7ce450c1fe50ebd1f21217d0cbeaa2c2b806e15a
[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_24((char *)msg, &fsp->file_id);
44         SIVAL(msg,24,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 == LEVEL_II_OPLOCK)
62             && koplocks && !(koplocks->flags & KOPLOCKS_LEVEL2_SUPPORTED)) {
63                 DEBUG(10, ("Refusing level2 oplock, kernel oplocks don't "
64                            "support them\n"));
65                 return false;
66         }
67         if ((fsp->oplock_type != NO_OPLOCK) &&
68             (fsp->oplock_type != FAKE_LEVEL_II_OPLOCK) &&
69             koplocks &&
70             !koplocks->ops->set_oplock(koplocks, fsp, oplock_type)) {
71                 return False;
72         }
73
74         fsp->oplock_type = oplock_type;
75         fsp->sent_oplock_break = NO_BREAK_SENT;
76         if (oplock_type == LEVEL_II_OPLOCK) {
77                 level_II_oplocks_open++;
78         } else if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
79                 exclusive_oplocks_open++;
80         }
81
82         DEBUG(5,("set_file_oplock: granted oplock on file %s, %s/%lu, "
83                     "tv_sec = %x, tv_usec = %x\n",
84                  fsp_str_dbg(fsp), file_id_string_tos(&fsp->file_id),
85                  fsp->fh->gen_id, (int)fsp->open_time.tv_sec,
86                  (int)fsp->open_time.tv_usec ));
87
88         return True;
89 }
90
91 /****************************************************************************
92  Attempt to release an oplock on a file. Decrements oplock count.
93 ****************************************************************************/
94
95 void release_file_oplock(files_struct *fsp)
96 {
97         if ((fsp->oplock_type != NO_OPLOCK) &&
98             (fsp->oplock_type != FAKE_LEVEL_II_OPLOCK) &&
99             koplocks) {
100                 koplocks->ops->release_oplock(koplocks, fsp, NO_OPLOCK);
101         }
102
103         if (fsp->oplock_type == LEVEL_II_OPLOCK) {
104                 level_II_oplocks_open--;
105         } else if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
106                 exclusive_oplocks_open--;
107         }
108
109         SMB_ASSERT(exclusive_oplocks_open>=0);
110         SMB_ASSERT(level_II_oplocks_open>=0);
111
112         if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
113                 /* This doesn't matter for close. */
114                 fsp->oplock_type = FAKE_LEVEL_II_OPLOCK;
115         } else {
116                 fsp->oplock_type = NO_OPLOCK;
117         }
118         fsp->sent_oplock_break = NO_BREAK_SENT;
119
120         flush_write_cache(fsp, OPLOCK_RELEASE_FLUSH);
121
122         TALLOC_FREE(fsp->oplock_timeout);
123 }
124
125 /****************************************************************************
126  Attempt to downgrade an oplock on a file. Doesn't decrement oplock count.
127 ****************************************************************************/
128
129 static void downgrade_file_oplock(files_struct *fsp)
130 {
131         if (!EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
132                 DEBUG(0, ("trying to downgrade an already-downgraded oplock!\n"));
133                 return;
134         }
135
136         if (koplocks) {
137                 koplocks->ops->release_oplock(koplocks, fsp, LEVEL_II_OPLOCK);
138         }
139         fsp->oplock_type = LEVEL_II_OPLOCK;
140         exclusive_oplocks_open--;
141         level_II_oplocks_open++;
142         fsp->sent_oplock_break = NO_BREAK_SENT;
143 }
144
145 /****************************************************************************
146  Remove a file oplock. Copes with level II and exclusive.
147  Locks then unlocks the share mode lock. Client can decide to go directly
148  to none even if a "break-to-level II" was sent.
149 ****************************************************************************/
150
151 bool remove_oplock(files_struct *fsp)
152 {
153         bool ret;
154         struct share_mode_lock *lck;
155
156         /* Remove the oplock flag from the sharemode. */
157         lck = get_share_mode_lock(talloc_tos(), fsp->file_id, NULL, NULL,
158                                   NULL);
159         if (lck == NULL) {
160                 DEBUG(0,("remove_oplock: failed to lock share entry for "
161                          "file %s\n", fsp_str_dbg(fsp)));
162                 return False;
163         }
164         ret = remove_share_oplock(lck, fsp);
165         if (!ret) {
166                 DEBUG(0,("remove_oplock: failed to remove share oplock for "
167                          "file %s fnum %d, %s\n",
168                          fsp_str_dbg(fsp), fsp->fnum,
169                          file_id_string_tos(&fsp->file_id)));
170         }
171         release_file_oplock(fsp);
172         TALLOC_FREE(lck);
173         return ret;
174 }
175
176 /*
177  * Deal with a reply when a break-to-level II was sent.
178  */
179 bool downgrade_oplock(files_struct *fsp)
180 {
181         bool ret;
182         struct share_mode_lock *lck;
183
184         lck = get_share_mode_lock(talloc_tos(), fsp->file_id, NULL, NULL,
185                                   NULL);
186         if (lck == NULL) {
187                 DEBUG(0,("downgrade_oplock: failed to lock share entry for "
188                          "file %s\n", fsp_str_dbg(fsp)));
189                 return False;
190         }
191         ret = downgrade_share_oplock(lck, fsp);
192         if (!ret) {
193                 DEBUG(0,("downgrade_oplock: failed to downgrade share oplock "
194                          "for file %s fnum %d, file_id %s\n",
195                          fsp_str_dbg(fsp), fsp->fnum,
196                          file_id_string_tos(&fsp->file_id)));
197         }
198
199         downgrade_file_oplock(fsp);
200         TALLOC_FREE(lck);
201         return ret;
202 }
203
204 /*
205  * Some kernel oplock implementations handle the notification themselves.
206  */
207 bool should_notify_deferred_opens()
208 {
209         return !(koplocks &&
210                 (koplocks->flags & KOPLOCKS_DEFERRED_OPEN_NOTIFICATION));
211 }
212
213 /****************************************************************************
214  Set up an oplock break message.
215 ****************************************************************************/
216
217 static char *new_break_smb_message(TALLOC_CTX *mem_ctx,
218                                    files_struct *fsp, uint8 cmd)
219 {
220         char *result = TALLOC_ARRAY(mem_ctx, char, smb_size + 8*2 + 0);
221
222         if (result == NULL) {
223                 DEBUG(0, ("talloc failed\n"));
224                 return NULL;
225         }
226
227         memset(result,'\0',smb_size);
228         srv_set_message(result,8,0,true);
229         SCVAL(result,smb_com,SMBlockingX);
230         SSVAL(result,smb_tid,fsp->conn->cnum);
231         SSVAL(result,smb_pid,0xFFFF);
232         SSVAL(result,smb_uid,0);
233         SSVAL(result,smb_mid,0xFFFF);
234         SCVAL(result,smb_vwv0,0xFF);
235         SSVAL(result,smb_vwv2,fsp->fnum);
236         SCVAL(result,smb_vwv3,LOCKING_ANDX_OPLOCK_RELEASE);
237         SCVAL(result,smb_vwv3+1,cmd);
238         return result;
239 }
240
241 /****************************************************************************
242  Function to do the waiting before sending a local break.
243 ****************************************************************************/
244
245 static void wait_before_sending_break(void)
246 {
247         long wait_time = (long)lp_oplock_break_wait_time();
248
249         if (wait_time) {
250                 smb_msleep(wait_time);
251         }
252 }
253
254 /****************************************************************************
255  Ensure that we have a valid oplock.
256 ****************************************************************************/
257
258 static files_struct *initial_break_processing(struct file_id id, unsigned long file_id)
259 {
260         files_struct *fsp = NULL;
261
262         if( DEBUGLVL( 3 ) ) {
263                 dbgtext( "initial_break_processing: called for %s/%u\n",
264                          file_id_string_tos(&id), (int)file_id);
265                 dbgtext( "Current oplocks_open (exclusive = %d, levelII = %d)\n",
266                         exclusive_oplocks_open, level_II_oplocks_open );
267         }
268
269         /*
270          * We need to search the file open table for the
271          * entry containing this dev and inode, and ensure
272          * we have an oplock on it.
273          */
274
275         fsp = file_find_dif(id, file_id);
276
277         if(fsp == NULL) {
278                 /* The file could have been closed in the meantime - return success. */
279                 if( DEBUGLVL( 3 ) ) {
280                         dbgtext( "initial_break_processing: cannot find open file with " );
281                         dbgtext( "file_id %s gen_id = %lu", file_id_string_tos(&id), file_id);
282                         dbgtext( "allowing break to succeed.\n" );
283                 }
284                 return NULL;
285         }
286
287         /* Ensure we have an oplock on the file */
288
289         /*
290          * There is a potential race condition in that an oplock could
291          * have been broken due to another udp request, and yet there are
292          * still oplock break messages being sent in the udp message
293          * queue for this file. So return true if we don't have an oplock,
294          * as we may have just freed it.
295          */
296
297         if(fsp->oplock_type == NO_OPLOCK) {
298                 if( DEBUGLVL( 3 ) ) {
299                         dbgtext( "initial_break_processing: file %s ",
300                                  fsp_str_dbg(fsp));
301                         dbgtext( "(file_id = %s gen_id = %lu) has no oplock.\n",
302                                  file_id_string_tos(&id), fsp->fh->gen_id );
303                         dbgtext( "Allowing break to succeed regardless.\n" );
304                 }
305                 return NULL;
306         }
307
308         return fsp;
309 }
310
311 static void oplock_timeout_handler(struct event_context *ctx,
312                                    struct timed_event *te,
313                                    struct timeval now,
314                                    void *private_data)
315 {
316         files_struct *fsp = (files_struct *)private_data;
317
318         /* Remove the timed event handler. */
319         TALLOC_FREE(fsp->oplock_timeout);
320         DEBUG(0, ("Oplock break failed for file %s -- replying anyway\n",
321                   fsp_str_dbg(fsp)));
322         global_client_failed_oplock_break = True;
323         remove_oplock(fsp);
324         reply_to_oplock_break_requests(fsp);
325 }
326
327 /*******************************************************************
328  Add a timeout handler waiting for the client reply.
329 *******************************************************************/
330
331 static void add_oplock_timeout_handler(files_struct *fsp)
332 {
333         /*
334          * If kernel oplocks already notifies smbds when an oplock break times
335          * out, just return.
336          */
337         if (koplocks &&
338             (koplocks->flags & KOPLOCKS_TIMEOUT_NOTIFICATION)) {
339                 return;
340         }
341
342         if (fsp->oplock_timeout != NULL) {
343                 DEBUG(0, ("Logic problem -- have an oplock event hanging "
344                           "around\n"));
345         }
346
347         fsp->oplock_timeout =
348                 event_add_timed(smbd_event_context(), NULL,
349                                 timeval_current_ofs(OPLOCK_BREAK_TIMEOUT, 0),
350                                 oplock_timeout_handler, fsp);
351
352         if (fsp->oplock_timeout == NULL) {
353                 DEBUG(0, ("Could not add oplock timeout handler\n"));
354         }
355 }
356
357 void break_level2_to_none_async(files_struct *fsp)
358 {
359         char *break_msg;
360
361         if (fsp->oplock_type == NO_OPLOCK) {
362                 /* We already got a "break to none" message and we've handled
363                  * it.  just ignore. */
364                 DEBUG(3, ("process_oplock_async_level2_break_message: already "
365                           "broken to none, ignoring.\n"));
366                 return;
367         }
368
369         if (fsp->oplock_type == FAKE_LEVEL_II_OPLOCK) {
370                 /* Don't tell the client, just downgrade. */
371                 DEBUG(3, ("process_oplock_async_level2_break_message: "
372                           "downgrading fake level 2 oplock.\n"));
373                 remove_oplock(fsp);
374                 return;
375         }
376
377         /* Ensure we're really at level2 state. */
378         SMB_ASSERT(fsp->oplock_type == LEVEL_II_OPLOCK);
379
380         DEBUG(10,("process_oplock_async_level2_break_message: sending break "
381                   "to none message for fid %d, file %s\n", fsp->fnum,
382                   fsp_str_dbg(fsp)));
383
384         /* Now send a break to none message to our client. */
385         break_msg = new_break_smb_message(NULL, fsp, OPLOCKLEVEL_NONE);
386         if (break_msg == NULL) {
387                 exit_server("Could not talloc break_msg\n");
388         }
389
390         show_msg(break_msg);
391         if (!srv_send_smb(smbd_server_fd(),
392                         break_msg, false, 0,
393                         IS_CONN_ENCRYPTED(fsp->conn),
394                         NULL)) {
395                 exit_server_cleanly("oplock_break: srv_send_smb failed.");
396         }
397
398         TALLOC_FREE(break_msg);
399
400         /* Async level2 request, don't send a reply, just remove the oplock. */
401         remove_oplock(fsp);
402
403 }
404
405 /*******************************************************************
406  This handles the case of a write triggering a break to none
407  message on a level2 oplock.
408  When we get this message we may be in any of three states :
409  NO_OPLOCK, LEVEL_II, FAKE_LEVEL2. We only send a message to
410  the client for LEVEL2.
411 *******************************************************************/
412
413 void process_oplock_async_level2_break_message(struct messaging_context *msg_ctx,
414                                                       void *private_data,
415                                                       uint32_t msg_type,
416                                                       struct server_id src,
417                                                       DATA_BLOB *data)
418 {
419         struct share_mode_entry msg;
420         files_struct *fsp;
421
422         if (data->data == NULL) {
423                 DEBUG(0, ("Got NULL buffer\n"));
424                 return;
425         }
426
427         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
428                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
429                 return;
430         }
431
432         /* De-linearize incoming message. */
433         message_to_share_mode_entry(&msg, (char *)data->data);
434
435         DEBUG(10, ("Got oplock async level 2 break message from pid %s: "
436                    "%s/%lu\n", procid_str(talloc_tos(), &src),
437                    file_id_string_tos(&msg.id), msg.share_file_id));
438
439         fsp = initial_break_processing(msg.id, msg.share_file_id);
440
441         if (fsp == NULL) {
442                 /* We hit a race here. Break messages are sent, and before we
443                  * get to process this message, we have closed the file. 
444                  * No need to reply as this is an async message. */
445                 DEBUG(3, ("process_oplock_async_level2_break_message: Did not find fsp, ignoring\n"));
446                 return;
447         }
448
449         break_level2_to_none_async(fsp);
450 }
451
452 /*******************************************************************
453  This handles the generic oplock break message from another smbd.
454 *******************************************************************/
455
456 static void process_oplock_break_message(struct messaging_context *msg_ctx,
457                                          void *private_data,
458                                          uint32_t msg_type,
459                                          struct server_id src,
460                                          DATA_BLOB *data)
461 {
462         struct share_mode_entry msg;
463         files_struct *fsp;
464         char *break_msg;
465         bool break_to_level2 = False;
466
467         if (data->data == NULL) {
468                 DEBUG(0, ("Got NULL buffer\n"));
469                 return;
470         }
471
472         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
473                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
474                 return;
475         }
476
477         /* De-linearize incoming message. */
478         message_to_share_mode_entry(&msg, (char *)data->data);
479
480         DEBUG(10, ("Got oplock break message from pid %s: %s/%lu\n",
481                    procid_str(talloc_tos(), &src), file_id_string_tos(&msg.id),
482                    msg.share_file_id));
483
484         fsp = initial_break_processing(msg.id, msg.share_file_id);
485
486         if (fsp == NULL) {
487                 /* a We hit race here. Break messages are sent, and before we
488                  * get to process this message, we have closed the file. Reply
489                  * with 'ok, oplock broken' */
490                 DEBUG(3, ("Did not find fsp\n"));
491
492                 /* We just send the same message back. */
493                 messaging_send_buf(msg_ctx, src, MSG_SMB_BREAK_RESPONSE,
494                                    (uint8 *)data->data,
495                                    MSG_SMB_SHARE_MODE_ENTRY_SIZE);
496                 return;
497         }
498
499         if (fsp->sent_oplock_break != NO_BREAK_SENT) {
500                 /* Remember we have to inform the requesting PID when the
501                  * client replies */
502                 msg.pid = src;
503                 ADD_TO_ARRAY(NULL, struct share_mode_entry, msg,
504                              &fsp->pending_break_messages,
505                              &fsp->num_pending_break_messages);
506                 return;
507         }
508
509         if (EXCLUSIVE_OPLOCK_TYPE(msg.op_type) &&
510             !EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
511                 DEBUG(3, ("Already downgraded oplock on %s: %s\n",
512                           file_id_string_tos(&fsp->file_id),
513                           fsp_str_dbg(fsp)));
514                 /* We just send the same message back. */
515                 messaging_send_buf(msg_ctx, src, MSG_SMB_BREAK_RESPONSE,
516                                    (uint8 *)data->data,
517                                    MSG_SMB_SHARE_MODE_ENTRY_SIZE);
518                 return;
519         }
520
521         if ((global_client_caps & CAP_LEVEL_II_OPLOCKS) && 
522             !(msg.op_type & FORCE_OPLOCK_BREAK_TO_NONE) &&
523             !(koplocks && !(koplocks->flags & KOPLOCKS_LEVEL2_SUPPORTED)) &&
524             lp_level2_oplocks(SNUM(fsp->conn))) {
525                 break_to_level2 = True;
526         }
527
528         break_msg = new_break_smb_message(NULL, fsp, break_to_level2 ?
529                                           OPLOCKLEVEL_II : OPLOCKLEVEL_NONE);
530         if (break_msg == NULL) {
531                 exit_server("Could not talloc break_msg\n");
532         }
533
534         /* Need to wait before sending a break message if we sent ourselves this message. */
535         if (procid_is_me(&src)) {
536                 wait_before_sending_break();
537         }
538
539         show_msg(break_msg);
540         if (!srv_send_smb(smbd_server_fd(),
541                         break_msg, false, 0,
542                         IS_CONN_ENCRYPTED(fsp->conn),
543                         NULL)) {
544                 exit_server_cleanly("oplock_break: srv_send_smb failed.");
545         }
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
574         if (data->data == NULL) {
575                 DEBUG(0, ("Got NULL buffer\n"));
576                 return;
577         }
578
579         if (data->length != MSG_SMB_KERNEL_BREAK_SIZE) {
580                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
581                 return;
582         }
583
584         /* Pull the data from the message. */
585         pull_file_id_24((char *)data->data, &id);
586         file_id = (unsigned long)IVAL(data->data, 24);
587
588         DEBUG(10, ("Got kernel oplock break message from pid %s: %s/%u\n",
589                    procid_str(talloc_tos(), &src), file_id_string_tos(&id),
590                    (unsigned int)file_id));
591
592         fsp = initial_break_processing(id, file_id);
593
594         if (fsp == NULL) {
595                 DEBUG(3, ("Got a kernel oplock break message for a file "
596                           "I don't know about\n"));
597                 return;
598         }
599
600         if (fsp->sent_oplock_break != NO_BREAK_SENT) {
601                 /* This is ok, kernel oplocks come in completely async */
602                 DEBUG(3, ("Got a kernel oplock request while waiting for a "
603                           "break reply\n"));
604                 return;
605         }
606
607         break_msg = new_break_smb_message(NULL, fsp, OPLOCKLEVEL_NONE);
608         if (break_msg == NULL) {
609                 exit_server("Could not talloc break_msg\n");
610         }
611
612         show_msg(break_msg);
613         if (!srv_send_smb(smbd_server_fd(),
614                         break_msg, false, 0,
615                         IS_CONN_ENCRYPTED(fsp->conn),
616                         NULL)) {
617                 exit_server_cleanly("oplock_break: srv_send_smb failed.");
618         }
619
620         TALLOC_FREE(break_msg);
621
622         fsp->sent_oplock_break = BREAK_TO_NONE_SENT;
623
624         add_oplock_timeout_handler(fsp);
625 }
626
627 void reply_to_oplock_break_requests(files_struct *fsp)
628 {
629         int i;
630
631         /*
632          * If kernel oplocks already notifies smbds when oplocks are
633          * broken/removed, just return.
634          */
635         if (koplocks &&
636             (koplocks->flags & KOPLOCKS_OPLOCK_BROKEN_NOTIFICATION)) {
637                 return;
638         }
639
640         for (i=0; i<fsp->num_pending_break_messages; i++) {
641                 struct share_mode_entry *e = &fsp->pending_break_messages[i];
642                 char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
643
644                 share_mode_entry_to_message(msg, e);
645
646                 messaging_send_buf(smbd_messaging_context(), e->pid,
647                                    MSG_SMB_BREAK_RESPONSE,
648                                    (uint8 *)msg,
649                                    MSG_SMB_SHARE_MODE_ENTRY_SIZE);
650         }
651
652         SAFE_FREE(fsp->pending_break_messages);
653         fsp->num_pending_break_messages = 0;
654         if (fsp->oplock_timeout != NULL) {
655                 /* Remove the timed event handler. */
656                 TALLOC_FREE(fsp->oplock_timeout);
657                 fsp->oplock_timeout = NULL;
658         }
659         return;
660 }
661
662 static void process_oplock_break_response(struct messaging_context *msg_ctx,
663                                           void *private_data,
664                                           uint32_t msg_type,
665                                           struct server_id src,
666                                           DATA_BLOB *data)
667 {
668         struct share_mode_entry msg;
669
670         if (data->data == NULL) {
671                 DEBUG(0, ("Got NULL buffer\n"));
672                 return;
673         }
674
675         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
676                 DEBUG(0, ("Got invalid msg len %u\n",
677                           (unsigned int)data->length));
678                 return;
679         }
680
681         /* De-linearize incoming message. */
682         message_to_share_mode_entry(&msg, (char *)data->data);
683
684         DEBUG(10, ("Got oplock break response from pid %s: %s/%lu mid %u\n",
685                    procid_str(talloc_tos(), &src), file_id_string_tos(&msg.id),
686                    msg.share_file_id, (unsigned int)msg.op_mid));
687
688         schedule_deferred_open_smb_message(msg.op_mid);
689 }
690
691 static void process_open_retry_message(struct messaging_context *msg_ctx,
692                                        void *private_data,
693                                        uint32_t msg_type,
694                                        struct server_id src,
695                                        DATA_BLOB *data)
696 {
697         struct share_mode_entry msg;
698         
699         if (data->data == NULL) {
700                 DEBUG(0, ("Got NULL buffer\n"));
701                 return;
702         }
703
704         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
705                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
706                 return;
707         }
708
709         /* De-linearize incoming message. */
710         message_to_share_mode_entry(&msg, (char *)data->data);
711
712         DEBUG(10, ("Got open retry msg from pid %s: %s mid %u\n",
713                    procid_str(talloc_tos(), &src), file_id_string_tos(&msg.id),
714                    (unsigned int)msg.op_mid));
715
716         schedule_deferred_open_smb_message(msg.op_mid);
717 }
718
719 /****************************************************************************
720  This function is called on any file modification or lock request. If a file
721  is level 2 oplocked then it must tell all other level 2 holders to break to
722  none.
723 ****************************************************************************/
724
725 static void contend_level2_oplocks_begin_default(files_struct *fsp,
726                                               enum level2_contention_type type)
727 {
728         int i;
729         struct share_mode_lock *lck;
730
731         /*
732          * If this file is level II oplocked then we need
733          * to grab the shared memory lock and inform all
734          * other files with a level II lock that they need
735          * to flush their read caches. We keep the lock over
736          * the shared memory area whilst doing this.
737          */
738
739         if (!LEVEL_II_OPLOCK_TYPE(fsp->oplock_type))
740                 return;
741
742         lck = get_share_mode_lock(talloc_tos(), fsp->file_id, NULL, NULL,
743                                   NULL);
744         if (lck == NULL) {
745                 DEBUG(0,("release_level_2_oplocks_on_change: failed to lock "
746                          "share mode entry for file %s.\n", fsp_str_dbg(fsp)));
747                 return;
748         }
749
750         DEBUG(10,("release_level_2_oplocks_on_change: num_share_modes = %d\n", 
751                   lck->num_share_modes ));
752
753         for(i = 0; i < lck->num_share_modes; i++) {
754                 struct share_mode_entry *share_entry = &lck->share_modes[i];
755                 char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
756
757                 if (!is_valid_share_mode_entry(share_entry)) {
758                         continue;
759                 }
760
761                 /*
762                  * As there could have been multiple writes waiting at the
763                  * lock_share_entry gate we may not be the first to
764                  * enter. Hence the state of the op_types in the share mode
765                  * entries may be partly NO_OPLOCK and partly LEVEL_II or FAKE_LEVEL_II
766                  * oplock. It will do no harm to re-send break messages to
767                  * those smbd's that are still waiting their turn to remove
768                  * their LEVEL_II state, and also no harm to ignore existing
769                  * NO_OPLOCK states. JRA.
770                  */
771
772                 DEBUG(10,("release_level_2_oplocks_on_change: "
773                           "share_entry[%i]->op_type == %d\n",
774                           i, share_entry->op_type ));
775
776                 if (share_entry->op_type == NO_OPLOCK) {
777                         continue;
778                 }
779
780                 /* Paranoia .... */
781                 if (EXCLUSIVE_OPLOCK_TYPE(share_entry->op_type)) {
782                         DEBUG(0,("release_level_2_oplocks_on_change: PANIC. "
783                                  "share mode entry %d is an exlusive "
784                                  "oplock !\n", i ));
785                         TALLOC_FREE(lck);
786                         abort();
787                 }
788
789                 share_mode_entry_to_message(msg, share_entry);
790
791                 /*
792                  * Deal with a race condition when breaking level2
793                  * oplocks. Don't send all the messages and release
794                  * the lock, this allows someone else to come in and
795                  * get a level2 lock before any of the messages are
796                  * processed, and thus miss getting a break message.
797                  * Ensure at least one entry (the one we're breaking)
798                  * is processed immediately under the lock and becomes
799                  * set as NO_OPLOCK to stop any waiter getting a level2.
800                  * Bugid #5980.
801                  */
802
803                 if (procid_is_me(&share_entry->pid)) {
804                         wait_before_sending_break();
805                         break_level2_to_none_async(fsp);
806                 } else {
807                         messaging_send_buf(smbd_messaging_context(),
808                                         share_entry->pid,
809                                         MSG_SMB_ASYNC_LEVEL2_BREAK,
810                                         (uint8 *)msg,
811                                         MSG_SMB_SHARE_MODE_ENTRY_SIZE);
812                 }
813         }
814
815         /* We let the message receivers handle removing the oplock state
816            in the share mode lock db. */
817
818         TALLOC_FREE(lck);
819 }
820
821 void contend_level2_oplocks_begin(files_struct *fsp,
822                                   enum level2_contention_type type)
823 {
824         if (koplocks && koplocks->ops->contend_level2_oplocks_begin) {
825                 koplocks->ops->contend_level2_oplocks_begin(fsp, type);
826                 return;
827         }
828
829         contend_level2_oplocks_begin_default(fsp, type);
830 }
831
832 void contend_level2_oplocks_end(files_struct *fsp,
833                                 enum level2_contention_type type)
834 {
835         /* Only kernel oplocks implement this so far */
836         if (koplocks && koplocks->ops->contend_level2_oplocks_end) {
837                 koplocks->ops->contend_level2_oplocks_end(fsp, type);
838         }
839 }
840
841 /****************************************************************************
842  Linearize a share mode entry struct to an internal oplock break message.
843 ****************************************************************************/
844
845 void share_mode_entry_to_message(char *msg, const struct share_mode_entry *e)
846 {
847         SIVAL(msg,0,(uint32)e->pid.pid);
848         SSVAL(msg,4,e->op_mid);
849         SSVAL(msg,6,e->op_type);
850         SIVAL(msg,8,e->access_mask);
851         SIVAL(msg,12,e->share_access);
852         SIVAL(msg,16,e->private_options);
853         SIVAL(msg,20,(uint32)e->time.tv_sec);
854         SIVAL(msg,24,(uint32)e->time.tv_usec);
855         push_file_id_24(msg+28, &e->id);
856         SIVAL(msg,52,e->share_file_id);
857         SIVAL(msg,56,e->uid);
858         SSVAL(msg,60,e->flags);
859 #ifdef CLUSTER_SUPPORT
860         SIVAL(msg,62,e->pid.vnn);
861 #endif
862 }
863
864 /****************************************************************************
865  De-linearize an internal oplock break message to a share mode entry struct.
866 ****************************************************************************/
867
868 void message_to_share_mode_entry(struct share_mode_entry *e, char *msg)
869 {
870         e->pid.pid = (pid_t)IVAL(msg,0);
871         e->op_mid = SVAL(msg,4);
872         e->op_type = SVAL(msg,6);
873         e->access_mask = IVAL(msg,8);
874         e->share_access = IVAL(msg,12);
875         e->private_options = IVAL(msg,16);
876         e->time.tv_sec = (time_t)IVAL(msg,20);
877         e->time.tv_usec = (int)IVAL(msg,24);
878         pull_file_id_24(msg+28, &e->id);
879         e->share_file_id = (unsigned long)IVAL(msg,52);
880         e->uid = (uint32)IVAL(msg,56);
881         e->flags = (uint16)SVAL(msg,60);
882 #ifdef CLUSTER_SUPPORT
883         e->pid.vnn = IVAL(msg,62);
884 #endif
885 }
886
887 /****************************************************************************
888  Setup oplocks for this process.
889 ****************************************************************************/
890
891 bool init_oplocks(struct messaging_context *msg_ctx)
892 {
893         DEBUG(3,("init_oplocks: initializing messages.\n"));
894
895         messaging_register(msg_ctx, NULL, MSG_SMB_BREAK_REQUEST,
896                            process_oplock_break_message);
897         messaging_register(msg_ctx, NULL, MSG_SMB_ASYNC_LEVEL2_BREAK,
898                            process_oplock_async_level2_break_message);
899         messaging_register(msg_ctx, NULL, MSG_SMB_BREAK_RESPONSE,
900                            process_oplock_break_response);
901         messaging_register(msg_ctx, NULL, MSG_SMB_KERNEL_BREAK,
902                            process_kernel_oplock_break);
903         messaging_register(msg_ctx, NULL, MSG_SMB_OPEN_RETRY,
904                            process_open_retry_message);
905
906         if (lp_kernel_oplocks()) {
907 #if HAVE_KERNEL_OPLOCKS_IRIX
908                 koplocks = irix_init_kernel_oplocks(talloc_autofree_context());
909 #elif HAVE_KERNEL_OPLOCKS_LINUX
910                 koplocks = linux_init_kernel_oplocks(talloc_autofree_context());
911 #elif HAVE_ONEFS
912                 koplocks = onefs_init_kernel_oplocks(talloc_autofree_context());
913 #endif
914         }
915
916         return True;
917 }