Initial import
[samba] / source / smbd / oplock.c
diff --git a/source/smbd/oplock.c b/source/smbd/oplock.c
new file mode 100644 (file)
index 0000000..5cc027f
--- /dev/null
@@ -0,0 +1,893 @@
+/* 
+   Unix SMB/CIFS implementation.
+   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
+   the Free Software Foundation; either version 2 of the License, or
+   (at your option) any later version.
+   
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+   
+   You should have received a copy of the GNU General Public License
+   along with this program; if not, write to the Free Software
+   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+*/
+
+#include "includes.h"
+
+/* 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;
+
+extern struct timeval smb_last_time;
+extern uint32 global_client_caps;
+extern int smb_read_error;
+
+static struct kernel_oplocks *koplocks;
+
+/****************************************************************************
+ Get the number of current exclusive oplocks.
+****************************************************************************/
+
+int32 get_number_of_exclusive_open_oplocks(void)
+{
+  return exclusive_oplocks_open;
+}
+
+/****************************************************************************
+ Return True if an oplock message is pending.
+****************************************************************************/
+
+BOOL oplock_message_waiting(fd_set *fds)
+{
+       if (koplocks && koplocks->msg_waiting(fds)) {
+               return True;
+       }
+
+       return False;
+}
+
+/****************************************************************************
+ 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.
+****************************************************************************/
+
+void process_kernel_oplocks(fd_set *pfds)
+{
+       /*
+        * 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) {
+               return;
+       }
+
+       while (koplocks->msg_waiting(pfds)) { 
+               files_struct *fsp;
+               char msg[MSG_SMB_KERNEL_BREAK_SIZE];
+
+               fsp = koplocks->receive_message(pfds);
+
+               if (fsp == NULL) {
+                       DEBUG(3, ("Kernel oplock message announced, but none "
+                                 "received\n"));
+                       return;
+               }
+
+               /* Put the kernel break info into the message. */
+               SDEV_T_VAL(msg,0,fsp->dev);
+               SINO_T_VAL(msg,8,fsp->inode);
+               SIVAL(msg,16,fsp->file_id);
+
+               /* Don't need to be root here as we're only ever
+                  sending to ourselves. */
+
+               message_send_pid(pid_to_procid(sys_getpid()),
+                                MSG_SMB_KERNEL_BREAK,
+                                &msg, MSG_SMB_KERNEL_BREAK_SIZE, True);
+       }
+}
+
+/****************************************************************************
+ Attempt to set an oplock on a file. Always succeeds if kernel oplocks are
+ disabled (just sets flags). Returns True if oplock set.
+****************************************************************************/
+
+BOOL set_file_oplock(files_struct *fsp, int 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) {
+               level_II_oplocks_open++;
+       } 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 ));
+
+       return True;
+}
+
+/****************************************************************************
+ Attempt to release an oplock on a file. Decrements oplock count.
+****************************************************************************/
+
+void release_file_oplock(files_struct *fsp)
+{
+       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 (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;
+       
+       flush_write_cache(fsp, OPLOCK_RELEASE_FLUSH);
+}
+
+/****************************************************************************
+ Attempt to downgrade an oplock on a file. Doesn't decrement oplock count.
+****************************************************************************/
+
+static void downgrade_file_oplock(files_struct *fsp)
+{
+       if (koplocks) {
+               koplocks->release_oplock(fsp);
+       }
+       fsp->oplock_type = LEVEL_II_OPLOCK;
+       exclusive_oplocks_open--;
+       level_II_oplocks_open++;
+       fsp->sent_oplock_break = NO_BREAK_SENT;
+}
+
+/****************************************************************************
+ Remove a file oplock. Copes with level II and exclusive.
+ Locks then unlocks the share mode lock. Client can decide to go directly
+ to none even if a "break-to-level II" was sent.
+****************************************************************************/
+
+BOOL remove_oplock(files_struct *fsp)
+{
+       SMB_DEV_T dev = fsp->dev;
+       SMB_INO_T inode = fsp->inode;
+       BOOL ret;
+       struct share_mode_lock *lck;
+
+       /* Remove the oplock flag from the sharemode. */
+       lck = get_share_mode_lock(NULL, fsp->dev, fsp->inode, NULL, 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;
+}
+
+/*
+ * 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;
+
+       lck = get_share_mode_lock(NULL, fsp->dev, fsp->inode, 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, dev = %x, inode = %.0f\n",
+                        fsp->fsp_name, fsp->fnum, (unsigned int)dev,
+                        (double)inode));
+       }
+
+       downgrade_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)
+{
+       int maxfd = 0;
+
+       if (koplocks && koplocks->notification_fd != -1) {
+               FD_SET(koplocks->notification_fd, fds);
+               maxfd = MAX(maxfd, koplocks->notification_fd);
+       }
+
+       return maxfd;
+}
+
+/****************************************************************************
+ Set up an oplock break message.
+****************************************************************************/
+
+static char *new_break_smb_message(TALLOC_CTX *mem_ctx,
+                                  files_struct *fsp, uint8 cmd)
+{
+       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(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(void)
+{
+       struct timeval cur_tv;
+       long wait_left = (long)lp_oplock_break_wait_time();
+
+       if (wait_left == 0) {
+               return;
+       }
+
+       GetTimeOfDay(&cur_tv);
+
+       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);
+       }
+}
+
+/****************************************************************************
+ 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)
+{
+       files_struct *fsp = NULL;
+
+       if( DEBUGLVL( 3 ) ) {
+               dbgtext( "initial_break_processing: called for dev = 0x%x, inode = %.0f file_id = %lu\n",
+                       (unsigned int)dev, (double)inode, file_id);
+               dbgtext( "Current oplocks_open (exclusive = %d, levelII = %d)\n",
+                       exclusive_oplocks_open, level_II_oplocks_open );
+       }
+
+       /*
+        * We need to search the file open table for the
+        * entry containing this dev and inode, and ensure
+        * we have an oplock on it.
+        */
+
+       fsp = file_find_dif(dev, inode, 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 = 0x%x, inode = %.0f file_id = %lu", (unsigned int)dev,
+                               (double)inode, file_id);
+                       dbgtext( "allowing break to succeed.\n" );
+               }
+               return NULL;
+       }
+
+       /* Ensure we have an oplock on the file */
+
+       /*
+        * There is a potential race condition in that an oplock could
+        * have been broken due to another udp request, and yet there are
+        * still oplock break messages being sent in the udp message
+        * queue for this file. So return true if we don't have an oplock,
+        * as we may have just freed it.
+        */
+
+       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( "Allowing break to succeed regardless.\n" );
+               }
+               return NULL;
+       }
+
+       return fsp;
+}
+
+static void oplock_timeout_handler(struct timed_event *te,
+                                  const struct timeval *now,
+                                  void *private_data)
+{
+       files_struct *fsp = private_data;
+
+       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.
+*******************************************************************/
+
+static void add_oplock_timeout_handler(files_struct *fsp)
+{
+       if (fsp->oplock_timeout != NULL) {
+               DEBUG(0, ("Logic problem -- have an oplock event hanging "
+                         "around\n"));
+       }
+
+       fsp->oplock_timeout =
+               add_timed_event(NULL,
+                               timeval_current_ofs(OPLOCK_BREAK_TIMEOUT, 0),
+                               "oplock_timeout_handler",
+                               oplock_timeout_handler, fsp);
+
+       if (fsp->oplock_timeout == NULL) {
+               DEBUG(0, ("Could not add oplock timeout handler\n"));
+       }
+}
+
+/*******************************************************************
+ 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(int msg_type, struct process_id src,
+                                        void *buf, size_t len)
+{
+       struct share_mode_entry msg;
+       files_struct *fsp;
+       char *break_msg;
+       BOOL sign_state;
+
+       if (buf == NULL) {
+               DEBUG(0, ("Got NULL buffer\n"));
+               return;
+       }
+
+       if (len != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
+               DEBUG(0, ("Got invalid msg len %d\n", (int)len));
+               return;
+       }
+
+       /* De-linearize incoming message. */
+       message_to_share_mode_entry(&msg, buf);
+
+       DEBUG(10, ("Got oplock async level 2 break message from pid %d: 0x%x/%.0f/%d\n",
+                  (int)procid_to_pid(&src), (unsigned int)msg.dev, (double)msg.inode,
+                  (int)msg.share_file_id));
+
+       fsp = initial_break_processing(msg.dev, msg.inode,
+                                      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 (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 (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;
+       }
+
+       /* 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. */
+
+       break_msg = new_break_smb_message(NULL, fsp, OPLOCKLEVEL_NONE);
+       if (break_msg == NULL) {
+               exit_server("Could not talloc break_msg\n");
+       }
+
+       /* 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();
+       }
+
+       /* Save the server smb signing state. */
+       sign_state = srv_oplock_set_signing(False);
+
+       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);
+
+       talloc_free(break_msg);
+
+       /* Async level2 request, don't send a reply, just remove the oplock. */
+       remove_oplock(fsp);
+}
+
+/*******************************************************************
+ This handles the generic oplock break message from another smbd.
+*******************************************************************/
+
+static void process_oplock_break_message(int msg_type, struct process_id src,
+                                        void *buf, size_t len)
+{
+       struct share_mode_entry msg;
+       files_struct *fsp;
+       char *break_msg;
+       BOOL break_to_level2 = False;
+       BOOL sign_state;
+
+       if (buf == NULL) {
+               DEBUG(0, ("Got NULL buffer\n"));
+               return;
+       }
+
+       if (len != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
+               DEBUG(0, ("Got invalid msg len %d\n", (int)len));
+               return;
+       }
+
+       /* De-linearize incoming message. */
+       message_to_share_mode_entry(&msg, buf);
+
+       DEBUG(10, ("Got oplock break message from pid %d: 0x%x/%.0f/%d\n",
+                  (int)procid_to_pid(&src), (unsigned int)msg.dev, (double)msg.inode,
+                  (int)msg.share_file_id));
+
+       fsp = initial_break_processing(msg.dev, msg.inode,
+                                      msg.share_file_id);
+
+       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"));
+               become_root();
+
+               /* We just send the same message back. */
+               message_send_pid(src, MSG_SMB_BREAK_RESPONSE,
+                                buf, MSG_SMB_SHARE_MODE_ENTRY_SIZE, True);
+
+               unbecome_root();
+               return;
+       }
+
+       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 0x%x/%.0f: %s\n",
+                         (unsigned int)fsp->dev, (double)fsp->inode,
+                         fsp->fsp_name));
+               become_root();
+
+               /* We just send the same message back. */
+               message_send_pid(src, MSG_SMB_BREAK_RESPONSE,
+                                buf, MSG_SMB_SHARE_MODE_ENTRY_SIZE, True);
+
+               unbecome_root();
+               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))) {
+               break_to_level2 = True;
+       }
+
+       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");
+       }
+
+       /* 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();
+       }
+
+       /* Save the server smb signing state. */
+       sign_state = srv_oplock_set_signing(False);
+
+       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);
+
+       talloc_free(break_msg);
+
+       fsp->sent_oplock_break = break_to_level2 ? LEVEL_II_BREAK_SENT:BREAK_TO_NONE_SENT;
+
+       msg.pid = src;
+       ADD_TO_ARRAY(NULL, struct share_mode_entry, msg,
+                    &fsp->pending_break_messages,
+                    &fsp->num_pending_break_messages);
+
+       add_oplock_timeout_handler(fsp);
+}
+
+/*******************************************************************
+ This handles the kernel oplock break message.
+*******************************************************************/
+
+static void process_kernel_oplock_break(int msg_type, struct process_id src,
+                                       void *buf, size_t len)
+{
+       SMB_DEV_T dev;
+       SMB_INO_T inode;
+       unsigned long file_id;
+       files_struct *fsp;
+       char *break_msg;
+       BOOL sign_state;
+
+       if (buf == NULL) {
+               DEBUG(0, ("Got NULL buffer\n"));
+               return;
+       }
+
+       if (len != MSG_SMB_KERNEL_BREAK_SIZE) {
+               DEBUG(0, ("Got invalid msg len %d\n", (int)len));
+               return;
+       }
+
+       /* Pull the data from the message. */
+       dev = DEV_T_VAL(buf, 0);
+       inode = INO_T_VAL(buf, 8);
+       file_id = (unsigned long)IVAL(buf, 16);
+
+       DEBUG(10, ("Got kernel oplock break message from pid %d: 0x%x/%.0f/%u\n",
+                  (int)procid_to_pid(&src), (unsigned int)dev, (double)inode,
+                  (unsigned int)file_id));
+
+       fsp = initial_break_processing(dev, inode, file_id);
+
+       if (fsp == NULL) {
+               DEBUG(3, ("Got a kernel oplock break message for a file "
+                         "I don't know about\n"));
+               return;
+       }
+
+       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;
+       }
+
+       break_msg = new_break_smb_message(NULL, fsp, OPLOCKLEVEL_NONE);
+       if (break_msg == NULL) {
+               exit_server("Could not talloc break_msg\n");
+       }
+
+       /* Save the server smb signing state. */
+       sign_state = srv_oplock_set_signing(False);
+
+       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);
+
+       talloc_free(break_msg);
+
+       fsp->sent_oplock_break = BREAK_TO_NONE_SENT;
+
+       add_oplock_timeout_handler(fsp);
+}
+
+void reply_to_oplock_break_requests(files_struct *fsp)
+{
+       int i;
+
+       become_root();
+       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];
+
+               share_mode_entry_to_message(msg, e);
+
+               message_send_pid(e->pid, MSG_SMB_BREAK_RESPONSE,
+                                msg, MSG_SMB_SHARE_MODE_ENTRY_SIZE, True);
+       }
+       unbecome_root();
+
+       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;
+}
+
+static void process_oplock_break_response(int msg_type, struct process_id src,
+                                         void *buf, size_t len)
+{
+       struct share_mode_entry msg;
+
+       if (buf == NULL) {
+               DEBUG(0, ("Got NULL buffer\n"));
+               return;
+       }
+
+       if (len != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
+               DEBUG(0, ("Got invalid msg len %u\n", (unsigned int)len));
+               return;
+       }
+
+       /* De-linearize incoming message. */
+       message_to_share_mode_entry(&msg, buf);
+
+       DEBUG(10, ("Got oplock break response from pid %d: 0x%x/%.0f/%u mid %u\n",
+                  (int)procid_to_pid(&src), (unsigned int)msg.dev, (double)msg.inode,
+                  (unsigned int)msg.share_file_id, (unsigned int)msg.op_mid));
+
+       /* Here's the hack from open.c, store the mid in the 'port' field */
+       schedule_deferred_open_smb_message(msg.op_mid);
+}
+
+static void process_open_retry_message(int msg_type, struct process_id src,
+                                      void *buf, size_t len)
+{
+       struct share_mode_entry msg;
+       
+       if (buf == NULL) {
+               DEBUG(0, ("Got NULL buffer\n"));
+               return;
+       }
+
+       if (len != MSG_SMB_SHARE_MODE_ENTRY_SIZE) {
+               DEBUG(0, ("Got invalid msg len %d\n", (int)len));
+               return;
+       }
+
+       /* De-linearize incoming message. */
+       message_to_share_mode_entry(&msg, buf);
+
+       DEBUG(10, ("Got open retry msg from pid %d: 0x%x/%.0f mid %u\n",
+                  (int)procid_to_pid(&src), (unsigned int)msg.dev, (double)msg.inode,
+                  (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.
+****************************************************************************/
+
+void release_level_2_oplocks_on_change(files_struct *fsp)
+{
+       int i;
+       struct share_mode_lock *lck;
+
+       /*
+        * If this file is level II oplocked then we need
+        * to grab the shared memory lock and inform all
+        * other files with a level II lock that they need
+        * to flush their read caches. We keep the lock over
+        * the shared memory area whilst doing this.
+        */
+
+       if (!LEVEL_II_OPLOCK_TYPE(fsp->oplock_type))
+               return;
+
+       lck = get_share_mode_lock(NULL, fsp->dev, fsp->inode, 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 ));
+       }
+
+       DEBUG(10,("release_level_2_oplocks_on_change: num_share_modes = %d\n", 
+                 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];
+
+               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 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 ));
+
+               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 ));
+                       talloc_free(lck);
+                       abort();
+               }
+
+               share_mode_entry_to_message(msg, share_entry);
+
+               become_root();
+               message_send_pid(share_entry->pid, MSG_SMB_ASYNC_LEVEL2_BREAK,
+                                msg, MSG_SMB_SHARE_MODE_ENTRY_SIZE, True);
+               unbecome_root();
+       }
+
+       /* We let the message receivers handle removing the oplock state
+          in the share mode lock db. */
+
+       talloc_free(lck);
+}
+
+/****************************************************************************
+ Linearize a share mode entry struct to an internal oplock break message.
+****************************************************************************/
+
+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);
+       SDEV_T_VAL(msg,28,e->dev);
+       SINO_T_VAL(msg,36,e->inode);
+       SIVAL(msg,44,e->share_file_id);
+}
+
+/****************************************************************************
+ De-linearize an internal oplock break message to a share mode entry struct.
+****************************************************************************/
+
+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);
+       e->dev = DEV_T_VAL(msg,28);
+       e->inode = INO_T_VAL(msg,36);
+       e->share_file_id = (unsigned long)IVAL(msg,44);
+}
+
+/****************************************************************************
+ Setup oplocks for this process.
+****************************************************************************/
+
+BOOL init_oplocks(void)
+{
+       DEBUG(3,("open_oplock_ipc: initializing messages.\n"));
+
+       message_register(MSG_SMB_BREAK_REQUEST,
+                        process_oplock_break_message);
+       message_register(MSG_SMB_ASYNC_LEVEL2_BREAK,
+                        process_oplock_async_level2_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
+               koplocks = irix_init_kernel_oplocks();
+#elif HAVE_KERNEL_OPLOCKS_LINUX
+               koplocks = linux_init_kernel_oplocks();
+#endif
+       }
+
+       return True;
+}