Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]
Groups > linux.kernel > #1303618 > unrolled thread
| Started by | Nikolay Borisov <kernel@kyup.com> |
|---|---|
| First post | 2016-01-07 15:40 +0100 |
| Last post | 2016-01-10 23:40 +0100 |
| Articles | 3 — 3 participants |
Back to article view | Back to linux.kernel
[PATCH 0/3] Namespaceify tcp keepalive machinery Nikolay Borisov <kernel@kyup.com> - 2016-01-07 15:40 +0100
Re: [PATCH 0/3] Namespaceify tcp keepalive machinery ebiederm@xmission.com (Eric W. Biederman) - 2016-01-07 22:50 +0100
Re: [PATCH 0/3] Namespaceify tcp keepalive machinery David Miller <davem@davemloft.net> - 2016-01-10 23:40 +0100
| From | Nikolay Borisov <kernel@kyup.com> |
|---|---|
| Date | 2016-01-07 15:40 +0100 |
| Subject | [PATCH 0/3] Namespaceify tcp keepalive machinery |
| Message-ID | <qOjTj-SP-3@gated-at.bofh.it> |
The following patch series enables the tcp keepalive mechanism to be configured per net namespace. This is especially useful if you have multiple containers hosted on one node and one of them is under DoS- in such situations one thing which could be done is to configure the tcp keepalive settings such that connections for that particular container are being reset faster. Another scenario where not being able to control those knob comes per container is problematic is occurs the value of net.netfilter.nf_conntrack_tcp_timeout_established is set below the keepalive interval, in such situations the server won't send an RST packet resulting in applications not trying to reconnect and stale connection waiting. Changing the global keepalive value is a possible solution but it might interfere with other containers. The three patches gradually convert each of the affected knobs to be per netns. I thought it would be easier for review than put everything in one patch. If people deem it more appropriate to squash everything in one patch (maybe after review) I'd be more than happy to do it. The patches have been compile-tested on 4.4 and functionally tested on 3.12 and they work as expected. These are based off 4.4-rc8 Nikolay Borisov (3): ipv4: Namespaceify tcp_keepalive_time sysctl knob ipv4: Namespecify tcp_keepalive_probes sysctl knob ipv4: Namespecify the tcp_keepalive_intvl sysctl knob include/net/netns/ipv4.h | 4 ++++ include/net/tcp.h | 15 +++++++++------ net/ipv4/sysctl_net_ipv4.c | 42 +++++++++++++++++++++--------------------- net/ipv4/tcp_ipv4.c | 4 ++++ net/ipv4/tcp_timer.c | 3 --- 5 files changed, 38 insertions(+), 30 deletions(-) -- 2.5.0 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
[toc] | [next] | [standalone]
| From | ebiederm@xmission.com (Eric W. Biederman) |
|---|---|
| Date | 2016-01-07 22:50 +0100 |
| Message-ID | <qOqBt-5uC-9@gated-at.bofh.it> |
| In reply to | #1303618 |
Nikolay Borisov <kernel@kyup.com> writes: > The following patch series enables the tcp keepalive mechanism > to be configured per net namespace. This is especially useful > if you have multiple containers hosted on one node and one of > them is under DoS- in such situations one thing which could > be done is to configure the tcp keepalive settings such that > connections for that particular container are being reset > faster. > > Another scenario where not being able to control those knob > comes per container is problematic is occurs the value of > net.netfilter.nf_conntrack_tcp_timeout_established is set > below the keepalive interval, in such situations the server won't > send an RST packet resulting in applications not trying to > reconnect and stale connection waiting. Changing the global > keepalive value is a possible solution but it might interfere > with other containers. > > The three patches gradually convert each of the affected knobs > to be per netns. I thought it would be easier for review than > put everything in one patch. If people deem it more appropriate > to squash everything in one patch (maybe after review) I'd > be more than happy to do it. > > The patches have been compile-tested on 4.4 and functionally > tested on 3.12 and they work as expected. > > These are based off 4.4-rc8 Acked-by: "Eric W. Biederman" <ebiederm@xmission.com> I took a quick skim and there appears to be nothing scary in your patches and the separation of the patches did make the review easy. All of the knobs are already per socket with a global default. Moving that global to be per network namespace appears straight forward in your patches. > Nikolay Borisov (3): > ipv4: Namespaceify tcp_keepalive_time sysctl knob > ipv4: Namespecify tcp_keepalive_probes sysctl knob > ipv4: Namespecify the tcp_keepalive_intvl sysctl knob > > include/net/netns/ipv4.h | 4 ++++ > include/net/tcp.h | 15 +++++++++------ > net/ipv4/sysctl_net_ipv4.c | 42 +++++++++++++++++++++--------------------- > net/ipv4/tcp_ipv4.c | 4 ++++ > net/ipv4/tcp_timer.c | 3 --- > 5 files changed, 38 insertions(+), 30 deletions(-)
[toc] | [prev] | [next] | [standalone]
| From | David Miller <davem@davemloft.net> |
|---|---|
| Date | 2016-01-10 23:40 +0100 |
| Message-ID | <qPwOt-1Sj-17@gated-at.bofh.it> |
| In reply to | #1303618 |
From: Nikolay Borisov <kernel@kyup.com> Date: Thu, 7 Jan 2016 16:38:42 +0200 > The following patch series enables the tcp keepalive mechanism > to be configured per net namespace. This is especially useful > if you have multiple containers hosted on one node and one of > them is under DoS- in such situations one thing which could > be done is to configure the tcp keepalive settings such that > connections for that particular container are being reset > faster. > > Another scenario where not being able to control those knob > comes per container is problematic is occurs the value of > net.netfilter.nf_conntrack_tcp_timeout_established is set > below the keepalive interval, in such situations the server won't > send an RST packet resulting in applications not trying to > reconnect and stale connection waiting. Changing the global > keepalive value is a possible solution but it might interfere > with other containers. > > The three patches gradually convert each of the affected knobs > to be per netns. I thought it would be easier for review than > put everything in one patch. If people deem it more appropriate > to squash everything in one patch (maybe after review) I'd > be more than happy to do it. > > The patches have been compile-tested on 4.4 and functionally > tested on 3.12 and they work as expected. > > These are based off 4.4-rc8 Series applied, thanks.
[toc] | [prev] | [standalone]
Back to top | Article view | linux.kernel
csiph-web