A good point actually - from a legal perspective this is tricky. Generally the publisher claims copyright over the arrangement that is "published". Photocopying the printed version would therefore be problematic.
With digital formats, the general approach is to export a PDF from LaTeX (or whatever is used to write the paper) as an author pre-print copy (i.e. not in the final form used by the journal), and publish that one. From a technical legal perspective you would likely need to look at what agreement you signed with the publisher at the time (if any), since that may restrict distribution of the text.
This raises a good point though for works that were pre-open access, and the risk of them "rotting" behind paywalls out of being in a state of legal limbo. Newer science will be more accessible due to open access rules, but older science may end up effectively "lost" to those without access to these paywalled repositories.
There's no real incentive for those publishers to open things up, as they rely on institutions paying large subscriptions to preserve access to the legacy archived content, and that keeps their many thousands of staff employed.