279
 280         /*
 281          * We need to search the file open table for the
 282          * entry containing this dev and inode, and ensure
 283          * we have an oplock on it.
 284          */
 285
 286         fsp = file_find_dif(sconn, id, file_id);
 287
 288         if(fsp == NULL) {
 289                 /* The file could have been closed in the meantime - return success. */
 290                 if( DEBUGLVL( 3 ) ) {
 291                         dbgtext( "initial_break_processing: cannot find open file with " );
 292                         dbgtext( "file_id %s gen_id = %lu", file_id_string_tos(&id), file_id);
 293                         dbgtext( "allowing break to succeed.\n" );
 294                 }
 295                 return NULL;
 296         }
 297
 298         /* Ensure we have an oplock on the file */
 299
 300         /*
 301          * There is a potential race condition in that an oplock could
 302          * have been broken due to another udp request, and yet there are
 303          * still oplock break messages being sent in the udp message
 304          * queue for this file. So return true if we don't have an oplock,
 305          * as we may have just freed it.
 306          */
 307
 308         if(fsp->oplock_type == NO_OPLOCK) {
 309                 if( DEBUGLVL( 3 ) ) {
 310                         dbgtext( "initial_break_processing: file %s ",
 311                                  fsp_str_dbg(fsp));
 312                         dbgtext( "(file_id = %s gen_id = %lu) has no oplock.\n",
 313                                  file_id_string_tos(&id), fsp->fh->gen_id );
 314                         dbgtext( "Allowing break to succeed regardless.\n" );
 315                 }
 316                 return NULL;
 317         }
 318
 319         return fsp;
 320 }
 321
 322 static void oplock_timeout_handler(struct event_context *ctx,
 323                                    struct timed_event *te,
 324                                    struct timeval now,
 325                                    void *private_data)
 326 {
 327         files_struct *fsp = (files_struct *)private_data;
 328
 329         /* Remove the timed event handler. */
 330         TALLOC_FREE(fsp->oplock_timeout);
 331         DEBUG(0, ("Oplock break failed for file %s -- replying anyway\n",
 332                   fsp_str_dbg(fsp)));
 333         remove_oplock(fsp);
 334         reply_to_oplock_break_requests(fsp);
 335 }
 336
 337 /*******************************************************************
 338  Add a timeout handler waiting for the client reply.
 339 *******************************************************************/
 340
 341 static void add_oplock_timeout_handler(files_struct *fsp)
 342 {
 343         /*
 344          * If kernel oplocks already notifies smbds when an oplock break times
 345          * out, just return.
 346          */
 347         if (koplocks &&
 348             (koplocks->flags & KOPLOCKS_TIMEOUT_NOTIFICATION)) {
 349                 return;
 350         }
 351
 352         if (fsp->oplock_timeout != NULL) {
 353                 DEBUG(0, ("Logic problem -- have an oplock event hanging "
 354                           "around\n"));
 355         }
 356
 357         fsp->oplock_timeout =
 358                 tevent_add_timer(fsp->conn->sconn->ev_ctx, fsp,
 359                                  timeval_current_ofs(OPLOCK_BREAK_TIMEOUT, 0),
 360                                  oplock_timeout_handler, fsp);
 361
 362         if (fsp->oplock_timeout == NULL) {
 363                 DEBUG(0, ("Could not add oplock timeout handler\n"));
 364         }
 365 }
 366
 367 static void send_break_message_smb1(files_struct *fsp, int level)
 368 {
 369         char *break_msg = new_break_message_smb1(talloc_tos(),
 370                                         fsp,
 371                                         level);
 372         if (break_msg == NULL) {
 373                 exit_server("Could not talloc break_msg\n");
 374         }
 375
 376         show_msg(break_msg);
 377         if (!srv_send_smb(fsp->conn->sconn,
 378                         break_msg, false, 0,
 379                         IS_CONN_ENCRYPTED(fsp->conn),
 380                         NULL)) {
 381                 exit_server_cleanly("send_break_message_smb1: "
 382                         "srv_send_smb failed.");
 383         }
 384
 385         TALLOC_FREE(break_msg);
 386 }
 387
 388 void break_level2_to_none_async(files_struct *fsp)
 389 {
 390         struct smbd_server_connection *sconn = fsp->conn->sconn;
 391
 392         if (fsp->oplock_type == NO_OPLOCK) {
 393                 /* We already got a "break to none" message and we've handled
 394                  * it.  just ignore. */
 395                 DEBUG(3, ("process_oplock_async_level2_break_message: already "
 396                           "broken to none, ignoring.\n"));
 397                 return;
 398         }
 399
 400         if (fsp->oplock_type == FAKE_LEVEL_II_OPLOCK) {
 401                 /* Don't tell the client, just downgrade. */
 402                 DEBUG(3, ("process_oplock_async_level2_break_message: "
 403                           "downgrading fake level 2 oplock.\n"));
 404                 remove_oplock(fsp);
 405                 return;
 406         }
 407
 408         /* Ensure we're really at level2 state. */
 409         SMB_ASSERT(fsp->oplock_type == LEVEL_II_OPLOCK);
 410
 411         DEBUG(10,("process_oplock_async_level2_break_message: sending break "
 412                   "to none message for fid %d, file %s\n", fsp->fnum,
 413                   fsp_str_dbg(fsp)));
 414
 415         /* Now send a break to none message to our client. */
 416         if (sconn->using_smb2) {
 417                 send_break_message_smb2(fsp, OPLOCKLEVEL_NONE);
 418         } else {
 419                 send_break_message_smb1(fsp, OPLOCKLEVEL_NONE);
 420         }
 421
 422         /* Async level2 request, don't send a reply, just remove the oplock. */
 423         remove_oplock(fsp);
 424 }
 425
 426 /*******************************************************************
 427  This handles the case of a write triggering a break to none
 428  message on a level2 oplock.
 429  When we get this message we may be in any of three states :
 430  NO_OPLOCK, LEVEL_II, FAKE_LEVEL2. We only send a message to
 431  the client for LEVEL2.
 432 *******************************************************************/
 433
 434 static void process_oplock_async_level2_break_message(struct messaging_context *msg_ctx,
 435                                                       void *private_data,
 436                                                       uint32_t msg_type,
 437                                                       struct server_id src,
 438                                                       DATA_BLOB *data)
 439 {
 440         struct share_mode_entry msg;
 441         files_struct *fsp;
 442         struct smbd_server_connection *sconn =
 443                 talloc_get_type_abort(private_data,
 444                 struct smbd_server_connection);
 445
 446         if (data->data == NULL) {
 447                 DEBUG(0, ("Got NULL buffer\n"));
 448                 return;
 449         }
 450
 451         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
 452                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
 453                 return;
 454         }
 455
 456         /* De-linearize incoming message. */
 457         message_to_share_mode_entry(&msg, (char *)data->data);
 458
 459         DEBUG(10, ("Got oplock async level 2 break message from pid %s: "
 460                    "%s/%llu\n", server_id_str(talloc_tos(), &src),
 461                    file_id_string_tos(&msg.id),
 462                    (unsigned long long)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 share_mode_entry msg;
 488         files_struct *fsp;
 489         bool break_to_level2 = False;
 490         bool use_kernel;
 491         struct smbd_server_connection *sconn =
 492                 talloc_get_type_abort(private_data,
 493                 struct smbd_server_connection);
 494
 495         if (data->data == NULL) {
 496                 DEBUG(0, ("Got NULL buffer\n"));
 497                 return;
 498         }
 499
 500         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
 501                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
 502                 return;
 503         }
 504
 505         /* De-linearize incoming message. */
 506         message_to_share_mode_entry(&msg, (char *)data->data);
 507
 508         DEBUG(10, ("Got oplock break message from pid %s: %s/%llu\n",
 509                    server_id_str(talloc_tos(), &src),
 510                    file_id_string_tos(&msg.id),
 511                    (unsigned long long)msg.share_file_id));
 512
 513         fsp = initial_break_processing(sconn, msg.id, msg.share_file_id);
 514
 515         if (fsp == NULL) {
 516                 /* We hit a race here. Break messages are sent, and before we
 517                  * get to process this message, we have closed the file. Reply
 518                  * with 'ok, oplock broken' */
 519                 DEBUG(3, ("Did not find fsp\n"));
 520
 521                 /* We just send the same message back. */
 522                 messaging_send_buf(msg_ctx, src, MSG_SMB_BREAK_RESPONSE,
 523                                    (uint8 *)data->data,
 524                                    MSG_SMB_SHARE_MODE_ENTRY_SIZE);
 525                 return;
 526         }
 527
 528         if (fsp->sent_oplock_break != NO_BREAK_SENT) {
 529                 /* Remember we have to inform the requesting PID when the
 530                  * client replies */
 531                 msg.pid = src;
 532                 ADD_TO_ARRAY(NULL, struct share_mode_entry, msg,
 533                              &fsp->pending_break_messages,
 534                              &fsp->num_pending_break_messages);
 535                 return;
 536         }
 537
 538         if (EXCLUSIVE_OPLOCK_TYPE(msg.op_type) &&
 539             !EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
 540                 DEBUG(3, ("Already downgraded oplock on %s: %s\n",
 541                           file_id_string_tos(&fsp->file_id),
 542                           fsp_str_dbg(fsp)));
 543                 /* We just send the same message back. */
 544                 messaging_send_buf(msg_ctx, src, MSG_SMB_BREAK_RESPONSE,
 545                                    (uint8 *)data->data,
 546                                    MSG_SMB_SHARE_MODE_ENTRY_SIZE);
 547                 return;
 548         }
 549
 550         use_kernel = lp_kernel_oplocks(SNUM(fsp->conn)) && koplocks;
 551
 552         if ((global_client_caps & CAP_LEVEL_II_OPLOCKS) &&
 553             !(msg.op_type & FORCE_OPLOCK_BREAK_TO_NONE) &&
 554             !(use_kernel && !(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 file_id id;
 594         unsigned long file_id;
 595         files_struct *fsp;
 596         struct smbd_server_connection *sconn =
 597                 talloc_get_type_abort(private_data,
 598                 struct smbd_server_connection);
 599
 600         if (data->data == NULL) {
 601                 DEBUG(0, ("Got NULL buffer\n"));
 602                 return;
 603         }
 604
 605         if (data->length != MSG_SMB_KERNEL_BREAK_SIZE) {
 606                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
 607                 return;
 608         }
 609
 610         /* Pull the data from the message. */
 611         pull_file_id_24((char *)data->data, &id);
 612         file_id = (unsigned long)IVAL(data->data, 24);
 613
 614         DEBUG(10, ("Got kernel oplock break message from pid %s: %s/%u\n",
 615                    server_id_str(talloc_tos(), &src), file_id_string_tos(&id),
 616                    (unsigned int)file_id));
 617
 618         fsp = initial_break_processing(sconn, id, file_id);
 619
 620         if (fsp == NULL) {
 621                 DEBUG(3, ("Got a kernel oplock break message for a file "
 622                           "I don't know about\n"));
 623                 return;
 624         }
 625
 626         if (fsp->sent_oplock_break != NO_BREAK_SENT) {
 627                 /* This is ok, kernel oplocks come in completely async */
 628                 DEBUG(3, ("Got a kernel oplock request while waiting for a "
 629                           "break reply\n"));
 630                 return;
 631         }
 632
 633         if (sconn->using_smb2) {
 634                 send_break_message_smb2(fsp, OPLOCKLEVEL_NONE);
 635         } else {
 636                 send_break_message_smb1(fsp, OPLOCKLEVEL_NONE);
 637         }
 638
 639         fsp->sent_oplock_break = BREAK_TO_NONE_SENT;
 640
 641         add_oplock_timeout_handler(fsp);
 642 }
 643
 644 void reply_to_oplock_break_requests(files_struct *fsp)
 645 {
 646         int i;
 647
 648         /*
 649          * If kernel oplocks already notifies smbds when oplocks are
 650          * broken/removed, just return.
 651          */
 652         if (koplocks &&
 653             (koplocks->flags & KOPLOCKS_OPLOCK_BROKEN_NOTIFICATION)) {
 654                 return;
 655         }
 656
 657         for (i=0; i<fsp->num_pending_break_messages; i++) {
 658                 struct share_mode_entry *e = &fsp->pending_break_messages[i];
 659                 char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
 660
 661                 share_mode_entry_to_message(msg, e);
 662
 663                 messaging_send_buf(fsp->conn->sconn->msg_ctx, e->pid,
 664                                    MSG_SMB_BREAK_RESPONSE,
 665                                    (uint8 *)msg,
 666                                    MSG_SMB_SHARE_MODE_ENTRY_SIZE);
 667         }
 668
 669         SAFE_FREE(fsp->pending_break_messages);
 670         fsp->num_pending_break_messages = 0;
 671         TALLOC_FREE(fsp->oplock_timeout);
 672         return;
 673 }
 674
 675 static void process_oplock_break_response(struct messaging_context *msg_ctx,
 676                                           void *private_data,
 677                                           uint32_t msg_type,
 678                                           struct server_id src,
 679                                           DATA_BLOB *data)
 680 {
 681         struct share_mode_entry msg;
 682         struct smbd_server_connection *sconn =
 683                 talloc_get_type_abort(private_data,
 684                 struct smbd_server_connection);
 685
 686         if (data->data == NULL) {
 687                 DEBUG(0, ("Got NULL buffer\n"));
 688                 return;
 689         }
 690
 691         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
 692                 DEBUG(0, ("Got invalid msg len %u\n",
 693                           (unsigned int)data->length));
 694                 return;
 695         }
 696
 697         /* De-linearize incoming message. */
 698         message_to_share_mode_entry(&msg, (char *)data->data);
 699
 700         DEBUG(10, ("Got oplock break response from pid %s: %s/%llu mid %llu\n",
 701                    server_id_str(talloc_tos(), &src),
 702                    file_id_string_tos(&msg.id),
 703                    (unsigned long long)msg.share_file_id,
 704                    (unsigned long long)msg.op_mid));
 705
 706         schedule_deferred_open_message_smb(sconn, msg.op_mid);
 707 }
 708
 709 static void process_open_retry_message(struct messaging_context *msg_ctx,
 710                                        void *private_data,
 711                                        uint32_t msg_type,
 712                                        struct server_id src,
 713                                        DATA_BLOB *data)
 714 {
 715         struct share_mode_entry msg;
 716         struct smbd_server_connection *sconn =
 717                 talloc_get_type_abort(private_data,
 718                 struct smbd_server_connection);
 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                    server_id_str(talloc_tos(), &src), file_id_string_tos(&msg.id),
 735                    (unsigned long long)msg.op_mid));
 736
 737         schedule_deferred_open_message_smb(sconn, msg.op_mid);
 738 }
 739
 740 struct break_to_none_state {
 741         struct smbd_server_connection *sconn;
 742         struct file_id id;
 743 };
 744 static void do_break_to_none(struct tevent_req *req);
 745
 746 /****************************************************************************
 747  This function is called on any file modification or lock request. If a file
 748  is level 2 oplocked then it must tell all other level 2 holders to break to
 749  none.
 750 ****************************************************************************/
 751
 752 static void contend_level2_oplocks_begin_default(files_struct *fsp,
 753                                               enum level2_contention_type type)
 754 {
 755         struct smbd_server_connection *sconn = fsp->conn->sconn;
 756         struct tevent_req *req;
 757         struct break_to_none_state *state;
 758
 759         /*
 760          * If this file is level II oplocked then we need
 761          * to grab the shared memory lock and inform all
 762          * other files with a level II lock that they need
 763          * to flush their read caches. We keep the lock over
 764          * the shared memory area whilst doing this.
 765          */
 766
 767         if (!LEVEL_II_OPLOCK_TYPE(fsp->oplock_type))
 768                 return;
 769
 770         /*
 771          * When we get here we might have a brlock entry locked. Also
 772          * locking the share mode entry would violate the locking
 773          * order. Breaking level2 oplocks to none is asynchronous
 774          * anyway, so we postpone this into an immediate timed event.
 775          */
 776
 777         state = talloc(sconn, struct break_to_none_state);
 778         if (state == NULL) {
 779                 DEBUG(1, ("talloc failed\n"));
 780                 return;
 781         }
 782         state->sconn = sconn;
 783         state->id = fsp->file_id;
 784
 785         req = tevent_wakeup_send(state, sconn->ev_ctx, timeval_set(0, 0));
 786         if (req == NULL) {
 787                 DEBUG(1, ("tevent_wakeup_send failed\n"));
 788                 TALLOC_FREE(state);
 789                 return;
 790         }
 791         tevent_req_set_callback(req, do_break_to_none, state);
 792         return;
 793 }
 794
 795 static void do_break_to_none(struct tevent_req *req)
 796 {
 797         struct break_to_none_state *state = tevent_req_callback_data(
 798                 req, struct break_to_none_state);
 799         bool ret;
 800         int i;
 801         struct share_mode_lock *lck;
 802
 803         ret = tevent_wakeup_recv(req);
 804         TALLOC_FREE(req);
 805         if (!ret) {
 806                 DEBUG(1, ("tevent_wakeup_recv failed\n"));
 807                 goto done;
 808         }
 809         lck = get_existing_share_mode_lock(talloc_tos(), state->id);
 810         if (lck == NULL) {
 811                 DEBUG(1, ("release_level_2_oplocks_on_change: failed to lock "
 812                           "share mode entry for file %s.\n",
 813                           file_id_string_tos(&state->id)));
 814                 goto done;
 815         }
 816
 817         DEBUG(10,("release_level_2_oplocks_on_change: num_share_modes = %d\n",
 818                   lck->data->num_share_modes ));
 819
 820         for(i = 0; i < lck->data->num_share_modes; i++) {
 821                 struct share_mode_entry *share_entry = &lck->data->share_modes[i];
 822                 char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
 823
 824                 if (!is_valid_share_mode_entry(share_entry)) {
 825                         continue;
 826                 }
 827
 828                 /*
 829                  * As there could have been multiple writes waiting at the
 830                  * lock_share_entry gate we may not be the first to
 831                  * enter. Hence the state of the op_types in the share mode
 832                  * entries may be partly NO_OPLOCK and partly LEVEL_II or FAKE_LEVEL_II
 833                  * oplock. It will do no harm to re-send break messages to
 834                  * those smbd's that are still waiting their turn to remove
 835                  * their LEVEL_II state, and also no harm to ignore existing
 836                  * NO_OPLOCK states. JRA.
 837                  */
 838
 839                 DEBUG(10,("release_level_2_oplocks_on_change: "
 840                           "share_entry[%i]->op_type == %d\n",
 841                           i, share_entry->op_type ));
 842
 843                 if (share_entry->op_type == NO_OPLOCK) {
 844                         continue;
 845                 }
 846
 847                 /* Paranoia .... */
 848                 if (EXCLUSIVE_OPLOCK_TYPE(share_entry->op_type)) {
 849                         DEBUG(0,("release_level_2_oplocks_on_change: PANIC. "
 850                                  "share mode entry %d is an exlusive "
 851                                  "oplock !\n", i ));
 852                         TALLOC_FREE(lck);
 853                         abort();
 854                 }
 855
 856                 share_mode_entry_to_message(msg, share_entry);
 857
 858                 /*
 859                  * Deal with a race condition when breaking level2
 860                  * oplocks. Don't send all the messages and release
 861                  * the lock, this allows someone else to come in and
 862                  * get a level2 lock before any of the messages are
 863                  * processed, and thus miss getting a break message.
 864                  * Ensure at least one entry (the one we're breaking)
 865                  * is processed immediately under the lock and becomes
 866                  * set as NO_OPLOCK to stop any waiter getting a level2.
 867                  * Bugid #5980.
 868                  */
 869
 870                 if (procid_is_me(&share_entry->pid)) {
 871                         struct files_struct *cur_fsp =
 872                                 initial_break_processing(state->sconn,
 873                                         share_entry->id,
 874                                         share_entry->share_file_id);
 875                         wait_before_sending_break();
 876                         if (cur_fsp != NULL) {
 877                                 break_level2_to_none_async(cur_fsp);
 878                         } else {
 879                                 DEBUG(3, ("release_level_2_oplocks_on_change: "
 880                                 "Did not find fsp, ignoring\n"));
 881                         }
 882                 } else {
 883                         messaging_send_buf(state->sconn->msg_ctx,
 884                                         share_entry->pid,
 885                                         MSG_SMB_ASYNC_LEVEL2_BREAK,
 886                                         (uint8 *)msg,
 887                                         MSG_SMB_SHARE_MODE_ENTRY_SIZE);
 888                 }
 889         }
 890
 891         /* We let the message receivers handle removing the oplock state
 892            in the share mode lock db. */
 893
 894         TALLOC_FREE(lck);
 895 done:
 896         TALLOC_FREE(state);
 897         return;
 898 }
 899
 900 void smbd_contend_level2_oplocks_begin(files_struct *fsp,
 901                                   enum level2_contention_type type)
 902 {
 903         if (koplocks && koplocks->ops->contend_level2_oplocks_begin) {
 904                 koplocks->ops->contend_level2_oplocks_begin(fsp, type);
 905                 return;
 906         }
 907
 908         contend_level2_oplocks_begin_default(fsp, type);
 909 }
 910
 911 void smbd_contend_level2_oplocks_end(files_struct *fsp,
 912                                 enum level2_contention_type type)
 913 {
 914         /* Only kernel oplocks implement this so far */
 915         if (koplocks && koplocks->ops->contend_level2_oplocks_end) {
 916                 koplocks->ops->contend_level2_oplocks_end(fsp, type);
 917         }
 918 }
 919
 920 /****************************************************************************
 921  Linearize a share mode entry struct to an internal oplock break message.
 922 ****************************************************************************/
 923
 924 void share_mode_entry_to_message(char *msg, const struct share_mode_entry *e)
 925 {
 926         SIVAL(msg,OP_BREAK_MSG_PID_OFFSET,(uint32)e->pid.pid);
 927         SBVAL(msg,OP_BREAK_MSG_MID_OFFSET,e->op_mid);
 928         SSVAL(msg,OP_BREAK_MSG_OP_TYPE_OFFSET,e->op_type);
 929         SIVAL(msg,OP_BREAK_MSG_ACCESS_MASK_OFFSET,e->access_mask);
 930         SIVAL(msg,OP_BREAK_MSG_SHARE_ACCESS_OFFSET,e->share_access);
 931         SIVAL(msg,OP_BREAK_MSG_PRIV_OFFSET,e->private_options);
 932         SIVAL(msg,OP_BREAK_MSG_TIME_SEC_OFFSET,(uint32_t)e->time.tv_sec);
 933         SIVAL(msg,OP_BREAK_MSG_TIME_USEC_OFFSET,(uint32_t)e->time.tv_usec);
 934         push_file_id_24(msg+OP_BREAK_MSG_DEV_OFFSET, &e->id);
 935         SIVAL(msg,OP_BREAK_MSG_FILE_ID_OFFSET,e->share_file_id);
 936         SIVAL(msg,OP_BREAK_MSG_UID_OFFSET,e->uid);
 937         SSVAL(msg,OP_BREAK_MSG_FLAGS_OFFSET,e->flags);
 938         SIVAL(msg,OP_BREAK_MSG_NAME_HASH_OFFSET,e->name_hash);
 939         SIVAL(msg,OP_BREAK_MSG_VNN_OFFSET,e->pid.vnn);
 940 }
 941
 942 /****************************************************************************
 943  De-linearize an internal oplock break message to a share mode entry struct.
 944 ****************************************************************************/
 945
 946 void message_to_share_mode_entry(struct share_mode_entry *e, char *msg)
 947 {
 948         e->pid.pid = (pid_t)IVAL(msg,OP_BREAK_MSG_PID_OFFSET);
 949         e->op_mid = BVAL(msg,OP_BREAK_MSG_MID_OFFSET);
 950         e->op_type = SVAL(msg,OP_BREAK_MSG_OP_TYPE_OFFSET);
 951         e->access_mask = IVAL(msg,OP_BREAK_MSG_ACCESS_MASK_OFFSET);
 952         e->share_access = IVAL(msg,OP_BREAK_MSG_SHARE_ACCESS_OFFSET);
 953         e->private_options = IVAL(msg,OP_BREAK_MSG_PRIV_OFFSET);
 954         e->time.tv_sec = (time_t)IVAL(msg,OP_BREAK_MSG_TIME_SEC_OFFSET);
 955         e->time.tv_usec = (int)IVAL(msg,OP_BREAK_MSG_TIME_USEC_OFFSET);
 956         pull_file_id_24(msg+OP_BREAK_MSG_DEV_OFFSET, &e->id);
 957         e->share_file_id = (unsigned long)IVAL(msg,OP_BREAK_MSG_FILE_ID_OFFSET);
 958         e->uid = (uint32)IVAL(msg,OP_BREAK_MSG_UID_OFFSET);
 959         e->flags = (uint16)SVAL(msg,OP_BREAK_MSG_FLAGS_OFFSET);
 960         e->name_hash = IVAL(msg,OP_BREAK_MSG_NAME_HASH_OFFSET);
 961         e->pid.vnn = IVAL(msg,OP_BREAK_MSG_VNN_OFFSET);
 962 }
 963
 964 /****************************************************************************
 965  Setup oplocks for this process.
 966 ****************************************************************************/
 967
 968 bool init_oplocks(struct smbd_server_connection *sconn)
 969 {
 970         DEBUG(3,("init_oplocks: initializing messages.\n"));
 971
 972         messaging_register(sconn->msg_ctx, sconn, MSG_SMB_BREAK_REQUEST,
 973                            process_oplock_break_message);
 974         messaging_register(sconn->msg_ctx, sconn, MSG_SMB_ASYNC_LEVEL2_BREAK,
 975                            process_oplock_async_level2_break_message);
 976         messaging_register(sconn->msg_ctx, sconn, MSG_SMB_BREAK_RESPONSE,
 977                            process_oplock_break_response);
 978         messaging_register(sconn->msg_ctx, sconn, MSG_SMB_KERNEL_BREAK,
 979                            process_kernel_oplock_break);
 980         messaging_register(sconn->msg_ctx, sconn, MSG_SMB_OPEN_RETRY,
 981                            process_open_retry_message);
 982
 983         return true;
 984 }
 985
 986 void init_kernel_oplocks(struct smbd_server_connection *sconn)
 987 {
 988         /* only initialize once */
 989         if (koplocks == NULL) {
 990 #if HAVE_KERNEL_OPLOCKS_IRIX
 991                 koplocks = irix_init_kernel_oplocks(sconn);
 992 #elif HAVE_KERNEL_OPLOCKS_LINUX
 993                 koplocks = linux_init_kernel_oplocks(sconn);
 994 #elif HAVE_ONEFS
 995 #error Isilon, please check if the NULL context is okay here. Thanks!
 996                 koplocks = onefs_init_kernel_oplocks(sconn);
 997 #endif
 998         }
 999 }