At 23:29 25.03.2003 +0100, Jelmer Vernooij wrote: I took Jelmer's patch and modifed it,
to support an easy way to have per connection default_vfs_ops and per connection private data in the vfs modules
so every module registers and got a unique vfs module number for te current smbd or vfstest process.
now the module only regiter the vfsops_tuple_struct array and the vfs_init() and vfs_done() fn's are gone.
vfs_init_custom() hang the default ops for each module on the vfs_handle and mark this handle with the unique vfs module number
#define VFS_HANDLE_GET_OPS(conn, ret) \
if ((default_vfs_ops=vfs_handle_get_ops(conn,vfs_module_number))==NULL) { \
DEBUG(0,("%s() failed to get default_vfs_ops!\n",FUNCTION_MACRO)); \
ret; \
}
this macro helps the module to get its default_vfs_ops
struct vfs_ops *vfs_handle_get_ops(connection_struct *conn, int vfs_num)
{
struct smb_vfs_handle_struct *handle = NULL;
if (!conn) {
DEBUG(0,("vfs_handle_get_ops() called with NULL pointer.\n"));
return NULL;
}
handle = conn->vfs_handles;
while (handle) {
if (handle->vfs_num == vfs_num) {
return &handle->vfs_def_ops;
}
handle = handle->next;
}
return NULL;
}
I have converted vfs_recycle.c for this setup:
all options are now resolved in run time (if smb.conf changed and smbd got -HUP)
the modified options are used
What's left todo for this fix up pramateric options a bit....
basicly prevent us from memory leeks when we alloc the lp_params values on each call
maybe we should cache the values and make the return values const
But that will be another patch...
I hope you see how the vfs stuff work.
Note: this patch is only a preview to get feedback it's not tested and there maybe some typos and merging bugs
but here smbd/vfs.o and modules/vfs_recycle.o are build fine (expect some cast warnings in vfs_recycle.o)
metze
-----------------------------------------------------------------------------
Stefan "metze" Metzmacher <[EMAIL PROTECTED]>
