| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
Commit 7bd509e311f4 ("bpf: add prog_digest and expose it via
fdinfo/netlink") was recently discussed, partially due to
admittedly suboptimal name of "prog_digest" in combination
with sha1 hash usage, thus inevitably and rightfully concerns
about its security in terms of collision resistance were
raised with regards to use-cases.
The intended use cases are for debugging resp. introspection
only for providing a stable "tag" over the instruction sequence
that both kernel and user space can calculate independently.
It's not usable at all for making a security relevant decision.
So collisions where two different instruction sequences generate
the same tag can happen, but ideally at a rather low rate. The
"tag" will be dumped in hex and is short enough to introspect
in tracepoints or kallsyms output along with other data such
as stack trace, etc. Thus, this patch performs a rename into
prog_tag and truncates the tag to a short output (64 bits) to
make it obvious it's not collision-free.
Should in future a hash or facility be needed with a security
relevant focus, then we can think about requirements, constraints,
etc that would fit to that situation. For now, rework the exposed
parts for the current use cases as long as nothing has been
released yet. Tested on x86_64 and s390x.
Fixes: 7bd509e311f4 ("bpf: add prog_digest and expose it via fdinfo/netlink")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Cc: Andy Lutomirski <luto@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
When loading a BPF program via bpf(2), calculate the digest over
the program's instruction stream and store it in struct bpf_prog's
digest member. This is done at a point in time before any instructions
are rewritten by the verifier. Any unstable map file descriptor
number part of the imm field will be zeroed for the hash.
fdinfo example output for progs:
# cat /proc/1590/fdinfo/5
pos: 0
flags: 02000002
mnt_id: 11
prog_type: 1
prog_jited: 1
prog_digest: b27e8b06da22707513aa97363dfb11c7c3675d28
memlock: 4096
When programs are pinned and retrieved by an ELF loader, the loader
can check the program's digest through fdinfo and compare it against
one that was generated over the ELF file's program section to see
if the program needs to be reloaded. Furthermore, this can also be
exposed through other means such as netlink in case of a tc cls/act
dump (or xdp in future), but also through tracepoints or other
facilities to identify the program. Other than that, the digest can
also serve as a base name for the work in progress kallsyms support
of programs. The digest doesn't depend/select the crypto layer, since
we need to keep dependencies to a minimum. iproute2 will get support
for this facility.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Commit 18cdb37ebf4c ("net: sched: do not use tcf_proto 'tp' argument from
call_rcu") removed the last usage of tp from cls_bpf_delete_prog(), so also
remove it from the function as argument to not give a wrong impression. tp
is illegal to access from this callback, since it could already have been
freed.
Refactor the deletion code a bit, so that cls_bpf_destroy() can call into
the same code for prog deletion as cls_bpf_delete() op, instead of having
it unnecessarily duplicated.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Couple conflicts resolved here:
1) In the MACB driver, a bug fix to properly initialize the
RX tail pointer properly overlapped with some changes
to support variable sized rings.
2) In XGBE we had a "CONFIG_PM" --> "CONFIG_PM_SLEEP" fix
overlapping with a reorganization of the driver to support
ACPI, OF, as well as PCI variants of the chip.
3) In 'net' we had several probe error path bug fixes to the
stmmac driver, meanwhile a lot of this code was cleaned up
and reorganized in 'net-next'.
4) The cls_flower classifier obtained a helper function in
'net-next' called __fl_delete() and this overlapped with
Daniel Borkamann's bug fix to use RCU for object destruction
in 'net'. It also overlapped with Jiri's change to guard
the rhashtable_remove_fast() call with a check against
tc_skip_sw().
5) In mlx4, a revert bug fix in 'net' overlapped with some
unrelated changes in 'net-next'.
6) In geneve, a stale header pointer after pskb_expand_head()
bug fix in 'net' overlapped with a large reorganization of
the same code in 'net-next'. Since the 'net-next' code no
longer had the bug in question, there was nothing to do
other than to simply take the 'net-next' hunks.
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Roi reported a crash in flower where tp->root was NULL in ->classify()
callbacks. Reason is that in ->destroy() tp->root is set to NULL via
RCU_INIT_POINTER(). It's problematic for some of the classifiers, because
this doesn't respect RCU grace period for them, and as a result, still
outstanding readers from tc_classify() will try to blindly dereference
a NULL tp->root.
The tp->root object is strictly private to the classifier implementation
and holds internal data the core such as tc_ctl_tfilter() doesn't know
about. Within some classifiers, such as cls_bpf, cls_basic, etc, tp->root
is only checked for NULL in ->get() callback, but nowhere else. This is
misleading and seemed to be copied from old classifier code that was not
cleaned up properly. For example, d3fa76ee6b4a ("[NET_SCHED]: cls_basic:
fix NULL pointer dereference") moved tp->root initialization into ->init()
routine, where before it was part of ->change(), so ->get() had to deal
with tp->root being NULL back then, so that was indeed a valid case, after
d3fa76ee6b4a, not really anymore. We used to set tp->root to NULL long
ago in ->destroy(), see 47a1a1d4be29 ("pkt_sched: remove unnecessary xchg()
in packet classifiers"); but the NULLifying was reintroduced with the
RCUification, but it's not correct for every classifier implementation.
In the cases that are fixed here with one exception of cls_cgroup, tp->root
object is allocated and initialized inside ->init() callback, which is always
performed at a point in time after we allocate a new tp, which means tp and
thus tp->root was not globally visible in the tp chain yet (see tc_ctl_tfilter()).
Also, on destruction tp->root is strictly kfree_rcu()'ed in ->destroy()
handler, same for the tp which is kfree_rcu()'ed right when we return
from ->destroy() in tcf_destroy(). This means, the head object's lifetime
for such classifiers is always tied to the tp lifetime. The RCU callback
invocation for the two kfree_rcu() could be out of order, but that's fine
since both are independent.
Dropping the RCU_INIT_POINTER(tp->root, NULL) for these classifiers here
means that 1) we don't need a useless NULL check in fast-path and, 2) that
outstanding readers of that tp in tc_classify() can still execute under
respect with RCU grace period as it is actually expected.
Things that haven't been touched here: cls_fw and cls_route. They each
handle tp->root being NULL in ->classify() path for historic reasons, so
their ->destroy() implementation can stay as is. If someone actually
cares, they could get cleaned up at some point to avoid the test in fast
path. cls_u32 doesn't set tp->root to NULL. For cls_rsvp, I just added a
!head should anyone actually be using/testing it, so it at least aligns with
cls_fw and cls_route. For cls_flower we additionally need to defer rhashtable
destruction (to a sleepable context) after RCU grace period as concurrent
readers might still access it. (Note that in this case we need to hold module
reference to keep work callback address intact, since we only wait on module
unload for all call_rcu()s to finish.)
This fixes one race to bring RCU grace period guarantees back. Next step
as worked on by Cong however is to fix 1e052be69d04 ("net_sched: destroy
proto tp when all filters are gone") to get the order of unlinking the tp
in tc_ctl_tfilter() for the RTM_DELTFILTER case right by moving
RCU_INIT_POINTER() before tcf_destroy() and let the notification for
removal be done through the prior ->delete() callback. Both are independant
issues. Once we have that right, we can then clean tp->root up for a number
of classifiers by not making them RCU pointers, which requires a new callback
(->uninit) that is triggered from tp's RCU callback, where we just kfree()
tp->root from there.
Fixes: 1f947bf151e9 ("net: sched: rcu'ify cls_bpf")
Fixes: 9888faefe132 ("net: sched: cls_basic use RCU")
Fixes: 70da9f0bf999 ("net: sched: cls_flow use RCU")
Fixes: 77b9900ef53a ("tc: introduce Flower classifier")
Fixes: bf3994d2ed31 ("net/sched: introduce Match-all classifier")
Fixes: 952313bd6258 ("net: sched: cls_cgroup use RCU")
Reported-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Roi Dayan <roid@mellanox.com>
Cc: Jiri Pirko <jiri@mellanox.com>
Acked-by: John Fastabend <john.r.fastabend@intel.com>
Acked-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
After setup we don't need to keep user space fd number around anymore, as
it also has no useful meaning for anyone, just remove it.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Wrap several common instances of:
kmemdup(nla_data(attr), nla_len(attr), GFP_KERNEL);
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Acked-by: Johannes Berg <johannes@sipsolutions.net>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Call into offloaded filters to update stats.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Add cls_bpf support for the TCA_CLS_FLAGS_SKIP_SW flag.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Add cls_bpf support for the TCA_CLS_FLAGS_SKIP_HW flag.
Unlike U32 and flower cls_bpf already has some netlink
flags defined. Create a new attribute to be able to use
the same flag values as the above.
Unlike U32 and flower reject unknown flags.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This patch adds hardware offload capability to cls_bpf classifier,
similar to what have been done with U32 and flower.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The skb_mac_header_was_set() test in cls_bpf's and act_bpf's fast-path is
actually unnecessary and can be removed altogether. This was added by
commit a166151cbe33 ("bpf: fix bpf helpers to use skb->mac_header relative
offsets"), which was later on improved by 3431205e0397 ("bpf: make programs
see skb->data == L2 for ingress and egress"). We're always guaranteed to
have valid mac header at the time we invoke cls_bpf_classify() or tcf_bpf().
Reason is that since 6d1ccff62780 ("net: reset mac header in dev_start_xmit()")
we do skb_reset_mac_header() in __dev_queue_xmit() before we could call
into sch_handle_egress() or any subsequent enqueue. sch_handle_ingress()
always sees a valid mac header as well (things like skb_reset_mac_len()
would badly fail otherwise). Thus, drop the unnecessary test in classifier
and action case.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
After commit 22dc13c837c3 ("net_sched: convert tcf_exts from list to pointer array")
we do dynamic allocation in tcf_exts_init(), therefore we need
to handle the ENOMEM case properly.
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Since bpf_prog_get() and program type check is used in a couple of places,
refactor this into a small helper function that we can make use of. Since
the non RO prog->aux part is not used in performance critical paths and a
program destruction via RCU is rather very unlikley when doing the put, we
shouldn't have an issue just doing the bpf_prog_get() + prog->type != type
check, but actually not taking the ref at all (due to being in fdget() /
fdput() section of the bpf fd) is even cleaner and makes the diff smaller
as well, so just go for that. Callsites are changed to make use of the new
helper where possible.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
allow cls_bpf and act_bpf programs access skb->data and skb->data_end pointers.
The bpf helpers that change skb->data need to update data_end pointer as well.
The verifier checks that programs always reload data, data_end pointers
after calls to such bpf helpers.
We cannot add 'data_end' pointer to struct qdisc_skb_cb directly,
since it's embedded as-is by infiniband ipoib, so wrapper struct is needed.
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
There are two issues with the current code. First one is that we need
to set res->class to 0 in case we use non-default classid matching.
This is important for the case where cls_bpf was initially set up with
an optional binding to a default class with tcf_bind_filter(), where
the underlying qdisc implements bind_tcf() that fills res->class and
tests for it later on when doing the classification. Convention for
these cases is that after tc_classify() was called, such qdiscs (atm,
drr, qfq, cbq, hfsc, htb) first test class, and if 0, then they lookup
based on classid.
Second, there's a bug with da mode, where res->classid is only assigned
a 16 bit minor, but it needs to expand to the full 32 bit major/minor
combination instead, therefore we need to expand with the bound major.
This is fine as classes belonging to a classful qdisc must share the
same major.
Fixes: 045efa82ff56 ("cls_bpf: introduce integrated actions")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This work adds a generalization of the ingress qdisc as a qdisc holding
only classifiers. The clsact qdisc works on ingress, but also on egress.
In both cases, it's execution happens without taking the qdisc lock, and
the main difference for the egress part compared to prior version of [1]
is that this can be applied with _any_ underlying real egress qdisc (also
classless ones).
Besides solving the use-case of [1], that is, allowing for more programmability
on assigning skb->priority for the mqprio case that is supported by most
popular 10G+ NICs, it also opens up a lot more flexibility for other tc
applications. The main work on classification can already be done at clsact
egress time if the use-case allows and state stored for later retrieval
f.e. again in skb->priority with major/minors (which is checked by most
classful qdiscs before consulting tc_classify()) and/or in other skb fields
like skb->tc_index for some light-weight post-processing to get to the
eventual classid in case of a classful qdisc. Another use case is that
the clsact egress part allows to have a central egress counterpart to
the ingress classifiers, so that classifiers can easily share state (e.g.
in cls_bpf via eBPF maps) for ingress and egress.
Currently, default setups like mq + pfifo_fast would require for this to
use, for example, prio qdisc instead (to get a tc_classify() run) and to
duplicate the egress classifier for each queue. With clsact, it allows
for leaving the setup as is, it can additionally assign skb->priority to
put the skb in one of pfifo_fast's bands and it can share state with maps.
Moreover, we can access the skb's dst entry (f.e. to retrieve tclassid)
w/o the need to perform a skb_dst_force() to hold on to it any longer. In
lwt case, we can also use this facility to setup dst metadata via cls_bpf
(bpf_skb_set_tunnel_key()) without needing a real egress qdisc just for
that (case of IFF_NO_QUEUE devices, for example).
The realization can be done without any changes to the scheduler core
framework. All it takes is that we have two a-priori defined minors/child
classes, where we can mux between ingress and egress classifier list
(dev->ingress_cl_list and dev->egress_cl_list, latter stored close to
dev->_tx to avoid extra cacheline miss for moderate loads). The egress
part is a bit similar modelled to handle_ing() and patched to a noop in
case the functionality is not used. Both handlers are now called
sch_handle_ingress() and sch_handle_egress(), code sharing among the two
doesn't seem practical as there are various minor differences in both
paths, so that making them conditional in a single handler would rather
slow things down.
Full compatibility to ingress qdisc is provided as well. Since both
piggyback on TC_H_CLSACT, only one of them (ingress/clsact) can exist
per netdevice, and thus ingress qdisc specific behaviour can be retained
for user space. This means, either a user does 'tc qdisc add dev foo ingress'
and configures ingress qdisc as usual, or the 'tc qdisc add dev foo clsact'
alternative, where both, ingress and egress classifier can be configured
as in the below example. ingress qdisc supports attaching classifier to any
minor number whereas clsact has two fixed minors for muxing between the
lists, therefore to not break user space setups, they are better done as
two separate qdiscs.
I decided to extend the sch_ingress module with clsact functionality so
that commonly used code can be reused, the module is being aliased with
sch_clsact so that it can be auto-loaded properly. Alternative would have been
to add a flag when initializing ingress to alter its behaviour plus aliasing
to a different name (as it's more than just ingress). However, the first would
end up, based on the flag, choosing the new/old behaviour by calling different
function implementations to handle each anyway, the latter would require to
register ingress qdisc once again under different alias. So, this really begs
to provide a minimal, cleaner approach to have Qdisc_ops and Qdisc_class_ops
by its own that share callbacks used by both.
Example, adding qdisc:
# tc qdisc add dev foo clsact
# tc qdisc show dev foo
qdisc mq 0: root
qdisc pfifo_fast 0: parent :1 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1
qdisc pfifo_fast 0: parent :2 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1
qdisc pfifo_fast 0: parent :3 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1
qdisc pfifo_fast 0: parent :4 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1
qdisc clsact ffff: parent ffff:fff1
Adding filters (deleting, etc works analogous by specifying ingress/egress):
# tc filter add dev foo ingress bpf da obj bar.o sec ingress
# tc filter add dev foo egress bpf da obj bar.o sec egress
# tc filter show dev foo ingress
filter protocol all pref 49152 bpf
filter protocol all pref 49152 bpf handle 0x1 bar.o:[ingress] direct-action
# tc filter show dev foo egress
filter protocol all pref 49152 bpf
filter protocol all pref 49152 bpf handle 0x1 bar.o:[egress] direct-action
A 'tc filter show dev foo' or 'tc filter show dev foo parent ffff:' will
show an empty list for clsact. Either using the parent names (ingress/egress)
or specifying the full major/minor will then show the related filter lists.
Prior work on a mqprio prequeue() facility [1] was done mainly by John Fastabend.
[1] http://patchwork.ozlabs.org/patch/512949/
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Add a skb_at_tc_ingress() as this will be needed elsewhere as well and
can hide the ugly ifdef.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Using routing realms as part of the classifier is quite useful, it
can be viewed as a tag for one or multiple routing entries (think of
an analogy to net_cls cgroup for processes), set by user space routing
daemons or via iproute2 as an indicator for traffic classifiers and
later on processed in the eBPF program.
Unlike actions, the classifier can inspect device flags and enable
netif_keep_dst() if necessary. tc actions don't have that possibility,
but in case people know what they are doing, it can be used from there
as well (e.g. via devs that must keep dsts by design anyway).
If a realm is set, the handler returns the non-zero realm. User space
can set the full 32bit realm for the dst.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Jamal suggested to further limit the currently allowed subset of opcodes
that may be used by a direct action return code as the intention is not
to replace the full action engine, but rather to have a minimal set that
can be used in the fast-path on things like ingress for some features
that cls_bpf supports.
Classifiers can, of course, still be chained together that have direct
action mode with those that have a full exec pass. For more complex
scenarios that go beyond this minimal set here, the full tcf_exts_exec()
path must be used.
Suggested-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The binding to a particular classid was so far always mandatory for
cls_bpf, but it doesn't need to be. Therefore, lift this restriction
as similarly done in other classifiers.
Only a couple of qdiscs make use of class from the tcf_result, others
don't strictly care, so let the user choose his needs (those that read
out class can handle situations where it could be NULL).
An explicit check for tcf_unbind_filter() is also not needed here, as
the previous r->class was 0, so the xchg() will return that and
therefore a callback to the qdisc's unbind_tcf() is skipped.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
In commit 43388da42a49 ("cls_bpf: introduce integrated actions") we
have added TCA_BPF_FLAGS. We can also retrieve this information from
the prog, dump it back to user space as well. It's useful in tc when
displaying/dumping filter info.
Also, remove tp from cls_bpf_prog_from_efd(), came in as a conflict
from a rebase and it's unused here (later work may add it along with
a real user).
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Existing bpf_clone_redirect() helper clones skb before redirecting
it to RX or TX of destination netdev.
Introduce bpf_redirect() helper that does that without cloning.
Benchmarked with two hosts using 10G ixgbe NICs.
One host is doing line rate pktgen.
Another host is configured as:
$ tc qdisc add dev $dev ingress
$ tc filter add dev $dev root pref 10 u32 match u32 0 0 flowid 1:2 \
action bpf run object-file tcbpf1_kern.o section clone_redirect_xmit drop
so it receives the packet on $dev and immediately xmits it on $dev + 1
The section 'clone_redirect_xmit' in tcbpf1_kern.o file has the program
that does bpf_clone_redirect() and performance is 2.0 Mpps
$ tc filter add dev $dev root pref 10 u32 match u32 0 0 flowid 1:2 \
action bpf run object-file tcbpf1_kern.o section redirect_xmit drop
which is using bpf_redirect() - 2.4 Mpps
and using cls_bpf with integrated actions as:
$ tc filter add dev $dev root pref 10 \
bpf run object-file tcbpf1_kern.o section redirect_xmit integ_act classid 1
performance is 2.5 Mpps
To summarize:
u32+act_bpf using clone_redirect - 2.0 Mpps
u32+act_bpf using redirect - 2.4 Mpps
cls_bpf using redirect - 2.5 Mpps
For comparison linux bridge in this setup is doing 2.1 Mpps
and ixgbe rx + drop in ip_rcv - 7.8 Mpps
Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Often cls_bpf classifier is used with single action drop attached.
Optimize this use case and let cls_bpf return both classid and action.
For backwards compatibility reasons enable this feature under
TCA_BPF_FLAG_ACT_DIRECT flag.
Then more interesting programs like the following are easier to write:
int cls_bpf_prog(struct __sk_buff *skb)
{
/* classify arp, ip, ipv6 into different traffic classes
* and drop all other packets
*/
switch (skb->protocol) {
case htons(ETH_P_ARP):
skb->tc_classid = 1;
break;
case htons(ETH_P_IP):
skb->tc_classid = 2;
break;
case htons(ETH_P_IPV6):
skb->tc_classid = 3;
break;
default:
return TC_ACT_SHOT;
}
return TC_ACT_OK;
}
Joint work with Daniel Borkmann.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The following test case causes a NULL pointer dereference in cls_bpf:
FOO="1,6 0 0 4294967295,"
tc filter add dev foo parent 1: bpf bytecode "$FOO" flowid 1:1 action ok
tc filter replace dev foo parent 1: pref 49152 handle 0x1 \
bpf bytecode "$FOO" flowid 1:1 action drop
The problem is that commit 1f947bf151e9 ("net: sched: rcu'ify cls_bpf")
accidentally swapped the arguments of list_replace_rcu(), the old
element needs to be the first argument and the new element the second.
Fixes: 1f947bf151e9 ("net: sched: rcu'ify cls_bpf")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.r.fastabend@intel.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
eBPF programs attached to ingress and egress qdiscs see inconsistent skb->data.
For ingress L2 header is already pulled, whereas for egress it's present.
This is known to program writers which are currently forced to use
BPF_LL_OFF workaround.
Since programs don't change skb internal pointers it is safe to do
pull/push right around invocation of the program and earlier taps and
later pt->func() will not be affected.
Multiple taps via packet_rcv(), tpacket_rcv() are doing the same trick
around run_filter/BPF_PROG_RUN even if skb_shared.
This fix finally allows programs to use optimized LD_ABS/IND instructions
without BPF_LL_OFF for higher performance.
tc ingress + cls_bpf + samples/bpf/tcbpf1_kern.o
w/o JIT w/JIT
before 20.5 23.6 Mpps
after 21.8 26.6 Mpps
Old programs with BPF_LL_OFF will still work as-is.
We can now undo most of the earlier workaround commit:
a166151cbe33 ("bpf: fix bpf helpers to use skb->mac_header relative offsets")
Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
For the short-term solution, lets fix bpf helper functions to use
skb->mac_header relative offsets instead of skb->data in order to
get the same eBPF programs with cls_bpf and act_bpf work on ingress
and egress qdisc path. We need to ensure that mac_header is set
before calling into programs. This is effectively the first option
from below referenced discussion.
More long term solution for LD_ABS|LD_IND instructions will be more
intrusive but also more beneficial than this, and implemented later
as it's too risky at this point in time.
I.e., we plan to look into the option of moving skb_pull() out of
eth_type_trans() and into netif_receive_skb() as has been suggested
as second option. Meanwhile, this solution ensures ingress can be
used with eBPF, too, and that we won't run into ABI troubles later.
For dealing with negative offsets inside eBPF helper functions,
we've implemented bpf_skb_clone_unwritable() to test for unwriteable
headers.
Reference: http://thread.gmane.org/gmane.linux.network/359129/focus=359694
Fixes: 608cd71a9c7c ("tc: bpf: generalize pedit action")
Fixes: 91bc4822c3d6 ("tc: bpf: add checksum helpers")
Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Currently, it is possible in cls_bpf to access eBPF maps only under
rcu_read_lock_bh() variants: while on ingress side, that is, handle_ing(),
the classifier would be called from __netif_receive_skb_core() under
rcu_read_lock(); on egress side, however, it's rcu_read_lock_bh() via
__dev_queue_xmit().
This rcu/rcu_bh mix doesn't work together with eBPF maps as they require
soley to be called under rcu_read_lock(). eBPF maps could also be shared
among various other eBPF programs (possibly even with other eBPF program
types, f.e. tracing) and user space processes, so any context is assumed.
Therefore, a possible fix for cls_bpf is to wrap/nest eBPF program
invocation under non-bh RCU lock variant.
Fixes: e2e9b6541dd4 ("cls_bpf: add initial eBPF support for programmable classifiers")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Kernel automatically creates a tp for each
(kind, protocol, priority) tuple, which has handle 0,
when we add a new filter, but it still is left there
after we remove our own, unless we don't specify the
handle (literally means all the filters under
the tuple). For example this one is left:
# tc filter show dev eth0
filter parent 8001: protocol arp pref 49152 basic
The user-space is hard to clean up these for kernel
because filters like u32 are organized in a complex way.
So kernel is responsible to remove it after all filters
are gone. Each type of filter has its own way to
store the filters, so each type has to provide its
way to check if all filters are gone.
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Cong Wang <cwang@twopensource.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Jamal Hadi Salim<jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This work extends the "classic" BPF programmable tc classifier by
extending its scope also to native eBPF code!
This allows for user space to implement own custom, 'safe' C like
classifiers (or whatever other frontend language LLVM et al may
provide in future), that can then be compiled with the LLVM eBPF
backend to an eBPF elf file. The result of this can be loaded into
the kernel via iproute2's tc. In the kernel, they can be JITed on
major archs and thus run in native performance.
Simple, minimal toy example to demonstrate the workflow:
#include <linux/ip.h>
#include <linux/if_ether.h>
#include <linux/bpf.h>
#include "tc_bpf_api.h"
__section("classify")
int cls_main(struct sk_buff *skb)
{
return (0x800 << 16) | load_byte(skb, ETH_HLEN + __builtin_offsetof(struct iphdr, tos));
}
char __license[] __section("license") = "GPL";
The classifier can then be compiled into eBPF opcodes and loaded
via tc, for example:
clang -O2 -emit-llvm -c cls.c -o - | llc -march=bpf -filetype=obj -o cls.o
tc filter add dev em1 parent 1: bpf cls.o [...]
As it has been demonstrated, the scope can even reach up to a fully
fledged flow dissector (similarly as in samples/bpf/sockex2_kern.c).
For tc, maps are allowed to be used, but from kernel context only,
in other words, eBPF code can keep state across filter invocations.
In future, we perhaps may reattach from a different application to
those maps e.g., to read out collected statistics/state.
Similarly as in socket filters, we may extend functionality for eBPF
classifiers over time depending on the use cases. For that purpose,
cls_bpf programs are using BPF_PROG_TYPE_SCHED_CLS program type, so
we can allow additional functions/accessors (e.g. an ABI compatible
offset translation to skb fields/metadata). For an initial cls_bpf
support, we allow the same set of helper functions as eBPF socket
filters, but we could diverge at some point in time w/o problem.
I was wondering whether cls_bpf and act_bpf could share C programs,
I can imagine that at some point, we introduce i) further common
handlers for both (or even beyond their scope), and/or if truly needed
ii) some restricted function space for each of them. Both can be
abstracted easily through struct bpf_verifier_ops in future.
The context of cls_bpf versus act_bpf is slightly different though:
a cls_bpf program will return a specific classid whereas act_bpf a
drop/non-drop return code, latter may also in future mangle skbs.
That said, we can surely have a "classify" and "action" section in
a single object file, or considered mentioned constraint add a
possibility of a shared section.
The workflow for getting native eBPF running from tc [1] is as
follows: for f_bpf, I've added a slightly modified ELF parser code
from Alexei's kernel sample, which reads out the LLVM compiled
object, sets up maps (and dynamically fixes up map fds) if any, and
loads the eBPF instructions all centrally through the bpf syscall.
The resulting fd from the loaded program itself is being passed down
to cls_bpf, which looks up struct bpf_prog from the fd store, and
holds reference, so that it stays available also after tc program
lifetime. On tc filter destruction, it will then drop its reference.
Moreover, I've also added the optional possibility to annotate an
eBPF filter with a name (e.g. path to object file, or something
else if preferred) so that when tc dumps currently installed filters,
some more context can be given to an admin for a given instance (as
opposed to just the file descriptor number).
Last but not least, bpf_prog_get() and bpf_prog_put() needed to be
exported, so that eBPF can be used from cls_bpf built as a module.
Thanks to 60a3b2253c41 ("net: bpf: make eBPF interpreter images
read-only") I think this is of no concern since anything wanting to
alter eBPF opcode after verification stage would crash the kernel.
[1] http://git.breakpoint.cc/cgit/dborkman/iproute2.git/log/?h=ebpf
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Conflicts:
arch/arm/boot/dts/imx6sx-sdb.dts
net/sched/cls_bpf.c
Two simple sets of overlapping changes.
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
When creating a bpf classifier in tc with priority collisions and
invoking automatic unique handle assignment, cls_bpf_grab_new_handle()
will return a wrong handle id which in fact is non-unique. Usually
altering of specific filters is being addressed over major id, but
in case of collisions we result in a filter chain, where handle ids
address individual cls_bpf_progs inside the classifier.
Issue is, in cls_bpf_grab_new_handle() we probe for head->hgen handle
in cls_bpf_get() and in case we found a free handle, we're supposed
to use exactly head->hgen. In case of insufficient numbers of handles,
we bail out later as handle id 0 is not allowed.
Fixes: 7d1d65cb84e1 ("net: sched: cls_bpf: add BPF-based classifier")
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
In cls_bpf_modify_existing(), we read out the number of filter blocks,
do some sanity checks, allocate a block on that size, and copy over the
BPF instruction blob from user space, then pass everything through the
classic BPF checker prior to installation of the classifier.
We should reject mismatches here, there are 2 scenarios: the number of
filter blocks could be smaller than the provided instruction blob, so
we do a partial copy of the BPF program, and thus the instructions will
either be rejected from the verifier or a valid BPF program will be run;
in the other case, we'll end up copying more than we're supposed to,
and most likely the trailing garbage will be rejected by the verifier
as well (i.e. we need to fit instruction pattern, ret {A,K} needs to be
last instruction, load/stores must be correct, etc); in case not, we
would leak memory when dumping back instruction patterns. The code should
have only used nla_len() as Dave noted to avoid this from the beginning.
Anyway, lets fix it by rejecting such load attempts.
Fixes: 7d1d65cb84e1 ("net: sched: cls_bpf: add BPF-based classifier")
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
It was suggested by DaveM to change the name as "len" might indicate
unit bytes.
Suggested-by: David Miller <davem@davemloft.net>
Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
It is never called and implementations are void. So just remove it.
Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
rcu variant is not correct here. The code is called by updater (rtnl
lock is held), not by reader (no rcu_read_lock is held).
Signed-off-by: Jiri Pirko <jiri@resnulli.us>
ACKed-by: Jamal Hadi Salim <jhs@mojatatu.com>
Acked-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Acked-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Using the tcf_proto pointer 'tp' from inside the classifiers callback
is not valid because it may have been cleaned up by another call_rcu
occuring on another CPU.
'tp' is currently being used by tcf_unbind_filter() in this patch we
move instances of tcf_unbind_filter outside of the call_rcu() context.
This is safe to do because any running schedulers will either read the
valid class field or it will be zeroed.
And all schedulers today when the class is 0 do a lookup using the
same call used by the tcf_exts_bind(). So even if we have a running
classifier hit the null class pointer it will do a lookup and get
to the same result. This is particularly fragile at the moment because
the only way to verify this is to audit the schedulers call sites.
Reported-by: Cong Wang <xiyou.wangconf@gmail.com>
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Acked-by: Cong Wang <cwang@twopensource.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Jamal Hadi Salim <hadi@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Fixes: commit 1f947bf151e90ec0baad2948 ("net: sched: rcu'ify cls_bpf")
Cc: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This patch makes the cls_bpf classifier RCU safe. The tcf_lock
was being used to protect a list of cls_bpf_prog now this list
is RCU safe and updates occur with rcu_replace.
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
clean up names related to socket filtering and bpf in the following way:
- everything that deals with sockets keeps 'sk_*' prefix
- everything that is pure BPF is changed to 'bpf_*' prefix
split 'struct sk_filter' into
struct sk_filter {
atomic_t refcnt;
struct rcu_head rcu;
struct bpf_prog *prog;
};
and
struct bpf_prog {
u32 jited:1,
len:31;
struct sock_fprog_kern *orig_prog;
unsigned int (*bpf_func)(const struct sk_buff *skb,
const struct bpf_insn *filter);
union {
struct sock_filter insns[0];
struct bpf_insn insnsi[0];
struct work_struct work;
};
};
so that 'struct bpf_prog' can be used independent of sockets and cleans up
'unattached' bpf use cases
split SK_RUN_FILTER macro into:
SK_RUN_FILTER to be used with 'struct sk_filter *' and
BPF_PROG_RUN to be used with 'struct bpf_prog *'
__sk_filter_release(struct sk_filter *) gains
__bpf_prog_release(struct bpf_prog *) helper function
also perform related renames for the functions that work
with 'struct bpf_prog *', since they're on the same lines:
sk_filter_size -> bpf_prog_size
sk_filter_select_runtime -> bpf_prog_select_runtime
sk_filter_free -> bpf_prog_free
sk_unattached_filter_create -> bpf_prog_create
sk_unattached_filter_destroy -> bpf_prog_destroy
sk_store_orig_filter -> bpf_prog_store_orig_filter
sk_release_orig_filter -> bpf_release_orig_filter
__sk_migrate_filter -> bpf_migrate_filter
__sk_prepare_filter -> bpf_prepare_filter
API for attaching classic BPF to a socket stays the same:
sk_attach_filter(prog, struct sock *)/sk_detach_filter(struct sock *)
and SK_RUN_FILTER(struct sk_filter *, ctx) to execute a program
which is used by sockets, tun, af_packet
API for 'unattached' BPF programs becomes:
bpf_prog_create(struct bpf_prog **)/bpf_prog_destroy(struct bpf_prog *)
and BPF_PROG_RUN(struct bpf_prog *, ctx) to execute a program
which is used by isdn, ppp, team, seccomp, ptp, xt_bpf, cls_bpf, test_bpf
Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The sk_unattached_filter_create() API is used by BPF filters that
are not directly attached or related to sockets, and are used in
team, ptp, xt_bpf, cls_bpf, etc. As such all users do their own
internal managment of obtaining filter blocks and thus already
have them in kernel memory and set up before calling into
sk_unattached_filter_create(). As a result, due to __user annotation
in sock_fprog, sparse triggers false positives (incorrect type in
assignment [different address space]) when filters are set up before
passing them to sk_unattached_filter_create(). Therefore, let
sk_unattached_filter_create() API use sock_fprog_kern to overcome
this issue.
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
When actions are attached to a filter, they are a part of the filter
itself, so when changing a filter we should allow to overwrite the actions
inside as well.
In my specific case, when I tried to _append_ a new action to an existing
filter which already has an action, I got EEXIST since kernel refused
to overwrite the existing one in kernel.
This patch checks if we are changing the filter checking NLM_F_CREATE flag
(Sigh, filters don't use NLM_F_REPLACE...) and then passes the boolean down
to actions. This fixes the problem above.
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Cong Wang <cwang@twopensource.com>
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
It will be needed by the next patch.
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
These information can be saved in tcf_exts, and this will
simplify the code.
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Currently actions are chained by a singly linked list,
therefore it is a bit hard to add and remove a specific
entry. Convert it to struct list_head so that in the
latter patch we can remove an action without finding
its head.
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Code indent should use tabs where possible
Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This work contains a lightweight BPF-based traffic classifier that can
serve as a flexible alternative to ematch-based tree classification, i.e.
now that BPF filter engine can also be JITed in the kernel. Naturally, tc
actions and policies are supported as well with cls_bpf. Multiple BPF
programs/filter can be attached for a class, or they can just as well be
written within a single BPF program, that's really up to the user how he
wishes to run/optimize the code, e.g. also for inversion of verdicts etc.
The notion of a BPF program's return/exit codes is being kept as follows:
0: No match
-1: Select classid given in "tc filter ..." command
else: flowid, overwrite the default one
As a minimal usage example with iproute2, we use a 3 band prio root qdisc
on a router with sfq each as leave, and assign ssh and icmp bpf-based
filters to band 1, http traffic to band 2 and the rest to band 3. For the
first two bands we load the bytecode from a file, in the 2nd we load it
inline as an example:
echo 1 > /proc/sys/net/core/bpf_jit_enable
tc qdisc del dev em1 root
tc qdisc add dev em1 root handle 1: prio bands 3 priomap 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
tc qdisc add dev em1 parent 1:1 sfq perturb 16
tc qdisc add dev em1 parent 1:2 sfq perturb 16
tc qdisc add dev em1 parent 1:3 sfq perturb 16
tc filter add dev em1 parent 1: bpf run bytecode-file /etc/tc/ssh.bpf flowid 1:1
tc filter add dev em1 parent 1: bpf run bytecode-file /etc/tc/icmp.bpf flowid 1:1
tc filter add dev em1 parent 1: bpf run bytecode-file /etc/tc/http.bpf flowid 1:2
tc filter add dev em1 parent 1: bpf run bytecode "`bpfc -f tc -i misc.ops`" flowid 1:3
BPF programs can be easily created and passed to tc, either as inline
'bytecode' or 'bytecode-file'. There are a couple of front-ends that can
compile opcodes, for example:
1) People familiar with tcpdump-like filters:
tcpdump -iem1 -ddd port 22 | tr '\n' ',' > /etc/tc/ssh.bpf
2) People that want to low-level program their filters or use BPF
extensions that lack support by libpcap's compiler:
bpfc -f tc -i ssh.ops > /etc/tc/ssh.bpf
ssh.ops example code:
ldh [12]
jne #0x800, drop
ldb [23]
jneq #6, drop
ldh [20]
jset #0x1fff, drop
ldxb 4 * ([14] & 0xf)
ldh [%x + 14]
jeq #0x16, pass
ldh [%x + 16]
jne #0x16, drop
pass: ret #-1
drop: ret #0
It was chosen to load bytecode into tc, since the reverse operation,
tc filter list dev em1, is then able to show the exact commands again.
Possible follow-up work could also include a small expression compiler
for iproute2. Tested with the help of bmon. This idea came up during
the Netfilter Workshop 2013 in Copenhagen. Also thanks to feedback from
Eric Dumazet!
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Cc: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
Daniel Borkmann
David S. Miller
Thomas Graf
Jakub Kicinski
Jamal Hadi Salim
WANG Cong
Alexei Starovoitov
Alexei Starovoitov
Cong Wang
Daniel Borkmann
Jiri Pirko
John Fastabend
Yang Yingliang