r10656: BIG merge from trunk. Features not copied over
[samba.git] / source3 / smbd / oplock.c
index c0c9e989a9970abe5938e7c5bdc482adc7ec8077..385f998b1ce31ac46a8cc19c056c984175b0b3c8 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
 
 #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 struct timeval smb_last_time;
 extern uint32 global_client_caps;
-extern struct current_user current_user;
 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.
 ****************************************************************************/
@@ -58,9 +51,6 @@ BOOL oplock_message_waiting(fd_set *fds)
        if (koplocks && koplocks->msg_waiting(fds))
                return True;
 
-       if (FD_ISSET(oplock_sock, fds))
-               return True;
-
        return False;
 }
 
@@ -75,13 +65,9 @@ BOOL oplock_message_waiting(fd_set *fds)
 
 ****************************************************************************/
 
-BOOL receive_local_message( char *buffer, int buffer_len, int timeout)
+void process_kernel_oplocks(void)
 {
-       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;
@@ -92,110 +78,29 @@ BOOL receive_local_message( char *buffer, int buffer_len, int timeout)
         * 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);
+       while (koplocks->msg_waiting(&fds)) { 
+               files_struct *fsp;
+               struct kernel_oplock_message msg;
 
-               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);
-                       }
+               fsp = koplocks->receive_message(&fds);
 
-                       /*
-                        * 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;
-                       }
-
-                       /* 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;
-
-                       DEBUG(5,("receive_local_message: EINTR : new timeout %d ms\n", timeout));
-                       continue;
+               if (fsp == NULL) {
+                       DEBUG(3, ("Kernel oplock message announced, but none "
+                                 "received\n"));
+                       return;
                }
 
-               /* 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 (koplocks && koplocks->msg_waiting(&fds)) {
-               return koplocks->receive_message(&fds, buffer, buffer_len);
+               msg.dev = fsp->dev;
+               msg.inode = fsp->inode;
+               msg.file_id = fsp->file_id;
+               message_send_pid(pid_to_procid(sys_getpid()),
+                                MSG_SMB_KERNEL_BREAK,
+                                &msg, sizeof(msg), True);
        }
-
-       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;
-       }
-
-       /* 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;
-       }
-
-       /* 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;
-       }
-
-       /* Setup the message header */
-       SIVAL(buffer,OPBRK_CMD_LEN_OFFSET,msg_len);
-       SSVAL(buffer,OPBRK_CMD_PORT_OFFSET,ntohs(from.sin_port));
-
-       return True;
 }
 
 /****************************************************************************
@@ -229,13 +134,19 @@ tv_sec = %x, tv_usec = %x\n",
 
 void release_file_oplock(files_struct *fsp)
 {
-       if ((fsp->oplock_type != NO_OPLOCK) && 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)
                level_II_oplocks_open--;
-       else if (fsp->oplock_type)
+       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;
@@ -263,45 +174,58 @@ 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 ));
+       lck = get_share_mode_lock(NULL, fsp->dev, fsp->inode, NULL);
+       if (lck == NULL) {
+               DEBUG(0,("remove_oplock: failed to lock share entry for "
+                        "file %s\n", fsp->fsp_name ));
                return False;
        }
+       ret = remove_share_oplock(lck, fsp);
+       if (!ret) {
+               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));
+       }
+       release_file_oplock(fsp);
+       talloc_free(lck);
+       return ret;
+}
 
-       if (fsp->sent_oplock_break == BREAK_TO_NONE_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;
-               }
+/*
+ * Deal with a reply when a break-to-level II was sent.
+ */
+BOOL downgrade_oplock(files_struct *fsp)
+{
+       SMB_DEV_T dev = fsp->dev;
+       SMB_INO_T inode = fsp->inode;
+       BOOL ret;
+       struct share_mode_lock *lck;
 
-               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);
+       lck = get_share_mode_lock(NULL, fsp->dev, fsp->inode, NULL);
+       if (lck == NULL) {
+               DEBUG(0,("downgrade_oplock: failed to lock share entry for "
+                        "file %s\n", fsp->fsp_name ));
+               return False;
        }
-
-       unlock_share_entry_fsp(fsp);
+       ret = downgrade_share_oplock(lck, fsp);
+       if (!ret) {
+               DEBUG(0,("downgrade_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));
+       }
+               
+       downgrade_file_oplock(fsp);
+       talloc_free(lck);
        return ret;
 }
 
@@ -313,12 +237,7 @@ dev = %x, inode = %.0f\n", fsp->fsp_name, fsp->fnum, (unsigned int)dev, (double)
 
 int setup_oplock_select_set( fd_set *fds)
 {
-       int maxfd = oplock_sock;
-
-       if(oplock_sock == -1)
-               return 0;
-
-       FD_SET(oplock_sock,fds);
+       int maxfd = 0;
 
        if (koplocks && koplocks->notification_fd != -1) {
                FD_SET(koplocks->notification_fd, fds);
@@ -329,197 +248,31 @@ int setup_oplock_select_set( fd_set *fds)
 }
 
 /****************************************************************************
- Process an oplock break message - whether it came from the UDP socket
- or from the kernel.
+ Set up an oplock break message.
 ****************************************************************************/
 
-BOOL process_local_message(char *buffer, int buf_size)
+static char *new_break_smb_message(TALLOC_CTX *mem_ctx,
+                                  files_struct *fsp, uint8_t cmd)
 {
-       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;
-       struct sockaddr_in toaddr;
-
-       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"));
-                               return False;
-                       }
-                       break;
-
-               case OPLOCK_BREAK_CMD:
-               case LEVEL_II_OPLOCK_BREAK_CMD:
-               case ASYNC_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;
-
-               case RETRY_DEFERRED_OPEN_CMD:
-
-                       /* Request to retry and open that would return SHARING_VIOLATION. */
-                       if (msg_len != DEFERRED_OPEN_MSG_LEN) {
-                               DEBUG(0,("process_local_message: incorrect length for RETRY_DEFERRED_OPEN_CMD (was %d, should be %d).\n",
-                                       (int)msg_len, (int)DEFERRED_OPEN_MSG_LEN));
-                               return False;
-                       }
-                       {
-                               uint16 mid;
-
-                               memcpy((char *)&remotepid, msg_start+DEFERRED_OPEN_PID_OFFSET,sizeof(remotepid));
-                               memcpy((char *)&inode, msg_start+DEFERRED_OPEN_INODE_OFFSET,sizeof(inode));
-                               memcpy((char *)&dev, msg_start+DEFERRED_OPEN_DEV_OFFSET,sizeof(dev));
-                               memcpy((char *)&mid, msg_start+DEFERRED_OPEN_MID_OFFSET,sizeof(mid));
-
-                               DEBUG(5,("process_local_message: RETRY_DEFERRED_OPEN from \
-pid %d, port %d, dev = %x, inode = %.0f, mid = %u\n",
-                                       (int)remotepid, from_port, (unsigned int)dev, (double)inode, (unsigned int)mid));
-
-                               schedule_sharing_violation_open_smb_message(mid);
-                       }
-                       return True;
-
-               /* 
-                * 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));
+       char *result = TALLOC_ARRAY(mem_ctx, char, smb_size + 8*2 + 0);
 
-                       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 == 0) &&
-           (level_II_oplocks_open == 0)) {
-               /*
-                * 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"));
-
-       } else {
-               if (!oplock_break(dev, inode, file_id, False)) {
-                       DEBUG(0,("process_local_message: oplock break failed.\n"));
-                       return False;
-               }
-       }
-
-       /* 
-        * Do the appropriate reply - none in the kernel or async level II
-        * case.
-        */
-
-       if (!((break_cmd_type == OPLOCK_BREAK_CMD) ||
-             (break_cmd_type == LEVEL_II_OPLOCK_BREAK_CMD))) {
-               return True;
-       }
-
-       /* Send the message back after OR'ing in the 'REPLY' bit. */
-       SSVAL(msg_start,OPBRK_MESSAGE_CMD_OFFSET,break_cmd_type | 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;
+       if (result == NULL) {
+               DEBUG(0, ("talloc failed\n"));
+               return NULL;
        }
 
-       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;
-}
-
-/****************************************************************************
- Set up an oplock break message.
-****************************************************************************/
-
-static void prepare_break_message(char *outbuf, files_struct *fsp, BOOL level2)
-{
-       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);
+       memset(result,'\0',smb_size);
+       set_message(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;
 }
 
 /****************************************************************************
@@ -602,639 +355,266 @@ static files_struct *initial_break_processing(SMB_DEV_T dev, SMB_INO_T inode, un
        return fsp;
 }
 
-/****************************************************************************
- Process a level II oplock break directly.
- We must call this function with the share mode entry locked.
-****************************************************************************/
-
-static BOOL oplock_break_level2(files_struct *fsp, BOOL local_request)
+static void oplock_timeout_handler(struct timed_event *te,
+                                  const struct timeval *now,
+                                  void *private_data)
 {
-       char outbuf[128];
-       SMB_DEV_T dev = fsp->dev;
-       SMB_INO_T inode = fsp->inode;
-
-       /*
-        * 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.
-        */
-
-       if (global_client_caps & CAP_LEVEL_II_OPLOCKS) {
-               BOOL sign_state;
-
-               /*
-                * 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 (local_request) {
-                       wait_before_sending_break();
-               }
-
-               /* Prepare the SMBlockingX message. */
-               prepare_break_message( outbuf, fsp, False);
-
-               /* Save the server smb signing state. */
-               sign_state = srv_oplock_set_signing(False);
-
-               show_msg(outbuf);
-               if (!send_smb(smbd_server_fd(), outbuf))
-                       exit_server("oplock_break_level2: send_smb failed.");
+       files_struct *fsp = private_data;
 
-               /* Restore the sign state to what it was. */
-               srv_oplock_set_signing(sign_state);
-       }
-
-       /*
-        * Now we must update the shared memory structure to tell
-        * everyone else we no longer have a level II oplock on 
-        * this open file. We must call this function with the share mode
-        * entry locked so we can change the entry directly.
-        */
-
-       if(remove_share_oplock(fsp)==False) {
-               DEBUG(0,("oplock_break_level2: unable to remove level II oplock for file %s\n", fsp->fsp_name ));
-       }
-
-       release_file_oplock(fsp);
-
-       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();
-       }
-
-       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 );
-       }
-
-       return True;
+       DEBUG(0, ("Oplock break failed -- replying anyway\n"));
+       global_client_failed_oplock_break = True;
+       remove_oplock(fsp);
+       reply_to_oplock_break_requests(fsp);
 }
 
-/****************************************************************************
- Process an oplock break directly.
- This is always called with the share mode lock *NOT* held.
-****************************************************************************/
-
-static BOOL oplock_break(SMB_DEV_T dev, SMB_INO_T inode, unsigned long file_id, BOOL local_request)
+static void process_oplock_break_message(int msg_type, struct process_id src,
+                                        void *buf, size_t len)
 {
-       char *inbuf = NULL;
-       char *saved_inbuf = NULL;
-       char *outbuf = NULL;
-       char *saved_outbuf = NULL;
-       files_struct *fsp = NULL;
-       time_t start_time;
-       BOOL shutdown_server = False;
-       BOOL oplock_timeout = False;
+       struct share_mode_entry *msg = buf;
+       files_struct *fsp;
+       char *break_msg;
+       BOOL break_to_level2 = False;
        BOOL sign_state;
-       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;
 
-       /*
-        * Deal with a level II oplock going break to none separately.
-        */
-
-       if (LEVEL_II_OPLOCK_TYPE(fsp->oplock_type)) {
-               BOOL ret;
-               /* We must always call oplock_break_level2() with
-                  the share mode entry locked. */
-               if (lock_share_entry_fsp(fsp) == False) {
-                       DEBUG(0,("oplock_break: unable to lock share entry for file %s\n", fsp->fsp_name ));
-                       return False;
-               }
-               ret = oplock_break_level2(fsp, local_request);
-               unlock_share_entry_fsp(fsp);
-               return ret;
+       if (buf == NULL) {
+               DEBUG(0, ("Got NULL buffer\n"));
+               return;
        }
 
-       /* 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 );
-               }
-
-               /*
-                * 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;
+       if (len != sizeof(*msg)) {
+               DEBUG(0, ("Got invalid msg len %d\n", (int)len));
+               return;
        }
 
-       if(global_oplock_break) {
-               DEBUG(0,("ABORT : ABORT : recursion in oplock_break !!!!!\n"));
-               abort();
-       }
+       DEBUG(10, ("Got oplock break message from pid %d: %d/%d/%d\n",
+                  (int)procid_to_pid(&src), (int)msg->dev, (int)msg->inode,
+                  (int)msg->share_file_id));
 
-       /*
-        * 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.
-        */
+       fsp = initial_break_processing(msg->dev, msg->inode,
+                                      msg->share_file_id);
 
-       if((inbuf = NewInBuffer(&saved_inbuf))==NULL) {
-               DEBUG(0,("oplock_break: malloc fail for input buffer.\n"));
-               return False;
+       if (fsp == NULL) {
+               /* 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"));
+               message_send_pid(src, MSG_SMB_BREAK_RESPONSE,
+                                msg, sizeof(*msg), True);
+               return;
        }
 
-       if((outbuf = NewOutBuffer(&saved_outbuf))==NULL) {
-               DEBUG(0,("oplock_break: malloc fail for output buffer.\n"));
-               /* Free must be done before set.. */
-               free_InBuffer(inbuf);
-               set_InBuffer(saved_inbuf);
-               return False;
+       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 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 (local_request) {
-               wait_before_sending_break();
+       if (EXCLUSIVE_OPLOCK_TYPE(msg->op_type) &&
+           !EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
+               DEBUG(3, ("Already downgraded oplock on %.0f/%.0f: %s\n",
+                         (double)fsp->dev, (double)fsp->inode,
+                         fsp->fsp_name));
+               message_send_pid(src, MSG_SMB_BREAK_RESPONSE,
+                                msg, sizeof(*msg), True);
+               return;
        }
 
-       /* Prepare the SMBlockingX message. */
+       if ((msg_type == MSG_SMB_BREAK_REQUEST) &&
+           (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))) {
+               break_to_level2 = True;
+       }
 
-       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;
+       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");
        }
 
-       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:BREAK_TO_NONE_SENT;
+       /* Need to wait before sending a break message to a file of our own */
+       if (procid_to_pid(&src) == sys_getpid()) {
+               wait_before_sending_break();
+       }
 
        /* Save the server smb signing state. */
        sign_state = srv_oplock_set_signing(False);
 
-       show_msg(outbuf);
-       if (!send_smb(smbd_server_fd(), outbuf)) {
-               srv_oplock_set_signing(sign_state);
+       show_msg(break_msg);
+       if (!send_smb(smbd_server_fd(), break_msg)) {
                exit_server("oplock_break: send_smb failed.");
        }
 
        /* Restore the sign state to what it was. */
        srv_oplock_set_signing(sign_state);
 
-       /* We need this in case a readraw crosses on the wire. */
-       global_oplock_break = True;
-       /* Process incoming messages. */
-
-       /*
-        * JRA - If we don't get a break from the client in OPLOCK_BREAK_TIMEOUT
-        * seconds we should just die....
-        */
-
-       start_time = time(NULL);
+       talloc_free(break_msg);
 
-       /*
-        * 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;
-       /*
-        * Initialize saved_dir to something sensible: vfs_GetWd may not work well
-        * for root: the directory may be NFS-mounted and exported with root_squash
-        * (so has no root access).
-        */
-       pstrcpy(saved_dir,saved_fsp_conn->connectpath);
-       vfs_GetWd(saved_fsp_conn,saved_dir);
-       /* Save the chain fnum. */
-       file_chain_save();
-
-       pstrcpy(file_name, fsp->fsp_name);
-
-       change_to_root_user();
-
-       /*
-        * 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. With async I/O, write calls may steal the global InBuffer,
-        * so ensure we're using the correct one each time around the loop.
-        */
-
-       while((fsp = initial_break_processing(dev, inode, file_id)) &&
-                       OPEN_FSP(fsp) && EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
-
-               inbuf = get_InBuffer();
-               outbuf = get_OutBuffer();
-
-               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_BAD_SIG) {
-                               DEBUG( 0, ("oplock_break: bad signature from client\n" ));
-                               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));
+       if (msg_type == MSG_SMB_BREAK_REQUEST) {
+               fsp->sent_oplock_break = break_to_level2 ?
+                       LEVEL_II_BREAK_SENT:BREAK_TO_NONE_SENT;
+       } else {
+               /* Async level2 request, don't send a reply */
+               fsp->sent_oplock_break = ASYNC_LEVEL_II_BREAK_SENT;
+       }
+       msg->pid = src;
+       ADD_TO_ARRAY(NULL, struct share_mode_entry, *msg,
+                    &fsp->pending_break_messages,
+                    &fsp->num_pending_break_messages);
 
-                       break;
-               }
+       if (fsp->oplock_timeout != NULL) {
+               DEBUG(0, ("Logic problem -- have an oplock event hanging "
+                         "around\n"));
+       }
 
-               /*
-                * 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.
-                */
+       fsp->oplock_timeout =
+               add_timed_event(NULL,
+                               timeval_current_ofs(OPLOCK_BREAK_TIMEOUT, 0),
+                               "oplock_timeout_handler",
+                               oplock_timeout_handler, fsp);
 
-               process_smb(inbuf, outbuf);
+       if (fsp->oplock_timeout == NULL) {
+               DEBUG(0, ("Could not add oplock timeout handler\n"));
+       }
+}
 
-               /*
-                * Die if we go over the time limit.
-                */
+static void process_kernel_oplock_break(int msg_type, struct process_id src,
+                                       void *buf, size_t len)
+{
+       struct kernel_oplock_message *msg = buf;
+       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 (buf == 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 (len != sizeof(*msg)) {
+               DEBUG(0, ("Got invalid msg len %d\n", (int)len));
+               return;
        }
 
-       /* Including the directory. */
-       vfs_ChDir(saved_fsp_conn,saved_dir);
-
-       /* Restore the chain fnum. */
-       file_chain_restore();
+       DEBUG(10, ("Got kernel oplock break message from pid %d: %d/%d/%d\n",
+                  (int)procid_to_pid(&src), (int)msg->dev, (int)msg->inode,
+                  (int)msg->file_id));
 
-       /* Free the buffers we've been using to recurse. */
-       /* Free must be done before set.. */
-       free_InBuffer(inbuf);
-       free_OutBuffer(outbuf);
+       fsp = initial_break_processing(msg->dev, msg->inode, msg->file_id);
 
-       /* Restore the global In/Out buffers. */
-       set_InBuffer(saved_inbuf);
-       set_OutBuffer(saved_outbuf);
-
-       /* 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");
        }
 
-       /* We know we have no saved errors here. */
-       set_saved_error_triple(0, 0, NT_STATUS_OK);
+       /* Save the server smb signing state. */
+       sign_state = srv_oplock_set_signing(False);
 
-       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 );
+       show_msg(break_msg);
+       if (!send_smb(smbd_server_fd(), break_msg)) {
+               exit_server("oplock_break: send_smb failed.");
        }
 
-       return True;
-}
-
-/****************************************************************************
- 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.
- This function is called with no share locks held.
-****************************************************************************/
+       /* 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;
-       uint16 break_cmd_type;
-
-       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"));
+       fsp->sent_oplock_break = BREAK_TO_NONE_SENT;
+}
 
-#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... */
+void reply_to_oplock_break_requests(files_struct *fsp)
+{
+       int i;
 
-               /* Call oplock break direct. */
-               return oplock_break(dev, inode, file_id, True);
+       for (i=0; i<fsp->num_pending_break_messages; i++) {
+               struct share_mode_entry *msg = &fsp->pending_break_messages[i];
+               message_send_pid(msg->pid, MSG_SMB_BREAK_RESPONSE,
+                                msg, sizeof(*msg), True);
        }
 
-       /* We need to send a OPLOCK_BREAK_CMD message to the port in the share mode entry. */
-
-       if (LEVEL_II_OPLOCK_TYPE(share_entry->op_type)) {
-               break_cmd_type = LEVEL_II_OPLOCK_BREAK_CMD;
-       } else {
-               break_cmd_type = OPLOCK_BREAK_CMD;
+       SAFE_FREE(fsp->pending_break_messages);
+       fsp->num_pending_break_messages = 0;
+       if (fsp->oplock_timeout != NULL) {
+               talloc_free(fsp->oplock_timeout);
+               fsp->oplock_timeout = NULL;
        }
+       return;
+}
 
-       SSVAL(op_break_msg,OPBRK_MESSAGE_CMD_OFFSET,break_cmd_type);
-       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));
-
-       /* 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 synchronous 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 );
-       }
+static void process_oplock_break_response(int msg_type, struct process_id src,
+                                         void *buf, size_t len)
+{
+       struct share_mode_entry *msg = buf;
 
-       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;
+       if (buf == 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 (ie. the one we sent with the CMD_REPLY flag OR'ed in).
-                */
-               if((SVAL(reply_msg_start,OPBRK_MESSAGE_CMD_OFFSET) & CMD_REPLY) &&
-                       ((SVAL(reply_msg_start,OPBRK_MESSAGE_CMD_OFFSET) & ~CMD_REPLY) == break_cmd_type) &&
-                       (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 (len != sizeof(*msg)) {
+               DEBUG(0, ("Got invalid msg len %d\n", (int)len));
+               return;
        }
 
-       DEBUG(3,("request_oplock_break: broke oplock.\n"));
+       DEBUG(10, ("Got oplock break response from pid %d: %d/%d/%d mid %d\n",
+                  (int)procid_to_pid(&src), (int)msg->dev, (int)msg->inode,
+                  (int)msg->share_file_id, (int)msg->op_mid));
 
-       return True;
+       /* Here's the hack from open.c, store the mid in the 'port' field */
+       schedule_deferred_open_smb_message(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.
-****************************************************************************/
-
-BOOL attempt_close_oplocked_file(files_struct *fsp)
+static void process_open_retry_message(int msg_type, struct process_id src,
+                                      void *buf, size_t len)
 {
-       DEBUG(5,("attempt_close_oplocked_file: checking file %s.\n", fsp->fsp_name));
-
-       if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type) && !fsp->sent_oplock_break && (fsp->fh->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;
-               }
+       struct share_mode_entry *msg = buf;
+       
+       if (buf == NULL) {
+               DEBUG(0, ("Got NULL buffer\n"));
+               return;
        }
 
-       return False;
-}
-
-/****************************************************************************
- Send an asynchronous oplock break message to another smbd process.
-****************************************************************************/
-
-static BOOL request_remote_level2_async_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();
-       SMB_DEV_T dev = share_entry->dev;
-       SMB_INO_T inode = share_entry->inode;
-       unsigned long file_id = share_entry->share_file_id;
-
-       /* We need to send a ASYNC_LEVEL_II_OPLOCK_BREAK_CMD message to the port in the share mode entry. */
-
-       SSVAL(op_break_msg,OPBRK_MESSAGE_CMD_OFFSET,ASYNC_LEVEL_II_OPLOCK_BREAK_CMD);
-       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));
-
-       /* 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_remote_level2_async_oplock_break: sending an asynchronous 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 );
+       if (len != sizeof(*msg)) {
+               DEBUG(0, ("Got invalid msg len %d\n", (int)len));
+               return;
        }
 
-       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_remote_level2_async_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;
-       }
+       DEBUG(10, ("Got open retry msg from pid %d: %d/%d mid %d\n",
+                  (int)procid_to_pid(&src), (int)msg->dev, (int)msg->inode,
+                  (int)msg->op_mid));
 
-       DEBUG(3,("request_remote_level2_async_oplock_break: sent async break message to level II entry.\n"));
-       return True;
+       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 num_share_modes = 0;
        int i;
-       BOOL dummy;
+       struct share_mode_lock *lck;
 
        /*
         * If this file is level II oplocked then we need
@@ -1247,125 +627,71 @@ 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->dev, fsp->inode, 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 ));
        }
 
-       num_share_modes = get_share_modes(fsp->dev, fsp->inode, &share_list,
-                                         &dummy);
-
        DEBUG(10,("release_level_2_oplocks_on_change: num_share_modes = %d\n", 
-                       num_share_modes ));
+                 lck->num_share_modes ));
+
+       if (fsp->oplock_type == FAKE_LEVEL_II_OPLOCK) {
+               /* See if someone else has already downgraded us, then we
+                  don't have to do anything */
+               for (i=0; i<lck->num_share_modes; i++) {
+                       struct share_mode_entry *e = &lck->share_modes[i];
+                       if ((e->op_type == NO_OPLOCK) &&
+                           (e->share_file_id == fsp->file_id) &&
+                           (e->dev == fsp->dev) &&
+                           (e->inode == fsp->inode) &&
+                           (procid_is_me(&e->pid))) {
+                               /* We're done */
+                               fsp->oplock_type = NO_OPLOCK;
+                               talloc_free(lck);
+                               return;
+                       }
+               }
+       }
 
-       for(i = 0; i < num_share_modes; i++) {
-               share_mode_entry *share_entry = &share_list[i];
+       for(i = 0; i < lck->num_share_modes; i++) {
+               struct share_mode_entry *share_entry = &lck->share_modes[i];
 
                /*
-                * 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
+                * 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) ||
+                   (share_entry->op_type == FAKE_LEVEL_II_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.
-                */
-
-               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"));
-
-                       oplock_break_level2(new_fsp, True);
-
-               } else {
-
-                       /*
-                        * This is a remote file and so we send an asynchronous
-                        * message.
-                        */
-
-                       DEBUG(10,("release_level_2_oplocks_on_change: breaking remote oplock (async).\n"));
-                       request_remote_level2_async_oplock_break(share_entry);
-               }
-       }
-
-       SAFE_FREE(share_list);
-       unlock_share_entry_fsp(fsp);
-
-       /* 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");
-       }
-}
-
-/****************************************************************************
- Send a 'retry your open' message to a process with a deferred open entry.
-****************************************************************************/
-
-BOOL send_deferred_open_retry_message(deferred_open_entry *entry)
-{
-       char de_msg[DEFERRED_OPEN_MSG_LEN];
-       struct sockaddr_in addr_out;
-       pid_t pid = sys_getpid();
-
-       memset(de_msg, '\0', DEFERRED_OPEN_MSG_LEN);
-       SSVAL(de_msg,DEFERRED_OPEN_CMD_OFFSET,RETRY_DEFERRED_OPEN_CMD);
-       memcpy(de_msg+DEFERRED_OPEN_PID_OFFSET,(char *)&pid,sizeof(pid));
-       memcpy(de_msg+DEFERRED_OPEN_DEV_OFFSET,(char *)&entry->dev,sizeof(entry->dev));
-       memcpy(de_msg+DEFERRED_OPEN_INODE_OFFSET,(char *)&entry->inode,sizeof(entry->inode));
-       memcpy(de_msg+DEFERRED_OPEN_MID_OFFSET,(char *)&entry->mid,sizeof(entry->mid));
-
-       /* 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( entry->port );
-       addr_out.sin_family = AF_INET;
-   
-       if( DEBUGLVL( 3 ) ) {
-               dbgtext( "send_deferred_open_retry_message: sending a message to ");
-               dbgtext( "pid %d on port %d ", (int)entry->pid, entry->port );
-               dbgtext( "for dev = %x, inode = %.0f, mid = %u\n",
-                       (unsigned int)entry->dev, (double)entry->inode, (unsigned int)entry->mid );
+               message_send_pid(share_entry->pid, MSG_SMB_ASYNC_LEVEL2_BREAK,
+                                share_entry, sizeof(*share_entry), True);
        }
 
-       if(sys_sendto(oplock_sock,de_msg,DEFERRED_OPEN_MSG_LEN,0,
-                       (struct sockaddr *)&addr_out,sizeof(addr_out)) < 0) {
-               if( DEBUGLVL( 0 ) ) {
-                       dbgtext( "send_deferred_open_retry_message: failed sending a message to ");
-                       dbgtext( "pid %d on port %d ", (int)entry->pid, entry->port );
-                       dbgtext( "for dev = %x, inode = %.0f, mid = %u\n",
-                               (unsigned int)entry->dev, (double)entry->inode, (unsigned int)entry->mid );
-                       dbgtext( "Error was %s\n", strerror(errno) );
-               }
-               return False;
-       }
-       return True;
+       remove_all_share_oplocks(lck, fsp);
+       talloc_free(lck);
 }
 
 /****************************************************************************
@@ -1374,30 +700,18 @@ BOOL send_deferred_open_retry_message(deferred_open_entry *entry)
 
 BOOL init_oplocks(void)
 {
-       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);
+       message_register(MSG_SMB_BREAK_REQUEST,
+                        process_oplock_break_message);
+       message_register(MSG_SMB_ASYNC_LEVEL2_BREAK,
+                        process_oplock_break_message);
+       message_register(MSG_SMB_BREAK_RESPONSE,
+                        process_oplock_break_response);
+       message_register(MSG_SMB_KERNEL_BREAK,
+                        process_kernel_oplock_break);
+       message_register(MSG_SMB_OPEN_RETRY,
+                        process_open_retry_message);
 
        if (lp_kernel_oplocks()) {
 #if HAVE_KERNEL_OPLOCKS_IRIX
@@ -1407,8 +721,5 @@ 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;
 }