21792bdee40675a3c8f407568069f0f8a1b96b0f
[mat/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 #include "../librpc/gen_ndr/open_files.h"
28
29 /*
30  * helper function used by the kernel oplock backends to post the break message
31  */
32 void break_kernel_oplock(struct messaging_context *msg_ctx, files_struct *fsp)
33 {
34         uint8_t msg[MSG_SMB_KERNEL_BREAK_SIZE];
35
36         /* Put the kernel break info into the message. */
37         push_file_id_24((char *)msg, &fsp->file_id);
38         SIVAL(msg,24,fsp->fh->gen_id);
39
40         /* Don't need to be root here as we're only ever
41            sending to ourselves. */
42
43         messaging_send_buf(msg_ctx, messaging_server_id(msg_ctx),
44                            MSG_SMB_KERNEL_BREAK,
45                            msg, MSG_SMB_KERNEL_BREAK_SIZE);
46 }
47
48 /****************************************************************************
49  Attempt to set an oplock on a file. Succeeds if kernel oplocks are
50  disabled (just sets flags) and no byte-range locks in the file. Returns True
51  if oplock set.
52 ****************************************************************************/
53
54 NTSTATUS set_file_oplock(files_struct *fsp, int oplock_type)
55 {
56         struct smbd_server_connection *sconn = fsp->conn->sconn;
57         struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
58         bool use_kernel = lp_kernel_oplocks(SNUM(fsp->conn)) && koplocks;
59
60         if (fsp->oplock_type == LEVEL_II_OPLOCK) {
61                 if (use_kernel &&
62                     !(koplocks->flags & KOPLOCKS_LEVEL2_SUPPORTED)) {
63                         DEBUG(10, ("Refusing level2 oplock, kernel oplocks "
64                                    "don't support them\n"));
65                         return NT_STATUS_NOT_SUPPORTED;
66                 }
67         }
68
69         if ((fsp->oplock_type != NO_OPLOCK) &&
70             (fsp->oplock_type != FAKE_LEVEL_II_OPLOCK) &&
71             use_kernel &&
72             !koplocks->ops->set_oplock(koplocks, fsp, oplock_type))
73         {
74                 return map_nt_error_from_unix(errno);
75         }
76
77         fsp->oplock_type = oplock_type;
78         fsp->sent_oplock_break = NO_BREAK_SENT;
79         if (oplock_type == LEVEL_II_OPLOCK) {
80                 sconn->oplocks.level_II_open++;
81         } else if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
82                 sconn->oplocks.exclusive_open++;
83         }
84
85         DEBUG(5,("set_file_oplock: granted oplock on file %s, %s/%lu, "
86                     "tv_sec = %x, tv_usec = %x\n",
87                  fsp_str_dbg(fsp), file_id_string_tos(&fsp->file_id),
88                  fsp->fh->gen_id, (int)fsp->open_time.tv_sec,
89                  (int)fsp->open_time.tv_usec ));
90
91         return NT_STATUS_OK;
92 }
93
94 /****************************************************************************
95  Attempt to release an oplock on a file. Decrements oplock count.
96 ****************************************************************************/
97
98 void release_file_oplock(files_struct *fsp)
99 {
100         struct smbd_server_connection *sconn = fsp->conn->sconn;
101         struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
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                 sconn->oplocks.level_II_open--;
111         } else if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
112                 sconn->oplocks.exclusive_open--;
113         }
114
115         SMB_ASSERT(sconn->oplocks.exclusive_open>=0);
116         SMB_ASSERT(sconn->oplocks.level_II_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         struct smbd_server_connection *sconn = fsp->conn->sconn;
139         struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
140
141         if (!EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
142                 DEBUG(0, ("trying to downgrade an already-downgraded oplock!\n"));
143                 return;
144         }
145
146         if (koplocks) {
147                 koplocks->ops->release_oplock(koplocks, fsp, LEVEL_II_OPLOCK);
148         }
149         fsp->oplock_type = LEVEL_II_OPLOCK;
150         sconn->oplocks.exclusive_open--;
151         sconn->oplocks.level_II_open++;
152         fsp->sent_oplock_break = NO_BREAK_SENT;
153 }
154
155 /****************************************************************************
156  Remove a file oplock. Copes with level II and exclusive.
157  Locks then unlocks the share mode lock. Client can decide to go directly
158  to none even if a "break-to-level II" was sent.
159 ****************************************************************************/
160
161 bool remove_oplock(files_struct *fsp)
162 {
163         bool ret;
164         struct share_mode_lock *lck;
165
166         /* Remove the oplock flag from the sharemode. */
167         lck = get_existing_share_mode_lock(talloc_tos(), fsp->file_id);
168         if (lck == NULL) {
169                 DEBUG(0,("remove_oplock: failed to lock share entry for "
170                          "file %s\n", fsp_str_dbg(fsp)));
171                 return False;
172         }
173         ret = remove_share_oplock(lck, fsp);
174         if (!ret) {
175                 DEBUG(0,("remove_oplock: failed to remove share oplock for "
176                          "file %s, %s, %s\n",
177                          fsp_str_dbg(fsp), fsp_fnum_dbg(fsp),
178                          file_id_string_tos(&fsp->file_id)));
179         }
180         release_file_oplock(fsp);
181         TALLOC_FREE(lck);
182         return ret;
183 }
184
185 /*
186  * Deal with a reply when a break-to-level II was sent.
187  */
188 bool downgrade_oplock(files_struct *fsp)
189 {
190         bool ret;
191         struct share_mode_lock *lck;
192
193         lck = get_existing_share_mode_lock(talloc_tos(), fsp->file_id);
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, %s, file_id %s\n",
203                          fsp_str_dbg(fsp), fsp_fnum_dbg(fsp),
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(struct smbd_server_connection *sconn)
216 {
217         struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
218         return !(koplocks &&
219                 (koplocks->flags & KOPLOCKS_DEFERRED_OPEN_NOTIFICATION));
220 }
221
222 /****************************************************************************
223  Set up an oplock break message.
224 ****************************************************************************/
225
226 #define SMB1_BREAK_MESSAGE_LENGTH (smb_size + 8*2)
227
228 static void new_break_message_smb1(files_struct *fsp, int cmd,
229                                    char result[SMB1_BREAK_MESSAGE_LENGTH])
230 {
231         memset(result,'\0',smb_size);
232         srv_set_message(result,8,0,true);
233         SCVAL(result,smb_com,SMBlockingX);
234         SSVAL(result,smb_tid,fsp->conn->cnum);
235         SSVAL(result,smb_pid,0xFFFF);
236         SSVAL(result,smb_uid,0);
237         SSVAL(result,smb_mid,0xFFFF);
238         SCVAL(result,smb_vwv0,0xFF);
239         SSVAL(result,smb_vwv2,fsp->fnum);
240         SCVAL(result,smb_vwv3,LOCKING_ANDX_OPLOCK_RELEASE);
241         SCVAL(result,smb_vwv3+1,cmd);
242 }
243
244 /****************************************************************************
245  Function to do the waiting before sending a local break.
246 ****************************************************************************/
247
248 static void wait_before_sending_break(void)
249 {
250         long wait_time = (long)lp_oplock_break_wait_time();
251
252         if (wait_time) {
253                 smb_msleep(wait_time);
254         }
255 }
256
257 /****************************************************************************
258  Ensure that we have a valid oplock.
259 ****************************************************************************/
260
261 static files_struct *initial_break_processing(
262         struct smbd_server_connection *sconn, struct file_id id,
263         unsigned long file_id)
264 {
265         files_struct *fsp = NULL;
266
267         if( DEBUGLVL( 3 ) ) {
268                 dbgtext( "initial_break_processing: called for %s/%u\n",
269                          file_id_string_tos(&id), (int)file_id);
270                 dbgtext( "Current oplocks_open (exclusive = %d, levelII = %d)\n",
271                         sconn->oplocks.exclusive_open,
272                         sconn->oplocks.level_II_open);
273         }
274
275         /*
276          * We need to search the file open table for the
277          * entry containing this dev and inode, and ensure
278          * we have an oplock on it.
279          */
280
281         fsp = file_find_dif(sconn, id, file_id);
282
283         if(fsp == NULL) {
284                 /* The file could have been closed in the meantime - return success. */
285                 if( DEBUGLVL( 3 ) ) {
286                         dbgtext( "initial_break_processing: cannot find open file with " );
287                         dbgtext( "file_id %s gen_id = %lu, ", file_id_string_tos(&id), file_id);
288                         dbgtext( "allowing break to succeed.\n" );
289                 }
290                 return NULL;
291         }
292
293         /* Ensure we have an oplock on the file */
294
295         /*
296          * There is a potential race condition in that an oplock could
297          * have been broken due to another udp request, and yet there are
298          * still oplock break messages being sent in the udp message
299          * queue for this file. So return true if we don't have an oplock,
300          * as we may have just freed it.
301          */
302
303         if(fsp->oplock_type == NO_OPLOCK) {
304                 if( DEBUGLVL( 3 ) ) {
305                         dbgtext( "initial_break_processing: file %s ",
306                                  fsp_str_dbg(fsp));
307                         dbgtext( "(file_id = %s gen_id = %lu) has no oplock.\n",
308                                  file_id_string_tos(&id), fsp->fh->gen_id );
309                         dbgtext( "Allowing break to succeed regardless.\n" );
310                 }
311                 return NULL;
312         }
313
314         return fsp;
315 }
316
317 static void oplock_timeout_handler(struct tevent_context *ctx,
318                                    struct tevent_timer *te,
319                                    struct timeval now,
320                                    void *private_data)
321 {
322         files_struct *fsp = (files_struct *)private_data;
323
324         /* Remove the timed event handler. */
325         TALLOC_FREE(fsp->oplock_timeout);
326         DEBUG(0, ("Oplock break failed for file %s -- replying anyway\n",
327                   fsp_str_dbg(fsp)));
328         remove_oplock(fsp);
329 }
330
331 /*******************************************************************
332  Add a timeout handler waiting for the client reply.
333 *******************************************************************/
334
335 static void add_oplock_timeout_handler(files_struct *fsp)
336 {
337         struct smbd_server_connection *sconn = fsp->conn->sconn;
338         struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
339
340         /*
341          * If kernel oplocks already notifies smbds when an oplock break times
342          * out, just return.
343          */
344         if (koplocks &&
345             (koplocks->flags & KOPLOCKS_TIMEOUT_NOTIFICATION)) {
346                 return;
347         }
348
349         if (fsp->oplock_timeout != NULL) {
350                 DEBUG(0, ("Logic problem -- have an oplock event hanging "
351                           "around\n"));
352         }
353
354         fsp->oplock_timeout =
355                 tevent_add_timer(fsp->conn->sconn->ev_ctx, fsp,
356                                  timeval_current_ofs(OPLOCK_BREAK_TIMEOUT, 0),
357                                  oplock_timeout_handler, fsp);
358
359         if (fsp->oplock_timeout == NULL) {
360                 DEBUG(0, ("Could not add oplock timeout handler\n"));
361         }
362 }
363
364 static void send_break_message_smb1(files_struct *fsp, int level)
365 {
366         char break_msg[SMB1_BREAK_MESSAGE_LENGTH];
367
368         new_break_message_smb1(fsp, level, break_msg);
369
370         show_msg(break_msg);
371         if (!srv_send_smb(fsp->conn->sconn,
372                         break_msg, false, 0,
373                         IS_CONN_ENCRYPTED(fsp->conn),
374                         NULL)) {
375                 exit_server_cleanly("send_break_message_smb1: "
376                         "srv_send_smb failed.");
377         }
378 }
379
380 static void break_level2_to_none_async(files_struct *fsp)
381 {
382         struct smbd_server_connection *sconn = fsp->conn->sconn;
383
384         if (fsp->oplock_type == NO_OPLOCK) {
385                 /* We already got a "break to none" message and we've handled
386                  * it.  just ignore. */
387                 DEBUG(3, ("process_oplock_async_level2_break_message: already "
388                           "broken to none, ignoring.\n"));
389                 return;
390         }
391
392         if (fsp->oplock_type == FAKE_LEVEL_II_OPLOCK) {
393                 /* Don't tell the client, just downgrade. */
394                 DEBUG(3, ("process_oplock_async_level2_break_message: "
395                           "downgrading fake level 2 oplock.\n"));
396                 remove_oplock(fsp);
397                 return;
398         }
399
400         /* Ensure we're really at level2 state. */
401         SMB_ASSERT(fsp->oplock_type == LEVEL_II_OPLOCK);
402
403         DEBUG(10,("process_oplock_async_level2_break_message: sending break "
404                   "to none message for %s, file %s\n", fsp_fnum_dbg(fsp),
405                   fsp_str_dbg(fsp)));
406
407         /* Now send a break to none message to our client. */
408         if (sconn->using_smb2) {
409                 send_break_message_smb2(fsp, OPLOCKLEVEL_NONE);
410         } else {
411                 send_break_message_smb1(fsp, OPLOCKLEVEL_NONE);
412         }
413
414         /* Async level2 request, don't send a reply, just remove the oplock. */
415         remove_oplock(fsp);
416 }
417
418 /*******************************************************************
419  This handles the case of a write triggering a break to none
420  message on a level2 oplock.
421  When we get this message we may be in any of three states :
422  NO_OPLOCK, LEVEL_II, FAKE_LEVEL2. We only send a message to
423  the client for LEVEL2.
424 *******************************************************************/
425
426 static void process_oplock_async_level2_break_message(struct messaging_context *msg_ctx,
427                                                       void *private_data,
428                                                       uint32_t msg_type,
429                                                       struct server_id src,
430                                                       DATA_BLOB *data)
431 {
432         struct share_mode_entry msg;
433         files_struct *fsp;
434         struct smbd_server_connection *sconn =
435                 talloc_get_type_abort(private_data,
436                 struct smbd_server_connection);
437
438         if (data->data == NULL) {
439                 DEBUG(0, ("Got NULL buffer\n"));
440                 return;
441         }
442
443         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
444                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
445                 return;
446         }
447
448         /* De-linearize incoming message. */
449         message_to_share_mode_entry(&msg, (char *)data->data);
450
451         DEBUG(10, ("Got oplock async level 2 break message from pid %s: "
452                    "%s/%llu\n", server_id_str(talloc_tos(), &src),
453                    file_id_string_tos(&msg.id),
454                    (unsigned long long)msg.share_file_id));
455
456         fsp = initial_break_processing(sconn, msg.id, msg.share_file_id);
457
458         if (fsp == NULL) {
459                 /* We hit a race here. Break messages are sent, and before we
460                  * get to process this message, we have closed the file. 
461                  * No need to reply as this is an async message. */
462                 DEBUG(3, ("process_oplock_async_level2_break_message: Did not find fsp, ignoring\n"));
463                 return;
464         }
465
466         break_level2_to_none_async(fsp);
467 }
468
469 /*******************************************************************
470  This handles the generic oplock break message from another smbd.
471 *******************************************************************/
472
473 static void process_oplock_break_message(struct messaging_context *msg_ctx,
474                                          void *private_data,
475                                          uint32_t msg_type,
476                                          struct server_id src,
477                                          DATA_BLOB *data)
478 {
479         struct share_mode_entry msg;
480         files_struct *fsp;
481         bool break_to_level2 = False;
482         bool use_kernel;
483         struct smbd_server_connection *sconn =
484                 talloc_get_type_abort(private_data,
485                 struct smbd_server_connection);
486         struct server_id self = messaging_server_id(sconn->msg_ctx);
487         struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
488
489         if (data->data == NULL) {
490                 DEBUG(0, ("Got NULL buffer\n"));
491                 return;
492         }
493
494         if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
495                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
496                 return;
497         }
498
499         /* De-linearize incoming message. */
500         message_to_share_mode_entry(&msg, (char *)data->data);
501
502         DEBUG(10, ("Got oplock break message from pid %s: %s/%llu\n",
503                    server_id_str(talloc_tos(), &src),
504                    file_id_string_tos(&msg.id),
505                    (unsigned long long)msg.share_file_id));
506
507         fsp = initial_break_processing(sconn, msg.id, msg.share_file_id);
508
509         if (fsp == NULL) {
510                 /* We hit a race here. Break messages are sent, and before we
511                  * get to process this message, we have closed the file. */
512                 DEBUG(3, ("Did not find fsp\n"));
513                 return;
514         }
515
516         if (fsp->sent_oplock_break != NO_BREAK_SENT) {
517                 /*
518                  * Nothing to do anymore
519                  */
520                 return;
521         }
522
523         if (EXCLUSIVE_OPLOCK_TYPE(msg.op_type) &&
524             !EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
525                 DEBUG(3, ("Already downgraded oplock on %s: %s\n",
526                           file_id_string_tos(&fsp->file_id),
527                           fsp_str_dbg(fsp)));
528                 return;
529         }
530
531         use_kernel = lp_kernel_oplocks(SNUM(fsp->conn)) && koplocks;
532
533         if ((global_client_caps & CAP_LEVEL_II_OPLOCKS) &&
534             !(msg.op_type & FORCE_OPLOCK_BREAK_TO_NONE) &&
535             !(use_kernel && !(koplocks->flags & KOPLOCKS_LEVEL2_SUPPORTED)) &&
536             lp_level2_oplocks(SNUM(fsp->conn))) {
537                 break_to_level2 = True;
538         }
539
540         /* Need to wait before sending a break
541            message if we sent ourselves this message. */
542         if (serverid_equal(&self, &src)) {
543                 wait_before_sending_break();
544         }
545
546         if (sconn->using_smb2) {
547                 send_break_message_smb2(fsp, break_to_level2 ?
548                         OPLOCKLEVEL_II : OPLOCKLEVEL_NONE);
549         } else {
550                 send_break_message_smb1(fsp, break_to_level2 ?
551                         OPLOCKLEVEL_II : OPLOCKLEVEL_NONE);
552         }
553
554         fsp->sent_oplock_break = break_to_level2 ? LEVEL_II_BREAK_SENT:BREAK_TO_NONE_SENT;
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         struct smbd_server_connection *sconn =
573                 talloc_get_type_abort(private_data,
574                 struct smbd_server_connection);
575
576         if (data->data == NULL) {
577                 DEBUG(0, ("Got NULL buffer\n"));
578                 return;
579         }
580
581         if (data->length != MSG_SMB_KERNEL_BREAK_SIZE) {
582                 DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
583                 return;
584         }
585
586         /* Pull the data from the message. */
587         pull_file_id_24((char *)data->data, &id);
588         file_id = (unsigned long)IVAL(data->data, 24);
589
590         DEBUG(10, ("Got kernel oplock break message from pid %s: %s/%u\n",
591                    server_id_str(talloc_tos(), &src), file_id_string_tos(&id),
592                    (unsigned int)file_id));
593
594         fsp = initial_break_processing(sconn, id, file_id);
595
596         if (fsp == NULL) {
597                 DEBUG(3, ("Got a kernel oplock break message for a file "
598                           "I don't know about\n"));
599                 return;
600         }
601
602         if (fsp->sent_oplock_break != NO_BREAK_SENT) {
603                 /* This is ok, kernel oplocks come in completely async */
604                 DEBUG(3, ("Got a kernel oplock request while waiting for a "
605                           "break reply\n"));
606                 return;
607         }
608
609         if (sconn->using_smb2) {
610                 send_break_message_smb2(fsp, OPLOCKLEVEL_NONE);
611         } else {
612                 send_break_message_smb1(fsp, OPLOCKLEVEL_NONE);
613         }
614
615         fsp->sent_oplock_break = BREAK_TO_NONE_SENT;
616
617         add_oplock_timeout_handler(fsp);
618 }
619
620 struct break_to_none_state {
621         struct smbd_server_connection *sconn;
622         struct file_id id;
623 };
624 static void do_break_to_none(struct tevent_context *ctx,
625                              struct tevent_immediate *im,
626                              void *private_data);
627
628 /****************************************************************************
629  This function is called on any file modification or lock request. If a file
630  is level 2 oplocked then it must tell all other level 2 holders to break to
631  none.
632 ****************************************************************************/
633
634 static void contend_level2_oplocks_begin_default(files_struct *fsp,
635                                               enum level2_contention_type type)
636 {
637         struct smbd_server_connection *sconn = fsp->conn->sconn;
638         struct tevent_immediate *im;
639         struct break_to_none_state *state;
640
641         /*
642          * If this file is level II oplocked then we need
643          * to grab the shared memory lock and inform all
644          * other files with a level II lock that they need
645          * to flush their read caches. We keep the lock over
646          * the shared memory area whilst doing this.
647          */
648
649         if (!LEVEL_II_OPLOCK_TYPE(fsp->oplock_type))
650                 return;
651
652         /*
653          * When we get here we might have a brlock entry locked. Also
654          * locking the share mode entry would violate the locking
655          * order. Breaking level2 oplocks to none is asynchronous
656          * anyway, so we postpone this into an immediate timed event.
657          */
658
659         state = talloc(sconn, struct break_to_none_state);
660         if (state == NULL) {
661                 DEBUG(1, ("talloc failed\n"));
662                 return;
663         }
664         state->sconn = sconn;
665         state->id = fsp->file_id;
666
667         im = tevent_create_immediate(state);
668         if (im == NULL) {
669                 DEBUG(1, ("tevent_create_immediate failed\n"));
670                 TALLOC_FREE(state);
671                 return;
672         }
673         tevent_schedule_immediate(im, sconn->ev_ctx, do_break_to_none, state);
674 }
675
676 static void do_break_to_none(struct tevent_context *ctx,
677                              struct tevent_immediate *im,
678                              void *private_data)
679 {
680         struct break_to_none_state *state = talloc_get_type_abort(
681                 private_data, struct break_to_none_state);
682         struct server_id self = messaging_server_id(state->sconn->msg_ctx);
683         int i;
684         struct share_mode_lock *lck;
685
686         lck = get_existing_share_mode_lock(talloc_tos(), state->id);
687         if (lck == NULL) {
688                 DEBUG(1, ("release_level_2_oplocks_on_change: failed to lock "
689                           "share mode entry for file %s.\n",
690                           file_id_string_tos(&state->id)));
691                 goto done;
692         }
693
694         DEBUG(10,("release_level_2_oplocks_on_change: num_share_modes = %d\n", 
695                   lck->data->num_share_modes ));
696
697         for(i = 0; i < lck->data->num_share_modes; i++) {
698                 struct share_mode_entry *share_entry = &lck->data->share_modes[i];
699                 char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
700
701                 if (!is_valid_share_mode_entry(share_entry)) {
702                         continue;
703                 }
704
705                 /*
706                  * As there could have been multiple writes waiting at the
707                  * lock_share_entry gate we may not be the first to
708                  * enter. Hence the state of the op_types in the share mode
709                  * entries may be partly NO_OPLOCK and partly LEVEL_II or FAKE_LEVEL_II
710                  * oplock. It will do no harm to re-send break messages to
711                  * those smbd's that are still waiting their turn to remove
712                  * their LEVEL_II state, and also no harm to ignore existing
713                  * NO_OPLOCK states. JRA.
714                  */
715
716                 DEBUG(10,("release_level_2_oplocks_on_change: "
717                           "share_entry[%i]->op_type == %d\n",
718                           i, share_entry->op_type ));
719
720                 if (share_entry->op_type == NO_OPLOCK) {
721                         continue;
722                 }
723
724                 /* Paranoia .... */
725                 if (EXCLUSIVE_OPLOCK_TYPE(share_entry->op_type)) {
726                         DEBUG(0,("release_level_2_oplocks_on_change: PANIC. "
727                                  "share mode entry %d is an exlusive "
728                                  "oplock !\n", i ));
729                         TALLOC_FREE(lck);
730                         abort();
731                 }
732
733                 share_mode_entry_to_message(msg, share_entry);
734
735                 /*
736                  * Deal with a race condition when breaking level2
737                  * oplocks. Don't send all the messages and release
738                  * the lock, this allows someone else to come in and
739                  * get a level2 lock before any of the messages are
740                  * processed, and thus miss getting a break message.
741                  * Ensure at least one entry (the one we're breaking)
742                  * is processed immediately under the lock and becomes
743                  * set as NO_OPLOCK to stop any waiter getting a level2.
744                  * Bugid #5980.
745                  */
746
747                 if (serverid_equal(&self, &share_entry->pid)) {
748                         struct files_struct *cur_fsp =
749                                 initial_break_processing(state->sconn,
750                                         share_entry->id,
751                                         share_entry->share_file_id);
752                         if (cur_fsp != NULL) {
753                                 wait_before_sending_break();
754                                 break_level2_to_none_async(cur_fsp);
755                         } else {
756                                 DEBUG(3, ("release_level_2_oplocks_on_change: "
757                                 "Did not find fsp, ignoring\n"));
758                         }
759                 } else {
760                         messaging_send_buf(state->sconn->msg_ctx,
761                                         share_entry->pid,
762                                         MSG_SMB_ASYNC_LEVEL2_BREAK,
763                                         (uint8 *)msg,
764                                         MSG_SMB_SHARE_MODE_ENTRY_SIZE);
765                 }
766         }
767
768         /* We let the message receivers handle removing the oplock state
769            in the share mode lock db. */
770
771         TALLOC_FREE(lck);
772 done:
773         TALLOC_FREE(state);
774         return;
775 }
776
777 void smbd_contend_level2_oplocks_begin(files_struct *fsp,
778                                   enum level2_contention_type type)
779 {
780         struct smbd_server_connection *sconn = fsp->conn->sconn;
781         struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
782
783         if (koplocks && koplocks->ops->contend_level2_oplocks_begin) {
784                 koplocks->ops->contend_level2_oplocks_begin(fsp, type);
785                 return;
786         }
787
788         contend_level2_oplocks_begin_default(fsp, type);
789 }
790
791 void smbd_contend_level2_oplocks_end(files_struct *fsp,
792                                 enum level2_contention_type type)
793 {
794         struct smbd_server_connection *sconn = fsp->conn->sconn;
795         struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
796
797         /* Only kernel oplocks implement this so far */
798         if (koplocks && koplocks->ops->contend_level2_oplocks_end) {
799                 koplocks->ops->contend_level2_oplocks_end(fsp, type);
800         }
801 }
802
803 /****************************************************************************
804  Linearize a share mode entry struct to an internal oplock break message.
805 ****************************************************************************/
806
807 void share_mode_entry_to_message(char *msg, const struct share_mode_entry *e)
808 {
809         SIVAL(msg,OP_BREAK_MSG_PID_OFFSET,(uint32)e->pid.pid);
810         SBVAL(msg,OP_BREAK_MSG_MID_OFFSET,e->op_mid);
811         SSVAL(msg,OP_BREAK_MSG_OP_TYPE_OFFSET,e->op_type);
812         SIVAL(msg,OP_BREAK_MSG_ACCESS_MASK_OFFSET,e->access_mask);
813         SIVAL(msg,OP_BREAK_MSG_SHARE_ACCESS_OFFSET,e->share_access);
814         SIVAL(msg,OP_BREAK_MSG_PRIV_OFFSET,e->private_options);
815         SIVAL(msg,OP_BREAK_MSG_TIME_SEC_OFFSET,(uint32_t)e->time.tv_sec);
816         SIVAL(msg,OP_BREAK_MSG_TIME_USEC_OFFSET,(uint32_t)e->time.tv_usec);
817         push_file_id_24(msg+OP_BREAK_MSG_DEV_OFFSET, &e->id);
818         SIVAL(msg,OP_BREAK_MSG_FILE_ID_OFFSET,e->share_file_id);
819         SIVAL(msg,OP_BREAK_MSG_UID_OFFSET,e->uid);
820         SSVAL(msg,OP_BREAK_MSG_FLAGS_OFFSET,e->flags);
821         SIVAL(msg,OP_BREAK_MSG_NAME_HASH_OFFSET,e->name_hash);
822         SIVAL(msg,OP_BREAK_MSG_VNN_OFFSET,e->pid.vnn);
823 }
824
825 /****************************************************************************
826  De-linearize an internal oplock break message to a share mode entry struct.
827 ****************************************************************************/
828
829 void message_to_share_mode_entry(struct share_mode_entry *e, char *msg)
830 {
831         e->pid.pid = (pid_t)IVAL(msg,OP_BREAK_MSG_PID_OFFSET);
832         e->op_mid = BVAL(msg,OP_BREAK_MSG_MID_OFFSET);
833         e->op_type = SVAL(msg,OP_BREAK_MSG_OP_TYPE_OFFSET);
834         e->access_mask = IVAL(msg,OP_BREAK_MSG_ACCESS_MASK_OFFSET);
835         e->share_access = IVAL(msg,OP_BREAK_MSG_SHARE_ACCESS_OFFSET);
836         e->private_options = IVAL(msg,OP_BREAK_MSG_PRIV_OFFSET);
837         e->time.tv_sec = (time_t)IVAL(msg,OP_BREAK_MSG_TIME_SEC_OFFSET);
838         e->time.tv_usec = (int)IVAL(msg,OP_BREAK_MSG_TIME_USEC_OFFSET);
839         pull_file_id_24(msg+OP_BREAK_MSG_DEV_OFFSET, &e->id);
840         e->share_file_id = (unsigned long)IVAL(msg,OP_BREAK_MSG_FILE_ID_OFFSET);
841         e->uid = (uint32)IVAL(msg,OP_BREAK_MSG_UID_OFFSET);
842         e->flags = (uint16)SVAL(msg,OP_BREAK_MSG_FLAGS_OFFSET);
843         e->name_hash = IVAL(msg,OP_BREAK_MSG_NAME_HASH_OFFSET);
844         e->pid.vnn = IVAL(msg,OP_BREAK_MSG_VNN_OFFSET);
845 }
846
847 /****************************************************************************
848  Setup oplocks for this process.
849 ****************************************************************************/
850
851 bool init_oplocks(struct smbd_server_connection *sconn)
852 {
853         DEBUG(3,("init_oplocks: initializing messages.\n"));
854
855         messaging_register(sconn->msg_ctx, sconn, MSG_SMB_BREAK_REQUEST,
856                            process_oplock_break_message);
857         messaging_register(sconn->msg_ctx, sconn, MSG_SMB_ASYNC_LEVEL2_BREAK,
858                            process_oplock_async_level2_break_message);
859         messaging_register(sconn->msg_ctx, sconn, MSG_SMB_KERNEL_BREAK,
860                            process_kernel_oplock_break);
861         return true;
862 }
863
864 void init_kernel_oplocks(struct smbd_server_connection *sconn)
865 {
866         struct kernel_oplocks *koplocks = sconn->oplocks.kernel_ops;
867
868         /* only initialize once */
869         if (koplocks == NULL) {
870 #if HAVE_KERNEL_OPLOCKS_IRIX
871                 koplocks = irix_init_kernel_oplocks(sconn);
872 #elif HAVE_KERNEL_OPLOCKS_LINUX
873                 koplocks = linux_init_kernel_oplocks(sconn);
874 #endif
875                 sconn->oplocks.kernel_ops = koplocks;
876         }
877 }