On Wed, May 04, 2022 at 02:21:11PM +0200, Alexandr Nedvedicky wrote: > I'm not sure flipping a flag is a right change. In general we don't want > to hold NET_LOCK()/PF_LOCK() while waiting for memory.
- We must not wait for memory when in the packet processing hot path. Drop the packet instead. - We should not wait for memory when holding pf lock. Then we can replace rw lock with a mutex or something else later. - Waiting for memory with net lock is unavoidable. We have to wait when coming from system call. Doing preallocation may be possible in some cases but code may get too complicated elsewhere. If getting the allocation out of the locks is doable here, it could be the best solution. bluhm
