talloc_tos in shadow_copy2_insert_string

[Andrew Klaassen]

--- On Tue, 3/27/12, Jeremy Allison <jra at samba.org> wrote:

> This is just straight C programming.
> 
> static void bar_alloc_string(char *barstring)
> {
>     barstring = talloc_asprintf(talloc_tos(),
> "%s/%s", "foo", "bar");
> }
> 
> Overwrites the passed in barstring (which was NULL) with
> the talloc'd value. This is then not returned (although
> kept on the talloc_tos() pointer). It will be kept around
> and only freed when the current talloc_tos() pointer
> frame is freed.

Apologies for beating this horse some more, but it looks like I'm still not understanding parts of talloc's behaviour.  If you can spare a few more cycles I'd very much appreciate it.

In this pair of functions, I'd expect clientFname to stick around until the second function returns, since all talloc operations on it use the talloc_tos() from the second function for context.  But, instead, it gets freed as soon as the first function returns:

static int do_the_clone_of_smb_fname(struct vfs_handle_struct *handle,
                TALLOC_CTX *mem_ctx,
                struct smb_filename *smb_fname,
                struct smb_filename *clientFname)
{
        /* Use mem_ctx from caller. */
        copy_smb_filename(mem_ctx, smb_fname,
                        &clientFname);

        DEBUG(0, ("1 clientFname->base_name '%s'\n",
                 clientFname->base_name));
        /* prints expected basename. */

}

static int clone_smb_fname_stat(vfs_handle_struct *handle,
                struct smb_filename *smb_fname)
{
        struct smb_filename *clientFname = NULL;

        TALLOC_CTX *mem_ctx = talloc_tos();

        do_the_clone_of_smb_fname(handle, mem_ctx,
                        smb_fname, clientFname);

        DEBUG(0, ("2 clientFname '%s'\n", clientFname));
        /* prints (null)
         * Shouldn't clientFname still be allocated and accessible,
         * since all talloc operations on it use the talloc_tos()
         * from this function as the context?  But, instead, we
         * get null...
         */        
}

Thanks again.

Andrew