This change intends to run the whole network stack in softint context
(or normal LWP), not hardware interrupt context. Note that the work is
still incomplete by this change; to that end, we also have to softint-ify
if_link_state_change (and bpf) which can still run in hardware interrupt.
This change softint-ifies at ifp->if_input that is called from
each device driver (and ieee80211_input) to ensure Layer 2 runs
in softint (e.g., ether_input and bridge_input). To this end,
we provide a framework (called percpuq) that utlizes softint(9)
and percpu ifqueues. With this patch, rxintr of most drivers just
queues received packets and schedules a softint, and the softint
dequeues packets and does rest packet processing.
To minimize changes to each driver, percpuq is allocated in struct
ifnet for now and that is initialized by default (in if_attach).
We probably have to move percpuq to softc of each driver, but it's
future work. At this point, only wm(4) has percpuq in its softc
as a reference implementation.
Additional information including performance numbers can be found
in the thread at tech-kern@ and tech-net@:
http://mail-index.netbsd.org/tech-kern/2016/01/14/msg019997.html
Acknowledgment: riastradh@ greatly helped this work.
Thank you very much!
designated initializers.
I have not built every extant kernel so I have probably broken at
least one build; however I've also found and fixed some wrong
cdevsw/bdevsw entries so even if so I think we come out ahead.
mode. Some return EINVAL when they are dying, but others like USB return EIO.
Downgrade to a DIAGNOSTIC printf. Same should be done for the malloc/NOWAIT,
but this is rarely hit.
which add a capability to call external functions in a predetermined way.
It can be thought as a BPF "coprocessor" -- a generic mechanism to offload
more complex packet inspection operations. There is no default coprocessor
and this functionality is not targeted to the /dev/bpf. This is primarily
targeted to the kernel subsystems, therefore there is no way to set a custom
coprocessor at the userlevel.
Discussed on: tech-net@
OK: core@
- When handling contiguous buffer in _bpf_tap(), pass its real size
rather than 0 to avoid reading packet data as mbuf struct on
out-of-bounds loads.
- Correctly pass pktlen and buflen arguments from bpf_deliver() to
bpf_filter() to avoid reading mbuf struct as packet data.
JIT case is still broken.
Also, test pointers againts NULL.
returned to userland by read(2) also needs to be converted.
For this, the bpf descriptor is flagged as compat32 (or not) in the
open and ioctl functions (where the user process's pid is also updated
in the descriptor). When the bpf buffer is filled in, the 32bits or native
header is used depending on the information stored in the descriptor.
This won't work if a 64bit binary does the open and ioctls, and then
exec a 32bit program which will do the read. But this is very
unlikely to happen in real life ...
Tested on i386 and loongson; with these changes my loongson can run
dhclient and tcpdump with a n32 userland.
M_NOWAIT cause dhcpd on a low-memory server with lots of interfaces to
occasionally fail to start with ENOBUFS; (M_WAITOK | M_CANFAIL) seems to
fix this.
Tested on 3 different dhcp servers.
#if NBPFILTER is no longer required in the client. This change
doesn't yet add support for loading bpf as a module, since drivers
can register before bpf is attached. However, callers of bpf can
now be modularized.
Dynamically loadable bpf could probably be done fairly easily with
coordination from the stub driver and the real driver by registering
attachments in the stub before the real driver is loaded and doing
a handoff. ... and I'm not going to ponder the depths of unload
here.
Tested with i386/MONOLITHIC, modified MONOLITHIC without bpf and rump.
read/write/accept, then the expectation is that the blocked thread will
exit and the close complete.
Since only one fd is affected, but many fd can refer to the same file,
the close code can only request the fs code unblock with ERESTART.
Fixed for pipes and sockets, ERESTART will only be generated after such
a close - so there should be no change for other programs.
Also rename fo_abort() to fo_restart() (this used to be fo_drain()).
Fixes PR/26567
do drain' in many places, whereas fo_drain() was called in order to force
blocking read()/write() etc calls to return to userspace so that a close()
call from a different thread can complete.
In the sockets code comment out the broken code in the inner function,
it was being called from compat code.
