Lorenz recently reported:
In our TC classifier cls_redirect [0], we use the following sequence of
helper calls to decapsulate a GUE (basically IP + UDP + custom header)
encapsulated packet:
bpf_skb_adjust_room(skb, -encap_len, BPF_ADJ_ROOM_MAC, BPF_F_ADJ_ROOM_FIXED_GSO)
bpf_redirect(skb->ifindex, BPF_F_INGRESS)
It seems like some checksums of the inner headers are not validated in
this case. For example, a TCP SYN packet with invalid TCP checksum is
still accepted by the network stack and elicits a SYN ACK. [...]
That is, we receive the following packet from the driver:
| ETH | IP | UDP | GUE | IP | TCP |
skb->ip_summed == CHECKSUM_UNNECESSARY
ip_summed is CHECKSUM_UNNECESSARY because our NICs do rx checksum offloading.
On this packet we run skb_adjust_room_mac(-encap_len), and get the following:
| ETH | IP | TCP |
skb->ip_summed == CHECKSUM_UNNECESSARY
Note that ip_summed is still CHECKSUM_UNNECESSARY. After bpf_redirect()'ing
into the ingress, we end up in tcp_v4_rcv(). There, skb_checksum_init() is
turned into a no-op due to CHECKSUM_UNNECESSARY.
The bpf_skb_adjust_room() helper is not aware of protocol specifics. Internally,
it handles the CHECKSUM_COMPLETE case via skb_postpull_rcsum(), but that does
not cover CHECKSUM_UNNECESSARY. In this case skb->csum_level of the original
skb prior to bpf_skb_adjust_room() call was 0, that is, covering UDP. Right now
there is no way to adjust the skb->csum_level. NICs that have checksum offload
disabled (CHECKSUM_NONE) or that support CHECKSUM_COMPLETE are not affected.
Use a safe default for CHECKSUM_UNNECESSARY by resetting to CHECKSUM_NONE and
add a flag to the helper called BPF_F_ADJ_ROOM_NO_CSUM_RESET that allows users
from opting out. Opting out is useful for the case where we don't remove/add
full protocol headers, or for the case where a user wants to adjust the csum
level manually e.g. through bpf_csum_level() helper that is added in subsequent
patch.
The bpf_skb_proto_{4_to_6,6_to_4}() for NAT64/46 translation from the BPF
bpf_skb_change_proto() helper uses bpf_skb_net_hdr_{push,pop}() pair internally
as well but doesn't change layers, only transitions between v4 to v6 and vice
versa, therefore no adoption is required there.
[0] https://lore.kernel.org/bpf/20200424185556.7358-1-lmb@cloudflare.com/
Fixes: 2be7e212d5 ("bpf: add bpf_skb_adjust_room helper")
Reported-by: Lorenz Bauer <lmb@cloudflare.com>
Reported-by: Alan Maguire <alan.maguire@oracle.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Lorenz Bauer <lmb@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Reviewed-by: Alan Maguire <alan.maguire@oracle.com>
Link: https://lore.kernel.org/bpf/CACAyw9-uU_52esMd1JjuA80fRPHJv5vsSg8GnfW3t_qDU4aVKQ@mail.gmail.com/
Link: https://lore.kernel.org/bpf/11a90472e7cce83e76ddbfce81fdfce7bfc68808.1591108731.git.daniel@iogearbox.net
Alexei Starovoitov says:
====================
pull-request: bpf-next 2020-06-01
The following pull-request contains BPF updates for your *net-next* tree.
We've added 55 non-merge commits during the last 1 day(s) which contain
a total of 91 files changed, 4986 insertions(+), 463 deletions(-).
The main changes are:
1) Add rx_queue_mapping to bpf_sock from Amritha.
2) Add BPF ring buffer, from Andrii.
3) Attach and run programs through devmap, from David.
4) Allow SO_BINDTODEVICE opt in bpf_setsockopt, from Ferenc.
5) link based flow_dissector, from Jakub.
6) Use tracing helpers for lsm programs, from Jiri.
7) Several sk_msg fixes and extensions, from John.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Sparse reports a warning at efx_ef10_try_update_nic_stats_vf()
warning: context imbalance in efx_ef10_try_update_nic_stats_vf()
- unexpected unlock
The root cause is the missing annotation at
efx_ef10_try_update_nic_stats_vf()
Add the missing _must_hold(&efx->stats_lock) annotation
Signed-off-by: Jules Irenge <jbi.octave@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Extends support to IPv6 for Inline TLS server.
Signed-off-by: Vinay Kumar Yadav <vinay.yadav@chelsio.com>
v1->v2:
- cc'd tcp folks.
v2->v3:
- changed EXPORT_SYMBOL() to EXPORT_SYMBOL_GPL()
Signed-off-by: David S. Miller <davem@davemloft.net>
Ayush Sawal says:
====================
Fixing compilation warnings and errors
Patch 1: Fixes the warnings seen when compiling using sparse tool.
Patch 2: Fixes a cocci check error introduced after commit
567be3a5d2 ("crypto: chelsio -
Use multiple txq/rxq per tfm to process the requests").
V1->V2
patch1: Avoid type casting by using get_unaligned_be32() and
put_unaligned_be16/32() functions.
patch2: Modified subject of the patch.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
This fixes an error observed after running coccinile check.
drivers/crypto/chelsio/chcr_algo.c:1462:5-8: Unneeded variable:
"err". Return "0" on line 1480
This line is missed in the commit 567be3a5d2 ("crypto:
chelsio - Use multiple txq/rxq per tfm to process the requests").
Fixes: 567be3a5d2 ("crypto:
chelsio - Use multiple txq/rxq per tfm to process the requests").
V1->V2
-Modified subject.
Signed-off-by: Ayush Sawal <ayush.sawal@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch fixes the compilation warnings displayed by sparse tool for
chcr driver.
V1->V2
Avoid type casting by using get_unaligned_be32() and
put_unaligned_be16/32() functions.
The key which comes from stack is an u8 byte stream so we store it in
an unsigned char array(ablkctx->key). The function get_aes_decrypt_key()
is a used to calculate the reverse round key for decryption, for this
operation the key has to be divided into 4 bytes, so to extract 4 bytes
from an u8 byte stream and store it in an u32 variable, get_aligned_be32()
is used. Similarly for copying back the key from u32 variable to the
original u8 key stream, put_aligned_be32() is used.
Signed-off-by: Ayush Sawal <ayush.sawal@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Error messages seen while building kernel with CONFIG_IPV6
disabled.
Signed-off-by: Rohit Maheshwari <rohitm@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Current design enables ktls setting from start, which is not
efficient. Now the feature will be enabled when user demands
TLS offload on any interface.
v1->v2:
- taking ULD module refcount till any single connection exists.
- taking rtnl_lock() before clearing tls_devops.
v2->v3:
- cxgb4 is now registering to tlsdev_ops.
- module refcount inc/dec in chcr.
- refcount is only for connections.
- removed new code from cxgb_set_feature().
v3->v4:
- fixed warning message.
Signed-off-by: Rohit Maheshwari <rohitm@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Socket option IPV6_ADDRFORM supports UDP/UDPLITE and TCP at present.
Previously the checking logic looks like:
if (sk->sk_protocol == IPPROTO_UDP || sk->sk_protocol == IPPROTO_UDPLITE)
do_some_check;
else if (sk->sk_protocol != IPPROTO_TCP)
break;
After commit b6f6118901 ("ipv6: restrict IPV6_ADDRFORM operation"), TCP
was blocked as the logic changed to:
if (sk->sk_protocol == IPPROTO_UDP || sk->sk_protocol == IPPROTO_UDPLITE)
do_some_check;
else if (sk->sk_protocol == IPPROTO_TCP)
do_some_check;
break;
else
break;
Then after commit 82c9ae4408 ("ipv6: fix restrict IPV6_ADDRFORM operation")
UDP/UDPLITE were blocked as the logic changed to:
if (sk->sk_protocol == IPPROTO_UDP || sk->sk_protocol == IPPROTO_UDPLITE)
do_some_check;
if (sk->sk_protocol == IPPROTO_TCP)
do_some_check;
if (sk->sk_protocol != IPPROTO_TCP)
break;
Fix it by using Eric's code and simply remove the break in TCP check, which
looks like:
if (sk->sk_protocol == IPPROTO_UDP || sk->sk_protocol == IPPROTO_UDPLITE)
do_some_check;
else if (sk->sk_protocol == IPPROTO_TCP)
do_some_check;
else
break;
Fixes: 82c9ae4408 ("ipv6: fix restrict IPV6_ADDRFORM operation")
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
tipc_sendstream() may send zero length packet, then tipc_msg_append()
do not alloc skb, skb_peek_tail() will get NULL, msg_set_ack_required
will trigger NULL pointer dereference.
Reported-by: syzbot+8eac6d030e7807c21d32@syzkaller.appspotmail.com
Fixes: 0a3e060f34 ("tipc: add test for Nagle algorithm effectiveness")
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jakub Sitnicki says:
====================
One of the pieces of feedback from recent review of BPF hooks for socket
lookup [0] was that new program types should use bpf_link-based
attachment.
This series introduces new bpf_link type for attaching to network
namespace. All link operations are supported. Errors returned from ops
follow cgroup example. Patch 4 description goes into error semantics.
The major change in v2 is a switch away from RCU to mutex-only
synchronization. Andrii pointed out that it is not needed, and it makes
sense to keep locking straightforward.
Also, there were a couple of bugs in update_prog and fill_info initial
implementation, one picked up by kbuild. Those are now fixed. Tests have
been extended to cover them. Full changelog below.
Series is organized as so:
Patches 1-3 prepare a space in struct net to keep state for attached BPF
programs, and massage the code in flow_dissector to make it attach type
agnostic, to finally move it under kernel/bpf/.
Patch 4, the most important one, introduces new bpf_link link type for
attaching to network namespace.
Patch 5 unifies the update error (ENOLINK) between BPF cgroup and netns.
Patches 6-8 make libbpf and bpftool aware of the new link type.
Patches 9-12 Add and extend tests to check that link low- and high-level
API for operating on links to netns works as intended.
Thanks to Alexei, Andrii, Lorenz, Marek, and Stanislav for feedback.
-jkbs
[0] https://lore.kernel.org/bpf/20200511185218.1422406-1-jakub@cloudflare.com/
Cc: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Cc: Andrii Nakryiko <andrii.nakryiko@gmail.com>
Cc: Lorenz Bauer <lmb@cloudflare.com>
Cc: Marek Majkowski <marek@cloudflare.com>
Cc: Stanislav Fomichev <sdf@google.com>
v1 -> v2:
- Switch to mutex-only synchronization. Don't rely on RCU grace period
guarantee when accessing struct net from link release / update /
fill_info, and when accessing bpf_link from pernet pre_exit
callback. (Andrii)
- Drop patch 1, no longer needed with mutex-only synchronization.
- Don't leak uninitialized variable contents from fill_info callback
when link is in defunct state. (kbuild)
- Make fill_info treat the link as defunct (i.e. no attached netns) when
struct net refcount is 0, but link has not been yet auto-detached.
- Add missing BPF_LINK_TYPE define in bpf_types.h for new link type.
- Fix link update_prog callback to update the prog that will run, and
not just the link itself.
- Return EEXIST on prog attach when link already exists, and on link
create when prog is already attached directly. (Andrii)
- Return EINVAL on prog detach when link is attached. (Andrii)
- Fold __netns_bpf_link_attach into its only caller. (Stanislav)
- Get rid of a wrapper around container_of() (Andrii)
- Use rcu_dereference_protected instead of rcu_access_pointer on
update-side. (Stanislav)
- Make return-on-success from netns_bpf_link_create less
confusing. (Andrii)
- Adapt bpf_link for cgroup to return ENOLINK when updating a defunct
link. (Andrii, Alexei)
- Order new exported symbols in libbpf.map alphabetically (Andrii)
- Keep libbpf's "failed to attach link" warning message clear as to what
we failed to attach to (cgroup vs netns). (Andrii)
- Extract helpers for printing link attach type. (bpftool, Andrii)
- Switch flow_dissector tests to BPF skeleton and extend them to
exercise link-based flow dissector attachment. (Andrii)
- Harden flow dissector attachment tests with prog query checks after
prog attach/detach, or link create/update/close.
- Extend flow dissector tests to cover fill_info for defunct links.
- Rebase onto recent bpf-next
====================
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Extend the existing flow_dissector test case to run tests once using direct
prog attachments, and then for the second time using indirect attachment
via link.
The intention is to exercises the newly added high-level API for attaching
programs to network namespace with links (bpf_program__attach_netns).
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200531082846.2117903-13-jakub@cloudflare.com
Switch flow dissector test setup from custom BPF object loader to BPF
skeleton to save boilerplate and prepare for testing higher-level API for
attaching flow dissector with bpf_link.
To avoid depending on program order in the BPF object when populating the
flow dissector PROG_ARRAY map, change the program section names to contain
the program index into the map. This follows the example set by tailcall
tests.
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200531082846.2117903-12-jakub@cloudflare.com
test_flow_dissector leaves a TAP device after it's finished, potentially
interfering with other tests that will run after it. Fix it by closing the
TAP descriptor on cleanup.
Fixes: 0905beec9f ("selftests/bpf: run flow dissector tests in skb-less mode")
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200531082846.2117903-11-jakub@cloudflare.com
Extend the existing test case for flow dissector attaching to cover:
- link creation,
- link updates,
- link info querying,
- mixing links with direct prog attachment.
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200531082846.2117903-10-jakub@cloudflare.com
Code for printing link attach_type is duplicated in a couple of places, and
likely will be duplicated for future link types as well. Create helpers to
prevent duplication.
Suggested-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200531082846.2117903-8-jakub@cloudflare.com
Add bpf_program__attach_nets(), which uses LINK_CREATE subcommand to create
an FD-based kernel bpf_link, for attach types tied to network namespace,
that is BPF_FLOW_DISSECTOR for the moment.
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200531082846.2117903-7-jakub@cloudflare.com
Failure to update a bpf_link because it has been auto-detached by a dying
cgroup currently results in EINVAL error, even though the arguments passed
to bpf() syscall are not wrong.
bpf_links attaching to netns in this case will return ENOLINK, which
carries the message that the link is no longer attached to anything.
Change cgroup bpf_links to do the same to keep the uAPI errors consistent.
Fixes: 0c991ebc8c ("bpf: Implement bpf_prog replacement for an active bpf_cgroup_link")
Suggested-by: Lorenz Bauer <lmb@cloudflare.com>
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200531082846.2117903-6-jakub@cloudflare.com
Extend bpf() syscall subcommands that operate on bpf_link, that is
LINK_CREATE, LINK_UPDATE, OBJ_GET_INFO, to accept attach types tied to
network namespaces (only flow dissector at the moment).
Link-based and prog-based attachment can be used interchangeably, but only
one can exist at a time. Attempts to attach a link when a prog is already
attached directly, and the other way around, will be met with -EEXIST.
Attempts to detach a program when link exists result in -EINVAL.
Attachment of multiple links of same attach type to one netns is not
supported with the intention to lift the restriction when a use-case
presents itself. Because of that link create returns -E2BIG when trying to
create another netns link, when one already exists.
Link-based attachments to netns don't keep a netns alive by holding a ref
to it. Instead links get auto-detached from netns when the latter is being
destroyed, using a pernet pre_exit callback.
When auto-detached, link lives in defunct state as long there are open FDs
for it. -ENOLINK is returned if a user tries to update a defunct link.
Because bpf_link to netns doesn't hold a ref to struct net, special care is
taken when releasing, updating, or filling link info. The netns might be
getting torn down when any of these link operations are in progress. That
is why auto-detach and update/release/fill_info are synchronized by the
same mutex. Also, link ops have to always check if auto-detach has not
happened yet and if netns is still alive (refcnt > 0).
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200531082846.2117903-5-jakub@cloudflare.com
Move functions to manage BPF programs attached to netns that are not
specific to flow dissector to a dedicated module named
bpf/net_namespace.c.
The set of functions will grow with the addition of bpf_link support for
netns attached programs. This patch prepares ground by creating a place
for it.
This is a code move with no functional changes intended.
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200531082846.2117903-4-jakub@cloudflare.com
In order to:
(1) attach more than one BPF program type to netns, or
(2) support attaching BPF programs to netns with bpf_link, or
(3) support multi-prog attach points for netns
we will need to keep more state per netns than a single pointer like we
have now for BPF flow dissector program.
Prepare for the above by extracting netns_bpf that is part of struct net,
for storing all state related to BPF programs attached to netns.
Turn flow dissector callbacks for querying/attaching/detaching a program
into generic ones that operate on netns_bpf. Next patch will move the
generic callbacks into their own module.
This is similar to how it is organized for cgroup with cgroup_bpf.
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Cc: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/bpf/20200531082846.2117903-3-jakub@cloudflare.com
Split out the part of attach callback that happens with attach/detach lock
acquired. This structures the prog attach callback in a way that opens up
doors for moving the locking out of flow_dissector and into generic
callbacks for attaching/detaching progs to netns in subsequent patches.
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Reviewed-by: Stanislav Fomichev <sdf@google.com>
Link: https://lore.kernel.org/bpf/20200531082846.2117903-2-jakub@cloudflare.com
On systems with recent enough glibc, reallocarray compat won't kick in, so
reallocarray() itself has to come from stdlib.h include. But _GNU_SOURCE is
necessary to enable it. So add it.
Fixes: bf99c936f9 ("libbpf: Add BPF ring buffer support")
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/bpf/20200601202601.2139477-1-andriin@fb.com
Currenty lsm uses bpf_tracing_func_proto helpers which do
not include stack trace or perf event output. It's useful
to have those for bpftrace lsm support [1].
Using tracing_prog_func_proto helpers for lsm programs.
[1] https://github.com/iovisor/bpftrace/pull/1347
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Cc: KP Singh <kpsingh@google.com>
Link: https://lore.kernel.org/bpf/20200531154255.896551-1-jolsa@kernel.org
In order to use standard 'xdp' prefix, rename convert_to_xdp_frame
utility routine in xdp_convert_buff_to_frame and replace all the
occurrences
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Link: https://lore.kernel.org/bpf/6344f739be0d1a08ab2b9607584c4d5478c8c083.1590698295.git.lorenzo@kernel.org
Ferenc Fejes says:
====================
This option makes it possible to programatically bind sockets
to netdevices. With the help of this option sockets
of VRF unaware applications could be distributed between
multiple VRFs with an eBPF program. This lets the applications
benefit from multiple possible routes.
v2:
- splitting up the patch to three parts
- lock_sk parameter for optional locking in sock_bindtoindex - Stanislav Fomichev
- testing the SO_BINDTODEVICE option - Andrii Nakryiko
====================
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
This test intended to verify if SO_BINDTODEVICE option works in
bpf_setsockopt. Because we already in the SOL_SOCKET level in this
connect bpf prog its safe to verify the sanity in the beginning of
the connect_v4_prog by calling the bind_to_device test helper.
The testing environment already created by the test_sock_addr.sh
script so this test assume that two netdevices already existing in
the system: veth pair with names test_sock_addr1 and test_sock_addr2.
The test will try to bind the socket to those devices first.
Then the test assume there are no netdevice with "nonexistent_dev"
name so the bpf_setsockopt will give use ENODEV error.
At the end the test remove the device binding from the socket
by binding it to an empty name.
Signed-off-by: Ferenc Fejes <fejes@inf.elte.hu>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/3f055b8e45c65639c5c73d0b4b6c589e60b86f15.1590871065.git.fejes@inf.elte.hu
The sock_bindtoindex intended for kernel wide usage however
it will lock the socket regardless of the context. This modification
relax this behavior optionally: locking the socket will be optional
by calling the sock_bindtoindex with lock_sk = true.
The modification applied to all users of the sock_bindtoindex.
Signed-off-by: Ferenc Fejes <fejes@inf.elte.hu>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/bee6355da40d9e991b2f2d12b67d55ebb5f5b207.1590871065.git.fejes@inf.elte.hu
buf_prevkey in generic_map_lookup_batch() is allocated with
kmalloc(). It's safe to free it with kfree().
Fixes: cb4d03ab49 ("bpf: Add generic support for lookup batch op")
Signed-off-by: Denis Efremov <efremov@linux.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/bpf/20200601162814.17426-1-efremov@linux.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
John Fastabend says:
====================
If a socket is running a BPF_SK_SKB_SREAM_VERDICT program and KTLS is
enabled the data stream may be broken if both TLS stream parser and
BPF stream parser try to handle data. Fix this here by making KTLS
stream parser run first to ensure TLS messages are received correctly
and then calling the verdict program. This analogous to how we handle
a similar conflict on the TX side.
Note, this is a fix but it doesn't make sense to push this late to
bpf tree so targeting bpf-next and keeping fixes tags.
====================
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
This adds a test for bpf ingress policy. To ensure data writes happen
as expected with extra TLS headers we run these tests with data
verification enabled by default. This will test receive packets have
"PASS" stamped into the front of the payload.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/159079363965.5745.3390806911628980210.stgit@john-Precision-5820-Tower
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
David Ahern says:
====================
Implementation of Daniel's proposal for allowing DEVMAP entries to be
a device index, program fd pair.
Programs are run after XDP_REDIRECT and have access to both Rx device
and Tx device.
v4
- moved struct bpf_devmap_val from uapi to devmap.c, named the union
and dropped the prefix from the elements - Jesper
- fixed 2 bugs in selftests
v3
- renamed struct to bpf_devmap_val
- used offsetofend to check for expected map size, modification of
Toke's comment
- check for explicit value sizes
- adjusted switch statement in dev_map_run_prog per Andrii's comment
- changed SEC shortcut to xdp_devmap
- changed selftests to use skeleton and new map declaration
v2
- moved dev_map_ext_val definition to uapi to formalize the API for devmap
extensions; add bpf_ prefix to the prog_fd and prog_id entries
- changed devmap code to handle struct in a way that it can support future
extensions
- fixed subject in libbpf patch
v1
- fixed prog put on invalid program - Toke
- changed write value from id to fd per Toke's comments about capabilities
- add test cases
====================
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
KTLS uses a stream parser to collect TLS messages and send them to
the upper layer tls receive handler. This ensures the tls receiver
has a full TLS header to parse when it is run. However, when a
socket has BPF_SK_SKB_STREAM_VERDICT program attached before KTLS
is enabled we end up with two stream parsers running on the same
socket.
The result is both try to run on the same socket. First the KTLS
stream parser runs and calls read_sock() which will tcp_read_sock
which in turn calls tcp_rcv_skb(). This dequeues the skb from the
sk_receive_queue. When this is done KTLS code then data_ready()
callback which because we stacked KTLS on top of the bpf stream
verdict program has been replaced with sk_psock_start_strp(). This
will in turn kick the stream parser again and eventually do the
same thing KTLS did above calling into tcp_rcv_skb() and dequeuing
a skb from the sk_receive_queue.
At this point the data stream is broke. Part of the stream was
handled by the KTLS side some other bytes may have been handled
by the BPF side. Generally this results in either missing data
or more likely a "Bad Message" complaint from the kTLS receive
handler as the BPF program steals some bytes meant to be in a
TLS header and/or the TLS header length is no longer correct.
We've already broke the idealized model where we can stack ULPs
in any order with generic callbacks on the TX side to handle this.
So in this patch we do the same thing but for RX side. We add
a sk_psock_strp_enabled() helper so TLS can learn a BPF verdict
program is running and add a tls_sw_has_ctx_rx() helper so BPF
side can learn there is a TLS ULP on the socket.
Then on BPF side we omit calling our stream parser to avoid
breaking the data stream for the KTLS receiver. Then on the
KTLS side we call BPF_SK_SKB_STREAM_VERDICT once the KTLS
receiver is done with the packet but before it posts the
msg to userspace. This gives us symmetry between the TX and
RX halfs and IMO makes it usable again. On the TX side we
process packets in this order BPF -> TLS -> TCP and on
the receive side in the reverse order TCP -> TLS -> BPF.
Discovered while testing OpenSSL 3.0 Alpha2.0 release.
Fixes: d829e9c411 ("tls: convert to generic sk_msg interface")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/159079361946.5745.605854335665044485.stgit@john-Precision-5820-Tower
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Add tests to verify ability to add an XDP program to a
entry in a DEVMAP.
Add negative tests to show DEVMAP programs can not be
attached to devices as a normal XDP program, and accesses
to egress_ifindex require BPF_XDP_DEVMAP attach type.
Signed-off-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20200529220716.75383-6-dsahern@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
We will need this block of code called from tls context shortly
lets refactor the redirect logic so its easy to use. This also
cleans up the switch stmt so we have fewer fallthrough cases.
No logic changes are intended.
Fixes: d829e9c411 ("tls: convert to generic sk_msg interface")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Reviewed-by: Jakub Sitnicki <jakub@cloudflare.com>
Acked-by: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/bpf/159079360110.5745.7024009076049029819.stgit@john-Precision-5820-Tower
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Support SEC("xdp_devmap*") as a short cut for loading the program with
type BPF_PROG_TYPE_XDP and expected attach type BPF_XDP_DEVMAP.
Signed-off-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20200529220716.75383-5-dsahern@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Add xdp_txq_info as the Tx counterpart to xdp_rxq_info. At the
moment only the device is added. Other fields (queue_index)
can be added as use cases arise.
>From a UAPI perspective, add egress_ifindex to xdp context for
bpf programs to see the Tx device.
Update the verifier to only allow accesses to egress_ifindex by
XDP programs with BPF_XDP_DEVMAP expected attach type.
Signed-off-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20200529220716.75383-4-dsahern@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Add BPF_XDP_DEVMAP attach type for use with programs associated with a
DEVMAP entry.
Allow DEVMAPs to associate a program with a device entry by adding
a bpf_prog.fd to 'struct bpf_devmap_val'. Values read show the program
id, so the fd and id are a union. bpf programs can get access to the
struct via vmlinux.h.
The program associated with the fd must have type XDP with expected
attach type BPF_XDP_DEVMAP. When a program is associated with a device
index, the program is run on an XDP_REDIRECT and before the buffer is
added to the per-cpu queue. At this point rxq data is still valid; the
next patch adds tx device information allowing the prorgam to see both
ingress and egress device indices.
XDP generic is skb based and XDP programs do not work with skb's. Block
the use case by walking maps used by a program that is to be attached
via xdpgeneric and fail if any of them are DEVMAP / DEVMAP_HASH with
Block attach of BPF_XDP_DEVMAP programs to devices.
Signed-off-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20200529220716.75383-3-dsahern@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
In bpf_seq_printf() helper, when user specified a "%s" in the
format string, strncpy_from_unsafe() is used to read the actual string
to a buffer. The string could be a format string or a string in
the kernel data structure. It is really unlikely that the string
will reside in the user memory.
This is different from Commit b2a5212fb6 ("bpf: Restrict bpf_trace_printk()'s %s
usage and add %pks, %pus specifier") which still used
strncpy_from_unsafe() for "%s" to preserve the old behavior.
If in the future, bpf_seq_printf() indeed needs to read user
memory, we can implement "%pus" format string.
Based on discussion in [1], if the intent is to read kernel memory,
strncpy_from_unsafe_strict() should be used. So this patch
changed to use strncpy_from_unsafe_strict().
[1]: https://lore.kernel.org/bpf/20200521152301.2587579-1-hch@lst.de/T/
Fixes: 492e639f0c ("bpf: Add bpf_seq_printf and bpf_seq_write helpers")
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Song Liu <songliubraving@fb.com>
Cc: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/bpf/20200529004810.3352219-1-yhs@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Add 'struct bpf_devmap_val' to formalize the expected values that can
be passed in for a DEVMAP. Update devmap code to use the struct.
Signed-off-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20200529220716.75383-2-dsahern@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Add "rx_queue_mapping" to bpf_sock. This gives read access for the
existing field (sk_rx_queue_mapping) of struct sock from bpf_sock.
Semantics for the bpf_sock rx_queue_mapping access are similar to
sk_rx_queue_get(), i.e the value NO_QUEUE_MAPPING is not allowed
and -1 is returned in that case. This is useful for transmit queue
selection based on the received queue index which is cached in the
socket in the receive path.
v3: Addressed review comments to add usecase in patch description,
and fixed default value for rx_queue_mapping.
v2: fixed build error for CONFIG_XPS wrapping, reported by
kbuild test robot <lkp@intel.com>
Signed-off-by: Amritha Nambiar <amritha.nambiar@intel.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Andrii Nakryiko says:
====================
Implement a new BPF ring buffer, as presented at BPF virtual conference ([0]).
It presents an alternative to perf buffer, following its semantics closely,
but allowing sharing same instance of ring buffer across multiple CPUs
efficiently.
Most patches have extensive commentary explaining various aspects, so I'll
keep cover letter short. Overall structure of the patch set:
- patch #1 adds BPF ring buffer implementation to kernel and necessary
verifier support;
- patch #2 adds libbpf consumer implementation for BPF ringbuf;
- patch #3 adds selftest, both for single BPF ring buf use case, as well as
using it with array/hash of maps;
- patch #4 adds extensive benchmarks and provide some analysis in commit
message, it builds upon selftests/bpf's bench runner.
- patch #5 adds most of patch #1 commit message as a doc under
Documentation/bpf/ringbuf.rst.
Litmus tests, validating consumer/producer protocols and memory orderings,
were moved out as discussed in [1] and are going to be posted against -rcu
tree and put under Documentation/litmus-tests/bpf-rb.
[0] https://docs.google.com/presentation/d/18ITdg77Bj6YDOH2LghxrnFxiPWe0fAqcmJY95t_qr0w
[1] https://lkml.org/lkml/2020/5/22/1011
v3->v4:
- fix ringbuf freeing (vunmap, __free_page); verified with a trivial loop
creating and closing ringbuf map endlessly (Daniel);
v2->v3:
- dropped unnecessary smp_wmb() (Paul);
- verifier reference type enhancement patch was dropped (Alexei);
- better verifier message for various memory access checks (Alexei);
- clarified a bit roundup_len() bit shifting (Alexei);
- converted doc to .rst (Alexei);
- fixed warning on 32-bit arches regarding tautological ring area size check.
v1->v2:
- commit()/discard()/output() accept flags (NO_WAKEUP/FORCE_WAKEUP) (Stanislav);
- bpf_ringbuf_query() added, returning available data size, ringbuf size,
consumer/producer positions, needed to implement smarter notification policy
(Stanislav);
- added ringbuf UAPI constants to include/uapi/linux/bpf.h (Jonathan);
- fixed sample size check, added proper ringbuf size check (Jonathan, Alexei);
- wake_up_all() is done through irq_work (Alexei);
- consistent use of smp_load_acquire/smp_store_release, no
READ_ONCE/WRITE_ONCE (Alexei);
- added Documentation/bpf/ringbuf.txt (Stanislav);
- updated litmus test with smp_load_acquire/smp_store_release changes;
- added ring_buffer__consume() API to libbpf for busy-polling;
- ring_buffer__poll() on success returns number of records consumed;
- fixed EPOLL notifications, don't assume available data, done similarly to
perfbuf's implementation;
- both ringbuf and perfbuf now have --rb-sampled mode, instead of
pb-raw/pb-custom mode, updated benchmark results;
- extended ringbuf selftests to validate epoll logic/manual notification
logic, as well as bpf_ringbuf_query().
====================
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
For write-only stacks and queues bpf_map_update_elem should be allowed, but
bpf_map_lookup_elem and bpf_map_lookup_and_delete_elem should fail with EPERM.
Signed-off-by: Anton Protopopov <a.s.protopopov@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200527185700.14658-6-a.s.protopopov@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Add commit description from patch #1 as a stand-alone documentation under
Documentation/bpf, as it might be more convenient format, in long term
perspective.
Suggested-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200529075424.3139988-6-andriin@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Extend bench framework with ability to have benchmark-provided child argument
parser for custom benchmark-specific parameters. This makes bench generic code
modular and independent from any specific benchmark.
Also implement a set of benchmarks for new BPF ring buffer and existing perf
buffer. 4 benchmarks were implemented: 2 variations for each of BPF ringbuf
and perfbuf:,
- rb-libbpf utilizes stock libbpf ring_buffer manager for reading data;
- rb-custom implements custom ring buffer setup and reading code, to
eliminate overheads inherent in generic libbpf code due to callback
functions and the need to update consumer position after each consumed
record, instead of batching updates (due to pessimistic assumption that
user callback might take long time and thus could unnecessarily hold ring
buffer space for too long);
- pb-libbpf uses stock libbpf perf_buffer code with all the default
settings, though uses higher-performance raw event callback to minimize
unnecessary overhead;
- pb-custom implements its own custom consumer code to minimize any possible
overhead of generic libbpf implementation and indirect function calls.
All of the test support default, no data notification skipped, mode, as well
as sampled mode (with --rb-sampled flag), which allows to trigger epoll
notification less frequently and reduce overhead. As will be shown, this mode
is especially critical for perf buffer, which suffers from high overhead of
wakeups in kernel.
Otherwise, all benchamrks implement similar way to generate a batch of records
by using fentry/sys_getpgid BPF program, which pushes a bunch of records in
a tight loop and records number of successful and dropped samples. Each record
is a small 8-byte integer, to minimize the effect of memory copying with
bpf_perf_event_output() and bpf_ringbuf_output().
Benchmarks that have only one producer implement optional back-to-back mode,
in which record production and consumption is alternating on the same CPU.
This is the highest-throughput happy case, showing ultimate performance
achievable with either BPF ringbuf or perfbuf.
All the below scenarios are implemented in a script in
benchs/run_bench_ringbufs.sh. Tests were performed on 28-core/56-thread
Intel Xeon CPU E5-2680 v4 @ 2.40GHz CPU.
Single-producer, parallel producer
==================================
rb-libbpf 12.054 ± 0.320M/s (drops 0.000 ± 0.000M/s)
rb-custom 8.158 ± 0.118M/s (drops 0.001 ± 0.003M/s)
pb-libbpf 0.931 ± 0.007M/s (drops 0.000 ± 0.000M/s)
pb-custom 0.965 ± 0.003M/s (drops 0.000 ± 0.000M/s)
Single-producer, parallel producer, sampled notification
========================================================
rb-libbpf 11.563 ± 0.067M/s (drops 0.000 ± 0.000M/s)
rb-custom 15.895 ± 0.076M/s (drops 0.000 ± 0.000M/s)
pb-libbpf 9.889 ± 0.032M/s (drops 0.000 ± 0.000M/s)
pb-custom 9.866 ± 0.028M/s (drops 0.000 ± 0.000M/s)
Single producer on one CPU, consumer on another one, both running at full
speed. Curiously, rb-libbpf has higher throughput than objectively faster (due
to more lightweight consumer code path) rb-custom. It appears that faster
consumer causes kernel to send notifications more frequently, because consumer
appears to be caught up more frequently. Performance of perfbuf suffers from
default "no sampling" policy and huge overhead that causes.
In sampled mode, rb-custom is winning very significantly eliminating too
frequent in-kernel wakeups, the gain appears to be more than 2x.
Perf buffer achieves even more impressive wins, compared to stock perfbuf
settings, with 10x improvements in throughput with 1:500 sampling rate. The
trade-off is that with sampling, application might not get next X events until
X+1st arrives, which is not always acceptable. With steady influx of events,
though, this shouldn't be a problem.
Overall, single-producer performance of ring buffers seems to be better no
matter the sampled/non-sampled modes, but it especially beats ring buffer
without sampling due to its adaptive notification approach.
Single-producer, back-to-back mode
==================================
rb-libbpf 15.507 ± 0.247M/s (drops 0.000 ± 0.000M/s)
rb-libbpf-sampled 14.692 ± 0.195M/s (drops 0.000 ± 0.000M/s)
rb-custom 21.449 ± 0.157M/s (drops 0.000 ± 0.000M/s)
rb-custom-sampled 20.024 ± 0.386M/s (drops 0.000 ± 0.000M/s)
pb-libbpf 1.601 ± 0.015M/s (drops 0.000 ± 0.000M/s)
pb-libbpf-sampled 8.545 ± 0.064M/s (drops 0.000 ± 0.000M/s)
pb-custom 1.607 ± 0.022M/s (drops 0.000 ± 0.000M/s)
pb-custom-sampled 8.988 ± 0.144M/s (drops 0.000 ± 0.000M/s)
Here we test a back-to-back mode, which is arguably best-case scenario both
for BPF ringbuf and perfbuf, because there is no contention and for ringbuf
also no excessive notification, because consumer appears to be behind after
the first record. For ringbuf, custom consumer code clearly wins with 21.5 vs
16 million records per second exchanged between producer and consumer. Sampled
mode actually hurts a bit due to slightly slower producer logic (it needs to
fetch amount of data available to decide whether to skip or force notification).
Perfbuf with wakeup sampling gets 5.5x throughput increase, compared to
no-sampling version. There also doesn't seem to be noticeable overhead from
generic libbpf handling code.
Perfbuf back-to-back, effect of sample rate
===========================================
pb-sampled-1 1.035 ± 0.012M/s (drops 0.000 ± 0.000M/s)
pb-sampled-5 3.476 ± 0.087M/s (drops 0.000 ± 0.000M/s)
pb-sampled-10 5.094 ± 0.136M/s (drops 0.000 ± 0.000M/s)
pb-sampled-25 7.118 ± 0.153M/s (drops 0.000 ± 0.000M/s)
pb-sampled-50 8.169 ± 0.156M/s (drops 0.000 ± 0.000M/s)
pb-sampled-100 8.887 ± 0.136M/s (drops 0.000 ± 0.000M/s)
pb-sampled-250 9.180 ± 0.209M/s (drops 0.000 ± 0.000M/s)
pb-sampled-500 9.353 ± 0.281M/s (drops 0.000 ± 0.000M/s)
pb-sampled-1000 9.411 ± 0.217M/s (drops 0.000 ± 0.000M/s)
pb-sampled-2000 9.464 ± 0.167M/s (drops 0.000 ± 0.000M/s)
pb-sampled-3000 9.575 ± 0.273M/s (drops 0.000 ± 0.000M/s)
This benchmark shows the effect of event sampling for perfbuf. Back-to-back
mode for highest throughput. Just doing every 5th record notification gives
3.5x speed up. 250-500 appears to be the point of diminishing return, with
almost 9x speed up. Most benchmarks use 500 as the default sampling for pb-raw
and pb-custom.
Ringbuf back-to-back, effect of sample rate
===========================================
rb-sampled-1 1.106 ± 0.010M/s (drops 0.000 ± 0.000M/s)
rb-sampled-5 4.746 ± 0.149M/s (drops 0.000 ± 0.000M/s)
rb-sampled-10 7.706 ± 0.164M/s (drops 0.000 ± 0.000M/s)
rb-sampled-25 12.893 ± 0.273M/s (drops 0.000 ± 0.000M/s)
rb-sampled-50 15.961 ± 0.361M/s (drops 0.000 ± 0.000M/s)
rb-sampled-100 18.203 ± 0.445M/s (drops 0.000 ± 0.000M/s)
rb-sampled-250 19.962 ± 0.786M/s (drops 0.000 ± 0.000M/s)
rb-sampled-500 20.881 ± 0.551M/s (drops 0.000 ± 0.000M/s)
rb-sampled-1000 21.317 ± 0.532M/s (drops 0.000 ± 0.000M/s)
rb-sampled-2000 21.331 ± 0.535M/s (drops 0.000 ± 0.000M/s)
rb-sampled-3000 21.688 ± 0.392M/s (drops 0.000 ± 0.000M/s)
Similar benchmark for ring buffer also shows a great advantage (in terms of
throughput) of skipping notifications. Skipping every 5th one gives 4x boost.
Also similar to perfbuf case, 250-500 seems to be the point of diminishing
returns, giving roughly 20x better results.
Keep in mind, for this test, notifications are controlled manually with
BPF_RB_NO_WAKEUP and BPF_RB_FORCE_WAKEUP. As can be seen from previous
benchmarks, adaptive notifications based on consumer's positions provides same
(or even slightly better due to simpler load generator on BPF side) benefits in
favorable back-to-back scenario. Over zealous and fast consumer, which is
almost always caught up, will make thoughput numbers smaller. That's the case
when manual notification control might prove to be extremely beneficial.
Ringbuf back-to-back, reserve+commit vs output
==============================================
reserve 22.819 ± 0.503M/s (drops 0.000 ± 0.000M/s)
output 18.906 ± 0.433M/s (drops 0.000 ± 0.000M/s)
Ringbuf sampled, reserve+commit vs output
=========================================
reserve-sampled 15.350 ± 0.132M/s (drops 0.000 ± 0.000M/s)
output-sampled 14.195 ± 0.144M/s (drops 0.000 ± 0.000M/s)
BPF ringbuf supports two sets of APIs with various usability and performance
tradeoffs: bpf_ringbuf_reserve()+bpf_ringbuf_commit() vs bpf_ringbuf_output().
This benchmark clearly shows superiority of reserve+commit approach, despite
using a small 8-byte record size.
Single-producer, consumer/producer competing on the same CPU, low batch count
=============================================================================
rb-libbpf 3.045 ± 0.020M/s (drops 3.536 ± 0.148M/s)
rb-custom 3.055 ± 0.022M/s (drops 3.893 ± 0.066M/s)
pb-libbpf 1.393 ± 0.024M/s (drops 0.000 ± 0.000M/s)
pb-custom 1.407 ± 0.016M/s (drops 0.000 ± 0.000M/s)
This benchmark shows one of the worst-case scenarios, in which producer and
consumer do not coordinate *and* fight for the same CPU. No batch count and
sampling settings were able to eliminate drops for ringbuffer, producer is
just too fast for consumer to keep up. But ringbuf and perfbuf still able to
pass through quite a lot of messages, which is more than enough for a lot of
applications.
Ringbuf, multi-producer contention
==================================
rb-libbpf nr_prod 1 10.916 ± 0.399M/s (drops 0.000 ± 0.000M/s)
rb-libbpf nr_prod 2 4.931 ± 0.030M/s (drops 0.000 ± 0.000M/s)
rb-libbpf nr_prod 3 4.880 ± 0.006M/s (drops 0.000 ± 0.000M/s)
rb-libbpf nr_prod 4 3.926 ± 0.004M/s (drops 0.000 ± 0.000M/s)
rb-libbpf nr_prod 8 4.011 ± 0.004M/s (drops 0.000 ± 0.000M/s)
rb-libbpf nr_prod 12 3.967 ± 0.016M/s (drops 0.000 ± 0.000M/s)
rb-libbpf nr_prod 16 2.604 ± 0.030M/s (drops 0.001 ± 0.002M/s)
rb-libbpf nr_prod 20 2.233 ± 0.003M/s (drops 0.000 ± 0.000M/s)
rb-libbpf nr_prod 24 2.085 ± 0.015M/s (drops 0.000 ± 0.000M/s)
rb-libbpf nr_prod 28 2.055 ± 0.004M/s (drops 0.000 ± 0.000M/s)
rb-libbpf nr_prod 32 1.962 ± 0.004M/s (drops 0.000 ± 0.000M/s)
rb-libbpf nr_prod 36 2.089 ± 0.005M/s (drops 0.000 ± 0.000M/s)
rb-libbpf nr_prod 40 2.118 ± 0.006M/s (drops 0.000 ± 0.000M/s)
rb-libbpf nr_prod 44 2.105 ± 0.004M/s (drops 0.000 ± 0.000M/s)
rb-libbpf nr_prod 48 2.120 ± 0.058M/s (drops 0.000 ± 0.001M/s)
rb-libbpf nr_prod 52 2.074 ± 0.024M/s (drops 0.007 ± 0.014M/s)
Ringbuf uses a very short-duration spinlock during reservation phase, to check
few invariants, increment producer count and set record header. This is the
biggest point of contention for ringbuf implementation. This benchmark
evaluates the effect of multiple competing writers on overall throughput of
a single shared ringbuffer.
Overall throughput drops almost 2x when going from single to two
highly-contended producers, gradually dropping with additional competing
producers. Performance drop stabilizes at around 20 producers and hovers
around 2mln even with 50+ fighting producers, which is a 5x drop compared to
non-contended case. Good kernel implementation in kernel helps maintain decent
performance here.
Note, that in the intended real-world scenarios, it's not expected to get even
close to such a high levels of contention. But if contention will become
a problem, there is always an option of sharding few ring buffers across a set
of CPUs.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200529075424.3139988-5-andriin@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
