r8219: Merge the new open code from HEAD to 3.0. Haven't yet run the torture
[samba.git] / source / 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    
7    This program is free software; you can redistribute it and/or modify
8    it under the terms of the GNU General Public License as published by
9    the Free Software Foundation; either version 2 of the License, or
10    (at your option) any later version.
11    
12    This program is distributed in the hope that it will be useful,
13    but WITHOUT ANY WARRANTY; without even the implied warranty of
14    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15    GNU General Public License for more details.
16    
17    You should have received a copy of the GNU General Public License
18    along with this program; if not, write to the Free Software
19    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 */
21
22 #include "includes.h"
23
24 /* Oplock ipc UDP socket. */
25 static int oplock_sock = -1;
26 uint16 global_oplock_port = 0;
27
28 /* Current number of oplocks we have outstanding. */
29 static int32 exclusive_oplocks_open = 0;
30 static int32 level_II_oplocks_open = 0;
31 BOOL global_client_failed_oplock_break = False;
32 BOOL global_oplock_break = False;
33
34 extern struct timeval smb_last_time;
35 extern uint32 global_client_caps;
36 extern struct current_user current_user;
37 extern int smb_read_error;
38
39 static struct kernel_oplocks *koplocks;
40
41 static BOOL oplock_break(SMB_DEV_T dev, SMB_INO_T inode, unsigned long file_id, BOOL local);
42
43 /****************************************************************************
44  Get the number of current exclusive oplocks.
45 ****************************************************************************/
46
47 int32 get_number_of_exclusive_open_oplocks(void)
48 {
49   return exclusive_oplocks_open;
50 }
51
52 /****************************************************************************
53  Return True if an oplock message is pending.
54 ****************************************************************************/
55
56 BOOL oplock_message_waiting(fd_set *fds)
57 {
58         if (koplocks && koplocks->msg_waiting(fds))
59                 return True;
60
61         if (FD_ISSET(oplock_sock, fds))
62                 return True;
63
64         return False;
65 }
66
67 /****************************************************************************
68  Read an oplock break message from either the oplock UDP fd or the
69  kernel (if kernel oplocks are supported).
70
71  If timeout is zero then *fds contains the file descriptors that
72  are ready to be read and acted upon. If timeout is non-zero then
73  *fds contains the file descriptors to be selected on for read.
74  The timeout is in milliseconds
75
76 ****************************************************************************/
77
78 BOOL receive_local_message( char *buffer, int buffer_len, int timeout)
79 {
80         struct sockaddr_in from;
81         socklen_t fromlen = sizeof(from);
82         int32 msg_len = 0;
83         fd_set fds;
84         int selrtn = -1;
85
86         FD_ZERO(&fds);
87         smb_read_error = 0;
88
89         /*
90          * We need to check for kernel oplocks before going into the select
91          * here, as the EINTR generated by the linux kernel oplock may have
92          * already been eaten. JRA.
93          */
94
95         if (koplocks && koplocks->msg_waiting(&fds)) {
96                 return koplocks->receive_message(&fds, buffer, buffer_len);
97         }
98
99         while (timeout > 0 && selrtn == -1) {
100                 struct timeval to;
101                 int maxfd = oplock_sock;
102                 time_t starttime = time(NULL);
103
104                 FD_ZERO(&fds);
105                 maxfd = setup_oplock_select_set(&fds);
106
107                 to.tv_sec = timeout / 1000;
108                 to.tv_usec = (timeout % 1000) * 1000;
109
110                 DEBUG(5,("receive_local_message: doing select with timeout of %d ms\n", timeout));
111
112                 selrtn = sys_select(maxfd+1,&fds,NULL,NULL,&to);
113
114                 if (selrtn == -1 && errno == EINTR) {
115
116                         /* could be a kernel oplock interrupt */
117                         if (koplocks && koplocks->msg_waiting(&fds)) {
118                                 return koplocks->receive_message(&fds, buffer, buffer_len);
119                         }
120
121                         /*
122                          * Linux 2.0.x seems to have a bug in that
123                          * it can return -1, EINTR with a timeout of zero.
124                          * Make sure we bail out here with a read timeout
125                          * if we got EINTR on a timeout of 1 or less.
126                          */
127
128                         if (timeout <= 1) {
129                                 smb_read_error = READ_TIMEOUT;
130                                 return False;
131                         }
132
133                         /* Not a kernel interrupt - could be a SIGUSR1 message. We must restart. */
134                         /* We need to decrement the timeout here. */
135                         timeout -= ((time(NULL) - starttime)*1000);
136                         if (timeout < 0)
137                                 timeout = 1;
138
139                         DEBUG(5,("receive_local_message: EINTR : new timeout %d ms\n", timeout));
140                         continue;
141                 }
142
143                 /* Check if error */
144                 if(selrtn == -1) {
145                         /* something is wrong. Maybe the socket is dead? */
146                         smb_read_error = READ_ERROR;
147                         return False;
148                 }
149
150                 /* Did we timeout ? */
151                 if (selrtn == 0) {
152                         smb_read_error = READ_TIMEOUT;
153                         return False;
154                 }
155         }
156
157         if (koplocks && koplocks->msg_waiting(&fds)) {
158                 return koplocks->receive_message(&fds, buffer, buffer_len);
159         }
160
161         if (!FD_ISSET(oplock_sock, &fds))
162                 return False;
163
164         /*
165          * From here down we deal with the smbd <--> smbd
166          * oplock break protocol only.
167          */
168
169         /*
170          * Read a loopback udp message.
171          */
172         msg_len = sys_recvfrom(oplock_sock, &buffer[OPBRK_CMD_HEADER_LEN],
173                                                 buffer_len - OPBRK_CMD_HEADER_LEN, 0, (struct sockaddr *)&from, &fromlen);
174
175         if(msg_len < 0) {
176                 DEBUG(0,("receive_local_message. Error in recvfrom. (%s).\n",strerror(errno)));
177                 return False;
178         }
179
180         /* Validate message length. */
181         if(msg_len > (buffer_len - OPBRK_CMD_HEADER_LEN)) {
182                 DEBUG(0,("receive_local_message: invalid msg_len (%d) max can be %d\n", msg_len,
183                         buffer_len  - OPBRK_CMD_HEADER_LEN));
184                 return False;
185         }
186
187         /* Validate message from address (must be localhost). */
188         if(from.sin_addr.s_addr != htonl(INADDR_LOOPBACK)) {
189                 DEBUG(0,("receive_local_message: invalid 'from' address \
190 (was %lx should be 127.0.0.1)\n", (long)from.sin_addr.s_addr));
191                 return False;
192         }
193
194         /* Setup the message header */
195         SIVAL(buffer,OPBRK_CMD_LEN_OFFSET,msg_len);
196         SSVAL(buffer,OPBRK_CMD_PORT_OFFSET,ntohs(from.sin_port));
197
198         return True;
199 }
200
201 /****************************************************************************
202  Attempt to set an oplock on a file. Always succeeds if kernel oplocks are
203  disabled (just sets flags). Returns True if oplock set.
204 ****************************************************************************/
205
206 BOOL set_file_oplock(files_struct *fsp, int oplock_type)
207 {
208         if (koplocks && !koplocks->set_oplock(fsp, oplock_type))
209                 return False;
210
211         fsp->oplock_type = oplock_type;
212         fsp->sent_oplock_break = NO_BREAK_SENT;
213         if (oplock_type == LEVEL_II_OPLOCK)
214                 level_II_oplocks_open++;
215         else
216                 exclusive_oplocks_open++;
217
218         DEBUG(5,("set_file_oplock: granted oplock on file %s, dev = %x, inode = %.0f, file_id = %lu, \
219 tv_sec = %x, tv_usec = %x\n",
220                  fsp->fsp_name, (unsigned int)fsp->dev, (double)fsp->inode, fsp->file_id,
221                  (int)fsp->open_time.tv_sec, (int)fsp->open_time.tv_usec ));
222
223         return True;
224 }
225
226 /****************************************************************************
227  Attempt to release an oplock on a file. Decrements oplock count.
228 ****************************************************************************/
229
230 void release_file_oplock(files_struct *fsp)
231 {
232         if ((fsp->oplock_type != NO_OPLOCK) && koplocks)
233                 koplocks->release_oplock(fsp);
234
235         if (fsp->oplock_type == LEVEL_II_OPLOCK)
236                 level_II_oplocks_open--;
237         else if (fsp->oplock_type)
238                 exclusive_oplocks_open--;
239         
240         fsp->oplock_type = NO_OPLOCK;
241         fsp->sent_oplock_break = NO_BREAK_SENT;
242         
243         flush_write_cache(fsp, OPLOCK_RELEASE_FLUSH);
244 }
245
246 /****************************************************************************
247  Attempt to downgrade an oplock on a file. Doesn't decrement oplock count.
248 ****************************************************************************/
249
250 static void downgrade_file_oplock(files_struct *fsp)
251 {
252         if (koplocks)
253                 koplocks->release_oplock(fsp);
254         fsp->oplock_type = LEVEL_II_OPLOCK;
255         exclusive_oplocks_open--;
256         level_II_oplocks_open++;
257         fsp->sent_oplock_break = NO_BREAK_SENT;
258 }
259
260 /****************************************************************************
261  Remove a file oplock. Copes with level II and exclusive.
262  Locks then unlocks the share mode lock. Client can decide to go directly
263  to none even if a "break-to-level II" was sent.
264 ****************************************************************************/
265
266 BOOL remove_oplock(files_struct *fsp, BOOL break_to_none)
267 {
268         SMB_DEV_T dev = fsp->dev;
269         SMB_INO_T inode = fsp->inode;
270         BOOL ret = True;
271
272         /* Remove the oplock flag from the sharemode. */
273         if (lock_share_entry_fsp(fsp) == False) {
274                 DEBUG(0,("remove_oplock: failed to lock share entry for file %s\n",
275                          fsp->fsp_name ));
276                 return False;
277         }
278
279         if (fsp->sent_oplock_break == EXCLUSIVE_BREAK_SENT || break_to_none) {
280                 /*
281                  * Deal with a reply when a break-to-none was sent.
282                  */
283
284                 if(remove_share_oplock(fsp)==False) {
285                         DEBUG(0,("remove_oplock: failed to remove share oplock for file %s fnum %d, \
286 dev = %x, inode = %.0f\n", fsp->fsp_name, fsp->fnum, (unsigned int)dev, (double)inode));
287                         ret = False;
288                 }
289
290                 release_file_oplock(fsp);
291         } else {
292                 /*
293                  * Deal with a reply when a break-to-level II was sent.
294                  */
295                 if(downgrade_share_oplock(fsp)==False) {
296                         DEBUG(0,("remove_oplock: failed to downgrade share oplock for file %s fnum %d, \
297 dev = %x, inode = %.0f\n", fsp->fsp_name, fsp->fnum, (unsigned int)dev, (double)inode));
298                         ret = False;
299                 }
300                 
301                 downgrade_file_oplock(fsp);
302         }
303
304         unlock_share_entry_fsp(fsp);
305         return ret;
306 }
307
308 /****************************************************************************
309  Setup the listening set of file descriptors for an oplock break
310  message either from the UDP socket or from the kernel. Returns the maximum
311  fd used.
312 ****************************************************************************/
313
314 int setup_oplock_select_set( fd_set *fds)
315 {
316         int maxfd = oplock_sock;
317
318         if(oplock_sock == -1)
319                 return 0;
320
321         FD_SET(oplock_sock,fds);
322
323         if (koplocks && koplocks->notification_fd != -1) {
324                 FD_SET(koplocks->notification_fd, fds);
325                 maxfd = MAX(maxfd, koplocks->notification_fd);
326         }
327
328         return maxfd;
329 }
330
331 /****************************************************************************
332  Process an oplock break message - whether it came from the UDP socket
333  or from the kernel.
334 ****************************************************************************/
335
336 BOOL process_local_message(char *buffer, int buf_size)
337 {
338         int32 msg_len;
339         uint16 from_port;
340         char *msg_start;
341         pid_t remotepid;
342         SMB_DEV_T dev;
343         SMB_INO_T inode;
344         unsigned long file_id;
345         uint16 break_cmd_type;
346         struct sockaddr_in toaddr;
347
348         msg_len = IVAL(buffer,OPBRK_CMD_LEN_OFFSET);
349         from_port = SVAL(buffer,OPBRK_CMD_PORT_OFFSET);
350
351         msg_start = &buffer[OPBRK_CMD_HEADER_LEN];
352
353         DEBUG(5,("process_local_message: Got a message of length %d from port (%d)\n", 
354                 msg_len, from_port));
355
356         /* 
357          * Pull the info out of the requesting packet.
358          */
359
360         break_cmd_type = SVAL(msg_start,OPBRK_MESSAGE_CMD_OFFSET);
361
362         switch(break_cmd_type) {
363                 case KERNEL_OPLOCK_BREAK_CMD:
364                         if (!koplocks) {
365                                 DEBUG(0,("unexpected kernel oplock break!\n"));
366                                 break;
367                         } 
368                         if (!koplocks->parse_message(msg_start, msg_len, &inode, &dev, &file_id)) {
369                                 DEBUG(0,("kernel oplock break parse failure!\n"));
370                                 return False;
371                         }
372                         break;
373
374                 case OPLOCK_BREAK_CMD:
375                 case LEVEL_II_OPLOCK_BREAK_CMD:
376                 case ASYNC_LEVEL_II_OPLOCK_BREAK_CMD:
377
378                         /* Ensure that the msg length is correct. */
379                         if(msg_len != OPLOCK_BREAK_MSG_LEN) {
380                                 DEBUG(0,("process_local_message: incorrect length for OPLOCK_BREAK_CMD (was %d, should be %d).\n",
381                                         (int)msg_len, (int)OPLOCK_BREAK_MSG_LEN));
382                                 return False;
383                         }
384
385                         memcpy((char *)&remotepid, msg_start+OPLOCK_BREAK_PID_OFFSET,sizeof(remotepid));
386                         memcpy((char *)&inode, msg_start+OPLOCK_BREAK_INODE_OFFSET,sizeof(inode));
387                         memcpy((char *)&dev, msg_start+OPLOCK_BREAK_DEV_OFFSET,sizeof(dev));
388                         memcpy((char *)&file_id, msg_start+OPLOCK_BREAK_FILEID_OFFSET,sizeof(file_id));
389
390                         DEBUG(5,("process_local_message: (%s) oplock break request from \
391 pid %d, port %d, dev = %x, inode = %.0f, file_id = %lu\n",
392                                 (break_cmd_type == OPLOCK_BREAK_CMD) ? "exclusive" : "level II",
393                                 (int)remotepid, from_port, (unsigned int)dev, (double)inode, file_id));
394                         break;
395
396                 case RETRY_DEFERRED_OPEN_CMD:
397
398                         /* Request to retry and open that would return SHARING_VIOLATION. */
399                         if (msg_len != DEFERRED_OPEN_MSG_LEN) {
400                                 DEBUG(0,("process_local_message: incorrect length for RETRY_DEFERRED_OPEN_CMD (was %d, should be %d).\n",
401                                         (int)msg_len, (int)DEFERRED_OPEN_MSG_LEN));
402                                 return False;
403                         }
404                         {
405                                 uint16 mid;
406
407                                 memcpy((char *)&remotepid, msg_start+DEFERRED_OPEN_PID_OFFSET,sizeof(remotepid));
408                                 memcpy((char *)&inode, msg_start+DEFERRED_OPEN_INODE_OFFSET,sizeof(inode));
409                                 memcpy((char *)&dev, msg_start+DEFERRED_OPEN_DEV_OFFSET,sizeof(dev));
410                                 memcpy((char *)&mid, msg_start+DEFERRED_OPEN_MID_OFFSET,sizeof(mid));
411
412                                 DEBUG(5,("process_local_message: RETRY_DEFERRED_OPEN from \
413 pid %d, port %d, dev = %x, inode = %.0f, mid = %u\n",
414                                         (int)remotepid, from_port, (unsigned int)dev, (double)inode, (unsigned int)mid));
415
416                                 schedule_sharing_violation_open_smb_message(mid);
417                         }
418                         return True;
419
420                 /* 
421                  * Keep this as a debug case - eventually we can remove it.
422                  */
423                 case 0x8001:
424                         DEBUG(0,("process_local_message: Received unsolicited break \
425 reply - dumping info.\n"));
426
427                         if(msg_len != OPLOCK_BREAK_MSG_LEN) {
428                                 DEBUG(0,("process_local_message: ubr: incorrect length for reply \
429 (was %d, should be %d).\n", (int)msg_len, (int)OPLOCK_BREAK_MSG_LEN));
430                                 return False;
431                         }
432
433                         memcpy((char *)&inode, msg_start+OPLOCK_BREAK_INODE_OFFSET,sizeof(inode));
434                         memcpy((char *)&remotepid, msg_start+OPLOCK_BREAK_PID_OFFSET,sizeof(remotepid));
435                         memcpy((char *)&dev, msg_start+OPLOCK_BREAK_DEV_OFFSET,sizeof(dev));
436                         memcpy((char *)&file_id, msg_start+OPLOCK_BREAK_FILEID_OFFSET,sizeof(file_id));
437
438                         DEBUG(0,("process_local_message: unsolicited oplock break reply from \
439 pid %d, port %d, dev = %x, inode = %.0f, file_id = %lu\n",
440                                 (int)remotepid, from_port, (unsigned int)dev, (double)inode, file_id));
441
442                         return False;
443
444                 default:
445                         DEBUG(0,("process_local_message: unknown UDP message command code (%x) - ignoring.\n",
446                                 (unsigned int)SVAL(msg_start,0)));
447                         return False;
448         }
449
450         /*
451          * Now actually process the break request.
452          */
453
454         if ((exclusive_oplocks_open == 0) &&
455             (level_II_oplocks_open == 0)) {
456                 /*
457                  * If we have no record of any currently open oplocks,
458                  * it's not an error, as a close command may have
459                  * just been issued on the file that was oplocked.
460                  * Just log a message and return success in this case.
461                  */
462                 DEBUG(3,("process_local_message: oplock break requested with "
463                          "no outstanding oplocks. Returning success.\n"));
464                 return True;
465         }
466
467         if (!oplock_break(dev, inode, file_id, False)) {
468                 DEBUG(0,("process_local_message: oplock break failed.\n"));
469                 return False;
470         }
471
472         /* 
473          * Do the appropriate reply - none in the kernel or async level II
474          * case.
475          */
476
477         if (!((break_cmd_type == OPLOCK_BREAK_CMD) ||
478               (break_cmd_type == LEVEL_II_OPLOCK_BREAK_CMD))) {
479                 return True;
480         }
481
482         /* Send the message back after OR'ing in the 'REPLY' bit. */
483         SSVAL(msg_start,OPBRK_MESSAGE_CMD_OFFSET,break_cmd_type | CMD_REPLY);
484
485         memset((char *)&toaddr,'\0',sizeof(toaddr));
486         toaddr.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
487         toaddr.sin_port = htons(from_port);
488         toaddr.sin_family = AF_INET;
489
490         if(sys_sendto( oplock_sock, msg_start, OPLOCK_BREAK_MSG_LEN, 0,
491                        (struct sockaddr *)&toaddr, sizeof(toaddr)) < 0) {
492                 DEBUG(0,("process_local_message: sendto process %d failed. "
493                          "Errno was %s\n", (int)remotepid, strerror(errno)));
494                 return False;
495         }
496
497         DEBUG(5,("process_local_message: oplock break reply sent to pid %d, "
498                  "port %d, for file dev = %x, inode = %.0f, file_id = %lu\n",
499                  (int)remotepid, from_port, (unsigned int)dev,
500                  (double)inode, file_id));
501
502         return True;
503 }
504
505 /****************************************************************************
506  Set up an oplock break message.
507 ****************************************************************************/
508
509 static void prepare_break_message(char *outbuf, files_struct *fsp, BOOL level2)
510 {
511         memset(outbuf,'\0',smb_size);
512         set_message(outbuf,8,0,True);
513
514         SCVAL(outbuf,smb_com,SMBlockingX);
515         SSVAL(outbuf,smb_tid,fsp->conn->cnum);
516         SSVAL(outbuf,smb_pid,0xFFFF);
517         SSVAL(outbuf,smb_uid,0);
518         SSVAL(outbuf,smb_mid,0xFFFF);
519         SCVAL(outbuf,smb_vwv0,0xFF);
520         SSVAL(outbuf,smb_vwv2,fsp->fnum);
521         SCVAL(outbuf,smb_vwv3,LOCKING_ANDX_OPLOCK_RELEASE);
522         SCVAL(outbuf,smb_vwv3+1,level2 ? OPLOCKLEVEL_II : OPLOCKLEVEL_NONE);
523 }
524
525 /****************************************************************************
526  Function to do the waiting before sending a local break.
527 ****************************************************************************/
528
529 static void wait_before_sending_break(BOOL local_request)
530 {
531         if(local_request) {
532                 struct timeval cur_tv;
533                 long wait_left = (long)lp_oplock_break_wait_time();
534
535                 if (wait_left == 0)
536                         return;
537
538                 GetTimeOfDay(&cur_tv);
539
540                 wait_left -= ((cur_tv.tv_sec - smb_last_time.tv_sec)*1000) +
541                 ((cur_tv.tv_usec - smb_last_time.tv_usec)/1000);
542
543                 if(wait_left > 0) {
544                         wait_left = MIN(wait_left, 1000);
545                         sys_usleep(wait_left * 1000);
546                 }
547         }
548 }
549
550 /****************************************************************************
551  Ensure that we have a valid oplock.
552 ****************************************************************************/
553
554 static files_struct *initial_break_processing(SMB_DEV_T dev, SMB_INO_T inode, unsigned long file_id)
555 {
556         files_struct *fsp = NULL;
557
558         if( DEBUGLVL( 3 ) ) {
559                 dbgtext( "initial_break_processing: called for dev = %x, inode = %.0f file_id = %lu\n",
560                         (unsigned int)dev, (double)inode, file_id);
561                 dbgtext( "Current oplocks_open (exclusive = %d, levelII = %d)\n",
562                         exclusive_oplocks_open, level_II_oplocks_open );
563         }
564
565         /*
566          * We need to search the file open table for the
567          * entry containing this dev and inode, and ensure
568          * we have an oplock on it.
569          */
570
571         fsp = file_find_dif(dev, inode, file_id);
572
573         if(fsp == NULL) {
574                 /* The file could have been closed in the meantime - return success. */
575                 if( DEBUGLVL( 3 ) ) {
576                         dbgtext( "initial_break_processing: cannot find open file with " );
577                         dbgtext( "dev = %x, inode = %.0f file_id = %lu", (unsigned int)dev,
578                                 (double)inode, file_id);
579                         dbgtext( "allowing break to succeed.\n" );
580                 }
581                 return NULL;
582         }
583
584         /* Ensure we have an oplock on the file */
585
586         /*
587          * There is a potential race condition in that an oplock could
588          * have been broken due to another udp request, and yet there are
589          * still oplock break messages being sent in the udp message
590          * queue for this file. So return true if we don't have an oplock,
591          * as we may have just freed it.
592          */
593
594         if(fsp->oplock_type == NO_OPLOCK) {
595                 if( DEBUGLVL( 3 ) ) {
596                         dbgtext( "initial_break_processing: file %s ", fsp->fsp_name );
597                         dbgtext( "(dev = %x, inode = %.0f, file_id = %lu) has no oplock.\n",
598                                 (unsigned int)dev, (double)inode, fsp->file_id );
599                         dbgtext( "Allowing break to succeed regardless.\n" );
600                 }
601                 return NULL;
602         }
603
604         return fsp;
605 }
606
607 /****************************************************************************
608  Process a level II oplock break directly.
609  We must call this function with the share mode entry locked.
610 ****************************************************************************/
611
612 static BOOL oplock_break_level2(files_struct *fsp, BOOL local_request)
613 {
614         char outbuf[128];
615         SMB_DEV_T dev = fsp->dev;
616         SMB_INO_T inode = fsp->inode;
617
618         /*
619          * We can have a level II oplock even if the client is not
620          * level II oplock aware. In this case just remove the
621          * flags and don't send the break-to-none message to
622          * the client.
623          */
624
625         if (global_client_caps & CAP_LEVEL_II_OPLOCKS) {
626                 BOOL sign_state;
627
628                 /*
629                  * If we are sending an oplock break due to an SMB sent
630                  * by our own client we ensure that we wait at leat
631                  * lp_oplock_break_wait_time() milliseconds before sending
632                  * the packet. Sending the packet sooner can break Win9x
633                  * and has reported to cause problems on NT. JRA.
634                  */
635
636                 wait_before_sending_break(local_request);
637
638                 /* Prepare the SMBlockingX message. */
639                 prepare_break_message( outbuf, fsp, False);
640
641                 /* Save the server smb signing state. */
642                 sign_state = srv_oplock_set_signing(False);
643
644                 show_msg(outbuf);
645                 if (!send_smb(smbd_server_fd(), outbuf))
646                         exit_server("oplock_break_level2: send_smb failed.");
647
648                 /* Restore the sign state to what it was. */
649                 srv_oplock_set_signing(sign_state);
650         }
651
652         /*
653          * Now we must update the shared memory structure to tell
654          * everyone else we no longer have a level II oplock on 
655          * this open file. We must call this function with the share mode
656          * entry locked so we can change the entry directly.
657          */
658
659         if(remove_share_oplock(fsp)==False) {
660                 DEBUG(0,("oplock_break_level2: unable to remove level II oplock for file %s\n", fsp->fsp_name ));
661         }
662
663         release_file_oplock(fsp);
664
665         if(level_II_oplocks_open < 0) {
666                 DEBUG(0,("oplock_break_level2: level_II_oplocks_open < 0 (%d). PANIC ERROR\n",
667                         level_II_oplocks_open));
668                 abort();
669         }
670
671         if( DEBUGLVL( 3 ) ) {
672                 dbgtext( "oplock_break_level2: returning success for " );
673                 dbgtext( "dev = %x, inode = %.0f, file_id = %lu\n", (unsigned int)dev, (double)inode, fsp->file_id );
674                 dbgtext( "Current level II oplocks_open = %d\n", level_II_oplocks_open );
675         }
676
677         return True;
678 }
679
680 /****************************************************************************
681  Process an oplock break directly.
682  This is always called with the share mode lock *NOT* held.
683 ****************************************************************************/
684
685 static BOOL oplock_break(SMB_DEV_T dev, SMB_INO_T inode, unsigned long file_id, BOOL local_request)
686 {
687         char *inbuf = NULL;
688         char *saved_inbuf = NULL;
689         char *outbuf = NULL;
690         char *saved_outbuf = NULL;
691         files_struct *fsp = NULL;
692         time_t start_time;
693         BOOL shutdown_server = False;
694         BOOL oplock_timeout = False;
695         BOOL sign_state;
696         connection_struct *saved_user_conn;
697         connection_struct *saved_fsp_conn;
698         int saved_vuid;
699         pstring saved_dir; 
700         int timeout = (OPLOCK_BREAK_TIMEOUT * 1000);
701         pstring file_name;
702         BOOL using_levelII;
703
704         if((fsp = initial_break_processing(dev, inode, file_id)) == NULL)
705                 return True;
706
707         /*
708          * Deal with a level II oplock going break to none separately.
709          */
710
711         if (LEVEL_II_OPLOCK_TYPE(fsp->oplock_type)) {
712                 BOOL ret;
713                 /* We must always call oplock_break_level2() with
714                    the share mode entry locked. */
715                 if (lock_share_entry_fsp(fsp) == False) {
716                         DEBUG(0,("oplock_break: unable to lock share entry for file %s\n", fsp->fsp_name ));
717                         return False;
718                 }
719                 ret = oplock_break_level2(fsp, local_request);
720                 unlock_share_entry_fsp(fsp);
721                 return ret;
722         }
723
724         /* Mark the oplock break as sent - we don't want to send twice! */
725         if (fsp->sent_oplock_break) {
726                 if( DEBUGLVL( 0 ) ) {
727                         dbgtext( "oplock_break: ERROR: oplock_break already sent for " );
728                         dbgtext( "file %s ", fsp->fsp_name);
729                         dbgtext( "(dev = %x, inode = %.0f, file_id = %lu)\n", (unsigned int)dev, (double)inode, fsp->file_id );
730                 }
731
732                 /*
733                  * We have to fail the open here as we cannot send another oplock break on
734                  * this file whilst we are awaiting a response from the client - neither
735                  * can we allow another open to succeed while we are waiting for the client.
736                  */
737                 return False;
738         }
739
740         if(global_oplock_break) {
741                 DEBUG(0,("ABORT : ABORT : recursion in oplock_break !!!!!\n"));
742                 abort();
743         }
744
745         /*
746          * Now comes the horrid part. We must send an oplock break to the client,
747          * and then process incoming messages until we get a close or oplock release.
748          * At this point we know we need a new inbuf/outbuf buffer pair.
749          * We cannot use these staticaly as we may recurse into here due to
750          * messages crossing on the wire.
751          */
752
753         if((inbuf = NewInBuffer(&saved_inbuf))==NULL) {
754                 DEBUG(0,("oplock_break: malloc fail for input buffer.\n"));
755                 return False;
756         }
757
758         if((outbuf = NewOutBuffer(&saved_outbuf))==NULL) {
759                 DEBUG(0,("oplock_break: malloc fail for output buffer.\n"));
760                 set_InBuffer(saved_inbuf);
761                 free_InBuffer(inbuf);
762                 return False;
763         }
764
765         /*
766          * If we are sending an oplock break due to an SMB sent
767          * by our own client we ensure that we wait at leat
768          * lp_oplock_break_wait_time() milliseconds before sending
769          * the packet. Sending the packet sooner can break Win9x
770          * and has reported to cause problems on NT. JRA.
771          */
772
773         wait_before_sending_break(local_request);
774
775         /* Prepare the SMBlockingX message. */
776
777         if ((global_client_caps & CAP_LEVEL_II_OPLOCKS) && 
778                         !koplocks && /* NOTE: we force levelII off for kernel oplocks - this will change when it is supported */
779                         lp_level2_oplocks(SNUM(fsp->conn))) {
780                 using_levelII = True;
781         } else {
782                 using_levelII = False;
783         }
784
785         prepare_break_message( outbuf, fsp, using_levelII);
786         /* Remember if we just sent a break to level II on this file. */
787         fsp->sent_oplock_break = using_levelII? LEVEL_II_BREAK_SENT:EXCLUSIVE_BREAK_SENT;
788
789         /* Save the server smb signing state. */
790         sign_state = srv_oplock_set_signing(False);
791
792         show_msg(outbuf);
793         if (!send_smb(smbd_server_fd(), outbuf)) {
794                 srv_oplock_set_signing(sign_state);
795                 exit_server("oplock_break: send_smb failed.");
796         }
797
798         /* Restore the sign state to what it was. */
799         srv_oplock_set_signing(sign_state);
800
801         /* We need this in case a readraw crosses on the wire. */
802         global_oplock_break = True;
803  
804         /* Process incoming messages. */
805
806         /*
807          * JRA - If we don't get a break from the client in OPLOCK_BREAK_TIMEOUT
808          * seconds we should just die....
809          */
810
811         start_time = time(NULL);
812
813         /*
814          * Save the information we need to re-become the
815          * user, then unbecome the user whilst we're doing this.
816          */
817         saved_user_conn = current_user.conn;
818         saved_vuid = current_user.vuid;
819         saved_fsp_conn = fsp->conn;
820         /*
821          * Initialize saved_dir to something sensible: vfs_GetWd may not work well
822          * for root: the directory may be NFS-mounted and exported with root_squash
823          * (so has no root access).
824          */
825         pstrcpy(saved_dir,saved_fsp_conn->connectpath);
826         vfs_GetWd(saved_fsp_conn,saved_dir);
827         /* Save the chain fnum. */
828         file_chain_save();
829
830         pstrcpy(file_name, fsp->fsp_name);
831
832         change_to_root_user();
833
834         /*
835          * From Charles Hoch <hoch@exemplary.com>. If the break processing
836          * code closes the file (as it often does), then the fsp pointer here
837          * points to free()'d memory. We *must* revalidate fsp each time
838          * around the loop. With async I/O, write calls may steal the global InBuffer,
839          * so ensure we're using the correct one each time around the loop.
840          */
841
842         while((fsp = initial_break_processing(dev, inode, file_id)) &&
843                         OPEN_FSP(fsp) && EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
844
845                 inbuf = get_InBuffer();
846                 outbuf = get_OutBuffer();
847
848                 if(receive_smb(smbd_server_fd(),inbuf, timeout) == False) {
849                         /*
850                          * Die if we got an error.
851                          */
852
853                         if (smb_read_error == READ_EOF) {
854                                 DEBUG( 0, ( "oplock_break: end of file from client\n" ) );
855                                 shutdown_server = True;
856                         } else if (smb_read_error == READ_ERROR) {
857                                 DEBUG( 0, ("oplock_break: receive_smb error (%s)\n", strerror(errno)) );
858                                 shutdown_server = True;
859                         } else if (smb_read_error == READ_BAD_SIG) {
860                                 DEBUG( 0, ("oplock_break: bad signature from client\n" ));
861                                 shutdown_server = True;
862                         } else if (smb_read_error == READ_TIMEOUT) {
863                                 DEBUG( 0, ( "oplock_break: receive_smb timed out after %d seconds.\n", OPLOCK_BREAK_TIMEOUT ) );
864                                 oplock_timeout = True;
865                         }
866
867                         DEBUGADD( 0, ( "oplock_break failed for file %s ", file_name ) );
868                         DEBUGADD( 0, ( "(dev = %x, inode = %.0f, file_id = %lu).\n",
869                                 (unsigned int)dev, (double)inode, file_id));
870
871                         break;
872                 }
873
874                 /*
875                  * There are certain SMB requests that we shouldn't allow
876                  * to recurse. opens, renames and deletes are the obvious
877                  * ones. This is handled in the switch_message() function.
878                  * If global_oplock_break is set they will push the packet onto
879                  * the pending smb queue and return -1 (no reply).
880                  * JRA.
881                  */
882
883                 process_smb(inbuf, outbuf);
884
885                 /*
886                  * Die if we go over the time limit.
887                  */
888
889                 if((time(NULL) - start_time) > OPLOCK_BREAK_TIMEOUT) {
890                         if( DEBUGLVL( 0 ) ) {
891                                 dbgtext( "oplock_break: no break received from client " );
892                                 dbgtext( "within %d seconds.\n", OPLOCK_BREAK_TIMEOUT );
893                                 dbgtext( "oplock_break failed for file %s ", fsp->fsp_name );
894                                 dbgtext( "(dev = %x, inode = %.0f, file_id = %lu).\n",  
895                                         (unsigned int)dev, (double)inode, file_id );
896                         }
897                         oplock_timeout = True;
898                         break;
899                 }
900         }
901
902         /*
903          * Go back to being the user who requested the oplock
904          * break.
905          */
906         if((saved_user_conn != NULL) && (saved_vuid != UID_FIELD_INVALID) && !change_to_user(saved_user_conn, saved_vuid)) {
907                 DEBUG( 0, ( "oplock_break: unable to re-become user!" ) );
908                 DEBUGADD( 0, ( "Shutting down server\n" ) );
909                 close(oplock_sock);
910                 exit_server("unable to re-become user");
911         }
912
913         /* Including the directory. */
914         vfs_ChDir(saved_fsp_conn,saved_dir);
915
916         /* Restore the chain fnum. */
917         file_chain_restore();
918
919         /* Restore the global In/Out buffers. */
920         set_InBuffer(saved_inbuf);
921         set_OutBuffer(saved_outbuf);
922
923         /* Free the buffers we've been using to recurse. */
924         free_InBuffer(inbuf);
925         free_OutBuffer(outbuf);
926
927         /* We need this in case a readraw crossed on the wire. */
928         if(global_oplock_break)
929                 global_oplock_break = False;
930
931         /*
932          * If the client timed out then clear the oplock (or go to level II)
933          * and continue. This seems to be what NT does and is better than dropping
934          * the connection.
935          */
936
937         if(oplock_timeout && (fsp = initial_break_processing(dev, inode, file_id)) &&
938                         OPEN_FSP(fsp) && EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
939                 DEBUG(0,("oplock_break: client failure in oplock break in file %s\n", fsp->fsp_name));
940                 remove_oplock(fsp,True);
941 #if FASCIST_OPLOCK_BACKOFF
942                 global_client_failed_oplock_break = True; /* Never grant this client an oplock again. */
943 #endif
944         }
945
946         /*
947          * If the client had an error we must die.
948          */
949
950         if(shutdown_server) {
951                 DEBUG( 0, ( "oplock_break: client failure in break - " ) );
952                 DEBUGADD( 0, ( "shutting down this smbd.\n" ) );
953                 close(oplock_sock);
954                 exit_server("oplock break failure");
955         }
956
957         /* Santity check - remove this later. JRA */
958         if(exclusive_oplocks_open < 0) {
959                 DEBUG(0,("oplock_break: exclusive_oplocks_open < 0 (%d). PANIC ERROR\n", exclusive_oplocks_open));
960                 abort();
961         }
962
963         /* We know we have no saved errors here. */
964         set_saved_error_triple(0, 0, NT_STATUS_OK);
965
966         if( DEBUGLVL( 3 ) ) {
967                 dbgtext( "oplock_break: returning success for " );
968                 dbgtext( "dev = %x, inode = %.0f, file_id = %lu\n", (unsigned int)dev, (double)inode, file_id );
969                 dbgtext( "Current exclusive_oplocks_open = %d\n", exclusive_oplocks_open );
970         }
971
972         return True;
973 }
974
975 /****************************************************************************
976  Send an oplock break message to another smbd process. If the oplock is held 
977  by the local smbd then call the oplock break function directly.
978  This function is called with no share locks held.
979 ****************************************************************************/
980
981 BOOL request_oplock_break(share_mode_entry *share_entry)
982 {
983         char op_break_msg[OPLOCK_BREAK_MSG_LEN];
984         struct sockaddr_in addr_out;
985         pid_t pid = sys_getpid();
986         time_t start_time;
987         int time_left;
988         SMB_DEV_T dev = share_entry->dev;
989         SMB_INO_T inode = share_entry->inode;
990         unsigned long file_id = share_entry->share_file_id;
991         uint16 break_cmd_type;
992
993         if(pid == share_entry->pid) {
994                 /* We are breaking our own oplock, make sure it's us. */
995                 if(share_entry->op_port != global_oplock_port) {
996                         DEBUG(0,("request_oplock_break: corrupt share mode entry - pid = %d, port = %d \
997 should be %d\n", (int)pid, share_entry->op_port, global_oplock_port));
998                         return False;
999                 }
1000
1001                 DEBUG(5,("request_oplock_break: breaking our own oplock\n"));
1002
1003 #if 1 /* JRA PARANOIA TEST.... */
1004                 {
1005                         files_struct *fsp = file_find_dif(dev, inode, file_id);
1006                         if (!fsp) {
1007                                 DEBUG(0,("request_oplock_break: PANIC : breaking our own oplock requested for \
1008 dev = %x, inode = %.0f, file_id = %lu and no fsp found !\n",
1009             (unsigned int)dev, (double)inode, file_id ));
1010                                 smb_panic("request_oplock_break: no fsp found for our own oplock\n");
1011                         }
1012                 }
1013 #endif /* END JRA PARANOIA TEST... */
1014
1015                 /* Call oplock break direct. */
1016                 return oplock_break(dev, inode, file_id, True);
1017         }
1018
1019         /* We need to send a OPLOCK_BREAK_CMD message to the port in the share mode entry. */
1020
1021         if (LEVEL_II_OPLOCK_TYPE(share_entry->op_type)) {
1022                 break_cmd_type = LEVEL_II_OPLOCK_BREAK_CMD;
1023         } else {
1024                 break_cmd_type = OPLOCK_BREAK_CMD;
1025         }
1026
1027         SSVAL(op_break_msg,OPBRK_MESSAGE_CMD_OFFSET,break_cmd_type);
1028         memcpy(op_break_msg+OPLOCK_BREAK_PID_OFFSET,(char *)&pid,sizeof(pid));
1029         memcpy(op_break_msg+OPLOCK_BREAK_DEV_OFFSET,(char *)&dev,sizeof(dev));
1030         memcpy(op_break_msg+OPLOCK_BREAK_INODE_OFFSET,(char *)&inode,sizeof(inode));
1031         memcpy(op_break_msg+OPLOCK_BREAK_FILEID_OFFSET,(char *)&file_id,sizeof(file_id));
1032
1033         /* Set the address and port. */
1034         memset((char *)&addr_out,'\0',sizeof(addr_out));
1035         addr_out.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1036         addr_out.sin_port = htons( share_entry->op_port );
1037         addr_out.sin_family = AF_INET;
1038    
1039         if( DEBUGLVL( 3 ) ) {
1040                 dbgtext( "request_oplock_break: sending a synchronous oplock break message to " );
1041                 dbgtext( "pid %d on port %d ", (int)share_entry->pid, share_entry->op_port );
1042                 dbgtext( "for dev = %x, inode = %.0f, file_id = %lu\n",
1043             (unsigned int)dev, (double)inode, file_id );
1044         }
1045
1046         if(sys_sendto(oplock_sock,op_break_msg,OPLOCK_BREAK_MSG_LEN,0,
1047                         (struct sockaddr *)&addr_out,sizeof(addr_out)) < 0) {
1048                 if( DEBUGLVL( 0 ) ) {
1049                         dbgtext( "request_oplock_break: failed when sending a oplock " );
1050                         dbgtext( "break message to pid %d ", (int)share_entry->pid );
1051                         dbgtext( "on port %d ", share_entry->op_port );
1052                         dbgtext( "for dev = %x, inode = %.0f, file_id = %lu\n",
1053                                 (unsigned int)dev, (double)inode, file_id );
1054                         dbgtext( "Error was %s\n", strerror(errno) );
1055                 }
1056                 return False;
1057         }
1058
1059         /*
1060          * Now we must await the oplock broken message coming back
1061          * from the target smbd process. Timeout if it fails to
1062          * return in (OPLOCK_BREAK_TIMEOUT + OPLOCK_BREAK_TIMEOUT_FUDGEFACTOR) seconds.
1063          * While we get messages that aren't ours, loop.
1064          */
1065
1066         start_time = time(NULL);
1067         time_left = OPLOCK_BREAK_TIMEOUT+OPLOCK_BREAK_TIMEOUT_FUDGEFACTOR;
1068
1069         while(time_left >= 0) {
1070                 char op_break_reply[OPBRK_CMD_HEADER_LEN+OPLOCK_BREAK_MSG_LEN];
1071                 uint16 reply_from_port;
1072                 char *reply_msg_start;
1073
1074                 if(receive_local_message(op_break_reply, sizeof(op_break_reply),
1075                                 time_left ? time_left * 1000 : 1) == False) {
1076                         if(smb_read_error == READ_TIMEOUT) {
1077                                 if( DEBUGLVL( 0 ) ) {
1078                                         dbgtext( "request_oplock_break: no response received to oplock " );
1079                                         dbgtext( "break request to pid %d ", (int)share_entry->pid );
1080                                         dbgtext( "on port %d ", share_entry->op_port );
1081                                         dbgtext( "for dev = %x, inode = %.0f, file_id = %lu\n",
1082                                                         (unsigned int)dev, (double)inode, file_id );
1083                                 }
1084
1085                                 /*
1086                                  * This is a hack to make handling of failing clients more robust.
1087                                  * If a oplock break response message is not received in the timeout
1088                                  * period we may assume that the smbd servicing that client holding
1089                                  * the oplock has died and the client changes were lost anyway, so
1090                                  * we should continue to try and open the file.
1091                                  */
1092                                 break;
1093                         } else {
1094                                 if( DEBUGLVL( 0 ) ) {
1095                                         dbgtext( "request_oplock_break: error in response received " );
1096                                         dbgtext( "to oplock break request to pid %d ", (int)share_entry->pid );
1097                                         dbgtext( "on port %d ", share_entry->op_port );
1098                                         dbgtext( "for dev = %x, inode = %.0f, file_id = %lu\n",
1099                                                 (unsigned int)dev, (double)inode, file_id );
1100                                         dbgtext( "Error was (%s).\n", strerror(errno) );
1101                                 }
1102                         }
1103                         return False;
1104                 }
1105
1106                 reply_from_port = SVAL(op_break_reply,OPBRK_CMD_PORT_OFFSET);
1107                 reply_msg_start = &op_break_reply[OPBRK_CMD_HEADER_LEN];
1108
1109                 /*
1110                  * Test to see if this is the reply we are awaiting (ie. the one we sent with the CMD_REPLY flag OR'ed in).
1111                  */
1112                 if((SVAL(reply_msg_start,OPBRK_MESSAGE_CMD_OFFSET) & CMD_REPLY) &&
1113                         ((SVAL(reply_msg_start,OPBRK_MESSAGE_CMD_OFFSET) & ~CMD_REPLY) == break_cmd_type) &&
1114                         (reply_from_port == share_entry->op_port) && 
1115                         (memcmp(&reply_msg_start[OPLOCK_BREAK_PID_OFFSET], &op_break_msg[OPLOCK_BREAK_PID_OFFSET],
1116                                 OPLOCK_BREAK_MSG_LEN - OPLOCK_BREAK_PID_OFFSET) == 0)) {
1117
1118                         /*
1119                          * This is the reply we've been waiting for.
1120                          */
1121                         break;
1122                 } else {
1123                         /*
1124                          * This is another message - a break request.
1125                          * Note that both kernel oplock break requests
1126                          * and UDP inter-smbd oplock break requests will
1127                          * be processed here.
1128                          *
1129                          * Process it to prevent potential deadlock.
1130                          * Note that the code in switch_message() prevents
1131                          * us from recursing into here as any SMB requests
1132                          * we might process that would cause another oplock
1133                          * break request to be made will be queued.
1134                          * JRA.
1135                          */
1136
1137                         process_local_message(op_break_reply, sizeof(op_break_reply));
1138                 }
1139
1140                 time_left -= (time(NULL) - start_time);
1141         }
1142
1143         DEBUG(3,("request_oplock_break: broke oplock.\n"));
1144
1145         return True;
1146 }
1147
1148 /****************************************************************************
1149   Attempt to break an oplock on a file (if oplocked).
1150   Returns True if the file was closed as a result of
1151   the oplock break, False otherwise.
1152   Used as a last ditch attempt to free a space in the 
1153   file table when we have run out.
1154 ****************************************************************************/
1155
1156 BOOL attempt_close_oplocked_file(files_struct *fsp)
1157 {
1158         DEBUG(5,("attempt_close_oplocked_file: checking file %s.\n", fsp->fsp_name));
1159
1160         if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type) && !fsp->sent_oplock_break && (fsp->fh->fd != -1)) {
1161                 /* Try and break the oplock. */
1162                 if (oplock_break(fsp->dev, fsp->inode, fsp->file_id, True)) {
1163                         if(file_find_fsp(fsp) == NULL) /* Did the oplock break close the file ? */
1164                                 return True;
1165                 }
1166         }
1167
1168         return False;
1169 }
1170
1171 /****************************************************************************
1172  Send an asynchronous oplock break message to another smbd process.
1173 ****************************************************************************/
1174
1175 static BOOL request_remote_level2_async_oplock_break(share_mode_entry *share_entry)
1176 {
1177         char op_break_msg[OPLOCK_BREAK_MSG_LEN];
1178         struct sockaddr_in addr_out;
1179         pid_t pid = sys_getpid();
1180         SMB_DEV_T dev = share_entry->dev;
1181         SMB_INO_T inode = share_entry->inode;
1182         unsigned long file_id = share_entry->share_file_id;
1183
1184         /* We need to send a ASYNC_LEVEL_II_OPLOCK_BREAK_CMD message to the port in the share mode entry. */
1185
1186         SSVAL(op_break_msg,OPBRK_MESSAGE_CMD_OFFSET,ASYNC_LEVEL_II_OPLOCK_BREAK_CMD);
1187         memcpy(op_break_msg+OPLOCK_BREAK_PID_OFFSET,(char *)&pid,sizeof(pid));
1188         memcpy(op_break_msg+OPLOCK_BREAK_DEV_OFFSET,(char *)&dev,sizeof(dev));
1189         memcpy(op_break_msg+OPLOCK_BREAK_INODE_OFFSET,(char *)&inode,sizeof(inode));
1190         memcpy(op_break_msg+OPLOCK_BREAK_FILEID_OFFSET,(char *)&file_id,sizeof(file_id));
1191
1192         /* Set the address and port. */
1193         memset((char *)&addr_out,'\0',sizeof(addr_out));
1194         addr_out.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1195         addr_out.sin_port = htons( share_entry->op_port );
1196         addr_out.sin_family = AF_INET;
1197    
1198         if( DEBUGLVL( 3 ) ) {
1199                 dbgtext( "request_remote_level2_async_oplock_break: sending an asynchronous oplock break message to ");
1200                 dbgtext( "pid %d on port %d ", (int)share_entry->pid, share_entry->op_port );
1201                 dbgtext( "for dev = %x, inode = %.0f, file_id = %lu\n",
1202             (unsigned int)dev, (double)inode, file_id );
1203         }
1204
1205         if(sys_sendto(oplock_sock,op_break_msg,OPLOCK_BREAK_MSG_LEN,0,
1206                         (struct sockaddr *)&addr_out,sizeof(addr_out)) < 0) {
1207                 if( DEBUGLVL( 0 ) ) {
1208                         dbgtext( "request_remote_level2_async_oplock_break: failed when sending a oplock " );
1209                         dbgtext( "break message to pid %d ", (int)share_entry->pid );
1210                         dbgtext( "on port %d ", share_entry->op_port );
1211                         dbgtext( "for dev = %x, inode = %.0f, file_id = %lu\n",
1212                                 (unsigned int)dev, (double)inode, file_id );
1213                         dbgtext( "Error was %s\n", strerror(errno) );
1214                 }
1215                 return False;
1216         }
1217
1218         DEBUG(3,("request_remote_level2_async_oplock_break: sent async break message to level II entry.\n"));
1219         return True;
1220 }
1221
1222 /****************************************************************************
1223  This function is called on any file modification or lock request. If a file
1224  is level 2 oplocked then it must tell all other level 2 holders to break to none.
1225 ****************************************************************************/
1226
1227 void release_level_2_oplocks_on_change(files_struct *fsp)
1228 {
1229         share_mode_entry *share_list = NULL;
1230         pid_t pid = sys_getpid();
1231         int num_share_modes = 0;
1232         int i;
1233         BOOL dummy;
1234
1235         /*
1236          * If this file is level II oplocked then we need
1237          * to grab the shared memory lock and inform all
1238          * other files with a level II lock that they need
1239          * to flush their read caches. We keep the lock over
1240          * the shared memory area whilst doing this.
1241          */
1242
1243         if (!LEVEL_II_OPLOCK_TYPE(fsp->oplock_type))
1244                 return;
1245
1246         if (lock_share_entry_fsp(fsp) == False) {
1247                 DEBUG(0,("release_level_2_oplocks_on_change: failed to lock share mode entry for file %s.\n", fsp->fsp_name ));
1248         }
1249
1250         num_share_modes = get_share_modes(fsp->dev, fsp->inode, &share_list,
1251                                           &dummy);
1252
1253         DEBUG(10,("release_level_2_oplocks_on_change: num_share_modes = %d\n", 
1254                         num_share_modes ));
1255
1256         for(i = 0; i < num_share_modes; i++) {
1257                 share_mode_entry *share_entry = &share_list[i];
1258
1259                 /*
1260                  * As there could have been multiple writes waiting at the lock_share_entry
1261                  * gate we may not be the first to enter. Hence the state of the op_types
1262                  * in the share mode entries may be partly NO_OPLOCK and partly LEVEL_II
1263                  * oplock. It will do no harm to re-send break messages to those smbd's
1264                  * that are still waiting their turn to remove their LEVEL_II state, and
1265                  * also no harm to ignore existing NO_OPLOCK states. JRA.
1266                  */
1267
1268                 DEBUG(10,("release_level_2_oplocks_on_change: share_entry[%i]->op_type == %d\n",
1269                                 i, share_entry->op_type ));
1270
1271                 if (share_entry->op_type == NO_OPLOCK)
1272                         continue;
1273
1274                 /* Paranoia .... */
1275                 if (EXCLUSIVE_OPLOCK_TYPE(share_entry->op_type)) {
1276                         DEBUG(0,("release_level_2_oplocks_on_change: PANIC. share mode entry %d is an exlusive oplock !\n", i ));
1277                         unlock_share_entry(fsp->conn, fsp->dev, fsp->inode);
1278                         abort();
1279                 }
1280
1281                 /*
1282                  * Check if this is a file we have open (including the
1283                  * file we've been called to do write_file on. If so
1284                  * then break it directly without releasing the lock.
1285                  */
1286
1287                 if (pid == share_entry->pid) {
1288                         files_struct *new_fsp = file_find_dif(share_entry->dev, share_entry->inode, share_entry->share_file_id);
1289
1290                         /* Paranoia check... */
1291                         if(new_fsp == NULL) {
1292                                 DEBUG(0,("release_level_2_oplocks_on_change: PANIC. share mode entry %d is not a local file !\n", i ));
1293                                 unlock_share_entry(fsp->conn, fsp->dev, fsp->inode);
1294                                 abort();
1295                         }
1296
1297                         DEBUG(10,("release_level_2_oplocks_on_change: breaking our own oplock.\n"));
1298
1299                         oplock_break_level2(new_fsp, True);
1300
1301                 } else {
1302
1303                         /*
1304                          * This is a remote file and so we send an asynchronous
1305                          * message.
1306                          */
1307
1308                         DEBUG(10,("release_level_2_oplocks_on_change: breaking remote oplock (async).\n"));
1309                         request_remote_level2_async_oplock_break(share_entry);
1310                 }
1311         }
1312
1313         SAFE_FREE(share_list);
1314         unlock_share_entry_fsp(fsp);
1315
1316         /* Paranoia check... */
1317         if (LEVEL_II_OPLOCK_TYPE(fsp->oplock_type)) {
1318                 DEBUG(0,("release_level_2_oplocks_on_change: PANIC. File %s still has a level II oplock.\n", fsp->fsp_name));
1319                 smb_panic("release_level_2_oplocks_on_change");
1320         }
1321 }
1322
1323 /****************************************************************************
1324  Send a 'retry your open' message to a process with a deferred open entry.
1325 ****************************************************************************/
1326
1327 BOOL send_deferred_open_retry_message(deferred_open_entry *entry)
1328 {
1329         char de_msg[DEFERRED_OPEN_MSG_LEN];
1330         struct sockaddr_in addr_out;
1331         pid_t pid = sys_getpid();
1332
1333         memset(de_msg, '\0', DEFERRED_OPEN_MSG_LEN);
1334         SSVAL(de_msg,DEFERRED_OPEN_CMD_OFFSET,RETRY_DEFERRED_OPEN_CMD);
1335         memcpy(de_msg+DEFERRED_OPEN_PID_OFFSET,(char *)&pid,sizeof(pid));
1336         memcpy(de_msg+DEFERRED_OPEN_DEV_OFFSET,(char *)&entry->dev,sizeof(entry->dev));
1337         memcpy(de_msg+DEFERRED_OPEN_INODE_OFFSET,(char *)&entry->inode,sizeof(entry->inode));
1338         memcpy(de_msg+DEFERRED_OPEN_MID_OFFSET,(char *)&entry->mid,sizeof(entry->mid));
1339
1340         /* Set the address and port. */
1341         memset((char *)&addr_out,'\0',sizeof(addr_out));
1342         addr_out.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1343         addr_out.sin_port = htons( entry->port );
1344         addr_out.sin_family = AF_INET;
1345    
1346         if( DEBUGLVL( 3 ) ) {
1347                 dbgtext( "send_deferred_open_retry_message: sending a message to ");
1348                 dbgtext( "pid %d on port %d ", (int)entry->pid, entry->port );
1349                 dbgtext( "for dev = %x, inode = %.0f, mid = %u\n",
1350                         (unsigned int)entry->dev, (double)entry->inode, (unsigned int)entry->mid );
1351         }
1352
1353         if(sys_sendto(oplock_sock,de_msg,DEFERRED_OPEN_MSG_LEN,0,
1354                         (struct sockaddr *)&addr_out,sizeof(addr_out)) < 0) {
1355                 if( DEBUGLVL( 0 ) ) {
1356                         dbgtext( "send_deferred_open_retry_message: failed sending a message to ");
1357                         dbgtext( "pid %d on port %d ", (int)entry->pid, entry->port );
1358                         dbgtext( "for dev = %x, inode = %.0f, mid = %u\n",
1359                                 (unsigned int)entry->dev, (double)entry->inode, (unsigned int)entry->mid );
1360                         dbgtext( "Error was %s\n", strerror(errno) );
1361                 }
1362                 return False;
1363         }
1364         return True;
1365 }
1366
1367 /****************************************************************************
1368  Setup oplocks for this process.
1369 ****************************************************************************/
1370
1371 BOOL init_oplocks(void)
1372 {
1373         struct sockaddr_in sock_name;
1374         socklen_t len = sizeof(sock_name);
1375
1376         DEBUG(3,("open_oplock_ipc: opening loopback UDP socket.\n"));
1377
1378         /* Open a lookback UDP socket on a random port. */
1379         oplock_sock = open_socket_in(SOCK_DGRAM, 0, 0, htonl(INADDR_LOOPBACK),False);
1380         if (oplock_sock == -1) {
1381                 DEBUG(0,("open_oplock_ipc: Failed to get local UDP socket for \
1382 address %lx. Error was %s\n", (long)htonl(INADDR_LOOPBACK), strerror(errno)));
1383                 global_oplock_port = 0;
1384                 return(False);
1385         }
1386
1387         /* Find out the transient UDP port we have been allocated. */
1388         if(getsockname(oplock_sock, (struct sockaddr *)&sock_name, &len)<0) {
1389                 DEBUG(0,("open_oplock_ipc: Failed to get local UDP port. Error was %s\n",
1390                          strerror(errno)));
1391                 close(oplock_sock);
1392                 oplock_sock = -1;
1393                 global_oplock_port = 0;
1394                 return False;
1395         }
1396         global_oplock_port = ntohs(sock_name.sin_port);
1397
1398         if (lp_kernel_oplocks()) {
1399 #if HAVE_KERNEL_OPLOCKS_IRIX
1400                 koplocks = irix_init_kernel_oplocks();
1401 #elif HAVE_KERNEL_OPLOCKS_LINUX
1402                 koplocks = linux_init_kernel_oplocks();
1403 #endif
1404         }
1405
1406         DEBUG(3,("open_oplock ipc: pid = %d, global_oplock_port = %u\n", 
1407                  (int)sys_getpid(), global_oplock_port));
1408
1409         return True;
1410 }