[email protected] writes: > I just installed a NetBSD 9.0 and transmission-daemon on an amd64 > computer. When I start the daemon, it gives the following warnings: > > UDP Failed to set receive buffer: requested 4194304, got ... > UDP Failed to set send buffer: requested 1048576, got ... > > so I set > > net.inet.udp.recvspace=419430 > net.inet.udp.sendspace=1048576 > > in the /etc/sysctl.conf. But when I reboot, my whole networking is > dead, even ifconfig gives: > > ifconfig: status: getsock: No buffer space available
Those are large (compared to how memory used to be), and I suspect that setting them overall, times the number of UDP sockets, exceeds the number of mbuf clusters. So two thoughts: increase nmbclusters I would expect it's ok for a program to set a high limit on one socket, more than the systemwide value. To work on the second approach, I'd use ktrace to see the failing syscall. And then find the code that handles the ioctl. I looked in src/sys/netinet/udp_usrreq.c and am beginning to think per-udp socket buffer sizes aren't supported, which seems surprising.
