If the preferred lifetime was less than the minimum required lifetime,
ipv6_create_tempaddr would error out without creating any new address.
On my machine and network, this error happened immediately with the
preferred lifetime set to 5 seconds or less, after a few minutes with
the preferred lifetime set to 6 seconds, and not at all with the
preferred lifetime set to 7 seconds. During my investigation, I found a
Stack Exchange post from another person who seems to have had the same
problem: They stopped getting new addresses if they lowered the
preferred lifetime below 3 seconds, and they didn't really know why.
The preferred lifetime is a preference, not a hard requirement. The
kernel does not strictly forbid new connections on a deprecated address,
nor does it guarantee that the address will be disposed of the instant
its total valid lifetime expires. So rather than disable IPv6 privacy
extensions altogether if the minimum required lifetime swells above the
preferred lifetime, it is more in keeping with the user's intent to
increase the temporary address's lifetime to the minimum necessary for
the current network conditions.
With these fixes, setting the preferred lifetime to 5 or 6 seconds "just
works" because the extra fraction of a second is practically
unnoticeable. It's even possible to reduce the time before deprecation
to 1 or 2 seconds by setting /proc/sys/net/ipv6/conf/*/regen_min_advance
and /proc/sys/net/ipv6/conf/*/dad_transmits to 0. I realize that that is
a pretty niche use case, but I know at least one person who would gladly
sacrifice performance and convenience to be sure that they are getting
the maximum possible level of privacy.
Link: https://serverfault.com/a/1031168/310447
Signed-off-by: Alex Henrie <alexhenrie24@gmail.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
In RFC 8981, REGEN_ADVANCE cannot be less than 2 seconds, and the RFC
does not permit the creation of temporary addresses with lifetimes
shorter than that:
> When processing a Router Advertisement with a
> Prefix Information option carrying a prefix for the purposes of
> address autoconfiguration (i.e., the A bit is set), the host MUST
> perform the following steps:
> 5. A temporary address is created only if this calculated preferred
> lifetime is greater than REGEN_ADVANCE time units.
However, some users want to change their IPv6 address as frequently as
possible regardless of the RFC's arbitrary minimum lifetime. For the
benefit of those users, add a regen_min_advance sysctl parameter that
can be set to below or above 2 seconds.
Link: https://datatracker.ietf.org/doc/html/rfc8981
Signed-off-by: Alex Henrie <alexhenrie24@gmail.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
RFC 8981 defines REGEN_ADVANCE as follows:
REGEN_ADVANCE = 2 + (TEMP_IDGEN_RETRIES * DupAddrDetectTransmits * RetransTimer / 1000)
Thus, allowing it to be less than 2 seconds is technically a protocol
violation.
Link: https://datatracker.ietf.org/doc/html/rfc8981#name-defined-protocol-parameters
Signed-off-by: Alex Henrie <alexhenrie24@gmail.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
TCP pingpong threshold is 1 by default. But some applications, like SQL DB
may prefer a higher pingpong threshold to activate delayed acks in quick
ack mode for better performance.
The pingpong threshold and related code were changed to 3 in the year
2019 in:
commit 4a41f453be ("tcp: change pingpong threshold to 3")
And reverted to 1 in the year 2022 in:
commit 4d8f24eeed ("Revert "tcp: change pingpong threshold to 3"")
There is no single value that fits all applications.
Add net.ipv4.tcp_pingpong_thresh sysctl tunable, so it can be tuned for
optimal performance based on the application needs.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/1697056244-21888-1-git-send-email-haiyangz@microsoft.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This change adds a sysctl to opt-out of RFC4862 section 5.5.3e's valid
lifetime derivation mechanism.
RFC4862 section 5.5.3e prescribes that the valid lifetime in a Router
Advertisement PIO shall be ignored if it less than 2 hours and to reset
the lifetime of the corresponding address to 2 hours. An in-progress
6man draft (see draft-ietf-6man-slaac-renum-07 section 4.2) is currently
looking to remove this mechanism. While this draft has not been moving
particularly quickly for other reasons, there is widespread consensus on
section 4.2 which updates RFC4862 section 5.5.3e.
Cc: Maciej Żenczykowski <maze@google.com>
Cc: Lorenzo Colitti <lorenzo@google.com>
Cc: Jen Linkova <furry@google.com>
Signed-off-by: Patrick Rohr <prohr@google.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20230925214711.959704-1-prohr@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This idea came after a particular workload requested
the quickack attribute set on routes, and a performance
drop was noticed for large bulk transfers.
For high throughput flows, it is best to use one cpu
running the user thread issuing socket system calls,
and a separate cpu to process incoming packets from BH context.
(With TSO/GRO, bottleneck is usually the 'user' cpu)
Problem is the user thread can spend a lot of time while holding
the socket lock, forcing BH handler to queue most of incoming
packets in the socket backlog.
Whenever the user thread releases the socket lock, it must first
process all accumulated packets in the backlog, potentially
adding latency spikes. Due to flood mitigation, having too many
packets in the backlog increases chance of unexpected drops.
Backlog processing unfortunately shifts a fair amount of cpu cycles
from the BH cpu to the 'user' cpu, thus reducing max throughput.
This patch takes advantage of the backlog processing,
and the fact that ACK are mostly cumulative.
The idea is to detect we are in the backlog processing
and defer all eligible ACK into a single one,
sent from tcp_release_cb().
This saves cpu cycles on both sides, and network resources.
Performance of a single TCP flow on a 200Gbit NIC:
- Throughput is increased by 20% (100Gbit -> 120Gbit).
- Number of generated ACK per second shrinks from 240,000 to 40,000.
- Number of backlog drops per second shrinks from 230 to 0.
Benchmark context:
- Regular netperf TCP_STREAM (no zerocopy)
- Intel(R) Xeon(R) Platinum 8481C (Saphire Rapids)
- MAX_SKB_FRAGS = 17 (~60KB per GRO packet)
This feature is guarded by a new sysctl, and enabled by default:
/proc/sys/net/ipv4/tcp_backlog_ack_defer
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Dave Taht <dave.taht@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
accept_ra_min_rtr_lft only considered the lifetime of the default route
and discarded entire RAs accordingly.
This change renames accept_ra_min_rtr_lft to accept_ra_min_lft, and
applies the value to individual RA sections; in particular, router
lifetime, PIO preferred lifetime, and RIO lifetime. If any of those
lifetimes are lower than the configured value, the specific RA section
is ignored.
In order for the sysctl to be useful to Android, it should really apply
to all lifetimes in the RA, since that is what determines the minimum
frequency at which RAs must be processed by the kernel. Android uses
hardware offloads to drop RAs for a fraction of the minimum of all
lifetimes present in the RA (some networks have very frequent RAs (5s)
with high lifetimes (2h)). Despite this, we have encountered networks
that set the router lifetime to 30s which results in very frequent CPU
wakeups. Instead of disabling IPv6 (and dropping IPv6 ethertype in the
WiFi firmware) entirely on such networks, it seems better to ignore the
misconfigured routers while still processing RAs from other IPv6 routers
on the same network (i.e. to support IoT applications).
The previous implementation dropped the entire RA based on router
lifetime. This turned out to be hard to expand to the other lifetimes
present in the RA in a consistent manner; dropping the entire RA based
on RIO/PIO lifetimes would essentially require parsing the whole thing
twice.
Fixes: 1671bcfd76 ("net: add sysctl accept_ra_min_rtr_lft")
Cc: Lorenzo Colitti <lorenzo@google.com>
Signed-off-by: Patrick Rohr <prohr@google.com>
Reviewed-by: Maciej Żenczykowski <maze@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20230726230701.919212-1-prohr@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This change adds a new sysctl accept_ra_min_rtr_lft to specify the
minimum acceptable router lifetime in an RA. If the received RA router
lifetime is less than the configured value (and not 0), the RA is
ignored.
This is useful for mobile devices, whose battery life can be impacted
by networks that configure RAs with a short lifetime. On such networks,
the device should never gain IPv6 provisioning and should attempt to
drop RAs via hardware offload, if available.
Signed-off-by: Patrick Rohr <prohr@google.com>
Cc: Maciej Żenczykowski <maze@google.com>
Cc: Lorenzo Colitti <lorenzo@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
With modern NIC drivers shifting to full page allocations per
received frame, we face the following issue:
TCP has one per-netns sysctl used to tweak how to translate
a memory use into an expected payload (RWIN), in RX path.
tcp_win_from_space() implementation is limited to few cases.
For hosts dealing with various MSS, we either under estimate
or over estimate the RWIN we send to the remote peers.
For instance with the default sysctl_tcp_adv_win_scale value,
we expect to store 50% of payload per allocated chunk of memory.
For the typical use of MTU=1500 traffic, and order-0 pages allocations
by NIC drivers, we are sending too big RWIN, leading to potential
tcp collapse operations, which are extremely expensive and source
of latency spikes.
This patch makes sysctl_tcp_adv_win_scale obsolete, and instead
uses a per socket scaling factor, so that we can precisely
adjust the RWIN based on effective skb->len/skb->truesize ratio.
This patch alone can double TCP receive performance when receivers
are too slow to drain their receive queue, or by allowing
a bigger RWIN when MSS is close to PAGE_SIZE.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Link: https://lore.kernel.org/r/20230717152917.751987-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Under certain circumstances, the tcp receive buffer memory limit
set by autotuning (sk_rcvbuf) is increased due to incoming data
packets as a result of the window not closing when it should be.
This can result in the receive buffer growing all the way up to
tcp_rmem[2], even for tcp sessions with a low BDP.
To reproduce: Connect a TCP session with the receiver doing
nothing and the sender sending small packets (an infinite loop
of socket send() with 4 bytes of payload with a sleep of 1 ms
in between each send()). This will cause the tcp receive buffer
to grow all the way up to tcp_rmem[2].
As a result, a host can have individual tcp sessions with receive
buffers of size tcp_rmem[2], and the host itself can reach tcp_mem
limits, causing the host to go into tcp memory pressure mode.
The fundamental issue is the relationship between the granularity
of the window scaling factor and the number of byte ACKed back
to the sender. This problem has previously been identified in
RFC 7323, appendix F [1].
The Linux kernel currently adheres to never shrinking the window.
In addition to the overallocation of memory mentioned above, the
current behavior is functionally incorrect, because once tcp_rmem[2]
is reached when no remediations remain (i.e. tcp collapse fails to
free up any more memory and there are no packets to prune from the
out-of-order queue), the receiver will drop in-window packets
resulting in retransmissions and an eventual timeout of the tcp
session. A receive buffer full condition should instead result
in a zero window and an indefinite wait.
In practice, this problem is largely hidden for most flows. It
is not applicable to mice flows. Elephant flows can send data
fast enough to "overrun" the sk_rcvbuf limit (in a single ACK),
triggering a zero window.
But this problem does show up for other types of flows. Examples
are websockets and other type of flows that send small amounts of
data spaced apart slightly in time. In these cases, we directly
encounter the problem described in [1].
RFC 7323, section 2.4 [2], says there are instances when a retracted
window can be offered, and that TCP implementations MUST ensure
that they handle a shrinking window, as specified in RFC 1122,
section 4.2.2.16 [3]. All prior RFCs on the topic of tcp window
management have made clear that sender must accept a shrunk window
from the receiver, including RFC 793 [4] and RFC 1323 [5].
This patch implements the functionality to shrink the tcp window
when necessary to keep the right edge within the memory limit by
autotuning (sk_rcvbuf). This new functionality is enabled with
the new sysctl: net.ipv4.tcp_shrink_window
Additional information can be found at:
https://blog.cloudflare.com/unbounded-memory-usage-by-tcp-for-receive-buffers-and-how-we-fixed-it/
[1] https://www.rfc-editor.org/rfc/rfc7323#appendix-F
[2] https://www.rfc-editor.org/rfc/rfc7323#section-2.4
[3] https://www.rfc-editor.org/rfc/rfc1122#page-91
[4] https://www.rfc-editor.org/rfc/rfc793
[5] https://www.rfc-editor.org/rfc/rfc1323
Signed-off-by: Mike Freemon <mfreemon@cloudflare.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Cross-merge networking fixes after downstream PR.
Conflicts:
net/sched/sch_taprio.c
d636fc5dd6 ("net: sched: add rcu annotations around qdisc->qdisc_sleeping")
dced11ef84 ("net/sched: taprio: don't overwrite "sch" variable in taprio_dump_class_stats()")
net/ipv4/sysctl_net_ipv4.c
e209fee411 ("net/ipv4: ping_group_range: allow GID from 2147483648 to 4294967294")
ccce324dab ("tcp: make the first N SYN RTO backoffs linear")
https://lore.kernel.org/all/20230605100816.08d41a7b@canb.auug.org.au/
No adjacent changes.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
With this commit, all the GIDs ("0 4294967294") can be written to the
"net.ipv4.ping_group_range" sysctl.
Note that 4294967295 (0xffffffff) is an invalid GID (see gid_valid() in
include/linux/uidgid.h), and an attempt to register this number will cause
-EINVAL.
Prior to this commit, only up to GID 2147483647 could be covered.
Documentation/networking/ip-sysctl.rst had "0 4294967295" as an example
value, but this example was wrong and causing -EINVAL.
Fixes: c319b4d76b ("net: ipv4: add IPPROTO_ICMP socket kind")
Co-developed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: Akihiro Suda <akihiro.suda.cz@hco.ntt.co.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently the SYN RTO schedule follows an exponential backoff
scheme, which can be unnecessarily conservative in cases where
there are link failures. In such cases, it's better to
aggressively try to retransmit packets, so it takes routers
less time to find a repath with a working link.
We chose a default value for this sysctl of 4, to follow
the macOS and IOS backoff scheme of 1,1,1,1,1,2,4,8, ...
MacOS and IOS have used this backoff schedule for over
a decade, since before this 2009 IETF presentation
discussed the behavior:
https://www.ietf.org/proceedings/75/slides/tcpm-1.pdf
This commit makes the SYN RTO schedule start with a number of
linear backoffs given by the following sysctl:
* tcp_syn_linear_timeouts
This changes the SYN RTO scheme to be: init_rto_val for
tcp_syn_linear_timeouts, exp backoff starting at init_rto_val
For example if init_rto_val = 1 and tcp_syn_linear_timeouts = 2, our
backoff scheme would be: 1, 1, 1, 2, 4, 8, 16, ...
Signed-off-by: David Morley <morleyd@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Tested-by: David Morley <morleyd@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20230509180558.2541885-1-morleyd.kernel@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
ICMPv6 error packets are not sent to the anycast destinations and this
prevents things like traceroute from working. So create a setting similar
to ECHO when dealing with Anycast sources (icmpv6_echo_ignore_anycast).
Signed-off-by: Mahesh Bandewar <maheshb@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Maciej Żenczykowski <maze@google.com>
Link: https://lore.kernel.org/r/20230419013238.2691167-1-maheshb@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
UBSAN: shift-out-of-bounds in net/ipv4/tcp_input.c:555:23
shift exponent 255 is too large for 32-bit type 'int'
CPU: 1 PID: 7907 Comm: ssh Not tainted 6.3.0-rc4-00161-g62bad54b26db-dirty #206
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.15.0-1 04/01/2014
Call Trace:
<TASK>
dump_stack_lvl+0x136/0x150
__ubsan_handle_shift_out_of_bounds+0x21f/0x5a0
tcp_init_transfer.cold+0x3a/0xb9
tcp_finish_connect+0x1d0/0x620
tcp_rcv_state_process+0xd78/0x4d60
tcp_v4_do_rcv+0x33d/0x9d0
__release_sock+0x133/0x3b0
release_sock+0x58/0x1b0
'maxwin' is int, shifting int for 32 or more bits is undefined behaviour.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When set to zero, the neighbor sysctl proxy_delay value
does not cause an immediate reply for ARP/ND requests
as expected, it instead causes a random delay between
[0, U32_MAX). Looking at this comment from
__get_random_u32_below() explains the reason:
/*
* This function is technically undefined for ceil == 0, and in fact
* for the non-underscored constant version in the header, we build bug
* on that. But for the non-constant case, it's convenient to have that
* evaluate to being a straight call to get_random_u32(), so that
* get_random_u32_inclusive() can work over its whole range without
* undefined behavior.
*/
Added helper function that does not call get_random_u32_below()
if proxy_delay is zero and just uses the current value of
jiffies instead, causing pneigh_enqueue() to respond
immediately.
Also added definition of proxy_delay to ip-sysctl.txt since
it was missing.
Signed-off-by: Brian Haley <haleyb.dev@gmail.com>
Link: https://lore.kernel.org/r/20230130171428.367111-1-haleyb.dev@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Correct spelling problems for Documentation/networking/ as reported
by codespell.
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: linux-doc@vger.kernel.org
Cc: Jiri Pirko <jiri@nvidia.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Paolo Abeni <pabeni@redhat.com>
Cc: netdev@vger.kernel.org
Link: https://lore.kernel.org/r/20230129231053.20863-5-rdunlap@infradead.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This patch is to add sysctl net.sctp.l3mdev_accept to allow
users to change the pernet global l3mdev_accept.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The maximum hash table size is 64K due to the nature of the protocol. [0]
It's smaller than TCP, and fewer sockets can cause a performance drop.
On an EC2 c5.24xlarge instance (192 GiB memory), after running iperf3 in
different netns, creating 32Mi sockets without data transfer in the root
netns causes regression for the iperf3's connection.
uhash_entries sockets length Gbps
64K 1 1 5.69
1Mi 16 5.27
2Mi 32 4.90
4Mi 64 4.09
8Mi 128 2.96
16Mi 256 2.06
32Mi 512 1.12
The per-netns hash table breaks the lengthy lists into shorter ones. It is
useful on a multi-tenant system with thousands of netns. With smaller hash
tables, we can look up sockets faster, isolate noisy neighbours, and reduce
lock contention.
The max size of the per-netns table is 64K as well. This is because the
possible hash range by udp_hashfn() always fits in 64K within the same
netns and we cannot make full use of the whole buckets larger than 64K.
/* 0 < num < 64K -> X < hash < X + 64K */
(num + net_hash_mix(net)) & mask;
Also, the min size is 128. We use a bitmap to search for an available
port in udp_lib_get_port(). To keep the bitmap on the stack and not
fire the CONFIG_FRAME_WARN error at build time, we round up the table
size to 128.
The sysctl usage is the same with TCP:
$ dmesg | cut -d ' ' -f 6- | grep "UDP hash"
UDP hash table entries: 65536 (order: 9, 2097152 bytes, vmalloc)
# sysctl net.ipv4.udp_hash_entries
net.ipv4.udp_hash_entries = 65536 # can be changed by uhash_entries
# sysctl net.ipv4.udp_child_hash_entries
net.ipv4.udp_child_hash_entries = 0 # disabled by default
# ip netns add test1
# ip netns exec test1 sysctl net.ipv4.udp_hash_entries
net.ipv4.udp_hash_entries = -65536 # share the global table
# sysctl -w net.ipv4.udp_child_hash_entries=100
net.ipv4.udp_child_hash_entries = 100
# ip netns add test2
# ip netns exec test2 sysctl net.ipv4.udp_hash_entries
net.ipv4.udp_hash_entries = 128 # own a per-netns table with 2^n buckets
We could optimise the hash table lookup/iteration further by removing
the netns comparison for the per-netns one in the future. Also, we
could optimise the sparse udp_hslot layout by putting it in udp_table.
[0]: https://lore.kernel.org/netdev/4ACC2815.7010101@gmail.com/
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
PLB (Protective Load Balancing) is a host based mechanism for load
balancing across switch links. It leverages congestion signals(e.g. ECN)
from transport layer to randomly change the path of the connection
experiencing congestion. PLB changes the path of the connection by
changing the outgoing IPv6 flow label for IPv6 connections (implemented
in Linux by calling sk_rethink_txhash()). Because of this implementation
mechanism, PLB can currently only work for IPv6 traffic. For more
information, see the SIGCOMM 2022 paper:
https://doi.org/10.1145/3544216.3544226
This commit adds new sysctl knobs and sets their default values for
TCP PLB.
Signed-off-by: Mubashir Adnan Qureshi <mubashirq@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The more sockets we have in the hash table, the longer we spend looking
up the socket. While running a number of small workloads on the same
host, they penalise each other and cause performance degradation.
The root cause might be a single workload that consumes much more
resources than the others. It often happens on a cloud service where
different workloads share the same computing resource.
On EC2 c5.24xlarge instance (196 GiB memory and 524288 (1Mi / 2) ehash
entries), after running iperf3 in different netns, creating 24Mi sockets
without data transfer in the root netns causes about 10% performance
regression for the iperf3's connection.
thash_entries sockets length Gbps
524288 1 1 50.7
24Mi 48 45.1
It is basically related to the length of the list of each hash bucket.
For testing purposes to see how performance drops along the length,
I set 131072 (1Mi / 8) to thash_entries, and here's the result.
thash_entries sockets length Gbps
131072 1 1 50.7
1Mi 8 49.9
2Mi 16 48.9
4Mi 32 47.3
8Mi 64 44.6
16Mi 128 40.6
24Mi 192 36.3
32Mi 256 32.5
40Mi 320 27.0
48Mi 384 25.0
To resolve the socket lookup degradation, we introduce an optional
per-netns hash table for TCP, but it's just ehash, and we still share
the global bhash, bhash2 and lhash2.
With a smaller ehash, we can look up non-listener sockets faster and
isolate such noisy neighbours. In addition, we can reduce lock contention.
We can control the ehash size by a new sysctl knob. However, depending
on workloads, it will require very sensitive tuning, so we disable the
feature by default (net.ipv4.tcp_child_ehash_entries == 0). Moreover,
we can fall back to using the global ehash in case we fail to allocate
enough memory for a new ehash. The maximum size is 16Mi, which is large
enough that even if we have 48Mi sockets, the average list length is 3,
and regression would be less than 1%.
We can check the current ehash size by another read-only sysctl knob,
net.ipv4.tcp_ehash_entries. A negative value means the netns shares
the global ehash (per-netns ehash is disabled or failed to allocate
memory).
# dmesg | cut -d ' ' -f 5- | grep "established hash"
TCP established hash table entries: 524288 (order: 10, 4194304 bytes, vmalloc hugepage)
# sysctl net.ipv4.tcp_ehash_entries
net.ipv4.tcp_ehash_entries = 524288 # can be changed by thash_entries
# sysctl net.ipv4.tcp_child_ehash_entries
net.ipv4.tcp_child_ehash_entries = 0 # disabled by default
# ip netns add test1
# ip netns exec test1 sysctl net.ipv4.tcp_ehash_entries
net.ipv4.tcp_ehash_entries = -524288 # share the global ehash
# sysctl -w net.ipv4.tcp_child_ehash_entries=100
net.ipv4.tcp_child_ehash_entries = 100
# ip netns add test2
# ip netns exec test2 sysctl net.ipv4.tcp_ehash_entries
net.ipv4.tcp_ehash_entries = 128 # own a per-netns ehash with 2^n buckets
When more than two processes in the same netns create per-netns ehash
concurrently with different sizes, we need to guarantee the size in
one of the following ways:
1) Share the global ehash and create per-netns ehash
First, unshare() with tcp_child_ehash_entries==0. It creates dedicated
netns sysctl knobs where we can safely change tcp_child_ehash_entries
and clone()/unshare() to create a per-netns ehash.
2) Control write on sysctl by BPF
We can use BPF_PROG_TYPE_CGROUP_SYSCTL to allow/deny read/write on
sysctl knobs.
Note that the global ehash allocated at the boot time is spread over
available NUMA nodes, but inet_pernet_hashinfo_alloc() will allocate
pages for each per-netns ehash depending on the current process's NUMA
policy. By default, the allocation is done in the local node only, so
the per-netns hash table could fully reside on a random node. Thus,
depending on the NUMA policy the netns is created with and the CPU the
current thread is running on, we could see some performance differences
for highly optimised networking applications.
Note also that the default values of two sysctl knobs depend on the ehash
size and should be tuned carefully:
tcp_max_tw_buckets : tcp_child_ehash_entries / 2
tcp_max_syn_backlog : max(128, tcp_child_ehash_entries / 128)
As a bonus, we can dismantle netns faster. Currently, while destroying
netns, we call inet_twsk_purge(), which walks through the global ehash.
It can be potentially big because it can have many sockets other than
TIME_WAIT in all netns. Splitting ehash changes that situation, where
it's only necessary for inet_twsk_purge() to clean up TIME_WAIT sockets
in each netns.
With regard to this, we do not free the per-netns ehash in inet_twsk_kill()
to avoid UAF while iterating the per-netns ehash in inet_twsk_purge().
Instead, we do it in tcp_sk_exit_batch() after calling tcp_twsk_purge() to
keep it protocol-family-independent.
In the future, we could optimise ehash lookup/iteration further by removing
netns comparison for the per-netns ehash.
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Because per host rate limiting has been proven problematic (side channel
attacks can be based on it), per host rate limiting of challenge acks ideally
should be per netns and turned off by default.
This is a long due followup of following commits:
083ae30828 ("tcp: enable per-socket rate limiting of all 'challenge acks'")
f2b2c582e8 ("tcp: mitigate ACK loops for connections as tcp_sock")
75ff39ccc1 ("tcp: make challenge acks less predictable")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Jason Baron <jbaron@akamai.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The tcp_reflect_tos option was introduced in Linux 5.10 but was still
undocumented.
Signed-off-by: Guillaume Nault <gnault@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since commit 1033990ac5 ("sctp: implement memory accounting on tx path"),
SCTP has supported memory accounting on tx path where 'sctp_wmem' is used
by sk_wmem_schedule(). So we should fix the description for this option in
ip-sysctl.rst accordingly.
v1->v2:
- Improve the description as Marcelo suggested.
Fixes: 1033990ac5 ("sctp: implement memory accounting on tx path")
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds a third knob, '2', which extends the
accept_untracked_na option to learn a neighbor only if the src ip is
in the same subnet as an address configured on the interface that
received the neighbor advertisement. This is similar to the arp_accept
configuration for ipv4.
Signed-off-by: Jaehee Park <jhpark1013@gmail.com>
Suggested-by: Roopa Prabhu <roopa@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In many deployments, we want the option to not learn a neighbor from
garp if the src ip is not in the same subnet as an address configured
on the interface that received the garp message. net.ipv4.arp_accept
sysctl is currently used to control creation of a neigh from a
received garp packet. This patch adds a new option '2' to
net.ipv4.arp_accept which extends option '1' by including the subnet
check.
Signed-off-by: Jaehee Park <jhpark1013@gmail.com>
Suggested-by: Roopa Prabhu <roopa@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
While reading sysctl_ip_dynaddr, it can be changed concurrently.
Thus, we need to add READ_ONCE() to its readers.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
While reading cipso sysctl variables, they can be changed concurrently.
So, we need to add READ_ONCE() to avoid data-races.
Fixes: 446fda4f26 ("[NetLabel]: CIPSOv4 engine")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Acked-by: Paul Moore <paul@paul-moore.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
commit ed6cd6a178 ("net, neigh: Set lower cap for neigh_managed_work rearming")
fixed a case when DELAY_PROBE_TIME is configured to 0, the processing of the
system work queue hog CPU to 100%, and further more we should introduce
a new option used by periodic probe
Signed-off-by: Yuwei Wang <wangyuweihx@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Describe it in networking/ip-sysctl.rst like other SCTP options.
Fixes: 2f5268a924 ("sctp: allow users to set netns ecn flag with sysctl")
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Describe it in networking/ip-sysctl.rst like other SCTP options.
We need to document this especially as when using the feature
of User Message Interleaving, some socket options also needs
to be set.
Fixes: 463118c34a ("sctp: support sysctl to allow users to use stream interleave")
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Describe it in networking/ip-sysctl.rst like other SCTP options.
Fixes: c0d8bab6ae ("sctp: add get and set sockopt for reconf_enable")
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
RFC 9131 changes default behaviour of handling RX of NA messages when the
corresponding entry is absent in the neighbour cache. The current
implementation is limited to accept just unsolicited NAs. However, the
RFC is more generic where it also accepts solicited NAs. Both types
should result in adding a STALE entry for this case.
Expand accept_untracked_na behaviour to also accept solicited NAs to
be compliant with the RFC and rename the sysctl knob to
accept_untracked_na.
Fixes: f9a2fb7331 ("net/ipv6: Introduce accept_unsolicited_na knob to implement router-side changes for RFC9131")
Signed-off-by: Arun Ajith S <aajith@arista.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20220530101414.65439-1-aajith@arista.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Let's describe this sysctl.
Fixes: 5cbf777cfd ("route: add support for directed broadcast forwarding")
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a new neighbour cache entry in STALE state for routers on receiving
an unsolicited (gratuitous) neighbour advertisement with
target link-layer-address option specified.
This is similar to the arp_accept configuration for IPv4.
A new sysctl endpoint is created to turn on this behaviour:
/proc/sys/net/ipv6/conf/interface/accept_unsolicited_na.
Signed-off-by: Arun Ajith S <aajith@arista.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Back when tcp_tso_autosize() and TCP pacing were introduced,
our focus was really to reduce burst sizes for long distance
flows.
The simple heuristic of using sk_pacing_rate/1024 has worked
well, but can lead to too small packets for hosts in the same
rack/cluster, when thousands of flows compete for the bottleneck.
Neal Cardwell had the idea of making the TSO burst size
a function of both sk_pacing_rate and tcp_min_rtt()
Indeed, for local flows, sending bigger bursts is better
to reduce cpu costs, as occasional losses can be repaired
quite fast.
This patch is based on Neal Cardwell implementation
done more than two years ago.
bbr is adjusting max_pacing_rate based on measured bandwidth,
while cubic would over estimate max_pacing_rate.
/proc/sys/net/ipv4/tcp_tso_rtt_log can be used to tune or disable
this new feature, in logarithmic steps.
Tested:
100Gbit NIC, two hosts in the same rack, 4K MTU.
600 flows rate-limited to 20000000 bytes per second.
Before patch: (TSO sizes would be limited to 20000000/1024/4096 -> 4 segments per TSO)
~# echo 0 >/proc/sys/net/ipv4/tcp_tso_rtt_log
~# nstat -n;perf stat ./super_netperf 600 -H otrv6 -l 20 -- -K dctcp -q 20000000;nstat|egrep "TcpInSegs|TcpOutSegs|TcpRetransSegs|Delivered"
96005
Performance counter stats for './super_netperf 600 -H otrv6 -l 20 -- -K dctcp -q 20000000':
65,945.29 msec task-clock # 2.845 CPUs utilized
1,314,632 context-switches # 19935.279 M/sec
5,292 cpu-migrations # 80.249 M/sec
940,641 page-faults # 14264.023 M/sec
201,117,030,926 cycles # 3049769.216 GHz (83.45%)
17,699,435,405 stalled-cycles-frontend # 8.80% frontend cycles idle (83.48%)
136,584,015,071 stalled-cycles-backend # 67.91% backend cycles idle (83.44%)
53,809,530,436 instructions # 0.27 insn per cycle
# 2.54 stalled cycles per insn (83.36%)
9,062,315,523 branches # 137422329.563 M/sec (83.22%)
153,008,621 branch-misses # 1.69% of all branches (83.32%)
23.182970846 seconds time elapsed
TcpInSegs 15648792 0.0
TcpOutSegs 58659110 0.0 # Average of 3.7 4K segments per TSO packet
TcpExtTCPDelivered 58654791 0.0
TcpExtTCPDeliveredCE 19 0.0
After patch:
~# echo 9 >/proc/sys/net/ipv4/tcp_tso_rtt_log
~# nstat -n;perf stat ./super_netperf 600 -H otrv6 -l 20 -- -K dctcp -q 20000000;nstat|egrep "TcpInSegs|TcpOutSegs|TcpRetransSegs|Delivered"
96046
Performance counter stats for './super_netperf 600 -H otrv6 -l 20 -- -K dctcp -q 20000000':
48,982.58 msec task-clock # 2.104 CPUs utilized
186,014 context-switches # 3797.599 M/sec
3,109 cpu-migrations # 63.472 M/sec
941,180 page-faults # 19214.814 M/sec
153,459,763,868 cycles # 3132982.807 GHz (83.56%)
12,069,861,356 stalled-cycles-frontend # 7.87% frontend cycles idle (83.32%)
120,485,917,953 stalled-cycles-backend # 78.51% backend cycles idle (83.24%)
36,803,672,106 instructions # 0.24 insn per cycle
# 3.27 stalled cycles per insn (83.18%)
5,947,266,275 branches # 121417383.427 M/sec (83.64%)
87,984,616 branch-misses # 1.48% of all branches (83.43%)
23.281200256 seconds time elapsed
TcpInSegs 1434706 0.0
TcpOutSegs 58883378 0.0 # Average of 41 4K segments per TSO packet
TcpExtTCPDelivered 58878971 0.0
TcpExtTCPDeliveredCE 9664 0.0
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Link: https://lore.kernel.org/r/20220309015757.2532973-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The updating way of pmtu has changed, but documentation is still in the
old way. So this patch updates the interpretation of ip_no_pmtu_disc and
min_pmtu.
See commit 28d35bcdd3 ("net: ipv4: don't let PMTU updates increase
route MTU")
Reported-by: Zeal Robot <zealci@zte.com.cn>
Signed-off-by: xu xin <xu.xin16@zte.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
udp_mem is a vector of 3 INTEGERs, which is used to limit the number of
pages allowed for queueing by all UDP sockets.
However, sk_has_memory_pressure() in __sk_mem_raise_allocated() always
return false for udp, as memory pressure is not supported by udp, which
means that __sk_mem_raise_allocated() will fail once pages allocated
for udp socket exceeds udp_mem[0].
Therefor, udp_mem[0] is the only one that limit the number of pages.
However, the document of udp_mem just express that udp_mem[2] is the
limitation. So, just fix it.
Signed-off-by: Menglong Dong <imagedong@tencent.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In most situations the neighbor discovery cache should be cleared on a
NOCARRIER event which is currently done unconditionally. But for wireless
roams the neighbor discovery cache can and should remain intact since
the underlying network has not changed.
This patch introduces a sysctl option ndisc_evict_nocarrier which can
be disabled by a wireless supplicant during a roam. This allows packets
to be sent after a roam immediately without having to wait for
neighbor discovery.
A user reported roughly a 1 second delay after a roam before packets
could be sent out (note, on IPv4). This delay was due to the ARP
cache being cleared. During testing of this same scenario using IPv6
no delay was noticed, but regardless there is no reason to clear
the ndisc cache for wireless roams.
Signed-off-by: James Prestwood <prestwoj@gmail.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This change introduces a new sysctl parameter, arp_evict_nocarrier.
When set (default) the ARP cache will be cleared on a NOCARRIER event.
This new option has been defaulted to '1' which maintains existing
behavior.
Clearing the ARP cache on NOCARRIER is relatively new, introduced by:
commit 859bd2ef1f
Author: David Ahern <dsahern@gmail.com>
Date: Thu Oct 11 20:33:49 2018 -0700
net: Evict neighbor entries on carrier down
The reason for this changes is to prevent the ARP cache from being
cleared when a wireless device roams. Specifically for wireless roams
the ARP cache should not be cleared because the underlying network has not
changed. Clearing the ARP cache in this case can introduce significant
delays sending out packets after a roam.
A user reported such a situation here:
https://lore.kernel.org/linux-wireless/CACsRnHWa47zpx3D1oDq9JYnZWniS8yBwW1h0WAVZ6vrbwL_S0w@mail.gmail.com/
After some investigation it was found that the kernel was holding onto
packets until ARP finished which resulted in this 1 second delay. It
was also found that the first ARP who-has was never responded to,
which is actually what caues the delay. This change is more or less
working around this behavior, but again, there is no reason to clear
the cache on a roam anyways.
As for the unanswered who-has, we know the packet made it OTA since
it was seen while monitoring. Why it never received a response is
unknown. In any case, since this is a problem on the AP side of things
all that can be done is to work around it until it is solved.
Some background on testing/reproducing the packet delay:
Hardware:
- 2 access points configured for Fast BSS Transition (Though I don't
see why regular reassociation wouldn't have the same behavior)
- Wireless station running IWD as supplicant
- A device on network able to respond to pings (I used one of the APs)
Procedure:
- Connect to first AP
- Ping once to establish an ARP entry
- Start a tcpdump
- Roam to second AP
- Wait for operstate UP event, and note the timestamp
- Start pinging
Results:
Below is the tcpdump after UP. It was recorded the interface went UP at
10:42:01.432875.
10:42:01.461871 ARP, Request who-has 192.168.254.1 tell 192.168.254.71, length 28
10:42:02.497976 ARP, Request who-has 192.168.254.1 tell 192.168.254.71, length 28
10:42:02.507162 ARP, Reply 192.168.254.1 is-at ac:86:74:55:b0:20, length 46
10:42:02.507185 IP 192.168.254.71 > 192.168.254.1: ICMP echo request, id 52792, seq 1, length 64
10:42:02.507205 IP 192.168.254.71 > 192.168.254.1: ICMP echo request, id 52792, seq 2, length 64
10:42:02.507212 IP 192.168.254.71 > 192.168.254.1: ICMP echo request, id 52792, seq 3, length 64
10:42:02.507219 IP 192.168.254.71 > 192.168.254.1: ICMP echo request, id 52792, seq 4, length 64
10:42:02.507225 IP 192.168.254.71 > 192.168.254.1: ICMP echo request, id 52792, seq 5, length 64
10:42:02.507232 IP 192.168.254.71 > 192.168.254.1: ICMP echo request, id 52792, seq 6, length 64
10:42:02.515373 IP 192.168.254.1 > 192.168.254.71: ICMP echo reply, id 52792, seq 1, length 64
10:42:02.521399 IP 192.168.254.1 > 192.168.254.71: ICMP echo reply, id 52792, seq 2, length 64
10:42:02.521612 IP 192.168.254.1 > 192.168.254.71: ICMP echo reply, id 52792, seq 3, length 64
10:42:02.521941 IP 192.168.254.1 > 192.168.254.71: ICMP echo reply, id 52792, seq 4, length 64
10:42:02.522419 IP 192.168.254.1 > 192.168.254.71: ICMP echo reply, id 52792, seq 5, length 64
10:42:02.523085 IP 192.168.254.1 > 192.168.254.71: ICMP echo reply, id 52792, seq 6, length 64
You can see the first ARP who-has went out very quickly after UP, but
was never responded to. Nearly a second later the kernel retries and
gets a response. Only then do the ping packets go out. If an ARP entry
is manually added prior to UP (after the cache is cleared) it is seen
that the first ping is never responded to, so its not only an issue with
ARP but with data packets in general.
As mentioned prior, the wireless interface was also monitored to verify
the ping/ARP packet made it OTA which was observed to be true.
Signed-off-by: James Prestwood <prestwoj@gmail.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This reverts the following patches :
- commit 2e05fcae83 ("tcp: fix compile error if !CONFIG_SYSCTL")
- commit 4f661542a4 ("tcp: fix zerocopy and notsent_lowat issues")
- commit 472c2e07ee ("tcp: add one skb cache for tx")
- commit 8b27dae5a2 ("tcp: add one skb cache for rx")
Having a cache of one skb (in each direction) per TCP socket is fragile,
since it can cause a significant increase of memory needs,
and not good enough for high speed flows anyway where more than one skb
is needed.
We want instead to add a generic infrastructure, with more flexible
per-cpu caches, for alien NUMA nodes.
Acked-by: Paolo Abeni <pabeni@redhat.com>
Acked-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Alex Henrie