Open Bug 726103 Opened 13 years ago Updated 2 years ago

theoretical deadlock between ocsp and http socket pool limit

Categories

(Core :: Networking: HTTP, defect, P3)

13 Branch
x86_64
Linux
defect

Tracking

()

People

(Reporter: mcmanus, Unassigned)

Details

(Whiteboard: [necko-backlog])

There are both global limits and per-Connection-Info limits on the number of HTTP sockets we can have open. existing connections trying to negotiatie SSL might do http based OCSP "on the stack" as a condition of moving forward. If they can't get a socket, and none of the used sockets can complete because they are in turn blocked on OCSP, we have a deadlock. This isn't a problem for the per-ci limits (6) because SSL and non-SSL (OCSP) are kept in different CI. But I think it is a theoretical problem for the global limit. I think the right thing to do here is to provide an interface ocsp can use to say ignore the connection limits. I thought this was the source of what the reporter in bug 724519 was seeing - but the http log he provided rules it out.
Whiteboard: [necko-backlog]
Priority: -- → P1
Priority: P1 → P3
Severity: normal → S3
You need to log in before you can comment on or make changes to this bug.