Package: libapr1t64
Version: 1.7.5-1
Severity: normal
Forwarded: https://bz.apache.org/bugzilla/show_bug.cgi?id=69542

Could you backport the upstream fix for APR bug #69542 to Trixie
and apply it to unstable while it still ships APR 1.7.6?

In support/unix/waitio.c, the poll() path in
apr_wait_for_io_or_timeout() stores a 64-bit timeout in microseconds
in a signed 32-bit int before converting it to milliseconds. A socket
timeout of 2148 seconds exceeds INT32_MAX at that assignment, so
poll() receives an incorrect timeout.

This affects long Apache reverse-proxy timeouts, including AJP
backends. Our AJP configuration uses this workaround:

    ProxyPass / ajp://127.0.0.1:8009/ retry=0 timeout=2147

The expected behaviour is for a configured timeout of 2148 seconds
or more to be honoured. The workaround caps it at 2147 seconds,
the largest whole-second value that fits before conversion.

The affected code is present in Trixie's apr 1.7.5-1 and in
Forky/Sid's apr 1.7.6-3. Bookworm's 1.7.2 divides by 1000 before
assigning to the int and predates the regression.

Upstream has already backported the fix to its 1.7.x branch and
lists it for APR 1.7.7. The backport is commit:
https://github.com/apache/apr/commit/5140fe6815a0a440ac822d85ddae87f9fd3835dd

Thanks.

Reply via email to