blaze-core, a library underlying http4s-blaze-server, accepts connections unboundedly on its selector pool. This has the net effect of amplifying degradation in services that are unable to handle their current request load, since incoming connections are still accepted and added to an unbounded queue. Each connection allocates a socket handle, which drains a scarce OS resource. This can also confound higher level circuit breakers which work based on detecting failed connections.
http4s provides a general MaxActiveRequests
middleware mechanism for limiting open connections, but it is enforced inside the Blaze accept loop, after the connection is accepted and the socket opened. Thus, the limit only prevents the number of connections which can be simultaneously processed, not the number of connections which can be held open.
In 0.21.18, 0.22.0-M3, and 1.0.0-M16, a newmaxConnections
property, with a default value of 1024, has been added to the BlazeServerBuilder
. Setting the value to a negative number restores unbounded behavior, but is strongly disrecommended.
The NIO2 backend does not respect maxConnections
. Its use is now deprecated in http4s-0.21, and the option is removed altogether starting in http4s-0.22.
The connections are bounded in 0.21.17, 0.22.0-M2, and 1.0.0-M14, but the maxConnections
parameter was passed incorrectly, making it impossible to change the Blaze default of 512.
See the Blaze GHSA for more on the underlying issue.
If you have any questions or comments about this advisory: * Open an issue in http4s/http4s * Contact us according to the http4s security policy
{ "nvd_published_at": "2021-02-02T22:15:00Z", "github_reviewed_at": "2021-02-02T21:42:25Z", "severity": "HIGH", "github_reviewed": true, "cwe_ids": [ "CWE-400", "CWE-770" ] }