openbsd changes of note 3
Things happened. In November.
Time marches on. So do timezones.
vmd can now boot kernels from inside the disk image. And it also has boot.conf support.
Starting up a privsep daemon with lots of children requires creating lots of pipes. Too many children means too many files and failure trying to start. Be more sparing in fd use in relayd.
512M should be enough for anyone. But some people need 768M.
sshd preallocates more memory for keys. The idea here is that even if you follow best practice and zero sensitive memory after use, additional copies may have been made when the buffer was resized by realloc. By taking back responsibility from libc, sshd can ensure that only one copy of the key is live in memory at any given time.
Like I said, 512M should be enough for anyone. vmd users included.
Build kernel with DEBUG=-g by default. Including type information in the kernel opens the possibility to doing fun introspective style things at runtime in the future.
Once a daemon, always a daemon. No need for sshd to do it twice.
New sshd option to disable all the forwarding. As well as anything else we might implement in the future. Because nothing is ever done.
Add a new implementation of MiRA, a rate scaling algorithm for 802.11n. And of course, stsp’s tech announcement. 20MHz should be enough for anyone.
And looking ahead...
TSC timecounters make a comeback. Using rdtsc for the clock is an old idea, but shelved because it wasn’t always reliable and not really necessary. Now it seems to be necessary. Modern machines usually use the acpihpet (ACPI High Precision Event Timer) driver, but this requires evaluating AML which really means doing just about anything. Presumably on previous generations of machines this would be something simple like reading a PCI register or two, but now it goes off an runs SMI or whatever? Who even knows? And since gettimeofday remains in the hot path of a surprising number of applications, it’s important to have a fast clock, which the newest gen of acpihpet is not. Anyway, but since the trouble with drifting TSC counters has been resolved on modern CPUs, we may have a new winner for best clock.
syspatch is now a thing you can test. Upgrade early and often for best results.