Microdescriptor cache consistency looks error-prone

On IRC, wanoskarnet notes that the microdesc_free() function requires that the microdescriptor not be held by any node_t , yet does not check it. He says this is a bug waiting to happen. I agree.

changed milestone to %Tor: 0.2.3.x-final

added component::core tor/tor milestone::Tor: 0.2.3.x-final priority::medium resolution::fixed status::closed tor-client type::defect labels

I'm adding some flags to track microdesc addition and deletion from the nodelist, and also from the microdesc_map. There are two options for how microdesc_free() should handle these flags: It can either assert that everything freed has been removed, or it can remove stuff for us itself, and log that there was a bug.

I'm going with the second approach for now to try to make 0.2.3.x be less crashy. If these warnings show up, though, we will want to track them down.

Please review branch bug3153 in my public repository.

Trac:
Status: new to needs_review

Wanoskarnet says on IRC that nothing actually calls nodelist_remove_microdesc, and nothing actually clears node->md during normal operation. Better look into that.

I think wanoskarnet is right here: right now, so long as a microdesc_t is held in a node_t, the microdesc should never actually get freed. That's fine. I'm okay with leaving that function in and unused for now.

Merging the sanity-test code above.

Trac:
Status: needs_review to closed
Resolution: N/A to fixed

Good thing we chose "fix and warn" rather than assert -- #6429 (moved) appears to be such a warning.

Trac:
Keywords: N/A deleted, tor-client added

Trac:
Component: Tor Client to Tor

closed

mentioned in issue #6429 (moved)

mentioned in issue #7164 (moved)

moved to tpo/core/tor#3153 (closed)