Hello,
This is an attempt to implement 13.5 Export session ids <https://curl.se/docs/todo.html#Export_session_ids> feature from the TODO list. It only works for OpenSSL, but adding support for other back-ends should be rather straightforward.
See PR#7467 <https://github.com/curl/curl/pull/7467>. How it is supposed to work: * Exporting: when a new SSL session was established, a user callback will be executed with the serialized info of the new session. The relevant new symbols here are: o CURLOPT_SSL_SESS_FUNCTION o CURLOPT_SSL_SESS_DATA o curl_ssl_sess_callback (prototype) * Importing: when SSL session cache was initialized, a user callback will be executed to allow user code to pass serialized session info blobs to curl. The relevant new symbols here are: o CURLOPT_SSL_CACHE_FUNCTION o CURLOPT_SSL_CACHE_DATA o curl_ssl_cache_callback (prototype) Things that probably need improving: * The patch is pretty big. If it is difficult to review, I guess it can be split in at least 2 parts: export, import. It got like this due to too many change attempts that could not really be isolated... * Naming: it might have been a better idea to use "export"/"import" rather than "dump"/"load", but this should be fairly quick to change. The same goes for any other names, so please let me know about any other terminology preference. * Some implementation decisions might not have been the right ones: o Dumping (export) and loading (import) work with the minimum amount of data: { back-end id, scheme, hostname, port, serialized blob, blob size }. Why: this seemed to be enough for the use cases here, also when using the same feature in other libs. Concerns: a sessionid loaded to cache (thus with some missing details) will have them filled in from the first connection that is trying to use it. Could that ever go wrong? o Memory handling, when loading sessionid-s: user code is supposed to allocate memory for the structures (and their members) for all instances that are too be loaded. Then curl code will release that memory after execution has returned from the callback. For this it required the proper release function, which needed to be set from the user code in the same callback. When not setting it, it will fallback to Curl_cfree(). Is this ok, or should this case always require the use of curl_global_init_mem(), or is there a simpler and better way that I've missed? * Documentation: I tried adding relevant comments to code, but if there is any other place where they need to be added please let me know. * Test Cases: I don't really know how a test case for this would work. Tests here have been based on breakpoints and log messages from curl. Regards, C
------------------------------------------------------------------- Unsubscribe: https://cool.haxx.se/list/listinfo/curl-library Etiquette: https://curl.se/mail/etiquette.html
