* Re: [PATCH net-next v2] Add Common Applications Kept Enhanced (cake) qdisc
From: kbuild test robot @ 2018-04-26 19:16 UTC (permalink / raw)
To: Toke Høiland-Jørgensen
Cc: kbuild-all, netdev, cake, Toke Høiland-Jørgensen,
Dave Taht
In-Reply-To: <20180424114407.5939-1-toke@toke.dk>
[-- Attachment #1: Type: text/plain, Size: 3367 bytes --]
Hi Toke,
Thank you for the patch! Perhaps something to improve:
[auto build test WARNING on net-next/master]
url: https://github.com/0day-ci/linux/commits/Toke-H-iland-J-rgensen/Add-Common-Applications-Kept-Enhanced-cake-qdisc/20180426-064653
config: parisc-allmodconfig (attached as .config)
compiler: hppa-linux-gnu-gcc (Debian 7.2.0-11) 7.2.0
reproduce:
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
chmod +x ~/bin/make.cross
# save the attached .config to linux build tree
make.cross ARCH=parisc
All warnings (new ones prefixed by >>):
vim +2589 net//sched/sch_cake.c
2525
2526 static int cake_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
2527 {
2528 struct cake_sched_data *q = qdisc_priv(sch);
2529 struct tc_cake_xstats *st;
2530 size_t size = (sizeof(*st) +
2531 sizeof(struct tc_cake_tin_stats) * q->tin_cnt);
2532 int i;
2533
2534 st = cake_zalloc(size);
2535
2536 if (!st)
2537 return -1;
2538
2539 st->version = 0x102; /* old userspace code discards versions > 0xFF */
2540 st->tin_stats_size = sizeof(struct tc_cake_tin_stats);
2541 st->tin_cnt = q->tin_cnt;
2542
2543 st->avg_trnoff = (q->avg_trnoff + 0x8000) >> 16;
2544 st->max_netlen = q->max_netlen;
2545 st->max_adjlen = q->max_adjlen;
2546 st->min_netlen = q->min_netlen;
2547 st->min_adjlen = q->min_adjlen;
2548
2549 for (i = 0; i < q->tin_cnt; i++) {
2550 struct cake_tin_data *b = &q->tins[q->tin_order[i]];
2551 struct tc_cake_tin_stats *tstat = &st->tin_stats[i];
2552
2553 tstat->threshold_rate = b->tin_rate_bps;
2554 tstat->target_us = cobalt_time_to_us(b->cparams.target);
2555 tstat->interval_us = cobalt_time_to_us(b->cparams.interval);
2556
2557 /* TODO FIXME: add missing aspects of these composite stats */
2558 tstat->sent.packets = b->packets;
2559 tstat->sent.bytes = b->bytes;
2560 tstat->dropped.packets = b->tin_dropped;
2561 tstat->ecn_marked.packets = b->tin_ecn_mark;
2562 tstat->backlog.bytes = b->tin_backlog;
2563 tstat->ack_drops.packets = b->ack_drops;
2564
2565 tstat->peak_delay_us = cobalt_time_to_us(b->peak_delay);
2566 tstat->avge_delay_us = cobalt_time_to_us(b->avge_delay);
2567 tstat->base_delay_us = cobalt_time_to_us(b->base_delay);
2568
2569 tstat->way_indirect_hits = b->way_hits;
2570 tstat->way_misses = b->way_misses;
2571 tstat->way_collisions = b->way_collisions;
2572
2573 tstat->sparse_flows = b->sparse_flow_count +
2574 b->decaying_flow_count;
2575 tstat->bulk_flows = b->bulk_flow_count;
2576 tstat->unresponse_flows = b->unresponsive_flow_count;
2577 tstat->spare = 0;
2578 tstat->max_skblen = b->max_skblen;
2579
2580 tstat->flow_quantum = b->flow_quantum;
2581 }
2582 st->capacity_estimate = q->avg_peak_bandwidth;
2583 st->memory_limit = q->buffer_limit;
2584 st->memory_used = q->buffer_max_used;
2585
2586 i = gnet_stats_copy_app(d, st, size);
2587 cake_free(st);
2588 return i;
> 2589 }
2590
---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation
[-- Attachment #2: .config.gz --]
[-- Type: application/gzip, Size: 52468 bytes --]
^ permalink raw reply
* Re: [dm-devel] [PATCH v5] fault-injection: introduce kvmalloc fallback options
From: Michael S. Tsirkin @ 2018-04-26 19:14 UTC (permalink / raw)
To: Mikulas Patocka
Cc: James Bottomley, Michal Hocko, David Rientjes, dm-devel,
eric.dumazet, netdev, jasowang, Randy Dunlap, linux-kernel,
Matthew Wilcox, linux-mm, edumazet, Andrew Morton, virtualization,
David Miller, Vlastimil Babka
In-Reply-To: <alpine.LRH.2.02.1804261451120.23716@file01.intranet.prod.int.rdu2.redhat.com>
On Thu, Apr 26, 2018 at 02:54:26PM -0400, Mikulas Patocka wrote:
>
>
> On Thu, 26 Apr 2018, Michael S. Tsirkin wrote:
>
> > On Thu, Apr 26, 2018 at 12:07:25PM -0400, Mikulas Patocka wrote:
> > > > IIUC debug kernels mainly exist so people who experience e.g. memory
> > > > corruption can try and debug the failure. In this case, CONFIG_DEBUG_SG
> > > > will *already* catch a failure early. Nothing special needs to be done.
> > >
> > > The patch helps people debug such memory coprruptions (such as using DMA
> > > API on the result of kvmalloc).
> >
> > That's my point. I don't think your patch helps debug any memory
> > corruptions. With CONFIG_DEBUG_SG using DMA API already causes a
> > BUG_ON, that's before any memory can get corrupted.
>
> The patch turns a hard-to-reproduce bug into an easy-to-reproduce bug.
It's still not a memory corruption. It's a BUG_ON the source of which -
should it trigger - can be typically found using grep.
> Obviously we don't want this in production kernels, but in the debug
> kernels it should be done.
>
> Mikulas
I'm not so sure. debug kernels should make debugging easier,
definitely.
Unfortunately they are already slower so some races don't trigger.
If they also start crashing more because we are injecting
memory allocation errors, people are even less likely to
be able to use them.
Just add a comment near the BUG_ON within DMA API telling people how
they can inject this error some more if the bug does not
reproduce, and leave it at that.
--
MST
^ permalink raw reply
* Re: [dm-devel] [PATCH v5] fault-injection: introduce kvmalloc fallback options
From: Michael S. Tsirkin @ 2018-04-26 19:05 UTC (permalink / raw)
To: Mikulas Patocka
Cc: James Bottomley, Michal Hocko, David Rientjes, dm-devel,
eric.dumazet, netdev, jasowang, Randy Dunlap, linux-kernel,
Matthew Wilcox, linux-mm, edumazet, Andrew Morton, virtualization,
David Miller, Vlastimil Babka
In-Reply-To: <alpine.LRH.2.02.1804261454380.23716@file01.intranet.prod.int.rdu2.redhat.com>
On Thu, Apr 26, 2018 at 02:58:08PM -0400, Mikulas Patocka wrote:
>
>
> On Thu, 26 Apr 2018, Michael S. Tsirkin wrote:
>
> > How do you make sure QA tests a specific corner case? Add it to
> > the test plan :)
>
> BTW. how many "lines of code" of corporate bureaucracy would that take? :-)
It's pretty easy at least here at Red Hat.
> > I don't speak for Red Hat, etc.
> >
> > --
> > MST
>
> Mikulas
^ permalink raw reply
* Re: [dm-devel] [PATCH v5] fault-injection: introduce kvmalloc fallback options
From: John Stoffel @ 2018-04-26 18:58 UTC (permalink / raw)
To: James Bottomley
Cc: Mikulas Patocka, Michal, eric.dumazet, mst, netdev, jasowang,
Randy Dunlap, linux-kernel, Matthew Wilcox, Hocko, linux-mm,
dm-devel, Vlastimil Babka, Andrew, David Rientjes, Morton,
virtualization, David Miller, edumazet
In-Reply-To: <1524697697.4100.23.camel@HansenPartnership.com>
>>>>> "James" == James Bottomley <James.Bottomley@HansenPartnership.com> writes:
James> On Wed, 2018-04-25 at 19:00 -0400, Mikulas Patocka wrote:
>>
>> On Wed, 25 Apr 2018, James Bottomley wrote:
>>
>> > > > Do we really need the new config option? This could just be
>> > > > manually tunable via fault injection IIUC.
>> > >
>> > > We do, because we want to enable it in RHEL and Fedora debugging
>> > > kernels, so that it will be tested by the users.
>> > >
>> > > The users won't use some extra magic kernel options or debugfs
>> files.
>> >
>> > If it can be enabled via a tunable, then the distro can turn it on
>> > without the user having to do anything. If you want to present the
>> > user with a different boot option, you can (just have the tunable
>> set
>> > on the command line), but being tunable driven means that you don't
>> > have to choose that option, you could automatically enable it under
>> a
>> > range of circumstances. I think most sane distributions would want
>> > that flexibility.
>> >
>> > Kconfig proliferation, conversely, is a bit of a nightmare from
>> both
>> > the user and the tester's point of view, so we're trying to avoid
>> it
>> > unless absolutely necessary.
>> >
>> > James
>>
>> BTW. even developers who compile their own kernel should have this
>> enabled by a CONFIG option - because if the developer sees the option
>> when browsing through menuconfig, he may enable it. If he doesn't see
>> the option, he won't even know that such an option exists.
James> I may be an atypical developer but I'd rather have a root canal
James> than browse through menuconfig options. The way to get people
James> to learn about new debugging options is to blog about it (or
James> write an lwn.net article) which google will find the next time
James> I ask it how I debug XXX. Google (probably as a service to
James> humanity) rarely turns up Kconfig options in response to a
James> query.
I agree with James here. Looking at the SLAB vs SLUB Kconfig entries
tells me *nothing* about why I should pick one or the other, as an
example.
John
^ permalink raw reply
* Re: [dm-devel] [PATCH v5] fault-injection: introduce kvmalloc fallback options
From: Mikulas Patocka @ 2018-04-26 18:58 UTC (permalink / raw)
To: Michael S. Tsirkin
Cc: James Bottomley, Michal Hocko, David Rientjes, dm-devel,
eric.dumazet, netdev, jasowang, Randy Dunlap, linux-kernel,
Matthew Wilcox, linux-mm, edumazet, Andrew Morton, virtualization,
David Miller, Vlastimil Babka
In-Reply-To: <20180426184845-mutt-send-email-mst@kernel.org>
On Thu, 26 Apr 2018, Michael S. Tsirkin wrote:
> How do you make sure QA tests a specific corner case? Add it to
> the test plan :)
BTW. how many "lines of code" of corporate bureaucracy would that take? :-)
> I don't speak for Red Hat, etc.
>
> --
> MST
Mikulas
^ permalink raw reply
* Re: [dm-devel] [PATCH v5] fault-injection: introduce kvmalloc fallback options
From: Mikulas Patocka @ 2018-04-26 18:54 UTC (permalink / raw)
To: Michael S. Tsirkin
Cc: eric.dumazet, netdev, Randy Dunlap, linux-kernel, Matthew Wilcox,
Michal Hocko, James Bottomley, linux-mm, dm-devel,
Vlastimil Babka, David Rientjes, Andrew Morton, virtualization,
David Miller, edumazet
In-Reply-To: <20180426214011-mutt-send-email-mst@kernel.org>
On Thu, 26 Apr 2018, Michael S. Tsirkin wrote:
> On Thu, Apr 26, 2018 at 12:07:25PM -0400, Mikulas Patocka wrote:
> > > IIUC debug kernels mainly exist so people who experience e.g. memory
> > > corruption can try and debug the failure. In this case, CONFIG_DEBUG_SG
> > > will *already* catch a failure early. Nothing special needs to be done.
> >
> > The patch helps people debug such memory coprruptions (such as using DMA
> > API on the result of kvmalloc).
>
> That's my point. I don't think your patch helps debug any memory
> corruptions. With CONFIG_DEBUG_SG using DMA API already causes a
> BUG_ON, that's before any memory can get corrupted.
The patch turns a hard-to-reproduce bug into an easy-to-reproduce bug.
Obviously we don't want this in production kernels, but in the debug
kernels it should be done.
Mikulas
^ permalink raw reply
* [PULL] virtio: fixups
From: Michael S. Tsirkin @ 2018-04-26 18:50 UTC (permalink / raw)
To: Linus Torvalds; +Cc: kvm, mst, netdev, linux-kernel, stable, virtualization
The following changes since commit 6d08b06e67cd117f6992c46611dfb4ce267cd71e:
Linux 4.17-rc2 (2018-04-22 19:20:09 -0700)
are available in the Git repository at:
git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost.git tags/for_linus
for you to fetch changes up to 5c60300d68da32ca77f7f978039dc72bfc78b06b:
virtio_console: reset on out of memory (2018-04-25 20:41:29 +0300)
----------------------------------------------------------------
virtio: fixups
Latest header update will break QEMU (if it's rebuilt with the new
header) - and it seems that the code there is so fragile that any change
in this header will break it. Add a better interface so users do not
need to change their code every time that header changes.
Fix virtio console for spec compliance.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
----------------------------------------------------------------
Michael S. Tsirkin (7):
virtio_balloon: add array of stat names
virtio_console: don't tie bufs to a vq
virtio: add ability to iterate over vqs
virtio_console: free buffers after reset
virtio_console: drop custom control queue cleanup
virtio_console: move removal code
virtio_console: reset on out of memory
drivers/char/virtio_console.c | 157 ++++++++++++++++--------------------
include/linux/virtio.h | 3 +
include/uapi/linux/virtio_balloon.h | 15 ++++
3 files changed, 89 insertions(+), 86 deletions(-)
^ permalink raw reply
* Re: [dm-devel] [PATCH v5] fault-injection: introduce kvmalloc fallback options
From: Michael S. Tsirkin @ 2018-04-26 18:49 UTC (permalink / raw)
To: Mikulas Patocka
Cc: James Bottomley, Michal Hocko, David Rientjes, dm-devel,
eric.dumazet, netdev, jasowang, Randy Dunlap, linux-kernel,
Matthew Wilcox, linux-mm, edumazet, Andrew Morton, virtualization,
David Miller, Vlastimil Babka
In-Reply-To: <alpine.LRH.2.02.1804261202350.24656@file01.intranet.prod.int.rdu2.redhat.com>
On Thu, Apr 26, 2018 at 12:07:25PM -0400, Mikulas Patocka wrote:
> > IIUC debug kernels mainly exist so people who experience e.g. memory
> > corruption can try and debug the failure. In this case, CONFIG_DEBUG_SG
> > will *already* catch a failure early. Nothing special needs to be done.
>
> The patch helps people debug such memory coprruptions (such as using DMA
> API on the result of kvmalloc).
That's my point. I don't think your patch helps debug any memory
corruptions. With CONFIG_DEBUG_SG using DMA API already causes a
BUG_ON, that's before any memory can get corrupted.
--
MST
^ permalink raw reply
* [PATCH net-next 1/1] inet_diag: fetch cong algo info when socket is destroyed
From: Jamal Hadi Salim @ 2018-04-26 17:58 UTC (permalink / raw)
To: davem
Cc: kraig, netdev, eric.dumazet, kernel, Jamal Hadi Salim,
Jamal Hadi Salim
From: Jamal Hadi Salim <hadi@mojatatu.com>
When a user dumps an existing established tcp socket state
via inet diag, it is possible to retrieve the congestion control
details.
When an the sock is destroyed, the generated event has all the
details available in the dump sans congestion control info.
This patch fixes it.
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
---
net/core/sock_diag.c | 3 +++
net/ipv4/inet_diag.c | 48 ++++++++++++++++++++++++++++++++++++++----------
2 files changed, 41 insertions(+), 10 deletions(-)
diff --git a/net/core/sock_diag.c b/net/core/sock_diag.c
index c37b5be7c5e4..0bf64dd70aee 100644
--- a/net/core/sock_diag.c
+++ b/net/core/sock_diag.c
@@ -7,6 +7,7 @@
#include <net/net_namespace.h>
#include <linux/module.h>
#include <net/sock.h>
+#include <net/tcp.h>
#include <linux/kernel.h>
#include <linux/tcp.h>
#include <linux/workqueue.h>
@@ -112,6 +113,8 @@ static size_t sock_diag_nlmsg_size(void)
{
return NLMSG_ALIGN(sizeof(struct inet_diag_msg)
+ nla_total_size(sizeof(u8)) /* INET_DIAG_PROTOCOL */
+ + nla_total_size(TCP_CA_NAME_MAX) /* INET_DIAG_CONG */
+ + nla_total_size(sizeof(union tcp_cc_info))
+ nla_total_size_64bit(sizeof(struct tcp_info))); /* INET_DIAG_INFO */
}
diff --git a/net/ipv4/inet_diag.c b/net/ipv4/inet_diag.c
index 4e5bc4b2f14e..9722f31cc9c5 100644
--- a/net/ipv4/inet_diag.c
+++ b/net/ipv4/inet_diag.c
@@ -159,6 +159,35 @@ int inet_diag_msg_attrs_fill(struct sock *sk, struct sk_buff *skb,
}
EXPORT_SYMBOL_GPL(inet_diag_msg_attrs_fill);
+static int inet_csk_cong_fill(struct sock *sk, struct sk_buff *skb, int ext)
+{
+ struct inet_connection_sock *icsk = inet_csk(sk);
+ const struct tcp_congestion_ops *ca_ops;
+ union tcp_cc_info info;
+ int attr, err = 0;
+ size_t sz = 0;
+
+ rcu_read_lock();
+ ca_ops = READ_ONCE(icsk->icsk_ca_ops);
+ if (ca_ops) {
+ if (ca_ops->get_info)
+ sz = ca_ops->get_info(sk, ext, &attr, &info);
+ if (ext & (1 << (INET_DIAG_CONG - 1))) {
+ err = nla_put_string(skb, INET_DIAG_CONG, ca_ops->name);
+ if (err < 0) {
+ rcu_read_unlock();
+ return err;
+ }
+ }
+ }
+ rcu_read_unlock();
+
+ if (sz)
+ err = nla_put(skb, attr, sz, &info);
+
+ return err;
+}
+
int inet_sk_diag_fill(struct sock *sk, struct inet_connection_sock *icsk,
struct sk_buff *skb, const struct inet_diag_req_v2 *req,
struct user_namespace *user_ns,
@@ -274,16 +303,7 @@ int inet_sk_diag_fill(struct sock *sk, struct inet_connection_sock *icsk,
goto errout;
if (sk->sk_state < TCP_TIME_WAIT) {
- union tcp_cc_info info;
- size_t sz = 0;
- int attr;
-
- rcu_read_lock();
- ca_ops = READ_ONCE(icsk->icsk_ca_ops);
- if (ca_ops && ca_ops->get_info)
- sz = ca_ops->get_info(sk, ext, &attr, &info);
- rcu_read_unlock();
- if (sz && nla_put(skb, attr, sz, &info) < 0)
+ if (inet_csk_cong_fill(sk, skb, ext))
goto errout;
}
@@ -1215,6 +1235,14 @@ int inet_diag_handler_get_info(struct sk_buff *skb, struct sock *sk)
if (attr)
info = nla_data(attr);
+#define EXT_MASK (1 << (INET_DIAG_VEGASINFO - 1) | 1 << (INET_DIAG_CONG - 1))
+ err = inet_csk_cong_fill(sk, skb, EXT_MASK);
+ if (err) {
+ inet_diag_unlock_handler(handler);
+ nlmsg_cancel(skb, nlh);
+ return err;
+ }
+
handler->idiag_get_info(sk, r, info);
inet_diag_unlock_handler(handler);
--
2.11.0
^ permalink raw reply related
* Re: [PATCH net-next 02/10] udp: add gso
From: Willem de Bruijn @ 2018-04-26 17:49 UTC (permalink / raw)
To: Alexander Duyck
Cc: Netdev, David Miller, Dimitris Michailidis, Willem de Bruijn
In-Reply-To: <CAKgT0UccfvfwLemZ5XcWoZPupQi6U2rfjbGjfjFfvJptMcNYeQ@mail.gmail.com>
>>> That way for things like GSO_PARTIAL we can update after segmentation
>>> since there are only going to be 2 segments most likely instead of
>>> multiple MSS sized segments.
>>
>> I don't quite follow. Which two segments?
>
> When we do GSO partial we end up with 2 segments. One really big one
> that is a multiple of MSS and the remainder assuming the frame is odd
> sized. The idea is we can just replicate all of the headers from the
> outer IP header to the inner transport header in hardware so we do all
> the updates based on that assumption and then we do the standard
> segmentation update on the tail skb.
Thanks for the explanation. That is a very cool feature. I clearly hadn't
read the GSO_PARTIAL code closely enough yet.
^ permalink raw reply
* Re: [PATCH net-next 02/10] udp: add gso
From: Willem de Bruijn @ 2018-04-26 17:48 UTC (permalink / raw)
To: Alexander Duyck
Cc: Netdev, David Miller, Dimitris Michailidis, Willem de Bruijn
In-Reply-To: <CAKgT0UfP7ztrtV7smQviAXZyaiPAwCSARuKnbKnc3SP_R1ogsQ@mail.gmail.com>
Sent a v2 with all but the below suggestion incorporated.
>> diff --git a/net/core/skbuff.c b/net/core/skbuff.c
>> index ff49e352deea..c647cfe114e0 100644
>> --- a/net/core/skbuff.c
>> +++ b/net/core/skbuff.c
>> @@ -4940,6 +4940,8 @@ static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
>> thlen = tcp_hdrlen(skb);
>> } else if (unlikely(skb_is_gso_sctp(skb))) {
>> thlen = sizeof(struct sctphdr);
>> + } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
>> + thlen = sizeof(struct udphdr);
>> }
>> /* UFO sets gso_size to the size of the fragmentation
>> * payload, i.e. the size of the L4 (UDP) header is already
>
> It might make more sense to look at converting this over to a switch
> statement based off of shinfo(skb)->gso_type & GSO_TRANSPORT_MASK,
> where the transport mask consists of the 4 bits that are supported.
I decided to skip this. The types SKB_GSO_{TCPV4, TCPV6, SCTP, UDP_L4}
are far apart in the enum namespace and the tests have to use & instead
of direct comparison. I did not see an obvious way to have the compiler
convert this into a jump table.
Doing so is also a bit out of scope of the feature, so even if feasible
without too much gymnastics I suggest doing so in a separate patch.
^ permalink raw reply
* [PATCH net-next v2 11/11] selftests: udp gso benchmark
From: Willem de Bruijn @ 2018-04-26 17:42 UTC (permalink / raw)
To: netdev; +Cc: davem, alexander.duyck, Willem de Bruijn
In-Reply-To: <20180426174225.246388-1-willemdebruijn.kernel@gmail.com>
From: Willem de Bruijn <willemb@google.com>
Send udp data between a source and sink, optionally with udp gso.
The two processes are expected to be run on separate hosts.
A script is included that runs them together over loopback in a
single namespace for functionality testing.
Signed-off-by: Willem de Bruijn <willemb@google.com>
---
tools/testing/selftests/net/.gitignore | 2 +
tools/testing/selftests/net/Makefile | 3 +-
tools/testing/selftests/net/udpgso_bench.sh | 74 +++
tools/testing/selftests/net/udpgso_bench_rx.c | 265 +++++++++++
tools/testing/selftests/net/udpgso_bench_tx.c | 420 ++++++++++++++++++
5 files changed, 763 insertions(+), 1 deletion(-)
create mode 100755 tools/testing/selftests/net/udpgso_bench.sh
create mode 100644 tools/testing/selftests/net/udpgso_bench_rx.c
create mode 100644 tools/testing/selftests/net/udpgso_bench_tx.c
diff --git a/tools/testing/selftests/net/.gitignore b/tools/testing/selftests/net/.gitignore
index 67a26240407c..f0e6c35a93ae 100644
--- a/tools/testing/selftests/net/.gitignore
+++ b/tools/testing/selftests/net/.gitignore
@@ -9,3 +9,5 @@ reuseport_dualstack
reuseaddr_conflict
tcp_mmap
udpgso
+udpgso_bench_rx
+udpgso_bench_tx
diff --git a/tools/testing/selftests/net/Makefile b/tools/testing/selftests/net/Makefile
index db3303348315..df9102ec7b7a 100644
--- a/tools/testing/selftests/net/Makefile
+++ b/tools/testing/selftests/net/Makefile
@@ -6,12 +6,13 @@ CFLAGS += -I../../../../usr/include/
TEST_PROGS := run_netsocktests run_afpackettests test_bpf.sh netdevice.sh rtnetlink.sh
TEST_PROGS += fib_tests.sh fib-onlink-tests.sh in_netns.sh pmtu.sh udpgso.sh
+TEST_PROGS += udpgso_bench.sh
TEST_GEN_FILES = socket
TEST_GEN_FILES += psock_fanout psock_tpacket msg_zerocopy
TEST_GEN_FILES += tcp_mmap
TEST_GEN_PROGS = reuseport_bpf reuseport_bpf_cpu reuseport_bpf_numa
TEST_GEN_PROGS += reuseport_dualstack reuseaddr_conflict
-TEST_GEN_PROGS += udpgso
+TEST_GEN_PROGS += udpgso udpgso_bench_tx udpgso_bench_rx
include ../lib.mk
diff --git a/tools/testing/selftests/net/udpgso_bench.sh b/tools/testing/selftests/net/udpgso_bench.sh
new file mode 100755
index 000000000000..792fa4d0285e
--- /dev/null
+++ b/tools/testing/selftests/net/udpgso_bench.sh
@@ -0,0 +1,74 @@
+#!/bin/sh
+# SPDX-License-Identifier: GPL-2.0
+#
+# Run a series of udpgso benchmarks
+
+wake_children() {
+ local -r jobs="$(jobs -p)"
+
+ if [[ "${jobs}" != "" ]]; then
+ kill -1 ${jobs} 2>/dev/null
+ fi
+}
+trap wake_children EXIT
+
+run_one() {
+ local -r args=$@
+
+ ./udpgso_bench_rx &
+ ./udpgso_bench_rx -t &
+
+ ./udpgso_bench_tx ${args}
+}
+
+run_in_netns() {
+ local -r args=$@
+
+ ./in_netns.sh $0 __subprocess ${args}
+}
+
+run_udp() {
+ local -r args=$@
+
+ echo "udp"
+ run_in_netns ${args}
+
+ echo "udp gso"
+ run_in_netns ${args} -S
+
+ echo "udp gso zerocopy"
+ run_in_netns ${args} -S -z
+}
+
+run_tcp() {
+ local -r args=$@
+
+ echo "tcp"
+ run_in_netns ${args} -t
+
+ echo "tcp zerocopy"
+ run_in_netns ${args} -t -z
+}
+
+run_all() {
+ local -r core_args="-l 4"
+ local -r ipv4_args="${core_args} -4 -D 127.0.0.1"
+ local -r ipv6_args="${core_args} -6 -D ::1"
+
+ echo "ipv4"
+ run_tcp "${ipv4_args}"
+ run_udp "${ipv4_args}"
+
+ echo "ipv6"
+ run_tcp "${ipv4_args}"
+ run_udp "${ipv6_args}"
+}
+
+if [[ $# -eq 0 ]]; then
+ run_all
+elif [[ $1 == "__subprocess" ]]; then
+ shift
+ run_one $@
+else
+ run_in_netns $@
+fi
diff --git a/tools/testing/selftests/net/udpgso_bench_rx.c b/tools/testing/selftests/net/udpgso_bench_rx.c
new file mode 100644
index 000000000000..727cf67a3f75
--- /dev/null
+++ b/tools/testing/selftests/net/udpgso_bench_rx.c
@@ -0,0 +1,265 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#define _GNU_SOURCE
+
+#include <arpa/inet.h>
+#include <error.h>
+#include <errno.h>
+#include <limits.h>
+#include <linux/errqueue.h>
+#include <linux/if_packet.h>
+#include <linux/socket.h>
+#include <linux/sockios.h>
+#include <net/ethernet.h>
+#include <net/if.h>
+#include <netinet/ip.h>
+#include <netinet/ip6.h>
+#include <netinet/tcp.h>
+#include <netinet/udp.h>
+#include <poll.h>
+#include <sched.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/ioctl.h>
+#include <sys/socket.h>
+#include <sys/stat.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <unistd.h>
+
+static int cfg_port = 8000;
+static bool cfg_tcp;
+static bool cfg_verify;
+
+static bool interrupted;
+static unsigned long packets, bytes;
+
+static void sigint_handler(int signum)
+{
+ if (signum == SIGINT)
+ interrupted = true;
+}
+
+static unsigned long gettimeofday_ms(void)
+{
+ struct timeval tv;
+
+ gettimeofday(&tv, NULL);
+ return (tv.tv_sec * 1000) + (tv.tv_usec / 1000);
+}
+
+static void do_poll(int fd)
+{
+ struct pollfd pfd;
+ int ret;
+
+ pfd.events = POLLIN;
+ pfd.revents = 0;
+ pfd.fd = fd;
+
+ do {
+ ret = poll(&pfd, 1, 10);
+ if (ret == -1)
+ error(1, errno, "poll");
+ if (ret == 0)
+ continue;
+ if (pfd.revents != POLLIN)
+ error(1, errno, "poll: 0x%x expected 0x%x\n",
+ pfd.revents, POLLIN);
+ } while (!ret && !interrupted);
+}
+
+static int do_socket(bool do_tcp)
+{
+ struct sockaddr_in6 addr = {0};
+ int fd, val;
+
+ fd = socket(PF_INET6, cfg_tcp ? SOCK_STREAM : SOCK_DGRAM, 0);
+ if (fd == -1)
+ error(1, errno, "socket");
+
+ val = 1 << 21;
+ if (setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)))
+ error(1, errno, "setsockopt rcvbuf");
+ val = 1;
+ if (setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &val, sizeof(val)))
+ error(1, errno, "setsockopt reuseport");
+
+ addr.sin6_family = PF_INET6;
+ addr.sin6_port = htons(cfg_port);
+ addr.sin6_addr = in6addr_any;
+ if (bind(fd, (void *) &addr, sizeof(addr)))
+ error(1, errno, "bind");
+
+ if (do_tcp) {
+ int accept_fd = fd;
+
+ if (listen(accept_fd, 1))
+ error(1, errno, "listen");
+
+ do_poll(accept_fd);
+
+ fd = accept(accept_fd, NULL, NULL);
+ if (fd == -1)
+ error(1, errno, "accept");
+ if (close(accept_fd))
+ error(1, errno, "close accept fd");
+ }
+
+ return fd;
+}
+
+/* Flush all outstanding bytes for the tcp receive queue */
+static void do_flush_tcp(int fd)
+{
+ int ret;
+
+ while (true) {
+ /* MSG_TRUNC flushes up to len bytes */
+ ret = recv(fd, NULL, 1 << 21, MSG_TRUNC | MSG_DONTWAIT);
+ if (ret == -1 && errno == EAGAIN)
+ return;
+ if (ret == -1)
+ error(1, errno, "flush");
+ if (ret == 0) {
+ /* client detached */
+ exit(0);
+ }
+
+ packets++;
+ bytes += ret;
+ }
+
+}
+
+static char sanitized_char(char val)
+{
+ return (val >= 'a' && val <= 'z') ? val : '.';
+}
+
+static void do_verify_udp(const char *data, int len)
+{
+ char cur = data[0];
+ int i;
+
+ /* verify contents */
+ if (cur < 'a' || cur > 'z')
+ error(1, 0, "data initial byte out of range");
+
+ for (i = 1; i < len; i++) {
+ if (cur == 'z')
+ cur = 'a';
+ else
+ cur++;
+
+ if (data[i] != cur)
+ error(1, 0, "data[%d]: len %d, %c(%hhu) != %c(%hhu)\n",
+ i, len,
+ sanitized_char(data[i]), data[i],
+ sanitized_char(cur), cur);
+ }
+}
+
+/* Flush all outstanding datagrams. Verify first few bytes of each. */
+static void do_flush_udp(int fd)
+{
+ static char rbuf[ETH_DATA_LEN];
+ int ret, len, budget = 256;
+
+ len = cfg_verify ? sizeof(rbuf) : 0;
+ while (budget--) {
+ /* MSG_TRUNC will make return value full datagram length */
+ ret = recv(fd, rbuf, len, MSG_TRUNC | MSG_DONTWAIT);
+ if (ret == -1 && errno == EAGAIN)
+ return;
+ if (ret == -1)
+ error(1, errno, "recv");
+ if (len) {
+ if (ret == 0)
+ error(1, errno, "recv: 0 byte datagram\n");
+
+ do_verify_udp(rbuf, ret);
+ }
+
+ packets++;
+ bytes += ret;
+ }
+}
+
+static void usage(const char *filepath)
+{
+ error(1, 0, "Usage: %s [-tv] [-p port]", filepath);
+}
+
+static void parse_opts(int argc, char **argv)
+{
+ int c;
+
+ while ((c = getopt(argc, argv, "ptv")) != -1) {
+ switch (c) {
+ case 'p':
+ cfg_port = htons(strtoul(optarg, NULL, 0));
+ break;
+ case 't':
+ cfg_tcp = true;
+ break;
+ case 'v':
+ cfg_verify = true;
+ break;
+ }
+ }
+
+ if (optind != argc)
+ usage(argv[0]);
+
+ if (cfg_tcp && cfg_verify)
+ error(1, 0, "TODO: implement verify mode for tcp");
+}
+
+static void do_recv(void)
+{
+ unsigned long tnow, treport;
+ int fd;
+
+ fd = do_socket(cfg_tcp);
+
+ treport = gettimeofday_ms() + 1000;
+ do {
+ do_poll(fd);
+
+ if (cfg_tcp)
+ do_flush_tcp(fd);
+ else
+ do_flush_udp(fd);
+
+ tnow = gettimeofday_ms();
+ if (tnow > treport) {
+ if (packets)
+ fprintf(stderr,
+ "%s rx: %6lu MB/s %8lu calls/s\n",
+ cfg_tcp ? "tcp" : "udp",
+ bytes >> 20, packets);
+ bytes = packets = 0;
+ treport = tnow + 1000;
+ }
+
+ } while (!interrupted);
+
+ if (close(fd))
+ error(1, errno, "close");
+}
+
+int main(int argc, char **argv)
+{
+ parse_opts(argc, argv);
+
+ signal(SIGINT, sigint_handler);
+
+ do_recv();
+
+ return 0;
+}
diff --git a/tools/testing/selftests/net/udpgso_bench_tx.c b/tools/testing/selftests/net/udpgso_bench_tx.c
new file mode 100644
index 000000000000..e821564053cf
--- /dev/null
+++ b/tools/testing/selftests/net/udpgso_bench_tx.c
@@ -0,0 +1,420 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#define _GNU_SOURCE
+
+#include <arpa/inet.h>
+#include <errno.h>
+#include <error.h>
+#include <netinet/if_ether.h>
+#include <netinet/in.h>
+#include <netinet/ip.h>
+#include <netinet/ip6.h>
+#include <netinet/udp.h>
+#include <poll.h>
+#include <sched.h>
+#include <signal.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/socket.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <unistd.h>
+
+#ifndef ETH_MAX_MTU
+#define ETH_MAX_MTU 0xFFFFU
+#endif
+
+#ifndef UDP_SEGMENT
+#define UDP_SEGMENT 103
+#endif
+
+#ifndef SO_ZEROCOPY
+#define SO_ZEROCOPY 60
+#endif
+
+#ifndef MSG_ZEROCOPY
+#define MSG_ZEROCOPY 0x4000000
+#endif
+
+#define NUM_PKT 100
+
+static bool cfg_cache_trash;
+static int cfg_cpu = -1;
+static int cfg_connected = true;
+static int cfg_family = PF_UNSPEC;
+static uint16_t cfg_mss;
+static int cfg_payload_len = (1472 * 42);
+static int cfg_port = 8000;
+static int cfg_runtime_ms = -1;
+static bool cfg_segment;
+static bool cfg_sendmmsg;
+static bool cfg_tcp;
+static bool cfg_zerocopy;
+
+static socklen_t cfg_alen;
+static struct sockaddr_storage cfg_dst_addr;
+
+static bool interrupted;
+static char buf[NUM_PKT][ETH_MAX_MTU];
+
+static void sigint_handler(int signum)
+{
+ if (signum == SIGINT)
+ interrupted = true;
+}
+
+static unsigned long gettimeofday_ms(void)
+{
+ struct timeval tv;
+
+ gettimeofday(&tv, NULL);
+ return (tv.tv_sec * 1000) + (tv.tv_usec / 1000);
+}
+
+static int set_cpu(int cpu)
+{
+ cpu_set_t mask;
+
+ CPU_ZERO(&mask);
+ CPU_SET(cpu, &mask);
+ if (sched_setaffinity(0, sizeof(mask), &mask))
+ error(1, 0, "setaffinity %d", cpu);
+
+ return 0;
+}
+
+static void setup_sockaddr(int domain, const char *str_addr, void *sockaddr)
+{
+ struct sockaddr_in6 *addr6 = (void *) sockaddr;
+ struct sockaddr_in *addr4 = (void *) sockaddr;
+
+ switch (domain) {
+ case PF_INET:
+ addr4->sin_family = AF_INET;
+ addr4->sin_port = htons(cfg_port);
+ if (inet_pton(AF_INET, str_addr, &(addr4->sin_addr)) != 1)
+ error(1, 0, "ipv4 parse error: %s", str_addr);
+ break;
+ case PF_INET6:
+ addr6->sin6_family = AF_INET6;
+ addr6->sin6_port = htons(cfg_port);
+ if (inet_pton(AF_INET6, str_addr, &(addr6->sin6_addr)) != 1)
+ error(1, 0, "ipv6 parse error: %s", str_addr);
+ break;
+ default:
+ error(1, 0, "illegal domain");
+ }
+}
+
+static void flush_zerocopy(int fd)
+{
+ struct msghdr msg = {0}; /* flush */
+ int ret;
+
+ while (1) {
+ ret = recvmsg(fd, &msg, MSG_ERRQUEUE);
+ if (ret == -1 && errno == EAGAIN)
+ break;
+ if (ret == -1)
+ error(1, errno, "errqueue");
+ if (msg.msg_flags != (MSG_ERRQUEUE | MSG_CTRUNC))
+ error(1, 0, "errqueue: flags 0x%x\n", msg.msg_flags);
+ msg.msg_flags = 0;
+ }
+}
+
+static int send_tcp(int fd, char *data)
+{
+ int ret, done = 0, count = 0;
+
+ while (done < cfg_payload_len) {
+ ret = send(fd, data + done, cfg_payload_len - done,
+ cfg_zerocopy ? MSG_ZEROCOPY : 0);
+ if (ret == -1)
+ error(1, errno, "write");
+
+ done += ret;
+ count++;
+ }
+
+ return count;
+}
+
+static int send_udp(int fd, char *data)
+{
+ int ret, total_len, len, count = 0;
+
+ total_len = cfg_payload_len;
+
+ while (total_len) {
+ len = total_len < cfg_mss ? total_len : cfg_mss;
+
+ ret = sendto(fd, data, len, cfg_zerocopy ? MSG_ZEROCOPY : 0,
+ cfg_connected ? NULL : (void *)&cfg_dst_addr,
+ cfg_connected ? 0 : cfg_alen);
+ if (ret == -1)
+ error(1, errno, "write");
+ if (ret != len)
+ error(1, errno, "write: %uB != %uB\n", ret, len);
+
+ total_len -= len;
+ count++;
+ }
+
+ return count;
+}
+
+static int send_udp_sendmmsg(int fd, char *data)
+{
+ const int max_nr_msg = ETH_MAX_MTU / ETH_DATA_LEN;
+ struct mmsghdr mmsgs[max_nr_msg];
+ struct iovec iov[max_nr_msg];
+ unsigned int off = 0, left;
+ int i = 0, ret;
+
+ memset(mmsgs, 0, sizeof(mmsgs));
+
+ left = cfg_payload_len;
+ while (left) {
+ if (i == max_nr_msg)
+ error(1, 0, "sendmmsg: exceeds max_nr_msg");
+
+ iov[i].iov_base = data + off;
+ iov[i].iov_len = cfg_mss < left ? cfg_mss : left;
+
+ mmsgs[i].msg_hdr.msg_iov = iov + i;
+ mmsgs[i].msg_hdr.msg_iovlen = 1;
+
+ off += iov[i].iov_len;
+ left -= iov[i].iov_len;
+ i++;
+ }
+
+ ret = sendmmsg(fd, mmsgs, i, cfg_zerocopy ? MSG_ZEROCOPY : 0);
+ if (ret == -1)
+ error(1, errno, "sendmmsg");
+
+ return ret;
+}
+
+static void send_udp_segment_cmsg(struct cmsghdr *cm)
+{
+ uint16_t *valp;
+
+ cm->cmsg_level = SOL_UDP;
+ cm->cmsg_type = UDP_SEGMENT;
+ cm->cmsg_len = CMSG_LEN(sizeof(cfg_mss));
+ valp = (void *)CMSG_DATA(cm);
+ *valp = cfg_mss;
+}
+
+static int send_udp_segment(int fd, char *data)
+{
+ char control[CMSG_SPACE(sizeof(cfg_mss))] = {0};
+ struct msghdr msg = {0};
+ struct iovec iov = {0};
+ int ret;
+
+ iov.iov_base = data;
+ iov.iov_len = cfg_payload_len;
+
+ msg.msg_iov = &iov;
+ msg.msg_iovlen = 1;
+
+ msg.msg_control = control;
+ msg.msg_controllen = sizeof(control);
+ send_udp_segment_cmsg(CMSG_FIRSTHDR(&msg));
+
+ msg.msg_name = (void *)&cfg_dst_addr;
+ msg.msg_namelen = cfg_alen;
+
+ ret = sendmsg(fd, &msg, cfg_zerocopy ? MSG_ZEROCOPY : 0);
+ if (ret == -1)
+ error(1, errno, "sendmsg");
+ if (ret != iov.iov_len)
+ error(1, 0, "sendmsg: %u != %lu\n", ret, iov.iov_len);
+
+ return 1;
+}
+
+static void usage(const char *filepath)
+{
+ error(1, 0, "Usage: %s [-46cmStuz] [-C cpu] [-D dst ip] [-l secs] [-p port] [-s sendsize]",
+ filepath);
+}
+
+static void parse_opts(int argc, char **argv)
+{
+ int max_len, hdrlen;
+ int c;
+
+ while ((c = getopt(argc, argv, "46cC:D:l:mp:s:Stuz")) != -1) {
+ switch (c) {
+ case '4':
+ if (cfg_family != PF_UNSPEC)
+ error(1, 0, "Pass one of -4 or -6");
+ cfg_family = PF_INET;
+ cfg_alen = sizeof(struct sockaddr_in);
+ break;
+ case '6':
+ if (cfg_family != PF_UNSPEC)
+ error(1, 0, "Pass one of -4 or -6");
+ cfg_family = PF_INET6;
+ cfg_alen = sizeof(struct sockaddr_in6);
+ break;
+ case 'c':
+ cfg_cache_trash = true;
+ break;
+ case 'C':
+ cfg_cpu = strtol(optarg, NULL, 0);
+ break;
+ case 'D':
+ setup_sockaddr(cfg_family, optarg, &cfg_dst_addr);
+ break;
+ case 'l':
+ cfg_runtime_ms = strtoul(optarg, NULL, 10) * 1000;
+ break;
+ case 'm':
+ cfg_sendmmsg = true;
+ break;
+ case 'p':
+ cfg_port = strtoul(optarg, NULL, 0);
+ break;
+ case 's':
+ cfg_payload_len = strtoul(optarg, NULL, 0);
+ break;
+ case 'S':
+ cfg_segment = true;
+ break;
+ case 't':
+ cfg_tcp = true;
+ break;
+ case 'u':
+ cfg_connected = false;
+ break;
+ case 'z':
+ cfg_zerocopy = true;
+ break;
+ }
+ }
+
+ if (optind != argc)
+ usage(argv[0]);
+
+ if (cfg_family == PF_UNSPEC)
+ error(1, 0, "must pass one of -4 or -6");
+ if (cfg_tcp && !cfg_connected)
+ error(1, 0, "connectionless tcp makes no sense");
+ if (cfg_segment && cfg_sendmmsg)
+ error(1, 0, "cannot combine segment offload and sendmmsg");
+
+ if (cfg_family == PF_INET)
+ hdrlen = sizeof(struct iphdr) + sizeof(struct udphdr);
+ else
+ hdrlen = sizeof(struct ip6_hdr) + sizeof(struct udphdr);
+
+ cfg_mss = ETH_DATA_LEN - hdrlen;
+ max_len = ETH_MAX_MTU - hdrlen;
+
+ if (cfg_payload_len > max_len)
+ error(1, 0, "payload length %u exceeds max %u",
+ cfg_payload_len, max_len);
+}
+
+static void set_pmtu_discover(int fd, bool is_ipv4)
+{
+ int level, name, val;
+
+ if (is_ipv4) {
+ level = SOL_IP;
+ name = IP_MTU_DISCOVER;
+ val = IP_PMTUDISC_DO;
+ } else {
+ level = SOL_IPV6;
+ name = IPV6_MTU_DISCOVER;
+ val = IPV6_PMTUDISC_DO;
+ }
+
+ if (setsockopt(fd, level, name, &val, sizeof(val)))
+ error(1, errno, "setsockopt path mtu");
+}
+
+int main(int argc, char **argv)
+{
+ unsigned long num_msgs, num_sends;
+ unsigned long tnow, treport, tstop;
+ int fd, i, val;
+
+ parse_opts(argc, argv);
+
+ if (cfg_cpu > 0)
+ set_cpu(cfg_cpu);
+
+ for (i = 0; i < sizeof(buf[0]); i++)
+ buf[0][i] = 'a' + (i % 26);
+ for (i = 1; i < NUM_PKT; i++)
+ memcpy(buf[i], buf[0], sizeof(buf[0]));
+
+ signal(SIGINT, sigint_handler);
+
+ fd = socket(cfg_family, cfg_tcp ? SOCK_STREAM : SOCK_DGRAM, 0);
+ if (fd == -1)
+ error(1, errno, "socket");
+
+ if (cfg_zerocopy) {
+ val = 1;
+ if (setsockopt(fd, SOL_SOCKET, SO_ZEROCOPY, &val, sizeof(val)))
+ error(1, errno, "setsockopt zerocopy");
+ }
+
+ if (cfg_connected &&
+ connect(fd, (void *)&cfg_dst_addr, cfg_alen))
+ error(1, errno, "connect");
+
+ if (cfg_segment)
+ set_pmtu_discover(fd, cfg_family == PF_INET);
+
+ num_msgs = num_sends = 0;
+ tnow = gettimeofday_ms();
+ tstop = tnow + cfg_runtime_ms;
+ treport = tnow + 1000;
+
+ i = 0;
+ do {
+ if (cfg_tcp)
+ num_sends += send_tcp(fd, buf[i]);
+ else if (cfg_segment)
+ num_sends += send_udp_segment(fd, buf[i]);
+ else if (cfg_sendmmsg)
+ num_sends += send_udp_sendmmsg(fd, buf[i]);
+ else
+ num_sends += send_udp(fd, buf[i]);
+ num_msgs++;
+
+ if (cfg_zerocopy && ((num_msgs & 0xF) == 0))
+ flush_zerocopy(fd);
+
+ tnow = gettimeofday_ms();
+ if (tnow > treport) {
+ fprintf(stderr,
+ "%s tx: %6lu MB/s %8lu calls/s %6lu msg/s\n",
+ cfg_tcp ? "tcp" : "udp",
+ (num_msgs * cfg_payload_len) >> 20,
+ num_sends, num_msgs);
+ num_msgs = num_sends = 0;
+ treport = tnow + 1000;
+ }
+
+ /* cold cache when writing buffer */
+ if (cfg_cache_trash)
+ i = ++i < NUM_PKT ? i : 0;
+
+ } while (!interrupted && (cfg_runtime_ms == -1 || tnow < tstop));
+
+ if (close(fd))
+ error(1, errno, "close");
+
+ return 0;
+}
--
2.17.0.484.g0c8726318c-goog
^ permalink raw reply related
* [PATCH net-next v2 09/11] selftests: udp gso with connected sockets
From: Willem de Bruijn @ 2018-04-26 17:42 UTC (permalink / raw)
To: netdev; +Cc: davem, alexander.duyck, Willem de Bruijn
In-Reply-To: <20180426174225.246388-1-willemdebruijn.kernel@gmail.com>
From: Willem de Bruijn <willemb@google.com>
Connected sockets use path mtu instead of device mtu.
Test this path by inserting a route mtu that is lower than the device
mtu. Verify that the path mtu for the connection matches this lower
number, then run the same test as in the connectionless case.
Signed-off-by: Willem de Bruijn <willemb@google.com>
---
tools/testing/selftests/net/udpgso.c | 117 +++++++++++++++++++++++++-
tools/testing/selftests/net/udpgso.sh | 7 ++
2 files changed, 122 insertions(+), 2 deletions(-)
diff --git a/tools/testing/selftests/net/udpgso.c b/tools/testing/selftests/net/udpgso.c
index 68a230dfee73..a47dca025346 100644
--- a/tools/testing/selftests/net/udpgso.c
+++ b/tools/testing/selftests/net/udpgso.c
@@ -47,6 +47,7 @@
static bool cfg_do_ipv4;
static bool cfg_do_ipv6;
+static bool cfg_do_connected;
static bool cfg_do_connectionless;
static bool cfg_do_setsockopt;
static int cfg_specific_test_id = -1;
@@ -273,6 +274,101 @@ static void set_pmtu_discover(int fd, bool is_ipv4)
error(1, errno, "setsockopt path mtu");
}
+static unsigned int get_path_mtu(int fd, bool is_ipv4)
+{
+ socklen_t vallen;
+ unsigned int mtu;
+ int ret;
+
+ vallen = sizeof(mtu);
+ if (is_ipv4)
+ ret = getsockopt(fd, SOL_IP, IP_MTU, &mtu, &vallen);
+ else
+ ret = getsockopt(fd, SOL_IPV6, IPV6_MTU, &mtu, &vallen);
+
+ if (ret)
+ error(1, errno, "getsockopt mtu");
+
+
+ fprintf(stderr, "path mtu (read): %u\n", mtu);
+ return mtu;
+}
+
+/* very wordy version of system("ip route add dev lo mtu 1500 127.0.0.3/32") */
+static void set_route_mtu(int mtu, bool is_ipv4)
+{
+ struct sockaddr_nl nladdr = { .nl_family = AF_NETLINK };
+ struct nlmsghdr *nh;
+ struct rtattr *rta;
+ struct rtmsg *rt;
+ char data[NLMSG_ALIGN(sizeof(*nh)) +
+ NLMSG_ALIGN(sizeof(*rt)) +
+ NLMSG_ALIGN(RTA_LENGTH(sizeof(addr6))) +
+ NLMSG_ALIGN(RTA_LENGTH(sizeof(int))) +
+ NLMSG_ALIGN(RTA_LENGTH(0) + RTA_LENGTH(sizeof(int)))];
+ int fd, ret, alen, off = 0;
+
+ alen = is_ipv4 ? sizeof(addr4) : sizeof(addr6);
+
+ fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
+ if (fd == -1)
+ error(1, errno, "socket netlink");
+
+ memset(data, 0, sizeof(data));
+
+ nh = (void *)data;
+ nh->nlmsg_type = RTM_NEWROUTE;
+ nh->nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE;
+ off += NLMSG_ALIGN(sizeof(*nh));
+
+ rt = (void *)(data + off);
+ rt->rtm_family = is_ipv4 ? AF_INET : AF_INET6;
+ rt->rtm_table = RT_TABLE_MAIN;
+ rt->rtm_dst_len = alen << 3;
+ rt->rtm_protocol = RTPROT_BOOT;
+ rt->rtm_scope = RT_SCOPE_UNIVERSE;
+ rt->rtm_type = RTN_UNICAST;
+ off += NLMSG_ALIGN(sizeof(*rt));
+
+ rta = (void *)(data + off);
+ rta->rta_type = RTA_DST;
+ rta->rta_len = RTA_LENGTH(alen);
+ if (is_ipv4)
+ memcpy(RTA_DATA(rta), &addr4, alen);
+ else
+ memcpy(RTA_DATA(rta), &addr6, alen);
+ off += NLMSG_ALIGN(rta->rta_len);
+
+ rta = (void *)(data + off);
+ rta->rta_type = RTA_OIF;
+ rta->rta_len = RTA_LENGTH(sizeof(int));
+ *((int *)(RTA_DATA(rta))) = 1; //if_nametoindex("lo");
+ off += NLMSG_ALIGN(rta->rta_len);
+
+ /* MTU is a subtype in a metrics type */
+ rta = (void *)(data + off);
+ rta->rta_type = RTA_METRICS;
+ rta->rta_len = RTA_LENGTH(0) + RTA_LENGTH(sizeof(int));
+ off += NLMSG_ALIGN(rta->rta_len);
+
+ /* now fill MTU subtype. Note that it fits within above rta_len */
+ rta = (void *)(((char *) rta) + RTA_LENGTH(0));
+ rta->rta_type = RTAX_MTU;
+ rta->rta_len = RTA_LENGTH(sizeof(int));
+ *((int *)(RTA_DATA(rta))) = mtu;
+
+ nh->nlmsg_len = off;
+
+ ret = sendto(fd, data, off, 0, (void *)&nladdr, sizeof(nladdr));
+ if (ret != off)
+ error(1, errno, "send netlink: %uB != %uB\n", ret, off);
+
+ if (close(fd))
+ error(1, errno, "close netlink");
+
+ fprintf(stderr, "route mtu (test): %u\n", mtu);
+}
+
static bool send_one(int fd, int len, int gso_len,
struct sockaddr *addr, socklen_t alen)
{
@@ -391,7 +487,7 @@ static void run_all(int fdt, int fdr, struct sockaddr *addr, socklen_t alen)
static void run_test(struct sockaddr *addr, socklen_t alen)
{
struct timeval tv = { .tv_usec = 100 * 1000 };
- int fdr, fdt;
+ int fdr, fdt, val;
fdr = socket(addr->sa_family, SOCK_DGRAM, 0);
if (fdr == -1)
@@ -416,6 +512,20 @@ static void run_test(struct sockaddr *addr, socklen_t alen)
run_all(fdt, fdr, addr, alen);
}
+ if (cfg_do_connected) {
+ set_device_mtu(fdt, CONST_MTU_TEST + 100);
+ set_route_mtu(CONST_MTU_TEST, addr->sa_family == AF_INET);
+
+ if (connect(fdt, addr, alen))
+ error(1, errno, "connect");
+
+ val = get_path_mtu(fdt, addr->sa_family == AF_INET);
+ if (val != CONST_MTU_TEST)
+ error(1, 0, "bad path mtu %u\n", val);
+
+ run_all(fdt, fdr, addr, 0 /* use connected addr */);
+ }
+
if (close(fdt))
error(1, errno, "close t");
if (close(fdr))
@@ -448,7 +558,7 @@ static void parse_opts(int argc, char **argv)
{
int c;
- while ((c = getopt(argc, argv, "46Cst:")) != -1) {
+ while ((c = getopt(argc, argv, "46cCst:")) != -1) {
switch (c) {
case '4':
cfg_do_ipv4 = true;
@@ -456,6 +566,9 @@ static void parse_opts(int argc, char **argv)
case '6':
cfg_do_ipv6 = true;
break;
+ case 'c':
+ cfg_do_connected = true;
+ break;
case 'C':
cfg_do_connectionless = true;
break;
diff --git a/tools/testing/selftests/net/udpgso.sh b/tools/testing/selftests/net/udpgso.sh
index 7977b97e060c..7cdf0e7c1dde 100755
--- a/tools/testing/selftests/net/udpgso.sh
+++ b/tools/testing/selftests/net/udpgso.sh
@@ -14,3 +14,10 @@ echo "ipv6 cmsg"
echo "ipv6 setsockopt"
./in_netns.sh ./udpgso -6 -C -s
+
+echo "ipv4 connected"
+./in_netns.sh ./udpgso -4 -c
+
+# blocked on 2nd loopback address
+# echo "ipv6 connected"
+# ./in_netns.sh ./udpgso -6 -c
--
2.17.0.484.g0c8726318c-goog
^ permalink raw reply related
* [PATCH net-next v2 10/11] selftests: udp gso with corking
From: Willem de Bruijn @ 2018-04-26 17:42 UTC (permalink / raw)
To: netdev; +Cc: davem, alexander.duyck, Willem de Bruijn
In-Reply-To: <20180426174225.246388-1-willemdebruijn.kernel@gmail.com>
From: Willem de Bruijn <willemb@google.com>
Corked sockets take a different path to construct a udp datagram than
the lockless fast path. Test this alternate path.
Signed-off-by: Willem de Bruijn <willemb@google.com>
---
tools/testing/selftests/net/udpgso.c | 42 ++++++++++++++++++++-------
tools/testing/selftests/net/udpgso.sh | 6 ++++
2 files changed, 38 insertions(+), 10 deletions(-)
diff --git a/tools/testing/selftests/net/udpgso.c b/tools/testing/selftests/net/udpgso.c
index a47dca025346..6b1998b9f7df 100644
--- a/tools/testing/selftests/net/udpgso.c
+++ b/tools/testing/selftests/net/udpgso.c
@@ -49,6 +49,7 @@ static bool cfg_do_ipv4;
static bool cfg_do_ipv6;
static bool cfg_do_connected;
static bool cfg_do_connectionless;
+static bool cfg_do_msgmore;
static bool cfg_do_setsockopt;
static int cfg_specific_test_id = -1;
@@ -369,6 +370,23 @@ static void set_route_mtu(int mtu, bool is_ipv4)
fprintf(stderr, "route mtu (test): %u\n", mtu);
}
+static bool __send_one(int fd, struct msghdr *msg, int flags)
+{
+ int ret;
+
+ ret = sendmsg(fd, msg, flags);
+ if (ret == -1 && (errno == EMSGSIZE || errno == ENOMEM))
+ return false;
+ if (ret == -1)
+ error(1, errno, "sendmsg");
+ if (ret != msg->msg_iov->iov_len)
+ error(1, 0, "sendto: %d != %lu", ret, msg->msg_iov->iov_len);
+ if (msg->msg_flags)
+ error(1, 0, "sendmsg: return flags 0x%x\n", msg->msg_flags);
+
+ return true;
+}
+
static bool send_one(int fd, int len, int gso_len,
struct sockaddr *addr, socklen_t alen)
{
@@ -376,7 +394,6 @@ static bool send_one(int fd, int len, int gso_len,
struct msghdr msg = {0};
struct iovec iov = {0};
struct cmsghdr *cm;
- int ret;
iov.iov_base = buf;
iov.iov_len = len;
@@ -398,15 +415,17 @@ static bool send_one(int fd, int len, int gso_len,
*((uint16_t *) CMSG_DATA(cm)) = gso_len;
}
- ret = sendmsg(fd, &msg, 0);
- if (ret == -1 && (errno == EMSGSIZE || errno == ENOMEM))
- return false;
- if (ret == -1)
- error(1, errno, "sendmsg");
- if (ret != len)
- error(1, 0, "sendto: %d != %u", ret, len);
+ /* If MSG_MORE, send 1 byte followed by remainder */
+ if (cfg_do_msgmore && len > 1) {
+ iov.iov_len = 1;
+ if (!__send_one(fd, &msg, MSG_MORE))
+ error(1, 0, "send 1B failed");
- return true;
+ iov.iov_base++;
+ iov.iov_len = len - 1;
+ }
+
+ return __send_one(fd, &msg, 0);
}
static int recv_one(int fd, int flags)
@@ -558,7 +577,7 @@ static void parse_opts(int argc, char **argv)
{
int c;
- while ((c = getopt(argc, argv, "46cCst:")) != -1) {
+ while ((c = getopt(argc, argv, "46cCmst:")) != -1) {
switch (c) {
case '4':
cfg_do_ipv4 = true;
@@ -572,6 +591,9 @@ static void parse_opts(int argc, char **argv)
case 'C':
cfg_do_connectionless = true;
break;
+ case 'm':
+ cfg_do_msgmore = true;
+ break;
case 's':
cfg_do_setsockopt = true;
break;
diff --git a/tools/testing/selftests/net/udpgso.sh b/tools/testing/selftests/net/udpgso.sh
index 7cdf0e7c1dde..fec24f584fe9 100755
--- a/tools/testing/selftests/net/udpgso.sh
+++ b/tools/testing/selftests/net/udpgso.sh
@@ -21,3 +21,9 @@ echo "ipv4 connected"
# blocked on 2nd loopback address
# echo "ipv6 connected"
# ./in_netns.sh ./udpgso -6 -c
+
+echo "ipv4 msg_more"
+./in_netns.sh ./udpgso -4 -C -m
+
+echo "ipv6 msg_more"
+./in_netns.sh ./udpgso -6 -C -m
--
2.17.0.484.g0c8726318c-goog
^ permalink raw reply related
* [PATCH net-next v2 08/11] selftests: udp gso
From: Willem de Bruijn @ 2018-04-26 17:42 UTC (permalink / raw)
To: netdev; +Cc: davem, alexander.duyck, Willem de Bruijn
In-Reply-To: <20180426174225.246388-1-willemdebruijn.kernel@gmail.com>
From: Willem de Bruijn <willemb@google.com>
Validate udp gso, including edge cases (such as min/max gso sizes).
Signed-off-by: Willem de Bruijn <willemb@google.com>
---
tools/testing/selftests/net/.gitignore | 1 +
tools/testing/selftests/net/Makefile | 3 +-
tools/testing/selftests/net/udpgso.c | 486 +++++++++++++++++++++++++
tools/testing/selftests/net/udpgso.sh | 16 +
4 files changed, 505 insertions(+), 1 deletion(-)
create mode 100644 tools/testing/selftests/net/udpgso.c
create mode 100755 tools/testing/selftests/net/udpgso.sh
diff --git a/tools/testing/selftests/net/.gitignore b/tools/testing/selftests/net/.gitignore
index 5559f6add046..67a26240407c 100644
--- a/tools/testing/selftests/net/.gitignore
+++ b/tools/testing/selftests/net/.gitignore
@@ -8,3 +8,4 @@ reuseport_bpf_numa
reuseport_dualstack
reuseaddr_conflict
tcp_mmap
+udpgso
diff --git a/tools/testing/selftests/net/Makefile b/tools/testing/selftests/net/Makefile
index c3761c35f542..db3303348315 100644
--- a/tools/testing/selftests/net/Makefile
+++ b/tools/testing/selftests/net/Makefile
@@ -5,12 +5,13 @@ CFLAGS = -Wall -Wl,--no-as-needed -O2 -g
CFLAGS += -I../../../../usr/include/
TEST_PROGS := run_netsocktests run_afpackettests test_bpf.sh netdevice.sh rtnetlink.sh
-TEST_PROGS += fib_tests.sh fib-onlink-tests.sh in_netns.sh pmtu.sh
+TEST_PROGS += fib_tests.sh fib-onlink-tests.sh in_netns.sh pmtu.sh udpgso.sh
TEST_GEN_FILES = socket
TEST_GEN_FILES += psock_fanout psock_tpacket msg_zerocopy
TEST_GEN_FILES += tcp_mmap
TEST_GEN_PROGS = reuseport_bpf reuseport_bpf_cpu reuseport_bpf_numa
TEST_GEN_PROGS += reuseport_dualstack reuseaddr_conflict
+TEST_GEN_PROGS += udpgso
include ../lib.mk
diff --git a/tools/testing/selftests/net/udpgso.c b/tools/testing/selftests/net/udpgso.c
new file mode 100644
index 000000000000..68a230dfee73
--- /dev/null
+++ b/tools/testing/selftests/net/udpgso.c
@@ -0,0 +1,486 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#define _GNU_SOURCE
+
+#include <stddef.h>
+#include <arpa/inet.h>
+#include <error.h>
+#include <errno.h>
+#include <net/if.h>
+#include <linux/in.h>
+#include <linux/netlink.h>
+#include <linux/rtnetlink.h>
+#include <netinet/if_ether.h>
+#include <netinet/ip.h>
+#include <netinet/ip6.h>
+#include <netinet/udp.h>
+#include <stdbool.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/ioctl.h>
+#include <sys/socket.h>
+#include <sys/stat.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <unistd.h>
+
+#ifndef ETH_MAX_MTU
+#define ETH_MAX_MTU 0xFFFFU
+#endif
+
+#ifndef UDP_SEGMENT
+#define UDP_SEGMENT 103
+#endif
+
+#define CONST_MTU_TEST 1500
+
+#define CONST_HDRLEN_V4 (sizeof(struct iphdr) + sizeof(struct udphdr))
+#define CONST_HDRLEN_V6 (sizeof(struct ip6_hdr) + sizeof(struct udphdr))
+
+#define CONST_MSS_V4 (CONST_MTU_TEST - CONST_HDRLEN_V4)
+#define CONST_MSS_V6 (CONST_MTU_TEST - CONST_HDRLEN_V6)
+
+#define CONST_MAX_SEGS_V4 (ETH_MAX_MTU / CONST_MSS_V4)
+#define CONST_MAX_SEGS_V6 (ETH_MAX_MTU / CONST_MSS_V6)
+
+static bool cfg_do_ipv4;
+static bool cfg_do_ipv6;
+static bool cfg_do_connectionless;
+static bool cfg_do_setsockopt;
+static int cfg_specific_test_id = -1;
+
+static const char cfg_ifname[] = "lo";
+static unsigned short cfg_port = 9000;
+
+static char buf[ETH_MAX_MTU];
+
+struct testcase {
+ int tlen; /* send() buffer size, may exceed mss */
+ bool tfail; /* send() call is expected to fail */
+ int gso_len; /* mss after applying gso */
+ int r_num_mss; /* recv(): number of calls of full mss */
+ int r_len_last; /* recv(): size of last non-mss dgram, if any */
+};
+
+const struct in6_addr addr6 = IN6ADDR_LOOPBACK_INIT;
+const struct in_addr addr4 = { .s_addr = __constant_htonl(INADDR_LOOPBACK + 2) };
+
+struct testcase testcases_v4[] = {
+ {
+ /* no GSO: send a single byte */
+ .tlen = 1,
+ .r_len_last = 1,
+ },
+ {
+ /* no GSO: send a single MSS */
+ .tlen = CONST_MSS_V4,
+ .r_num_mss = 1,
+ },
+ {
+ /* no GSO: send a single MSS + 1B: fail */
+ .tlen = CONST_MSS_V4 + 1,
+ .tfail = true,
+ },
+ {
+ /* send a single MSS: will fail with GSO, because the segment
+ * logic in udp4_ufo_fragment demands a gso skb to be > MTU
+ */
+ .tlen = CONST_MSS_V4,
+ .gso_len = CONST_MSS_V4,
+ .tfail = true,
+ .r_num_mss = 1,
+ },
+ {
+ /* send a single MSS + 1B */
+ .tlen = CONST_MSS_V4 + 1,
+ .gso_len = CONST_MSS_V4,
+ .r_num_mss = 1,
+ .r_len_last = 1,
+ },
+ {
+ /* send exactly 2 MSS */
+ .tlen = CONST_MSS_V4 * 2,
+ .gso_len = CONST_MSS_V4,
+ .r_num_mss = 2,
+ },
+ {
+ /* send 2 MSS + 1B */
+ .tlen = (CONST_MSS_V4 * 2) + 1,
+ .gso_len = CONST_MSS_V4,
+ .r_num_mss = 2,
+ .r_len_last = 1,
+ },
+ {
+ /* send MAX segs */
+ .tlen = (ETH_MAX_MTU / CONST_MSS_V4) * CONST_MSS_V4,
+ .gso_len = CONST_MSS_V4,
+ .r_num_mss = (ETH_MAX_MTU / CONST_MSS_V4),
+ },
+
+ {
+ /* send MAX bytes */
+ .tlen = ETH_MAX_MTU - CONST_HDRLEN_V4,
+ .gso_len = CONST_MSS_V4,
+ .r_num_mss = CONST_MAX_SEGS_V4,
+ .r_len_last = ETH_MAX_MTU - CONST_HDRLEN_V4 -
+ (CONST_MAX_SEGS_V4 * CONST_MSS_V4),
+ },
+ {
+ /* send MAX + 1: fail */
+ .tlen = ETH_MAX_MTU - CONST_HDRLEN_V4 + 1,
+ .gso_len = CONST_MSS_V4,
+ .tfail = true,
+ },
+ {
+ /* EOL */
+ }
+};
+
+#ifndef IP6_MAX_MTU
+#define IP6_MAX_MTU (ETH_MAX_MTU + sizeof(struct ip6_hdr))
+#endif
+
+struct testcase testcases_v6[] = {
+ {
+ /* no GSO: send a single byte */
+ .tlen = 1,
+ .r_len_last = 1,
+ },
+ {
+ /* no GSO: send a single MSS */
+ .tlen = CONST_MSS_V6,
+ .r_num_mss = 1,
+ },
+ {
+ /* no GSO: send a single MSS + 1B: fail */
+ .tlen = CONST_MSS_V6 + 1,
+ .tfail = true,
+ },
+ {
+ /* send a single MSS: will fail with GSO, because the segment
+ * logic in udp4_ufo_fragment demands a gso skb to be > MTU
+ */
+ .tlen = CONST_MSS_V6,
+ .gso_len = CONST_MSS_V6,
+ .tfail = true,
+ .r_num_mss = 1,
+ },
+ {
+ /* send a single MSS + 1B */
+ .tlen = CONST_MSS_V6 + 1,
+ .gso_len = CONST_MSS_V6,
+ .r_num_mss = 1,
+ .r_len_last = 1,
+ },
+ {
+ /* send exactly 2 MSS */
+ .tlen = CONST_MSS_V6 * 2,
+ .gso_len = CONST_MSS_V6,
+ .r_num_mss = 2,
+ },
+ {
+ /* send 2 MSS + 1B */
+ .tlen = (CONST_MSS_V6 * 2) + 1,
+ .gso_len = CONST_MSS_V6,
+ .r_num_mss = 2,
+ .r_len_last = 1,
+ },
+ {
+ /* send MAX segs */
+ .tlen = (IP6_MAX_MTU / CONST_MSS_V6) * CONST_MSS_V6,
+ .gso_len = CONST_MSS_V6,
+ .r_num_mss = (IP6_MAX_MTU / CONST_MSS_V6),
+ },
+
+ {
+ /* send MAX bytes */
+ .tlen = IP6_MAX_MTU - CONST_HDRLEN_V6,
+ .gso_len = CONST_MSS_V6,
+ .r_num_mss = CONST_MAX_SEGS_V6,
+ .r_len_last = IP6_MAX_MTU - CONST_HDRLEN_V6 -
+ (CONST_MAX_SEGS_V6 * CONST_MSS_V6),
+ },
+ {
+ /* send MAX + 1: fail */
+ .tlen = IP6_MAX_MTU - CONST_HDRLEN_V6 + 1,
+ .gso_len = CONST_MSS_V6,
+ .tfail = true,
+ },
+ {
+ /* EOL */
+ }
+};
+
+static unsigned int get_device_mtu(int fd, const char *ifname)
+{
+ struct ifreq ifr;
+
+ memset(&ifr, 0, sizeof(ifr));
+
+ strcpy(ifr.ifr_name, ifname);
+
+ if (ioctl(fd, SIOCGIFMTU, &ifr))
+ error(1, errno, "ioctl get mtu");
+
+ return ifr.ifr_mtu;
+}
+
+static void __set_device_mtu(int fd, const char *ifname, unsigned int mtu)
+{
+ struct ifreq ifr;
+
+ memset(&ifr, 0, sizeof(ifr));
+
+ ifr.ifr_mtu = mtu;
+ strcpy(ifr.ifr_name, ifname);
+
+ if (ioctl(fd, SIOCSIFMTU, &ifr))
+ error(1, errno, "ioctl set mtu");
+}
+
+static void set_device_mtu(int fd, int mtu)
+{
+ int val;
+
+ val = get_device_mtu(fd, cfg_ifname);
+ fprintf(stderr, "device mtu (orig): %u\n", val);
+
+ __set_device_mtu(fd, cfg_ifname, mtu);
+ val = get_device_mtu(fd, cfg_ifname);
+ if (val != mtu)
+ error(1, 0, "unable to set device mtu to %u\n", val);
+
+ fprintf(stderr, "device mtu (test): %u\n", val);
+}
+
+static void set_pmtu_discover(int fd, bool is_ipv4)
+{
+ int level, name, val;
+
+ if (is_ipv4) {
+ level = SOL_IP;
+ name = IP_MTU_DISCOVER;
+ val = IP_PMTUDISC_DO;
+ } else {
+ level = SOL_IPV6;
+ name = IPV6_MTU_DISCOVER;
+ val = IPV6_PMTUDISC_DO;
+ }
+
+ if (setsockopt(fd, level, name, &val, sizeof(val)))
+ error(1, errno, "setsockopt path mtu");
+}
+
+static bool send_one(int fd, int len, int gso_len,
+ struct sockaddr *addr, socklen_t alen)
+{
+ char control[CMSG_SPACE(sizeof(uint16_t))] = {0};
+ struct msghdr msg = {0};
+ struct iovec iov = {0};
+ struct cmsghdr *cm;
+ int ret;
+
+ iov.iov_base = buf;
+ iov.iov_len = len;
+
+ msg.msg_iov = &iov;
+ msg.msg_iovlen = 1;
+
+ msg.msg_name = addr;
+ msg.msg_namelen = alen;
+
+ if (gso_len && !cfg_do_setsockopt) {
+ msg.msg_control = control;
+ msg.msg_controllen = sizeof(control);
+
+ cm = CMSG_FIRSTHDR(&msg);
+ cm->cmsg_level = SOL_UDP;
+ cm->cmsg_type = UDP_SEGMENT;
+ cm->cmsg_len = CMSG_LEN(sizeof(uint16_t));
+ *((uint16_t *) CMSG_DATA(cm)) = gso_len;
+ }
+
+ ret = sendmsg(fd, &msg, 0);
+ if (ret == -1 && (errno == EMSGSIZE || errno == ENOMEM))
+ return false;
+ if (ret == -1)
+ error(1, errno, "sendmsg");
+ if (ret != len)
+ error(1, 0, "sendto: %d != %u", ret, len);
+
+ return true;
+}
+
+static int recv_one(int fd, int flags)
+{
+ int ret;
+
+ ret = recv(fd, buf, sizeof(buf), flags);
+ if (ret == -1 && errno == EAGAIN && (flags & MSG_DONTWAIT))
+ return 0;
+ if (ret == -1)
+ error(1, errno, "recv");
+
+ return ret;
+}
+
+static void run_one(struct testcase *test, int fdt, int fdr,
+ struct sockaddr *addr, socklen_t alen)
+{
+ int i, ret, val, mss;
+ bool sent;
+
+ fprintf(stderr, "ipv%d tx:%d gso:%d %s\n",
+ addr->sa_family == AF_INET ? 4 : 6,
+ test->tlen, test->gso_len,
+ test->tfail ? "(fail)" : "");
+
+ val = test->gso_len;
+ if (cfg_do_setsockopt) {
+ if (setsockopt(fdt, SOL_UDP, UDP_SEGMENT, &val, sizeof(val)))
+ error(1, errno, "setsockopt udp segment");
+ }
+
+ sent = send_one(fdt, test->tlen, test->gso_len, addr, alen);
+ if (sent && test->tfail)
+ error(1, 0, "send succeeded while expecting failure");
+ if (!sent && !test->tfail)
+ error(1, 0, "send failed while expecting success");
+ if (!sent)
+ return;
+
+ mss = addr->sa_family == AF_INET ? CONST_MSS_V4 : CONST_MSS_V6;
+
+ /* Recv all full MSS datagrams */
+ for (i = 0; i < test->r_num_mss; i++) {
+ ret = recv_one(fdr, 0);
+ if (ret != mss)
+ error(1, 0, "recv.%d: %d != %d", i, ret, mss);
+ }
+
+ /* Recv the non-full last datagram, if tlen was not a multiple of mss */
+ if (test->r_len_last) {
+ ret = recv_one(fdr, 0);
+ if (ret != test->r_len_last)
+ error(1, 0, "recv.%d: %d != %d (last)",
+ i, ret, test->r_len_last);
+ }
+
+ /* Verify received all data */
+ ret = recv_one(fdr, MSG_DONTWAIT);
+ if (ret)
+ error(1, 0, "recv: unexpected datagram");
+}
+
+static void run_all(int fdt, int fdr, struct sockaddr *addr, socklen_t alen)
+{
+ struct testcase *tests, *test;
+
+ tests = addr->sa_family == AF_INET ? testcases_v4 : testcases_v6;
+
+ for (test = tests; test->tlen; test++) {
+ /* if a specific test is given, then skip all others */
+ if (cfg_specific_test_id == -1 ||
+ cfg_specific_test_id == test - tests)
+ run_one(test, fdt, fdr, addr, alen);
+ }
+}
+
+static void run_test(struct sockaddr *addr, socklen_t alen)
+{
+ struct timeval tv = { .tv_usec = 100 * 1000 };
+ int fdr, fdt;
+
+ fdr = socket(addr->sa_family, SOCK_DGRAM, 0);
+ if (fdr == -1)
+ error(1, errno, "socket r");
+
+ if (bind(fdr, addr, alen))
+ error(1, errno, "bind");
+
+ /* Have tests fail quickly instead of hang */
+ if (setsockopt(fdr, SOL_SOCKET, SO_RCVTIMEO, &tv, sizeof(tv)))
+ error(1, errno, "setsockopt rcv timeout");
+
+ fdt = socket(addr->sa_family, SOCK_DGRAM, 0);
+ if (fdt == -1)
+ error(1, errno, "socket t");
+
+ /* Do not fragment these datagrams: only succeed if GSO works */
+ set_pmtu_discover(fdt, addr->sa_family == AF_INET);
+
+ if (cfg_do_connectionless) {
+ set_device_mtu(fdt, CONST_MTU_TEST);
+ run_all(fdt, fdr, addr, alen);
+ }
+
+ if (close(fdt))
+ error(1, errno, "close t");
+ if (close(fdr))
+ error(1, errno, "close r");
+}
+
+static void run_test_v4(void)
+{
+ struct sockaddr_in addr = {0};
+
+ addr.sin_family = AF_INET;
+ addr.sin_port = htons(cfg_port);
+ addr.sin_addr = addr4;
+
+ run_test((void *)&addr, sizeof(addr));
+}
+
+static void run_test_v6(void)
+{
+ struct sockaddr_in6 addr = {0};
+
+ addr.sin6_family = AF_INET6;
+ addr.sin6_port = htons(cfg_port);
+ addr.sin6_addr = addr6;
+
+ run_test((void *)&addr, sizeof(addr));
+}
+
+static void parse_opts(int argc, char **argv)
+{
+ int c;
+
+ while ((c = getopt(argc, argv, "46Cst:")) != -1) {
+ switch (c) {
+ case '4':
+ cfg_do_ipv4 = true;
+ break;
+ case '6':
+ cfg_do_ipv6 = true;
+ break;
+ case 'C':
+ cfg_do_connectionless = true;
+ break;
+ case 's':
+ cfg_do_setsockopt = true;
+ break;
+ case 't':
+ cfg_specific_test_id = strtoul(optarg, NULL, 0);
+ break;
+ default:
+ error(1, 0, "%s: parse error", argv[0]);
+ }
+ }
+}
+
+int main(int argc, char **argv)
+{
+ parse_opts(argc, argv);
+
+ if (cfg_do_ipv4)
+ run_test_v4();
+ if (cfg_do_ipv6)
+ run_test_v6();
+
+ fprintf(stderr, "OK\n");
+ return 0;
+}
+
diff --git a/tools/testing/selftests/net/udpgso.sh b/tools/testing/selftests/net/udpgso.sh
new file mode 100755
index 000000000000..7977b97e060c
--- /dev/null
+++ b/tools/testing/selftests/net/udpgso.sh
@@ -0,0 +1,16 @@
+#!/bin/sh
+# SPDX-License-Identifier: GPL-2.0
+#
+# Run a series of udpgso regression tests
+
+echo "ipv4 cmsg"
+./in_netns.sh ./udpgso -4 -C
+
+echo "ipv4 setsockopt"
+./in_netns.sh ./udpgso -4 -C -s
+
+echo "ipv6 cmsg"
+./in_netns.sh ./udpgso -6 -C
+
+echo "ipv6 setsockopt"
+./in_netns.sh ./udpgso -6 -C -s
--
2.17.0.484.g0c8726318c-goog
^ permalink raw reply related
* [PATCH net-next v2 07/11] udp: add gso support to virtual devices
From: Willem de Bruijn @ 2018-04-26 17:42 UTC (permalink / raw)
To: netdev; +Cc: davem, alexander.duyck, Willem de Bruijn
In-Reply-To: <20180426174225.246388-1-willemdebruijn.kernel@gmail.com>
From: Willem de Bruijn <willemb@google.com>
Virtual devices such as tunnels and bonding can handle large packets.
Only segment packets when reaching a physical or loopback device.
Signed-off-by: Willem de Bruijn <willemb@google.com>
---
Documentation/networking/netdev-features.txt | 7 +++++++
include/linux/netdev_features.h | 5 ++++-
include/linux/netdevice.h | 1 +
net/core/ethtool.c | 1 +
4 files changed, 13 insertions(+), 1 deletion(-)
diff --git a/Documentation/networking/netdev-features.txt b/Documentation/networking/netdev-features.txt
index c77f9d57eb91..c4a54c162547 100644
--- a/Documentation/networking/netdev-features.txt
+++ b/Documentation/networking/netdev-features.txt
@@ -113,6 +113,13 @@ whatever headers there might be.
NETIF_F_TSO_ECN means that hardware can properly split packets with CWR bit
set, be it TCPv4 (when NETIF_F_TSO is enabled) or TCPv6 (NETIF_F_TSO6).
+ * Transmit UDP segmentation offload
+
+NETIF_F_GSO_UDP_GSO_L4 accepts a single UDP header with a payload that exceeds
+gso_size. On segmentation, it segments the payload on gso_size boundaries and
+replicates the network and UDP headers (fixing up the last one if less than
+gso_size).
+
* Transmit DMA from high memory
On platforms where this is relevant, NETIF_F_HIGHDMA signals that
diff --git a/include/linux/netdev_features.h b/include/linux/netdev_features.h
index 35b79f47a13d..fe2f3b30960e 100644
--- a/include/linux/netdev_features.h
+++ b/include/linux/netdev_features.h
@@ -55,8 +55,9 @@ enum {
NETIF_F_GSO_SCTP_BIT, /* ... SCTP fragmentation */
NETIF_F_GSO_ESP_BIT, /* ... ESP with TSO */
NETIF_F_GSO_UDP_BIT, /* ... UFO, deprecated except tuntap */
+ NETIF_F_GSO_UDP_L4_BIT, /* ... UDP payload GSO (not UFO) */
/**/NETIF_F_GSO_LAST = /* last bit, see GSO_MASK */
- NETIF_F_GSO_UDP_BIT,
+ NETIF_F_GSO_UDP_L4_BIT,
NETIF_F_FCOE_CRC_BIT, /* FCoE CRC32 */
NETIF_F_SCTP_CRC_BIT, /* SCTP checksum offload */
@@ -147,6 +148,7 @@ enum {
#define NETIF_F_HW_ESP_TX_CSUM __NETIF_F(HW_ESP_TX_CSUM)
#define NETIF_F_RX_UDP_TUNNEL_PORT __NETIF_F(RX_UDP_TUNNEL_PORT)
#define NETIF_F_HW_TLS_RECORD __NETIF_F(HW_TLS_RECORD)
+#define NETIF_F_GSO_UDP_L4 __NETIF_F(GSO_UDP_L4)
#define for_each_netdev_feature(mask_addr, bit) \
for_each_set_bit(bit, (unsigned long *)mask_addr, NETDEV_FEATURE_COUNT)
@@ -216,6 +218,7 @@ enum {
NETIF_F_GSO_GRE_CSUM | \
NETIF_F_GSO_IPXIP4 | \
NETIF_F_GSO_IPXIP6 | \
+ NETIF_F_GSO_UDP_L4 | \
NETIF_F_GSO_UDP_TUNNEL | \
NETIF_F_GSO_UDP_TUNNEL_CSUM)
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 14e0777ffcfb..366c32891158 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -4186,6 +4186,7 @@ static inline bool net_gso_ok(netdev_features_t features, int gso_type)
BUILD_BUG_ON(SKB_GSO_SCTP != (NETIF_F_GSO_SCTP >> NETIF_F_GSO_SHIFT));
BUILD_BUG_ON(SKB_GSO_ESP != (NETIF_F_GSO_ESP >> NETIF_F_GSO_SHIFT));
BUILD_BUG_ON(SKB_GSO_UDP != (NETIF_F_GSO_UDP >> NETIF_F_GSO_SHIFT));
+ BUILD_BUG_ON(SKB_GSO_UDP_L4 != (NETIF_F_GSO_UDP_L4 >> NETIF_F_GSO_SHIFT));
return (features & feature) == feature;
}
diff --git a/net/core/ethtool.c b/net/core/ethtool.c
index 03416e6dd5d7..4650fd6d678c 100644
--- a/net/core/ethtool.c
+++ b/net/core/ethtool.c
@@ -92,6 +92,7 @@ static const char netdev_features_strings[NETDEV_FEATURE_COUNT][ETH_GSTRING_LEN]
[NETIF_F_GSO_PARTIAL_BIT] = "tx-gso-partial",
[NETIF_F_GSO_SCTP_BIT] = "tx-sctp-segmentation",
[NETIF_F_GSO_ESP_BIT] = "tx-esp-segmentation",
+ [NETIF_F_GSO_UDP_L4_BIT] = "tx-udp-segmentation",
[NETIF_F_FCOE_CRC_BIT] = "tx-checksum-fcoe-crc",
[NETIF_F_SCTP_CRC_BIT] = "tx-checksum-sctp",
--
2.17.0.484.g0c8726318c-goog
^ permalink raw reply related
* [PATCH net-next v2 06/11] udp: add gso segment cmsg
From: Willem de Bruijn @ 2018-04-26 17:42 UTC (permalink / raw)
To: netdev; +Cc: davem, alexander.duyck, Willem de Bruijn
In-Reply-To: <20180426174225.246388-1-willemdebruijn.kernel@gmail.com>
From: Willem de Bruijn <willemb@google.com>
Allow specifying segment size in the send call.
The new control message performs the same function as socket option
UDP_SEGMENT while avoiding the extra system call.
Signed-off-by: Willem de Bruijn <willemb@google.com>
---
include/net/udp.h | 1 +
net/ipv4/udp.c | 43 +++++++++++++++++++++++++++++++++++++++++--
net/ipv6/udp.c | 5 ++++-
3 files changed, 46 insertions(+), 3 deletions(-)
diff --git a/include/net/udp.h b/include/net/udp.h
index 741d888d0fdb..05990746810e 100644
--- a/include/net/udp.h
+++ b/include/net/udp.h
@@ -273,6 +273,7 @@ int udp_abort(struct sock *sk, int err);
int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len);
int udp_push_pending_frames(struct sock *sk);
void udp_flush_pending_frames(struct sock *sk);
+int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size);
void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst);
int udp_rcv(struct sk_buff *skb);
int udp_ioctl(struct sock *sk, int cmd, unsigned long arg);
diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
index bda022c5480b..350bdafc462b 100644
--- a/net/ipv4/udp.c
+++ b/net/ipv4/udp.c
@@ -853,6 +853,42 @@ int udp_push_pending_frames(struct sock *sk)
}
EXPORT_SYMBOL(udp_push_pending_frames);
+static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
+{
+ switch (cmsg->cmsg_type) {
+ case UDP_SEGMENT:
+ if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
+ return -EINVAL;
+ *gso_size = *(__u16 *)CMSG_DATA(cmsg);
+ return 0;
+ default:
+ return -EINVAL;
+ }
+}
+
+int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
+{
+ struct cmsghdr *cmsg;
+ bool need_ip = false;
+ int err;
+
+ for_each_cmsghdr(cmsg, msg) {
+ if (!CMSG_OK(msg, cmsg))
+ return -EINVAL;
+
+ if (cmsg->cmsg_level != SOL_UDP) {
+ need_ip = true;
+ continue;
+ }
+
+ err = __udp_cmsg_send(cmsg, gso_size);
+ if (err)
+ return err;
+ }
+
+ return need_ip;
+}
+
int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
{
struct inet_sock *inet = inet_sk(sk);
@@ -941,8 +977,11 @@ int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
ipc.gso_size = up->gso_size;
if (msg->msg_controllen) {
- err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6);
- if (unlikely(err)) {
+ err = udp_cmsg_send(sk, msg, &ipc.gso_size);
+ if (err > 0)
+ err = ip_cmsg_send(sk, msg, &ipc,
+ sk->sk_family == AF_INET6);
+ if (unlikely(err < 0)) {
kfree(ipc.opt);
return err;
}
diff --git a/net/ipv6/udp.c b/net/ipv6/udp.c
index 86b7dd58d4b4..6acfdd3e442b 100644
--- a/net/ipv6/udp.c
+++ b/net/ipv6/udp.c
@@ -1276,7 +1276,10 @@ int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
opt->tot_len = sizeof(*opt);
ipc6.opt = opt;
- err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, &ipc6, &sockc);
+ err = udp_cmsg_send(sk, msg, &ipc6.gso_size);
+ if (err > 0)
+ err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6,
+ &ipc6, &sockc);
if (err < 0) {
fl6_sock_release(flowlabel);
return err;
--
2.17.0.484.g0c8726318c-goog
^ permalink raw reply related
* [PATCH net-next v2 05/11] udp: paged allocation with gso
From: Willem de Bruijn @ 2018-04-26 17:42 UTC (permalink / raw)
To: netdev; +Cc: davem, alexander.duyck, Willem de Bruijn
In-Reply-To: <20180426174225.246388-1-willemdebruijn.kernel@gmail.com>
From: Willem de Bruijn <willemb@google.com>
When sending large datagrams that are later segmented, store data in
page frags to avoid copying from linear in skb_segment.
Signed-off-by: Willem de Bruijn <willemb@google.com>
---
net/ipv4/ip_output.c | 15 +++++++++++----
net/ipv6/ip6_output.c | 19 ++++++++++++++-----
2 files changed, 25 insertions(+), 9 deletions(-)
diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c
index da4abbee10f7..f2338e40c37d 100644
--- a/net/ipv4/ip_output.c
+++ b/net/ipv4/ip_output.c
@@ -878,11 +878,13 @@ static int __ip_append_data(struct sock *sk,
struct rtable *rt = (struct rtable *)cork->dst;
unsigned int wmem_alloc_delta = 0;
u32 tskey = 0;
+ bool paged;
skb = skb_peek_tail(queue);
exthdrlen = !skb ? rt->dst.header_len : 0;
mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
+ paged = !!cork->gso_size;
if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
@@ -934,6 +936,7 @@ static int __ip_append_data(struct sock *sk,
unsigned int fraglen;
unsigned int fraggap;
unsigned int alloclen;
+ unsigned int pagedlen = 0;
struct sk_buff *skb_prev;
alloc_new_skb:
skb_prev = skb;
@@ -954,8 +957,12 @@ static int __ip_append_data(struct sock *sk,
if ((flags & MSG_MORE) &&
!(rt->dst.dev->features&NETIF_F_SG))
alloclen = mtu;
- else
+ else if (!paged)
alloclen = fraglen;
+ else {
+ alloclen = min_t(int, fraglen, MAX_HEADER);
+ pagedlen = fraglen - alloclen;
+ }
alloclen += exthdrlen;
@@ -999,7 +1006,7 @@ static int __ip_append_data(struct sock *sk,
/*
* Find where to start putting bytes.
*/
- data = skb_put(skb, fraglen + exthdrlen);
+ data = skb_put(skb, fraglen + exthdrlen - pagedlen);
skb_set_network_header(skb, exthdrlen);
skb->transport_header = (skb->network_header +
fragheaderlen);
@@ -1015,7 +1022,7 @@ static int __ip_append_data(struct sock *sk,
pskb_trim_unique(skb_prev, maxfraglen);
}
- copy = datalen - transhdrlen - fraggap;
+ copy = datalen - transhdrlen - fraggap - pagedlen;
if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
err = -EFAULT;
kfree_skb(skb);
@@ -1023,7 +1030,7 @@ static int __ip_append_data(struct sock *sk,
}
offset += copy;
- length -= datalen - fraggap;
+ length -= copy + transhdrlen;
transhdrlen = 0;
exthdrlen = 0;
csummode = CHECKSUM_NONE;
diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c
index a1c4a78132d2..dfd8af41824e 100644
--- a/net/ipv6/ip6_output.c
+++ b/net/ipv6/ip6_output.c
@@ -1276,6 +1276,7 @@ static int __ip6_append_data(struct sock *sk,
int csummode = CHECKSUM_NONE;
unsigned int maxnonfragsize, headersize;
unsigned int wmem_alloc_delta = 0;
+ bool paged;
skb = skb_peek_tail(queue);
if (!skb) {
@@ -1283,6 +1284,7 @@ static int __ip6_append_data(struct sock *sk,
dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
}
+ paged = !!cork->gso_size;
mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize;
orig_mtu = mtu;
@@ -1374,6 +1376,7 @@ static int __ip6_append_data(struct sock *sk,
unsigned int fraglen;
unsigned int fraggap;
unsigned int alloclen;
+ unsigned int pagedlen = 0;
alloc_new_skb:
/* There's no room in the current skb */
if (skb)
@@ -1396,11 +1399,17 @@ static int __ip6_append_data(struct sock *sk,
if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
+ fraglen = datalen + fragheaderlen;
+
if ((flags & MSG_MORE) &&
!(rt->dst.dev->features&NETIF_F_SG))
alloclen = mtu;
- else
- alloclen = datalen + fragheaderlen;
+ else if (!paged)
+ alloclen = fraglen;
+ else {
+ alloclen = min_t(int, fraglen, MAX_HEADER);
+ pagedlen = fraglen - alloclen;
+ }
alloclen += dst_exthdrlen;
@@ -1422,7 +1431,7 @@ static int __ip6_append_data(struct sock *sk,
*/
alloclen += sizeof(struct frag_hdr);
- copy = datalen - transhdrlen - fraggap;
+ copy = datalen - transhdrlen - fraggap - pagedlen;
if (copy < 0) {
err = -EINVAL;
goto error;
@@ -1461,7 +1470,7 @@ static int __ip6_append_data(struct sock *sk,
/*
* Find where to start putting bytes
*/
- data = skb_put(skb, fraglen);
+ data = skb_put(skb, fraglen - pagedlen);
skb_set_network_header(skb, exthdrlen);
data += fragheaderlen;
skb->transport_header = (skb->network_header +
@@ -1484,7 +1493,7 @@ static int __ip6_append_data(struct sock *sk,
}
offset += copy;
- length -= datalen - fraggap;
+ length -= copy + transhdrlen;
transhdrlen = 0;
exthdrlen = 0;
dst_exthdrlen = 0;
--
2.17.0.484.g0c8726318c-goog
^ permalink raw reply related
* [PATCH net-next v2 04/11] udp: better wmem accounting on gso
From: Willem de Bruijn @ 2018-04-26 17:42 UTC (permalink / raw)
To: netdev; +Cc: davem, alexander.duyck, Willem de Bruijn
In-Reply-To: <20180426174225.246388-1-willemdebruijn.kernel@gmail.com>
From: Willem de Bruijn <willemb@google.com>
skb_segment by default transfers allocated wmem from the gso skb
to the tail of the segment list. This underreports real truesize
of the list, especially if the tail might be dropped.
Similar to tcp_gso_segment, update wmem_alloc with the aggregate
list truesize and make each segment responsible for its own
share by setting skb->destructor.
Clear gso_skb->destructor prior to calling skb_segment to skip
the default assignment to tail.
Signed-off-by: Willem de Bruijn <willemb@google.com>
---
net/ipv4/udp_offload.c | 16 +++++++++++++++-
1 file changed, 15 insertions(+), 1 deletion(-)
diff --git a/net/ipv4/udp_offload.c b/net/ipv4/udp_offload.c
index 6b09220ca407..d806529b9b3d 100644
--- a/net/ipv4/udp_offload.c
+++ b/net/ipv4/udp_offload.c
@@ -191,6 +191,8 @@ struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
netdev_features_t features,
unsigned int mss, __sum16 check)
{
+ struct sock *sk = gso_skb->sk;
+ unsigned int sum_truesize = 0;
struct sk_buff *segs, *seg;
unsigned int hdrlen;
struct udphdr *uh;
@@ -201,9 +203,15 @@ struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
hdrlen = gso_skb->data - skb_mac_header(gso_skb);
skb_pull(gso_skb, sizeof(*uh));
+ /* clear destructor to avoid skb_segment assigning it to tail */
+ WARN_ON_ONCE(gso_skb->destructor != sock_wfree);
+ gso_skb->destructor = NULL;
+
segs = skb_segment(gso_skb, features);
- if (unlikely(IS_ERR_OR_NULL(segs)))
+ if (unlikely(IS_ERR_OR_NULL(segs))) {
+ gso_skb->destructor = sock_wfree;
return segs;
+ }
for (seg = segs; seg; seg = seg->next) {
uh = udp_hdr(seg);
@@ -214,8 +222,14 @@ struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
if (!seg->next)
csum_replace2(&uh->check, htons(mss),
htons(seg->len - hdrlen - sizeof(*uh)));
+
+ seg->destructor = sock_wfree;
+ seg->sk = sk;
+ sum_truesize += seg->truesize;
}
+ refcount_add(sum_truesize - gso_skb->truesize, &sk->sk_wmem_alloc);
+
return segs;
}
--
2.17.0.484.g0c8726318c-goog
^ permalink raw reply related
* [PATCH net-next v2 03/11] udp: generate gso with UDP_SEGMENT
From: Willem de Bruijn @ 2018-04-26 17:42 UTC (permalink / raw)
To: netdev; +Cc: davem, alexander.duyck, Willem de Bruijn
In-Reply-To: <20180426174225.246388-1-willemdebruijn.kernel@gmail.com>
From: Willem de Bruijn <willemb@google.com>
Support generic segmentation offload for udp datagrams. Callers can
concatenate and send at once the payload of multiple datagrams with
the same destination.
To set segment size, the caller sets socket option UDP_SEGMENT to the
length of each discrete payload. This value must be smaller than or
equal to the relevant MTU.
A follow-up patch adds cmsg UDP_SEGMENT to specify segment size on a
per send call basis.
Total byte length may then exceed MTU. If not an exact multiple of
segment size, the last segment will be shorter.
The implementation adds a gso_size field to the udp socket, ip(v6)
cmsg cookie and inet_cork structure to be able to set the value at
setsockopt or cmsg time and to work with both lockless and corked
paths.
Initial benchmark numbers show UDP GSO about as expensive as TCP GSO.
tcp tso
3197 MB/s 54232 msg/s 54232 calls/s
6,457,754,262 cycles
tcp gso
1765 MB/s 29939 msg/s 29939 calls/s
11,203,021,806 cycles
tcp without tso/gso *
739 MB/s 12548 msg/s 12548 calls/s
11,205,483,630 cycles
udp
876 MB/s 14873 msg/s 624666 calls/s
11,205,777,429 cycles
udp gso
2139 MB/s 36282 msg/s 36282 calls/s
11,204,374,561 cycles
[*] after reverting commit 0a6b2a1dc2a2
("tcp: switch to GSO being always on")
Measured total system cycles ('-a') for one core while pinning both
the network receive path and benchmark process to that core:
perf stat -a -C 12 -e cycles \
./udpgso_bench_tx -C 12 -4 -D "$DST" -l 4
Note the reduction in calls/s with GSO. Bytes per syscall drops
increases from 1470 to 61818.
Signed-off-by: Willem de Bruijn <willemb@google.com>
---
include/linux/udp.h | 3 +++
include/net/inet_sock.h | 1 +
include/net/ip.h | 1 +
include/net/ipv6.h | 1 +
include/uapi/linux/udp.h | 1 +
net/ipv4/ip_output.c | 9 ++++++---
net/ipv4/udp.c | 33 ++++++++++++++++++++++++++++++---
net/ipv6/ip6_output.c | 6 ++++--
net/ipv6/udp.c | 23 ++++++++++++++++++++---
9 files changed, 67 insertions(+), 11 deletions(-)
diff --git a/include/linux/udp.h b/include/linux/udp.h
index eaea63bc79bb..ca840345571b 100644
--- a/include/linux/udp.h
+++ b/include/linux/udp.h
@@ -55,6 +55,7 @@ struct udp_sock {
* when the socket is uncorked.
*/
__u16 len; /* total length of pending frames */
+ __u16 gso_size;
/*
* Fields specific to UDP-Lite.
*/
@@ -87,6 +88,8 @@ struct udp_sock {
int forward_deficit;
};
+#define UDP_MAX_SEGMENTS (1 << 6UL)
+
static inline struct udp_sock *udp_sk(const struct sock *sk)
{
return (struct udp_sock *)sk;
diff --git a/include/net/inet_sock.h b/include/net/inet_sock.h
index 0a671c32d6b9..83d5b3c2ac42 100644
--- a/include/net/inet_sock.h
+++ b/include/net/inet_sock.h
@@ -147,6 +147,7 @@ struct inet_cork {
__u8 ttl;
__s16 tos;
char priority;
+ __u16 gso_size;
};
struct inet_cork_full {
diff --git a/include/net/ip.h b/include/net/ip.h
index 7ec543a64bbc..bada1f1f871e 100644
--- a/include/net/ip.h
+++ b/include/net/ip.h
@@ -76,6 +76,7 @@ struct ipcm_cookie {
__u8 ttl;
__s16 tos;
char priority;
+ __u16 gso_size;
};
#define IPCB(skb) ((struct inet_skb_parm*)((skb)->cb))
diff --git a/include/net/ipv6.h b/include/net/ipv6.h
index 0dd722cab037..0a872a7c33c8 100644
--- a/include/net/ipv6.h
+++ b/include/net/ipv6.h
@@ -298,6 +298,7 @@ struct ipcm6_cookie {
__s16 tclass;
__s8 dontfrag;
struct ipv6_txoptions *opt;
+ __u16 gso_size;
};
static inline struct ipv6_txoptions *txopt_get(const struct ipv6_pinfo *np)
diff --git a/include/uapi/linux/udp.h b/include/uapi/linux/udp.h
index efb7b5991c2f..09d00f8c442b 100644
--- a/include/uapi/linux/udp.h
+++ b/include/uapi/linux/udp.h
@@ -32,6 +32,7 @@ struct udphdr {
#define UDP_ENCAP 100 /* Set the socket to accept encapsulated packets */
#define UDP_NO_CHECK6_TX 101 /* Disable sending checksum for UDP6X */
#define UDP_NO_CHECK6_RX 102 /* Disable accpeting checksum for UDP6 */
+#define UDP_SEGMENT 103 /* Set GSO segmentation size */
/* UDP encapsulation types */
#define UDP_ENCAP_ESPINUDP_NON_IKE 1 /* draft-ietf-ipsec-nat-t-ike-00/01 */
diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c
index 2883ff1e909c..da4abbee10f7 100644
--- a/net/ipv4/ip_output.c
+++ b/net/ipv4/ip_output.c
@@ -882,7 +882,8 @@ static int __ip_append_data(struct sock *sk,
skb = skb_peek_tail(queue);
exthdrlen = !skb ? rt->dst.header_len : 0;
- mtu = cork->fragsize;
+ mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
+
if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
tskey = sk->sk_tskey++;
@@ -906,7 +907,7 @@ static int __ip_append_data(struct sock *sk,
if (transhdrlen &&
length + fragheaderlen <= mtu &&
rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) &&
- !(flags & MSG_MORE) &&
+ (!(flags & MSG_MORE) || cork->gso_size) &&
!exthdrlen)
csummode = CHECKSUM_PARTIAL;
@@ -1135,6 +1136,8 @@ static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
*rtp = NULL;
cork->fragsize = ip_sk_use_pmtu(sk) ?
dst_mtu(&rt->dst) : rt->dst.dev->mtu;
+
+ cork->gso_size = sk->sk_type == SOCK_DGRAM ? ipc->gso_size : 0;
cork->dst = &rt->dst;
cork->length = 0;
cork->ttl = ipc->ttl;
@@ -1214,7 +1217,7 @@ ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
return -EOPNOTSUPP;
hh_len = LL_RESERVED_SPACE(rt->dst.dev);
- mtu = cork->fragsize;
+ mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
index 6b9d8017b319..bda022c5480b 100644
--- a/net/ipv4/udp.c
+++ b/net/ipv4/udp.c
@@ -757,7 +757,8 @@ void udp_set_csum(bool nocheck, struct sk_buff *skb,
}
EXPORT_SYMBOL(udp_set_csum);
-static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
+static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
+ struct inet_cork *cork)
{
struct sock *sk = skb->sk;
struct inet_sock *inet = inet_sk(sk);
@@ -777,6 +778,21 @@ static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
uh->len = htons(len);
uh->check = 0;
+ if (cork->gso_size) {
+ const int hlen = skb_network_header_len(skb) +
+ sizeof(struct udphdr);
+
+ if (hlen + cork->gso_size > cork->fragsize)
+ return -EINVAL;
+ if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS)
+ return -EINVAL;
+ if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite)
+ return -EIO;
+
+ skb_shinfo(skb)->gso_size = cork->gso_size;
+ skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
+ }
+
if (is_udplite) /* UDP-Lite */
csum = udplite_csum(skb);
@@ -828,7 +844,7 @@ int udp_push_pending_frames(struct sock *sk)
if (!skb)
goto out;
- err = udp_send_skb(skb, fl4);
+ err = udp_send_skb(skb, fl4, &inet->cork.base);
out:
up->len = 0;
@@ -922,6 +938,7 @@ int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
ipc.sockc.tsflags = sk->sk_tsflags;
ipc.addr = inet->inet_saddr;
ipc.oif = sk->sk_bound_dev_if;
+ ipc.gso_size = up->gso_size;
if (msg->msg_controllen) {
err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6);
@@ -1037,7 +1054,7 @@ int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
&cork, msg->msg_flags);
err = PTR_ERR(skb);
if (!IS_ERR_OR_NULL(skb))
- err = udp_send_skb(skb, fl4);
+ err = udp_send_skb(skb, fl4, &cork);
goto out;
}
@@ -2367,6 +2384,12 @@ int udp_lib_setsockopt(struct sock *sk, int level, int optname,
up->no_check6_rx = valbool;
break;
+ case UDP_SEGMENT:
+ if (val < 0 || val > USHRT_MAX)
+ return -EINVAL;
+ up->gso_size = val;
+ break;
+
/*
* UDP-Lite's partial checksum coverage (RFC 3828).
*/
@@ -2457,6 +2480,10 @@ int udp_lib_getsockopt(struct sock *sk, int level, int optname,
val = up->no_check6_rx;
break;
+ case UDP_SEGMENT:
+ val = up->gso_size;
+ break;
+
/* The following two cannot be changed on UDP sockets, the return is
* always 0 (which corresponds to the full checksum coverage of UDP). */
case UDPLITE_SEND_CSCOV:
diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c
index 7fa1db447405..a1c4a78132d2 100644
--- a/net/ipv6/ip6_output.c
+++ b/net/ipv6/ip6_output.c
@@ -1240,6 +1240,8 @@ static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
if (mtu < IPV6_MIN_MTU)
return -EINVAL;
cork->base.fragsize = mtu;
+ cork->base.gso_size = sk->sk_type == SOCK_DGRAM ? ipc6->gso_size : 0;
+
if (dst_allfrag(xfrm_dst_path(&rt->dst)))
cork->base.flags |= IPCORK_ALLFRAG;
cork->base.length = 0;
@@ -1281,7 +1283,7 @@ static int __ip6_append_data(struct sock *sk,
dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
}
- mtu = cork->fragsize;
+ mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize;
orig_mtu = mtu;
hh_len = LL_RESERVED_SPACE(rt->dst.dev);
@@ -1329,7 +1331,7 @@ static int __ip6_append_data(struct sock *sk,
if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
headersize == sizeof(struct ipv6hdr) &&
length <= mtu - headersize &&
- !(flags & MSG_MORE) &&
+ (!(flags & MSG_MORE) || cork->gso_size) &&
rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
csummode = CHECKSUM_PARTIAL;
diff --git a/net/ipv6/udp.c b/net/ipv6/udp.c
index 824797f8d1ab..86b7dd58d4b4 100644
--- a/net/ipv6/udp.c
+++ b/net/ipv6/udp.c
@@ -1023,7 +1023,8 @@ static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
* Sending
*/
-static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6)
+static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6,
+ struct inet_cork *cork)
{
struct sock *sk = skb->sk;
struct udphdr *uh;
@@ -1042,6 +1043,21 @@ static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6)
uh->len = htons(len);
uh->check = 0;
+ if (cork->gso_size) {
+ const int hlen = skb_network_header_len(skb) +
+ sizeof(struct udphdr);
+
+ if (hlen + cork->gso_size > cork->fragsize)
+ return -EINVAL;
+ if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS)
+ return -EINVAL;
+ if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite)
+ return -EIO;
+
+ skb_shinfo(skb)->gso_size = cork->gso_size;
+ skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
+ }
+
if (is_udplite)
csum = udplite_csum(skb);
else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
@@ -1093,7 +1109,7 @@ static int udp_v6_push_pending_frames(struct sock *sk)
if (!skb)
goto out;
- err = udp_v6_send_skb(skb, &fl6);
+ err = udp_v6_send_skb(skb, &fl6, &inet_sk(sk)->cork.base);
out:
up->len = 0;
@@ -1127,6 +1143,7 @@ int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
ipc6.hlimit = -1;
ipc6.tclass = -1;
ipc6.dontfrag = -1;
+ ipc6.gso_size = up->gso_size;
sockc.tsflags = sk->sk_tsflags;
/* destination address check */
@@ -1333,7 +1350,7 @@ int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
msg->msg_flags, &cork, &sockc);
err = PTR_ERR(skb);
if (!IS_ERR_OR_NULL(skb))
- err = udp_v6_send_skb(skb, &fl6);
+ err = udp_v6_send_skb(skb, &fl6, &cork.base);
goto out;
}
--
2.17.0.484.g0c8726318c-goog
^ permalink raw reply related
* [PATCH net-next v2 02/11] udp: add udp gso
From: Willem de Bruijn @ 2018-04-26 17:42 UTC (permalink / raw)
To: netdev; +Cc: davem, alexander.duyck, Willem de Bruijn
In-Reply-To: <20180426174225.246388-1-willemdebruijn.kernel@gmail.com>
From: Willem de Bruijn <willemb@google.com>
Implement generic segmentation offload support for udp datagrams. A
follow-up patch adds support to the protocol stack to generate such
packets.
UDP GSO is not UFO. UFO fragments a single large datagram. GSO splits
a large payload into a number of discrete UDP datagrams.
The implementation adds a GSO type SKB_UDP_GSO_L4 to differentiate it
from UFO (SKB_UDP_GSO).
IPPROTO_UDPLITE is excluded, as that protocol has no gso handler
registered.
Signed-off-by: Willem de Bruijn <willemb@google.com>
---
include/linux/skbuff.h | 2 ++
include/net/udp.h | 4 ++++
net/core/skbuff.c | 2 ++
net/ipv4/udp_offload.c | 52 +++++++++++++++++++++++++++++++++++++++++-
net/ipv6/ip6_offload.c | 6 +++--
net/ipv6/udp_offload.c | 19 ++++++++++++++-
6 files changed, 81 insertions(+), 4 deletions(-)
diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index d274059529eb..a4a5c0c5cba8 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -573,6 +573,8 @@ enum {
SKB_GSO_ESP = 1 << 15,
SKB_GSO_UDP = 1 << 16,
+
+ SKB_GSO_UDP_L4 = 1 << 17,
};
#if BITS_PER_LONG > 32
diff --git a/include/net/udp.h b/include/net/udp.h
index 0676b272f6ac..741d888d0fdb 100644
--- a/include/net/udp.h
+++ b/include/net/udp.h
@@ -174,6 +174,10 @@ struct sk_buff **udp_gro_receive(struct sk_buff **head, struct sk_buff *skb,
struct udphdr *uh, udp_lookup_t lookup);
int udp_gro_complete(struct sk_buff *skb, int nhoff, udp_lookup_t lookup);
+struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
+ netdev_features_t features,
+ unsigned int mss, __sum16 check);
+
static inline struct udphdr *udp_gro_udphdr(struct sk_buff *skb)
{
struct udphdr *uh;
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index ff49e352deea..c647cfe114e0 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -4940,6 +4940,8 @@ static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
thlen = tcp_hdrlen(skb);
} else if (unlikely(skb_is_gso_sctp(skb))) {
thlen = sizeof(struct sctphdr);
+ } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
+ thlen = sizeof(struct udphdr);
}
/* UFO sets gso_size to the size of the fragmentation
* payload, i.e. the size of the L4 (UDP) header is already
diff --git a/net/ipv4/udp_offload.c b/net/ipv4/udp_offload.c
index ea6e6e7df0ee..6b09220ca407 100644
--- a/net/ipv4/udp_offload.c
+++ b/net/ipv4/udp_offload.c
@@ -187,6 +187,53 @@ struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
}
EXPORT_SYMBOL(skb_udp_tunnel_segment);
+struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
+ netdev_features_t features,
+ unsigned int mss, __sum16 check)
+{
+ struct sk_buff *segs, *seg;
+ unsigned int hdrlen;
+ struct udphdr *uh;
+
+ if (gso_skb->len <= sizeof(*uh) + mss)
+ return ERR_PTR(-EINVAL);
+
+ hdrlen = gso_skb->data - skb_mac_header(gso_skb);
+ skb_pull(gso_skb, sizeof(*uh));
+
+ segs = skb_segment(gso_skb, features);
+ if (unlikely(IS_ERR_OR_NULL(segs)))
+ return segs;
+
+ for (seg = segs; seg; seg = seg->next) {
+ uh = udp_hdr(seg);
+ uh->len = htons(seg->len - hdrlen);
+ uh->check = check;
+
+ /* last packet can be partial gso_size */
+ if (!seg->next)
+ csum_replace2(&uh->check, htons(mss),
+ htons(seg->len - hdrlen - sizeof(*uh)));
+ }
+
+ return segs;
+}
+
+static struct sk_buff *__udp4_gso_segment(struct sk_buff *gso_skb,
+ netdev_features_t features)
+{
+ const struct iphdr *iph = ip_hdr(gso_skb);
+ unsigned int mss = skb_shinfo(gso_skb)->gso_size;
+
+ if (!can_checksum_protocol(features, htons(ETH_P_IP)))
+ return ERR_PTR(-EIO);
+
+ return __udp_gso_segment(gso_skb, features, mss,
+ udp_v4_check(sizeof(struct udphdr) + mss,
+ iph->saddr, iph->daddr, 0));
+}
+EXPORT_SYMBOL_GPL(__udp4_gso_segment);
+
static struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
netdev_features_t features)
{
@@ -203,12 +250,15 @@ static struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
goto out;
}
- if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP))
+ if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_UDP | SKB_GSO_UDP_L4)))
goto out;
if (!pskb_may_pull(skb, sizeof(struct udphdr)))
goto out;
+ if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4)
+ return __udp4_gso_segment(skb, features);
+
mss = skb_shinfo(skb)->gso_size;
if (unlikely(skb->len <= mss))
goto out;
diff --git a/net/ipv6/ip6_offload.c b/net/ipv6/ip6_offload.c
index 4a87f9428ca5..5b3f2f89ef41 100644
--- a/net/ipv6/ip6_offload.c
+++ b/net/ipv6/ip6_offload.c
@@ -88,9 +88,11 @@ static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
if (skb->encapsulation &&
skb_shinfo(skb)->gso_type & (SKB_GSO_IPXIP4 | SKB_GSO_IPXIP6))
- udpfrag = proto == IPPROTO_UDP && encap;
+ udpfrag = proto == IPPROTO_UDP && encap &&
+ (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
else
- udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
+ udpfrag = proto == IPPROTO_UDP && !skb->encapsulation &&
+ (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
ops = rcu_dereference(inet6_offloads[proto]);
if (likely(ops && ops->callbacks.gso_segment)) {
diff --git a/net/ipv6/udp_offload.c b/net/ipv6/udp_offload.c
index 2a04dc9c781b..f7b85b1e6b3e 100644
--- a/net/ipv6/udp_offload.c
+++ b/net/ipv6/udp_offload.c
@@ -17,6 +17,20 @@
#include <net/ip6_checksum.h>
#include "ip6_offload.h"
+static struct sk_buff *__udp6_gso_segment(struct sk_buff *gso_skb,
+ netdev_features_t features)
+{
+ const struct ipv6hdr *ip6h = ipv6_hdr(gso_skb);
+ unsigned int mss = skb_shinfo(gso_skb)->gso_size;
+
+ if (!can_checksum_protocol(features, htons(ETH_P_IPV6)))
+ return ERR_PTR(-EIO);
+
+ return __udp_gso_segment(gso_skb, features, mss,
+ udp_v6_check(sizeof(struct udphdr) + mss,
+ &ip6h->saddr, &ip6h->daddr, 0));
+}
+
static struct sk_buff *udp6_ufo_fragment(struct sk_buff *skb,
netdev_features_t features)
{
@@ -42,12 +56,15 @@ static struct sk_buff *udp6_ufo_fragment(struct sk_buff *skb,
const struct ipv6hdr *ipv6h;
struct udphdr *uh;
- if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP))
+ if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_UDP | SKB_GSO_UDP_L4)))
goto out;
if (!pskb_may_pull(skb, sizeof(struct udphdr)))
goto out;
+ if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4)
+ return __udp6_gso_segment(skb, features);
+
/* Do software UFO. Complete and fill in the UDP checksum as HW cannot
* do checksum of UDP packets sent as multiple IP fragments.
*/
--
2.17.0.484.g0c8726318c-goog
^ permalink raw reply related
* [PATCH net-next v2 01/11] udp: expose inet cork to udp
From: Willem de Bruijn @ 2018-04-26 17:42 UTC (permalink / raw)
To: netdev; +Cc: davem, alexander.duyck, Willem de Bruijn
In-Reply-To: <20180426174225.246388-1-willemdebruijn.kernel@gmail.com>
From: Willem de Bruijn <willemb@google.com>
UDP segmentation offload needs access to inet_cork in the udp layer.
Pass the struct to ip(6)_make_skb instead of allocating it on the
stack in that function itself.
This patch is a noop otherwise.
Signed-off-by: Willem de Bruijn <willemb@google.com>
---
include/net/ip.h | 2 +-
include/net/ipv6.h | 1 +
net/ipv4/ip_output.c | 17 ++++++++---------
net/ipv4/udp.c | 4 +++-
net/ipv6/ip6_output.c | 20 ++++++++++----------
net/ipv6/udp.c | 3 ++-
6 files changed, 25 insertions(+), 22 deletions(-)
diff --git a/include/net/ip.h b/include/net/ip.h
index dc4a2d6e58a5..7ec543a64bbc 100644
--- a/include/net/ip.h
+++ b/include/net/ip.h
@@ -171,7 +171,7 @@ struct sk_buff *ip_make_skb(struct sock *sk, struct flowi4 *fl4,
int len, int odd, struct sk_buff *skb),
void *from, int length, int transhdrlen,
struct ipcm_cookie *ipc, struct rtable **rtp,
- unsigned int flags);
+ struct inet_cork *cork, unsigned int flags);
static inline struct sk_buff *ip_finish_skb(struct sock *sk, struct flowi4 *fl4)
{
diff --git a/include/net/ipv6.h b/include/net/ipv6.h
index 68b167d98879..0dd722cab037 100644
--- a/include/net/ipv6.h
+++ b/include/net/ipv6.h
@@ -950,6 +950,7 @@ struct sk_buff *ip6_make_skb(struct sock *sk,
void *from, int length, int transhdrlen,
struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
struct rt6_info *rt, unsigned int flags,
+ struct inet_cork_full *cork,
const struct sockcm_cookie *sockc);
static inline struct sk_buff *ip6_finish_skb(struct sock *sk)
diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c
index 83c73bab2c3d..2883ff1e909c 100644
--- a/net/ipv4/ip_output.c
+++ b/net/ipv4/ip_output.c
@@ -1470,9 +1470,8 @@ struct sk_buff *ip_make_skb(struct sock *sk,
int len, int odd, struct sk_buff *skb),
void *from, int length, int transhdrlen,
struct ipcm_cookie *ipc, struct rtable **rtp,
- unsigned int flags)
+ struct inet_cork *cork, unsigned int flags)
{
- struct inet_cork cork;
struct sk_buff_head queue;
int err;
@@ -1481,22 +1480,22 @@ struct sk_buff *ip_make_skb(struct sock *sk,
__skb_queue_head_init(&queue);
- cork.flags = 0;
- cork.addr = 0;
- cork.opt = NULL;
- err = ip_setup_cork(sk, &cork, ipc, rtp);
+ cork->flags = 0;
+ cork->addr = 0;
+ cork->opt = NULL;
+ err = ip_setup_cork(sk, cork, ipc, rtp);
if (err)
return ERR_PTR(err);
- err = __ip_append_data(sk, fl4, &queue, &cork,
+ err = __ip_append_data(sk, fl4, &queue, cork,
¤t->task_frag, getfrag,
from, length, transhdrlen, flags);
if (err) {
- __ip_flush_pending_frames(sk, &queue, &cork);
+ __ip_flush_pending_frames(sk, &queue, cork);
return ERR_PTR(err);
}
- return __ip_make_skb(sk, fl4, &queue, &cork);
+ return __ip_make_skb(sk, fl4, &queue, cork);
}
/*
diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
index 24b5c59b1c53..6b9d8017b319 100644
--- a/net/ipv4/udp.c
+++ b/net/ipv4/udp.c
@@ -1030,9 +1030,11 @@ int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
/* Lockless fast path for the non-corking case. */
if (!corkreq) {
+ struct inet_cork cork;
+
skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
sizeof(struct udphdr), &ipc, &rt,
- msg->msg_flags);
+ &cork, msg->msg_flags);
err = PTR_ERR(skb);
if (!IS_ERR_OR_NULL(skb))
err = udp_send_skb(skb, fl4);
diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c
index 6a477d54f8c7..7fa1db447405 100644
--- a/net/ipv6/ip6_output.c
+++ b/net/ipv6/ip6_output.c
@@ -1755,9 +1755,9 @@ struct sk_buff *ip6_make_skb(struct sock *sk,
void *from, int length, int transhdrlen,
struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
struct rt6_info *rt, unsigned int flags,
+ struct inet_cork_full *cork,
const struct sockcm_cookie *sockc)
{
- struct inet_cork_full cork;
struct inet6_cork v6_cork;
struct sk_buff_head queue;
int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
@@ -1768,27 +1768,27 @@ struct sk_buff *ip6_make_skb(struct sock *sk,
__skb_queue_head_init(&queue);
- cork.base.flags = 0;
- cork.base.addr = 0;
- cork.base.opt = NULL;
- cork.base.dst = NULL;
+ cork->base.flags = 0;
+ cork->base.addr = 0;
+ cork->base.opt = NULL;
+ cork->base.dst = NULL;
v6_cork.opt = NULL;
- err = ip6_setup_cork(sk, &cork, &v6_cork, ipc6, rt, fl6);
+ err = ip6_setup_cork(sk, cork, &v6_cork, ipc6, rt, fl6);
if (err) {
- ip6_cork_release(&cork, &v6_cork);
+ ip6_cork_release(cork, &v6_cork);
return ERR_PTR(err);
}
if (ipc6->dontfrag < 0)
ipc6->dontfrag = inet6_sk(sk)->dontfrag;
- err = __ip6_append_data(sk, fl6, &queue, &cork.base, &v6_cork,
+ err = __ip6_append_data(sk, fl6, &queue, &cork->base, &v6_cork,
¤t->task_frag, getfrag, from,
length + exthdrlen, transhdrlen + exthdrlen,
flags, ipc6, sockc);
if (err) {
- __ip6_flush_pending_frames(sk, &queue, &cork, &v6_cork);
+ __ip6_flush_pending_frames(sk, &queue, cork, &v6_cork);
return ERR_PTR(err);
}
- return __ip6_make_skb(sk, &queue, &cork, &v6_cork);
+ return __ip6_make_skb(sk, &queue, cork, &v6_cork);
}
diff --git a/net/ipv6/udp.c b/net/ipv6/udp.c
index 4ec76a87aeb8..824797f8d1ab 100644
--- a/net/ipv6/udp.c
+++ b/net/ipv6/udp.c
@@ -1324,12 +1324,13 @@ int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
/* Lockless fast path for the non-corking case */
if (!corkreq) {
+ struct inet_cork_full cork;
struct sk_buff *skb;
skb = ip6_make_skb(sk, getfrag, msg, ulen,
sizeof(struct udphdr), &ipc6,
&fl6, (struct rt6_info *)dst,
- msg->msg_flags, &sockc);
+ msg->msg_flags, &cork, &sockc);
err = PTR_ERR(skb);
if (!IS_ERR_OR_NULL(skb))
err = udp_v6_send_skb(skb, &fl6);
--
2.17.0.484.g0c8726318c-goog
^ permalink raw reply related
* [PATCH net-next v2 00/11] udp gso
From: Willem de Bruijn @ 2018-04-26 17:42 UTC (permalink / raw)
To: netdev; +Cc: davem, alexander.duyck, Willem de Bruijn
From: Willem de Bruijn <willemb@google.com>
Segmentation offload reduces cycles/byte for large packets by
amortizing the cost of protocol stack traversal.
This patchset implements GSO for UDP. A process can concatenate and
submit multiple datagrams to the same destination in one send call
by setting socket option SOL_UDP/UDP_SEGMENT with the segment size,
or passing an analogous cmsg at send time.
The stack will send the entire large (up to network layer max size)
datagram through the protocol layer. At the GSO layer, it is broken
up in individual segments. All receive the same network layer header
and UDP src and dst port. All but the last segment have the same UDP
header, but the last may differ in length and checksum.
Initial results show a significant reduction in UDP cycles/byte.
See the main patch for more details and benchmark results.
udp
876 MB/s 14873 msg/s 624666 calls/s
11,205,777,429 cycles
udp gso
2139 MB/s 36282 msg/s 36282 calls/s
11,204,374,561 cycles
The patch set is broken down as follows:
- patch 1 is a prerequisite: code rearrangement, noop otherwise
- patch 2 implements the gso logic
- patch 3 adds protocol stack support for UDP_SEGMENT
- patch 4,5,7 are refinements
- patch 6 adds the cmsg interface
- patch 8..11 are tests
This idea was presented previously at netconf 2017-2
http://vger.kernel.org/netconf2017_files/rx_hardening_and_udp_gso.pdf
Changes v1 -> v2
- Convert __udp_gso_segment to modify headers after skb_segment
- Split main patch into two, one for gso logic, one for UDP_SEGMENT
Changes RFC -> v1
- MSG_MORE:
fixed, by allowing checksum offload with corking if gso
- SKB_GSO_UDP_L4:
made independent from SKB_GSO_UDP
and removed skb_is_ufo() wrapper
- NETIF_F_GSO_UDP_L4:
add to netdev_features_string
and to netdev-features.txt
add BUILD_BUG_ON to match SKB_GSO_UDP_L4 value
- UDP_MAX_SEGMENTS:
introduce limit on number of segments per gso skb
to avoid extreme cases like IP_MAX_MTU/IPV4_MIN_MTU
- CHECKSUM_PARTIAL:
test against missing feature after ndo_features_check
if not supported return error, analogous to udp_send_check
- MSG_ZEROCOPY: removed, deferred for now
Willem de Bruijn (11):
udp: expose inet cork to udp
udp: add udp gso
udp: generate gso with UDP_SEGMENT
udp: better wmem accounting on gso
udp: paged allocation with gso
udp: add gso segment cmsg
udp: add gso support to virtual devices
selftests: udp gso
selftests: udp gso with connected sockets
selftests: udp gso with corking
selftests: udp gso benchmark
Documentation/networking/netdev-features.txt | 7 +
include/linux/netdev_features.h | 5 +-
include/linux/netdevice.h | 1 +
include/linux/skbuff.h | 2 +
include/linux/udp.h | 3 +
include/net/inet_sock.h | 1 +
include/net/ip.h | 3 +-
include/net/ipv6.h | 2 +
include/net/udp.h | 5 +
include/uapi/linux/udp.h | 1 +
net/core/ethtool.c | 1 +
net/core/skbuff.c | 2 +
net/ipv4/ip_output.c | 41 +-
net/ipv4/udp.c | 80 ++-
net/ipv4/udp_offload.c | 66 +-
net/ipv6/ip6_offload.c | 6 +-
net/ipv6/ip6_output.c | 45 +-
net/ipv6/udp.c | 31 +-
net/ipv6/udp_offload.c | 19 +-
tools/testing/selftests/net/.gitignore | 3 +
tools/testing/selftests/net/Makefile | 4 +-
tools/testing/selftests/net/udpgso.c | 621 ++++++++++++++++++
tools/testing/selftests/net/udpgso.sh | 29 +
tools/testing/selftests/net/udpgso_bench.sh | 74 +++
tools/testing/selftests/net/udpgso_bench_rx.c | 265 ++++++++
tools/testing/selftests/net/udpgso_bench_tx.c | 420 ++++++++++++
26 files changed, 1686 insertions(+), 51 deletions(-)
create mode 100644 tools/testing/selftests/net/udpgso.c
create mode 100755 tools/testing/selftests/net/udpgso.sh
create mode 100755 tools/testing/selftests/net/udpgso_bench.sh
create mode 100644 tools/testing/selftests/net/udpgso_bench_rx.c
create mode 100644 tools/testing/selftests/net/udpgso_bench_tx.c
--
2.17.0.484.g0c8726318c-goog
^ permalink raw reply
* [PATCH net-next] selftests: pmtu: Minimum MTU for vti6 is 68
From: Stefano Brivio @ 2018-04-26 17:41 UTC (permalink / raw)
To: David S . Miller
Cc: Steffen Klassert, Xin Long, Alexey Kodanev, Jarod Wilson,
Sabrina Dubroca, netdev
A vti6 interface can carry IPv4 packets too.
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
---
tools/testing/selftests/net/pmtu.sh | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/tools/testing/selftests/net/pmtu.sh b/tools/testing/selftests/net/pmtu.sh
index 1e428781a625..7651fd4d86fe 100755
--- a/tools/testing/selftests/net/pmtu.sh
+++ b/tools/testing/selftests/net/pmtu.sh
@@ -368,7 +368,7 @@ test_pmtu_vti6_link_add_mtu() {
fail=0
- min=1280
+ min=68 # vti6 can carry IPv4 packets too
max=$((65535 - 40))
# Check invalid values first
for v in $((min - 1)) $((max + 1)); do
@@ -384,7 +384,7 @@ test_pmtu_vti6_link_add_mtu() {
done
# Now check valid values
- for v in 1280 1300 $((65535 - 40)); do
+ for v in 68 1280 1300 $((65535 - 40)); do
${ns_a} ip link add vti6_a mtu ${v} type vti6 local ${veth6_a_addr} remote ${veth6_b_addr} key 10
mtu="$(link_get_mtu "${ns_a}" vti6_a)"
${ns_a} ip link del vti6_a
--
2.15.1
^ permalink raw reply related
* [PATCH ipsec] vti6: Change minimum MTU to IPV4_MIN_MTU, vti6 can carry IPv4 too
From: Stefano Brivio @ 2018-04-26 17:39 UTC (permalink / raw)
To: Steffen Klassert
Cc: Xin Long, Alexey Kodanev, Jarod Wilson, Sabrina Dubroca, netdev
A vti6 interface can carry IPv4 as well, so it makes no sense to
enforce a minimum MTU of IPV6_MIN_MTU.
If the user sets an MTU below IPV6_MIN_MTU, IPv6 will be
disabled on the interface, courtesy of addrconf_notify().
Reported-by: Xin Long <lucien.xin@gmail.com>
Fixes: b96f9afee4eb ("ipv4/6: use core net MTU range checking")
Fixes: c6741fbed6dc ("vti6: Properly adjust vti6 MTU from MTU of lower device")
Fixes: 53c81e95df17 ("ip6_vti: adjust vti mtu according to mtu of lower device")
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
---
net/ipv6/ip6_vti.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/net/ipv6/ip6_vti.c b/net/ipv6/ip6_vti.c
index c214ffec02f0..ca957dd93a29 100644
--- a/net/ipv6/ip6_vti.c
+++ b/net/ipv6/ip6_vti.c
@@ -669,7 +669,7 @@ static void vti6_link_config(struct ip6_tnl *t, bool keep_mtu)
else
mtu = ETH_DATA_LEN - LL_MAX_HEADER - sizeof(struct ipv6hdr);
- dev->mtu = max_t(int, mtu, IPV6_MIN_MTU);
+ dev->mtu = max_t(int, mtu, IPV4_MIN_MTU);
}
/**
@@ -881,7 +881,7 @@ static void vti6_dev_setup(struct net_device *dev)
dev->priv_destructor = vti6_dev_free;
dev->type = ARPHRD_TUNNEL6;
- dev->min_mtu = IPV6_MIN_MTU;
+ dev->min_mtu = IPV4_MIN_MTU;
dev->max_mtu = IP_MAX_MTU - sizeof(struct ipv6hdr);
dev->flags |= IFF_NOARP;
dev->addr_len = sizeof(struct in6_addr);
--
2.15.1
^ permalink raw reply related
page: next (older) | prev (newer) | latest
- recent:[subjects (threaded)|topics (new)|topics (active)]
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox