Upcoming congestion controls for TCP require usec resolution for RTT
estimations. Millisecond resolution is simply not enough these days.
FQ/pacing in DC environments also require this change for finer control
and removal of bimodal behavior due to the current hack in
tcp_update_pacing_rate() for 'small rtt'
TCP_CONG_RTT_STAMP is no longer needed.
As Julian Anastasov pointed out, we need to keep user compatibility :
tcp_metrics used to export RTT and RTTVAR in msec resolution,
so we added RTT_US and RTTVAR_US. An iproute2 patch is needed
to use the new attributes if provided by the kernel.
In this example ss command displays a srtt of 32 usecs (10Gbit link)
lpk51:~# ./ss -i dst lpk52
Netid State Recv-Q Send-Q Local Address:Port Peer
Address:Port
tcp ESTAB 0 1 10.246.11.51:42959
10.246.11.52:64614
cubic wscale:6,6 rto:201 rtt:0.032/0.001 ato:40 mss:1448
cwnd:10 send
3620.0Mbps pacing_rate 7240.0Mbps unacked:1 rcv_rtt:993 rcv_space:29559
Updated iproute2 ip command displays :
lpk51:~# ./ip tcp_metrics | grep 10.246.11.52
10.246.11.52 age 561.914sec cwnd 10 rtt 274us rttvar 213us source
10.246.11.51
Old binary displays :
lpk51:~# ip tcp_metrics | grep 10.246.11.52
10.246.11.52 age 561.914sec cwnd 10 rtt 250us rttvar 125us source
10.246.11.51
With help from Julian Anastasov, Stephen Hemminger and Yuchung Cheng
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Larry Brakmo <brakmo@google.com>
Cc: Julian Anastasov <ja@ssi.bg>
Signed-off-by: David S. Miller <davem@davemloft.net>
A socket may be v6/v4-mapped. In that case sk->sk_family is AF_INET6,
but the IP being used is actually an IPv4-address.
Current's tcp-metrics will thus represent it as an IPv6-address:
root@server:~# ip tcp_metrics
::ffff:10.1.1.2 age 22.920sec rtt 18750us rttvar 15000us cwnd 10
10.1.1.2 age 47.970sec rtt 16250us rttvar 10000us cwnd 10
This patch modifies the tcp-metrics so that they are able to handle the
v6/v4-mapped sockets correctly.
Signed-off-by: Christoph Paasch <christoph.paasch@uclouvain.be>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In bbf852b96e I introduced the tmlist, which allows to delete
multiple entries from the cache that match a specified destination if no
source-IP is specified.
However, as the cache is an RCU-list, we should not create this tmlist, as
it will change the tcpm_next pointer of the element that will be deleted
and so a thread iterating over the cache's entries while holding the
RCU-lock might get "redirected" to this tmlist.
This patch fixes this, by reverting back to the old behavior prior to
bbf852b96e, which means that we simply change the tcpm_next
pointer of the previous element (pp) to jump over the one we are
deleting.
The difference is that we call kfree_rcu() directly on the cache entry,
which allows us to delete multiple entries from the list.
Fixes: bbf852b96e (tcp: metrics: Delete all entries matching a certain destination)
Signed-off-by: Christoph Paasch <christoph.paasch@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c
net/ipv4/tcp_metrics.c
Overlapping changes between the "don't create two tcp metrics objects
with the same key" race fix in net and the addition of the destination
address in the lookup key in net-next.
Minor overlapping changes in bnx2x driver.
Signed-off-by: David S. Miller <davem@davemloft.net>
Because the tcp-metrics is an RCU-list, it may be that two
soft-interrupts are inside __tcp_get_metrics() for the same
destination-IP at the same time. If this destination-IP is not yet part of
the tcp-metrics, both soft-interrupts will end up in tcpm_new and create
a new entry for this IP.
So, we will have two tcp-metrics with the same destination-IP in the list.
This patch checks twice __tcp_get_metrics(). First without holding the
lock, then while holding the lock. The second one is there to confirm
that the entry has not been added by another soft-irq while waiting for
the spin-lock.
Fixes: 51c5d0c4b1 (tcp: Maintain dynamic metrics in local cache.)
Signed-off-by: Christoph Paasch <christoph.paasch@uclouvain.be>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We want to be able to get/del tcp-metrics based on the src IP. This
patch adds the necessary parsing of the netlink attribute and if the
source address is set, it will match on this one too.
Signed-off-by: Christoph Paasch <christoph.paasch@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
As we now can have multiple entries per destination-IP, the "ip
tcp_metrics delete address ADDRESS" command deletes all of them.
Signed-off-by: Christoph Paasch <christoph.paasch@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds a new netlink attribute for the source-IP and appends it
to the netlink reply. Now, iproute2 can have access to the source-IP.
Signed-off-by: Christoph Paasch <christoph.paasch@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
We add the source-address to the tcp-metrics, so that different metrics
will be used per source/destination-pair. We use the destination-hash to
store the metric inside the hash-table. That way, deleting and dumping
via "ip tcp_metrics" is easy.
Signed-off-by: Christoph Paasch <christoph.paasch@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
As we will add also the source-address, we rename all accesses to the
tcp-metrics address to use "daddr".
Signed-off-by: Christoph Paasch <christoph.paasch@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
As suggested by David Miller, make genl_register_family_with_ops()
a macro and pass only the array, evaluating ARRAY_SIZE() in the
macro, this is a little safer.
The openvswitch has some indirection, assing ops/n_ops directly in
that code. This might ultimately just assign the pointers in the
family initializations, saving the struct genl_family_and_ops and
code (once mcast groups are handled differently.)
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that genl_ops are no longer modified in place when
registering, they can be made const. This patch was done
mostly with spatch:
@@
identifier ops;
@@
+const
struct genl_ops ops[] = {
...
};
(except the struct thing in net/openvswitch/datapath.c)
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We had some reports of crashes using TCP fastopen, and Dave Jones
gave a nice stack trace pointing to the error.
Issue is that tcp_get_metrics() should not be called with a NULL dst
Fixes: 1fe4c481ba ("net-tcp: Fast Open client - cookie cache")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Dave Jones <davej@redhat.com>
Cc: Yuchung Cheng <ycheng@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Tested-by: Dave Jones <davej@fedoraproject.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fast Open currently has a fall back feature to address SYN-data being
dropped but it requires the middle-box to pass on regular SYN retry
after SYN-data. This is implemented in commit aab487435 ("net-tcp:
Fast Open client - detecting SYN-data drops")
However some NAT boxes will drop all subsequent packets after first
SYN-data and blackholes the entire connections. An example is in
commit 356d7d8 "netfilter: nf_conntrack: fix tcp_in_window for Fast
Open".
The sender should note such incidents and fall back to use the regular
TCP handshake on subsequent attempts temporarily as well: after the
second SYN timeouts the original Fast Open SYN is most likely lost.
When such an event recurs Fast Open is disabled based on the number of
recurrences exponentially.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
TCP listener refactoring, part 5 :
We want to be able to insert request sockets (SYN_RECV) into main
ehash table instead of the per listener hash table to allow RCU
lookups and remove listener lock contention.
This patch includes the needed struct sock_common in front
of struct request_sock
This means there is no more inet6_request_sock IPv6 specific
structure.
Following inet_request_sock fields were renamed as they became
macros to reference fields from struct sock_common.
Prefix ir_ was chosen to avoid name collisions.
loc_port -> ir_loc_port
loc_addr -> ir_loc_addr
rmt_addr -> ir_rmt_addr
rmt_port -> ir_rmt_port
iif -> ir_iif
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
CONFIG_IPV6=n is still a valid choice ;)
It appears we can remove dead code.
Reported-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
TCP listener refactoring, part 4 :
To speed up inet lookups, we moved IPv4 addresses from inet to struct
sock_common
Now is time to do the same for IPv6, because it permits us to have fast
lookups for all kind of sockets, including upcoming SYN_RECV.
Getting IPv6 addresses in TCP lookups currently requires two extra cache
lines, plus a dereference (and memory stall).
inet6_sk(sk) does the dereference of inet_sk(__sk)->pinet6
This patch is way bigger than its IPv4 counter part, because for IPv4,
we could add aliases (inet_daddr, inet_rcv_saddr), while on IPv6,
it's not doable easily.
inet6_sk(sk)->daddr becomes sk->sk_v6_daddr
inet6_sk(sk)->rcv_saddr becomes sk->sk_v6_rcv_saddr
And timewait socket also have tw->tw_v6_daddr & tw->tw_v6_rcv_saddr
at the same offset.
We get rid of INET6_TW_MATCH() as INET6_MATCH() is now the generic
macro.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 1b7fdd2ab5 ("tcp: do not use cached RTT for RTT estimation")
did not correctly account for the fact that crtt is the RTT shifted
left 3 bits. Fix the calculation to consistently reflect this fact.
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Acked-By: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 1b7fdd2ab585("tcp: do not use cached RTT for RTT estimation")
removes important comments on how RTO is initialized and updated.
Hopefully this patch puts those information back.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
RTT cached in the TCP metrics are valuable for the initial timeout
because SYN RTT usually does not account for serialization delays
on low BW path.
However using it to seed the RTT estimator maybe disruptive because
other components (e.g., pacing) require the smooth RTT to be obtained
from actual connection.
The solution is to use the higher cached RTT to set the first RTO
conservatively like tcp_rtt_estimator(), but avoid seeding the other
RTT estimator variables such as srtt. It is also a good idea to
keep RTO conservative to obtain the first RTT sample, and the
performance is insured by TCP loss probe if SYN RTT is available.
To keep the seeding formula consistent across SYN RTT and cached RTT,
the rttvar is twice the cached RTT instead of cached RTTVAR value. The
reason is because cached variation may be too small (near min RTO)
which defeats the purpose of being conservative on first RTO. However
the metrics still keep the RTT variations as they might be useful for
user applications (through ip).
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Tested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
TCP metric cache expires entries after one hour.
This probably make sense for TCP RTT/RTTVAR/CWND, but not
for TCP fastopen cookies.
Its better to try previous cookie. If it appears to be obsolete,
server will send us new cookie anyway.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
order-5 allocations can fail with current kernels, we should
try vmalloc() as well.
Reported-by: Julien Tinnes <jln@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix double sizeof when parsing IPv6 address from
user space because it breaks get/del by specific IPv6 address.
Problem noticed by David Binderman:
https://bugzilla.kernel.org/show_bug.cgi?id=49171
Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: David S. Miller <davem@davemloft.net>
It is a frequent mistake to confuse the netlink port identifier with a
process identifier. Try to reduce this confusion by renaming fields
that hold port identifiers portid instead of pid.
I have carefully avoided changing the structures exported to
userspace to avoid changing the userspace API.
I have successfully built an allyesconfig kernel with this change.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Acked-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add support for genl "tcp_metrics". No locking
is changed, only that now we can unlink and delete
entries after grace period. We implement get/del for
single entry and dump to support show/flush filtering
in user space. Del without address attribute causes
flush for all addresses, sadly under genl_mutex.
v2:
- remove rcu_assign_pointer as suggested by Eric Dumazet,
it is not needed because there are no other writes under lock
- move the flushing code in tcp_metrics_flush_all
v3:
- remove synchronize_rcu on flush as suggested by Eric Dumazet
Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: David S. Miller <davem@davemloft.net>
We currently leak all tcp metrics at struct net dismantle time.
tcp_net_metrics_exit() frees the hash table, we must first
iterate it to free all metrics.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In tcp_tw_remember_stamp we incorrectly checked tw
instead of tm, it can lead to oops if the cached entry is
not found.
tcpm_stamp was not updated in tcpm_check_stamp when
tcpm_suck_dst was called, move the update into tcpm_suck_dst,
so that we do not call it infinitely on every next cache hit
after TCP_METRICS_TIMEOUT.
Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix a missing roundup_pow_of_two(), since tcpmhash_entries is not
guaranteed to be a power of two.
Uses hash_32() instead of custom hash.
tcpmhash_entries should be an unsigned int.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
On paths with firewalls dropping SYN with data or experimental TCP options,
Fast Open connections will have experience SYN timeout and bad performance.
The solution is to track such incidents in the cookie cache and disables
Fast Open temporarily.
Since only the original SYN includes data and/or Fast Open option, the
SYN-ACK has some tell-tale sign (tcp_rcv_fastopen_synack()) to detect
such drops. If a path has recurring Fast Open SYN drops, Fast Open is
disabled for 2^(recurring_losses) minutes starting from four minutes up to
roughly one and half day. sendmsg with MSG_FASTOPEN flag will succeed but
it behaves as connect() then write().
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
With help from Eric Dumazet, add Fast Open metrics in tcp metrics cache.
The basic ones are MSS and the cookies. Later patch will cache more to
handle unfriendly middleboxes.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Introduce ipv6_addr_hash() helper doing a XOR on all bits
of an IPv6 address, with an optimized x86_64 version.
Use it in flow dissector, as suggested by Andrew McGregor,
to reduce hash collision probabilities in fq_codel (and other
users of flow dissector)
Use it in ip6_tunnel.c and use more bit shuffling, as suggested
by David Laight, as existing hash was ignoring most of them.
Use it in sunrpc and use more bit shuffling, using hash_32().
Use it in net/ipv6/addrconf.c, using hash_32() as well.
As a cleanup, use it in net/ipv4/tcp_metrics.c
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Andrew McGregor <andrewmcgr@gmail.com>
Cc: Dave Taht <dave.taht@gmail.com>
Cc: Tom Herbert <therbert@google.com>
Cc: David Laight <David.Laight@ACULAB.COM>
Cc: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The recent patch "tcp: Maintain dynamic metrics in local cache." introduced
an out of bounds access due to what appears to be a typo. I believe this
change should resolve the issue by replacing the access to RTAX_CWND with
TCP_METRIC_CWND.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Maintain a local hash table of TCP dynamic metrics blobs.
Computed TCP metrics are no longer maintained in the route metrics.
The table uses RCU and an extremely simple hash so that it has low
latency and low overhead. A simple hash is legitimate because we only
make metrics blobs for fully established connections.
Some tweaking of the default hash table sizes, metric timeouts, and
the hash chain length limit certainly could use some tweaking. But
the basic design seems sound.
With help from Eric Dumazet and Joe Perches.
Signed-off-by: David S. Miller <davem@davemloft.net>
