r23445: Fix suggested by Volker. Don't call rename_open_files
[tprouty/samba.git] / source / smbd / oplock.c
index 14b243b36edeca15c8b549130d540b1f5da296ac..ddb05d8b922e15cf6f27e86eb5b486ab372b3e8a 100644 (file)
@@ -3,6 +3,7 @@
    oplock processing
    Copyright (C) Andrew Tridgell 1992-1998
    Copyright (C) Jeremy Allison 1998 - 2001
+   Copyright (C) Volker Lendecke 2005
    
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */
 
+#define DBGC_CLASS DBGC_LOCKING
 #include "includes.h"
 
-/* Oplock ipc UDP socket. */
-static int oplock_sock = -1;
-uint16 global_oplock_port = 0;
-
 /* Current number of oplocks we have outstanding. */
 static int32 exclusive_oplocks_open = 0;
 static int32 level_II_oplocks_open = 0;
 BOOL global_client_failed_oplock_break = False;
-BOOL global_oplock_break = False;
 
+extern uint32 global_client_caps;
 extern int smb_read_error;
 
 static struct kernel_oplocks *koplocks;
 
-static BOOL oplock_break(SMB_DEV_T dev, SMB_INO_T inode, unsigned long file_id, BOOL local);
-
 /****************************************************************************
  Get the number of current exclusive oplocks.
 ****************************************************************************/
@@ -52,147 +48,55 @@ int32 get_number_of_exclusive_open_oplocks(void)
 
 BOOL oplock_message_waiting(fd_set *fds)
 {
-       if (koplocks && koplocks->msg_waiting(fds))
-               return True;
-
-       if (FD_ISSET(oplock_sock, fds))
+       if (koplocks && koplocks->msg_waiting(fds)) {
                return True;
+       }
 
        return False;
 }
 
 /****************************************************************************
- Read an oplock break message from either the oplock UDP fd or the
- kernel (if kernel oplocks are supported).
-
- If timeout is zero then *fds contains the file descriptors that
- are ready to be read and acted upon. If timeout is non-zero then
- *fds contains the file descriptors to be selected on for read.
- The timeout is in milliseconds
-
+ Find out if there are any kernel oplock messages waiting and process them
+ if so. pfds is the fd_set from the main select loop (which contains any
+ kernel oplock fd if that's what the system uses (IRIX). If may be NULL if
+ we're calling this in a shutting down state.
 ****************************************************************************/
 
-BOOL receive_local_message( char *buffer, int buffer_len, int timeout)
+void process_kernel_oplocks(struct messaging_context *msg_ctx, fd_set *pfds)
 {
-       struct sockaddr_in from;
-       socklen_t fromlen = sizeof(from);
-       int32 msg_len = 0;
-       fd_set fds;
-       int selrtn = -1;
-
-       FD_ZERO(&fds);
-       smb_read_error = 0;
-
        /*
         * We need to check for kernel oplocks before going into the select
         * here, as the EINTR generated by the linux kernel oplock may have
         * already been eaten. JRA.
         */
 
-       if (koplocks && koplocks->msg_waiting(&fds)) {
-               return koplocks->receive_message(&fds, buffer, buffer_len);
+       if (!koplocks) {
+               return;
        }
 
-       while (timeout > 0 && selrtn == -1) {
-               struct timeval to;
-               int maxfd = oplock_sock;
-               time_t starttime = time(NULL);
-
-               FD_ZERO(&fds);
-               maxfd = setup_oplock_select_set(&fds);
-
-               to.tv_sec = timeout / 1000;
-               to.tv_usec = (timeout % 1000) * 1000;
-
-               DEBUG(5,("receive_local_message: doing select with timeout of %d ms\n", timeout));
-
-               selrtn = sys_select(maxfd+1,&fds,NULL,NULL,&to);
-
-               if (selrtn == -1 && errno == EINTR) {
-
-                       /* could be a kernel oplock interrupt */
-                       if (koplocks && koplocks->msg_waiting(&fds)) {
-                               return koplocks->receive_message(&fds, buffer, buffer_len);
-                       }
-
-                       /*
-                        * Linux 2.0.x seems to have a bug in that
-                        * it can return -1, EINTR with a timeout of zero.
-                        * Make sure we bail out here with a read timeout
-                        * if we got EINTR on a timeout of 1 or less.
-                        */
-
-                       if (timeout <= 1) {
-                               smb_read_error = READ_TIMEOUT;
-                               return False;
-                       }
+       while (koplocks->msg_waiting(pfds)) { 
+               files_struct *fsp;
+               char msg[MSG_SMB_KERNEL_BREAK_SIZE];
 
-                       /* Not a kernel interrupt - could be a SIGUSR1 message. We must restart. */
-                       /* We need to decrement the timeout here. */
-                       timeout -= ((time(NULL) - starttime)*1000);
-                       if (timeout < 0)
-                               timeout = 1;
+               fsp = koplocks->receive_message(pfds);
 
-                       DEBUG(5,("receive_local_message: EINTR : new timeout %d ms\n", timeout));
-                       continue;
-               }
-
-               /* Check if error */
-               if(selrtn == -1) {
-                       /* something is wrong. Maybe the socket is dead? */
-                       smb_read_error = READ_ERROR;
-                       return False;
-               }
-
-               /* Did we timeout ? */
-               if (selrtn == 0) {
-                       smb_read_error = READ_TIMEOUT;
-                       return False;
+               if (fsp == NULL) {
+                       DEBUG(3, ("Kernel oplock message announced, but none "
+                                 "received\n"));
+                       return;
                }
-       }
-
-       if (koplocks && koplocks->msg_waiting(&fds)) {
-               return koplocks->receive_message(&fds, buffer, buffer_len);
-       }
-
-       if (!FD_ISSET(oplock_sock, &fds))
-               return False;
-
-       /*
-        * From here down we deal with the smbd <--> smbd
-        * oplock break protocol only.
-        */
-
-       /*
-        * Read a loopback udp message.
-        */
-       msg_len = sys_recvfrom(oplock_sock, &buffer[OPBRK_CMD_HEADER_LEN],
-                                               buffer_len - OPBRK_CMD_HEADER_LEN, 0, (struct sockaddr *)&from, &fromlen);
 
-       if(msg_len < 0) {
-               DEBUG(0,("receive_local_message. Error in recvfrom. (%s).\n",strerror(errno)));
-               return False;
-       }
+               /* Put the kernel break info into the message. */
+               push_file_id_16(msg, &fsp->file_id);
+               SIVAL(msg,16,fsp->fh->gen_id);
 
-       /* Validate message length. */
-       if(msg_len > (buffer_len - OPBRK_CMD_HEADER_LEN)) {
-               DEBUG(0,("receive_local_message: invalid msg_len (%d) max can be %d\n", msg_len,
-                       buffer_len  - OPBRK_CMD_HEADER_LEN));
-               return False;
-       }
+               /* Don't need to be root here as we're only ever
+                  sending to ourselves. */
 
-       /* Validate message from address (must be localhost). */
-       if(from.sin_addr.s_addr != htonl(INADDR_LOOPBACK)) {
-               DEBUG(0,("receive_local_message: invalid 'from' address \
-(was %lx should be 127.0.0.1)\n", (long)from.sin_addr.s_addr));
-               return False;
+               messaging_send_buf(msg_ctx, procid_self(),
+                                  MSG_SMB_KERNEL_BREAK,
+                                  (uint8 *)&msg, MSG_SMB_KERNEL_BREAK_SIZE);
        }
-
-       /* Setup the message header */
-       SIVAL(buffer,OPBRK_CMD_LEN_OFFSET,msg_len);
-       SSVAL(buffer,OPBRK_CMD_PORT_OFFSET,ntohs(from.sin_port));
-
-       return True;
 }
 
 /****************************************************************************
@@ -202,20 +106,23 @@ BOOL receive_local_message( char *buffer, int buffer_len, int timeout)
 
 BOOL set_file_oplock(files_struct *fsp, int oplock_type)
 {
-       if (koplocks && !koplocks->set_oplock(fsp, oplock_type))
+       if (koplocks && !koplocks->set_oplock(fsp, oplock_type)) {
                return False;
+       }
 
        fsp->oplock_type = oplock_type;
        fsp->sent_oplock_break = NO_BREAK_SENT;
-       if (oplock_type == LEVEL_II_OPLOCK)
+       if (oplock_type == LEVEL_II_OPLOCK) {
                level_II_oplocks_open++;
-       else
+       } else {
                exclusive_oplocks_open++;
+       }
 
-       DEBUG(5,("set_file_oplock: granted oplock on file %s, dev = %x, inode = %.0f, file_id = %lu, \
-tv_sec = %x, tv_usec = %x\n",
-                fsp->fsp_name, (unsigned int)fsp->dev, (double)fsp->inode, fsp->file_id,
-                (int)fsp->open_time.tv_sec, (int)fsp->open_time.tv_usec ));
+       DEBUG(5,("set_file_oplock: granted oplock on file %s, %s/%lu, "
+                   "tv_sec = %x, tv_usec = %x\n",
+                fsp->fsp_name, file_id_static_string(&fsp->file_id),
+                fsp->fh->gen_id, (int)fsp->open_time.tv_sec,
+                (int)fsp->open_time.tv_usec ));
 
        return True;
 }
@@ -226,13 +133,20 @@ tv_sec = %x, tv_usec = %x\n",
 
 void release_file_oplock(files_struct *fsp)
 {
-       if (koplocks)
+       if ((fsp->oplock_type != NO_OPLOCK) &&
+           (fsp->oplock_type != FAKE_LEVEL_II_OPLOCK) &&
+           koplocks) {
                koplocks->release_oplock(fsp);
+       }
 
-       if (fsp->oplock_type == LEVEL_II_OPLOCK)
+       if (fsp->oplock_type == LEVEL_II_OPLOCK) {
                level_II_oplocks_open--;
-       else
+       } else if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
                exclusive_oplocks_open--;
+       }
+
+       SMB_ASSERT(exclusive_oplocks_open>=0);
+       SMB_ASSERT(level_II_oplocks_open>=0);
        
        fsp->oplock_type = NO_OPLOCK;
        fsp->sent_oplock_break = NO_BREAK_SENT;
@@ -246,8 +160,9 @@ void release_file_oplock(files_struct *fsp)
 
 static void downgrade_file_oplock(files_struct *fsp)
 {
-       if (koplocks)
+       if (koplocks) {
                koplocks->release_oplock(fsp);
+       }
        fsp->oplock_type = LEVEL_II_OPLOCK;
        exclusive_oplocks_open--;
        level_II_oplocks_open++;
@@ -260,257 +175,106 @@ static void downgrade_file_oplock(files_struct *fsp)
  to none even if a "break-to-level II" was sent.
 ****************************************************************************/
 
-BOOL remove_oplock(files_struct *fsp, BOOL break_to_none)
+BOOL remove_oplock(files_struct *fsp)
 {
-       SMB_DEV_T dev = fsp->dev;
-       SMB_INO_T inode = fsp->inode;
-       BOOL ret = True;
+       BOOL ret;
+       struct share_mode_lock *lck;
 
        /* Remove the oplock flag from the sharemode. */
-       if (lock_share_entry_fsp(fsp) == False) {
-               DEBUG(0,("remove_oplock: failed to lock share entry for file %s\n",
-                        fsp->fsp_name ));
-               ret = False;
+       lck = get_share_mode_lock(NULL, fsp->file_id, NULL, NULL);
+       if (lck == NULL) {
+               DEBUG(0,("remove_oplock: failed to lock share entry for "
+                        "file %s\n", fsp->fsp_name ));
+               return False;
        }
-
-       if (fsp->sent_oplock_break == EXCLUSIVE_BREAK_SENT || break_to_none) {
-               /*
-                * Deal with a reply when a break-to-none was sent.
-                */
-
-               if(remove_share_oplock(fsp)==False) {
-                       DEBUG(0,("remove_oplock: failed to remove share oplock for file %s fnum %d, \
-dev = %x, inode = %.0f\n", fsp->fsp_name, fsp->fnum, (unsigned int)dev, (double)inode));
-                       ret = False;
-               }
-
-               release_file_oplock(fsp);
-       } else {
-               /*
-                * Deal with a reply when a break-to-level II was sent.
-                */
-               if(downgrade_share_oplock(fsp)==False) {
-                       DEBUG(0,("remove_oplock: failed to downgrade share oplock for file %s fnum %d, \
-dev = %x, inode = %.0f\n", fsp->fsp_name, fsp->fnum, (unsigned int)dev, (double)inode));
-                       ret = False;
-               }
-               
-               downgrade_file_oplock(fsp);
+       ret = remove_share_oplock(lck, fsp);
+       if (!ret) {
+               DEBUG(0,("remove_oplock: failed to remove share oplock for "
+                        "file %s fnum %d, %s\n",
+                        fsp->fsp_name, fsp->fnum, file_id_static_string(&fsp->file_id)));
        }
-
-       unlock_share_entry_fsp(fsp);
+       release_file_oplock(fsp);
+       TALLOC_FREE(lck);
        return ret;
 }
 
-/****************************************************************************
- Setup the listening set of file descriptors for an oplock break
- message either from the UDP socket or from the kernel. Returns the maximum
- fd used.
-****************************************************************************/
-
-int setup_oplock_select_set( fd_set *fds)
+/*
+ * Deal with a reply when a break-to-level II was sent.
+ */
+BOOL downgrade_oplock(files_struct *fsp)
 {
-       int maxfd = oplock_sock;
-
-       if(oplock_sock == -1)
-               return 0;
+       BOOL ret;
+       struct share_mode_lock *lck;
 
-       FD_SET(oplock_sock,fds);
-
-       if (koplocks && koplocks->notification_fd != -1) {
-               FD_SET(koplocks->notification_fd, fds);
-               maxfd = MAX(maxfd, koplocks->notification_fd);
+       lck = get_share_mode_lock(NULL, fsp->file_id, NULL, NULL);
+       if (lck == NULL) {
+               DEBUG(0,("downgrade_oplock: failed to lock share entry for "
+                        "file %s\n", fsp->fsp_name ));
+               return False;
+       }
+       ret = downgrade_share_oplock(lck, fsp);
+       if (!ret) {
+               DEBUG(0,("downgrade_oplock: failed to downgrade share oplock "
+                        "for file %s fnum %d, file_id %s\n",
+                        fsp->fsp_name, fsp->fnum, file_id_static_string(&fsp->file_id)));
        }
 
-       return maxfd;
+       downgrade_file_oplock(fsp);
+       TALLOC_FREE(lck);
+       return ret;
 }
 
 /****************************************************************************
- Process an oplock break message - whether it came from the UDP socket
- or from the kernel.
+ Return the fd (if any) used for receiving oplock notifications.
 ****************************************************************************/
 
-BOOL process_local_message(char *buffer, int buf_size)
+int oplock_notify_fd(void)
 {
-       int32 msg_len;
-       uint16 from_port;
-       char *msg_start;
-       pid_t remotepid;
-       SMB_DEV_T dev;
-       SMB_INO_T inode;
-       unsigned long file_id;
-       uint16 break_cmd_type;
-
-       msg_len = IVAL(buffer,OPBRK_CMD_LEN_OFFSET);
-       from_port = SVAL(buffer,OPBRK_CMD_PORT_OFFSET);
-
-       msg_start = &buffer[OPBRK_CMD_HEADER_LEN];
-
-       DEBUG(5,("process_local_message: Got a message of length %d from port (%d)\n", 
-               msg_len, from_port));
-
-       /* 
-        * Pull the info out of the requesting packet.
-        */
-
-       break_cmd_type = SVAL(msg_start,OPBRK_MESSAGE_CMD_OFFSET);
-
-       switch(break_cmd_type) {
-               case KERNEL_OPLOCK_BREAK_CMD:
-                       if (!koplocks) {
-                               DEBUG(0,("unexpected kernel oplock break!\n"));
-                               break;
-                       } 
-                       if (!koplocks->parse_message(msg_start, msg_len, &inode, &dev, &file_id)) {
-                               DEBUG(0,("kernel oplock break parse failure!\n"));
-                       }
-                       break;
-
-               case OPLOCK_BREAK_CMD:
-               case LEVEL_II_OPLOCK_BREAK_CMD:
-
-                       /* Ensure that the msg length is correct. */
-                       if(msg_len != OPLOCK_BREAK_MSG_LEN) {
-                               DEBUG(0,("process_local_message: incorrect length for OPLOCK_BREAK_CMD (was %d, should be %d).\n",
-                                       (int)msg_len, (int)OPLOCK_BREAK_MSG_LEN));
-                               return False;
-                       }
-
-                       memcpy((char *)&remotepid, msg_start+OPLOCK_BREAK_PID_OFFSET,sizeof(remotepid));
-                       memcpy((char *)&inode, msg_start+OPLOCK_BREAK_INODE_OFFSET,sizeof(inode));
-                       memcpy((char *)&dev, msg_start+OPLOCK_BREAK_DEV_OFFSET,sizeof(dev));
-                       memcpy((char *)&file_id, msg_start+OPLOCK_BREAK_FILEID_OFFSET,sizeof(file_id));
-
-                       DEBUG(5,("process_local_message: (%s) oplock break request from \
-pid %d, port %d, dev = %x, inode = %.0f, file_id = %lu\n",
-                               (break_cmd_type == OPLOCK_BREAK_CMD) ? "exclusive" : "level II",
-                               (int)remotepid, from_port, (unsigned int)dev, (double)inode, file_id));
-                       break;
-
-               /* 
-                * Keep this as a debug case - eventually we can remove it.
-                */
-               case 0x8001:
-                       DEBUG(0,("process_local_message: Received unsolicited break \
-reply - dumping info.\n"));
-
-                       if(msg_len != OPLOCK_BREAK_MSG_LEN) {
-                               DEBUG(0,("process_local_message: ubr: incorrect length for reply \
-(was %d, should be %d).\n", (int)msg_len, (int)OPLOCK_BREAK_MSG_LEN));
-                               return False;
-                       }
-
-                       memcpy((char *)&inode, msg_start+OPLOCK_BREAK_INODE_OFFSET,sizeof(inode));
-                       memcpy((char *)&remotepid, msg_start+OPLOCK_BREAK_PID_OFFSET,sizeof(remotepid));
-                       memcpy((char *)&dev, msg_start+OPLOCK_BREAK_DEV_OFFSET,sizeof(dev));
-                       memcpy((char *)&file_id, msg_start+OPLOCK_BREAK_FILEID_OFFSET,sizeof(file_id));
-
-                       DEBUG(0,("process_local_message: unsolicited oplock break reply from \
-pid %d, port %d, dev = %x, inode = %.0f, file_id = %lu\n",
-                               (int)remotepid, from_port, (unsigned int)dev, (double)inode, file_id));
-
-                       return False;
-
-               default:
-                       DEBUG(0,("process_local_message: unknown UDP message command code (%x) - ignoring.\n",
-                               (unsigned int)SVAL(msg_start,0)));
-                       return False;
-       }
-
-       /*
-        * Now actually process the break request.
-        */
-
-       if((exclusive_oplocks_open + level_II_oplocks_open) != 0) {
-               if (oplock_break(dev, inode, file_id, False) == False) {
-                       DEBUG(0,("process_local_message: oplock break failed.\n"));
-                       return False;
-               }
-       } else {
-               /*
-                * If we have no record of any currently open oplocks,
-                * it's not an error, as a close command may have
-                * just been issued on the file that was oplocked.
-                * Just log a message and return success in this case.
-                */
-               DEBUG(3,("process_local_message: oplock break requested with no outstanding \
-oplocks. Returning success.\n"));
+       if (koplocks) {
+               return koplocks->notification_fd;
        }
 
-       /* 
-        * Do the appropriate reply - none in the kernel or level II case.
-        */
-
-       if(SVAL(msg_start,OPBRK_MESSAGE_CMD_OFFSET) == OPLOCK_BREAK_CMD) {
-               struct sockaddr_in toaddr;
-
-               /* Send the message back after OR'ing in the 'REPLY' bit. */
-               SSVAL(msg_start,OPBRK_MESSAGE_CMD_OFFSET,OPLOCK_BREAK_CMD | CMD_REPLY);
-
-               memset((char *)&toaddr,'\0',sizeof(toaddr));
-               toaddr.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
-               toaddr.sin_port = htons(from_port);
-               toaddr.sin_family = AF_INET;
-
-               if(sys_sendto( oplock_sock, msg_start, OPLOCK_BREAK_MSG_LEN, 0,
-                               (struct sockaddr *)&toaddr, sizeof(toaddr)) < 0) {
-                       DEBUG(0,("process_local_message: sendto process %d failed. Errno was %s\n",
-                               (int)remotepid, strerror(errno)));
-                       return False;
-               }
-
-               DEBUG(5,("process_local_message: oplock break reply sent to \
-pid %d, port %d, for file dev = %x, inode = %.0f, file_id = %lu\n",
-                       (int)remotepid, from_port, (unsigned int)dev, (double)inode, file_id));
-       }
-
-       return True;
+       return -1;
 }
 
 /****************************************************************************
  Set up an oplock break message.
 ****************************************************************************/
 
-static void prepare_break_message(char *outbuf, files_struct *fsp, BOOL level2)
+static char *new_break_smb_message(TALLOC_CTX *mem_ctx,
+                                  files_struct *fsp, uint8 cmd)
 {
-       memset(outbuf,'\0',smb_size);
-       set_message(outbuf,8,0,True);
-
-       SCVAL(outbuf,smb_com,SMBlockingX);
-       SSVAL(outbuf,smb_tid,fsp->conn->cnum);
-       SSVAL(outbuf,smb_pid,0xFFFF);
-       SSVAL(outbuf,smb_uid,0);
-       SSVAL(outbuf,smb_mid,0xFFFF);
-       SCVAL(outbuf,smb_vwv0,0xFF);
-       SSVAL(outbuf,smb_vwv2,fsp->fnum);
-       SCVAL(outbuf,smb_vwv3,LOCKING_ANDX_OPLOCK_RELEASE);
-       SCVAL(outbuf,smb_vwv3+1,level2 ? OPLOCKLEVEL_II : OPLOCKLEVEL_NONE);
+       char *result = TALLOC_ARRAY(mem_ctx, char, smb_size + 8*2 + 0);
+
+       if (result == NULL) {
+               DEBUG(0, ("talloc failed\n"));
+               return NULL;
+       }
+
+       memset(result,'\0',smb_size);
+       set_message(NULL,result,8,0,True);
+       SCVAL(result,smb_com,SMBlockingX);
+       SSVAL(result,smb_tid,fsp->conn->cnum);
+       SSVAL(result,smb_pid,0xFFFF);
+       SSVAL(result,smb_uid,0);
+       SSVAL(result,smb_mid,0xFFFF);
+       SCVAL(result,smb_vwv0,0xFF);
+       SSVAL(result,smb_vwv2,fsp->fnum);
+       SCVAL(result,smb_vwv3,LOCKING_ANDX_OPLOCK_RELEASE);
+       SCVAL(result,smb_vwv3+1,cmd);
+       return result;
 }
 
 /****************************************************************************
  Function to do the waiting before sending a local break.
 ****************************************************************************/
 
-static void wait_before_sending_break(BOOL local_request)
+static void wait_before_sending_break(void)
 {
-       extern struct timeval smb_last_time;
-
-       if(local_request) {
-               struct timeval cur_tv;
-               long wait_left = (long)lp_oplock_break_wait_time();
-
-               if (wait_left == 0)
-                       return;
-
-               GetTimeOfDay(&cur_tv);
+       long wait_time = (long)lp_oplock_break_wait_time();
 
-               wait_left -= ((cur_tv.tv_sec - smb_last_time.tv_sec)*1000) +
-                ((cur_tv.tv_usec - smb_last_time.tv_usec)/1000);
-
-               if(wait_left > 0) {
-                       wait_left = MIN(wait_left, 1000);
-                       sys_usleep(wait_left * 1000);
-               }
+       if (wait_time) {
+               smb_msleep(wait_time);
        }
 }
 
@@ -518,13 +282,13 @@ static void wait_before_sending_break(BOOL local_request)
  Ensure that we have a valid oplock.
 ****************************************************************************/
 
-static files_struct *initial_break_processing(SMB_DEV_T dev, SMB_INO_T inode, unsigned long file_id)
+static files_struct *initial_break_processing(struct file_id id, unsigned long file_id)
 {
        files_struct *fsp = NULL;
 
        if( DEBUGLVL( 3 ) ) {
-               dbgtext( "initial_break_processing: called for dev = %x, inode = %.0f file_id = %lu\n",
-                       (unsigned int)dev, (double)inode, file_id);
+               dbgtext( "initial_break_processing: called for %s/%u\n",
+                        file_id_static_string(&id), (int)file_id);
                dbgtext( "Current oplocks_open (exclusive = %d, levelII = %d)\n",
                        exclusive_oplocks_open, level_II_oplocks_open );
        }
@@ -535,14 +299,13 @@ static files_struct *initial_break_processing(SMB_DEV_T dev, SMB_INO_T inode, un
         * we have an oplock on it.
         */
 
-       fsp = file_find_dif(dev, inode, file_id);
+       fsp = file_find_dif(id, file_id);
 
        if(fsp == NULL) {
                /* The file could have been closed in the meantime - return success. */
                if( DEBUGLVL( 3 ) ) {
                        dbgtext( "initial_break_processing: cannot find open file with " );
-                       dbgtext( "dev = %x, inode = %.0f file_id = %lu", (unsigned int)dev,
-                               (double)inode, file_id);
+                       dbgtext( "file_id %s gen_id = %lu", file_id_static_string(&id), file_id);
                        dbgtext( "allowing break to succeed.\n" );
                }
                return NULL;
@@ -561,8 +324,8 @@ static files_struct *initial_break_processing(SMB_DEV_T dev, SMB_INO_T inode, un
        if(fsp->oplock_type == NO_OPLOCK) {
                if( DEBUGLVL( 3 ) ) {
                        dbgtext( "initial_break_processing: file %s ", fsp->fsp_name );
-                       dbgtext( "(dev = %x, inode = %.0f, file_id = %lu) has no oplock.\n",
-                               (unsigned int)dev, (double)inode, fsp->file_id );
+                       dbgtext( "(file_id = %s gen_id = %lu) has no oplock.\n",
+                                file_id_static_string(&id), fsp->fh->gen_id );
                        dbgtext( "Allowing break to succeed regardless.\n" );
                }
                return NULL;
@@ -571,547 +334,415 @@ static files_struct *initial_break_processing(SMB_DEV_T dev, SMB_INO_T inode, un
        return fsp;
 }
 
-/****************************************************************************
- Process a level II oplock break directly.
-****************************************************************************/
+static void oplock_timeout_handler(struct event_context *ctx,
+                                  struct timed_event *te,
+                                  const struct timeval *now,
+                                  void *private_data)
+{
+       files_struct *fsp = (files_struct *)private_data;
+
+       /* Ensure we always remove this event. */
+       if (fsp->oplock_timeout != NULL) {
+               /* Remove the timed event handler. */
+               TALLOC_FREE(fsp->oplock_timeout);
+               fsp->oplock_timeout = NULL;
+       }
+       DEBUG(0, ("Oplock break failed for file %s -- replying anyway\n", fsp->fsp_name));
+       global_client_failed_oplock_break = True;
+       remove_oplock(fsp);
+       reply_to_oplock_break_requests(fsp);
+}
+
+/*******************************************************************
+ Add a timeout handler waiting for the client reply.
+*******************************************************************/
 
-BOOL oplock_break_level2(files_struct *fsp, BOOL local_request, int token)
+static void add_oplock_timeout_handler(files_struct *fsp)
 {
-       extern uint32 global_client_caps;
-       char outbuf[128];
-       BOOL got_lock = False;
-       SMB_DEV_T dev = fsp->dev;
-       SMB_INO_T inode = fsp->inode;
+       if (fsp->oplock_timeout != NULL) {
+               DEBUG(0, ("Logic problem -- have an oplock event hanging "
+                         "around\n"));
+       }
 
-       /*
-        * We can have a level II oplock even if the client is not
-        * level II oplock aware. In this case just remove the
-        * flags and don't send the break-to-none message to
-        * the client.
-        */
+       fsp->oplock_timeout =
+               event_add_timed(smbd_event_context(), NULL,
+                               timeval_current_ofs(OPLOCK_BREAK_TIMEOUT, 0),
+                               "oplock_timeout_handler",
+                               oplock_timeout_handler, fsp);
 
-       if (global_client_caps & CAP_LEVEL_II_OPLOCKS) {
-               /*
-                * If we are sending an oplock break due to an SMB sent
-                * by our own client we ensure that we wait at leat
-                * lp_oplock_break_wait_time() milliseconds before sending
-                * the packet. Sending the packet sooner can break Win9x
-                * and has reported to cause problems on NT. JRA.
-                */
+       if (fsp->oplock_timeout == NULL) {
+               DEBUG(0, ("Could not add oplock timeout handler\n"));
+       }
+}
 
-               wait_before_sending_break(local_request);
+/*******************************************************************
+ This handles the case of a write triggering a break to none
+ message on a level2 oplock.
+ When we get this message we may be in any of three states :
+ NO_OPLOCK, LEVEL_II, FAKE_LEVEL2. We only send a message to
+ the client for LEVEL2.
+*******************************************************************/
+
+static void process_oplock_async_level2_break_message(struct messaging_context *msg_ctx,
+                                                     void *private_data,
+                                                     uint32_t msg_type,
+                                                     struct server_id src,
+                                                     DATA_BLOB *data)
+{
+       struct share_mode_entry msg;
+       files_struct *fsp;
+       char *break_msg;
+       BOOL sign_state;
 
-               /* Prepare the SMBlockingX message. */
+       if (data->data == NULL) {
+               DEBUG(0, ("Got NULL buffer\n"));
+               return;
+       }
 
-               prepare_break_message( outbuf, fsp, False);
-               if (!send_smb(smbd_server_fd(), outbuf))
-                       exit_server("oplock_break_level2: send_smb failed.");
+       if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
+               DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
+               return;
        }
 
-       /*
-        * Now we must update the shared memory structure to tell
-        * everyone else we no longer have a level II oplock on 
-        * this open file. If local_request is true then token is
-        * the existing lock on the shared memory area.
-        */
+       /* De-linearize incoming message. */
+       message_to_share_mode_entry(&msg, (char *)data->data);
 
-       if(!local_request && lock_share_entry_fsp(fsp) == False) {
-               DEBUG(0,("oplock_break_level2: unable to lock share entry for file %s\n", fsp->fsp_name ));
-       } else {
-               got_lock = True;
+       DEBUG(10, ("Got oplock async level 2 break message from pid %d: %s/%lu\n",
+                  (int)procid_to_pid(&src), file_id_static_string(&msg.id), msg.share_file_id));
+
+       fsp = initial_break_processing(msg.id, msg.share_file_id);
+
+       if (fsp == NULL) {
+               /* We hit a race here. Break messages are sent, and before we
+                * get to process this message, we have closed the file. 
+                * No need to reply as this is an async message. */
+               DEBUG(3, ("process_oplock_async_level2_break_message: Did not find fsp, ignoring\n"));
+               return;
        }
 
-       if(remove_share_oplock(fsp)==False) {
-               DEBUG(0,("oplock_break_level2: unable to remove level II oplock for file %s\n", fsp->fsp_name ));
+       if (fsp->oplock_type == NO_OPLOCK) {
+               /* We already got a "break to none" message and we've handled it.
+                * just ignore. */
+               DEBUG(3, ("process_oplock_async_level2_break_message: already broken to none, ignoring.\n"));
+               return;
        }
 
-       if (!local_request && got_lock)
-               unlock_share_entry_fsp(fsp);
+       if (fsp->oplock_type == FAKE_LEVEL_II_OPLOCK) {
+               /* Don't tell the client, just downgrade. */
+               DEBUG(3, ("process_oplock_async_level2_break_message: downgrading fake level 2 oplock.\n"));
+               remove_oplock(fsp);
+               return;
+       }
 
-       fsp->oplock_type = NO_OPLOCK;
-       level_II_oplocks_open--;
+       /* Ensure we're really at level2 state. */
+       SMB_ASSERT(fsp->oplock_type == LEVEL_II_OPLOCK);
+
+       /* Now send a break to none message to our client. */
 
-       if(level_II_oplocks_open < 0) {
-               DEBUG(0,("oplock_break_level2: level_II_oplocks_open < 0 (%d). PANIC ERROR\n",
-                       level_II_oplocks_open));
-               abort();
+       break_msg = new_break_smb_message(NULL, fsp, OPLOCKLEVEL_NONE);
+       if (break_msg == NULL) {
+               exit_server("Could not talloc break_msg\n");
        }
 
-       if( DEBUGLVL( 3 ) ) {
-               dbgtext( "oplock_break_level2: returning success for " );
-               dbgtext( "dev = %x, inode = %.0f, file_id = %lu\n", (unsigned int)dev, (double)inode, fsp->file_id );
-               dbgtext( "Current level II oplocks_open = %d\n", level_II_oplocks_open );
+       /* Need to wait before sending a break message if we sent ourselves this message. */
+       if (procid_to_pid(&src) == sys_getpid()) {
+               wait_before_sending_break();
        }
 
-       return True;
-}
+       /* Save the server smb signing state. */
+       sign_state = srv_oplock_set_signing(False);
 
-/****************************************************************************
- Process an oplock break directly.
-****************************************************************************/
+       show_msg(break_msg);
+       if (!send_smb(smbd_server_fd(), break_msg)) {
+               exit_server_cleanly("oplock_break: send_smb failed.");
+       }
 
-static BOOL oplock_break(SMB_DEV_T dev, SMB_INO_T inode, unsigned long file_id, BOOL local_request)
-{
-       extern uint32 global_client_caps;
-       extern struct current_user current_user;
-       char *inbuf = NULL;
-       char *outbuf = NULL;
-       files_struct *fsp = NULL;
-       time_t start_time;
-       BOOL shutdown_server = False;
-       BOOL oplock_timeout = False;
-       connection_struct *saved_user_conn;
-       connection_struct *saved_fsp_conn;
-       int saved_vuid;
-       pstring saved_dir; 
-       int timeout = (OPLOCK_BREAK_TIMEOUT * 1000);
-       pstring file_name;
-       BOOL using_levelII;
-
-       if((fsp = initial_break_processing(dev, inode, file_id)) == NULL)
-               return True;
+       /* Restore the sign state to what it was. */
+       srv_oplock_set_signing(sign_state);
 
-       /*
-        * Deal with a level II oplock going break to none separately.
-        */
+       TALLOC_FREE(break_msg);
 
-       if (LEVEL_II_OPLOCK_TYPE(fsp->oplock_type))
-               return oplock_break_level2(fsp, local_request, -1);
+       /* Async level2 request, don't send a reply, just remove the oplock. */
+       remove_oplock(fsp);
+}
 
-       /* Mark the oplock break as sent - we don't want to send twice! */
-       if (fsp->sent_oplock_break) {
-               if( DEBUGLVL( 0 ) ) {
-                       dbgtext( "oplock_break: ERROR: oplock_break already sent for " );
-                       dbgtext( "file %s ", fsp->fsp_name);
-                       dbgtext( "(dev = %x, inode = %.0f, file_id = %lu)\n", (unsigned int)dev, (double)inode, fsp->file_id );
-               }
+/*******************************************************************
+ This handles the generic oplock break message from another smbd.
+*******************************************************************/
 
-               /*
-                * We have to fail the open here as we cannot send another oplock break on
-                * this file whilst we are awaiting a response from the client - neither
-                * can we allow another open to succeed while we are waiting for the client.
-                */
-               return False;
+static void process_oplock_break_message(struct messaging_context *msg_ctx,
+                                        void *private_data,
+                                        uint32_t msg_type,
+                                        struct server_id src,
+                                        DATA_BLOB *data)
+{
+       struct share_mode_entry msg;
+       files_struct *fsp;
+       char *break_msg;
+       BOOL break_to_level2 = False;
+       BOOL sign_state;
+
+       if (data->data == NULL) {
+               DEBUG(0, ("Got NULL buffer\n"));
+               return;
        }
 
-       if(global_oplock_break) {
-               DEBUG(0,("ABORT : ABORT : recursion in oplock_break !!!!!\n"));
-               abort();
+       if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
+               DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
+               return;
        }
 
-       /*
-        * Now comes the horrid part. We must send an oplock break to the client,
-        * and then process incoming messages until we get a close or oplock release.
-        * At this point we know we need a new inbuf/outbuf buffer pair.
-        * We cannot use these staticaly as we may recurse into here due to
-        * messages crossing on the wire.
-        */
+       /* De-linearize incoming message. */
+       message_to_share_mode_entry(&msg, (char *)data->data);
 
-       if((inbuf = (char *)malloc(BUFFER_SIZE + LARGE_WRITEX_HDR_SIZE + SAFETY_MARGIN))==NULL) {
-               DEBUG(0,("oplock_break: malloc fail for input buffer.\n"));
-               return False;
-       }
+       DEBUG(10, ("Got oplock break message from pid %d: %s/%lu\n",
+                  (int)procid_to_pid(&src), file_id_static_string(&msg.id), msg.share_file_id));
 
-       if((outbuf = (char *)malloc(BUFFER_SIZE + LARGE_WRITEX_HDR_SIZE + SAFETY_MARGIN))==NULL) {
-               DEBUG(0,("oplock_break: malloc fail for output buffer.\n"));
-               SAFE_FREE(inbuf);
-               return False;
-       }
+       fsp = initial_break_processing(msg.id, msg.share_file_id);
 
-       /*
-        * If we are sending an oplock break due to an SMB sent
-        * by our own client we ensure that we wait at leat
-        * lp_oplock_break_wait_time() milliseconds before sending
-        * the packet. Sending the packet sooner can break Win9x
-        * and has reported to cause problems on NT. JRA.
-        */
+       if (fsp == NULL) {
+               /* a We hit race here. Break messages are sent, and before we
+                * get to process this message, we have closed the file. Reply
+                * with 'ok, oplock broken' */
+               DEBUG(3, ("Did not find fsp\n"));
 
-       wait_before_sending_break(local_request);
+               /* We just send the same message back. */
+               messaging_send_buf(msg_ctx, src, MSG_SMB_BREAK_RESPONSE,
+                                  (uint8 *)data->data,
+                                  MSG_SMB_SHARE_MODE_ENTRY_SIZE);
+               return;
+       }
 
-       /* Prepare the SMBlockingX message. */
+       if (fsp->sent_oplock_break != NO_BREAK_SENT) {
+               /* Remember we have to inform the requesting PID when the
+                * client replies */
+               msg.pid = src;
+               ADD_TO_ARRAY(NULL, struct share_mode_entry, msg,
+                            &fsp->pending_break_messages,
+                            &fsp->num_pending_break_messages);
+               return;
+       }
+
+       if (EXCLUSIVE_OPLOCK_TYPE(msg.op_type) &&
+           !EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
+               DEBUG(3, ("Already downgraded oplock on %s: %s\n",
+                         file_id_static_string(&fsp->file_id),
+                         fsp->fsp_name));
+               /* We just send the same message back. */
+               messaging_send_buf(msg_ctx, src, MSG_SMB_BREAK_RESPONSE,
+                                  (uint8 *)data->data,
+                                  MSG_SMB_SHARE_MODE_ENTRY_SIZE);
+               return;
+       }
 
        if ((global_client_caps & CAP_LEVEL_II_OPLOCKS) && 
-                       !koplocks && /* NOTE: we force levelII off for kernel oplocks - this will change when it is supported */
-                       lp_level2_oplocks(SNUM(fsp->conn))) {
-               using_levelII = True;
-       } else {
-               using_levelII = False;
+           !(msg.op_type & FORCE_OPLOCK_BREAK_TO_NONE) &&
+           !koplocks && /* NOTE: we force levelII off for kernel oplocks -
+                         * this will change when it is supported */
+           lp_level2_oplocks(SNUM(fsp->conn))) {
+               break_to_level2 = True;
        }
 
-       prepare_break_message( outbuf, fsp, using_levelII);
-       /* Remember if we just sent a break to level II on this file. */
-       fsp->sent_oplock_break = using_levelII? LEVEL_II_BREAK_SENT:EXCLUSIVE_BREAK_SENT;
+       break_msg = new_break_smb_message(NULL, fsp, break_to_level2 ?
+                                         OPLOCKLEVEL_II : OPLOCKLEVEL_NONE);
+       if (break_msg == NULL) {
+               exit_server("Could not talloc break_msg\n");
+       }
 
-       if (!send_smb(smbd_server_fd(), outbuf))
-               exit_server("oplock_break: send_smb failed.");
+       /* Need to wait before sending a break message if we sent ourselves this message. */
+       if (procid_to_pid(&src) == sys_getpid()) {
+               wait_before_sending_break();
+       }
 
-       /* We need this in case a readraw crosses on the wire. */
-       global_oplock_break = True;
-       /* Process incoming messages. */
+       /* Save the server smb signing state. */
+       sign_state = srv_oplock_set_signing(False);
 
-       /*
-        * JRA - If we don't get a break from the client in OPLOCK_BREAK_TIMEOUT
-        * seconds we should just die....
-        */
+       show_msg(break_msg);
+       if (!send_smb(smbd_server_fd(), break_msg)) {
+               exit_server_cleanly("oplock_break: send_smb failed.");
+       }
 
-       start_time = time(NULL);
+       /* Restore the sign state to what it was. */
+       srv_oplock_set_signing(sign_state);
 
-       /*
-        * Save the information we need to re-become the
-        * user, then unbecome the user whilst we're doing this.
-        */
-       saved_user_conn = current_user.conn;
-       saved_vuid = current_user.vuid;
-       saved_fsp_conn = fsp->conn;
-       change_to_root_user();
-       vfs_GetWd(saved_fsp_conn,saved_dir);
-       /* Save the chain fnum. */
-       file_chain_save();
+       TALLOC_FREE(break_msg);
 
-       /*
-        * From Charles Hoch <hoch@exemplary.com>. If the break processing
-        * code closes the file (as it often does), then the fsp pointer here
-        * points to free()'d memory. We *must* revalidate fsp each time
-        * around the loop.
-        */
+       fsp->sent_oplock_break = break_to_level2 ? LEVEL_II_BREAK_SENT:BREAK_TO_NONE_SENT;
 
-       pstrcpy(file_name, fsp->fsp_name);
-
-       while((fsp = initial_break_processing(dev, inode, file_id)) &&
-                       OPEN_FSP(fsp) && EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
-               if(receive_smb(smbd_server_fd(),inbuf, timeout) == False) {
-                       /*
-                        * Die if we got an error.
-                        */
-
-                       if (smb_read_error == READ_EOF) {
-                               DEBUG( 0, ( "oplock_break: end of file from client\n" ) );
-                               shutdown_server = True;
-                       } else if (smb_read_error == READ_ERROR) {
-                               DEBUG( 0, ("oplock_break: receive_smb error (%s)\n", strerror(errno)) );
-                               shutdown_server = True;
-                       } else if (smb_read_error == READ_TIMEOUT) {
-                               DEBUG( 0, ( "oplock_break: receive_smb timed out after %d seconds.\n", OPLOCK_BREAK_TIMEOUT ) );
-                               oplock_timeout = True;
-                       }
-
-                       DEBUGADD( 0, ( "oplock_break failed for file %s ", file_name ) );
-                       DEBUGADD( 0, ( "(dev = %x, inode = %.0f, file_id = %lu).\n",
-                               (unsigned int)dev, (double)inode, file_id));
-
-                       break;
-               }
+       msg.pid = src;
+       ADD_TO_ARRAY(NULL, struct share_mode_entry, msg,
+                    &fsp->pending_break_messages,
+                    &fsp->num_pending_break_messages);
 
-               /*
-                * There are certain SMB requests that we shouldn't allow
-                * to recurse. opens, renames and deletes are the obvious
-                * ones. This is handled in the switch_message() function.
-                * If global_oplock_break is set they will push the packet onto
-                * the pending smb queue and return -1 (no reply).
-                * JRA.
-                */
+       add_oplock_timeout_handler(fsp);
+}
 
-               process_smb(inbuf, outbuf);
+/*******************************************************************
+ This handles the kernel oplock break message.
+*******************************************************************/
 
-               /*
-                * Die if we go over the time limit.
-                */
+static void process_kernel_oplock_break(struct messaging_context *msg_ctx,
+                                       void *private_data,
+                                       uint32_t msg_type,
+                                       struct server_id src,
+                                       DATA_BLOB *data)
+{
+       struct file_id id;
+       unsigned long file_id;
+       files_struct *fsp;
+       char *break_msg;
+       BOOL sign_state;
 
-               if((time(NULL) - start_time) > OPLOCK_BREAK_TIMEOUT) {
-                       if( DEBUGLVL( 0 ) ) {
-                               dbgtext( "oplock_break: no break received from client " );
-                               dbgtext( "within %d seconds.\n", OPLOCK_BREAK_TIMEOUT );
-                               dbgtext( "oplock_break failed for file %s ", fsp->fsp_name );
-                               dbgtext( "(dev = %x, inode = %.0f, file_id = %lu).\n",  
-                                       (unsigned int)dev, (double)inode, file_id );
-                       }
-                       oplock_timeout = True;
-                       break;
-               }
+       if (data->data == NULL) {
+               DEBUG(0, ("Got NULL buffer\n"));
+               return;
        }
 
-       /*
-        * Go back to being the user who requested the oplock
-        * break.
-        */
-       if((saved_user_conn != NULL) && (saved_vuid != UID_FIELD_INVALID) && !change_to_user(saved_user_conn, saved_vuid)) {
-               DEBUG( 0, ( "oplock_break: unable to re-become user!" ) );
-               DEBUGADD( 0, ( "Shutting down server\n" ) );
-               close(oplock_sock);
-               exit_server("unable to re-become user");
+       if (data->length != MSG_SMB_KERNEL_BREAK_SIZE) {
+               DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
+               return;
        }
 
-       /* Including the directory. */
-       vfs_ChDir(saved_fsp_conn,saved_dir);
+       /* Pull the data from the message. */
+       pull_file_id_16((char *)data->data, &id);
+       file_id = (unsigned long)IVAL(data->data, 16);
 
-       /* Restore the chain fnum. */
-       file_chain_restore();
+       DEBUG(10, ("Got kernel oplock break message from pid %d: %s/%u\n",
+                  (int)procid_to_pid(&src), file_id_static_string(&id),
+                  (unsigned int)file_id));
 
-       /* Free the buffers we've been using to recurse. */
-       SAFE_FREE(inbuf);
-       SAFE_FREE(outbuf);
+       fsp = initial_break_processing(id, file_id);
 
-       /* We need this in case a readraw crossed on the wire. */
-       if(global_oplock_break)
-               global_oplock_break = False;
-
-       /*
-        * If the client timed out then clear the oplock (or go to level II)
-        * and continue. This seems to be what NT does and is better than dropping
-        * the connection.
-        */
-
-       if(oplock_timeout && (fsp = initial_break_processing(dev, inode, file_id)) &&
-                       OPEN_FSP(fsp) && EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
-               DEBUG(0,("oplock_break: client failure in oplock break in file %s\n", fsp->fsp_name));
-               remove_oplock(fsp,True);
-#if FASCIST_OPLOCK_BACKOFF
-               global_client_failed_oplock_break = True; /* Never grant this client an oplock again. */
-#endif
+       if (fsp == NULL) {
+               DEBUG(3, ("Got a kernel oplock break message for a file "
+                         "I don't know about\n"));
+               return;
        }
 
-       /*
-        * If the client had an error we must die.
-        */
-
-       if(shutdown_server) {
-               DEBUG( 0, ( "oplock_break: client failure in break - " ) );
-               DEBUGADD( 0, ( "shutting down this smbd.\n" ) );
-               close(oplock_sock);
-               exit_server("oplock break failure");
+       if (fsp->sent_oplock_break != NO_BREAK_SENT) {
+               /* This is ok, kernel oplocks come in completely async */
+               DEBUG(3, ("Got a kernel oplock request while waiting for a "
+                         "break reply\n"));
+               return;
        }
 
-       /* Santity check - remove this later. JRA */
-       if(exclusive_oplocks_open < 0) {
-               DEBUG(0,("oplock_break: exclusive_oplocks_open < 0 (%d). PANIC ERROR\n", exclusive_oplocks_open));
-               abort();
+       break_msg = new_break_smb_message(NULL, fsp, OPLOCKLEVEL_NONE);
+       if (break_msg == NULL) {
+               exit_server("Could not talloc break_msg\n");
        }
 
-       if( DEBUGLVL( 3 ) ) {
-               dbgtext( "oplock_break: returning success for " );
-               dbgtext( "dev = %x, inode = %.0f, file_id = %lu\n", (unsigned int)dev, (double)inode, file_id );
-               dbgtext( "Current exclusive_oplocks_open = %d\n", exclusive_oplocks_open );
-       }
+       /* Save the server smb signing state. */
+       sign_state = srv_oplock_set_signing(False);
 
-       return True;
-}
+       show_msg(break_msg);
+       if (!send_smb(smbd_server_fd(), break_msg)) {
+               exit_server_cleanly("oplock_break: send_smb failed.");
+       }
 
-/****************************************************************************
-Send an oplock break message to another smbd process. If the oplock is held 
-by the local smbd then call the oplock break function directly.
-****************************************************************************/
+       /* Restore the sign state to what it was. */
+       srv_oplock_set_signing(sign_state);
 
-BOOL request_oplock_break(share_mode_entry *share_entry)
-{
-       char op_break_msg[OPLOCK_BREAK_MSG_LEN];
-       struct sockaddr_in addr_out;
-       pid_t pid = sys_getpid();
-       time_t start_time;
-       int time_left;
-       SMB_DEV_T dev = share_entry->dev;
-       SMB_INO_T inode = share_entry->inode;
-       unsigned long file_id = share_entry->share_file_id;
-
-       if(pid == share_entry->pid) {
-               /* We are breaking our own oplock, make sure it's us. */
-               if(share_entry->op_port != global_oplock_port) {
-                       DEBUG(0,("request_oplock_break: corrupt share mode entry - pid = %d, port = %d \
-should be %d\n", (int)pid, share_entry->op_port, global_oplock_port));
-                       return False;
-               }
+       TALLOC_FREE(break_msg);
 
-               DEBUG(5,("request_oplock_break: breaking our own oplock\n"));
-
-#if 1 /* JRA PARANOIA TEST.... */
-               {
-                       files_struct *fsp = file_find_dif(dev, inode, file_id);
-                       if (!fsp) {
-                               DEBUG(0,("request_oplock_break: PANIC : breaking our own oplock requested for \
-dev = %x, inode = %.0f, file_id = %lu and no fsp found !\n",
-            (unsigned int)dev, (double)inode, file_id ));
-                               smb_panic("request_oplock_break: no fsp found for our own oplock\n");
-                       }
-               }
-#endif /* END JRA PARANOIA TEST... */
+       fsp->sent_oplock_break = BREAK_TO_NONE_SENT;
 
-               /* Call oplock break direct. */
-               return oplock_break(dev, inode, file_id, True);
-       }
+       add_oplock_timeout_handler(fsp);
+}
 
-       /* We need to send a OPLOCK_BREAK_CMD message to the port in the share mode entry. */
+void reply_to_oplock_break_requests(files_struct *fsp)
+{
+       int i;
 
-       if (LEVEL_II_OPLOCK_TYPE(share_entry->op_type)) {
-               SSVAL(op_break_msg,OPBRK_MESSAGE_CMD_OFFSET,LEVEL_II_OPLOCK_BREAK_CMD);
-       } else {
-               SSVAL(op_break_msg,OPBRK_MESSAGE_CMD_OFFSET,OPLOCK_BREAK_CMD);
-       }
+       for (i=0; i<fsp->num_pending_break_messages; i++) {
+               struct share_mode_entry *e = &fsp->pending_break_messages[i];
+               char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
 
-       memcpy(op_break_msg+OPLOCK_BREAK_PID_OFFSET,(char *)&pid,sizeof(pid));
-       memcpy(op_break_msg+OPLOCK_BREAK_DEV_OFFSET,(char *)&dev,sizeof(dev));
-       memcpy(op_break_msg+OPLOCK_BREAK_INODE_OFFSET,(char *)&inode,sizeof(inode));
-       memcpy(op_break_msg+OPLOCK_BREAK_FILEID_OFFSET,(char *)&file_id,sizeof(file_id));
+               share_mode_entry_to_message(msg, e);
 
-       /* Set the address and port. */
-       memset((char *)&addr_out,'\0',sizeof(addr_out));
-       addr_out.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
-       addr_out.sin_port = htons( share_entry->op_port );
-       addr_out.sin_family = AF_INET;
-   
-       if( DEBUGLVL( 3 ) ) {
-               dbgtext( "request_oplock_break: sending a oplock break message to " );
-               dbgtext( "pid %d on port %d ", (int)share_entry->pid, share_entry->op_port );
-               dbgtext( "for dev = %x, inode = %.0f, file_id = %lu\n",
-            (unsigned int)dev, (double)inode, file_id );
+               messaging_send_buf(smbd_messaging_context(), e->pid,
+                                  MSG_SMB_BREAK_RESPONSE,
+                                  (uint8 *)msg,
+                                  MSG_SMB_SHARE_MODE_ENTRY_SIZE);
        }
 
-       if(sys_sendto(oplock_sock,op_break_msg,OPLOCK_BREAK_MSG_LEN,0,
-                       (struct sockaddr *)&addr_out,sizeof(addr_out)) < 0) {
-               if( DEBUGLVL( 0 ) ) {
-                       dbgtext( "request_oplock_break: failed when sending a oplock " );
-                       dbgtext( "break message to pid %d ", (int)share_entry->pid );
-                       dbgtext( "on port %d ", share_entry->op_port );
-                       dbgtext( "for dev = %x, inode = %.0f, file_id = %lu\n",
-          (unsigned int)dev, (double)inode, file_id );
-                       dbgtext( "Error was %s\n", strerror(errno) );
-               }
-               return False;
+       SAFE_FREE(fsp->pending_break_messages);
+       fsp->num_pending_break_messages = 0;
+       if (fsp->oplock_timeout != NULL) {
+               /* Remove the timed event handler. */
+               TALLOC_FREE(fsp->oplock_timeout);
+               fsp->oplock_timeout = NULL;
        }
+       return;
+}
 
-       /*
-        * If we just sent a message to a level II oplock share entry then
-        * we are done and may return.
-        */
+static void process_oplock_break_response(struct messaging_context *msg_ctx,
+                                         void *private_data,
+                                         uint32_t msg_type,
+                                         struct server_id src,
+                                         DATA_BLOB *data)
+{
+       struct share_mode_entry msg;
 
-       if (LEVEL_II_OPLOCK_TYPE(share_entry->op_type)) {
-               DEBUG(3,("request_oplock_break: sent break message to level II entry.\n"));
-               return True;
+       if (data->data == NULL) {
+               DEBUG(0, ("Got NULL buffer\n"));
+               return;
        }
 
-       /*
-        * Now we must await the oplock broken message coming back
-        * from the target smbd process. Timeout if it fails to
-        * return in (OPLOCK_BREAK_TIMEOUT + OPLOCK_BREAK_TIMEOUT_FUDGEFACTOR) seconds.
-        * While we get messages that aren't ours, loop.
-        */
-
-       start_time = time(NULL);
-       time_left = OPLOCK_BREAK_TIMEOUT+OPLOCK_BREAK_TIMEOUT_FUDGEFACTOR;
-
-       while(time_left >= 0) {
-               char op_break_reply[OPBRK_CMD_HEADER_LEN+OPLOCK_BREAK_MSG_LEN];
-               uint16 reply_from_port;
-               char *reply_msg_start;
-
-               if(receive_local_message(op_break_reply, sizeof(op_break_reply),
-                               time_left ? time_left * 1000 : 1) == False) {
-                       if(smb_read_error == READ_TIMEOUT) {
-                               if( DEBUGLVL( 0 ) ) {
-                                       dbgtext( "request_oplock_break: no response received to oplock " );
-                                       dbgtext( "break request to pid %d ", (int)share_entry->pid );
-                                       dbgtext( "on port %d ", share_entry->op_port );
-                                       dbgtext( "for dev = %x, inode = %.0f, file_id = %lu\n",
-                                                       (unsigned int)dev, (double)inode, file_id );
-                               }
-
-                               /*
-                                * This is a hack to make handling of failing clients more robust.
-                                * If a oplock break response message is not received in the timeout
-                                * period we may assume that the smbd servicing that client holding
-                                * the oplock has died and the client changes were lost anyway, so
-                                * we should continue to try and open the file.
-                                */
-                               break;
-                       } else {
-                               if( DEBUGLVL( 0 ) ) {
-                                       dbgtext( "request_oplock_break: error in response received " );
-                                       dbgtext( "to oplock break request to pid %d ", (int)share_entry->pid );
-                                       dbgtext( "on port %d ", share_entry->op_port );
-                                       dbgtext( "for dev = %x, inode = %.0f, file_id = %lu\n",
-                                               (unsigned int)dev, (double)inode, file_id );
-                                       dbgtext( "Error was (%s).\n", strerror(errno) );
-                               }
-                       }
-                       return False;
-               }
-
-               reply_from_port = SVAL(op_break_reply,OPBRK_CMD_PORT_OFFSET);
-               reply_msg_start = &op_break_reply[OPBRK_CMD_HEADER_LEN];
-
-               /*
-                * Test to see if this is the reply we are awaiting.
-                */
-               if((SVAL(reply_msg_start,OPBRK_MESSAGE_CMD_OFFSET) & CMD_REPLY) &&
-                       ((SVAL(reply_msg_start,OPBRK_MESSAGE_CMD_OFFSET) & ~CMD_REPLY) == OPLOCK_BREAK_CMD) &&
-                       (reply_from_port == share_entry->op_port) && 
-                       (memcmp(&reply_msg_start[OPLOCK_BREAK_PID_OFFSET], &op_break_msg[OPLOCK_BREAK_PID_OFFSET],
-                               OPLOCK_BREAK_MSG_LEN - OPLOCK_BREAK_PID_OFFSET) == 0)) {
-
-                       /*
-                        * This is the reply we've been waiting for.
-                        */
-                       break;
-               } else {
-                       /*
-                        * This is another message - a break request.
-                        * Note that both kernel oplock break requests
-                        * and UDP inter-smbd oplock break requests will
-                        * be processed here.
-                        *
-                        * Process it to prevent potential deadlock.
-                        * Note that the code in switch_message() prevents
-                        * us from recursing into here as any SMB requests
-                        * we might process that would cause another oplock
-                        * break request to be made will be queued.
-                        * JRA.
-                        */
-
-                       process_local_message(op_break_reply, sizeof(op_break_reply));
-               }
-
-               time_left -= (time(NULL) - start_time);
+       if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
+               DEBUG(0, ("Got invalid msg len %u\n",
+                         (unsigned int)data->length));
+               return;
        }
 
-       DEBUG(3,("request_oplock_break: broke oplock.\n"));
+       /* De-linearize incoming message. */
+       message_to_share_mode_entry(&msg, (char *)data->data);
 
-       return True;
-}
+       DEBUG(10, ("Got oplock break response from pid %d: %s/%lu mid %u\n",
+                  (int)procid_to_pid(&src), file_id_static_string(&msg.id), msg.share_file_id,
+                  (unsigned int)msg.op_mid));
 
-/****************************************************************************
-  Attempt to break an oplock on a file (if oplocked).
-  Returns True if the file was closed as a result of
-  the oplock break, False otherwise.
-  Used as a last ditch attempt to free a space in the 
-  file table when we have run out.
-****************************************************************************/
+       /* Here's the hack from open.c, store the mid in the 'port' field */
+       schedule_deferred_open_smb_message(msg.op_mid);
+}
 
-BOOL attempt_close_oplocked_file(files_struct *fsp)
+static void process_open_retry_message(struct messaging_context *msg_ctx,
+                                      void *private_data,
+                                      uint32_t msg_type,
+                                      struct server_id src,
+                                      DATA_BLOB *data)
 {
-       DEBUG(5,("attempt_close_oplocked_file: checking file %s.\n", fsp->fsp_name));
+       struct share_mode_entry msg;
+       
+       if (data->data == NULL) {
+               DEBUG(0, ("Got NULL buffer\n"));
+               return;
+       }
 
-       if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type) && !fsp->sent_oplock_break && (fsp->fd != -1)) {
-               /* Try and break the oplock. */
-               if (oplock_break(fsp->dev, fsp->inode, fsp->file_id, True)) {
-                       if(file_find_fsp(fsp) == NULL) /* Did the oplock break close the file ? */
-                               return True;
-               }
+       if (data->length != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
+               DEBUG(0, ("Got invalid msg len %d\n", (int)data->length));
+               return;
        }
 
-       return False;
+       /* De-linearize incoming message. */
+       message_to_share_mode_entry(&msg, (char *)data->data);
+
+       DEBUG(10, ("Got open retry msg from pid %d: %s mid %u\n",
+                  (int)procid_to_pid(&src), file_id_static_string(&msg.id),
+                  (unsigned int)msg.op_mid));
+
+       schedule_deferred_open_smb_message(msg.op_mid);
 }
 
 /****************************************************************************
  This function is called on any file modification or lock request. If a file
- is level 2 oplocked then it must tell all other level 2 holders to break to none.
+ is level 2 oplocked then it must tell all other level 2 holders to break to
+ none.
 ****************************************************************************/
 
 void release_level_2_oplocks_on_change(files_struct *fsp)
 {
-       share_mode_entry *share_list = NULL;
-       pid_t pid = sys_getpid();
-       int token = -1;
-       int num_share_modes = 0;
        int i;
+       struct share_mode_lock *lck;
 
        /*
         * If this file is level II oplocked then we need
@@ -1124,112 +755,130 @@ void release_level_2_oplocks_on_change(files_struct *fsp)
        if (!LEVEL_II_OPLOCK_TYPE(fsp->oplock_type))
                return;
 
-       if (lock_share_entry_fsp(fsp) == False) {
-               DEBUG(0,("release_level_2_oplocks_on_change: failed to lock share mode entry for file %s.\n", fsp->fsp_name ));
+       lck = get_share_mode_lock(NULL, fsp->file_id, NULL, NULL);
+       if (lck == NULL) {
+               DEBUG(0,("release_level_2_oplocks_on_change: failed to lock "
+                        "share mode entry for file %s.\n", fsp->fsp_name ));
+               return;
        }
 
-       num_share_modes = get_share_modes(fsp->conn, fsp->dev, fsp->inode, &share_list);
-
        DEBUG(10,("release_level_2_oplocks_on_change: num_share_modes = %d\n", 
-                       num_share_modes ));
+                 lck->num_share_modes ));
+
+       for(i = 0; i < lck->num_share_modes; i++) {
+               struct share_mode_entry *share_entry = &lck->share_modes[i];
+               char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
 
-       for(i = 0; i < num_share_modes; i++) {
-               share_mode_entry *share_entry = &share_list[i];
+               if (!is_valid_share_mode_entry(share_entry)) {
+                       continue;
+               }
 
                /*
-                * As there could have been multiple writes waiting at the lock_share_entry
-                * gate we may not be the first to enter. Hence the state of the op_types
-                * in the share mode entries may be partly NO_OPLOCK and partly LEVEL_II
-                * oplock. It will do no harm to re-send break messages to those smbd's
-                * that are still waiting their turn to remove their LEVEL_II state, and
-                * also no harm to ignore existing NO_OPLOCK states. JRA.
+                * As there could have been multiple writes waiting at the
+                * lock_share_entry gate we may not be the first to
+                * enter. Hence the state of the op_types in the share mode
+                * entries may be partly NO_OPLOCK and partly LEVEL_II or FAKE_LEVEL_II
+                * oplock. It will do no harm to re-send break messages to
+                * those smbd's that are still waiting their turn to remove
+                * their LEVEL_II state, and also no harm to ignore existing
+                * NO_OPLOCK states. JRA.
                 */
 
-               DEBUG(10,("release_level_2_oplocks_on_change: share_entry[%i]->op_type == %d\n",
-                               i, share_entry->op_type ));
+               DEBUG(10,("release_level_2_oplocks_on_change: "
+                         "share_entry[%i]->op_type == %d\n",
+                         i, share_entry->op_type ));
 
-               if (share_entry->op_type == NO_OPLOCK)
+               if (share_entry->op_type == NO_OPLOCK) {
                        continue;
+               }
 
                /* Paranoia .... */
                if (EXCLUSIVE_OPLOCK_TYPE(share_entry->op_type)) {
-                       DEBUG(0,("release_level_2_oplocks_on_change: PANIC. share mode entry %d is an exlusive oplock !\n", i ));
-                       unlock_share_entry(fsp->conn, fsp->dev, fsp->inode);
+                       DEBUG(0,("release_level_2_oplocks_on_change: PANIC. "
+                                "share mode entry %d is an exlusive "
+                                "oplock !\n", i ));
+                       TALLOC_FREE(lck);
                        abort();
                }
 
-               /*
-                * Check if this is a file we have open (including the
-                * file we've been called to do write_file on. If so
-                * then break it directly without releasing the lock.
-                */
+               share_mode_entry_to_message(msg, share_entry);
 
-               if (pid == share_entry->pid) {
-                       files_struct *new_fsp = file_find_dif(share_entry->dev, share_entry->inode, share_entry->share_file_id);
-
-                       /* Paranoia check... */
-                       if(new_fsp == NULL) {
-                               DEBUG(0,("release_level_2_oplocks_on_change: PANIC. share mode entry %d is not a local file !\n", i ));
-                               unlock_share_entry(fsp->conn, fsp->dev, fsp->inode);
-                               abort();
-                       }
-
-                       DEBUG(10,("release_level_2_oplocks_on_change: breaking our own oplock.\n"));
+               messaging_send_buf(smbd_messaging_context(), share_entry->pid,
+                                  MSG_SMB_ASYNC_LEVEL2_BREAK,
+                                  (uint8 *)msg,
+                                  MSG_SMB_SHARE_MODE_ENTRY_SIZE);
+       }
 
-                       oplock_break_level2(new_fsp, True, token);
+       /* We let the message receivers handle removing the oplock state
+          in the share mode lock db. */
 
-               } else {
+       TALLOC_FREE(lck);
+}
 
-                       /*
-                        * This is a remote file and so we send an asynchronous
-                        * message.
-                        */
+/****************************************************************************
+ Linearize a share mode entry struct to an internal oplock break message.
+****************************************************************************/
 
-                       DEBUG(10,("release_level_2_oplocks_on_change: breaking remote oplock.\n"));
-                       request_oplock_break(share_entry);
-               }
-       }
+void share_mode_entry_to_message(char *msg, struct share_mode_entry *e)
+{
+       SIVAL(msg,0,(uint32)e->pid.pid);
+       SSVAL(msg,4,e->op_mid);
+       SSVAL(msg,6,e->op_type);
+       SIVAL(msg,8,e->access_mask);
+       SIVAL(msg,12,e->share_access);
+       SIVAL(msg,16,e->private_options);
+       SIVAL(msg,20,(uint32)e->time.tv_sec);
+       SIVAL(msg,24,(uint32)e->time.tv_usec);
+       push_file_id_16(msg+28, &e->id);
+       SIVAL(msg,44,e->share_file_id);
+       SIVAL(msg,48,e->uid);
+       SSVAL(msg,52,e->flags);
+#ifdef CLUSTER_SUPPORT
+       SIVAL(msg,54,e->pid.vnn);
+#endif
+}
 
-       SAFE_FREE(share_list);
-       unlock_share_entry_fsp(fsp);
+/****************************************************************************
+ De-linearize an internal oplock break message to a share mode entry struct.
+****************************************************************************/
 
-       /* Paranoia check... */
-       if (LEVEL_II_OPLOCK_TYPE(fsp->oplock_type)) {
-               DEBUG(0,("release_level_2_oplocks_on_change: PANIC. File %s still has a level II oplock.\n", fsp->fsp_name));
-               smb_panic("release_level_2_oplocks_on_change");
-       }
+void message_to_share_mode_entry(struct share_mode_entry *e, char *msg)
+{
+       e->pid.pid = (pid_t)IVAL(msg,0);
+       e->op_mid = SVAL(msg,4);
+       e->op_type = SVAL(msg,6);
+       e->access_mask = IVAL(msg,8);
+       e->share_access = IVAL(msg,12);
+       e->private_options = IVAL(msg,16);
+       e->time.tv_sec = (time_t)IVAL(msg,20);
+       e->time.tv_usec = (int)IVAL(msg,24);
+       pull_file_id_16(msg+28, &e->id);
+       e->share_file_id = (unsigned long)IVAL(msg,44);
+       e->uid = (uint32)IVAL(msg,48);
+       e->flags = (uint16)SVAL(msg,52);
+#ifdef CLUSTER_SUPPORT
+       e->pid.vnn = IVAL(msg,54);
+#endif
 }
 
 /****************************************************************************
-setup oplocks for this process
+ Setup oplocks for this process.
 ****************************************************************************/
 
-BOOL init_oplocks(void)
+BOOL init_oplocks(struct messaging_context *msg_ctx)
 {
-       struct sockaddr_in sock_name;
-       socklen_t len = sizeof(sock_name);
-
-       DEBUG(3,("open_oplock_ipc: opening loopback UDP socket.\n"));
-
-       /* Open a lookback UDP socket on a random port. */
-       oplock_sock = open_socket_in(SOCK_DGRAM, 0, 0, htonl(INADDR_LOOPBACK),False);
-       if (oplock_sock == -1) {
-               DEBUG(0,("open_oplock_ipc: Failed to get local UDP socket for \
-address %lx. Error was %s\n", (long)htonl(INADDR_LOOPBACK), strerror(errno)));
-               global_oplock_port = 0;
-               return(False);
-       }
-
-       /* Find out the transient UDP port we have been allocated. */
-       if(getsockname(oplock_sock, (struct sockaddr *)&sock_name, &len)<0) {
-               DEBUG(0,("open_oplock_ipc: Failed to get local UDP port. Error was %s\n",
-                        strerror(errno)));
-               close(oplock_sock);
-               oplock_sock = -1;
-               global_oplock_port = 0;
-               return False;
-       }
-       global_oplock_port = ntohs(sock_name.sin_port);
+       DEBUG(3,("init_oplocks: initializing messages.\n"));
+
+       messaging_register(msg_ctx, NULL, MSG_SMB_BREAK_REQUEST,
+                          process_oplock_break_message);
+       messaging_register(msg_ctx, NULL, MSG_SMB_ASYNC_LEVEL2_BREAK,
+                          process_oplock_async_level2_break_message);
+       messaging_register(msg_ctx, NULL, MSG_SMB_BREAK_RESPONSE,
+                          process_oplock_break_response);
+       messaging_register(msg_ctx, NULL, MSG_SMB_KERNEL_BREAK,
+                          process_kernel_oplock_break);
+       messaging_register(msg_ctx, NULL, MSG_SMB_OPEN_RETRY,
+                          process_open_retry_message);
 
        if (lp_kernel_oplocks()) {
 #if HAVE_KERNEL_OPLOCKS_IRIX
@@ -1239,8 +888,5 @@ address %lx. Error was %s\n", (long)htonl(INADDR_LOOPBACK), strerror(errno)));
 #endif
        }
 
-       DEBUG(3,("open_oplock ipc: pid = %d, global_oplock_port = %u\n", 
-                (int)sys_getpid(), global_oplock_port));
-
        return True;
 }