On Mon, Apr 20, 2015 at 02:23:05PM -0400, David Miller wrote:
From: Martin KaFai Lau ka...@fb.com
Date: Fri, 10 Apr 2015 18:54:09 -0700
Before patch 'Allow pmtu update on /128 via gateway route',
RTF_CACHE route was not created for DST_HOST. It also requires changes on
both
delete
), all routes that allow pmtu update should have a RTF_CACHE
clone. Hence, stop updating MTU for any non RTF_CACHE route.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Signed-off-by: Steffen Klassert steffen.klass...@secunet.com
Reviewed-by: Hannes Frederic Sowa han...@stressinduktion.org
---
net
the garbage collector deletes the
invalid route. This typically happens if a host route
expires afer a pmtu event. Fix this by searching also
for routes with a lower priority metric.
Signed-off-by: Steffen Klassert steffen.klass...@secunet.com
Signed-off-by: Martin KaFai Lau ka...@fb.com
Reviewed
KaFai Lau ka...@fb.com
Reviewed-by: Hannes Frederic Sowa han...@stressinduktion.org
Cc: Steffen Klassert steffen.klass...@secunet.com
---
include/net/ip6_fib.h | 10 +
net/ipv6/route.c | 102 +-
2 files changed, 60 insertions(+), 52
The series is separated from another patch series,
'ipv6: Only create RTF_CACHE route after encountering pmtu exception',
which can be found here:
http://thread.gmane.org/gmane.linux.network/359140
This series focus on fixing the /128 route issues. It is currently targeted
for net-next due to
.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Reviewed-by: Hannes Frederic Sowa han...@stressinduktion.org
Cc: Michal Kubeček mkube...@suse.cz
Cc: Steffen Klassert steffen.klass...@secunet.com
---
include/net/dst.h | 6 --
include/net/ip6_fib.h | 31 ---
net
it.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Reviewed-by: Hannes Frederic Sowa han...@stressinduktion.org
Cc: Steffen Klassert steffen.klass...@secunet.com
---
net/ipv6/addrconf.c | 2 ++
net/ipv6/route.c| 6 ++
2 files changed, 8 insertions(+)
diff --git a/net/ipv6/addrconf.c b/net/ipv6
) later.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Reviewed-by: Hannes Frederic Sowa han...@stressinduktion.org
Cc: Steffen Klassert steffen.klass...@secunet.com
---
drivers/scsi/cxgbi/libcxgbi.c | 2 +-
include/net/ipv6.h | 3 ++-
net/ipv6/icmp.c | 2 +-
net
This patch creates a RTF_CACHE routes only after encountering a pmtu
exception.
After ip6_rt_update_pmtu() has inserted the RTF_CACHE route to the fib6
tree, the rt-rt6i_node-fn_sernum is bumped which will fail the
ip6_dst_check() and trigger a relookup.
Signed-off-by: Martin KaFai Lau ka
v1 - v2:
- Move the /128 route bug fixes to another series (posted).
- Create a function for checking (rt6i_flags (RTF_NONEXTHOP | RTF_GATEWAY)).
- Avoid shuffling the skb network_header. Instead, change the function
signature to take iph instead of skb.
The perf numbers do not change much
A prep work for creating RTF_CACHE on exception only. After this
patch, the same condition (rt-rt6i_flags (RTF_NONEXTHOP | RTF_GATEWAY))
is checked twice. This redundancy will be removed in the later patch.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Reviewed-by: Hannes Frederic Sowa han
on rt6i_gateway and
RTF_ANYCAST.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Reviewed-by: Hannes Frederic Sowa han...@stressinduktion.org
Cc: Steffen Klassert steffen.klass...@secunet.com
---
include/net/ip6_route.h| 14 +-
net/bluetooth/6lowpan.c| 2
After the patch
'ipv6: Only create RTF_CACHE routes after encountering pmtu exception',
we need to compensate the performance hit (bouncing dst-__refcnt).
Signed-off-by: Martin KaFai Lau ka...@fb.com
Reviewed-by: Hannes Frederic Sowa han...@stressinduktion.org
Cc: Steffen Klassert steffen.klass
This patch breaks up ip6_rt_copy() into ip6_rt_copy_init() and
ip6_rt_cache_alloc().
In the later patch, we need to create a percpu rt6_info copy. Hence,
refactor the common rt6_info init codes to ip6_rt_copy_init().
Signed-off-by: Martin KaFai Lau ka...@fb.com
Reviewed-by: Hannes Frederic Sowa
Hi,
On Wed, Apr 29, 2015 at 11:28:46AM +0300, Julian Anastasov wrote:
Hello,
On Tue, 28 Apr 2015, Martin KaFai Lau wrote:
-static inline struct in6_addr *rt6_nexthop(struct rt6_info *rt)
+static inline struct in6_addr *rt6_nexthop(struct rt6_info *rt
This patch breaks up ip6_rt_copy() into ip6_rt_copy_init() and
ip6_rt_cache_alloc().
In the later patch, we need to create a percpu rt6_info copy. Hence,
refactor the common rt6_info init codes to ip6_rt_copy_init().
Signed-off-by: Martin KaFai Lau ka...@fb.com
Cc: Hannes Frederic Sowa han
v3 - v4:
- Patch 8 is new. It keeps track of the DST_NOCACHE routes in a list to handle
the iface down/unregister event.
- Remove rcu from the newly added rt6i_pcpu variable. It is not needed
because it has already been protected by the existing reader/writer lock.
- Thanks to 'Julian
This patch always creates RTF_CACHE clone with DST_NOCACHE
when FLOWI_FLAG_KNOWN_NH is set so that the rt6i_dst is set to
the fl6-daddr.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Acked-by: Julian Anastasov j...@ssi.bg
Tested-by: Julian Anastasov j...@ssi.bg
Cc: Hannes Frederic Sowa han
A prep work for creating RTF_CACHE on exception only. After this
patch, the same condition (rt-rt6i_flags (RTF_NONEXTHOP | RTF_GATEWAY))
is checked twice. This redundancy will be removed in the later patch.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Cc: Hannes Frederic Sowa han
) later.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Reviewed-by: Hannes Frederic Sowa han...@stressinduktion.org
Cc: Steffen Klassert steffen.klass...@secunet.com
Cc: Julian Anastasov j...@ssi.bg
---
drivers/scsi/cxgbi/libcxgbi.c | 2 +-
include/net/ipv6.h | 3 ++-
net/ipv6
Instead of doing the rt6-rt6i_node check whenever we need
to get the route's cookie. Refactor it into rt6_get_cookie().
It is a prep work to handle FLOWI_FLAG_KNOWN_NH and also
percpu rt6_info later.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Cc: Hannes Frederic Sowa han...@stressinduktion.org
on rt6i_gateway and
RTF_ANYCAST.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Cc: Hannes Frederic Sowa han...@stressinduktion.org
Cc: Steffen Klassert steffen.klass...@secunet.com
Cc: Julian Anastasov j...@ssi.bg
---
include/net/ip6_route.h| 19 ++-
net/bluetooth
After the patch
'ipv6: Only create RTF_CACHE routes after encountering pmtu exception',
we need to compensate the performance hit (bouncing dst-__refcnt).
Signed-off-by: Martin KaFai Lau ka...@fb.com
Cc: Hannes Frederic Sowa han...@stressinduktion.org
Cc: Steffen Klassert steffen.klass
This patch creates a RTF_CACHE routes only after encountering a pmtu
exception.
After ip6_rt_update_pmtu() has inserted the RTF_CACHE route to the fib6
tree, the rt-rt6i_node-fn_sernum is bumped which will fail the
ip6_dst_check() and trigger a relookup.
Signed-off-by: Martin KaFai Lau ka
This patch breaks up ip6_rt_copy() into ip6_rt_copy_init() and
ip6_rt_cache_alloc().
In the later patch, we need to create a percpu rt6_info copy. Hence,
refactor the common rt6_info init codes to ip6_rt_copy_init().
Signed-off-by: Martin KaFai Lau ka...@fb.com
Cc: Hannes Frederic Sowa han
This patch keeps track of the DST_NOCACHE routes in a list and replaces its
dev with loopback during the iface down/unregister event.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Cc: Hannes Frederic Sowa han...@stressinduktion.org
Cc: Steffen Klassert steffen.klass...@secunet.com
Cc: Julian
.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Acked-by: Julian Anastasov j...@ssi.bg
Tested-by: Julian Anastasov j...@ssi.bg
Cc: Hannes Frederic Sowa han...@stressinduktion.org
Cc: Steffen Klassert steffen.klass...@secunet.com
---
net/ipv6/raw.c | 3 +++
net/netfilter/ipvs
On Thu, Jun 04, 2015 at 01:10:26PM -0700, Grant Zhang wrote:
Hi Martin,
Thank you! My net.ipv4.tcp_mtu_probing is 1. After turning it off,
the WARN_ON stack is gone.
Thanks for confirming it.
Could you elaborate a bit on why this setting relates to the WARN_ON
trace?
The WARN_ON is
On Fri, Jun 05, 2015 at 06:11:33PM -0700, Eric Dumazet wrote:
On Fri, 2015-06-05 at 17:46 -0700, Martin KaFai Lau wrote:
The problem is caught by this WARN_ON(len skb-len) in tcp_fragment():
[810510ca] warn_slowpath_null+0x1a/0x20
[8160ec90] tcp_fragment+0x2a0/0x2b0
.
This patch is to set the pcount after skb_pull() was called
in tcp_mtu_probe().
Signed-off-by: Martin KaFai Lau ka...@fb.com
Reported-by: Grant Zhang gzh...@fastly.com
Cc: Eric Dumazet eduma...@google.com
Cc: Neal Cardwell ncardw...@google.com
Cc: Yuchung Cheng ych...@google.com
---
net/ipv4
On Fri, Jun 05, 2015 at 09:53:51AM -0700, Eric Dumazet wrote:
Sounds good, although I would simply get rid of all this complexity in
this very unlikely path.
Would you instead try the following ?
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index
- Replace the skb slicing codes by the existing tcp_trim_head(),
suggested by Eric Dumazet.
v1
- Call tcp_set_skb_tso_segs() for all slicing cases.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Reported-by: Grant Zhang gzh...@fastly.com
Cc: Grant Zhang gzh...@fastly.com
Cc: Eric Dumazet eduma
On Fri, Jun 05, 2015 at 02:23:55PM -0700, Eric Dumazet wrote:
On Fri, 2015-06-05 at 11:02 -0700, Martin KaFai Lau wrote:
tcp_trim_head() does not take the mss_now.
Is it fine to have mss_now = tcp_skb_mss(skb)? or we can depend on
the tcp_init_tso_segs() in the tcp_write_xmit() to take
On Tue, Jun 09, 2015 at 10:06:25AM -0700, Eric Dumazet wrote:
I've been working on this, but still can get the bug triggering in
tcp_fragment(), no matter what (Neal patch , yours, mine...)
Can you describe the test case that can reproduce it?
--
To unsubscribe from this list: send the line
Hi Grant,
On Thu, Jun 04, 2015 at 09:35:04AM -0700, Grant Zhang wrote:
Hi Neal,
Unfortunately with the patch we still see the same stack trace.
Attached is the TcpExtTCPSACKReneging with the patch, captured with
60 seconds interval. Its value is incremented at an similar speed as
before,
Reported-by: kbuild test robot fengguang...@intel.com
Thanks for fixing it.
Acked-by: Martin KaFai Lau ka...@fb.com
--
To unsubscribe from this list: send the line unsubscribe netdev in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
This patch creates a RTF_CACHE routes only after encountering a pmtu
exception.
After ip6_rt_update_pmtu() has inserted the RTF_CACHE route to the fib6
tree, the rt-rt6i_node-fn_sernum is bumped which will fail the
ip6_dst_check() and trigger a relookup.
Signed-off-by: Martin KaFai Lau ka
on rt6i_gateway and
RTF_ANYCAST.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Cc: Hannes Frederic Sowa han...@stressinduktion.org
Cc: Steffen Klassert steffen.klass...@secunet.com
Cc: Julian Anastasov j...@ssi.bg
---
include/net/ip6_route.h| 19 ++-
net/bluetooth
.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Acked-by: Julian Anastasov j...@ssi.bg
Tested-by: Julian Anastasov j...@ssi.bg
Cc: Hannes Frederic Sowa han...@stressinduktion.org
Cc: Steffen Klassert steffen.klass...@secunet.com
---
net/ipv6/raw.c | 3 +++
net/netfilter/ipvs
After the patch
'ipv6: Only create RTF_CACHE routes after encountering pmtu exception',
we need to compensate the performance hit (bouncing dst-__refcnt).
Signed-off-by: Martin KaFai Lau ka...@fb.com
Cc: Hannes Frederic Sowa han...@stressinduktion.org
Cc: Steffen Klassert steffen.klass
Instead of doing the rt6-rt6i_node check whenever we need
to get the route's cookie. Refactor it into rt6_get_cookie().
It is a prep work to handle FLOWI_FLAG_KNOWN_NH and also
percpu rt6_info later.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Cc: Hannes Frederic Sowa han...@stressinduktion.org
has been generated or not.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Cc: Hannes Frederic Sowa han...@stressinduktion.org
Cc: Steffen Klassert steffen.klass...@secunet.com
Cc: Julian Anastasov j...@ssi.bg
---
include/net/ipv6.h | 3 +--
net/ipv6/ip6_output.c | 17 ++---
net/ipv6
This patch keeps track of the DST_NOCACHE routes in a list and replaces its
dev with loopback during the iface down/unregister event.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Cc: Hannes Frederic Sowa han...@stressinduktion.org
Cc: Steffen Klassert steffen.klass...@secunet.com
Cc: Julian
This patch breaks up ip6_rt_copy() into ip6_rt_copy_init() and
ip6_rt_cache_alloc().
In the later patch, we need to create a percpu rt6_info copy. Hence,
refactor the common rt6_info init codes to ip6_rt_copy_init().
Signed-off-by: Martin KaFai Lau ka...@fb.com
Cc: Hannes Frederic Sowa han
) later.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Reviewed-by: Hannes Frederic Sowa han...@stressinduktion.org
Cc: Steffen Klassert steffen.klass...@secunet.com
Cc: Julian Anastasov j...@ssi.bg
---
drivers/scsi/cxgbi/libcxgbi.c | 2 +-
include/net/ipv6.h | 4 +++-
net/ipv6
This patch always creates RTF_CACHE clone with DST_NOCACHE
when FLOWI_FLAG_KNOWN_NH is set so that the rt6i_dst is set to
the fl6-daddr.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Acked-by: Julian Anastasov j...@ssi.bg
Tested-by: Julian Anastasov j...@ssi.bg
Cc: Hannes Frederic Sowa han
v4 - v5:
- Patch 1 is new. Clean up the ipv6_select_ident() and ip6_fragment().
- Further simplify the newly added rt6_get_pcpu_route(). If there is a
'prev' after cmpxchg, return prev instead of the newly created percpu
clone.
v3 - v4:
- Patch 8 is new. It keeps track of the DST_NOCACHE
A prep work for creating RTF_CACHE on exception only. After this
patch, the same condition (rt-rt6i_flags (RTF_NONEXTHOP | RTF_GATEWAY))
is checked twice. This redundancy will be removed in the later patch.
Signed-off-by: Martin KaFai Lau ka...@fb.com
Cc: Hannes Frederic Sowa han
On Tue, May 26, 2015 at 11:20:53PM +0200, Hannes Frederic Sowa wrote:
I also went over the changes to the last version and such, albeit a bit
late:
Reviewed-by: Hannes Frederic Sowa han...@stressinduktion.org
Thanks for your help and review, Hannes!
--Martin
--
To unsubscribe from this list:
This patch series fixes a potential deadlock when creating a pcpu rt.
It happens when dst_alloc() decided to run gc. Something like this:
read_lock(table-tb6_lock);
ip6_rt_pcpu_alloc()
= dst_alloc()
= ip6_dst_gc()
= write_lock(table-tb6_lock); /* oops */
Patch 1 and 2 are some prep works.
Patch
+0x5e/0x67
[141625.827146] [8132c9f8] ? sock_common_setsockopt+0xf/0x11
[141625.833660] [8132c08c] ? SyS_setsockopt+0x81/0xa2
[141625.839565] [8140ac17] entry_SYSCALL_64_fastpath+0x12/0x6a
Fixes: d52d3997f843 (pv6: Create percpu rt6_info)
Signed-off-by: Martin KaFai Lau
It is a prep work for the potential deadlock. The current
rt6_get_pcpu_route() will also create a pcpu rt if one does not exist.
This patch moves the pcpu rt creation logic into another function,
rt6_make_pcpu_route().
Signed-off-by: Martin KaFai Lau ka...@fb.com
CC: Hannes Frederic Sowa han
After 4b32b5ad31a6 (ipv6: Stop rt6_info from using inet_peer's metrics),
ip6_dst_alloc() does not need the 'table' argument. This patch
cleans it up.
Signed-off-by: Martin KaFai Lau ka...@fb.com
CC: Hannes Frederic Sowa han...@stressinduktion.org
---
net/ipv6/route.c | 21
On Mon, Aug 10, 2015 at 02:35:37PM -0400, Neal Cardwell wrote:
On Mon, Aug 10, 2015 at 2:10 PM, Jovi Zhangwei j...@cloudflare.com wrote:
Ping?
We saw a lot of this warnings in our production system. It would be
great appreciate if someone can give us the fix on this warnings. :)
What
It is a prep work for fixing a potential deadlock when creating
a pcpu rt.
The current rt6_get_pcpu_route() will also create a pcpu rt if one does not
exist. This patch moves the pcpu rt creation logic into another function,
rt6_make_pcpu_route().
Signed-off-by: Martin KaFai Lau ka...@fb.com
CC
v1 - v2:
A minor change in the commit message of patch 2.
This patch series fixes a potential deadlock when creating a pcpu rt.
It happens when dst_alloc() decided to run gc. Something like this:
read_lock(table-tb6_lock);
ip6_rt_pcpu_alloc()
= dst_alloc()
= ip6_dst_gc()
=
On Thu, Aug 13, 2015 at 05:29:09PM -0700, David Miller wrote:
From: Martin KaFai Lau ka...@fb.com
Date: Thu, 13 Aug 2015 00:58:00 -0700
This patch series fixes a potential deadlock when creating a pcpu rt.
It happens when dst_alloc() decided to run gc. Something like this:
read_lock
+0x5e/0x67
[141625.827146] [8132c9f8] ? sock_common_setsockopt+0xf/0x11
[141625.833660] [8132c08c] ? SyS_setsockopt+0x81/0xa2
[141625.839565] [8140ac17] entry_SYSCALL_64_fastpath+0x12/0x6a
Fixes: d52d3997f843 (pv6: Create percpu rt6_info)
Signed-off-by: Martin KaFai Lau
After 4b32b5ad31a6 (ipv6: Stop rt6_info from using inet_peer's metrics),
ip6_dst_alloc() does not need the 'table' argument. This patch
cleans it up.
Signed-off-by: Martin KaFai Lau ka...@fb.com
CC: Hannes Frederic Sowa han...@stressinduktion.org
---
net/ipv6/route.c | 21
On Wed, Jul 22, 2015 at 11:55:35AM -0700, Jovi Zhangwei wrote:
Sorry for disturbing, our production system(3.14 and 3.18 stable
kernel) have many tcp_fragment warnings,
the trace is same as below one which you discussed before.
On Wed, Jul 22, 2015 at 11:10:59AM +0900, YOSHIFUJI Hideaki wrote:
You have to take some lock when accessing neigh-nud_state
theoretically.
I don't think read_lock can buy us a lot of extra protection either.
If it has missed the train, the next ip6_pol_route() call will
trigger rt6_probe().
KaFai Lau ka...@fb.com
Cc: Hannes Frederic Sowa han...@stressinduktion.org
Cc: Julian Anastasov j...@ssi.bg
Cc: YOSHIFUJI Hideaki hideaki.yoshif...@miraclelinux.com
---
net/ipv6/route.c | 44
1 file changed, 20 insertions(+), 24 deletions(-)
diff --git
():
Before: 55M
After: 95M
Signed-off-by: Martin KaFai Lau ka...@fb.com
Cc: Hannes Frederic Sowa han...@stressinduktion.org
CC: Julian Anastasov j...@ssi.bg
CC: YOSHIFUJI Hideaki hideaki.yoshif...@miraclelinux.com
---
net/ipv6/route.c | 4
1 file changed, 4 insertions(+)
diff --git a/net/ipv6
v1 - v2:
1. Separate the code re-arrangement into another patch
2. Fix style
--
To unsubscribe from this list: send the line unsubscribe netdev in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
.
At the end, the total number of finished sendto():
BeforeAfter
55M 95M
Signed-off-by: Martin KaFai Lau ka...@fb.com
Cc: Hannes Frederic Sowa han...@stressinduktion.org
---
net/ipv6/route.c | 41 -
1 file changed, 20 insertions(+), 21 deletions
This patchset fixes:
1. An oops during IPv6 pmtu update on a IPv4 GRE running
in an IPSec setup
2. Misc fixes on DST_NOCACHE route
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at
__rt6_check_expired() as
one of the condition check.
Signed-off-by: Martin KaFai Lau <ka...@fb.com>
Cc: Hannes Frederic Sowa <han...@stressinduktion.org>
---
net/ipv6/route.c | 11 ++-
1 file changed, 10 insertions(+), 1 deletion(-)
diff --git a/net/ipv6/route.c b/net/ipv6/r
Fixes: 8e3d5be73681 ("ipv6: Avoid double dst_free")
Signed-off-by: Martin KaFai Lau <ka...@fb.com>
Cc: Hannes Frederic Sowa <han...@stressinduktion.org>
---
include/net/ip6_fib.h | 3 ++-
net/ipv6/route.c | 3 ++-
2 files changed, 4 insertions(+), 2 deletions(-)
diff --
On Tue, Oct 13, 2015 at 09:26:41PM +0200, Phil Sutter wrote:
> I have backed up the rt pointer at top of the function and restored it
> before pr_err, this is the output:
>
> | rt6i_dst:2001:4dd0:ff3b:13::/64 rt6i_gateway::: rt6i_flags:4001
> dst.flags:
Hi Phil, Can you try the
This patchset ensures the rt6_info's fields are initialized properly
in ip6_blackhole_route() where xfrm_policy is the primarily user.
The first patch is a prep work. The second patch is the fix. It
fixes d52d3997f843 ("ipv6: Create percpu rt6_info").
Here is the oops reported by Phil Sutter
allocated blackhole route.
This bug triggers an oops (reported by Phil Sutter) in rt6_get_cookie().
It is because RTF_PCPU is set while rt->dst.from is NULL.
Fixes: d52d3997f843 ("ipv6: Create percpu rt6_info")
Signed-off-by: Martin KaFai Lau <ka...@fb.com>
Reported-by: Phil
Introduce rt6_info_init() to do the common init work for
'struct rt6_info' (after calling dst_alloc).
It is a prep work to fix the rt6_info init logic in the
ip6_blackhole_route().
Signed-off-by: Martin KaFai Lau <ka...@fb.com>
Cc: Hannes Frederic Sowa <han...@stressinduktion.org>
On Thu, Oct 15, 2015 at 12:34:13AM +0200, Phil Sutter wrote:
> Hi Martin,
>
> On Tue, Oct 13, 2015 at 11:14:21PM -0700, Martin KaFai Lau wrote:
> > On Tue, Oct 13, 2015 at 09:26:41PM +0200, Phil Sutter wrote:
> > > I have backed up the rt pointer at top of the function and
On Sat, Oct 10, 2015 at 03:24:37PM +0200, Phil Sutter wrote:
> Using printk-debugging I could track down the problem to
> rt6_get_cookie() function in include/net/ip6_fib.h:
>
> The conditional at the start of the function evaluates true, since
> 'rt->rt6i_flags & RTF_PCPU' is non-zero. Due to
> incorrect behavior.
Thanks for fixing it.
Reviewed-by: Martin KaFai Lau <ka...@fb.com>
I also tested the following cases with the presence of DST_NOCACHE entries:
1. rmmod e1000.ko while running netperf
2. unshare(CLONE_NEWNET) as reported by Dmitry
Tested-by: Martin KaFai Lau <ka...@f
On Tue, Oct 13, 2015 at 08:46:49PM +0800, Li RongQing wrote:
> 1. in a machine, configure 3000 ipv6 address in one interface
>
> for i in {1..3000}; do ip -6 addr add 4001:5013::$i/0 dev eth0; done
>
>
> 2. in other machine, ping6 the upper configured ipv6 address, then
> lots of lost packets
>
>
On Tue, Oct 13, 2015 at 09:10:39PM +0200, Phil Sutter wrote:
> Hi Martin,
>
> On Tue, Oct 13, 2015 at 11:14:43AM -0700, Martin KaFai Lau wrote:
> > On Sat, Oct 10, 2015 at 03:24:37PM +0200, Phil Sutter wrote:
> > > The conditional at the start of the function evaluat
On Sat, Oct 10, 2015 at 03:24:37PM +0200, Phil Sutter wrote:
> The conditional at the start of the function evaluates true, since
> 'rt->rt6i_flags & RTF_PCPU' is non-zero.
Hi Phil, can you try the following patch and capture the dmesg output
to confirm the value of rt->rt6i_flags and the
cached
> dst on non current cpu are not actually reset.
>
> This patch replaces raw_cpu_ptr with per_cpu_ptr, properly cleaning
> such storage.
Thanks for fixing it.
Acked-by: Martin KaFai Lau <ka...@fb.com>
>
> Fixes: cdf3464e6c6b ("ipv6: Fix dst_entry refcnt bugs in
On Mon, Aug 17, 2015 at 11:43:20AM +0200, Alexander Holler wrote:
That's why I vote to check out if it's possible/reasonable to backport this
series to the stable kernels.
I have backported to 4.0.y without major issue, so possible.
I did try on 3.1x and gave up.
It is a lot of changes, so I
with ip6_tnl_dst_get().
Signed-off-by: Martin KaFai Lau <ka...@fb.com>
---
include/net/ip6_tunnel.h | 4 ++--
net/ipv6/ip6_gre.c | 4 ++--
net/ipv6/ip6_tunnel.c| 12 ++--
3 files changed, 10 insertions(+), 10 deletions(-)
diff --git a/include/net/ip6_tunnel.h b/inclu
y's refcnt.
This patch:
1. Create a percpu dst_entry cache in ip6_tnl
2. Use a spinlock to protect the dst_cache operations
3. The outgoing skb always holds the dst_entry's refcnt
Signed-off-by: Martin KaFai Lau <ka...@fb.com>
---
include/net/ip6_tunnel.h | 11 -
net/ipv6/ip6_gre.c
This patch series is to fix the dst refcnt bugs in ip6_tunnel.
Patch 1 and 2 are the prep works. Patch 3 is the fix.
I can reproduce the bug by adding and removing the ip6gre tunnel
while running a super_netperf TCP_CRR test. I get the following
trace by adding WARN_ON_ONCE(newrefcnt < 0) to
It is a prep work to fix the dst_entry refcnt bugs in ip6_tunnel.
This patch refactors some common init codes used by both
ip6gre_tunnel_init and ip6gre_tap_init.
Signed-off-by: Martin KaFai Lau <ka...@fb.com>
---
net/ipv6/ip6_gre.c | 37 -
1 file chang
On Tue, Sep 01, 2015 at 03:38:36PM -0700, Eric Dumazet wrote:
> On Tue, 2015-09-01 at 15:25 -0700, Martin KaFai Lau wrote:
> > On Tue, Sep 01, 2015 at 02:26:58PM -0700, Eric Dumazet wrote:
> > > On Tue, 2015-09-01 at 13:55 -0700, Martin KaFai Lau wrote:
> > > > On T
On Tue, Sep 01, 2015 at 01:14:20PM -0700, Eric Dumazet wrote:
> On Tue, 2015-09-01 at 11:55 -0700, Martin KaFai Lau wrote:
> > Problems in the current dst_entry cache in the ip6_tunnel:
> >
> > 1. ip6_tnl_dst_set is racy. There is no lock to protect it:
> >- One ma
On Tue, Sep 01, 2015 at 01:14:20PM -0700, Eric Dumazet wrote:
> It should not be a problem. refcnt is taken when/if necessary (skb
> queued on a qdisc for example)
>
> We have other uses of skb_dst_set_noref()
>
> Please describe the problem ?
The current ip6_tnl_dst_get() does not take the dst
On Tue, Sep 01, 2015 at 05:31:44PM -0700, Martin KaFai Lau wrote:
> On Tue, Sep 01, 2015 at 03:38:36PM -0700, Eric Dumazet wrote:
> > On Tue, 2015-09-01 at 15:25 -0700, Martin KaFai Lau wrote:
> > > On Tue, Sep 01, 2015 at 02:26:58PM -0700, Eric Dumazet wrote:
> > > &g
On Tue, Sep 01, 2015 at 05:42:00PM -0700, Martin KaFai Lau wrote:
> I look a closer look at dst_rcu_free() and your commit pointers. I can see
> your point
> for DST_NOCACHE.
>
> However, dst_free() for not DST_NOCACHE is still an issue, I think.
oops. Ignore this ema
On Tue, Sep 01, 2015 at 02:26:58PM -0700, Eric Dumazet wrote:
> On Tue, 2015-09-01 at 13:55 -0700, Martin KaFai Lau wrote:
> > On Tue, Sep 01, 2015 at 01:14:20PM -0700, Eric Dumazet wrote:
> > > It should not be a problem. refcnt is taken when/if necessary (skb
> > > qu
On Tue, Sep 01, 2015 at 01:14:20PM -0700, Eric Dumazet wrote:
> > 2. Use a spinlock to protect the dst_cache operations
>
> Well, a seqlock would be better : No need for an atomic operation in
> fast path.
>
seqlock can ensure consistency between idst->dst and idst->cookie.
However, IPv6 dst
On Wed, Sep 02, 2015 at 03:48:57PM -0700, Eric Dumazet wrote:
> On Wed, 2015-09-02 at 14:52 -0700, Martin KaFai Lau wrote:
> > On Wed, Sep 02, 2015 at 02:30:45PM -0700, Eric Dumazet wrote:
> > > Object cannot be freed until all cpus have exited their RCU sections.
> > Y
On Wed, Sep 02, 2015 at 02:30:45PM -0700, Eric Dumazet wrote:
> Object cannot be freed until all cpus have exited their RCU sections.
You meant the dst_destroy() here will wait for all cpus exited their RCU
sections?
static inline void dst_free(struct dst_entry *dst)
{
if (dst->obsolete
On Fri, Sep 04, 2015 at 04:12:41PM -0700, Martin KaFai Lau wrote:
> @@ -1962,6 +1961,9 @@ static int __ip6_del_rt(struct rt6_info *rt, struct
> nl_info *info)
> if (rt == net->ipv6.ip6_null_entry) {
> err = -ENOENT;
> goto out;
> + } e
destroyed already.
3. If rt is a DST_NOCACHE, dst_free(rt) should not be called.
4. It is a stopper to make dst freeing from fib tree undergo a
rcu grace period.
This patch is to use a DST_NOCACHE flag to indicate a rt is
managed by the fib tree or not.
Signed-off-by: Martin KaFai Lau <ka...@fb.
with ip6_tnl_dst_get().
Signed-off-by: Martin KaFai Lau <ka...@fb.com>
---
include/net/ip6_tunnel.h | 4 ++--
net/ipv6/ip6_gre.c | 4 ++--
net/ipv6/ip6_tunnel.c| 12 ++--
3 files changed, 10 insertions(+), 10 deletions(-)
diff --git a/include/net/ip6_tunnel.h b/inclu
This patch uses a seqlock to ensure consistency between idst->dst and
idst->cookie. It also makes dst freeing from fib tree to undergo a
rcu grace period.
Signed-off-by: Martin KaFai Lau <ka...@fb.com>
---
include/net/ip6_tunnel.h | 4 ++--
net/ipv6/ip6_fib.c | 9 +++
v2:
- Add patch 4 and 5 to remove the spinlock
v1:
This patch series is to fix the dst refcnt bugs in ip6_tunnel.
Patch 1 and 2 are the prep works. Patch 3 is the fix.
I can reproduce the bug by adding and removing the ip6gre tunnel
while running a super_netperf TCP_CRR test. I get the
:
1. Create a percpu dst_entry cache in ip6_tnl
2. Use a spinlock to protect the dst_cache operations
3. ip6_tnl_dst_get always takes the dst refcnt before returning
Signed-off-by: Martin KaFai Lau <ka...@fb.com>
Conflicts:
net/ipv6/ip6_gre.c
net/ipv6/ip6_tunnel.c
---
in
It is a prep work to fix the dst_entry refcnt bugs in ip6_tunnel.
This patch refactors some common init codes used by both
ip6gre_tunnel_init and ip6gre_tap_init.
Signed-off-by: Martin KaFai Lau <ka...@fb.com>
---
net/ipv6/ip6_gre.c | 42 +-
1 - 100 of 724 matches
Mail list logo
