b39e5bf6349d852bca47a837ccf9365c81c2878d
[samba.git] / 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
26 /****************************************************************************
27  Get the number of current exclusive oplocks.
28 ****************************************************************************/
29
30 int32 get_number_of_exclusive_open_oplocks(void)
31 {
32   return exclusive_oplocks_open;
33 }
34
35 /*
36  * helper function used by the kernel oplock backends to post the break message
37  */
38 void break_kernel_oplock(struct messaging_context *msg_ctx, files_struct *fsp)
39 {
40         uint8_t msg[MSG_SMB_KERNEL_BREAK_SIZE];
41
42         /* Put the kernel break info into the message. */
43         push_file_id_16((char *)msg, &fsp->file_id);
44         SIVAL(msg,16,fsp->fh->gen_id);
45
46         /* Don't need to be root here as we're only ever
47            sending to ourselves. */
48
49         messaging_send_buf(msg_ctx, procid_self(),
50                            MSG_SMB_KERNEL_BREAK,
51                            msg, MSG_SMB_KERNEL_BREAK_SIZE);
52 }
53
54 /****************************************************************************
55  Attempt to set an oplock on a file. Always succeeds if kernel oplocks are
56  disabled (just sets flags). Returns True if oplock set.
57 ****************************************************************************/
58
59 bool set_file_oplock(files_struct *fsp, int oplock_type)
60 {
61         if ((fsp->oplock_type != NO_OPLOCK) &&
62             (fsp->oplock_type != FAKE_LEVEL_II_OPLOCK) &&
63             koplocks &&
64             !koplocks->ops->set_oplock(koplocks, fsp, oplock_type)) {
65                 return False;
66         }
67
68         fsp->oplock_type = oplock_type;
69         fsp->sent_oplock_break = NO_BREAK_SENT;
70         if (oplock_type == LEVEL_II_OPLOCK) {
71                 level_II_oplocks_open++;
72         } else if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
73                 exclusive_oplocks_open++;
74         }
75
76         DEBUG(5,("set_file_oplock: granted oplock on file %s, %s/%lu, "
77                     "tv_sec = %x, tv_usec = %x\n",
78                  fsp->fsp_name, file_id_string_tos(&fsp->file_id),
79                  fsp->fh->gen_id, (int)fsp->open_time.tv_sec,
80                  (int)fsp->open_time.tv_usec ));
81
82         return True;
83 }
84
85 /****************************************************************************
86  Attempt to release an oplock on a file. Decrements oplock count.
87 ****************************************************************************/
88
89 void release_file_oplock(files_struct *fsp)
90 {
91         if ((fsp->oplock_type != NO_OPLOCK) &&
92             (fsp->oplock_type != FAKE_LEVEL_II_OPLOCK) &&
93             koplocks) {
94                 koplocks->ops->release_oplock(koplocks, fsp, NO_OPLOCK);
95         }
96
97         if (fsp->oplock_type == LEVEL_II_OPLOCK) {
98                 level_II_oplocks_open--;
99         } else if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
100                 exclusive_oplocks_open--;
101         }
102
103         SMB_ASSERT(exclusive_oplocks_open>=0);
104         SMB_ASSERT(level_II_oplocks_open>=0);
105
106         if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
107                 /* This doesn't matter for close. */
108                 fsp->oplock_type = FAKE_LEVEL_II_OPLOCK;
109         } else {
110                 fsp->oplock_type = NO_OPLOCK;
111         }
112         fsp->sent_oplock_break = NO_BREAK_SENT;
113
114         flush_write_cache(fsp, OPLOCK_RELEASE_FLUSH);
115
116         TALLOC_FREE(fsp->oplock_timeout);
117 }
118
119 /****************************************************************************
120  Attempt to downgrade an oplock on a file. Doesn't decrement oplock count.
121 ****************************************************************************/
122
123 static void downgrade_file_oplock(files_struct *fsp)
124 {
125         if (!EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
126                 DEBUG(0, ("trying to downgrade an already-downgraded oplock!\n"));
127                 return;
128         }
129
130         if (koplocks) {
131                 koplocks->ops->release_oplock(koplocks, fsp, LEVEL_II_OPLOCK);
132         }
133         fsp->oplock_type = LEVEL_II_OPLOCK;
134         exclusive_oplocks_open--;
135         level_II_oplocks_open++;
136         fsp->sent_oplock_break = NO_BREAK_SENT;
137 }
138
139 /****************************************************************************
140  Remove a file oplock. Copes with level II and exclusive.
141  Locks then unlocks the share mode lock. Client can decide to go directly
142  to none even if a "break-to-level II" was sent.
143 ****************************************************************************/
144
145 bool remove_oplock(files_struct *fsp)
146 {
147         bool ret;
148         struct share_mode_lock *lck;
149
150         /* Remove the oplock flag from the sharemode. */
151         lck = get_share_mode_lock(talloc_tos(), fsp->file_id, NULL, NULL,
152                                   NULL);
153         if (lck == NULL) {
154                 DEBUG(0,("remove_oplock: failed to lock share entry for "
155                          "file %s\n", fsp->fsp_name ));
156                 return False;
157         }
158         ret = remove_share_oplock(lck, fsp);
159         if (!ret) {
160                 DEBUG(0,("remove_oplock: failed to remove share oplock for "
161                          "file %s fnum %d, %s\n",
162                          fsp->fsp_name, fsp->fnum, file_id_string_tos(&fsp->file_id)));
163         }
164         release_file_oplock(fsp);
165         TALLOC_FREE(lck);
166         return ret;
167 }
168
169 /*
170  * Deal with a reply when a break-to-level II was sent.
171  */
172 bool downgrade_oplock(files_struct *fsp)
173 {
174         bool ret;
175         struct share_mode_lock *lck;
176
177         lck = get_share_mode_lock(talloc_tos(), fsp->file_id, NULL, NULL,
178                                   NULL);
179         if (lck == NULL) {
180                 DEBUG(0,("downgrade_oplock: failed to lock share entry for "
181                          "file %s\n", fsp->fsp_name ));
182                 return False;
183         }
184         ret = downgrade_share_oplock(lck, fsp);
185         if (!ret) {
186                 DEBUG(0,("downgrade_oplock: failed to downgrade share oplock "
187                          "for file %s fnum %d, file_id %s\n",
188                          fsp->fsp_name, fsp->fnum, file_id_string_tos(&fsp->file_id)));
189         }
190
191         downgrade_file_oplock(fsp);
192         TALLOC_FREE(lck);
193         return ret;
194 }
195
196 /*
197  * Some kernel oplock implementations handle the notification themselves.
198  */
199 bool should_notify_deferred_opens()
200 {
201         return !(koplocks &&
202                 (koplocks->flags & KOPLOCKS_DEFERRED_OPEN_NOTIFICATION));
203 }
204
205 /****************************************************************************
206  Set up an oplock break message.
207 ****************************************************************************/
208
209 static char *new_break_smb_message(TALLOC_CTX *mem_ctx,
210                                    files_struct *fsp, uint8 cmd)
211 {
212         char *result = TALLOC_ARRAY(mem_ctx, char, smb_size + 8*2 + 0);
213
214         if (result == NULL) {
215                 DEBUG(0, ("talloc failed\n"));
216                 return NULL;
217         }
218
219         memset(result,'\0',smb_size);
220         srv_set_message(result,8,0,true);
221         SCVAL(result,smb_com,SMBlockingX);
222         SSVAL(result,smb_tid,fsp->conn->cnum);
223         SSVAL(result,smb_pid,0xFFFF);
224         SSVAL(result,smb_uid,0);
225         SSVAL(result,smb_mid,0xFFFF);
226         SCVAL(result,smb_vwv0,0xFF);
227         SSVAL(result,smb_vwv2,fsp->fnum);
228         SCVAL(result,smb_vwv3,LOCKING_ANDX_OPLOCK_RELEASE);
229         SCVAL(result,smb_vwv3+1,cmd);
230         return result;
231 }
232
233 /****************************************************************************
234  Function to do the waiting before sending a local break.
235 ****************************************************************************/
236
237 static void wait_before_sending_break(void)
238 {
239         long wait_time = (long)lp_oplock_break_wait_time();
240
241         if (wait_time) {
242                 smb_msleep(wait_time);
243         }
244 }
245
246 /****************************************************************************
247  Ensure that we have a valid oplock.
248 ****************************************************************************/
249
250 static files_struct *initial_break_processing(struct file_id id, unsigned long file_id)
251 {
252         files_struct *fsp = NULL;
253
254         if( DEBUGLVL( 3 ) ) {
255                 dbgtext( "initial_break_processing: called for %s/%u\n",
256                          file_id_string_tos(&id), (int)file_id);
257                 dbgtext( "Current oplocks_open (exclusive = %d, levelII = %d)\n",
258                         exclusive_oplocks_open, level_II_oplocks_open );
259         }
260
261         /*
262          * We need to search the file open table for the
263          * entry containing this dev and inode, and ensure
264          * we have an oplock on it.
265          */
266
267         fsp = file_find_dif(id, file_id);
268
269         if(fsp == NULL) {
270                 /* The file could have been closed in the meantime - return success. */
271                 if( DEBUGLVL( 3 ) ) {
272                         dbgtext( "initial_break_processing: cannot find open file with " );
273                         dbgtext( "file_id %s gen_id = %lu", file_id_string_tos(&id), file_id);
274                         dbgtext( "allowing break to succeed.\n" );
275                 }
276                 return NULL;
277         }
278
279         /* Ensure we have an oplock on the file */
280
281         /*
282          * There is a potential race condition in that an oplock could
283          * have been broken due to another udp request, and yet there are
284          * still oplock break messages being sent in the udp message
285          * queue for this file. So return true if we don't have an oplock,
286          * as we may have just freed it.
287          */
288
289         if(fsp->oplock_type == NO_OPLOCK) {
290                 if( DEBUGLVL( 3 ) ) {
291                         dbgtext( "initial_break_processing: file %s ", fsp->fsp_name );
292                         dbgtext( "(file_id = %s gen_id = %lu) has no oplock.\n",
293                                  file_id_string_tos(&id), fsp->fh->gen_id );
294                         dbgtext( "Allowing break to succeed regardless.\n" );
295                 }
296                 return NULL;
297         }
298
299         return fsp;
300 }
301
302 static void oplock_timeout_handler(struct event_context *ctx,
303                                    struct timed_event *te,
304                                    struct timeval now,
305                                    void *private_data)
306 {
307         files_struct *fsp = (files_struct *)private_data;
308
309         /* Remove the timed event handler. */
310         TALLOC_FREE(fsp->oplock_timeout);
311         DEBUG(0, ("Oplock break failed for file %s -- replying anyway\n", fsp->fsp_name));
312         global_client_failed_oplock_break = True;
313         remove_oplock(fsp);
314         reply_to_oplock_break_requests(fsp);
315 }
316
317 /*******************************************************************
318  Add a timeout handler waiting for the client reply.
319 *******************************************************************/
320
321 static void add_oplock_timeout_handler(files_struct *fsp)
322 {
323         /*
324          * If kernel oplocks already notifies smbds when an oplock break times
325          * out, just return.
326          */
327         if (koplocks &&
328             (koplocks->flags & KOPLOCKS_TIMEOUT_NOTIFICATION)) {
329                 return;
330         }
331
332         if (fsp->oplock_timeout != NULL) {
333                 DEBUG(0, ("Logic problem -- have an oplock event hanging "
334                           "around\n"));
335         }
336
337         fsp->oplock_timeout =
338                 event_add_timed(smbd_event_context(), NULL,
339                                 timeval_current_ofs(OPLOCK_BREAK_TIMEOUT, 0),
340                                 oplock_timeout_handler, fsp);
341
342         if (fsp->oplock_timeout == NULL) {
343                 DEBUG(0, ("Could not add oplock timeout handler\n"));
344         }
345 }
346
347 /*******************************************************************
348  This handles the case of a write triggering a break to none
349  message on a level2 oplock.
350  When we get this message we may be in any of three states :
351  NO_OPLOCK, LEVEL_II, FAKE_LEVEL2. We only send a message to
352  the client for LEVEL2.
353 *******************************************************************/
354
355 void process_oplock_async_level2_break_message(struct messaging_context *msg_ctx,
356                                                       void *private_data,
357                                                       uint32_t msg_type,
358                                                       struct server_id src,
359                                                       DATA_BLOB *data)
360 {
361         struct share_mode_entry msg;
362         files_struct *fsp;
363         char *break_msg;
364         bool sign_state;
365
366         if (data->data == NULL) {
367                 DEBUG(0, ("Got NULL buffer\n"));
368                 return;
369         }
370
371         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
372                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
373                 return;
374         }
375
376         /* De-linearize incoming message. */
377         message_to_share_mode_entry(&msg, (char *)data->data);
378
379         DEBUG(10, ("Got oplock async level 2 break message from pid %d: %s/%lu\n",
380                    (int)procid_to_pid(&src), file_id_string_tos(&msg.id), msg.share_file_id));
381
382         fsp = initial_break_processing(msg.id, msg.share_file_id);
383
384         if (fsp == NULL) {
385                 /* We hit a race here. Break messages are sent, and before we
386                  * get to process this message, we have closed the file. 
387                  * No need to reply as this is an async message. */
388                 DEBUG(3, ("process_oplock_async_level2_break_message: Did not find fsp, ignoring\n"));
389                 return;
390         }
391
392         if (fsp->oplock_type == NO_OPLOCK) {
393                 /* We already got a "break to none" message and we've handled it.
394                  * just ignore. */
395                 DEBUG(3, ("process_oplock_async_level2_break_message: already broken to none, ignoring.\n"));
396                 return;
397         }
398
399         if (fsp->oplock_type == FAKE_LEVEL_II_OPLOCK) {
400                 /* Don't tell the client, just downgrade. */
401                 DEBUG(3, ("process_oplock_async_level2_break_message: downgrading fake level 2 oplock.\n"));
402                 remove_oplock(fsp);
403                 return;
404         }
405
406         /* Ensure we're really at level2 state. */
407         SMB_ASSERT(fsp->oplock_type == LEVEL_II_OPLOCK);
408
409         DEBUG(10,("process_oplock_async_level2_break_message: sending break to "
410                 "none message for fid %d, file %s\n",
411                 fsp->fnum,
412                 fsp->fsp_name));
413
414         /* Now send a break to none message to our client. */
415
416         break_msg = new_break_smb_message(NULL, fsp, OPLOCKLEVEL_NONE);
417         if (break_msg == NULL) {
418                 exit_server("Could not talloc break_msg\n");
419         }
420
421         /* Need to wait before sending a break message if we sent ourselves this message. */
422         if (procid_to_pid(&src) == sys_getpid()) {
423                 wait_before_sending_break();
424         }
425
426         /* Save the server smb signing state. */
427         sign_state = srv_oplock_set_signing(False);
428
429         show_msg(break_msg);
430         if (!srv_send_smb(smbd_server_fd(),
431                         break_msg,
432                         IS_CONN_ENCRYPTED(fsp->conn),
433                         NULL)) {
434                 exit_server_cleanly("oplock_break: srv_send_smb failed.");
435         }
436
437         /* Restore the sign state to what it was. */
438         srv_oplock_set_signing(sign_state);
439
440         TALLOC_FREE(break_msg);
441
442         /* Async level2 request, don't send a reply, just remove the oplock. */
443         remove_oplock(fsp);
444 }
445
446 /*******************************************************************
447  This handles the generic oplock break message from another smbd.
448 *******************************************************************/
449
450 static void process_oplock_break_message(struct messaging_context *msg_ctx,
451                                          void *private_data,
452                                          uint32_t msg_type,
453                                          struct server_id src,
454                                          DATA_BLOB *data)
455 {
456         struct share_mode_entry msg;
457         files_struct *fsp;
458         char *break_msg;
459         bool break_to_level2 = False;
460         bool sign_state;
461
462         if (data->data == NULL) {
463                 DEBUG(0, ("Got NULL buffer\n"));
464                 return;
465         }
466
467         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
468                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
469                 return;
470         }
471
472         /* De-linearize incoming message. */
473         message_to_share_mode_entry(&msg, (char *)data->data);
474
475         DEBUG(10, ("Got oplock break message from pid %d: %s/%lu\n",
476                    (int)procid_to_pid(&src), file_id_string_tos(&msg.id), msg.share_file_id));
477
478         fsp = initial_break_processing(msg.id, msg.share_file_id);
479
480         if (fsp == NULL) {
481                 /* a We hit race here. Break messages are sent, and before we
482                  * get to process this message, we have closed the file. Reply
483                  * with 'ok, oplock broken' */
484                 DEBUG(3, ("Did not find fsp\n"));
485
486                 /* We just send the same message back. */
487                 messaging_send_buf(msg_ctx, src, MSG_SMB_BREAK_RESPONSE,
488                                    (uint8 *)data->data,
489                                    MSG_SMB_SHARE_MODE_ENTRY_SIZE);
490                 return;
491         }
492
493         if (fsp->sent_oplock_break != NO_BREAK_SENT) {
494                 /* Remember we have to inform the requesting PID when the
495                  * client replies */
496                 msg.pid = src;
497                 ADD_TO_ARRAY(NULL, struct share_mode_entry, msg,
498                              &fsp->pending_break_messages,
499                              &fsp->num_pending_break_messages);
500                 return;
501         }
502
503         if (EXCLUSIVE_OPLOCK_TYPE(msg.op_type) &&
504             !EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
505                 DEBUG(3, ("Already downgraded oplock on %s: %s\n",
506                           file_id_string_tos(&fsp->file_id),
507                           fsp->fsp_name));
508                 /* We just send the same message back. */
509                 messaging_send_buf(msg_ctx, src, MSG_SMB_BREAK_RESPONSE,
510                                    (uint8 *)data->data,
511                                    MSG_SMB_SHARE_MODE_ENTRY_SIZE);
512                 return;
513         }
514
515         if ((global_client_caps & CAP_LEVEL_II_OPLOCKS) && 
516             !(msg.op_type & FORCE_OPLOCK_BREAK_TO_NONE) &&
517             !(koplocks && !(koplocks->flags & KOPLOCKS_LEVEL2_SUPPORTED)) &&
518             lp_level2_oplocks(SNUM(fsp->conn))) {
519                 break_to_level2 = True;
520         }
521
522         break_msg = new_break_smb_message(NULL, fsp, break_to_level2 ?
523                                           OPLOCKLEVEL_II : OPLOCKLEVEL_NONE);
524         if (break_msg == NULL) {
525                 exit_server("Could not talloc break_msg\n");
526         }
527
528         /* Need to wait before sending a break message if we sent ourselves this message. */
529         if (procid_to_pid(&src) == sys_getpid()) {
530                 wait_before_sending_break();
531         }
532
533         /* Save the server smb signing state. */
534         sign_state = srv_oplock_set_signing(False);
535
536         show_msg(break_msg);
537         if (!srv_send_smb(smbd_server_fd(),
538                         break_msg,
539                         IS_CONN_ENCRYPTED(fsp->conn),
540                         NULL)) {
541                 exit_server_cleanly("oplock_break: srv_send_smb failed.");
542         }
543
544         /* Restore the sign state to what it was. */
545         srv_oplock_set_signing(sign_state);
546
547         TALLOC_FREE(break_msg);
548
549         fsp->sent_oplock_break = break_to_level2 ? LEVEL_II_BREAK_SENT:BREAK_TO_NONE_SENT;
550
551         msg.pid = src;
552         ADD_TO_ARRAY(NULL, struct share_mode_entry, msg,
553                      &fsp->pending_break_messages,
554                      &fsp->num_pending_break_messages);
555
556         add_oplock_timeout_handler(fsp);
557 }
558
559 /*******************************************************************
560  This handles the kernel oplock break message.
561 *******************************************************************/
562
563 static void process_kernel_oplock_break(struct messaging_context *msg_ctx,
564                                         void *private_data,
565                                         uint32_t msg_type,
566                                         struct server_id src,
567                                         DATA_BLOB *data)
568 {
569         struct file_id id;
570         unsigned long file_id;
571         files_struct *fsp;
572         char *break_msg;
573         bool sign_state;
574
575         if (data->data == NULL) {
576                 DEBUG(0, ("Got NULL buffer\n"));
577                 return;
578         }
579
580         if (data->length != MSG_SMB_KERNEL_BREAK_SIZE) {
581                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
582                 return;
583         }
584
585         /* Pull the data from the message. */
586         pull_file_id_16((char *)data->data, &id);
587         file_id = (unsigned long)IVAL(data->data, 16);
588
589         DEBUG(10, ("Got kernel oplock break message from pid %d: %s/%u\n",
590                    (int)procid_to_pid(&src), file_id_string_tos(&id),
591                    (unsigned int)file_id));
592
593         fsp = initial_break_processing(id, file_id);
594
595         if (fsp == NULL) {
596                 DEBUG(3, ("Got a kernel oplock break message for a file "
597                           "I don't know about\n"));
598                 return;
599         }
600
601         if (fsp->sent_oplock_break != NO_BREAK_SENT) {
602                 /* This is ok, kernel oplocks come in completely async */
603                 DEBUG(3, ("Got a kernel oplock request while waiting for a "
604                           "break reply\n"));
605                 return;
606         }
607
608         break_msg = new_break_smb_message(NULL, fsp, OPLOCKLEVEL_NONE);
609         if (break_msg == NULL) {
610                 exit_server("Could not talloc break_msg\n");
611         }
612
613         /* Save the server smb signing state. */
614         sign_state = srv_oplock_set_signing(False);
615
616         show_msg(break_msg);
617         if (!srv_send_smb(smbd_server_fd(),
618                         break_msg,
619                         IS_CONN_ENCRYPTED(fsp->conn),
620                         NULL)) {
621                 exit_server_cleanly("oplock_break: srv_send_smb failed.");
622         }
623
624         /* Restore the sign state to what it was. */
625         srv_oplock_set_signing(sign_state);
626
627         TALLOC_FREE(break_msg);
628
629         fsp->sent_oplock_break = BREAK_TO_NONE_SENT;
630
631         add_oplock_timeout_handler(fsp);
632 }
633
634 void reply_to_oplock_break_requests(files_struct *fsp)
635 {
636         int i;
637
638         /*
639          * If kernel oplocks already notifies smbds when oplocks are
640          * broken/removed, just return.
641          */
642         if (koplocks &&
643             (koplocks->flags & KOPLOCKS_OPLOCK_BROKEN_NOTIFICATION)) {
644                 return;
645         }
646
647         for (i=0; i<fsp->num_pending_break_messages; i++) {
648                 struct share_mode_entry *e = &fsp->pending_break_messages[i];
649                 char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
650
651                 share_mode_entry_to_message(msg, e);
652
653                 messaging_send_buf(smbd_messaging_context(), e->pid,
654                                    MSG_SMB_BREAK_RESPONSE,
655                                    (uint8 *)msg,
656                                    MSG_SMB_SHARE_MODE_ENTRY_SIZE);
657         }
658
659         SAFE_FREE(fsp->pending_break_messages);
660         fsp->num_pending_break_messages = 0;
661         if (fsp->oplock_timeout != NULL) {
662                 /* Remove the timed event handler. */
663                 TALLOC_FREE(fsp->oplock_timeout);
664                 fsp->oplock_timeout = NULL;
665         }
666         return;
667 }
668
669 static void process_oplock_break_response(struct messaging_context *msg_ctx,
670                                           void *private_data,
671                                           uint32_t msg_type,
672                                           struct server_id src,
673                                           DATA_BLOB *data)
674 {
675         struct share_mode_entry msg;
676
677         if (data->data == NULL) {
678                 DEBUG(0, ("Got NULL buffer\n"));
679                 return;
680         }
681
682         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
683                 DEBUG(0, ("Got invalid msg len %u\n",
684                           (unsigned int)data->length));
685                 return;
686         }
687
688         /* De-linearize incoming message. */
689         message_to_share_mode_entry(&msg, (char *)data->data);
690
691         DEBUG(10, ("Got oplock break response from pid %d: %s/%lu mid %u\n",
692                    (int)procid_to_pid(&src), file_id_string_tos(&msg.id), msg.share_file_id,
693                    (unsigned int)msg.op_mid));
694
695         /* Here's the hack from open.c, store the mid in the 'port' field */
696         schedule_deferred_open_smb_message(msg.op_mid);
697 }
698
699 static void process_open_retry_message(struct messaging_context *msg_ctx,
700                                        void *private_data,
701                                        uint32_t msg_type,
702                                        struct server_id src,
703                                        DATA_BLOB *data)
704 {
705         struct share_mode_entry msg;
706         
707         if (data->data == NULL) {
708                 DEBUG(0, ("Got NULL buffer\n"));
709                 return;
710         }
711
712         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
713                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
714                 return;
715         }
716
717         /* De-linearize incoming message. */
718         message_to_share_mode_entry(&msg, (char *)data->data);
719
720         DEBUG(10, ("Got open retry msg from pid %d: %s mid %u\n",
721                    (int)procid_to_pid(&src), file_id_string_tos(&msg.id),
722                    (unsigned int)msg.op_mid));
723
724         schedule_deferred_open_smb_message(msg.op_mid);
725 }
726
727 /****************************************************************************
728  This function is called on any file modification or lock request. If a file
729  is level 2 oplocked then it must tell all other level 2 holders to break to
730  none.
731 ****************************************************************************/
732
733 static void contend_level2_oplocks_begin_default(files_struct *fsp,
734                                               enum level2_contention_type type)
735 {
736         int i;
737         struct share_mode_lock *lck;
738
739         /*
740          * If this file is level II oplocked then we need
741          * to grab the shared memory lock and inform all
742          * other files with a level II lock that they need
743          * to flush their read caches. We keep the lock over
744          * the shared memory area whilst doing this.
745          */
746
747         if (!LEVEL_II_OPLOCK_TYPE(fsp->oplock_type))
748                 return;
749
750         lck = get_share_mode_lock(talloc_tos(), fsp->file_id, NULL, NULL,
751                                   NULL);
752         if (lck == NULL) {
753                 DEBUG(0,("release_level_2_oplocks_on_change: failed to lock "
754                          "share mode entry for file %s.\n", fsp->fsp_name ));
755                 return;
756         }
757
758         DEBUG(10,("release_level_2_oplocks_on_change: num_share_modes = %d\n", 
759                   lck->num_share_modes ));
760
761         for(i = 0; i < lck->num_share_modes; i++) {
762                 struct share_mode_entry *share_entry = &lck->share_modes[i];
763                 char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
764
765                 if (!is_valid_share_mode_entry(share_entry)) {
766                         continue;
767                 }
768
769                 /*
770                  * As there could have been multiple writes waiting at the
771                  * lock_share_entry gate we may not be the first to
772                  * enter. Hence the state of the op_types in the share mode
773                  * entries may be partly NO_OPLOCK and partly LEVEL_II or FAKE_LEVEL_II
774                  * oplock. It will do no harm to re-send break messages to
775                  * those smbd's that are still waiting their turn to remove
776                  * their LEVEL_II state, and also no harm to ignore existing
777                  * NO_OPLOCK states. JRA.
778                  */
779
780                 DEBUG(10,("release_level_2_oplocks_on_change: "
781                           "share_entry[%i]->op_type == %d\n",
782                           i, share_entry->op_type ));
783
784                 if (share_entry->op_type == NO_OPLOCK) {
785                         continue;
786                 }
787
788                 /* Paranoia .... */
789                 if (EXCLUSIVE_OPLOCK_TYPE(share_entry->op_type)) {
790                         DEBUG(0,("release_level_2_oplocks_on_change: PANIC. "
791                                  "share mode entry %d is an exlusive "
792                                  "oplock !\n", i ));
793                         TALLOC_FREE(lck);
794                         abort();
795                 }
796
797                 share_mode_entry_to_message(msg, share_entry);
798
799                 /*
800                  * Deal with a race condition when breaking level2
801                  * oplocks. Don't send all the messages and release
802                  * the lock, this allows someone else to come in and
803                  * get a level2 lock before any of the messages are
804                  * processed, and thus miss getting a break message.
805                  * Ensure at least one entry (the one we're breaking)
806                  * is processed immediately under the lock and becomes
807                  * set as NO_OPLOCK to stop any waiter getting a level2.
808                  * Bugid #5980.
809                  */
810
811                 if (procid_is_me(&share_entry->pid)) {
812                         DATA_BLOB blob = data_blob_const(msg,
813                                         MSG_SMB_SHARE_MODE_ENTRY_SIZE);
814                         process_oplock_async_level2_break_message(smbd_messaging_context(),
815                                                 NULL,
816                                                 MSG_SMB_ASYNC_LEVEL2_BREAK,
817                                                 share_entry->pid,
818                                                 &blob);
819                 } else {
820                         messaging_send_buf(smbd_messaging_context(),
821                                         share_entry->pid,
822                                         MSG_SMB_ASYNC_LEVEL2_BREAK,
823                                         (uint8 *)msg,
824                                         MSG_SMB_SHARE_MODE_ENTRY_SIZE);
825                 }
826         }
827
828         /* We let the message receivers handle removing the oplock state
829            in the share mode lock db. */
830
831         TALLOC_FREE(lck);
832 }
833
834 void contend_level2_oplocks_begin(files_struct *fsp,
835                                   enum level2_contention_type type)
836 {
837         if (koplocks && koplocks->ops->contend_level2_oplocks_begin) {
838                 koplocks->ops->contend_level2_oplocks_begin(fsp, type);
839                 return;
840         }
841
842         contend_level2_oplocks_begin_default(fsp, type);
843 }
844
845 void contend_level2_oplocks_end(files_struct *fsp,
846                                 enum level2_contention_type type)
847 {
848         /* Only kernel oplocks implement this so far */
849         if (koplocks && koplocks->ops->contend_level2_oplocks_end) {
850                 koplocks->ops->contend_level2_oplocks_end(fsp, type);
851         }
852 }
853
854 /****************************************************************************
855  Linearize a share mode entry struct to an internal oplock break message.
856 ****************************************************************************/
857
858 void share_mode_entry_to_message(char *msg, const struct share_mode_entry *e)
859 {
860         SIVAL(msg,0,(uint32)e->pid.pid);
861         SSVAL(msg,4,e->op_mid);
862         SSVAL(msg,6,e->op_type);
863         SIVAL(msg,8,e->access_mask);
864         SIVAL(msg,12,e->share_access);
865         SIVAL(msg,16,e->private_options);
866         SIVAL(msg,20,(uint32)e->time.tv_sec);
867         SIVAL(msg,24,(uint32)e->time.tv_usec);
868         push_file_id_16(msg+28, &e->id);
869         SIVAL(msg,44,e->share_file_id);
870         SIVAL(msg,48,e->uid);
871         SSVAL(msg,52,e->flags);
872 #ifdef CLUSTER_SUPPORT
873         SIVAL(msg,54,e->pid.vnn);
874 #endif
875 }
876
877 /****************************************************************************
878  De-linearize an internal oplock break message to a share mode entry struct.
879 ****************************************************************************/
880
881 void message_to_share_mode_entry(struct share_mode_entry *e, char *msg)
882 {
883         e->pid.pid = (pid_t)IVAL(msg,0);
884         e->op_mid = SVAL(msg,4);
885         e->op_type = SVAL(msg,6);
886         e->access_mask = IVAL(msg,8);
887         e->share_access = IVAL(msg,12);
888         e->private_options = IVAL(msg,16);
889         e->time.tv_sec = (time_t)IVAL(msg,20);
890         e->time.tv_usec = (int)IVAL(msg,24);
891         pull_file_id_16(msg+28, &e->id);
892         e->share_file_id = (unsigned long)IVAL(msg,44);
893         e->uid = (uint32)IVAL(msg,48);
894         e->flags = (uint16)SVAL(msg,52);
895 #ifdef CLUSTER_SUPPORT
896         e->pid.vnn = IVAL(msg,54);
897 #endif
898 }
899
900 /****************************************************************************
901  Setup oplocks for this process.
902 ****************************************************************************/
903
904 bool init_oplocks(struct messaging_context *msg_ctx)
905 {
906         DEBUG(3,("init_oplocks: initializing messages.\n"));
907
908         messaging_register(msg_ctx, NULL, MSG_SMB_BREAK_REQUEST,
909                            process_oplock_break_message);
910         messaging_register(msg_ctx, NULL, MSG_SMB_ASYNC_LEVEL2_BREAK,
911                            process_oplock_async_level2_break_message);
912         messaging_register(msg_ctx, NULL, MSG_SMB_BREAK_RESPONSE,
913                            process_oplock_break_response);
914         messaging_register(msg_ctx, NULL, MSG_SMB_KERNEL_BREAK,
915                            process_kernel_oplock_break);
916         messaging_register(msg_ctx, NULL, MSG_SMB_OPEN_RETRY,
917                            process_open_retry_message);
918
919         if (lp_kernel_oplocks()) {
920 #if HAVE_KERNEL_OPLOCKS_IRIX
921                 koplocks = irix_init_kernel_oplocks(talloc_autofree_context());
922 #elif HAVE_KERNEL_OPLOCKS_LINUX
923                 koplocks = linux_init_kernel_oplocks(talloc_autofree_context());
924 #elif HAVE_ONEFS
925                 koplocks = onefs_init_kernel_oplocks(talloc_autofree_context());
926 #endif
927         }
928
929         return True;
930 }