r23151: Re-arrange the logic so it looks identical to the
[sfrench/samba-autobuild/.git] / source / smbd / open.c
index 810913c025d83294ee21d9e59158a1285bfb76e6..987da7a94daa0bd79df39bafff3272f8587250f1 100644 (file)
 
 #include "includes.h"
 
+extern struct generic_mapping file_generic_mapping;
 extern struct current_user current_user;
 extern userdom_struct current_user_info;
-extern uint16 global_oplock_port;
 extern uint16 global_smbpid;
 extern BOOL global_client_failed_oplock_break;
 
-struct dev_inode_bundle {
+struct deferred_open_record {
+       BOOL delayed_for_oplocks;
        SMB_DEV_T dev;
        SMB_INO_T inode;
 };
@@ -37,178 +38,185 @@ struct dev_inode_bundle {
  fd support routines - attempt to do a dos_open.
 ****************************************************************************/
 
-static int fd_open(struct connection_struct *conn,
-                       const char *fname, 
-                       int flags,
-                       mode_t mode)
+static NTSTATUS fd_open(struct connection_struct *conn,
+                   const char *fname, 
+                   files_struct *fsp,
+                   int flags,
+                   mode_t mode)
 {
-       int fd;
+       NTSTATUS status = NT_STATUS_OK;
+
 #ifdef O_NOFOLLOW
-       if (!lp_symlinks(SNUM(conn))) {
+       /* 
+        * Never follow symlinks on a POSIX client. The
+        * client should be doing this.
+        */
+
+       if (fsp->posix_open || !lp_symlinks(SNUM(conn))) {
                flags |= O_NOFOLLOW;
        }
 #endif
 
-       fd = SMB_VFS_OPEN(conn,fname,flags,mode);
+       fsp->fh->fd = SMB_VFS_OPEN(conn,fname,fsp,flags,mode);
+       if (fsp->fh->fd == -1) {
+               status = map_nt_error_from_unix(errno);
+       }
 
-       DEBUG(10,("fd_open: name %s, flags = 0%o mode = 0%o, fd = %d. %s\n", fname,
-               flags, (int)mode, fd, (fd == -1) ? strerror(errno) : "" ));
+       DEBUG(10,("fd_open: name %s, flags = 0%o mode = 0%o, fd = %d. %s\n",
+                   fname, flags, (int)mode, fsp->fh->fd,
+               (fsp->fh->fd == -1) ? strerror(errno) : "" ));
 
-       return fd;
+       return status;
 }
 
 /****************************************************************************
  Close the file associated with a fsp.
 ****************************************************************************/
 
-int fd_close(struct connection_struct *conn,
-               files_struct *fsp)
+NTSTATUS fd_close(struct connection_struct *conn, files_struct *fsp)
 {
        if (fsp->fh->fd == -1) {
-               return 0; /* What we used to call a stat open. */
+               return NT_STATUS_OK; /* What we used to call a stat open. */
        }
        if (fsp->fh->ref_count > 1) {
-               return 0; /* Shared handle. Only close last reference. */
+               return NT_STATUS_OK; /* Shared handle. Only close last reference. */
        }
        return fd_close_posix(conn, fsp);
 }
 
-
-/****************************************************************************
- Check a filename for the pipe string.
-****************************************************************************/
-
-static void check_for_pipe(const char *fname)
-{
-       /* special case of pipe opens */
-       char s[10];
-       StrnCpy(s,fname,sizeof(s)-1);
-       strlower_m(s);
-       if (strstr(s,"pipe/")) {
-               DEBUG(3,("Rejecting named pipe open for %s\n",fname));
-               set_saved_error_triple(ERRSRV, ERRaccess, NT_STATUS_ACCESS_DENIED);
-       }
-}
-
 /****************************************************************************
  Change the ownership of a file to that of the parent directory.
  Do this by fd if possible.
 ****************************************************************************/
 
-void change_owner_to_parent(connection_struct *conn,
-                               files_struct *fsp,
-                               const char *fname,
-                               SMB_STRUCT_STAT *psbuf)
+static void change_file_owner_to_parent(connection_struct *conn,
+                                       const char *inherit_from_dir,
+                                       files_struct *fsp)
 {
-       const char *parent_path = parent_dirname(fname);
        SMB_STRUCT_STAT parent_st;
        int ret;
 
-       ret = SMB_VFS_STAT(conn, parent_path, &parent_st);
+       ret = SMB_VFS_STAT(conn, inherit_from_dir, &parent_st);
        if (ret == -1) {
-               DEBUG(0,("change_owner_to_parent: failed to stat parent "
+               DEBUG(0,("change_file_owner_to_parent: failed to stat parent "
                         "directory %s. Error was %s\n",
-                        parent_path, strerror(errno) ));
+                        inherit_from_dir, strerror(errno) ));
                return;
        }
 
-       if (fsp && fsp->fh->fd != -1) {
-               become_root();
-               ret = SMB_VFS_FCHOWN(fsp, fsp->fh->fd, parent_st.st_uid, (gid_t)-1);
-               unbecome_root();
-               if (ret == -1) {
-                       DEBUG(0,("change_owner_to_parent: failed to fchown "
-                                "file %s to parent directory uid %u. Error "
-                                "was %s\n", fname,
-                                (unsigned int)parent_st.st_uid,
-                                strerror(errno) ));
-               }
+       become_root();
+       ret = SMB_VFS_FCHOWN(fsp, fsp->fh->fd, parent_st.st_uid, (gid_t)-1);
+       unbecome_root();
+       if (ret == -1) {
+               DEBUG(0,("change_file_owner_to_parent: failed to fchown "
+                        "file %s to parent directory uid %u. Error "
+                        "was %s\n", fsp->fsp_name,
+                        (unsigned int)parent_st.st_uid,
+                        strerror(errno) ));
+       }
 
-               DEBUG(10,("change_owner_to_parent: changed new file %s to "
-                         "parent directory uid %u.\n", fname,
-                         (unsigned int)parent_st.st_uid ));
+       DEBUG(10,("change_file_owner_to_parent: changed new file %s to "
+                 "parent directory uid %u.\n", fsp->fsp_name,
+                 (unsigned int)parent_st.st_uid ));
+}
 
-       } else {
-               /* We've already done an lstat into psbuf, and we know it's a
-                  directory. If we can cd into the directory and the dev/ino
-                  are the same then we can safely chown without races as
-                  we're locking the directory in place by being in it.  This
-                  should work on any UNIX (thanks tridge :-). JRA.
-               */
-
-               pstring saved_dir;
-               SMB_STRUCT_STAT sbuf;
-
-               if (!vfs_GetWd(conn,saved_dir)) {
-                       DEBUG(0,("change_owner_to_parent: failed to get "
-                                "current working directory\n"));
-                       return;
-               }
+static void change_dir_owner_to_parent(connection_struct *conn,
+                                      const char *inherit_from_dir,
+                                      const char *fname,
+                                      SMB_STRUCT_STAT *psbuf)
+{
+       pstring saved_dir;
+       SMB_STRUCT_STAT sbuf;
+       SMB_STRUCT_STAT parent_st;
+       int ret;
 
-               /* Chdir into the new path. */
-               if (vfs_ChDir(conn, fname) == -1) {
-                       DEBUG(0,("change_owner_to_parent: failed to change "
-                                "current working directory to %s. Error "
-                                "was %s\n", fname, strerror(errno) ));
-                       goto out;
-               }
+       ret = SMB_VFS_STAT(conn, inherit_from_dir, &parent_st);
+       if (ret == -1) {
+               DEBUG(0,("change_dir_owner_to_parent: failed to stat parent "
+                        "directory %s. Error was %s\n",
+                        inherit_from_dir, strerror(errno) ));
+               return;
+       }
 
-               if (SMB_VFS_STAT(conn,".",&sbuf) == -1) {
-                       DEBUG(0,("change_owner_to_parent: failed to stat "
-                                "directory '.' (%s) Error was %s\n",
-                                fname, strerror(errno)));
-                       goto out;
-               }
+       /* We've already done an lstat into psbuf, and we know it's a
+          directory. If we can cd into the directory and the dev/ino
+          are the same then we can safely chown without races as
+          we're locking the directory in place by being in it.  This
+          should work on any UNIX (thanks tridge :-). JRA.
+       */
 
-               /* Ensure we're pointing at the same place. */
-               if (sbuf.st_dev != psbuf->st_dev ||
-                   sbuf.st_ino != psbuf->st_ino ||
-                   sbuf.st_mode != psbuf->st_mode ) {
-                       DEBUG(0,("change_owner_to_parent: "
-                                "device/inode/mode on directory %s changed. "
-                                "Refusing to chown !\n", fname ));
-                       goto out;
-               }
+       if (!vfs_GetWd(conn,saved_dir)) {
+               DEBUG(0,("change_dir_owner_to_parent: failed to get "
+                        "current working directory\n"));
+               return;
+       }
 
-               become_root();
-               ret = SMB_VFS_CHOWN(conn, ".", parent_st.st_uid, (gid_t)-1);
-               unbecome_root();
-               if (ret == -1) {
-                       DEBUG(10,("change_owner_to_parent: failed to chown "
-                                 "directory %s to parent directory uid %u. "
-                                 "Error was %s\n", fname,
-                                 (unsigned int)parent_st.st_uid, strerror(errno) ));
-                       goto out;
-               }
+       /* Chdir into the new path. */
+       if (vfs_ChDir(conn, fname) == -1) {
+               DEBUG(0,("change_dir_owner_to_parent: failed to change "
+                        "current working directory to %s. Error "
+                        "was %s\n", fname, strerror(errno) ));
+               goto out;
+       }
 
-               DEBUG(10,("change_owner_to_parent: changed ownership of new "
-                         "directory %s to parent directory uid %u.\n",
-                         fname, (unsigned int)parent_st.st_uid ));
+       if (SMB_VFS_STAT(conn,".",&sbuf) == -1) {
+               DEBUG(0,("change_dir_owner_to_parent: failed to stat "
+                        "directory '.' (%s) Error was %s\n",
+                        fname, strerror(errno)));
+               goto out;
+       }
 
-  out:
+       /* Ensure we're pointing at the same place. */
+       if (sbuf.st_dev != psbuf->st_dev ||
+           sbuf.st_ino != psbuf->st_ino ||
+           sbuf.st_mode != psbuf->st_mode ) {
+               DEBUG(0,("change_dir_owner_to_parent: "
+                        "device/inode/mode on directory %s changed. "
+                        "Refusing to chown !\n", fname ));
+               goto out;
+       }
 
-               vfs_ChDir(conn,saved_dir);
+       become_root();
+       ret = SMB_VFS_CHOWN(conn, ".", parent_st.st_uid, (gid_t)-1);
+       unbecome_root();
+       if (ret == -1) {
+               DEBUG(10,("change_dir_owner_to_parent: failed to chown "
+                         "directory %s to parent directory uid %u. "
+                         "Error was %s\n", fname,
+                         (unsigned int)parent_st.st_uid, strerror(errno) ));
+               goto out;
        }
+
+       DEBUG(10,("change_dir_owner_to_parent: changed ownership of new "
+                 "directory %s to parent directory uid %u.\n",
+                 fname, (unsigned int)parent_st.st_uid ));
+
+ out:
+
+       vfs_ChDir(conn,saved_dir);
 }
 
 /****************************************************************************
  Open a file.
 ****************************************************************************/
 
-static BOOL open_file(files_struct *fsp,
-                       connection_struct *conn,
-                       const char *fname,
-                       SMB_STRUCT_STAT *psbuf,
-                       int flags,
-                       mode_t unx_mode,
-                       uint32 access_mask)
+static NTSTATUS open_file(files_struct *fsp,
+                         connection_struct *conn,
+                         const char *parent_dir,
+                         const char *name,
+                         const char *path,
+                         SMB_STRUCT_STAT *psbuf,
+                         int flags,
+                         mode_t unx_mode,
+                         uint32 access_mask, /* client requested access mask. */
+                         uint32 open_access_mask) /* what we're actually using in the open. */
 {
+       NTSTATUS status = NT_STATUS_OK;
        int accmode = (flags & O_ACCMODE);
        int local_flags = flags;
        BOOL file_existed = VALID_STAT(*psbuf);
 
        fsp->fh->fd = -1;
-       fsp->oplock_type = NO_OPLOCK;
        errno = EPERM;
 
        /* Check permissions */
@@ -226,9 +234,8 @@ static BOOL open_file(files_struct *fsp,
        if (!CAN_WRITE(conn)) {
                /* It's a read-only share - fail if we wanted to write. */
                if(accmode != O_RDONLY) {
-                       DEBUG(3,("Permission denied opening %s\n",fname));
-                       check_for_pipe(fname);
-                       return False;
+                       DEBUG(3,("Permission denied opening %s\n", path));
+                       return NT_STATUS_ACCESS_DENIED;
                } else if(flags & O_CREAT) {
                        /* We don't want to write - but we must make sure that
                           O_CREAT doesn't create the file if we have write
@@ -253,17 +260,17 @@ static BOOL open_file(files_struct *fsp,
 
        if ((accmode == O_RDONLY) && ((flags & O_TRUNC) == O_TRUNC)) {
                DEBUG(10,("open_file: truncate requested on read-only open "
-                         "for file %s\n",fname ));
+                         "for file %s\n", path));
                local_flags = (flags & ~O_ACCMODE)|O_RDWR;
        }
 
-       if ((access_mask & (FILE_READ_DATA|FILE_WRITE_DATA|FILE_APPEND_DATA|FILE_EXECUTE)) ||
-           (local_flags & O_CREAT) ||
+       if ((open_access_mask & (FILE_READ_DATA|FILE_WRITE_DATA|FILE_APPEND_DATA|FILE_EXECUTE)) ||
+           (!file_existed && (local_flags & O_CREAT)) ||
            ((local_flags & O_TRUNC) == O_TRUNC) ) {
 
                /*
                 * We can't actually truncate here as the file may be locked.
-                * open_file_shared will take care of the truncate later. JRA.
+                * open_file_ntcreate will take care of the truncate later. JRA.
                 */
 
                local_flags &= ~O_TRUNC;
@@ -282,25 +289,35 @@ static BOOL open_file(files_struct *fsp,
 
                /* Don't create files with Microsoft wildcard characters. */
                if ((local_flags & O_CREAT) && !file_existed &&
-                   ms_has_wild(fname))  {
-                       set_saved_error_triple(ERRDOS, ERRinvalidname,
-                                              NT_STATUS_OBJECT_NAME_INVALID);
-                       return False;
+                   ms_has_wild(path))  {
+                       return NT_STATUS_OBJECT_NAME_INVALID;
                }
 
                /* Actually do the open */
-               fsp->fh->fd = fd_open(conn, fname, local_flags, unx_mode);
-               if (fsp->fh->fd == -1)  {
+               status = fd_open(conn, path, fsp, local_flags, unx_mode);
+               if (!NT_STATUS_IS_OK(status)) {
                        DEBUG(3,("Error opening file %s (%s) (local_flags=%d) "
                                 "(flags=%d)\n",
-                                fname,strerror(errno),local_flags,flags));
-                       check_for_pipe(fname);
-                       return False;
+                                path,nt_errstr(status),local_flags,flags));
+                       return status;
                }
 
-               /* Inherit the ACL if the file was created. */
                if ((local_flags & O_CREAT) && !file_existed) {
-                       inherit_access_acl(conn, fname, unx_mode);
+
+                       /* Inherit the ACL if required */
+                       if (lp_inherit_perms(SNUM(conn))) {
+                               inherit_access_acl(conn, parent_dir, path,
+                                                  unx_mode);
+                       }
+
+                       /* Change the owner if required. */
+                       if (lp_inherit_owner(SNUM(conn))) {
+                               change_file_owner_to_parent(conn, parent_dir,
+                                                           fsp);
+                       }
+
+                       notify_fname(conn, NOTIFY_ACTION_ADDED,
+                                    FILE_NOTIFY_CHANGE_FILE_NAME, path);
                }
 
        } else {
@@ -311,20 +328,21 @@ static BOOL open_file(files_struct *fsp,
                int ret;
 
                if (fsp->fh->fd == -1) {
-                       ret = SMB_VFS_STAT(conn, fname, psbuf);
+                       ret = SMB_VFS_STAT(conn, path, psbuf);
                } else {
                        ret = SMB_VFS_FSTAT(fsp,fsp->fh->fd,psbuf);
                        /* If we have an fd, this stat should succeed. */
                        if (ret == -1) {
                                DEBUG(0,("Error doing fstat on open file %s "
-                                        "(%s)\n", fname,strerror(errno) ));
+                                        "(%s)\n", path,strerror(errno) ));
                        }
                }
 
                /* For a non-io open, this stat failing means file not found. JRA */
                if (ret == -1) {
+                       status = map_nt_error_from_unix(errno);
                        fd_close(conn, fsp);
-                       return False;
+                       return status;
                }
        }
 
@@ -337,7 +355,7 @@ static BOOL open_file(files_struct *fsp,
        if(S_ISDIR(psbuf->st_mode)) {
                fd_close(conn, fsp);
                errno = EISDIR;
-               return False;
+               return NT_STATUS_FILE_IS_A_DIRECTORY;
        }
 
        fsp->mode = psbuf->st_mode;
@@ -350,29 +368,26 @@ static BOOL open_file(files_struct *fsp,
        if (!CAN_WRITE(conn)) {
                fsp->can_write = False;
        } else {
-               fsp->can_write = (access_mask & (FILE_WRITE_DATA | FILE_APPEND_DATA)) ? True : False;
+               fsp->can_write = (access_mask & (FILE_WRITE_DATA | FILE_APPEND_DATA)) ?
+                       True : False;
        }
        fsp->print_file = False;
        fsp->modified = False;
-       fsp->oplock_type = NO_OPLOCK;
        fsp->sent_oplock_break = NO_BREAK_SENT;
        fsp->is_directory = False;
        fsp->is_stat = False;
-       if (conn->aio_write_behind_list &&
-           is_in_path(fname, conn->aio_write_behind_list, conn->case_sensitive)) {
-               fsp->aio_write_behind = True;
-       }
 
-       string_set(&fsp->fsp_name,fname);
+       string_set(&fsp->fsp_name, path);
        fsp->wcp = NULL; /* Write cache pointer. */
 
        DEBUG(2,("%s opened file %s read=%s write=%s (numopen=%d)\n",
-                *current_user_info.smb_name ? current_user_info.smb_name : conn->user,fsp->fsp_name,
+                *current_user_info.smb_name ?
+                current_user_info.smb_name : conn->user,fsp->fsp_name,
                 BOOLSTR(fsp->can_read), BOOLSTR(fsp->can_write),
                 conn->num_files_open + 1));
 
        errno = 0;
-       return True;
+       return NT_STATUS_OK;
 }
 
 /*******************************************************************
@@ -397,7 +412,7 @@ static BOOL is_executable(const char *fname)
  Returns True if conflict, False if not.
 ****************************************************************************/
 
-static BOOL share_conflict(share_mode_entry *entry,
+static BOOL share_conflict(struct share_mode_entry *entry,
                           uint32 access_mask,
                           uint32 share_access)
 {
@@ -445,7 +460,6 @@ static BOOL share_conflict(share_mode_entry *entry,
                DEBUG(10,("share_conflict: check %d conflict am = 0x%x, right = 0x%x, \
 sa = 0x%x, share = 0x%x\n", (num), (unsigned int)(am), (unsigned int)(right), (unsigned int)(sa), \
                        (unsigned int)(share) )); \
-               set_saved_error_triple(ERRDOS, ERRbadshare, NT_STATUS_SHARING_VIOLATION); \
                return True; \
        }
 #else
@@ -454,7 +468,6 @@ sa = 0x%x, share = 0x%x\n", (num), (unsigned int)(am), (unsigned int)(right), (u
                DEBUG(10,("share_conflict: check %d conflict am = 0x%x, right = 0x%x, \
 sa = 0x%x, share = 0x%x\n", (num), (unsigned int)(am), (unsigned int)(right), (unsigned int)(sa), \
                        (unsigned int)(share) )); \
-               set_saved_error_triple(ERRDOS, ERRbadshare, NT_STATUS_SHARING_VIOLATION); \
                return True; \
        }
 #endif
@@ -480,11 +493,23 @@ sa = 0x%x, share = 0x%x\n", (num), (unsigned int)(am), (unsigned int)(right), (u
 
 #if defined(DEVELOPER)
 static void validate_my_share_entries(int num,
-                                       share_mode_entry *share_entry)
+                                     struct share_mode_entry *share_entry)
 {
        files_struct *fsp;
 
-       if (share_entry->pid != sys_getpid()) {
+       if (!procid_is_me(&share_entry->pid)) {
+               return;
+       }
+
+       if (is_deferred_open_entry(share_entry) &&
+           !open_was_deferred(share_entry->op_mid)) {
+               pstring str;
+               pstr_sprintf(str, "Got a deferred entry without a request: "
+                            "PANIC: %s\n", share_mode_str(num, share_entry));
+               smb_panic(str);
+       }
+
+       if (!is_valid_share_mode_entry(share_entry)) {
                return;
        }
 
@@ -497,7 +522,26 @@ static void validate_my_share_entries(int num,
                          "share entry with an open file\n");
        }
 
+       if (is_deferred_open_entry(share_entry) ||
+           is_unused_share_mode_entry(share_entry)) {
+               goto panic;
+       }
+
+       if ((share_entry->op_type == NO_OPLOCK) &&
+           (fsp->oplock_type == FAKE_LEVEL_II_OPLOCK)) {
+               /* Someone has already written to it, but I haven't yet
+                * noticed */
+               return;
+       }
+
        if (((uint16)fsp->oplock_type) != share_entry->op_type) {
+               goto panic;
+       }
+
+       return;
+
+ panic:
+       {
                pstring str;
                DEBUG(0,("validate_my_share_entries: PANIC : %s\n",
                         share_mode_str(num, share_entry) ));
@@ -510,376 +554,256 @@ static void validate_my_share_entries(int num,
 }
 #endif
 
-struct share_mode_entry_list {
-       struct share_mode_entry_list *next, *prev;
-       share_mode_entry entry;
-};
-
-static void free_broken_entry_list(struct share_mode_entry_list *broken_entry_list)
+static BOOL is_stat_open(uint32 access_mask)
 {
-       while (broken_entry_list) {
-               struct share_mode_entry_list *broken_entry = broken_entry_list;
-               DLIST_REMOVE(broken_entry_list, broken_entry);
-               SAFE_FREE(broken_entry);
-       }
-}
-
-static BOOL cause_oplock_break(int request, int existing, uint32 access_mask)
-{
-       if ((access_mask == DELETE_ACCESS) &&
-           (request == NO_OPLOCK)) {
-               /* This is a delete request */
-               return (BATCH_OPLOCK_TYPE(existing) != 0);
-       }
-
-       if (EXCLUSIVE_OPLOCK_TYPE(existing) && (request != NO_OPLOCK)) {
-               return True;
-       }
-
-       if ((existing != NO_OPLOCK) && (request == NO_OPLOCK)) {
-               return True;
-       }
-
-       return False;
+       return (access_mask &&
+               ((access_mask & ~(SYNCHRONIZE_ACCESS| FILE_READ_ATTRIBUTES|
+                                 FILE_WRITE_ATTRIBUTES))==0) &&
+               ((access_mask & (SYNCHRONIZE_ACCESS|FILE_READ_ATTRIBUTES|
+                                FILE_WRITE_ATTRIBUTES)) != 0));
 }
 
 /****************************************************************************
- Deal with open deny mode and oplock break processing.
+ Deal with share modes
  Invarient: Share mode must be locked on entry and exit.
  Returns -1 on error, or number of share modes on success (may be zero).
 ****************************************************************************/
 
-static int open_mode_check(connection_struct *conn,
-                          const char *fname,
-                          SMB_DEV_T dev,
-                          SMB_INO_T inode, 
-                          uint32 access_mask,
-                          uint32 share_access,
-                          uint32 create_options,
-                          int *p_flags,
-                          int *p_oplock_request,
-                          BOOL *p_all_current_opens_are_level_II)
+static NTSTATUS open_mode_check(connection_struct *conn,
+                               const char *fname,
+                               struct share_mode_lock *lck,
+                               uint32 access_mask,
+                               uint32 share_access,
+                               uint32 create_options,
+                               BOOL *file_existed)
 {
        int i;
-       int num_share_modes;
-       int oplock_contention_count = 0;
-       share_mode_entry *old_shares = NULL;
-       BOOL broke_oplock;
-       BOOL delete_on_close;
 
-       num_share_modes = get_share_modes(dev, inode, &old_shares, &delete_on_close);
-       
-       if(num_share_modes == 0) {
-               SAFE_FREE(old_shares);
-               return 0;
+       if(lck->num_share_modes == 0) {
+               return NT_STATUS_OK;
        }
+
+       *file_existed = True;
        
-       if (access_mask &&
-           ((access_mask & ~(SYNCHRONIZE_ACCESS| FILE_READ_ATTRIBUTES|
-                             FILE_WRITE_ATTRIBUTES))==0) &&
-           ((access_mask & (SYNCHRONIZE_ACCESS|FILE_READ_ATTRIBUTES|
-                            FILE_WRITE_ATTRIBUTES)) != 0)) {
+       if (is_stat_open(access_mask)) {
                /* Stat open that doesn't trigger oplock breaks or share mode
                 * checks... ! JRA. */
-               SAFE_FREE(old_shares);
-               return num_share_modes;
+               return NT_STATUS_OK;
        }
 
        /* A delete on close prohibits everything */
 
-       if (delete_on_close) {
-               SAFE_FREE(old_shares);
-               errno = EACCES;
-               return -1;
+       if (lck->delete_on_close) {
+               return NT_STATUS_DELETE_PENDING;
        }
 
        /*
         * Check if the share modes will give us access.
         */
        
-       do {
-               struct share_mode_entry_list *broken_entry_list = NULL;
-               struct share_mode_entry_list *broken_entry = NULL;
-
-               broke_oplock = False;
-               *p_all_current_opens_are_level_II = True;
-               
-               for(i = 0; i < num_share_modes; i++) {
-                       share_mode_entry *share_entry = &old_shares[i];
-                       BOOL opb_ret;
-                       
 #if defined(DEVELOPER)
-                       validate_my_share_entries(i, share_entry);
+       for(i = 0; i < lck->num_share_modes; i++) {
+               validate_my_share_entries(i, &lck->share_modes[i]);
+       }
 #endif
 
-                       /* 
-                        * By observation of NetBench, oplocks are broken
-                        * *before* share modes are checked. This allows a
-                        * file to be closed by the client if the share mode
-                        * would deny access and the client has an oplock.
-                        * Check if someone has an oplock on this file. If so
-                        * we must break it before continuing.
-                        */
+       if (!lp_share_modes(SNUM(conn))) {
+               return NT_STATUS_OK;
+       }
 
-                       if (!cause_oplock_break(*p_oplock_request,
-                                               share_entry->op_type,
-                                               access_mask)) {
-                               if (!LEVEL_II_OPLOCK_TYPE(share_entry->op_type)) {
-                                       *p_all_current_opens_are_level_II = False;
-                               }
-                               continue;
-                       }
+       /* Now we check the share modes, after any oplock breaks. */
+       for(i = 0; i < lck->num_share_modes; i++) {
 
-                       /* This is an oplock break */
+               if (!is_valid_share_mode_entry(&lck->share_modes[i])) {
+                       continue;
+               }
 
-                       DEBUG(5,("open_mode_check: oplock_request = %d, "
-                                "breaking oplock (%x) on file %s, "
-                                "dev = %x, inode = %.0f\n",
-                                *p_oplock_request, share_entry->op_type,
-                                fname, (unsigned int)dev, (double)inode));
-                               
-                       /* Ensure the reply for the open uses the correct
-                        * sequence number. */
-                       /* This isn't a real deferred packet as it's response
-                        * will also increment the sequence.
-                        */
-                       srv_defer_sign_response(get_current_mid());
+               /* someone else has a share lock on it, check to see if we can
+                * too */
+               if (share_conflict(&lck->share_modes[i],
+                                  access_mask, share_access)) {
+                       return NT_STATUS_SHARING_VIOLATION;
+               }
+       }
+       
+       return NT_STATUS_OK;
+}
 
-                       /* Oplock break - unlock to request it. */
-                       unlock_share_entry(conn, dev, inode);
-                               
-                       opb_ret = request_oplock_break(share_entry);
-                               
-                       /* Now relock. */
-                       lock_share_entry(conn, dev, inode);
-                               
-                       if (!opb_ret) {
-                               DEBUG(0,("open_mode_check: FAILED when breaking "
-                                        "oplock (%x) on file %s, dev = %x, "
-                                        "inode = %.0f\n",
-                                        old_shares[i].op_type, fname,
-                                        (unsigned int)dev, (double)inode));
-                               SAFE_FREE(old_shares);
-                               set_saved_error_triple(ERRDOS, ERRbadshare,
-                                                      NT_STATUS_SHARING_VIOLATION);
-                               return -1;
+static BOOL is_delete_request(files_struct *fsp) {
+       return ((fsp->access_mask == DELETE_ACCESS) &&
+               (fsp->oplock_type == NO_OPLOCK));
+}
+
+/*
+ * 1) No files open at all or internal open: Grant whatever the client wants.
+ *
+ * 2) Exclusive (or batch) oplock around: If the requested access is a delete
+ *    request, break if the oplock around is a batch oplock. If it's another
+ *    requested access type, break.
+ * 
+ * 3) Only level2 around: Grant level2 and do nothing else.
+ */
+
+static BOOL delay_for_oplocks(struct share_mode_lock *lck,
+                             files_struct *fsp,
+                             int pass_number,
+                             int oplock_request)
+{
+       int i;
+       struct share_mode_entry *exclusive = NULL;
+       BOOL valid_entry = False;
+       BOOL delay_it = False;
+       BOOL have_level2 = False;
+       NTSTATUS status;
+       char msg[MSG_SMB_SHARE_MODE_ENTRY_SIZE];
+
+       if (oplock_request & INTERNAL_OPEN_ONLY) {
+               fsp->oplock_type = NO_OPLOCK;
+       }
+
+       if ((oplock_request & INTERNAL_OPEN_ONLY) || is_stat_open(fsp->access_mask)) {
+               return False;
+       }
+
+       for (i=0; i<lck->num_share_modes; i++) {
+
+               if (!is_valid_share_mode_entry(&lck->share_modes[i])) {
+                       continue;
+               }
+
+               /* At least one entry is not an invalid or deferred entry. */
+               valid_entry = True;
+
+               if (pass_number == 1) {
+                       if (BATCH_OPLOCK_TYPE(lck->share_modes[i].op_type)) {
+                               SMB_ASSERT(exclusive == NULL);                  
+                               exclusive = &lck->share_modes[i];
                        }
-                               
-                       broken_entry = SMB_MALLOC_P(struct share_mode_entry_list);
-                       if (!broken_entry) {
-                               smb_panic("open_mode_check: malloc fail.\n");
+               } else {
+                       if (EXCLUSIVE_OPLOCK_TYPE(lck->share_modes[i].op_type)) {
+                               SMB_ASSERT(exclusive == NULL);                  
+                               exclusive = &lck->share_modes[i];
                        }
-                       broken_entry->entry = *share_entry;
-                       DLIST_ADD(broken_entry_list, broken_entry);
-                       broke_oplock = True;
-                               
-               } /* end for */
-               
-               if (broke_oplock) {
-                       /* Update the current open table. */
-                       SAFE_FREE(old_shares);
-                       num_share_modes = get_share_modes(dev, inode,
-                                                         &old_shares,
-                                                         &delete_on_close);
                }
 
-               if (lp_share_modes(SNUM(conn))) {
-                       /* Now we check the share modes, after any oplock breaks. */
-                       for(i = 0; i < num_share_modes; i++) {
-                               share_mode_entry *share_entry = &old_shares[i];
-
-                               /* someone else has a share lock on it, check to see
-                                * if we can too */
-                               if (share_conflict(share_entry, access_mask,
-                                                  share_access)) {
-                                       SAFE_FREE(old_shares);
-                                       free_broken_entry_list(broken_entry_list);
-                                       errno = EACCES;
-                                       return -1;
-                               }
-                       }
+               if (lck->share_modes[i].op_type == LEVEL_II_OPLOCK) {
+                       SMB_ASSERT(exclusive == NULL);                  
+                       have_level2 = True;
                }
+       }
 
-               for(broken_entry = broken_entry_list; broken_entry;
-                   broken_entry = broken_entry->next) {
-                       oplock_contention_count++;
-                       
-                       /* Paranoia check that this is no longer an exlusive entry. */
-                       for(i = 0; i < num_share_modes; i++) {
-                               share_mode_entry *share_entry = &old_shares[i];
-                               
-                               if (!(share_modes_identical(&broken_entry->entry,
-                                                           share_entry) && 
-                                     EXCLUSIVE_OPLOCK_TYPE(share_entry->op_type))) {
-                                       continue;
-                               }
-                                       
-                               /*
-                                * This should not happen. The target left this oplock
-                                * as exlusive.... The process *must* be dead.... 
-                                */
-                                       
-                               DEBUG(0,("open_mode_check: exlusive oplock left by "
-                                        "process %d after break ! For file %s, "
-                                        "dev = %x, inode = %.0f. Deleting it to "
-                                        "continue...\n",
-                                        (int)broken_entry->entry.pid, fname,
-                                        (unsigned int)dev, (double)inode));
-                                       
-                               if (process_exists(broken_entry->entry.pid)) {
-                                       DEBUG(0,("open_mode_check: Existent process "
-                                                "%lu left active oplock.\n",
-                                                (unsigned long)broken_entry->entry.pid ));
-                               }
-                                       
-                               if (del_share_entry(dev, inode, &broken_entry->entry,
-                                                   NULL, &delete_on_close) == -1) {
-                                       free_broken_entry_list(broken_entry_list);
-                                       errno = EACCES;
-                                       set_saved_error_triple(ERRDOS, ERRbadshare,
-                                                              NT_STATUS_SHARING_VIOLATION);
-                                       return -1;
-                               }
-                                       
-                               /*
-                                * We must reload the share modes after deleting the 
-                                * other process's entry.
-                                */
-                                       
-                               SAFE_FREE(old_shares);
-                               num_share_modes = get_share_modes(dev, inode,
-                                                                 &old_shares,
-                                                                 &delete_on_close);
-                               break;
-                       } /* end for paranoia... */
-               } /* end for broken_entry */
-               free_broken_entry_list(broken_entry_list);
-       } while(broke_oplock);
-       
-       /*
-        * Refuse to grant an oplock in case the contention limit is
-        * reached when going through the lock list multiple times.
-        */
-       
-       if(oplock_contention_count >= lp_oplock_contention_limit(SNUM(conn))) {
-               *p_oplock_request = 0;
-               DEBUG(4,("open_mode_check: oplock contention = %d. Not granting oplock.\n",
-                        oplock_contention_count ));
+       if (!valid_entry) {
+               /* All entries are placeholders or deferred.
+                * Directly grant whatever the client wants. */
+               if (fsp->oplock_type == NO_OPLOCK) {
+                       /* Store a level2 oplock, but don't tell the client */
+                       fsp->oplock_type = FAKE_LEVEL_II_OPLOCK;
+               }
+               return False;
        }
-       
-       SAFE_FREE(old_shares);
-       return num_share_modes;
-}
 
-/****************************************************************************
- Delete the record for a handled deferred open entry.
-****************************************************************************/
+       if (exclusive != NULL) { /* Found an exclusive oplock */
+               SMB_ASSERT(!have_level2);
+               delay_it = is_delete_request(fsp) ?
+                       BATCH_OPLOCK_TYPE(exclusive->op_type) : True;
+       }
 
-static void delete_defered_open_entry_record(connection_struct *conn,
-                                               SMB_DEV_T dev,
-                                               SMB_INO_T inode)
-{
-       uint16 mid = get_current_mid();
-       pid_t mypid = sys_getpid();
-       deferred_open_entry *de_array = NULL;
-       int num_de_entries, i;
+       if (EXCLUSIVE_OPLOCK_TYPE(fsp->oplock_type)) {
+               /* We can at most grant level2 as there are other
+                * level2 or NO_OPLOCK entries. */
+               fsp->oplock_type = LEVEL_II_OPLOCK;
+       }
 
-       if (!lp_defer_sharing_violations()) {
-               return;
+       if ((fsp->oplock_type == NO_OPLOCK) && have_level2) {
+               /* Store a level2 oplock, but don't tell the client */
+               fsp->oplock_type = FAKE_LEVEL_II_OPLOCK;
        }
 
-       num_de_entries = get_deferred_opens(conn, dev, inode, &de_array);
-       for (i = 0; i < num_de_entries; i++) {
-               deferred_open_entry *entry = &de_array[i];
-               if (entry->pid == mypid && entry->mid == mid && entry->dev == dev &&
-                               entry->inode == inode) {
+       if (!delay_it) {
+               return False;
+       }
 
-                       /* Remove the deferred open entry from the array. */
-                       delete_deferred_open_entry(entry);
-                       SAFE_FREE(de_array);
-                       return;
-               }
+       /*
+        * Send a break message to the oplock holder and delay the open for
+        * our client.
+        */
+
+       DEBUG(10, ("Sending break request to PID %s\n",
+                  procid_str_static(&exclusive->pid)));
+       exclusive->op_mid = get_current_mid();
+
+       /* Create the message. */
+       share_mode_entry_to_message(msg, exclusive);
+
+       /* Add in the FORCE_OPLOCK_BREAK_TO_NONE bit in the message if set. We
+          don't want this set in the share mode struct pointed to by lck. */
+
+       if (oplock_request & FORCE_OPLOCK_BREAK_TO_NONE) {
+               SSVAL(msg,6,exclusive->op_type | FORCE_OPLOCK_BREAK_TO_NONE);
+       }
+
+       status = messaging_send_buf(smbd_messaging_context(), exclusive->pid,
+                                   MSG_SMB_BREAK_REQUEST,
+                                   (uint8 *)msg,
+                                   MSG_SMB_SHARE_MODE_ENTRY_SIZE);
+       if (!NT_STATUS_IS_OK(status)) {
+               DEBUG(3, ("Could not send oplock break message: %s\n",
+                         nt_errstr(status)));
        }
-       SAFE_FREE(de_array);
+
+       return True;
+}
+
+static BOOL request_timed_out(struct timeval request_time,
+                             struct timeval timeout)
+{
+       struct timeval now, end_time;
+       GetTimeOfDay(&now);
+       end_time = timeval_sum(&request_time, &timeout);
+       return (timeval_compare(&end_time, &now) < 0);
 }
 
 /****************************************************************************
  Handle the 1 second delay in returning a SHARING_VIOLATION error.
 ****************************************************************************/
 
-static void defer_open_sharing_error(connection_struct *conn,
-                                    struct timeval *ptv,
-                                    const char *fname,
-                                    SMB_DEV_T dev,
-                                    SMB_INO_T inode)
+static void defer_open(struct share_mode_lock *lck,
+                      struct timeval request_time,
+                      struct timeval timeout,
+                      struct deferred_open_record *state)
 {
        uint16 mid = get_current_mid();
-       pid_t mypid = sys_getpid();
-       deferred_open_entry *de_array = NULL;
-       int num_de_entries, i;
-       struct dev_inode_bundle dib;
+       int i;
 
-       if (!lp_defer_sharing_violations()) {
-               return;
-       }
+       /* Paranoia check */
 
-       dib.dev = dev;
-       dib.inode = inode;
+       for (i=0; i<lck->num_share_modes; i++) {
+               struct share_mode_entry *e = &lck->share_modes[i];
 
-       num_de_entries = get_deferred_opens(conn, dev, inode, &de_array);
-       for (i = 0; i < num_de_entries; i++) {
-               deferred_open_entry *entry = &de_array[i];
-               if (entry->pid == mypid && entry->mid == mid) {
-                       /*
-                        * Check if a 1 second timeout has expired.
-                        */
-                       if (usec_time_diff(ptv, &entry->time) >
-                           SHARING_VIOLATION_USEC_WAIT) {
-                               DEBUG(10,("defer_open_sharing_error: Deleting "
-                                         "deferred open entry for mid %u, "
-                                         "file %s\n",
-                                         (unsigned int)mid, fname ));
-
-                               /* Expired, return a real error. */
-                               /* Remove the deferred open entry from the array. */
-
-                               delete_deferred_open_entry(entry);
-                               SAFE_FREE(de_array);
-                               return;
-                       }
-                       /*
-                        * If the timeout hasn't expired yet and we still have
-                        * a sharing violation, just leave the entry in the
-                        * deferred open array alone. We do need to reschedule
-                        * this open call though (with the original created
-                        * time).
-                        */
-                       DEBUG(10,("defer_open_sharing_error: time [%u.%06u] "
-                                 "updating deferred open entry for mid %u, file %s\n",
-                                 (unsigned int)entry->time.tv_sec,
-                                 (unsigned int)entry->time.tv_usec,
-                                 (unsigned int)mid, fname ));
-
-                       push_sharing_violation_open_smb_message(&entry->time,
-                                                               (char *)&dib,
-                                                               sizeof(dib));
-                       SAFE_FREE(de_array);
-                       return;
+               if (!is_deferred_open_entry(e)) {
+                       continue;
+               }
+
+               if (procid_is_me(&e->pid) && (e->op_mid == mid)) {
+                       DEBUG(0, ("Trying to defer an already deferred "
+                                 "request: mid=%d, exiting\n", mid));
+                       exit_server("attempt to defer a deferred request");
                }
        }
 
+       /* End paranoia check */
+
        DEBUG(10,("defer_open_sharing_error: time [%u.%06u] adding deferred "
-                 "open entry for mid %u, file %s\n",
-                 (unsigned int)ptv->tv_sec, (unsigned int)ptv->tv_usec,
-                 (unsigned int)mid, fname ));
+                 "open entry for mid %u\n",
+                 (unsigned int)request_time.tv_sec,
+                 (unsigned int)request_time.tv_usec,
+                 (unsigned int)mid));
 
-       if (!push_sharing_violation_open_smb_message(ptv, (char *)&dib, sizeof(dib))) {
-               SAFE_FREE(de_array);
-               return;
-       }
-       if (!add_deferred_open(mid, ptv, dev, inode, global_oplock_port, fname)) {
-               remove_sharing_violation_open_smb_message(mid);
+       if (!push_deferred_smb_message(mid, request_time, timeout,
+                                      (char *)state, sizeof(*state))) {
+               exit_server("push_deferred_smb_message failed");
        }
+       add_deferred_open(lck, mid, request_time, state->dev, state->inode);
 
        /*
         * Push the MID of this packet on the signing queue.
@@ -889,44 +813,20 @@ static void defer_open_sharing_error(connection_struct *conn,
         */
 
        srv_defer_sign_response(mid);
-
-       SAFE_FREE(de_array);
 }
 
-/****************************************************************************
- Set a kernel flock on a file for NFS interoperability.
- This requires a patch to Linux.
-****************************************************************************/
-
-static void kernel_flock(files_struct *fsp, uint32 share_mode)
-{
-#if HAVE_KERNEL_SHARE_MODES
-       int kernel_mode = 0;
-       if (share_mode == FILE_SHARE_WRITE) {
-               kernel_mode = LOCK_MAND|LOCK_WRITE;
-       } else if (share_mode == FILE_SHARE_READ) {
-               kernel_mode = LOCK_MAND|LOCK_READ;
-       } else if (share_mode == FILE_SHARE_NONE) {
-               kernel_mode = LOCK_MAND;
-       }
-       if (kernel_mode) {
-               flock(fsp->fh->fd, kernel_mode);
-       }
-#endif
-       ;
-}
 
 /****************************************************************************
  On overwrite open ensure that the attributes match.
 ****************************************************************************/
 
 static BOOL open_match_attributes(connection_struct *conn,
-                               const char *path,
-                               uint32 old_dos_attr,
-                               uint32 new_dos_attr,
-                               mode_t existing_unx_mode,
-                               mode_t new_unx_mode,
-                               mode_t *returned_unx_mode)
+                                 const char *path,
+                                 uint32 old_dos_attr,
+                                 uint32 new_dos_attr,
+                                 mode_t existing_unx_mode,
+                                 mode_t new_unx_mode,
+                                 mode_t *returned_unx_mode)
 {
        uint32 noarch_old_dos_attr, noarch_new_dos_attr;
 
@@ -1018,8 +918,8 @@ static files_struct *fcb_or_dos_open(connection_struct *conn,
        }
 
        /* We need to duplicate this fsp. */
-       dup_fsp = dup_file_fsp(fsp, access_mask, share_access, create_options);
-       if (!dup_fsp) {
+       if (!NT_STATUS_IS_OK(dup_file_fsp(fsp, access_mask, share_access,
+                                         create_options, &dup_fsp))) {
                return NULL;
        }
 
@@ -1031,10 +931,10 @@ static files_struct *fcb_or_dos_open(connection_struct *conn,
 ****************************************************************************/
 
 BOOL map_open_params_to_ntcreate(const char *fname, int deny_mode, int open_func,
-                               uint32 *paccess_mask,
-                               uint32 *pshare_mode,
-                               uint32 *pcreate_disposition,
-                               uint32 *pcreate_options)
+                                uint32 *paccess_mask,
+                                uint32 *pshare_mode,
+                                uint32 *pcreate_disposition,
+                                uint32 *pcreate_options)
 {
        uint32 access_mask;
        uint32 share_mode;
@@ -1047,13 +947,13 @@ BOOL map_open_params_to_ntcreate(const char *fname, int deny_mode, int open_func
 
        /* Create the NT compatible access_mask. */
        switch (GET_OPENX_MODE(deny_mode)) {
+               case DOS_OPEN_EXEC: /* Implies read-only - used to be FILE_READ_DATA */
                case DOS_OPEN_RDONLY:
                        access_mask = FILE_GENERIC_READ;
                        break;
                case DOS_OPEN_WRONLY:
                        access_mask = FILE_GENERIC_WRITE;
                        break;
-               case DOS_OPEN_EXEC: /* This used to be FILE_READ_DATA... */
                case DOS_OPEN_RDWR:
                case DOS_OPEN_FCB:
                        access_mask = FILE_GENERIC_READ|FILE_GENERIC_WRITE;
@@ -1165,49 +1065,82 @@ BOOL map_open_params_to_ntcreate(const char *fname, int deny_mode, int open_func
 
 }
 
-/* Map generic permissions to file object specific permissions */
-                                                                                                               
-struct generic_mapping file_generic_mapping = {
-       FILE_GENERIC_READ,
-       FILE_GENERIC_WRITE,
-       FILE_GENERIC_EXECUTE,
-       FILE_GENERIC_ALL
-};
+static void schedule_defer_open(struct share_mode_lock *lck, struct timeval request_time)
+{
+       struct deferred_open_record state;
+
+       /* This is a relative time, added to the absolute
+          request_time value to get the absolute timeout time.
+          Note that if this is the second or greater time we enter
+          this codepath for this particular request mid then
+          request_time is left as the absolute time of the *first*
+          time this request mid was processed. This is what allows
+          the request to eventually time out. */
+
+       struct timeval timeout;
+
+       /* Normally the smbd we asked should respond within
+        * OPLOCK_BREAK_TIMEOUT seconds regardless of whether
+        * the client did, give twice the timeout as a safety
+        * measure here in case the other smbd is stuck
+        * somewhere else. */
+
+       timeout = timeval_set(OPLOCK_BREAK_TIMEOUT*2, 0);
+
+       /* Nothing actually uses state.delayed_for_oplocks
+          but it's handy to differentiate in debug messages
+          between a 30 second delay due to oplock break, and
+          a 1 second delay for share mode conflicts. */
+
+       state.delayed_for_oplocks = True;
+       state.dev = lck->dev;
+       state.inode = lck->ino;
+
+       if (!request_timed_out(request_time, timeout)) {
+               defer_open(lck, request_time, timeout, &state);
+       }
+}
 
 /****************************************************************************
  Open a file with a share mode.
 ****************************************************************************/
 
-files_struct *open_file_ntcreate(connection_struct *conn,
-                                const char *fname,
-                                SMB_STRUCT_STAT *psbuf,
-                                uint32 access_mask,            /* access bits (FILE_READ_DATA etc.) */
-                                uint32 share_access,           /* share constants (FILE_SHARE_READ etc). */
-                                uint32 create_disposition,     /* FILE_OPEN_IF etc. */
-                                uint32 create_options,         /* options such as delete on close. */
-                                uint32 new_dos_attributes,     /* attributes used for new file. */
-                                int oplock_request,            /* internal Samba oplock codes. */
-                                                               /* Information (FILE_EXISTS etc.) */
-                                int *pinfo)
+NTSTATUS open_file_ntcreate(connection_struct *conn,
+                           const char *fname,
+                           SMB_STRUCT_STAT *psbuf,
+                           uint32 access_mask,         /* access bits (FILE_READ_DATA etc.) */
+                           uint32 share_access,        /* share constants (FILE_SHARE_READ etc) */
+                           uint32 create_disposition,  /* FILE_OPEN_IF etc. */
+                           uint32 create_options,      /* options such as delete on close. */
+                           uint32 new_dos_attributes,  /* attributes used for new file. */
+                           int oplock_request,         /* internal Samba oplock codes. */
+                                                       /* Information (FILE_EXISTS etc.) */
+                           int *pinfo,
+                           files_struct **result)
 {
        int flags=0;
        int flags2=0;
        BOOL file_existed = VALID_STAT(*psbuf);
        BOOL def_acl = False;
-       BOOL internal_only_open = False;
+       BOOL posix_open = False;
+       BOOL new_file_created = False;
        SMB_DEV_T dev = 0;
        SMB_INO_T inode = 0;
-       int num_share_modes = 0;
-       BOOL all_current_opens_are_level_II = False;
-       BOOL fsp_open = False;
+       NTSTATUS fsp_open = NT_STATUS_ACCESS_DENIED;
        files_struct *fsp = NULL;
        mode_t new_unx_mode = (mode_t)0;
        mode_t unx_mode = (mode_t)0;
        int info;
        uint32 existing_dos_attributes = 0;
        struct pending_message_list *pml = NULL;
-       uint16 port = 0;
        uint16 mid = get_current_mid();
+       struct timeval request_time = timeval_zero();
+       struct share_mode_lock *lck = NULL;
+       uint32 open_access_mask = access_mask;
+       NTSTATUS status;
+       int ret_flock;
+       char *parent_dir;
+       const char *newname;
 
        if (conn->printer) {
                /* 
@@ -1221,12 +1154,24 @@ files_struct *open_file_ntcreate(connection_struct *conn,
 
                DEBUG(10, ("open_file_ntcreate: printer open fname=%s\n", fname));
 
-               return print_fsp_open(conn, fname);
+               return print_fsp_open(conn, fname, result);
        }
 
-       /* We add aARCH to this as this mode is only used if the file is
-        * created new. */
-       unx_mode = unix_mode(conn, new_dos_attributes | aARCH,fname, True);
+       if (!parent_dirname_talloc(tmp_talloc_ctx(), fname, &parent_dir,
+                                  &newname)) {
+               return NT_STATUS_NO_MEMORY;
+       }
+
+       if (new_dos_attributes & FILE_FLAG_POSIX_SEMANTICS) {
+               posix_open = True;
+               unx_mode = (mode_t)(new_dos_attributes & ~FILE_FLAG_POSIX_SEMANTICS);
+               new_dos_attributes = 0;
+       } else {
+               /* We add aARCH to this as this mode is only used if the file is
+                * created new. */
+               unx_mode = unix_mode(conn, new_dos_attributes | aARCH, fname,
+                                    parent_dir);
+       }
 
        DEBUG(10, ("open_file_ntcreate: fname=%s, dos_attrs=0x%x "
                   "access_mask=0x%x share_access=0x%x "
@@ -1236,69 +1181,58 @@ files_struct *open_file_ntcreate(connection_struct *conn,
                   create_disposition, create_options, unx_mode,
                   oplock_request));
 
-       if (oplock_request == INTERNAL_OPEN_ONLY) {
-               internal_only_open = True;
-               oplock_request = 0;
-       }
-
        if ((pml = get_open_deferred_message(mid)) != NULL) {
-               struct dev_inode_bundle dib;
-
-               memcpy(&dib, pml->private_data.data, sizeof(dib));
-
-               /* There could be a race condition where the dev/inode pair
-                  has changed since we deferred the message. If so, just
-                  remove the deferred open entry and return sharing
-                  violation. */
-
-               /* If the timeout value is non-zero, we need to just return
-                  sharing violation. Don't retry the open as we were not
-                  notified of a close and we don't want to trigger another
-                  spurious oplock break. */
-
-               if (!file_existed || dib.dev != psbuf->st_dev ||
-                   dib.inode != psbuf->st_ino || pml->msg_time.tv_sec ||
-                   pml->msg_time.tv_usec) {
-                       /* Ensure we don't reprocess this message. */
-                       remove_sharing_violation_open_smb_message(mid);
-
-                       /* Now remove the deferred open entry under lock. */
-                       lock_share_entry(conn, dib.dev, dib.inode);
-                       delete_defered_open_entry_record(conn, dib.dev,
-                                                        dib.inode);
-                       unlock_share_entry(conn, dib.dev, dib.inode);
-
-                       set_saved_error_triple(ERRDOS, ERRbadshare,
-                                              NT_STATUS_SHARING_VIOLATION);
-                       return NULL;
+               struct deferred_open_record *state =
+                       (struct deferred_open_record *)pml->private_data.data;
+
+               /* Remember the absolute time of the original
+                  request with this mid. We'll use it later to
+                  see if this has timed out. */
+
+               request_time = pml->request_time;
+
+               /* Remove the deferred open entry under lock. */
+               lck = get_share_mode_lock(NULL, state->dev, state->inode, NULL, NULL);
+               if (lck == NULL) {
+                       DEBUG(0, ("could not get share mode lock\n"));
+               } else {
+                       del_deferred_open_entry(lck, mid);
+                       TALLOC_FREE(lck);
                }
+
                /* Ensure we don't reprocess this message. */
-               remove_sharing_violation_open_smb_message(mid);
+               remove_deferred_open_smb_message(mid);
        }
 
-       if (!check_name(fname,conn)) {
-               return NULL;
+       status = check_name(conn, fname);
+       if (!NT_STATUS_IS_OK(status)) {
+               return status;
        } 
 
-       new_dos_attributes &= SAMBA_ATTRIBUTES_MASK;
-       if (file_existed) {
-               existing_dos_attributes = dos_mode(conn, fname, psbuf);
+       if (!posix_open) {
+               new_dos_attributes &= SAMBA_ATTRIBUTES_MASK;
+               if (file_existed) {
+                       existing_dos_attributes = dos_mode(conn, fname, psbuf);
+               }
        }
 
        /* ignore any oplock requests if oplocks are disabled */
-       if (!lp_oplocks(SNUM(conn)) || global_client_failed_oplock_break) {
-               oplock_request = 0;
+       if (!lp_oplocks(SNUM(conn)) || global_client_failed_oplock_break ||
+           IS_VETO_OPLOCK_PATH(conn, fname)) {
+               /* Mask off everything except the private Samba bits. */
+               oplock_request &= SAMBA_PRIVATE_OPLOCK_MASK;
        }
 
        /* this is for OS/2 long file names - say we don't support them */
        if (!lp_posix_pathnames() && strstr(fname,".+,;=[].")) {
                /* OS/2 Workplace shell fix may be main code stream in a later
-                * release. */ 
-               set_saved_error_triple(ERRDOS, ERRcannotopen,
-                                      NT_STATUS_OBJECT_NAME_NOT_FOUND);
+                * release. */
                DEBUG(5,("open_file_ntcreate: OS/2 long filenames are not "
                         "supported.\n"));
-               return NULL;
+               if (use_nt_status()) {
+                       return NT_STATUS_OBJECT_NAME_NOT_FOUND;
+               }
+               return NT_STATUS_DOS(ERRDOS, ERRcannotopen);
        }
 
        switch( create_disposition ) {
@@ -1326,9 +1260,8 @@ files_struct *open_file_ntcreate(connection_struct *conn,
                                DEBUG(5,("open_file_ntcreate: FILE_OPEN "
                                         "requested for file %s and file "
                                         "doesn't exist.\n", fname ));
-                               set_saved_error_triple(ERRDOS, ERRbadfile, NT_STATUS_OBJECT_NAME_NOT_FOUND);
                                errno = ENOENT;
-                               return NULL;
+                               return NT_STATUS_OBJECT_NAME_NOT_FOUND;
                        }
                        break;
 
@@ -1339,9 +1272,8 @@ files_struct *open_file_ntcreate(connection_struct *conn,
                                DEBUG(5,("open_file_ntcreate: FILE_OVERWRITE "
                                         "requested for file %s and file "
                                         "doesn't exist.\n", fname ));
-                               set_saved_error_triple(ERRDOS, ERRbadfile, NT_STATUS_OBJECT_NAME_NOT_FOUND);
                                errno = ENOENT;
-                               return NULL;
+                               return NT_STATUS_OBJECT_NAME_NOT_FOUND;
                        }
                        flags2 |= O_TRUNC;
                        break;
@@ -1358,7 +1290,7 @@ files_struct *open_file_ntcreate(connection_struct *conn,
                                } else {
                                        errno = EEXIST;
                                }
-                               return NULL;
+                               return map_nt_error_from_unix(errno);
                        }
                        flags2 |= (O_CREAT|O_EXCL);
                        break;
@@ -1370,15 +1302,13 @@ files_struct *open_file_ntcreate(connection_struct *conn,
                        break;
 
                default:
-                       set_saved_error_triple(ERRDOS, ERRinvalidparam,
-                                              NT_STATUS_INVALID_PARAMETER);
-                       return NULL;
+                       return NT_STATUS_INVALID_PARAMETER;
        }
 
        /* We only care about matching attributes on file exists and
         * overwrite. */
 
-       if (file_existed && ((create_disposition == FILE_OVERWRITE) ||
+       if (!posix_open && file_existed && ((create_disposition == FILE_OVERWRITE) ||
                             (create_disposition == FILE_OVERWRITE_IF))) {
                if (!open_match_attributes(conn, fname,
                                           existing_dos_attributes,
@@ -1391,13 +1321,13 @@ files_struct *open_file_ntcreate(connection_struct *conn,
                                 (unsigned int)psbuf->st_mode,
                                 (unsigned int)unx_mode ));
                        errno = EACCES;
-                       return NULL;
+                       return NT_STATUS_ACCESS_DENIED;
                }
        }
 
        /* This is a nasty hack - must fix... JRA. */
        if (access_mask == MAXIMUM_ALLOWED_ACCESS) {
-               access_mask = FILE_GENERIC_ALL;
+               open_access_mask = access_mask = FILE_GENERIC_ALL;
        }
 
        /*
@@ -1405,6 +1335,11 @@ files_struct *open_file_ntcreate(connection_struct *conn,
         */
 
        se_map_generic(&access_mask, &file_generic_mapping);
+       open_access_mask = access_mask;
+
+       if (flags2 & O_TRUNC) {
+               open_access_mask |= FILE_WRITE_DATA; /* This will cause oplock breaks. */
+       }
 
        DEBUG(10, ("open_file_ntcreate: fname=%s, after mapping "
                   "access_mask=0x%x\n", fname, access_mask ));
@@ -1415,7 +1350,15 @@ files_struct *open_file_ntcreate(connection_struct *conn,
         */
 
        if (access_mask & (FILE_WRITE_DATA | FILE_APPEND_DATA)) {
-               flags = O_RDWR;
+               /* DENY_DOS opens are always underlying read-write on the
+                  file handle, no matter what the requested access mask
+                   says. */
+               if ((create_options & NTCREATEX_OPTIONS_PRIVATE_DENY_DOS) ||
+                       access_mask & (FILE_READ_ATTRIBUTES|FILE_READ_DATA|FILE_READ_EA|FILE_EXECUTE)) {
+                       flags = O_RDWR;
+               } else {
+                       flags = O_WRONLY;
+               }
        } else {
                flags = O_RDONLY;
        }
@@ -1425,12 +1368,16 @@ files_struct *open_file_ntcreate(connection_struct *conn,
         */
 
 #if defined(O_SYNC)
-       if (create_options & FILE_WRITE_THROUGH) {
+       if ((create_options & FILE_WRITE_THROUGH) && lp_strict_sync(SNUM(conn))) {
                flags2 |= O_SYNC;
        }
 #endif /* O_SYNC */
   
-       if (!CAN_WRITE(conn)) {
+       if (posix_open & (access_mask & FILE_APPEND_DATA)) {
+               flags2 |= O_APPEND;
+       }
+
+       if (!posix_open && !CAN_WRITE(conn)) {
                /*
                 * We should really return a permission denied error if either
                 * O_CREAT or O_TRUNC are set, but for compatibility with
@@ -1448,56 +1395,108 @@ files_struct *open_file_ntcreate(connection_struct *conn,
                DEBUG(5,("open_file_ntcreate: write access requested for "
                         "file %s on read only %s\n",
                         fname, !CAN_WRITE(conn) ? "share" : "file" ));
-               set_saved_error_triple(ERRDOS, ERRnoaccess,
-                                      NT_STATUS_ACCESS_DENIED);
                errno = EACCES;
-               return NULL;
+               return NT_STATUS_ACCESS_DENIED;
        }
 
-       fsp = file_new(conn);
-       if(!fsp) {
-               return NULL;
+       status = file_new(conn, &fsp);
+       if(!NT_STATUS_IS_OK(status)) {
+               return status;
        }
 
-       if (file_existed) {
+       fsp->dev = psbuf->st_dev;
+       fsp->inode = psbuf->st_ino;
+       fsp->share_access = share_access;
+       fsp->fh->private_options = create_options;
+       fsp->access_mask = open_access_mask; /* We change this to the
+                                             * requested access_mask after
+                                             * the open is done. */
+       fsp->posix_open = posix_open;
+
+       /* Ensure no SAMBA_PRIVATE bits can be set. */
+       fsp->oplock_type = (oplock_request & ~SAMBA_PRIVATE_OPLOCK_MASK);
+
+       if (timeval_is_zero(&request_time)) {
+               request_time = fsp->open_time;
+       }
 
+       if (file_existed) {
                dev = psbuf->st_dev;
                inode = psbuf->st_ino;
 
-               lock_share_entry(conn, dev, inode);
-
-               num_share_modes = open_mode_check(conn, fname, dev, inode,
-                                                 access_mask, share_access,
-                                                 create_options,
-                                                 &flags, &oplock_request,
-                                                 &all_current_opens_are_level_II);
-               if(num_share_modes == -1) {
-
-                       if (!internal_only_open) {
-                               NTSTATUS status;
-                               get_saved_error_triple(NULL, NULL, &status);
-                               if (NT_STATUS_EQUAL(status,NT_STATUS_SHARING_VIOLATION)) {
-                                       /* Check if this can be done with the
-                                        * deny_dos and fcb calls. */
-                                       if (create_options &
-                                           (NTCREATEX_OPTIONS_PRIVATE_DENY_DOS|
-                                            NTCREATEX_OPTIONS_PRIVATE_DENY_FCB)) {
-                                               files_struct *fsp_dup;
-                                               fsp_dup = fcb_or_dos_open(conn, fname, dev,
-                                                                         inode, access_mask,
-                                                                         share_access,
-                                                                         create_options);
-
-                                               if (fsp_dup) {
-                                                       unlock_share_entry(conn, dev, inode);
-                                                       file_free(fsp);
-                                                       if (pinfo) {
-                                                               *pinfo = FILE_WAS_OPENED;
-                                                       }
-                                                       conn->num_files_open++;
-                                                       return fsp_dup;
-                                               }
+               lck = get_share_mode_lock(NULL, dev, inode,
+                                         conn->connectpath,
+                                         fname);
+
+               if (lck == NULL) {
+                       file_free(fsp);
+                       DEBUG(0, ("Could not get share mode lock\n"));
+                       return NT_STATUS_SHARING_VIOLATION;
+               }
+
+               /* First pass - send break only on batch oplocks. */
+               if (delay_for_oplocks(lck, fsp, 1, oplock_request)) {
+                       schedule_defer_open(lck, request_time);
+                       TALLOC_FREE(lck);
+                       file_free(fsp);
+                       return NT_STATUS_SHARING_VIOLATION;
+               }
+
+               /* Use the client requested access mask here, not the one we
+                * open with. */
+               status = open_mode_check(conn, fname, lck,
+                                        access_mask, share_access,
+                                        create_options, &file_existed);
+
+               if (NT_STATUS_IS_OK(status)) {
+                       /* We might be going to allow this open. Check oplock
+                        * status again. */
+                       /* Second pass - send break for both batch or
+                        * exclusive oplocks. */
+                       if (delay_for_oplocks(lck, fsp, 2, oplock_request)) {
+                               schedule_defer_open(lck, request_time);
+                               TALLOC_FREE(lck);
+                               file_free(fsp);
+                               return NT_STATUS_SHARING_VIOLATION;
+                       }
+               }
+
+               if (NT_STATUS_EQUAL(status, NT_STATUS_DELETE_PENDING)) {
+                       /* DELETE_PENDING is not deferred for a second */
+                       TALLOC_FREE(lck);
+                       file_free(fsp);
+                       return status;
+               }
+
+               if (!NT_STATUS_IS_OK(status)) {
+                       uint32 can_access_mask;
+                       BOOL can_access = True;
+
+                       SMB_ASSERT(NT_STATUS_EQUAL(status, NT_STATUS_SHARING_VIOLATION));
+
+                       /* Check if this can be done with the deny_dos and fcb
+                        * calls. */
+                       if (create_options &
+                           (NTCREATEX_OPTIONS_PRIVATE_DENY_DOS|
+                            NTCREATEX_OPTIONS_PRIVATE_DENY_FCB)) {
+                               files_struct *fsp_dup;
+
+                               /* Use the client requested access mask here,
+                                * not the one we open with. */
+                               fsp_dup = fcb_or_dos_open(conn, fname, dev,
+                                                         inode, access_mask,
+                                                         share_access,
+                                                         create_options);
+
+                               if (fsp_dup) {
+                                       TALLOC_FREE(lck);
+                                       file_free(fsp);
+                                       if (pinfo) {
+                                               *pinfo = FILE_WAS_OPENED;
                                        }
+                                       conn->num_files_open++;
+                                       *result = fsp_dup;
+                                       return NT_STATUS_OK;
                                }
                        }
 
@@ -1506,30 +1505,22 @@ files_struct *open_file_ntcreate(connection_struct *conn,
                         * MS-Access. If a file open will fail due to share
                         * permissions and also for security (access) reasons,
                         * we need to return the access failed error, not the
-                        * share error. This means we must attempt to open the
-                        * file anyway in order to get the UNIX access error -
-                        * even if we're going to fail the open for share
-                        * reasons. This is bad, as we're burning another fd
-                        * if there are existing locks but there's nothing
-                        * else we can do. We also ensure we're not going to
-                        * create or tuncate the file as we only want an
-                        * access decision at this stage. JRA.
+                        * share error. We can't open the file due to kernel
+                        * oplock deadlock (it's possible we failed above on
+                        * the open_mode_check()) so use a userspace check.
                         */
-                       errno = 0;
-                       fsp_open = open_file(fsp,conn,fname,psbuf,
-                                            flags|(flags2&~(O_TRUNC|O_CREAT)),
-                                            unx_mode,access_mask);
-
-                       DEBUG(4,("open_file_ntcreate : share_mode deny - "
-                                "calling open_file with flags=0x%X "
-                                "flags2=0x%X mode=0%o returned %d\n",
-                                flags, (flags2&~(O_TRUNC|O_CREAT)),
-                                (unsigned int)unx_mode, (int)fsp_open ));
-
-                       if (!fsp_open && errno) {
-                               /* Default error. */
-                               set_saved_error_triple(ERRDOS, ERRnoaccess,
-                                                      NT_STATUS_ACCESS_DENIED);
+
+                       if (flags & O_RDWR) {
+                               can_access_mask = FILE_READ_DATA|FILE_WRITE_DATA;
+                       } else if (flags & O_WRONLY) {
+                               can_access_mask = FILE_WRITE_DATA;
+                       } else {
+                               can_access_mask = FILE_READ_DATA;
+                       }
+
+                       if (((can_access_mask & FILE_WRITE_DATA) && !CAN_WRITE(conn)) ||
+                           !can_access_file(conn,fname,psbuf,can_access_mask)) {
+                               can_access = False;
                        }
 
                        /* 
@@ -1537,30 +1528,56 @@ files_struct *open_file_ntcreate(connection_struct *conn,
                         * cope with the braindead 1 second delay.
                         */
 
-                       if (!internal_only_open) {
-                               NTSTATUS status;
-                               get_saved_error_triple(NULL, NULL, &status);
-                               if (NT_STATUS_EQUAL(status,NT_STATUS_SHARING_VIOLATION)) {
-                                       /* The fsp->open_time here represents
-                                        * the current time of day. */
-                                       defer_open_sharing_error(conn,
-                                                                &fsp->open_time,
-                                                                fname, dev, inode);
+                       if (!(oplock_request & INTERNAL_OPEN_ONLY) &&
+                           lp_defer_sharing_violations()) {
+                               struct timeval timeout;
+                               struct deferred_open_record state;
+                               int timeout_usecs;
+
+                               /* this is a hack to speed up torture tests
+                                  in 'make test' */
+                               timeout_usecs = lp_parm_int(SNUM(conn),
+                                                           "smbd","sharedelay",
+                                                           SHARING_VIOLATION_USEC_WAIT);
+
+                               /* This is a relative time, added to the absolute
+                                  request_time value to get the absolute timeout time.
+                                  Note that if this is the second or greater time we enter
+                                  this codepath for this particular request mid then
+                                  request_time is left as the absolute time of the *first*
+                                  time this request mid was processed. This is what allows
+                                  the request to eventually time out. */
+
+                               timeout = timeval_set(0, timeout_usecs);
+
+                               /* Nothing actually uses state.delayed_for_oplocks
+                                  but it's handy to differentiate in debug messages
+                                  between a 30 second delay due to oplock break, and
+                                  a 1 second delay for share mode conflicts. */
+
+                               state.delayed_for_oplocks = False;
+                               state.dev = dev;
+                               state.inode = inode;
+
+                               if (!request_timed_out(request_time,
+                                                      timeout)) {
+                                       defer_open(lck, request_time, timeout,
+                                                  &state);
                                }
                        }
 
-                       unlock_share_entry(conn, dev, inode);
-                       if (fsp_open) {
-                               fd_close(conn, fsp);
+                       TALLOC_FREE(lck);
+                       if (can_access) {
                                /*
                                 * We have detected a sharing violation here
                                 * so return the correct error code
                                 */
-                               set_saved_error_triple(ERRDOS, ERRbadshare,
-                                                      NT_STATUS_SHARING_VIOLATION);
+                               status = NT_STATUS_SHARING_VIOLATION;
+                       } else {
+                               status = NT_STATUS_ACCESS_DENIED;
                        }
                        file_free(fsp);
-                       return NULL;
+                       return status;
                }
 
                /*
@@ -1568,47 +1585,48 @@ files_struct *open_file_ntcreate(connection_struct *conn,
                 */
        }
 
+       SMB_ASSERT(!file_existed || (lck != NULL));
+
        /*
         * Ensure we pay attention to default ACLs on directories if required.
         */
 
         if ((flags2 & O_CREAT) && lp_inherit_acls(SNUM(conn)) &&
-                       (def_acl = directory_has_default_acl(conn, parent_dirname(fname)))) {
+           (def_acl = directory_has_default_acl(conn, parent_dir))) {
                unx_mode = 0777;
        }
 
-       DEBUG(4,("calling open_file with flags=0x%X flags2=0x%X mode=0%o\n",
-                       (unsigned int)flags,(unsigned int)flags2,(unsigned int)unx_mode));
+       DEBUG(4,("calling open_file with flags=0x%X flags2=0x%X mode=0%o, "
+               "access_mask = 0x%x, open_access_mask = 0x%x\n",
+                (unsigned int)flags, (unsigned int)flags2,
+                (unsigned int)unx_mode, (unsigned int)access_mask,
+                (unsigned int)open_access_mask));
 
        /*
         * open_file strips any O_TRUNC flags itself.
         */
 
-       fsp_open = open_file(fsp,conn,fname,psbuf,flags|flags2,unx_mode,access_mask);
+       fsp_open = open_file(fsp, conn, parent_dir, newname, fname, psbuf,
+                            flags|flags2, unx_mode, access_mask,
+                            open_access_mask);
 
-       if (!fsp_open && (flags == O_RDWR) && (errno != ENOENT)) {
-               if((fsp_open = open_file(fsp,conn,fname,psbuf,
-                                        O_RDONLY,unx_mode,access_mask)) == True) {
-                       flags = O_RDONLY;
-               }
-       }
-
-       if (!fsp_open) {
-               if(file_existed) {
-                       unlock_share_entry(conn, dev, inode);
+       if (!NT_STATUS_IS_OK(fsp_open)) {
+               if (lck != NULL) {
+                       TALLOC_FREE(lck);
                }
                file_free(fsp);
-               return NULL;
+               return fsp_open;
        }
 
-       /*
-        * Deal with the race condition where two smbd's detect the file
-        * doesn't exist and do the create at the same time. One of them will
-        * win and set a share mode, the other (ie. this one) should check if
-        * the requested share mode for this create is allowed.
-        */
+       if (!file_existed) {
 
-       if (!file_existed) { 
+               /*
+                * Deal with the race condition where two smbd's detect the
+                * file doesn't exist and do the create at the same time. One
+                * of them will win and set a share mode, the other (ie. this
+                * one) should check if the requested share mode for this
+                * create is allowed.
+                */
 
                /*
                 * Now the file exists and fsp is successfully opened,
@@ -1620,85 +1638,93 @@ files_struct *open_file_ntcreate(connection_struct *conn,
                dev = fsp->dev;
                inode = fsp->inode;
 
-               lock_share_entry_fsp(fsp);
-
-               num_share_modes = open_mode_check(conn, fname, dev, inode,
-                                                 access_mask, share_access,
-                                                 create_options,
-                                                 &flags, &oplock_request,
-                                                 &all_current_opens_are_level_II);
-
-               if(num_share_modes == -1) {
-                       NTSTATUS status;
-                       get_saved_error_triple(NULL, NULL, &status);
-                       if (NT_STATUS_EQUAL(status,NT_STATUS_SHARING_VIOLATION)) {
-                               /* Check if this can be done with the deny_dos
-                                * and fcb calls. */
-                               if (create_options &
-                                   (NTCREATEX_OPTIONS_PRIVATE_DENY_DOS|
-                                    NTCREATEX_OPTIONS_PRIVATE_DENY_FCB)) {
-                                       files_struct *fsp_dup;
-                                       fsp_dup = fcb_or_dos_open(conn, fname, dev, inode,
-                                                                 access_mask, share_access,
-                                                                 create_options);
-                                       if (fsp_dup) {
-                                               unlock_share_entry(conn, dev, inode);
-                                               fd_close(conn, fsp);
-                                               file_free(fsp);
-                                               if (pinfo) {
-                                                       *pinfo = FILE_WAS_OPENED;
-                                               }
-                                               conn->num_files_open++;
-                                               return fsp_dup;
-                                       }
-                               }
-
-                               /* 
-                                * If we're returning a share violation,
-                                * ensure we cope with the braindead 1 second
-                                * delay.
-                                */
+               lck = get_share_mode_lock(NULL, dev, inode,
+                                         conn->connectpath,
+                                         fname);
 
-                               /* The fsp->open_time here represents the
-                                * current time of day. */
-                               defer_open_sharing_error(conn, &fsp->open_time,
-                                                        fname, dev, inode);
-                       }
+               if (lck == NULL) {
+                       DEBUG(0, ("open_file_ntcreate: Could not get share "
+                                 "mode lock for %s\n", fname));
+                       fd_close(conn, fsp);
+                       file_free(fsp);
+                       return NT_STATUS_SHARING_VIOLATION;
+               }
 
-                       unlock_share_entry_fsp(fsp);
-                       fd_close(conn,fsp);
+               /* First pass - send break only on batch oplocks. */
+               if (delay_for_oplocks(lck, fsp, 1, oplock_request)) {
+                       schedule_defer_open(lck, request_time);
+                       TALLOC_FREE(lck);
+                       fd_close(conn, fsp);
                        file_free(fsp);
-                       /*
-                        * We have detected a sharing violation here, so
-                        * return the correct code.
-                        */
-                       set_saved_error_triple(ERRDOS, ERRbadshare,
-                                              NT_STATUS_SHARING_VIOLATION);
-                       return NULL;
+                       return NT_STATUS_SHARING_VIOLATION;
                }
 
-               /*
-                * If there are any share modes set then the file *did*
-                * exist. Ensure we return the correct value for action.
-                */
+               status = open_mode_check(conn, fname, lck,
+                                        access_mask, share_access,
+                                        create_options, &file_existed);
+
+               if (NT_STATUS_IS_OK(status)) {
+                       /* We might be going to allow this open. Check oplock
+                        * status again. */
+                       /* Second pass - send break for both batch or
+                        * exclusive oplocks. */
+                       if (delay_for_oplocks(lck, fsp, 2, oplock_request)) {
+                               schedule_defer_open(lck, request_time);
+                               TALLOC_FREE(lck);
+                               fd_close(conn, fsp);
+                               file_free(fsp);
+                               return NT_STATUS_SHARING_VIOLATION;
+                       }
+               }
+
+               if (!NT_STATUS_IS_OK(status)) {
+                       struct deferred_open_record state;
 
-               if (num_share_modes > 0) {
-                       file_existed = True;
+                       fd_close(conn, fsp);
+                       file_free(fsp);
+
+                       state.delayed_for_oplocks = False;
+                       state.dev = dev;
+                       state.inode = inode;
+
+                       /* Do it all over again immediately. In the second
+                        * round we will find that the file existed and handle
+                        * the DELETE_PENDING and FCB cases correctly. No need
+                        * to duplicate the code here. Essentially this is a
+                        * "goto top of this function", but don't tell
+                        * anybody... */
+
+                       defer_open(lck, request_time, timeval_zero(),
+                                  &state);
+                       TALLOC_FREE(lck);
+                       return status;
                }
 
                /*
                 * We exit this block with the share entry *locked*.....
                 */
+
        }
 
+       SMB_ASSERT(lck != NULL);
+
        /* note that we ignore failure for the following. It is
            basically a hack for NFS, and NFS will never set one of
            these only read them. Nobody but Samba can ever set a deny
            mode and we have already checked our more authoritative
            locking database for permission to set this deny mode. If
-           the kernel refuses the operations then the kernel is wrong */
+           the kernel refuses the operations then the kernel is wrong.
+          note that GPFS supports it as well - jmcd */
 
-       kernel_flock(fsp, share_access);
+       ret_flock = SMB_VFS_KERNEL_FLOCK(fsp, fsp->fh->fd, share_access);
+       if(ret_flock == -1 ){
+
+               TALLOC_FREE(lck);
+               fd_close(conn, fsp);
+               file_free(fsp);
+               
+               return NT_STATUS_SHARING_VIOLATION;
+       }
 
        /*
         * At this point onwards, we can guarentee that the share entry
@@ -1717,10 +1743,11 @@ files_struct *open_file_ntcreate(connection_struct *conn,
                 */
                if ((SMB_VFS_FTRUNCATE(fsp,fsp->fh->fd,0) == -1) ||
                    (SMB_VFS_FSTAT(fsp,fsp->fh->fd,psbuf)==-1)) {
-                       unlock_share_entry_fsp(fsp);
+                       status = map_nt_error_from_unix(errno);
+                       TALLOC_FREE(lck);
                        fd_close(conn,fsp);
                        file_free(fsp);
-                       return NULL;
+                       return status;
                }
        }
 
@@ -1730,6 +1757,11 @@ files_struct *open_file_ntcreate(connection_struct *conn,
        fsp->access_mask = access_mask;
 
        if (file_existed) {
+               /* stat opens on existing files don't get oplocks. */
+               if (is_stat_open(open_access_mask)) {
+                       fsp->oplock_type = NO_OPLOCK;
+               }
+
                if (!(flags2 & O_TRUNC)) {
                        info = FILE_WAS_OPENED;
                } else {
@@ -1737,11 +1769,6 @@ files_struct *open_file_ntcreate(connection_struct *conn,
                }
        } else {
                info = FILE_WAS_CREATED;
-               /* Change the owner if required. */
-               if (lp_inherit_owner(SNUM(conn))) {
-                       change_owner_to_parent(conn, fsp, fsp->fsp_name,
-                                              psbuf);
-               }
        }
 
        if (pinfo) {
@@ -1753,56 +1780,46 @@ files_struct *open_file_ntcreate(connection_struct *conn,
         * file structs.
         */
 
-       if(oplock_request && (num_share_modes == 0) && 
-          !IS_VETO_OPLOCK_PATH(conn,fname) &&
-          set_file_oplock(fsp, oplock_request) ) {
-               port = global_oplock_port;
-       } else if (oplock_request && all_current_opens_are_level_II) {
-               port = global_oplock_port;
-               oplock_request = LEVEL_II_OPLOCK;
-               set_file_oplock(fsp, oplock_request);
-       } else {
-               port = 0;
-               oplock_request = 0;
+       if ((fsp->oplock_type != NO_OPLOCK) &&
+           (fsp->oplock_type != FAKE_LEVEL_II_OPLOCK)) {
+               if (!set_file_oplock(fsp, fsp->oplock_type)) {
+                       /* Could not get the kernel oplock */
+                       fsp->oplock_type = NO_OPLOCK;
+               }
        }
 
-       set_share_mode(fsp, port, oplock_request);
-
-       if (create_options & FILE_DELETE_ON_CLOSE) {
-               uint32 dosattr= existing_dos_attributes;
-               NTSTATUS result;
+       if (info == FILE_WAS_OVERWRITTEN || info == FILE_WAS_CREATED || info == FILE_WAS_SUPERSEDED) {
+               new_file_created = True;
+       }
 
-               if (info == FILE_WAS_OVERWRITTEN || info == FILE_WAS_CREATED ||
-                               info == FILE_WAS_SUPERSEDED) {
-                       dosattr = new_dos_attributes;
-               }
+       set_share_mode(lck, fsp, current_user.ut.uid, 0, fsp->oplock_type, new_file_created);
 
-               result = can_set_delete_on_close(fsp, True, dosattr);
+       /* Handle strange delete on close create semantics. */
+       if ((create_options & FILE_DELETE_ON_CLOSE) && can_set_initial_delete_on_close(lck)) {
+               status = can_set_delete_on_close(fsp, True, new_dos_attributes);
 
-               if (!NT_STATUS_IS_OK(result)) {
-                       uint8 u_e_c;
-                       uint32 u_e_code;
-                       BOOL dummy_del_on_close;
+               if (!NT_STATUS_IS_OK(status)) {
                        /* Remember to delete the mode we just added. */
-                       del_share_mode(fsp, NULL, &dummy_del_on_close);
-                       unlock_share_entry_fsp(fsp);
+                       del_share_mode(lck, fsp);
+                       TALLOC_FREE(lck);
                        fd_close(conn,fsp);
                        file_free(fsp);
-                       ntstatus_to_dos(result, &u_e_c, &u_e_code);
-                       set_saved_error_triple(u_e_c, u_e_code, result);
-                       return NULL;
+                       return status;
                }
-               set_delete_on_close(fsp, True);
+               /* Note that here we set the *inital* delete on close flag,
+                  not the regular one. The magic gets handled in close. */
+               fsp->initial_delete_on_close = True;
        }
        
-       if (info == FILE_WAS_OVERWRITTEN || info == FILE_WAS_CREATED ||
-                               info == FILE_WAS_SUPERSEDED) {
+       if (new_file_created) {
                /* Files should be initially set as archive */
                if (lp_map_archive(SNUM(conn)) ||
                    lp_store_dos_attributes(SNUM(conn))) {
-                       file_set_dosmode(conn, fname,
+                       if (!posix_open) {
+                               file_set_dosmode(conn, fname,
                                         new_dos_attributes | aARCH, NULL,
-                                        True);
+                                        parent_dir);
+                       }
                }
        }
 
@@ -1811,13 +1828,13 @@ files_struct *open_file_ntcreate(connection_struct *conn,
         * selected.
         */
 
-       if (!file_existed && !def_acl) {
+       if (!posix_open && !file_existed && !def_acl) {
 
                int saved_errno = errno; /* We might get ENOSYS in the next
                                          * call.. */
 
-               if (SMB_VFS_FCHMOD_ACL(fsp, fsp->fh->fd, unx_mode) == -1
-                   && errno == ENOSYS) {
+               if (SMB_VFS_FCHMOD_ACL(fsp, fsp->fh->fd, unx_mode) == -1 &&
+                   errno == ENOSYS) {
                        errno = saved_errno; /* Ignore ENOSYS */
                }
 
@@ -1836,52 +1853,54 @@ files_struct *open_file_ntcreate(connection_struct *conn,
                        if (ret == -1 && errno == ENOSYS) {
                                errno = saved_errno; /* Ignore ENOSYS */
                        } else {
-                               DEBUG(5, ("open_file_shared: failed to reset "
+                               DEBUG(5, ("open_file_ntcreate: reset "
                                          "attributes of file %s to 0%o\n",
-                                       fname, (unsigned int)new_unx_mode));
+                                         fname, (unsigned int)new_unx_mode));
                                ret = 0; /* Don't do the fchmod below. */
                        }
                }
 
                if ((ret == -1) &&
                    (SMB_VFS_FCHMOD(fsp, fsp->fh->fd, new_unx_mode) == -1))
-                       DEBUG(5, ("open_file_shared: failed to reset "
+                       DEBUG(5, ("open_file_ntcreate: failed to reset "
                                  "attributes of file %s to 0%o\n",
-                               fname, (unsigned int)new_unx_mode));
+                                 fname, (unsigned int)new_unx_mode));
        }
 
        /* If this is a successful open, we must remove any deferred open
         * records. */
-       delete_defered_open_entry_record(conn, fsp->dev, fsp->inode);
-       unlock_share_entry_fsp(fsp);
+       del_deferred_open_entry(lck, mid);
+       TALLOC_FREE(lck);
 
        conn->num_files_open++;
 
-       return fsp;
+       *result = fsp;
+       return NT_STATUS_OK;
 }
 
 /****************************************************************************
  Open a file for for write to ensure that we can fchmod it.
 ****************************************************************************/
 
-files_struct *open_file_fchmod(connection_struct *conn, const char *fname,
-                              SMB_STRUCT_STAT *psbuf)
+NTSTATUS open_file_fchmod(connection_struct *conn, const char *fname,
+                         SMB_STRUCT_STAT *psbuf, files_struct **result)
 {
        files_struct *fsp = NULL;
-       BOOL fsp_open;
+       NTSTATUS status;
 
        if (!VALID_STAT(*psbuf)) {
-               return NULL;
+               return NT_STATUS_INVALID_PARAMETER;
        }
 
-       fsp = file_new(conn);
-       if(!fsp) {
-               return NULL;
+       status = file_new(conn, &fsp);
+       if(!NT_STATUS_IS_OK(status)) {
+               return status;
        }
 
        /* note! we must use a non-zero desired access or we don't get
            a real file descriptor. Oh what a twisted web we weave. */
-       fsp_open = open_file(fsp,conn,fname,psbuf,O_WRONLY,0,FILE_WRITE_DATA);
+       status = open_file(fsp, conn, NULL, NULL, fname, psbuf, O_WRONLY, 0,
+                          FILE_WRITE_DATA, FILE_WRITE_DATA);
 
        /* 
         * This is not a user visible file open.
@@ -1889,107 +1908,197 @@ files_struct *open_file_fchmod(connection_struct *conn, const char *fname,
         * the conn->num_files_open.
         */
 
-       if (!fsp_open) {
+       if (!NT_STATUS_IS_OK(status)) {
                file_free(fsp);
-               return NULL;
+               return status;
        }
 
-       return fsp;
+       *result = fsp;
+       return NT_STATUS_OK;
 }
 
 /****************************************************************************
  Close the fchmod file fd - ensure no locks are lost.
 ****************************************************************************/
 
-int close_file_fchmod(files_struct *fsp)
+NTSTATUS close_file_fchmod(files_struct *fsp)
 {
-       int ret = fd_close(fsp->conn, fsp);
+       NTSTATUS status = fd_close(fsp->conn, fsp);
        file_free(fsp);
-       return ret;
+       return status;
+}
+
+static NTSTATUS mkdir_internal(connection_struct *conn,
+                               const char *name,
+                               uint32 file_attributes,
+                               SMB_STRUCT_STAT *psbuf)
+{
+       mode_t mode;
+       char *parent_dir;
+       const char *dirname;
+       NTSTATUS status;
+
+       if(!CAN_WRITE(conn)) {
+               DEBUG(5,("mkdir_internal: failing create on read-only share "
+                        "%s\n", lp_servicename(SNUM(conn))));
+               return NT_STATUS_ACCESS_DENIED;
+       }
+
+       status = check_name(conn, name);
+       if (!NT_STATUS_IS_OK(status)) {
+               return status;
+       }
+
+       if (!parent_dirname_talloc(tmp_talloc_ctx(), name, &parent_dir,
+                                  &dirname)) {
+               return NT_STATUS_NO_MEMORY;
+       }
+
+       if (file_attributes & FILE_FLAG_POSIX_SEMANTICS) {
+               mode = (mode_t)(file_attributes & ~FILE_FLAG_POSIX_SEMANTICS);
+       } else {
+               mode = unix_mode(conn, aDIR, name, parent_dir);
+       }
+
+       if (SMB_VFS_MKDIR(conn, name, mode) != 0) {
+               return map_nt_error_from_unix(errno);
+       }
+
+       /* Ensure we're checking for a symlink here.... */
+       /* We don't want to get caught by a symlink racer. */
+
+       if (SMB_VFS_LSTAT(conn, name, psbuf) == -1) {
+               DEBUG(2, ("Could not stat directory '%s' just created: %s\n",
+                         name, strerror(errno)));
+               return map_nt_error_from_unix(errno);
+       }
+
+       if (!S_ISDIR(psbuf->st_mode)) {
+               DEBUG(0, ("Directory just '%s' created is not a directory\n",
+                         name));
+               return NT_STATUS_ACCESS_DENIED;
+       }
+
+       if (lp_inherit_perms(SNUM(conn))) {
+               inherit_access_acl(conn, parent_dir, name, mode);
+       }
+
+       if (!(file_attributes & FILE_FLAG_POSIX_SEMANTICS)) {
+               /*
+                * Check if high bits should have been set,
+                * then (if bits are missing): add them.
+                * Consider bits automagically set by UNIX, i.e. SGID bit from parent
+                * dir.
+                */
+               if (mode & ~(S_IRWXU|S_IRWXG|S_IRWXO) && (mode & ~psbuf->st_mode)) {
+                       SMB_VFS_CHMOD(conn, name,
+                                     psbuf->st_mode | (mode & ~psbuf->st_mode));
+               }
+       }
+
+       /* Change the owner if required. */
+       if (lp_inherit_owner(SNUM(conn))) {
+               change_dir_owner_to_parent(conn, parent_dir, name, psbuf);
+       }
+
+       notify_fname(conn, NOTIFY_ACTION_ADDED, FILE_NOTIFY_CHANGE_DIR_NAME,
+                    name);
+
+       return NT_STATUS_OK;
 }
 
 /****************************************************************************
  Open a directory from an NT SMB call.
 ****************************************************************************/
 
-files_struct *open_directory(connection_struct *conn,
-                               const char *fname,
-                               SMB_STRUCT_STAT *psbuf,
-                               uint32 access_mask,
-                               uint32 share_access,
-                               uint32 create_disposition,
-                               uint32 create_options,
-                               int *pinfo)
+NTSTATUS open_directory(connection_struct *conn,
+                       const char *fname,
+                       SMB_STRUCT_STAT *psbuf,
+                       uint32 access_mask,
+                       uint32 share_access,
+                       uint32 create_disposition,
+                       uint32 create_options,
+                       uint32 file_attributes,
+                       int *pinfo,
+                       files_struct **result)
 {
        files_struct *fsp = NULL;
        BOOL dir_existed = VALID_STAT(*psbuf) ? True : False;
-       BOOL create_dir = False;
+       struct share_mode_lock *lck = NULL;
+       NTSTATUS status;
        int info = 0;
 
        DEBUG(5,("open_directory: opening directory %s, access_mask = 0x%x, "
                 "share_access = 0x%x create_options = 0x%x, "
-                "create_disposition = 0x%x\n",
+                "create_disposition = 0x%x, file_attributes = 0x%x\n",
                 fname,
                 (unsigned int)access_mask,
                 (unsigned int)share_access,
                 (unsigned int)create_options,
-                (unsigned int)create_disposition));
+                (unsigned int)create_disposition,
+                (unsigned int)file_attributes));
 
        if (is_ntfs_stream_name(fname)) {
                DEBUG(0,("open_directory: %s is a stream name!\n", fname ));
-               /* NB. Is the DOS error ERRbadpath or ERRbaddirectory ? */
-               set_saved_error_triple(ERRDOS, ERRbadpath,
-                                      NT_STATUS_NOT_A_DIRECTORY);
-               return NULL;
-       }
-
-       if (dir_existed && !S_ISDIR(psbuf->st_mode)) {
-               DEBUG(0,("open_directory: %s is not a directory !\n", fname ));
-               /* NB. Is the DOS error ERRbadpath or ERRbaddirectory ? */
-               set_saved_error_triple(ERRDOS, ERRbadpath,
-                                      NT_STATUS_NOT_A_DIRECTORY);
-               return NULL;
+               return NT_STATUS_NOT_A_DIRECTORY;
        }
 
        switch( create_disposition ) {
                case FILE_OPEN:
-                       /* If directory exists open. If directory doesn't
-                        * exist error. */
-                       if (!dir_existed) {
-                               DEBUG(5,("open_directory: FILE_OPEN requested "
-                                        "for directory %s and it doesn't "
-                                        "exist.\n", fname ));
-                               set_saved_error_triple(ERRDOS, ERRbadfile,
-                                                      NT_STATUS_OBJECT_NAME_NOT_FOUND);
-                               return NULL;
-                       }
+
                        info = FILE_WAS_OPENED;
+
+                       /*
+                        * We want to follow symlinks here.
+                        */
+
+                       if (SMB_VFS_STAT(conn, fname, psbuf) != 0) {
+                               return map_nt_error_from_unix(errno);
+                       }
+                               
                        break;
 
                case FILE_CREATE:
+
                        /* If directory exists error. If directory doesn't
                         * exist create. */
-                       if (dir_existed) {
-                               DEBUG(5,("open_directory: FILE_CREATE "
-                                        "requested for directory %s and it "
-                                        "already exists.\n", fname ));
-                               set_saved_error_triple(ERRDOS, ERRfilexists,
-                                                      NT_STATUS_OBJECT_NAME_COLLISION);
-                               return NULL;
+
+                       status = mkdir_internal(conn,
+                                               fname,
+                                               file_attributes,
+                                               psbuf);
+
+                       if (!NT_STATUS_IS_OK(status)) {
+                               DEBUG(2, ("open_directory: unable to create "
+                                         "%s. Error was %s\n", fname,
+                                         nt_errstr(status)));
+                               return status;
                        }
-                       create_dir = True;
+
                        info = FILE_WAS_CREATED;
                        break;
 
                case FILE_OPEN_IF:
-                       /* If directory exists open. If directory doesn't
-                        * exist create. */
-                       if (!dir_existed) {
-                               create_dir = True;
+                       /*
+                        * If directory exists open. If directory doesn't
+                        * exist create.
+                        */
+
+                       status = mkdir_internal(conn,
+                                               fname,
+                                               file_attributes,
+                                               psbuf);
+
+                       if (NT_STATUS_IS_OK(status)) {
                                info = FILE_WAS_CREATED;
-                       } else {
+                       }
+
+                       if (NT_STATUS_EQUAL(status,
+                                           NT_STATUS_OBJECT_NAME_COLLISION)) {
                                info = FILE_WAS_OPENED;
+                               status = NT_STATUS_OK;
                        }
+                               
                        break;
 
                case FILE_SUPERSEDE:
@@ -1999,47 +2108,18 @@ files_struct *open_directory(connection_struct *conn,
                        DEBUG(5,("open_directory: invalid create_disposition "
                                 "0x%x for directory %s\n",
                                 (unsigned int)create_disposition, fname));
-                       file_free(fsp);
-                       set_saved_error_triple(ERRDOS, ERRinvalidparam,
-                                              NT_STATUS_INVALID_PARAMETER);
-                       return NULL;
+                       return NT_STATUS_INVALID_PARAMETER;
        }
 
-       if (create_dir) {
-               /*
-                * Try and create the directory.
-                */
-
-               /* We know bad_path is false as it's caught earlier. */
-
-               NTSTATUS status = mkdir_internal(conn, fname, False);
-
-               if (!NT_STATUS_IS_OK(status)) {
-                       DEBUG(2,("open_directory: unable to create %s. "
-                                "Error was %s\n", fname, strerror(errno) ));
-                       /* Ensure we return the correct NT status to the
-                        * client. */
-                       set_saved_error_triple(0, 0, status);
-                       return NULL;
-               }
-
-               /* Ensure we're checking for a symlink here.... */
-               /* We don't want to get caught by a symlink racer. */
-
-               if(SMB_VFS_LSTAT(conn,fname, psbuf) != 0) {
-                       return NULL;
-               }
-
-               if(!S_ISDIR(psbuf->st_mode)) {
-                       DEBUG(0,("open_directory: %s is not a directory !\n",
-                                fname ));
-                       return NULL;
-               }
+       if(!S_ISDIR(psbuf->st_mode)) {
+               DEBUG(5,("open_directory: %s is not a directory !\n",
+                        fname ));
+               return NT_STATUS_NOT_A_DIRECTORY;
        }
 
-       fsp = file_new(conn);
-       if(!fsp) {
-               return NULL;
+       status = file_new(conn, &fsp);
+       if(!NT_STATUS_IS_OK(status)) {
+               return status;
        }
 
        /*
@@ -2051,7 +2131,7 @@ files_struct *open_directory(connection_struct *conn,
        fsp->dev = psbuf->st_dev;
        fsp->vuid = current_user.vuid;
        fsp->file_pid = global_smbpid;
-       fsp->can_lock = True;
+       fsp->can_lock = False;
        fsp->can_read = False;
        fsp->can_write = False;
 
@@ -2065,49 +2145,108 @@ files_struct *open_directory(connection_struct *conn,
        fsp->sent_oplock_break = NO_BREAK_SENT;
        fsp->is_directory = True;
        fsp->is_stat = False;
+       fsp->posix_open = (file_attributes & FILE_FLAG_POSIX_SEMANTICS) ? True : False;
+
        string_set(&fsp->fsp_name,fname);
 
+       lck = get_share_mode_lock(NULL, fsp->dev, fsp->inode,
+                                 conn->connectpath,
+                                 fname);
+
+       if (lck == NULL) {
+               DEBUG(0, ("open_directory: Could not get share mode lock for %s\n", fname));
+               file_free(fsp);
+               return NT_STATUS_SHARING_VIOLATION;
+       }
+
+       status = open_mode_check(conn, fname, lck,
+                               access_mask, share_access,
+                               create_options, &dir_existed);
+
+       if (!NT_STATUS_IS_OK(status)) {
+               TALLOC_FREE(lck);
+               file_free(fsp);
+               return status;
+       }
+
+       set_share_mode(lck, fsp, current_user.ut.uid, 0, NO_OPLOCK, True);
+
+       /* For directories the delete on close bit at open time seems
+          always to be honored on close... See test 19 in Samba4 BASE-DELETE. */
        if (create_options & FILE_DELETE_ON_CLOSE) {
-               NTSTATUS status = can_set_delete_on_close(fsp, True, 0);
-               if (!NT_STATUS_IS_OK(status)) {
+               status = can_set_delete_on_close(fsp, True, 0);
+               if (!NT_STATUS_IS_OK(status) && !NT_STATUS_EQUAL(status, NT_STATUS_DIRECTORY_NOT_EMPTY)) {
+                       TALLOC_FREE(lck);
                        file_free(fsp);
-                       return NULL;
+                       return status;
                }
-       }
 
-       /* Change the owner if required. */
-       if ((info == FILE_WAS_CREATED) && lp_inherit_owner(SNUM(conn))) {
-               change_owner_to_parent(conn, fsp, fsp->fsp_name, psbuf);
+               if (NT_STATUS_IS_OK(status)) {
+                       /* Note that here we set the *inital* delete on close flag,
+                          not the regular one. The magic gets handled in close. */
+                       fsp->initial_delete_on_close = True;
+               }
        }
 
+       TALLOC_FREE(lck);
+
        if (pinfo) {
                *pinfo = info;
        }
 
        conn->num_files_open++;
 
-       return fsp;
+       *result = fsp;
+       return NT_STATUS_OK;
+}
+
+NTSTATUS create_directory(connection_struct *conn, const char *directory)
+{
+       NTSTATUS status;
+       SMB_STRUCT_STAT sbuf;
+       files_struct *fsp;
+
+       SET_STAT_INVALID(sbuf);
+       
+       status = open_directory(conn, directory, &sbuf,
+                               FILE_READ_ATTRIBUTES, /* Just a stat open */
+                               FILE_SHARE_NONE, /* Ignored for stat opens */
+                               FILE_CREATE,
+                               0,
+                               FILE_ATTRIBUTE_DIRECTORY,
+                               NULL,
+                               &fsp);
+
+       if (NT_STATUS_IS_OK(status)) {
+               close_file(fsp, NORMAL_CLOSE);
+       }
+
+       return status;
 }
 
 /****************************************************************************
  Open a pseudo-file (no locking checks - a 'stat' open).
 ****************************************************************************/
 
-files_struct *open_file_stat(connection_struct *conn, char *fname,
-                            SMB_STRUCT_STAT *psbuf)
+NTSTATUS open_file_stat(connection_struct *conn, const char *fname,
+                       SMB_STRUCT_STAT *psbuf, files_struct **result)
 {
        files_struct *fsp = NULL;
+       NTSTATUS status;
 
-       if (!VALID_STAT(*psbuf))
-               return NULL;
+       if (!VALID_STAT(*psbuf)) {
+               return NT_STATUS_INVALID_PARAMETER;
+       }
 
        /* Can't 'stat' open directories. */
-       if(S_ISDIR(psbuf->st_mode))
-               return NULL;
+       if(S_ISDIR(psbuf->st_mode)) {
+               return NT_STATUS_FILE_IS_A_DIRECTORY;
+       }
 
-       fsp = file_new(conn);
-       if(!fsp)
-               return NULL;
+       status = file_new(conn, &fsp);
+       if(!NT_STATUS_IS_OK(status)) {
+               return status;
+       }
 
        DEBUG(5,("open_file_stat: 'opening' file %s\n", fname));
 
@@ -2133,5 +2272,64 @@ files_struct *open_file_stat(connection_struct *conn, char *fname,
 
        conn->num_files_open++;
 
-       return fsp;
+       *result = fsp;
+       return NT_STATUS_OK;
+}
+
+/****************************************************************************
+ Receive notification that one of our open files has been renamed by another
+ smbd process.
+****************************************************************************/
+
+void msg_file_was_renamed(struct messaging_context *msg,
+                         void *private_data,
+                         uint32_t msg_type,
+                         struct server_id server_id,
+                         DATA_BLOB *data)
+{
+       files_struct *fsp;
+       char *frm = (char *)data->data;
+       SMB_DEV_T dev;
+       SMB_INO_T inode;
+       const char *sharepath;
+       const char *newname;
+       size_t sp_len;
+
+       if (data->data == NULL
+           || data->length < MSG_FILE_RENAMED_MIN_SIZE + 2) {
+                DEBUG(0, ("msg_file_was_renamed: Got invalid msg len %d\n",
+                         data->length));
+                return;
+        }
+
+       /* Unpack the message. */
+       dev = DEV_T_VAL(frm,0);
+       inode = INO_T_VAL(frm,8);
+       sharepath = &frm[16];
+       newname = sharepath + strlen(sharepath) + 1;
+       sp_len = strlen(sharepath);
+
+       DEBUG(10,("msg_file_was_renamed: Got rename message for sharepath %s, new name %s, "
+               "dev %x, inode  %.0f\n",
+               sharepath, newname, (unsigned int)dev, (double)inode ));
+
+       for(fsp = file_find_di_first(dev, inode); fsp; fsp = file_find_di_next(fsp)) {
+               if (memcmp(fsp->conn->connectpath, sharepath, sp_len) == 0) {
+                       DEBUG(10,("msg_file_was_renamed: renaming file fnum %d from %s -> %s\n",
+                               fsp->fnum, fsp->fsp_name, newname ));
+                       string_set(&fsp->fsp_name, newname);
+               } else {
+                       /* TODO. JRA. */
+                       /* Now we have the complete path we can work out if this is
+                          actually within this share and adjust newname accordingly. */
+                       DEBUG(10,("msg_file_was_renamed: share mismatch (sharepath %s "
+                               "not sharepath %s) "
+                               "fnum %d from %s -> %s\n",
+                               fsp->conn->connectpath,
+                               sharepath,
+                               fsp->fnum,
+                               fsp->fsp_name,
+                               newname ));
+               }
+        }
 }