s3:smbd/oplock: pass sconn as private_data to process_kernel_oplock_break()
[garming/samba-autobuild/.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 #include "../librpc/gen_ndr/open_files.h"
28
29 /****************************************************************************
30  Get the number of current exclusive oplocks.
31 ****************************************************************************/
32
33 int32 get_number_of_exclusive_open_oplocks(void)
34 {
35   return exclusive_oplocks_open;
36 }
37
38 /*
39  * helper function used by the kernel oplock backends to post the break message
40  */
41 void break_kernel_oplock(struct messaging_context *msg_ctx, files_struct *fsp)
42 {
43         uint8_t msg[MSG_SMB_KERNEL_BREAK_SIZE];
44
45         /* Put the kernel break info into the message. */
46         push_file_id_24((char *)msg, &fsp->file_id);
47         SIVAL(msg,24,fsp->fh->gen_id);
48
49         /* Don't need to be root here as we're only ever
50            sending to ourselves. */
51
52         messaging_send_buf(msg_ctx, messaging_server_id(msg_ctx),
53                            MSG_SMB_KERNEL_BREAK,
54                            msg, MSG_SMB_KERNEL_BREAK_SIZE);
55 }
56
57 /****************************************************************************
58  Attempt to set an oplock on a file. Succeeds if kernel oplocks are
59  disabled (just sets flags) and no byte-range locks in the file. Returns True
60  if oplock set.
61 ****************************************************************************/
62
63 bool set_file_oplock(files_struct *fsp, int oplock_type)
64 {
65         if (fsp->oplock_type == LEVEL_II_OPLOCK) {
66                 if (koplocks &&
67                     !(koplocks->flags & KOPLOCKS_LEVEL2_SUPPORTED)) {
68                         DEBUG(10, ("Refusing level2 oplock, kernel oplocks "
69                                    "don't support them\n"));
70                         return false;
71                 }
72         }
73
74         if ((fsp->oplock_type != NO_OPLOCK) &&
75             (fsp->oplock_type != FAKE_LEVEL_II_OPLOCK) &&
76             koplocks &&
77             !koplocks->ops->set_oplock(koplocks, fsp, oplock_type)) {
78                 return False;
79         }
80
81         fsp->oplock_type = oplock_type;
82         fsp->sent_oplock_break = NO_BREAK_SENT;
83         if (oplock_type == LEVEL_II_OPLOCK) {
84                 level_II_oplocks_open++;
85         } else if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
86                 exclusive_oplocks_open++;
87         }
88
89         DEBUG(5,("set_file_oplock: granted oplock on file %s, %s/%lu, "
90                     "tv_sec = %x, tv_usec = %x\n",
91                  fsp_str_dbg(fsp), file_id_string_tos(&fsp->file_id),
92                  fsp->fh->gen_id, (int)fsp->open_time.tv_sec,
93                  (int)fsp->open_time.tv_usec ));
94
95         return True;
96 }
97
98 /****************************************************************************
99  Attempt to release an oplock on a file. Decrements oplock count.
100 ****************************************************************************/
101
102 void release_file_oplock(files_struct *fsp)
103 {
104         if ((fsp->oplock_type != NO_OPLOCK) &&
105             (fsp->oplock_type != FAKE_LEVEL_II_OPLOCK) &&
106             koplocks) {
107                 koplocks->ops->release_oplock(koplocks, fsp, NO_OPLOCK);
108         }
109
110         if (fsp->oplock_type == LEVEL_II_OPLOCK) {
111                 level_II_oplocks_open--;
112         } else if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
113                 exclusive_oplocks_open--;
114         }
115
116         SMB_ASSERT(exclusive_oplocks_open>=0);
117         SMB_ASSERT(level_II_oplocks_open>=0);
118
119         if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
120                 /* This doesn't matter for close. */
121                 fsp->oplock_type = FAKE_LEVEL_II_OPLOCK;
122         } else {
123                 fsp->oplock_type = NO_OPLOCK;
124         }
125         fsp->sent_oplock_break = NO_BREAK_SENT;
126
127         flush_write_cache(fsp, OPLOCK_RELEASE_FLUSH);
128         delete_write_cache(fsp);
129
130         TALLOC_FREE(fsp->oplock_timeout);
131 }
132
133 /****************************************************************************
134  Attempt to downgrade an oplock on a file. Doesn't decrement oplock count.
135 ****************************************************************************/
136
137 static void downgrade_file_oplock(files_struct *fsp)
138 {
139         if (!EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
140                 DEBUG(0, ("trying to downgrade an already-downgraded oplock!\n"));
141                 return;
142         }
143
144         if (koplocks) {
145                 koplocks->ops->release_oplock(koplocks, fsp, LEVEL_II_OPLOCK);
146         }
147         fsp->oplock_type = LEVEL_II_OPLOCK;
148         exclusive_oplocks_open--;
149         level_II_oplocks_open++;
150         fsp->sent_oplock_break = NO_BREAK_SENT;
151 }
152
153 /****************************************************************************
154  Remove a file oplock. Copes with level II and exclusive.
155  Locks then unlocks the share mode lock. Client can decide to go directly
156  to none even if a "break-to-level II" was sent.
157 ****************************************************************************/
158
159 bool remove_oplock(files_struct *fsp)
160 {
161         bool ret;
162         struct share_mode_lock *lck;
163
164         /* Remove the oplock flag from the sharemode. */
165         lck = get_share_mode_lock(talloc_tos(), fsp->file_id, NULL, NULL,
166                                   NULL);
167         if (lck == NULL) {
168                 DEBUG(0,("remove_oplock: failed to lock share entry for "
169                          "file %s\n", fsp_str_dbg(fsp)));
170                 return False;
171         }
172         ret = remove_share_oplock(lck, fsp);
173         if (!ret) {
174                 DEBUG(0,("remove_oplock: failed to remove share oplock for "
175                          "file %s fnum %d, %s\n",
176                          fsp_str_dbg(fsp), fsp->fnum,
177                          file_id_string_tos(&fsp->file_id)));
178         }
179         release_file_oplock(fsp);
180         TALLOC_FREE(lck);
181         return ret;
182 }
183
184 /*
185  * Deal with a reply when a break-to-level II was sent.
186  */
187 bool downgrade_oplock(files_struct *fsp)
188 {
189         bool ret;
190         struct share_mode_lock *lck;
191
192         lck = get_share_mode_lock(talloc_tos(), fsp->file_id, NULL, NULL,
193                                   NULL);
194         if (lck == NULL) {
195                 DEBUG(0,("downgrade_oplock: failed to lock share entry for "
196                          "file %s\n", fsp_str_dbg(fsp)));
197                 return False;
198         }
199         ret = downgrade_share_oplock(lck, fsp);
200         if (!ret) {
201                 DEBUG(0,("downgrade_oplock: failed to downgrade share oplock "
202                          "for file %s fnum %d, file_id %s\n",
203                          fsp_str_dbg(fsp), fsp->fnum,
204                          file_id_string_tos(&fsp->file_id)));
205         }
206
207         downgrade_file_oplock(fsp);
208         TALLOC_FREE(lck);
209         return ret;
210 }
211
212 /*
213  * Some kernel oplock implementations handle the notification themselves.
214  */
215 bool should_notify_deferred_opens()
216 {
217         return !(koplocks &&
218                 (koplocks->flags & KOPLOCKS_DEFERRED_OPEN_NOTIFICATION));
219 }
220
221 /****************************************************************************
222  Set up an oplock break message.
223 ****************************************************************************/
224
225 static char *new_break_message_smb1(TALLOC_CTX *mem_ctx,
226                                    files_struct *fsp, int cmd)
227 {
228         char *result = talloc_array(mem_ctx, char, smb_size + 8*2 + 0);
229
230         if (result == NULL) {
231                 DEBUG(0, ("talloc failed\n"));
232                 return NULL;
233         }
234
235         memset(result,'\0',smb_size);
236         srv_set_message(result,8,0,true);
237         SCVAL(result,smb_com,SMBlockingX);
238         SSVAL(result,smb_tid,fsp->conn->cnum);
239         SSVAL(result,smb_pid,0xFFFF);
240         SSVAL(result,smb_uid,0);
241         SSVAL(result,smb_mid,0xFFFF);
242         SCVAL(result,smb_vwv0,0xFF);
243         SSVAL(result,smb_vwv2,fsp->fnum);
244         SCVAL(result,smb_vwv3,LOCKING_ANDX_OPLOCK_RELEASE);
245         SCVAL(result,smb_vwv3+1,cmd);
246         return result;
247 }
248
249 /****************************************************************************
250  Function to do the waiting before sending a local break.
251 ****************************************************************************/
252
253 static void wait_before_sending_break(void)
254 {
255         long wait_time = (long)lp_oplock_break_wait_time();
256
257         if (wait_time) {
258                 smb_msleep(wait_time);
259         }
260 }
261
262 /****************************************************************************
263  Ensure that we have a valid oplock.
264 ****************************************************************************/
265
266 static files_struct *initial_break_processing(
267         struct smbd_server_connection *sconn, struct file_id id,
268         unsigned long file_id)
269 {
270         files_struct *fsp = NULL;
271
272         if( DEBUGLVL( 3 ) ) {
273                 dbgtext( "initial_break_processing: called for %s/%u\n",
274                          file_id_string_tos(&id), (int)file_id);
275                 dbgtext( "Current oplocks_open (exclusive = %d, levelII = %d)\n",
276                         exclusive_oplocks_open, level_II_oplocks_open );
277         }
278
279         /*
280          * We need to search the file open table for the
281          * entry containing this dev and inode, and ensure
282          * we have an oplock on it.
283          */
284
285         fsp = file_find_dif(sconn, id, file_id);
286
287         if(fsp == NULL) {
288                 /* The file could have been closed in the meantime - return success. */
289                 if( DEBUGLVL( 3 ) ) {
290                         dbgtext( "initial_break_processing: cannot find open file with " );
291                         dbgtext( "file_id %s gen_id = %lu", file_id_string_tos(&id), file_id);
292                         dbgtext( "allowing break to succeed.\n" );
293                 }
294                 return NULL;
295         }
296
297         /* Ensure we have an oplock on the file */
298
299         /*
300          * There is a potential race condition in that an oplock could
301          * have been broken due to another udp request, and yet there are
302          * still oplock break messages being sent in the udp message
303          * queue for this file. So return true if we don't have an oplock,
304          * as we may have just freed it.
305          */
306
307         if(fsp->oplock_type == NO_OPLOCK) {
308                 if( DEBUGLVL( 3 ) ) {
309                         dbgtext( "initial_break_processing: file %s ",
310                                  fsp_str_dbg(fsp));
311                         dbgtext( "(file_id = %s gen_id = %lu) has no oplock.\n",
312                                  file_id_string_tos(&id), fsp->fh->gen_id );
313                         dbgtext( "Allowing break to succeed regardless.\n" );
314                 }
315                 return NULL;
316         }
317
318         return fsp;
319 }
320
321 static void oplock_timeout_handler(struct event_context *ctx,
322                                    struct timed_event *te,
323                                    struct timeval now,
324                                    void *private_data)
325 {
326         files_struct *fsp = (files_struct *)private_data;
327
328         /* Remove the timed event handler. */
329         TALLOC_FREE(fsp->oplock_timeout);
330         DEBUG(0, ("Oplock break failed for file %s -- replying anyway\n",
331                   fsp_str_dbg(fsp)));
332         remove_oplock(fsp);
333         reply_to_oplock_break_requests(fsp);
334 }
335
336 /*******************************************************************
337  Add a timeout handler waiting for the client reply.
338 *******************************************************************/
339
340 static void add_oplock_timeout_handler(files_struct *fsp)
341 {
342         /*
343          * If kernel oplocks already notifies smbds when an oplock break times
344          * out, just return.
345          */
346         if (koplocks &&
347             (koplocks->flags & KOPLOCKS_TIMEOUT_NOTIFICATION)) {
348                 return;
349         }
350
351         if (fsp->oplock_timeout != NULL) {
352                 DEBUG(0, ("Logic problem -- have an oplock event hanging "
353                           "around\n"));
354         }
355
356         fsp->oplock_timeout =
357                 tevent_add_timer(fsp->conn->sconn->ev_ctx, fsp,
358                                  timeval_current_ofs(OPLOCK_BREAK_TIMEOUT, 0),
359                                  oplock_timeout_handler, fsp);
360
361         if (fsp->oplock_timeout == NULL) {
362                 DEBUG(0, ("Could not add oplock timeout handler\n"));
363         }
364 }
365
366 static void send_break_message_smb1(files_struct *fsp, int level)
367 {
368         char *break_msg = new_break_message_smb1(talloc_tos(),
369                                         fsp,
370                                         level);
371         if (break_msg == NULL) {
372                 exit_server("Could not talloc break_msg\n");
373         }
374
375         show_msg(break_msg);
376         if (!srv_send_smb(fsp->conn->sconn,
377                         break_msg, false, 0,
378                         IS_CONN_ENCRYPTED(fsp->conn),
379                         NULL)) {
380                 exit_server_cleanly("send_break_message_smb1: "
381                         "srv_send_smb failed.");
382         }
383
384         TALLOC_FREE(break_msg);
385 }
386
387 void break_level2_to_none_async(files_struct *fsp)
388 {
389         struct smbd_server_connection *sconn = fsp->conn->sconn;
390
391         if (fsp->oplock_type == NO_OPLOCK) {
392                 /* We already got a "break to none" message and we've handled
393                  * it.  just ignore. */
394                 DEBUG(3, ("process_oplock_async_level2_break_message: already "
395                           "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: "
402                           "downgrading fake level 2 oplock.\n"));
403                 remove_oplock(fsp);
404                 return;
405         }
406
407         /* Ensure we're really at level2 state. */
408         SMB_ASSERT(fsp->oplock_type == LEVEL_II_OPLOCK);
409
410         DEBUG(10,("process_oplock_async_level2_break_message: sending break "
411                   "to none message for fid %d, file %s\n", fsp->fnum,
412                   fsp_str_dbg(fsp)));
413
414         /* Now send a break to none message to our client. */
415         if (sconn->using_smb2) {
416                 send_break_message_smb2(fsp, OPLOCKLEVEL_NONE);
417         } else {
418                 send_break_message_smb1(fsp, OPLOCKLEVEL_NONE);
419         }
420
421         /* Async level2 request, don't send a reply, just remove the oplock. */
422         remove_oplock(fsp);
423 }
424
425 /*******************************************************************
426  This handles the case of a write triggering a break to none
427  message on a level2 oplock.
428  When we get this message we may be in any of three states :
429  NO_OPLOCK, LEVEL_II, FAKE_LEVEL2. We only send a message to
430  the client for LEVEL2.
431 *******************************************************************/
432
433 void process_oplock_async_level2_break_message(struct messaging_context *msg_ctx,
434                                                       void *private_data,
435                                                       uint32_t msg_type,
436                                                       struct server_id src,
437                                                       DATA_BLOB *data)
438 {
439         struct share_mode_entry msg;
440         files_struct *fsp;
441         struct smbd_server_connection *sconn =
442                 talloc_get_type(private_data,
443                 struct smbd_server_connection);
444
445         if (sconn == NULL) {
446                 return;
447         }
448
449         if (data->data == NULL) {
450                 DEBUG(0, ("Got NULL buffer\n"));
451                 return;
452         }
453
454         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
455                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
456                 return;
457         }
458
459         /* De-linearize incoming message. */
460         message_to_share_mode_entry(&msg, (char *)data->data);
461
462         DEBUG(10, ("Got oplock async level 2 break message from pid %s: "
463                    "%s/%llu\n", server_id_str(talloc_tos(), &src),
464                    file_id_string_tos(&msg.id),
465                    (unsigned long long)msg.share_file_id));
466
467         fsp = initial_break_processing(sconn, msg.id, msg.share_file_id);
468
469         if (fsp == NULL) {
470                 /* We hit a race here. Break messages are sent, and before we
471                  * get to process this message, we have closed the file. 
472                  * No need to reply as this is an async message. */
473                 DEBUG(3, ("process_oplock_async_level2_break_message: Did not find fsp, ignoring\n"));
474                 return;
475         }
476
477         break_level2_to_none_async(fsp);
478 }
479
480 /*******************************************************************
481  This handles the generic oplock break message from another smbd.
482 *******************************************************************/
483
484 static void process_oplock_break_message(struct messaging_context *msg_ctx,
485                                          void *private_data,
486                                          uint32_t msg_type,
487                                          struct server_id src,
488                                          DATA_BLOB *data)
489 {
490         struct share_mode_entry msg;
491         files_struct *fsp;
492         bool break_to_level2 = False;
493         struct smbd_server_connection *sconn =
494                 talloc_get_type(private_data,
495                 struct smbd_server_connection);
496
497         if (sconn == NULL) {
498                 return;
499         }
500
501         if (data->data == NULL) {
502                 DEBUG(0, ("Got NULL buffer\n"));
503                 return;
504         }
505
506         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
507                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
508                 return;
509         }
510
511         /* De-linearize incoming message. */
512         message_to_share_mode_entry(&msg, (char *)data->data);
513
514         DEBUG(10, ("Got oplock break message from pid %s: %s/%llu\n",
515                    server_id_str(talloc_tos(), &src),
516                    file_id_string_tos(&msg.id),
517                    (unsigned long long)msg.share_file_id));
518
519         fsp = initial_break_processing(sconn, msg.id, msg.share_file_id);
520
521         if (fsp == NULL) {
522                 /* We hit a race here. Break messages are sent, and before we
523                  * get to process this message, we have closed the file. Reply
524                  * with 'ok, oplock broken' */
525                 DEBUG(3, ("Did not find fsp\n"));
526
527                 /* We just send the same message back. */
528                 messaging_send_buf(msg_ctx, src, MSG_SMB_BREAK_RESPONSE,
529                                    (uint8 *)data->data,
530                                    MSG_SMB_SHARE_MODE_ENTRY_SIZE);
531                 return;
532         }
533
534         if (fsp->sent_oplock_break != NO_BREAK_SENT) {
535                 /* Remember we have to inform the requesting PID when the
536                  * client replies */
537                 msg.pid = src;
538                 ADD_TO_ARRAY(NULL, struct share_mode_entry, msg,
539                              &fsp->pending_break_messages,
540                              &fsp->num_pending_break_messages);
541                 return;
542         }
543
544         if (EXCLUSIVE_OPLOCK_TYPE(msg.op_type) &&
545             !EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
546                 DEBUG(3, ("Already downgraded oplock on %s: %s\n",
547                           file_id_string_tos(&fsp->file_id),
548                           fsp_str_dbg(fsp)));
549                 /* We just send the same message back. */
550                 messaging_send_buf(msg_ctx, src, MSG_SMB_BREAK_RESPONSE,
551                                    (uint8 *)data->data,
552                                    MSG_SMB_SHARE_MODE_ENTRY_SIZE);
553                 return;
554         }
555
556         if ((global_client_caps & CAP_LEVEL_II_OPLOCKS) && 
557             !(msg.op_type & FORCE_OPLOCK_BREAK_TO_NONE) &&
558             !(koplocks && !(koplocks->flags & KOPLOCKS_LEVEL2_SUPPORTED)) &&
559             lp_level2_oplocks(SNUM(fsp->conn))) {
560                 break_to_level2 = True;
561         }
562
563         /* Need to wait before sending a break
564            message if we sent ourselves this message. */
565         if (procid_is_me(&src)) {
566                 wait_before_sending_break();
567         }
568
569         if (sconn->using_smb2) {
570                 send_break_message_smb2(fsp, break_to_level2 ?
571                         OPLOCKLEVEL_II : OPLOCKLEVEL_NONE);
572         } else {
573                 send_break_message_smb1(fsp, break_to_level2 ?
574                         OPLOCKLEVEL_II : OPLOCKLEVEL_NONE);
575         }
576
577         fsp->sent_oplock_break = break_to_level2 ? LEVEL_II_BREAK_SENT:BREAK_TO_NONE_SENT;
578
579         msg.pid = src;
580         ADD_TO_ARRAY(NULL, struct share_mode_entry, msg,
581                      &fsp->pending_break_messages,
582                      &fsp->num_pending_break_messages);
583
584         add_oplock_timeout_handler(fsp);
585 }
586
587 /*******************************************************************
588  This handles the kernel oplock break message.
589 *******************************************************************/
590
591 static void process_kernel_oplock_break(struct messaging_context *msg_ctx,
592                                         void *private_data,
593                                         uint32_t msg_type,
594                                         struct server_id src,
595                                         DATA_BLOB *data)
596 {
597         struct file_id id;
598         unsigned long file_id;
599         files_struct *fsp;
600         struct smbd_server_connection *sconn =
601                 talloc_get_type(private_data,
602                 struct smbd_server_connection);
603
604         if (sconn == NULL) {
605                 return;
606         }
607
608         if (data->data == NULL) {
609                 DEBUG(0, ("Got NULL buffer\n"));
610                 return;
611         }
612
613         if (data->length != MSG_SMB_KERNEL_BREAK_SIZE) {
614                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
615                 return;
616         }
617
618         /* Pull the data from the message. */
619         pull_file_id_24((char *)data->data, &id);
620         file_id = (unsigned long)IVAL(data->data, 24);
621
622         DEBUG(10, ("Got kernel oplock break message from pid %s: %s/%u\n",
623                    server_id_str(talloc_tos(), &src), file_id_string_tos(&id),
624                    (unsigned int)file_id));
625
626         fsp = initial_break_processing(sconn, id, file_id);
627
628         if (fsp == NULL) {
629                 DEBUG(3, ("Got a kernel oplock break message for a file "
630                           "I don't know about\n"));
631                 return;
632         }
633
634         if (fsp->sent_oplock_break != NO_BREAK_SENT) {
635                 /* This is ok, kernel oplocks come in completely async */
636                 DEBUG(3, ("Got a kernel oplock request while waiting for a "
637                           "break reply\n"));
638                 return;
639         }
640
641         if (sconn->using_smb2) {
642                 send_break_message_smb2(fsp, OPLOCKLEVEL_NONE);
643         } else {
644                 send_break_message_smb1(fsp, OPLOCKLEVEL_NONE);
645         }
646
647         fsp->sent_oplock_break = BREAK_TO_NONE_SENT;
648
649         add_oplock_timeout_handler(fsp);
650 }
651
652 void reply_to_oplock_break_requests(files_struct *fsp)
653 {
654         int i;
655
656         /*
657          * If kernel oplocks already notifies smbds when oplocks are
658          * broken/removed, just return.
659          */
660         if (koplocks &&
661             (koplocks->flags & KOPLOCKS_OPLOCK_BROKEN_NOTIFICATION)) {
662                 return;
663         }
664
665         for (i=0; i<fsp->num_pending_break_messages; i++) {
666                 struct share_mode_entry *e = &fsp->pending_break_messages[i];
667                 char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
668
669                 share_mode_entry_to_message(msg, e);
670
671                 messaging_send_buf(fsp->conn->sconn->msg_ctx, e->pid,
672                                    MSG_SMB_BREAK_RESPONSE,
673                                    (uint8 *)msg,
674                                    MSG_SMB_SHARE_MODE_ENTRY_SIZE);
675         }
676
677         SAFE_FREE(fsp->pending_break_messages);
678         fsp->num_pending_break_messages = 0;
679         if (fsp->oplock_timeout != NULL) {
680                 /* Remove the timed event handler. */
681                 TALLOC_FREE(fsp->oplock_timeout);
682                 fsp->oplock_timeout = NULL;
683         }
684         return;
685 }
686
687 static void process_oplock_break_response(struct messaging_context *msg_ctx,
688                                           void *private_data,
689                                           uint32_t msg_type,
690                                           struct server_id src,
691                                           DATA_BLOB *data)
692 {
693         struct share_mode_entry msg;
694         struct smbd_server_connection *sconn =
695                 talloc_get_type(private_data,
696                 struct smbd_server_connection);
697
698         if (sconn == NULL) {
699                 return;
700         }
701
702         if (data->data == NULL) {
703                 DEBUG(0, ("Got NULL buffer\n"));
704                 return;
705         }
706
707         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
708                 DEBUG(0, ("Got invalid msg len %u\n",
709                           (unsigned int)data->length));
710                 return;
711         }
712
713         /* De-linearize incoming message. */
714         message_to_share_mode_entry(&msg, (char *)data->data);
715
716         DEBUG(10, ("Got oplock break response from pid %s: %s/%llu mid %llu\n",
717                    server_id_str(talloc_tos(), &src),
718                    file_id_string_tos(&msg.id),
719                    (unsigned long long)msg.share_file_id,
720                    (unsigned long long)msg.op_mid));
721
722         schedule_deferred_open_message_smb(sconn, msg.op_mid);
723 }
724
725 static void process_open_retry_message(struct messaging_context *msg_ctx,
726                                        void *private_data,
727                                        uint32_t msg_type,
728                                        struct server_id src,
729                                        DATA_BLOB *data)
730 {
731         struct share_mode_entry msg;
732         struct smbd_server_connection *sconn;
733
734         if (data->data == NULL) {
735                 DEBUG(0, ("Got NULL buffer\n"));
736                 return;
737         }
738
739         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
740                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
741                 return;
742         }
743
744         /* De-linearize incoming message. */
745         message_to_share_mode_entry(&msg, (char *)data->data);
746
747         DEBUG(10, ("Got open retry msg from pid %s: %s mid %llu\n",
748                    server_id_str(talloc_tos(), &src), file_id_string_tos(&msg.id),
749                    (unsigned long long)msg.op_mid));
750
751         sconn = msg_ctx_to_sconn(msg_ctx);
752         if (sconn != NULL) {
753                 schedule_deferred_open_message_smb(sconn, msg.op_mid);
754         }
755 }
756
757 /****************************************************************************
758  This function is called on any file modification or lock request. If a file
759  is level 2 oplocked then it must tell all other level 2 holders to break to
760  none.
761 ****************************************************************************/
762
763 static void contend_level2_oplocks_begin_default(files_struct *fsp,
764                                               enum level2_contention_type type)
765 {
766         int i;
767         struct share_mode_lock *lck;
768
769         /*
770          * If this file is level II oplocked then we need
771          * to grab the shared memory lock and inform all
772          * other files with a level II lock that they need
773          * to flush their read caches. We keep the lock over
774          * the shared memory area whilst doing this.
775          */
776
777         if (!LEVEL_II_OPLOCK_TYPE(fsp->oplock_type))
778                 return;
779
780         lck = get_share_mode_lock(talloc_tos(), fsp->file_id, NULL, NULL,
781                                   NULL);
782         if (lck == NULL) {
783                 DEBUG(0,("release_level_2_oplocks_on_change: failed to lock "
784                          "share mode entry for file %s.\n", fsp_str_dbg(fsp)));
785                 return;
786         }
787
788         DEBUG(10,("release_level_2_oplocks_on_change: num_share_modes = %d\n", 
789                   lck->num_share_modes ));
790
791         for(i = 0; i < lck->num_share_modes; i++) {
792                 struct share_mode_entry *share_entry = &lck->share_modes[i];
793                 char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
794
795                 if (!is_valid_share_mode_entry(share_entry)) {
796                         continue;
797                 }
798
799                 /*
800                  * As there could have been multiple writes waiting at the
801                  * lock_share_entry gate we may not be the first to
802                  * enter. Hence the state of the op_types in the share mode
803                  * entries may be partly NO_OPLOCK and partly LEVEL_II or FAKE_LEVEL_II
804                  * oplock. It will do no harm to re-send break messages to
805                  * those smbd's that are still waiting their turn to remove
806                  * their LEVEL_II state, and also no harm to ignore existing
807                  * NO_OPLOCK states. JRA.
808                  */
809
810                 DEBUG(10,("release_level_2_oplocks_on_change: "
811                           "share_entry[%i]->op_type == %d\n",
812                           i, share_entry->op_type ));
813
814                 if (share_entry->op_type == NO_OPLOCK) {
815                         continue;
816                 }
817
818                 /* Paranoia .... */
819                 if (EXCLUSIVE_OPLOCK_TYPE(share_entry->op_type)) {
820                         DEBUG(0,("release_level_2_oplocks_on_change: PANIC. "
821                                  "share mode entry %d is an exlusive "
822                                  "oplock !\n", i ));
823                         TALLOC_FREE(lck);
824                         abort();
825                 }
826
827                 share_mode_entry_to_message(msg, share_entry);
828
829                 /*
830                  * Deal with a race condition when breaking level2
831                  * oplocks. Don't send all the messages and release
832                  * the lock, this allows someone else to come in and
833                  * get a level2 lock before any of the messages are
834                  * processed, and thus miss getting a break message.
835                  * Ensure at least one entry (the one we're breaking)
836                  * is processed immediately under the lock and becomes
837                  * set as NO_OPLOCK to stop any waiter getting a level2.
838                  * Bugid #5980.
839                  */
840
841                 if (procid_is_me(&share_entry->pid)) {
842                         struct files_struct *cur_fsp =
843                                 initial_break_processing(fsp->conn->sconn,
844                                         share_entry->id,
845                                         share_entry->share_file_id);
846                         wait_before_sending_break();
847                         if (cur_fsp != NULL) {
848                                 break_level2_to_none_async(cur_fsp);
849                         } else {
850                                 DEBUG(3, ("release_level_2_oplocks_on_change: "
851                                 "Did not find fsp, ignoring\n"));
852                         }
853                 } else {
854                         messaging_send_buf(fsp->conn->sconn->msg_ctx,
855                                         share_entry->pid,
856                                         MSG_SMB_ASYNC_LEVEL2_BREAK,
857                                         (uint8 *)msg,
858                                         MSG_SMB_SHARE_MODE_ENTRY_SIZE);
859                 }
860         }
861
862         /* We let the message receivers handle removing the oplock state
863            in the share mode lock db. */
864
865         TALLOC_FREE(lck);
866 }
867
868 void smbd_contend_level2_oplocks_begin(files_struct *fsp,
869                                   enum level2_contention_type type)
870 {
871         if (koplocks && koplocks->ops->contend_level2_oplocks_begin) {
872                 koplocks->ops->contend_level2_oplocks_begin(fsp, type);
873                 return;
874         }
875
876         contend_level2_oplocks_begin_default(fsp, type);
877 }
878
879 void smbd_contend_level2_oplocks_end(files_struct *fsp,
880                                 enum level2_contention_type type)
881 {
882         /* Only kernel oplocks implement this so far */
883         if (koplocks && koplocks->ops->contend_level2_oplocks_end) {
884                 koplocks->ops->contend_level2_oplocks_end(fsp, type);
885         }
886 }
887
888 /****************************************************************************
889  Linearize a share mode entry struct to an internal oplock break message.
890 ****************************************************************************/
891
892 void share_mode_entry_to_message(char *msg, const struct share_mode_entry *e)
893 {
894         SIVAL(msg,OP_BREAK_MSG_PID_OFFSET,(uint32)e->pid.pid);
895         SBVAL(msg,OP_BREAK_MSG_MID_OFFSET,e->op_mid);
896         SSVAL(msg,OP_BREAK_MSG_OP_TYPE_OFFSET,e->op_type);
897         SIVAL(msg,OP_BREAK_MSG_ACCESS_MASK_OFFSET,e->access_mask);
898         SIVAL(msg,OP_BREAK_MSG_SHARE_ACCESS_OFFSET,e->share_access);
899         SIVAL(msg,OP_BREAK_MSG_PRIV_OFFSET,e->private_options);
900         SIVAL(msg,OP_BREAK_MSG_TIME_SEC_OFFSET,(uint32_t)e->time.tv_sec);
901         SIVAL(msg,OP_BREAK_MSG_TIME_USEC_OFFSET,(uint32_t)e->time.tv_usec);
902         push_file_id_24(msg+OP_BREAK_MSG_DEV_OFFSET, &e->id);
903         SIVAL(msg,OP_BREAK_MSG_FILE_ID_OFFSET,e->share_file_id);
904         SIVAL(msg,OP_BREAK_MSG_UID_OFFSET,e->uid);
905         SSVAL(msg,OP_BREAK_MSG_FLAGS_OFFSET,e->flags);
906         SIVAL(msg,OP_BREAK_MSG_NAME_HASH_OFFSET,e->name_hash);
907         SIVAL(msg,OP_BREAK_MSG_VNN_OFFSET,e->pid.vnn);
908 }
909
910 /****************************************************************************
911  De-linearize an internal oplock break message to a share mode entry struct.
912 ****************************************************************************/
913
914 void message_to_share_mode_entry(struct share_mode_entry *e, char *msg)
915 {
916         e->pid.pid = (pid_t)IVAL(msg,OP_BREAK_MSG_PID_OFFSET);
917         e->op_mid = BVAL(msg,OP_BREAK_MSG_MID_OFFSET);
918         e->op_type = SVAL(msg,OP_BREAK_MSG_OP_TYPE_OFFSET);
919         e->access_mask = IVAL(msg,OP_BREAK_MSG_ACCESS_MASK_OFFSET);
920         e->share_access = IVAL(msg,OP_BREAK_MSG_SHARE_ACCESS_OFFSET);
921         e->private_options = IVAL(msg,OP_BREAK_MSG_PRIV_OFFSET);
922         e->time.tv_sec = (time_t)IVAL(msg,OP_BREAK_MSG_TIME_SEC_OFFSET);
923         e->time.tv_usec = (int)IVAL(msg,OP_BREAK_MSG_TIME_USEC_OFFSET);
924         pull_file_id_24(msg+OP_BREAK_MSG_DEV_OFFSET, &e->id);
925         e->share_file_id = (unsigned long)IVAL(msg,OP_BREAK_MSG_FILE_ID_OFFSET);
926         e->uid = (uint32)IVAL(msg,OP_BREAK_MSG_UID_OFFSET);
927         e->flags = (uint16)SVAL(msg,OP_BREAK_MSG_FLAGS_OFFSET);
928         e->name_hash = IVAL(msg,OP_BREAK_MSG_NAME_HASH_OFFSET);
929         e->pid.vnn = IVAL(msg,OP_BREAK_MSG_VNN_OFFSET);
930 }
931
932 /****************************************************************************
933  Setup oplocks for this process.
934 ****************************************************************************/
935
936 bool init_oplocks(struct smbd_server_connection *sconn)
937 {
938         DEBUG(3,("init_oplocks: initializing messages.\n"));
939
940         messaging_register(sconn->msg_ctx, sconn, MSG_SMB_BREAK_REQUEST,
941                            process_oplock_break_message);
942         messaging_register(sconn->msg_ctx, sconn, MSG_SMB_ASYNC_LEVEL2_BREAK,
943                            process_oplock_async_level2_break_message);
944         messaging_register(sconn->msg_ctx, sconn, MSG_SMB_BREAK_RESPONSE,
945                            process_oplock_break_response);
946         messaging_register(sconn->msg_ctx, sconn, MSG_SMB_KERNEL_BREAK,
947                            process_kernel_oplock_break);
948         messaging_register(sconn->msg_ctx, NULL, MSG_SMB_OPEN_RETRY,
949                            process_open_retry_message);
950
951         if (lp_kernel_oplocks()) {
952 #if HAVE_KERNEL_OPLOCKS_IRIX
953                 koplocks = irix_init_kernel_oplocks(NULL);
954 #elif HAVE_KERNEL_OPLOCKS_LINUX
955                 koplocks = linux_init_kernel_oplocks(NULL);
956 #elif HAVE_ONEFS
957 #error Isilon, please check if the NULL context is okay here. Thanks!
958                 koplocks = onefs_init_kernel_oplocks(NULL);
959 #endif
960         }
961
962         return True;
963 }