search

OpenAoE / vblade

The Minimal Userland AoE Target
GNU General Public License v2.0
51 stars 16 forks source link

FreeBSD write to network: Input/output error with jumbo frames #7

Open andrewbasterfield opened 8 years ago

andrewbasterfield commented 8 years ago

vblade running on FreeBSD 10.2-RELEASE x86_64, client is Linux. The 2 boxes have jumbo frames enabled and will happily ping at ~9000 bytes

When I set the MTU to 4131 all is fine; I get 3584 byte data frames When I set the MTU to 4132 the frame size jumps to 4096 and vblade exits with write to network: Input/output error

Presumably there is some limitation on FreeBSD with >4096 byte writes to BPF device; I haven't been able to find any reference to this limitation and if it is tweakable. Any suggestions?

ecashin commented 8 years ago

I don't have much experience with AoE on FreeBSD. You might be able to find some information by searching the archives of the aoetools-discuss mailing list, hosted at sourceforge and archived elsewhere, including gmane, I believe.

On 10/10/2015 03:37 PM, Andrew Basterfield wrote:

vblade running on FreeBSD 10.2-RELEASE x86_64, client is Linux. The 2 boxes have jumbo frames enabled and will happily ping at ~9000 bytes

When I set the MTU to 4131 all is fine; I get 3584 byte data frames When I set the MTU to 4132 the frame size jumps to 4096 and vblade exits with write to network: Input/output error

Presumably there is some limitation on FreeBSD with >4096 byte writes to BPF device; I haven't been able to find any reference to this limitation and if it is tweakable. Any suggestions?

— Reply to this email directly or view it on GitHub https://github.com/OpenAoE/vblade/issues/7.

andrewbasterfield commented 8 years ago

Have taken it to FreeBSD mailing lists; will update with any progress

andrewbasterfield commented 8 years ago

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=205164

carton-of-mice commented 6 years ago

Hello. I haven't tracked the project lately. Due to insufficient dog-food-ing, this slipped by. FreeBSD indeed, as the other BSDs, limits BPF writes to one mbuf (have not checked reads, but behaviour probably matches). I'm going to file a PR to cap MTU at page size, but I'm not sure if a full-sized jumbo frame workaround is available.