c18n: Duplicate string when adding compartment name

[dpgao]

If the name argument is used, it needs to be duplicated to prevent it from being subsequently freed by others.

[bsdjhb]

Hmm, I do prefer the existing flow a bit, and a comment about why we strdup() is probably warranted. Maybe something like:

    /* Prefer DT_SONAME as the compartment base name if present. */
    if (obj->soname != NULL) {
        name = obj->soname;
    } else if (name != NULL) {
        /* If a name was provided, allocate a duplicate copy. */
        name = strdup(name);
    } else {
        /* Try to find a name if none provided. */
        ... (existing cases)
    }

[bsdjhb]

Might also be good to note in the commit message why supplied names are transient storage and require duplication. obj_main->path is not transient for example, so I guess it's the do_load_object case? If so, another approach might be to add strdup() explicitly in that call instead of doing it here? (I wonder if we could also just pass NULL when registering obj_main to avoid the strdup for that case?)

[dpgao]

I realize my fix is flawed because now I cannot decide whether to call free(obj->name) in obj_free any more.

Note that object_add_name in do_load_object already unconditionally duplicates name, so we can piggyback on that copy and there's then no need to do anything extra in obj_free.

[dpgao]

I also added an error message because this failure mode might not be obvious to the user.

This failure should be rare, but may happen if the user preloads an SONAME-less object using a file descriptor.

[bsdjhb]

I think this is fine. The old code was giving preference to the do_load_object name rather than some other name in &obj->names but it's not clear to me that that was really meaningful.