s3-talloc Change TALLOC_ARRAY() to talloc_array()
[kai/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/smbd.h"
25 #include "smbd/globals.h"
26 #include "messages.h"
27
28 /****************************************************************************
29  Get the number of current exclusive oplocks.
30 ****************************************************************************/
31
32 int32 get_number_of_exclusive_open_oplocks(void)
33 {
34   return exclusive_oplocks_open;
35 }
36
37 /*
38  * helper function used by the kernel oplock backends to post the break message
39  */
40 void break_kernel_oplock(struct messaging_context *msg_ctx, files_struct *fsp)
41 {
42         uint8_t msg[MSG_SMB_KERNEL_BREAK_SIZE];
43
44         /* Put the kernel break info into the message. */
45         push_file_id_24((char *)msg, &fsp->file_id);
46         SIVAL(msg,24,fsp->fh->gen_id);
47
48         /* Don't need to be root here as we're only ever
49            sending to ourselves. */
50
51         messaging_send_buf(msg_ctx, messaging_server_id(msg_ctx),
52                            MSG_SMB_KERNEL_BREAK,
53                            msg, MSG_SMB_KERNEL_BREAK_SIZE);
54 }
55
56 /****************************************************************************
57  Attempt to set an oplock on a file. Succeeds if kernel oplocks are
58  disabled (just sets flags) and no byte-range locks in the file. Returns True
59  if oplock set.
60 ****************************************************************************/
61
62 bool set_file_oplock(files_struct *fsp, int oplock_type)
63 {
64         if (fsp->oplock_type == LEVEL_II_OPLOCK) {
65                 if (koplocks &&
66                     !(koplocks->flags & KOPLOCKS_LEVEL2_SUPPORTED)) {
67                         DEBUG(10, ("Refusing level2 oplock, kernel oplocks "
68                                    "don't support them\n"));
69                         return false;
70                 }
71         }
72
73         if ((fsp->oplock_type != NO_OPLOCK) &&
74             (fsp->oplock_type != FAKE_LEVEL_II_OPLOCK) &&
75             koplocks &&
76             !koplocks->ops->set_oplock(koplocks, fsp, oplock_type)) {
77                 return False;
78         }
79
80         fsp->oplock_type = oplock_type;
81         fsp->sent_oplock_break = NO_BREAK_SENT;
82         if (oplock_type == LEVEL_II_OPLOCK) {
83                 level_II_oplocks_open++;
84         } else if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
85                 exclusive_oplocks_open++;
86         }
87
88         DEBUG(5,("set_file_oplock: granted oplock on file %s, %s/%lu, "
89                     "tv_sec = %x, tv_usec = %x\n",
90                  fsp_str_dbg(fsp), file_id_string_tos(&fsp->file_id),
91                  fsp->fh->gen_id, (int)fsp->open_time.tv_sec,
92                  (int)fsp->open_time.tv_usec ));
93
94         return True;
95 }
96
97 /****************************************************************************
98  Attempt to release an oplock on a file. Decrements oplock count.
99 ****************************************************************************/
100
101 void release_file_oplock(files_struct *fsp)
102 {
103         if ((fsp->oplock_type != NO_OPLOCK) &&
104             (fsp->oplock_type != FAKE_LEVEL_II_OPLOCK) &&
105             koplocks) {
106                 koplocks->ops->release_oplock(koplocks, fsp, NO_OPLOCK);
107         }
108
109         if (fsp->oplock_type == LEVEL_II_OPLOCK) {
110                 level_II_oplocks_open--;
111         } else if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
112                 exclusive_oplocks_open--;
113         }
114
115         SMB_ASSERT(exclusive_oplocks_open>=0);
116         SMB_ASSERT(level_II_oplocks_open>=0);
117
118         if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
119                 /* This doesn't matter for close. */
120                 fsp->oplock_type = FAKE_LEVEL_II_OPLOCK;
121         } else {
122                 fsp->oplock_type = NO_OPLOCK;
123         }
124         fsp->sent_oplock_break = NO_BREAK_SENT;
125
126         flush_write_cache(fsp, OPLOCK_RELEASE_FLUSH);
127         delete_write_cache(fsp);
128
129         TALLOC_FREE(fsp->oplock_timeout);
130 }
131
132 /****************************************************************************
133  Attempt to downgrade an oplock on a file. Doesn't decrement oplock count.
134 ****************************************************************************/
135
136 static void downgrade_file_oplock(files_struct *fsp)
137 {
138         if (!EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
139                 DEBUG(0, ("trying to downgrade an already-downgraded oplock!\n"));
140                 return;
141         }
142
143         if (koplocks) {
144                 koplocks->ops->release_oplock(koplocks, fsp, LEVEL_II_OPLOCK);
145         }
146         fsp->oplock_type = LEVEL_II_OPLOCK;
147         exclusive_oplocks_open--;
148         level_II_oplocks_open++;
149         fsp->sent_oplock_break = NO_BREAK_SENT;
150 }
151
152 /****************************************************************************
153  Remove a file oplock. Copes with level II and exclusive.
154  Locks then unlocks the share mode lock. Client can decide to go directly
155  to none even if a "break-to-level II" was sent.
156 ****************************************************************************/
157
158 bool remove_oplock(files_struct *fsp)
159 {
160         bool ret;
161         struct share_mode_lock *lck;
162
163         /* Remove the oplock flag from the sharemode. */
164         lck = get_share_mode_lock(talloc_tos(), fsp->file_id, NULL, NULL,
165                                   NULL);
166         if (lck == NULL) {
167                 DEBUG(0,("remove_oplock: failed to lock share entry for "
168                          "file %s\n", fsp_str_dbg(fsp)));
169                 return False;
170         }
171         ret = remove_share_oplock(lck, fsp);
172         if (!ret) {
173                 DEBUG(0,("remove_oplock: failed to remove share oplock for "
174                          "file %s fnum %d, %s\n",
175                          fsp_str_dbg(fsp), fsp->fnum,
176                          file_id_string_tos(&fsp->file_id)));
177         }
178         release_file_oplock(fsp);
179         TALLOC_FREE(lck);
180         return ret;
181 }
182
183 /*
184  * Deal with a reply when a break-to-level II was sent.
185  */
186 bool downgrade_oplock(files_struct *fsp)
187 {
188         bool ret;
189         struct share_mode_lock *lck;
190
191         lck = get_share_mode_lock(talloc_tos(), fsp->file_id, NULL, NULL,
192                                   NULL);
193         if (lck == NULL) {
194                 DEBUG(0,("downgrade_oplock: failed to lock share entry for "
195                          "file %s\n", fsp_str_dbg(fsp)));
196                 return False;
197         }
198         ret = downgrade_share_oplock(lck, fsp);
199         if (!ret) {
200                 DEBUG(0,("downgrade_oplock: failed to downgrade share oplock "
201                          "for file %s fnum %d, file_id %s\n",
202                          fsp_str_dbg(fsp), fsp->fnum,
203                          file_id_string_tos(&fsp->file_id)));
204         }
205
206         downgrade_file_oplock(fsp);
207         TALLOC_FREE(lck);
208         return ret;
209 }
210
211 /*
212  * Some kernel oplock implementations handle the notification themselves.
213  */
214 bool should_notify_deferred_opens()
215 {
216         return !(koplocks &&
217                 (koplocks->flags & KOPLOCKS_DEFERRED_OPEN_NOTIFICATION));
218 }
219
220 /****************************************************************************
221  Set up an oplock break message.
222 ****************************************************************************/
223
224 static char *new_break_message_smb1(TALLOC_CTX *mem_ctx,
225                                    files_struct *fsp, int cmd)
226 {
227         char *result = talloc_array(mem_ctx, char, smb_size + 8*2 + 0);
228
229         if (result == NULL) {
230                 DEBUG(0, ("talloc failed\n"));
231                 return NULL;
232         }
233
234         memset(result,'\0',smb_size);
235         srv_set_message(result,8,0,true);
236         SCVAL(result,smb_com,SMBlockingX);
237         SSVAL(result,smb_tid,fsp->conn->cnum);
238         SSVAL(result,smb_pid,0xFFFF);
239         SSVAL(result,smb_uid,0);
240         SSVAL(result,smb_mid,0xFFFF);
241         SCVAL(result,smb_vwv0,0xFF);
242         SSVAL(result,smb_vwv2,fsp->fnum);
243         SCVAL(result,smb_vwv3,LOCKING_ANDX_OPLOCK_RELEASE);
244         SCVAL(result,smb_vwv3+1,cmd);
245         return result;
246 }
247
248 /****************************************************************************
249  Function to do the waiting before sending a local break.
250 ****************************************************************************/
251
252 static void wait_before_sending_break(void)
253 {
254         long wait_time = (long)lp_oplock_break_wait_time();
255
256         if (wait_time) {
257                 smb_msleep(wait_time);
258         }
259 }
260
261 /****************************************************************************
262  Ensure that we have a valid oplock.
263 ****************************************************************************/
264
265 static files_struct *initial_break_processing(
266         struct smbd_server_connection *sconn, struct file_id id,
267         unsigned long file_id)
268 {
269         files_struct *fsp = NULL;
270
271         if( DEBUGLVL( 3 ) ) {
272                 dbgtext( "initial_break_processing: called for %s/%u\n",
273                          file_id_string_tos(&id), (int)file_id);
274                 dbgtext( "Current oplocks_open (exclusive = %d, levelII = %d)\n",
275                         exclusive_oplocks_open, level_II_oplocks_open );
276         }
277
278         /*
279          * We need to search the file open table for the
280          * entry containing this dev and inode, and ensure
281          * we have an oplock on it.
282          */
283
284         fsp = file_find_dif(sconn, id, file_id);
285
286         if(fsp == NULL) {
287                 /* The file could have been closed in the meantime - return success. */
288                 if( DEBUGLVL( 3 ) ) {
289                         dbgtext( "initial_break_processing: cannot find open file with " );
290                         dbgtext( "file_id %s gen_id = %lu", file_id_string_tos(&id), file_id);
291                         dbgtext( "allowing break to succeed.\n" );
292                 }
293                 return NULL;
294         }
295
296         /* Ensure we have an oplock on the file */
297
298         /*
299          * There is a potential race condition in that an oplock could
300          * have been broken due to another udp request, and yet there are
301          * still oplock break messages being sent in the udp message
302          * queue for this file. So return true if we don't have an oplock,
303          * as we may have just freed it.
304          */
305
306         if(fsp->oplock_type == NO_OPLOCK) {
307                 if( DEBUGLVL( 3 ) ) {
308                         dbgtext( "initial_break_processing: file %s ",
309                                  fsp_str_dbg(fsp));
310                         dbgtext( "(file_id = %s gen_id = %lu) has no oplock.\n",
311                                  file_id_string_tos(&id), fsp->fh->gen_id );
312                         dbgtext( "Allowing break to succeed regardless.\n" );
313                 }
314                 return NULL;
315         }
316
317         return fsp;
318 }
319
320 static void oplock_timeout_handler(struct event_context *ctx,
321                                    struct timed_event *te,
322                                    struct timeval now,
323                                    void *private_data)
324 {
325         files_struct *fsp = (files_struct *)private_data;
326
327         /* Remove the timed event handler. */
328         TALLOC_FREE(fsp->oplock_timeout);
329         DEBUG(0, ("Oplock break failed for file %s -- replying anyway\n",
330                   fsp_str_dbg(fsp)));
331         remove_oplock(fsp);
332         reply_to_oplock_break_requests(fsp);
333 }
334
335 /*******************************************************************
336  Add a timeout handler waiting for the client reply.
337 *******************************************************************/
338
339 static void add_oplock_timeout_handler(files_struct *fsp)
340 {
341         /*
342          * If kernel oplocks already notifies smbds when an oplock break times
343          * out, just return.
344          */
345         if (koplocks &&
346             (koplocks->flags & KOPLOCKS_TIMEOUT_NOTIFICATION)) {
347                 return;
348         }
349
350         if (fsp->oplock_timeout != NULL) {
351                 DEBUG(0, ("Logic problem -- have an oplock event hanging "
352                           "around\n"));
353         }
354
355         fsp->oplock_timeout =
356                 event_add_timed(server_event_context(), fsp,
357                                 timeval_current_ofs(OPLOCK_BREAK_TIMEOUT, 0),
358                                 oplock_timeout_handler, fsp);
359
360         if (fsp->oplock_timeout == NULL) {
361                 DEBUG(0, ("Could not add oplock timeout handler\n"));
362         }
363 }
364
365 static void send_break_message_smb1(files_struct *fsp, int level)
366 {
367         char *break_msg = new_break_message_smb1(talloc_tos(),
368                                         fsp,
369                                         level);
370         if (break_msg == NULL) {
371                 exit_server("Could not talloc break_msg\n");
372         }
373
374         show_msg(break_msg);
375         if (!srv_send_smb(fsp->conn->sconn,
376                         break_msg, false, 0,
377                         IS_CONN_ENCRYPTED(fsp->conn),
378                         NULL)) {
379                 exit_server_cleanly("send_break_message_smb1: "
380                         "srv_send_smb failed.");
381         }
382
383         TALLOC_FREE(break_msg);
384 }
385
386 void break_level2_to_none_async(files_struct *fsp)
387 {
388         struct smbd_server_connection *sconn = fsp->conn->sconn;
389
390         if (fsp->oplock_type == NO_OPLOCK) {
391                 /* We already got a "break to none" message and we've handled
392                  * it.  just ignore. */
393                 DEBUG(3, ("process_oplock_async_level2_break_message: already "
394                           "broken to none, ignoring.\n"));
395                 return;
396         }
397
398         if (fsp->oplock_type == FAKE_LEVEL_II_OPLOCK) {
399                 /* Don't tell the client, just downgrade. */
400                 DEBUG(3, ("process_oplock_async_level2_break_message: "
401                           "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 "
410                   "to none message for fid %d, file %s\n", fsp->fnum,
411                   fsp_str_dbg(fsp)));
412
413         /* Now send a break to none message to our client. */
414         if (sconn->using_smb2) {
415                 send_break_message_smb2(fsp, OPLOCKLEVEL_NONE);
416         } else {
417                 send_break_message_smb1(fsp, OPLOCKLEVEL_NONE);
418         }
419
420         /* Async level2 request, don't send a reply, just remove the oplock. */
421         remove_oplock(fsp);
422 }
423
424 /*******************************************************************
425  This handles the case of a write triggering a break to none
426  message on a level2 oplock.
427  When we get this message we may be in any of three states :
428  NO_OPLOCK, LEVEL_II, FAKE_LEVEL2. We only send a message to
429  the client for LEVEL2.
430 *******************************************************************/
431
432 void process_oplock_async_level2_break_message(struct messaging_context *msg_ctx,
433                                                       void *private_data,
434                                                       uint32_t msg_type,
435                                                       struct server_id src,
436                                                       DATA_BLOB *data)
437 {
438         struct smbd_server_connection *sconn;
439         struct share_mode_entry msg;
440         files_struct *fsp;
441
442         if (data->data == NULL) {
443                 DEBUG(0, ("Got NULL buffer\n"));
444                 return;
445         }
446
447         sconn = msg_ctx_to_sconn(msg_ctx);
448         if (sconn == NULL) {
449                 DEBUG(1, ("could not find sconn\n"));
450                 return;
451         }
452
453         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
454                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
455                 return;
456         }
457
458         /* De-linearize incoming message. */
459         message_to_share_mode_entry(&msg, (char *)data->data);
460
461         DEBUG(10, ("Got oplock async level 2 break message from pid %s: "
462                    "%s/%lu\n", procid_str(talloc_tos(), &src),
463                    file_id_string_tos(&msg.id), msg.share_file_id));
464
465         fsp = initial_break_processing(sconn, msg.id, msg.share_file_id);
466
467         if (fsp == NULL) {
468                 /* We hit a race here. Break messages are sent, and before we
469                  * get to process this message, we have closed the file. 
470                  * No need to reply as this is an async message. */
471                 DEBUG(3, ("process_oplock_async_level2_break_message: Did not find fsp, ignoring\n"));
472                 return;
473         }
474
475         break_level2_to_none_async(fsp);
476 }
477
478 /*******************************************************************
479  This handles the generic oplock break message from another smbd.
480 *******************************************************************/
481
482 static void process_oplock_break_message(struct messaging_context *msg_ctx,
483                                          void *private_data,
484                                          uint32_t msg_type,
485                                          struct server_id src,
486                                          DATA_BLOB *data)
487 {
488         struct smbd_server_connection *sconn;
489         struct share_mode_entry msg;
490         files_struct *fsp;
491         bool break_to_level2 = False;
492
493         if (data->data == NULL) {
494                 DEBUG(0, ("Got NULL buffer\n"));
495                 return;
496         }
497
498         sconn = msg_ctx_to_sconn(msg_ctx);
499         if (sconn == NULL) {
500                 DEBUG(1, ("could not find sconn\n"));
501                 return;
502         }
503
504         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
505                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
506                 return;
507         }
508
509         /* De-linearize incoming message. */
510         message_to_share_mode_entry(&msg, (char *)data->data);
511
512         DEBUG(10, ("Got oplock break message from pid %s: %s/%lu\n",
513                    procid_str(talloc_tos(), &src), file_id_string_tos(&msg.id),
514                    msg.share_file_id));
515
516         fsp = initial_break_processing(sconn, msg.id, msg.share_file_id);
517
518         if (fsp == NULL) {
519                 /* We hit a race here. Break messages are sent, and before we
520                  * get to process this message, we have closed the file. Reply
521                  * with 'ok, oplock broken' */
522                 DEBUG(3, ("Did not find fsp\n"));
523
524                 /* We just send the same message back. */
525                 messaging_send_buf(msg_ctx, src, MSG_SMB_BREAK_RESPONSE,
526                                    (uint8 *)data->data,
527                                    MSG_SMB_SHARE_MODE_ENTRY_SIZE);
528                 return;
529         }
530
531         if (fsp->sent_oplock_break != NO_BREAK_SENT) {
532                 /* Remember we have to inform the requesting PID when the
533                  * client replies */
534                 msg.pid = src;
535                 ADD_TO_ARRAY(NULL, struct share_mode_entry, msg,
536                              &fsp->pending_break_messages,
537                              &fsp->num_pending_break_messages);
538                 return;
539         }
540
541         if (EXCLUSIVE_OPLOCK_TYPE(msg.op_type) &&
542             !EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
543                 DEBUG(3, ("Already downgraded oplock on %s: %s\n",
544                           file_id_string_tos(&fsp->file_id),
545                           fsp_str_dbg(fsp)));
546                 /* We just send the same message back. */
547                 messaging_send_buf(msg_ctx, src, MSG_SMB_BREAK_RESPONSE,
548                                    (uint8 *)data->data,
549                                    MSG_SMB_SHARE_MODE_ENTRY_SIZE);
550                 return;
551         }
552
553         if ((global_client_caps & CAP_LEVEL_II_OPLOCKS) && 
554             !(msg.op_type & FORCE_OPLOCK_BREAK_TO_NONE) &&
555             !(koplocks && !(koplocks->flags & KOPLOCKS_LEVEL2_SUPPORTED)) &&
556             lp_level2_oplocks(SNUM(fsp->conn))) {
557                 break_to_level2 = True;
558         }
559
560         /* Need to wait before sending a break
561            message if we sent ourselves this message. */
562         if (procid_is_me(&src)) {
563                 wait_before_sending_break();
564         }
565
566         if (sconn->using_smb2) {
567                 send_break_message_smb2(fsp, break_to_level2 ?
568                         OPLOCKLEVEL_II : OPLOCKLEVEL_NONE);
569         } else {
570                 send_break_message_smb1(fsp, break_to_level2 ?
571                         OPLOCKLEVEL_II : OPLOCKLEVEL_NONE);
572         }
573
574         fsp->sent_oplock_break = break_to_level2 ? LEVEL_II_BREAK_SENT:BREAK_TO_NONE_SENT;
575
576         msg.pid = src;
577         ADD_TO_ARRAY(NULL, struct share_mode_entry, msg,
578                      &fsp->pending_break_messages,
579                      &fsp->num_pending_break_messages);
580
581         add_oplock_timeout_handler(fsp);
582 }
583
584 /*******************************************************************
585  This handles the kernel oplock break message.
586 *******************************************************************/
587
588 static void process_kernel_oplock_break(struct messaging_context *msg_ctx,
589                                         void *private_data,
590                                         uint32_t msg_type,
591                                         struct server_id src,
592                                         DATA_BLOB *data)
593 {
594         struct smbd_server_connection *sconn;
595         struct file_id id;
596         unsigned long file_id;
597         files_struct *fsp;
598
599         if (data->data == NULL) {
600                 DEBUG(0, ("Got NULL buffer\n"));
601                 return;
602         }
603
604         if (data->length != MSG_SMB_KERNEL_BREAK_SIZE) {
605                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
606                 return;
607         }
608
609         sconn = msg_ctx_to_sconn(msg_ctx);
610         if (sconn == NULL) {
611                 DEBUG(1, ("could not find sconn\n"));
612                 return;
613         }
614
615         /* Pull the data from the message. */
616         pull_file_id_24((char *)data->data, &id);
617         file_id = (unsigned long)IVAL(data->data, 24);
618
619         DEBUG(10, ("Got kernel oplock break message from pid %s: %s/%u\n",
620                    procid_str(talloc_tos(), &src), file_id_string_tos(&id),
621                    (unsigned int)file_id));
622
623         fsp = initial_break_processing(sconn, id, file_id);
624
625         if (fsp == NULL) {
626                 DEBUG(3, ("Got a kernel oplock break message for a file "
627                           "I don't know about\n"));
628                 return;
629         }
630
631         if (fsp->sent_oplock_break != NO_BREAK_SENT) {
632                 /* This is ok, kernel oplocks come in completely async */
633                 DEBUG(3, ("Got a kernel oplock request while waiting for a "
634                           "break reply\n"));
635                 return;
636         }
637
638         if (sconn->using_smb2) {
639                 send_break_message_smb2(fsp, OPLOCKLEVEL_NONE);
640         } else {
641                 send_break_message_smb1(fsp, OPLOCKLEVEL_NONE);
642         }
643
644         fsp->sent_oplock_break = BREAK_TO_NONE_SENT;
645
646         add_oplock_timeout_handler(fsp);
647 }
648
649 void reply_to_oplock_break_requests(files_struct *fsp)
650 {
651         int i;
652
653         /*
654          * If kernel oplocks already notifies smbds when oplocks are
655          * broken/removed, just return.
656          */
657         if (koplocks &&
658             (koplocks->flags & KOPLOCKS_OPLOCK_BROKEN_NOTIFICATION)) {
659                 return;
660         }
661
662         for (i=0; i<fsp->num_pending_break_messages; i++) {
663                 struct share_mode_entry *e = &fsp->pending_break_messages[i];
664                 char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
665
666                 share_mode_entry_to_message(msg, e);
667
668                 messaging_send_buf(fsp->conn->sconn->msg_ctx, e->pid,
669                                    MSG_SMB_BREAK_RESPONSE,
670                                    (uint8 *)msg,
671                                    MSG_SMB_SHARE_MODE_ENTRY_SIZE);
672         }
673
674         SAFE_FREE(fsp->pending_break_messages);
675         fsp->num_pending_break_messages = 0;
676         if (fsp->oplock_timeout != NULL) {
677                 /* Remove the timed event handler. */
678                 TALLOC_FREE(fsp->oplock_timeout);
679                 fsp->oplock_timeout = NULL;
680         }
681         return;
682 }
683
684 static void process_oplock_break_response(struct messaging_context *msg_ctx,
685                                           void *private_data,
686                                           uint32_t msg_type,
687                                           struct server_id src,
688                                           DATA_BLOB *data)
689 {
690         struct share_mode_entry msg;
691
692         if (data->data == NULL) {
693                 DEBUG(0, ("Got NULL buffer\n"));
694                 return;
695         }
696
697         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
698                 DEBUG(0, ("Got invalid msg len %u\n",
699                           (unsigned int)data->length));
700                 return;
701         }
702
703         /* De-linearize incoming message. */
704         message_to_share_mode_entry(&msg, (char *)data->data);
705
706         DEBUG(10, ("Got oplock break response from pid %s: %s/%lu mid %llu\n",
707                    procid_str(talloc_tos(), &src), file_id_string_tos(&msg.id),
708                    msg.share_file_id, (unsigned long long)msg.op_mid));
709
710         schedule_deferred_open_message_smb(msg.op_mid);
711 }
712
713 static void process_open_retry_message(struct messaging_context *msg_ctx,
714                                        void *private_data,
715                                        uint32_t msg_type,
716                                        struct server_id src,
717                                        DATA_BLOB *data)
718 {
719         struct share_mode_entry msg;
720         
721         if (data->data == NULL) {
722                 DEBUG(0, ("Got NULL buffer\n"));
723                 return;
724         }
725
726         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
727                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
728                 return;
729         }
730
731         /* De-linearize incoming message. */
732         message_to_share_mode_entry(&msg, (char *)data->data);
733
734         DEBUG(10, ("Got open retry msg from pid %s: %s mid %llu\n",
735                    procid_str(talloc_tos(), &src), file_id_string_tos(&msg.id),
736                    (unsigned long long)msg.op_mid));
737
738         schedule_deferred_open_message_smb(msg.op_mid);
739 }
740
741 /****************************************************************************
742  This function is called on any file modification or lock request. If a file
743  is level 2 oplocked then it must tell all other level 2 holders to break to
744  none.
745 ****************************************************************************/
746
747 static void contend_level2_oplocks_begin_default(files_struct *fsp,
748                                               enum level2_contention_type type)
749 {
750         int i;
751         struct share_mode_lock *lck;
752
753         /*
754          * If this file is level II oplocked then we need
755          * to grab the shared memory lock and inform all
756          * other files with a level II lock that they need
757          * to flush their read caches. We keep the lock over
758          * the shared memory area whilst doing this.
759          */
760
761         if (!LEVEL_II_OPLOCK_TYPE(fsp->oplock_type))
762                 return;
763
764         lck = get_share_mode_lock(talloc_tos(), fsp->file_id, NULL, NULL,
765                                   NULL);
766         if (lck == NULL) {
767                 DEBUG(0,("release_level_2_oplocks_on_change: failed to lock "
768                          "share mode entry for file %s.\n", fsp_str_dbg(fsp)));
769                 return;
770         }
771
772         DEBUG(10,("release_level_2_oplocks_on_change: num_share_modes = %d\n", 
773                   lck->num_share_modes ));
774
775         for(i = 0; i < lck->num_share_modes; i++) {
776                 struct share_mode_entry *share_entry = &lck->share_modes[i];
777                 char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
778
779                 if (!is_valid_share_mode_entry(share_entry)) {
780                         continue;
781                 }
782
783                 /*
784                  * As there could have been multiple writes waiting at the
785                  * lock_share_entry gate we may not be the first to
786                  * enter. Hence the state of the op_types in the share mode
787                  * entries may be partly NO_OPLOCK and partly LEVEL_II or FAKE_LEVEL_II
788                  * oplock. It will do no harm to re-send break messages to
789                  * those smbd's that are still waiting their turn to remove
790                  * their LEVEL_II state, and also no harm to ignore existing
791                  * NO_OPLOCK states. JRA.
792                  */
793
794                 DEBUG(10,("release_level_2_oplocks_on_change: "
795                           "share_entry[%i]->op_type == %d\n",
796                           i, share_entry->op_type ));
797
798                 if (share_entry->op_type == NO_OPLOCK) {
799                         continue;
800                 }
801
802                 /* Paranoia .... */
803                 if (EXCLUSIVE_OPLOCK_TYPE(share_entry->op_type)) {
804                         DEBUG(0,("release_level_2_oplocks_on_change: PANIC. "
805                                  "share mode entry %d is an exlusive "
806                                  "oplock !\n", i ));
807                         TALLOC_FREE(lck);
808                         abort();
809                 }
810
811                 share_mode_entry_to_message(msg, share_entry);
812
813                 /*
814                  * Deal with a race condition when breaking level2
815                  * oplocks. Don't send all the messages and release
816                  * the lock, this allows someone else to come in and
817                  * get a level2 lock before any of the messages are
818                  * processed, and thus miss getting a break message.
819                  * Ensure at least one entry (the one we're breaking)
820                  * is processed immediately under the lock and becomes
821                  * set as NO_OPLOCK to stop any waiter getting a level2.
822                  * Bugid #5980.
823                  */
824
825                 if (procid_is_me(&share_entry->pid)) {
826                         struct files_struct *cur_fsp =
827                                 initial_break_processing(fsp->conn->sconn,
828                                         share_entry->id,
829                                         share_entry->share_file_id);
830                         wait_before_sending_break();
831                         if (cur_fsp != NULL) {
832                                 break_level2_to_none_async(cur_fsp);
833                         } else {
834                                 DEBUG(3, ("release_level_2_oplocks_on_change: "
835                                 "Did not find fsp, ignoring\n"));
836                         }
837                 } else {
838                         messaging_send_buf(fsp->conn->sconn->msg_ctx,
839                                         share_entry->pid,
840                                         MSG_SMB_ASYNC_LEVEL2_BREAK,
841                                         (uint8 *)msg,
842                                         MSG_SMB_SHARE_MODE_ENTRY_SIZE);
843                 }
844         }
845
846         /* We let the message receivers handle removing the oplock state
847            in the share mode lock db. */
848
849         TALLOC_FREE(lck);
850 }
851
852 void contend_level2_oplocks_begin(files_struct *fsp,
853                                   enum level2_contention_type type)
854 {
855         if (koplocks && koplocks->ops->contend_level2_oplocks_begin) {
856                 koplocks->ops->contend_level2_oplocks_begin(fsp, type);
857                 return;
858         }
859
860         contend_level2_oplocks_begin_default(fsp, type);
861 }
862
863 void contend_level2_oplocks_end(files_struct *fsp,
864                                 enum level2_contention_type type)
865 {
866         /* Only kernel oplocks implement this so far */
867         if (koplocks && koplocks->ops->contend_level2_oplocks_end) {
868                 koplocks->ops->contend_level2_oplocks_end(fsp, type);
869         }
870 }
871
872 /****************************************************************************
873  Linearize a share mode entry struct to an internal oplock break message.
874 ****************************************************************************/
875
876 void share_mode_entry_to_message(char *msg, const struct share_mode_entry *e)
877 {
878         SIVAL(msg,OP_BREAK_MSG_PID_OFFSET,(uint32)e->pid.pid);
879         SBVAL(msg,OP_BREAK_MSG_MID_OFFSET,e->op_mid);
880         SSVAL(msg,OP_BREAK_MSG_OP_TYPE_OFFSET,e->op_type);
881         SIVAL(msg,OP_BREAK_MSG_ACCESS_MASK_OFFSET,e->access_mask);
882         SIVAL(msg,OP_BREAK_MSG_SHARE_ACCESS_OFFSET,e->share_access);
883         SIVAL(msg,OP_BREAK_MSG_PRIV_OFFSET,e->private_options);
884         SIVAL(msg,OP_BREAK_MSG_TIME_SEC_OFFSET,(uint32_t)e->time.tv_sec);
885         SIVAL(msg,OP_BREAK_MSG_TIME_USEC_OFFSET,(uint32_t)e->time.tv_usec);
886         push_file_id_24(msg+OP_BREAK_MSG_DEV_OFFSET, &e->id);
887         SIVAL(msg,OP_BREAK_MSG_FILE_ID_OFFSET,e->share_file_id);
888         SIVAL(msg,OP_BREAK_MSG_UID_OFFSET,e->uid);
889         SSVAL(msg,OP_BREAK_MSG_FLAGS_OFFSET,e->flags);
890         SIVAL(msg,OP_BREAK_MSG_NAME_HASH_OFFSET,e->name_hash);
891         SIVAL(msg,OP_BREAK_MSG_VNN_OFFSET,e->pid.vnn);
892 }
893
894 /****************************************************************************
895  De-linearize an internal oplock break message to a share mode entry struct.
896 ****************************************************************************/
897
898 void message_to_share_mode_entry(struct share_mode_entry *e, char *msg)
899 {
900         e->pid.pid = (pid_t)IVAL(msg,OP_BREAK_MSG_PID_OFFSET);
901         e->op_mid = BVAL(msg,OP_BREAK_MSG_MID_OFFSET);
902         e->op_type = SVAL(msg,OP_BREAK_MSG_OP_TYPE_OFFSET);
903         e->access_mask = IVAL(msg,OP_BREAK_MSG_ACCESS_MASK_OFFSET);
904         e->share_access = IVAL(msg,OP_BREAK_MSG_SHARE_ACCESS_OFFSET);
905         e->private_options = IVAL(msg,OP_BREAK_MSG_PRIV_OFFSET);
906         e->time.tv_sec = (time_t)IVAL(msg,OP_BREAK_MSG_TIME_SEC_OFFSET);
907         e->time.tv_usec = (int)IVAL(msg,OP_BREAK_MSG_TIME_USEC_OFFSET);
908         pull_file_id_24(msg+OP_BREAK_MSG_DEV_OFFSET, &e->id);
909         e->share_file_id = (unsigned long)IVAL(msg,OP_BREAK_MSG_FILE_ID_OFFSET);
910         e->uid = (uint32)IVAL(msg,OP_BREAK_MSG_UID_OFFSET);
911         e->flags = (uint16)SVAL(msg,OP_BREAK_MSG_FLAGS_OFFSET);
912         e->name_hash = IVAL(msg,OP_BREAK_MSG_NAME_HASH_OFFSET);
913         e->pid.vnn = IVAL(msg,OP_BREAK_MSG_VNN_OFFSET);
914 }
915
916 /****************************************************************************
917  Setup oplocks for this process.
918 ****************************************************************************/
919
920 bool init_oplocks(struct messaging_context *msg_ctx)
921 {
922         DEBUG(3,("init_oplocks: initializing messages.\n"));
923
924         messaging_register(msg_ctx, NULL, MSG_SMB_BREAK_REQUEST,
925                            process_oplock_break_message);
926         messaging_register(msg_ctx, NULL, MSG_SMB_ASYNC_LEVEL2_BREAK,
927                            process_oplock_async_level2_break_message);
928         messaging_register(msg_ctx, NULL, MSG_SMB_BREAK_RESPONSE,
929                            process_oplock_break_response);
930         messaging_register(msg_ctx, NULL, MSG_SMB_KERNEL_BREAK,
931                            process_kernel_oplock_break);
932         messaging_register(msg_ctx, NULL, MSG_SMB_OPEN_RETRY,
933                            process_open_retry_message);
934
935         if (lp_kernel_oplocks()) {
936 #if HAVE_KERNEL_OPLOCKS_IRIX
937                 koplocks = irix_init_kernel_oplocks(NULL);
938 #elif HAVE_KERNEL_OPLOCKS_LINUX
939                 koplocks = linux_init_kernel_oplocks(NULL);
940 #elif HAVE_ONEFS
941 #error Isilon, please check if the NULL context is okay here. Thanks!
942                 koplocks = onefs_init_kernel_oplocks(NULL);
943 #endif
944         }
945
946         return True;
947 }