Stanislav Fomichev wrote:
I was gonna do it, but looks like you volunteered yourself? We do
get this almost every day now:
https://lore.kernel.org/netdev/20260918150858.3c98728d@kernel.org/ (local)
Thanks for the pointer - I see the same flag patch was posted just
a few days ago. Sorry for adding to the pile, I should have
searched the list harder before sending.
That leaves the accounting side as my one remaining question. With
bind-tx unprivileged on purpose, is there anything that stops a
user from exhausting memory and IOMMU mappings by binding over and
over? As far as I can see there's no per-socket or per-user limit,
the allocations don't go to memcg (GFP_KERNEL, devmem.c:214), and
repeated binds of the same dmabuf fd aren't deduplicated - but I
may well be missing a guard somewhere.
If there isn't one, is that a gap you'd want a patch for, or is it
considered acceptable in practice? Happy to help either way, just
tell me which direction you'd prefer.
Thanks,
Dairui Zhang