IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
Commits a602456 ("udp: Add GRO functions to UDP socket") and 57c67ff ("udp:
additional GRO support") introduce incorrect usage of {ip,ipv6}_hdr in the
complete phase of gro. The functions always return skb->network_header,
which in the case of encapsulated packets at the gro complete phase, is
always set to the innermost L3 of the packet. That means that calling
{ip,ipv6}_hdr for skbs which completed the GRO receive phase (both in
gro_list and *_gro_complete) when parsing an encapsulated packet's _outer_
L3/L4 may return an unexpected value.
This incorrect usage leads to a bug in GRO's UDP socket lookup.
udp{4,6}_lib_lookup_skb functions use ip_hdr/ipv6_hdr respectively. These
*_hdr functions return network_header which will point to the innermost L3,
resulting in the wrong offset being used in __udp{4,6}_lib_lookup with
encapsulated packets.
This patch adds network_offset and inner_network_offset to napi_gro_cb, and
makes sure both are set correctly.
To fix the issue, network_offsets union is used inside napi_gro_cb, in
which both the outer and the inner network offsets are saved.
Reproduction example:
Endpoint configuration example (fou + local address bind)
# ip fou add port 6666 ipproto 4
# ip link add name tun1 type ipip remote 2.2.2.1 local 2.2.2.2 encap fou encap-dport 5555 encap-sport 6666 mode ipip
# ip link set tun1 up
# ip a add 1.1.1.2/24 dev tun1
Netperf TCP_STREAM result on net-next before patch is applied:
net-next main, GRO enabled:
$ netperf -H 1.1.1.2 -t TCP_STREAM -l 5
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
131072 16384 16384 5.28 2.37
net-next main, GRO disabled:
$ netperf -H 1.1.1.2 -t TCP_STREAM -l 5
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
131072 16384 16384 5.01 2745.06
patch applied, GRO enabled:
$ netperf -H 1.1.1.2 -t TCP_STREAM -l 5
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
131072 16384 16384 5.01 2877.38
Fixes: a6024562ffd7 ("udp: Add GRO functions to UDP socket")
Signed-off-by: Richard Gobert <richardbgobert@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
KMSAN reported uninit-value access in __ip_make_skb() [1]. __ip_make_skb()
tests HDRINCL to know if the skb has icmphdr. However, HDRINCL can cause a
race condition. If calling setsockopt(2) with IP_HDRINCL changes HDRINCL
while __ip_make_skb() is running, the function will access icmphdr in the
skb even if it is not included. This causes the issue reported by KMSAN.
Check FLOWI_FLAG_KNOWN_NH on fl4->flowi4_flags instead of testing HDRINCL
on the socket.
Also, fl4->fl4_icmp_type and fl4->fl4_icmp_code are not initialized. These
are union in struct flowi4 and are implicitly initialized by
flowi4_init_output(), but we should not rely on specific union layout.
Initialize these explicitly in raw_sendmsg().
[1]
BUG: KMSAN: uninit-value in __ip_make_skb+0x2b74/0x2d20 net/ipv4/ip_output.c:1481
__ip_make_skb+0x2b74/0x2d20 net/ipv4/ip_output.c:1481
ip_finish_skb include/net/ip.h:243 [inline]
ip_push_pending_frames+0x4c/0x5c0 net/ipv4/ip_output.c:1508
raw_sendmsg+0x2381/0x2690 net/ipv4/raw.c:654
inet_sendmsg+0x27b/0x2a0 net/ipv4/af_inet.c:851
sock_sendmsg_nosec net/socket.c:730 [inline]
__sock_sendmsg+0x274/0x3c0 net/socket.c:745
__sys_sendto+0x62c/0x7b0 net/socket.c:2191
__do_sys_sendto net/socket.c:2203 [inline]
__se_sys_sendto net/socket.c:2199 [inline]
__x64_sys_sendto+0x130/0x200 net/socket.c:2199
do_syscall_64+0xd8/0x1f0 arch/x86/entry/common.c:83
entry_SYSCALL_64_after_hwframe+0x6d/0x75
Uninit was created at:
slab_post_alloc_hook mm/slub.c:3804 [inline]
slab_alloc_node mm/slub.c:3845 [inline]
kmem_cache_alloc_node+0x5f6/0xc50 mm/slub.c:3888
kmalloc_reserve+0x13c/0x4a0 net/core/skbuff.c:577
__alloc_skb+0x35a/0x7c0 net/core/skbuff.c:668
alloc_skb include/linux/skbuff.h:1318 [inline]
__ip_append_data+0x49ab/0x68c0 net/ipv4/ip_output.c:1128
ip_append_data+0x1e7/0x260 net/ipv4/ip_output.c:1365
raw_sendmsg+0x22b1/0x2690 net/ipv4/raw.c:648
inet_sendmsg+0x27b/0x2a0 net/ipv4/af_inet.c:851
sock_sendmsg_nosec net/socket.c:730 [inline]
__sock_sendmsg+0x274/0x3c0 net/socket.c:745
__sys_sendto+0x62c/0x7b0 net/socket.c:2191
__do_sys_sendto net/socket.c:2203 [inline]
__se_sys_sendto net/socket.c:2199 [inline]
__x64_sys_sendto+0x130/0x200 net/socket.c:2199
do_syscall_64+0xd8/0x1f0 arch/x86/entry/common.c:83
entry_SYSCALL_64_after_hwframe+0x6d/0x75
CPU: 1 PID: 15709 Comm: syz-executor.7 Not tainted 6.8.0-11567-gb3603fcb79b1 #25
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.3-1.fc39 04/01/2014
Fixes: 99e5acae193e ("ipv4: Fix potential uninit variable access bug in __ip_make_skb()")
Reported-by: syzkaller <syzkaller@googlegroups.com>
Signed-off-by: Shigeru Yoshida <syoshida@redhat.com>
Link: https://lore.kernel.org/r/20240430123945.2057348-1-syoshida@redhat.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
ioctl(SIOCGARP) holds rtnl_lock() to get netdev by __dev_get_by_name()
and copy dev->name safely and calls neigh_lookup() later, which looks
up a neighbour entry under RCU.
Let's replace __dev_get_by_name() with dev_get_by_name_rcu() and strscpy()
with netdev_copy_name() to avoid locking rtnl_lock().
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/20240430015813.71143-8-kuniyu@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
arp_ioctl() holds rtnl_lock() first regardless of cmd (SIOCDARP,
SIOCSARP, and SIOCGARP) to get net_device by __dev_get_by_name()
and copy dev->name safely.
In the SIOCGARP path, arp_req_get() calls neigh_lookup(), which
looks up a neighbour entry under RCU.
We will extend the RCU section not to take rtnl_lock() and instead
use dev_get_by_name_rcu() for SIOCGARP.
As a preparation, let's move __dev_get_by_name() into another
function and call it from arp_req_delete(), arp_req_set(), and
arp_req_get().
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/20240430015813.71143-6-kuniyu@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When ioctl(SIOCDARP/SIOCSARP) is issued for non-proxy entry (no ATF_COM)
without arpreq.arp_dev[] set, arp_req_set() and arp_req_delete() looks up
dev based on IPv4 address by ip_route_output().
Let's factorise the same code as arp_req_dev().
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/20240430015813.71143-4-kuniyu@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When ioctl(SIOCDARP/SIOCSARP) is issued with ATF_PUBL, r.arp_netmask
must be 0.0.0.0 or 255.255.255.255.
Currently, the netmask is validated in arp_req_delete_public() or
arp_req_set_public() under rtnl_lock().
We have ATF_NETMASK test in arp_ioctl() before holding rtnl_lock(),
so let's move the netmask validation there.
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/20240430015813.71143-3-kuniyu@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In arp_req_set(), if ATF_PERM is set in arpreq.arp_flags,
ATF_COM is set automatically.
The flag will be used later for neigh_update() only when
a neighbour entry is found.
Let's set ATF_COM just before calling neigh_update().
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/20240430015813.71143-2-kuniyu@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Move some proto memory definitions out of <net/sock.h>
Very few files need them, and following patch
will include <net/hotdata.h> from <net/proto_memory.h>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240429134025.1233626-5-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tcp_out_of_memory() has a single caller: tcp_check_oom().
Following patch will also make sk_memory_allocated()
not anymore visible from <net/sock.h> and <net/tcp.h>
Add const qualifier to sock argument of tcp_out_of_memory()
and tcp_check_oom().
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240429134025.1233626-4-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
sysctl_max_skb_frags is used in TCP and MPTCP fast paths,
move it to net_hodata for better cache locality.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20240429134025.1233626-2-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
I added dst_rt6_info() in commit
e8dfd42c17fa ("ipv6: introduce dst_rt6_info() helper")
This patch does a similar change for IPv4.
Instead of (struct rtable *)dst casts, we can use :
#define dst_rtable(_ptr) \
container_of_const(_ptr, struct rtable, dst)
Patch is smaller than IPv6 one, because IPv4 has skb_rtable() helper.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://lore.kernel.org/r/20240429133009.1227754-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
-----BEGIN PGP SIGNATURE-----
iHUEABYIAB0WIQTFp0I1jqZrAX+hPRXbK58LschIgwUCZi9+AAAKCRDbK58LschI
g0nEAP487m7L0nLVriC2oIOWsi29tklW3etm6DO7gmGRGIHgrgEAnMyV1xBj3bGj
v6jJwDcybCym1hLx+1x1JCZ4eoAFswE=
=xbna
-----END PGP SIGNATURE-----
Merge tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next
Daniel Borkmann says:
====================
pull-request: bpf-next 2024-04-29
We've added 147 non-merge commits during the last 32 day(s) which contain
a total of 158 files changed, 9400 insertions(+), 2213 deletions(-).
The main changes are:
1) Add an internal-only BPF per-CPU instruction for resolving per-CPU
memory addresses and implement support in x86 BPF JIT. This allows
inlining per-CPU array and hashmap lookups
and the bpf_get_smp_processor_id() helper, from Andrii Nakryiko.
2) Add BPF link support for sk_msg and sk_skb programs, from Yonghong Song.
3) Optimize x86 BPF JIT's emit_mov_imm64, and add support for various
atomics in bpf_arena which can be JITed as a single x86 instruction,
from Alexei Starovoitov.
4) Add support for passing mark with bpf_fib_lookup helper,
from Anton Protopopov.
5) Add a new bpf_wq API for deferring events and refactor sleepable
bpf_timer code to keep common code where possible,
from Benjamin Tissoires.
6) Fix BPF_PROG_TEST_RUN infra with regards to bpf_dummy_struct_ops programs
to check when NULL is passed for non-NULLable parameters,
from Eduard Zingerman.
7) Harden the BPF verifier's and/or/xor value tracking,
from Harishankar Vishwanathan.
8) Introduce crypto kfuncs to make BPF programs able to utilize the kernel
crypto subsystem, from Vadim Fedorenko.
9) Various improvements to the BPF instruction set standardization doc,
from Dave Thaler.
10) Extend libbpf APIs to partially consume items from the BPF ringbuffer,
from Andrea Righi.
11) Bigger batch of BPF selftests refactoring to use common network helpers
and to drop duplicate code, from Geliang Tang.
12) Support bpf_tail_call_static() helper for BPF programs with GCC 13,
from Jose E. Marchesi.
13) Add bpf_preempt_{disable,enable}() kfuncs in order to allow a BPF
program to have code sections where preemption is disabled,
from Kumar Kartikeya Dwivedi.
14) Allow invoking BPF kfuncs from BPF_PROG_TYPE_SYSCALL programs,
from David Vernet.
15) Extend the BPF verifier to allow different input maps for a given
bpf_for_each_map_elem() helper call in a BPF program, from Philo Lu.
16) Add support for PROBE_MEM32 and bpf_addr_space_cast instructions
for riscv64 and arm64 JITs to enable BPF Arena, from Puranjay Mohan.
17) Shut up a false-positive KMSAN splat in interpreter mode by unpoison
the stack memory, from Martin KaFai Lau.
18) Improve xsk selftest coverage with new tests on maximum and minimum
hardware ring size configurations, from Tushar Vyavahare.
19) Various ReST man pages fixes as well as documentation and bash completion
improvements for bpftool, from Rameez Rehman & Quentin Monnet.
20) Fix libbpf with regards to dumping subsequent char arrays,
from Quentin Deslandes.
* tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next: (147 commits)
bpf, docs: Clarify PC use in instruction-set.rst
bpf_helpers.h: Define bpf_tail_call_static when building with GCC
bpf, docs: Add introduction for use in the ISA Internet Draft
selftests/bpf: extend BPF_SOCK_OPS_RTT_CB test for srtt and mrtt_us
bpf: add mrtt and srtt as BPF_SOCK_OPS_RTT_CB args
selftests/bpf: dummy_st_ops should reject 0 for non-nullable params
bpf: check bpf_dummy_struct_ops program params for test runs
selftests/bpf: do not pass NULL for non-nullable params in dummy_st_ops
selftests/bpf: adjust dummy_st_ops_success to detect additional error
bpf: mark bpf_dummy_struct_ops.test_1 parameter as nullable
selftests/bpf: Add ring_buffer__consume_n test.
bpf: Add bpf_guard_preempt() convenience macro
selftests: bpf: crypto: add benchmark for crypto functions
selftests: bpf: crypto skcipher algo selftests
bpf: crypto: add skcipher to bpf crypto
bpf: make common crypto API for TC/XDP programs
bpf: update the comment for BTF_FIELDS_MAX
selftests/bpf: Fix wq test.
selftests/bpf: Use make_sockaddr in test_sock_addr
selftests/bpf: Use connect_to_addr in test_sock_addr
...
====================
Link: https://lore.kernel.org/r/20240429131657.19423-1-daniel@iogearbox.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Instead of (struct rt6_info *)dst casts, we can use :
#define dst_rt6_info(_ptr) \
container_of_const(_ptr, struct rt6_info, dst)
Some places needed missing const qualifiers :
ip6_confirm_neigh(), ipv6_anycast_destination(),
ipv6_unicast_destination(), has_gateway()
v2: added missing parts (David Ahern)
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is a followup of commit c4e86b4363ac ("net: add two more
call_rcu_hurry()")
Our reference to ifa->ifa_dev must be freed ASAP
to release the reference to the netdev the same way.
inet_rcu_free_ifa()
in_dev_put()
-> in_dev_finish_destroy()
-> netdev_put()
This should speedup device/netns dismantles when CONFIG_RCU_LAZY=y
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
I forgot to call tcp_skb_collapse_tstamp() in the
case we consume the second skb in write queue.
Neal suggested to create a common helper used by tcp_mtu_probe()
and tcp_grow_skb().
Fixes: 8ee602c63520 ("tcp: try to send bigger TSO packets")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Link: https://lore.kernel.org/r/20240425193450.411640-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
At last, we should let it work by introducing this reset reason in
trace world.
One of the possible expected outputs is:
... tcp_send_reset: skbaddr=xxx skaddr=xxx src=xxx dest=xxx
state=TCP_ESTABLISHED reason=NOT_SPECIFIED
Signed-off-by: Jason Xing <kernelxing@tencent.com>
Reviewed-by: Steven Rostedt (Google) <rostedt@goodmis.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reuse the dropreason logic to show the exact reason of tcp reset,
so we can finally display the corresponding item in enum sk_reset_reason
instead of reinventing new reset reasons. This patch replaces all
the prior NOT_SPECIFIED reasons.
Signed-off-by: Jason Xing <kernelxing@tencent.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Like what we did to passive reset:
only passing possible reset reason in each active reset path.
No functional changes.
Signed-off-by: Jason Xing <kernelxing@tencent.com>
Acked-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Adjust the parameter and support passing reason of reset which
is for now NOT_SPECIFIED. No functional changes.
Signed-off-by: Jason Xing <kernelxing@tencent.com>
Acked-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The software GRO path for esp transport mode uses skb_mac_header_rebuild
prior to re-injecting the packet via the xfrm_napi_dev. This only
copies skb->mac_len bytes of header which may not be sufficient if the
packet contains 802.1Q tags or other VLAN tags. Worse copying only the
initial header will leave a packet marked as being VLAN tagged but
without the corresponding tag leading to mangling when it is later
untagged.
The VLAN tags are important when receiving the decrypted esp transport
mode packet after GRO processing to ensure it is received on the correct
interface.
Therefore record the full mac header length in xfrm*_transport_input for
later use in corresponding xfrm*_transport_finish to copy the entire mac
header when rebuilding the mac header for GRO. The skb->data pointer is
left pointing skb->mac_header bytes after the start of the mac header as
is expected by the network stack and network and transport header
offsets reset to this location.
Fixes: 7785bba299a8 ("esp: Add a software GRO codepath")
Signed-off-by: Paul Davey <paul.davey@alliedtelesis.co.nz>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Two important arguments in RTT estimation, mrtt and srtt, are passed to
tcp_bpf_rtt(), so that bpf programs get more information about RTT
computation in BPF_SOCK_OPS_RTT_CB.
The difference between bpf_sock_ops->srtt_us and the srtt here is: the
former is an old rtt before update, while srtt passed by tcp_bpf_rtt()
is that after update.
Signed-off-by: Philo Lu <lulie@linux.alibaba.com>
Link: https://lore.kernel.org/r/20240425161724.73707-2-lulie@linux.alibaba.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
While testing TCP performance with latest trees,
I saw suspect SOCKET_BACKLOG drops.
tcp_add_backlog() computes its limit with :
limit = (u32)READ_ONCE(sk->sk_rcvbuf) +
(u32)(READ_ONCE(sk->sk_sndbuf) >> 1);
limit += 64 * 1024;
This does not take into account that sk->sk_backlog.len
is reset only at the very end of __release_sock().
Both sk->sk_backlog.len and sk->sk_rmem_alloc could reach
sk_rcvbuf in normal conditions.
We should double sk->sk_rcvbuf contribution in the formula
to absorb bubbles in the backlog, which happen more often
for very fast flows.
This change maintains decent protection against abuses.
Fixes: c377411f2494 ("net: sk_add_backlog() take rmem_alloc into account")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20240423125620.3309458-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Marking TCP_SKB_CB(skb)->sacked with TCPCB_EVER_RETRANS after the
traceopint (trace_tcp_retransmit_skb), then we can get the
retransmission efficiency by counting skbs w/ and w/o TCPCB_EVER_RETRANS
mark in this tracepoint.
We have discussed to achieve this with BPF_SOCK_OPS in [0], and using
tracepoint is thought to be a better solution.
[0]
https://lore.kernel.org/all/20240417124622.35333-1-lulie@linux.alibaba.com/
Signed-off-by: Philo Lu <lulie@linux.alibaba.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since call_rcu, which is called in the hlist_for_each_entry_rcu traversal
of tcp_ao_connect_init, is not part of the RCU read critical section, it
is possible that the RCU grace period will pass during the traversal and
the key will be free.
To prevent this, it should be changed to hlist_for_each_entry_safe.
Fixes: 7c2ffaf21bd6 ("net/tcp: Calculate TCP-AO traffic keys")
Signed-off-by: Hyunwoo Kim <v4bel@theori.io>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Dmitry Safonov <0x7f454c46@gmail.com>
Link: https://lore.kernel.org/r/ZiYu9NJ/ClR8uSkH@v4bel-B760M-AORUS-ELITE-AX
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
While investigating TCP performance, I found that TCP would
sometimes send big skbs followed by a single MSS skb,
in a 'locked' pattern.
For instance, BIG TCP is enabled, MSS is set to have 4096 bytes
of payload per segment. gso_max_size is set to 181000.
This means that an optimal TCP packet size should contain
44 * 4096 = 180224 bytes of payload,
However, I was seeing packets sizes interleaved in this pattern:
172032, 8192, 172032, 8192, 172032, 8192, <repeat>
tcp_tso_should_defer() heuristic is defeated, because after a split of
a packet in write queue for whatever reason (this might be a too small
CWND or a small enough pacing_rate),
the leftover packet in the queue is smaller than the optimal size.
It is time to try to make 'leftover packets' bigger so that
tcp_tso_should_defer() can give its full potential.
After this patch, we can see the following output:
14:13:34.009273 IP6 sender > receiver: Flags [P.], seq 4048380:4098360, ack 1, win 256, options [nop,nop,TS val 3425678144 ecr 1561784500], length 49980
14:13:34.010272 IP6 sender > receiver: Flags [P.], seq 4098360:4148340, ack 1, win 256, options [nop,nop,TS val 3425678145 ecr 1561784501], length 49980
14:13:34.011271 IP6 sender > receiver: Flags [P.], seq 4148340:4198320, ack 1, win 256, options [nop,nop,TS val 3425678146 ecr 1561784502], length 49980
14:13:34.012271 IP6 sender > receiver: Flags [P.], seq 4198320:4248300, ack 1, win 256, options [nop,nop,TS val 3425678147 ecr 1561784503], length 49980
14:13:34.013272 IP6 sender > receiver: Flags [P.], seq 4248300:4298280, ack 1, win 256, options [nop,nop,TS val 3425678148 ecr 1561784504], length 49980
14:13:34.014271 IP6 sender > receiver: Flags [P.], seq 4298280:4348260, ack 1, win 256, options [nop,nop,TS val 3425678149 ecr 1561784505], length 49980
14:13:34.015272 IP6 sender > receiver: Flags [P.], seq 4348260:4398240, ack 1, win 256, options [nop,nop,TS val 3425678150 ecr 1561784506], length 49980
14:13:34.016270 IP6 sender > receiver: Flags [P.], seq 4398240:4448220, ack 1, win 256, options [nop,nop,TS val 3425678151 ecr 1561784507], length 49980
14:13:34.017269 IP6 sender > receiver: Flags [P.], seq 4448220:4498200, ack 1, win 256, options [nop,nop,TS val 3425678152 ecr 1561784508], length 49980
14:13:34.018276 IP6 sender > receiver: Flags [P.], seq 4498200:4548180, ack 1, win 256, options [nop,nop,TS val 3425678153 ecr 1561784509], length 49980
14:13:34.019259 IP6 sender > receiver: Flags [P.], seq 4548180:4598160, ack 1, win 256, options [nop,nop,TS val 3425678154 ecr 1561784510], length 49980
With 200 concurrent flows on a 100Gbit NIC, we can see a reduction
of TSO packets (and ACK packets) of about 30 %.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20240418214600.1291486-4-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tcp_write_xmit() calls tcp_init_tso_segs()
to set gso_size and gso_segs on the packet.
tcp_init_tso_segs() requires the stack to maintain
an up to date tcp_skb_pcount(), and this makes sense
for packets in rtx queue. Not so much for packets
still in the write queue.
In the following patch, we don't want to deal with
tcp_skb_pcount() when moving payload from 2nd
skb to 1st skb in the write queue.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20240418214600.1291486-3-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tcp_cwnd_test() has a special handing for the last packet in
the write queue if it is smaller than one MSS and has the FIN flag.
This is in violation of TCP RFC, and seems quite dubious.
This packet can be sent only if the current CWND is bigger
than the number of packets in flight.
Making tcp_cwnd_test() result independent of the first skb
in the write queue is needed for the last patch of the series.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20240418214600.1291486-2-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
After commit 1eeb50435739 ("tcp/dccp: do not care about
families in inet_twsk_purge()") tcp_twsk_purge() is
no longer potentially called from a module.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
First problem is a double call to __in_dev_get_rcu(), because
the second one could return NULL.
if (__in_dev_get_rcu(dev) && __in_dev_get_rcu(dev)->ifa_list)
Second problem is a read from dev->ip6_ptr with no NULL check:
if (!list_empty(&rcu_dereference(dev->ip6_ptr)->addr_list))
Use the correct RCU API to fix these.
v2: add missing include <net/addrconf.h>
Fixes: d329ea5bd884 ("icmp: add response to RFC 8335 PROBE messages")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Andreas Roeseler <andreas.a.roeseler@gmail.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
To be able to constify instances of struct ctl_tables it is necessary to
remove ways through which non-const versions are exposed from the
sysctl core.
One of these is the ctl_table_arg member of struct ctl_table_header.
Constify this reference as a prerequisite for the full constification of
struct ctl_table instances.
No functional change.
Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Reviewed-by: Kees Cook <keescook@chromium.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
If "udp_cmsg_send()" returned 0 (i.e. only UDP cmsg),
"connected" should not be set to 0. Otherwise it stops
the connected socket from using the cached route.
Fixes: 2e8de8576343 ("udp: add gso segment cmsg")
Signed-off-by: Yick Xie <yick.xie@gmail.com>
Cc: stable@vger.kernel.org
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://lore.kernel.org/r/20240418170610.867084-1-yick.xie@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The UDP_ENCAP_ESPINUDP_NON_IKE mode, introduced into the Linux kernel
in 2004 [2], has remained inactive and obsolete for an extended period.
This mode was originally defined in an early version of an IETF draft
[1] from 2001. By the time it was integrated into the kernel in 2004 [2],
it had already been replaced by UDP_ENCAP_ESPINUDP [3] in later
versions of draft-ietf-ipsec-udp-encaps, particularly in version 06.
Over time, UDP_ENCAP_ESPINUDP_NON_IKE has lost its relevance, with no
known use cases.
With this commit, we remove support for UDP_ENCAP_ESPINUDP_NON_IKE,
simplifying the codebase and eliminating unnecessary complexity.
Kernel will return an error -ENOPROTOOPT if the userspace tries to set
this option.
References:
[1] https://datatracker.ietf.org/doc/html/draft-ietf-ipsec-udp-encaps-00.txt
[2] Commit that added UDP_ENCAP_ESPINUDP_NON_IKE to the Linux historic
repository.
Author: Andreas Gruenbacher <agruen@suse.de>
Date: Fri Apr 9 01:47:47 2004 -0700
[IPSEC]: Support draft-ietf-ipsec-udp-encaps-00/01, some ipec impls need it.
[3] Commit that added UDP_ENCAP_ESPINUDP to the Linux historic
repository.
Author: Derek Atkins <derek@ihtfp.com>
Date: Wed Apr 2 13:21:02 2003 -0800
[IPSEC]: Implement UDP Encapsulation framework.
Signed-off-by: Antony Antony <antony.antony@secunet.com>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
TCP_METRICS_CMD_GET and TCP_METRICS_CMD_DEL use their
own locking (tcp_metrics_lock and RCU),
they do not need genl_mutex protection.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20240416162025.1251547-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Change tcp_metrics_nl_dump() to return 0 at the end
of a dump so that NLMSG_DONE can be appended
to the current skb, saving one recvmsg() system call.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20240416161112.1199265-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Andrew Oates reported that some macOS hosts could repeatedly
send FIN packets even if the remote peer drops them and
send back DUP ACK RWIN 0 packets.
<quoting Andrew>
20:27:16.968254 gif0 In IP macos > victim: Flags [SEW], seq 1950399762, win 65535, options [mss 1460,nop,wscale 6,nop,nop,TS val 501897188 ecr 0,sackOK,eol], length 0
20:27:16.968339 gif0 Out IP victim > macos: Flags [S.E], seq 2995489058, ack 1950399763, win 1448, options [mss 1460,sackOK,TS val 3829877593 ecr 501897188,nop,wscale 0], length 0
20:27:16.968833 gif0 In IP macos > victim: Flags [.], ack 1, win 2058, options [nop,nop,TS val 501897188 ecr 3829877593], length 0
20:27:16.968885 gif0 In IP macos > victim: Flags [P.], seq 1:1449, ack 1, win 2058, options [nop,nop,TS val 501897188 ecr 3829877593], length 1448
20:27:16.968896 gif0 Out IP victim > macos: Flags [.], ack 1449, win 0, options [nop,nop,TS val 3829877593 ecr 501897188], length 0
20:27:19.454593 gif0 In IP macos > victim: Flags [F.], seq 1449, ack 1, win 2058, options [nop,nop,TS val 501899674 ecr 3829877593], length 0
20:27:19.454675 gif0 Out IP victim > macos: Flags [.], ack 1449, win 0, options [nop,nop,TS val 3829880079 ecr 501899674], length 0
20:27:19.455116 gif0 In IP macos > victim: Flags [F.], seq 1449, ack 1, win 2058, options [nop,nop,TS val 501899674 ecr 3829880079], length 0
The retransmits/dup-ACKs then repeat in a tight loop.
</quoting Andrew>
RFC 9293 3.4. Sequence Numbers states :
Note that when the receive window is zero no segments should be
acceptable except ACK segments. Thus, it is be possible for a TCP to
maintain a zero receive window while transmitting data and receiving
ACKs. However, even when the receive window is zero, a TCP must
process the RST and URG fields of all incoming segments.
Even if we could consider a bare FIN.ACK packet to be an ACK in RFC terms,
the retransmits should use exponential backoff.
Accepting the FIN in linux does not add extra memory costs,
because the FIN flag will simply be merged to the tail skb in
the receive queue, and incoming packet is freed.
Reported-by: Andrew Oates <aoates@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Christoph Paasch <cpaasch@apple.com>
Cc: Vidhi Goel <vidhi_goel@apple.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We've observed a 7-12% performance regression in iperf3 UDP ipv4 and
ipv6 tests with multiple sockets on Zen3 cpus, which we traced back to
commit f0ea27e7bfe1 ("udp: re-score reuseport groups when connected
sockets are present"). The failing tests were those that would spawn
UDP sockets per-cpu on systems that have a high number of cpus.
Unsurprisingly, it is not caused by the extra re-scoring of the reused
socket, but due to the compiler no longer inlining compute_score, once
it has the extra call site in udp4_lib_lookup2. This is augmented by
the "Safe RET" mitigation for SRSO, needed in our Zen3 cpus.
We could just explicitly inline it, but compute_score() is quite a large
function, around 300b. Inlining in two sites would almost double
udp4_lib_lookup2, which is a silly thing to do just to workaround a
mitigation. Instead, this patch shuffles the code a bit to avoid the
multiple calls to compute_score. Since it is a static function used in
one spot, the compiler can safely fold it in, as it did before, without
increasing the text size.
With this patch applied I ran my original iperf3 testcases. The failing
cases all looked like this (ipv4):
iperf3 -c 127.0.0.1 --udp -4 -f K -b $R -l 8920 -t 30 -i 5 -P 64 -O 2
where $R is either 1G/10G/0 (max, unlimited). I ran 3 times each.
baseline is v6.9-rc3. harmean == harmonic mean; CV == coefficient of
variation.
ipv4:
1G 10G MAX
HARMEAN (CV) HARMEAN (CV) HARMEAN (CV)
baseline 1743852.66(0.0208) 1725933.02(0.0167) 1705203.78(0.0386)
patched 1968727.61(0.0035) 1962283.22(0.0195) 1923853.50(0.0256)
ipv6:
1G 10G MAX
HARMEAN (CV) HARMEAN (CV) HARMEAN (CV)
baseline 1729020.03(0.0028) 1691704.49(0.0243) 1692251.34(0.0083)
patched 1900422.19(0.0067) 1900968.01(0.0067) 1568532.72(0.1519)
This restores the performance we had before the change above with this
benchmark. We obviously don't expect any real impact when mitigations
are disabled, but just to be sure it also doesn't regresses:
mitigations=off ipv4:
1G 10G MAX
HARMEAN (CV) HARMEAN (CV) HARMEAN (CV)
baseline 3230279.97(0.0066) 3229320.91(0.0060) 2605693.19(0.0697)
patched 3242802.36(0.0073) 3239310.71(0.0035) 2502427.19(0.0882)
Cc: Lorenz Bauer <lmb@isovalent.com>
Fixes: f0ea27e7bfe1 ("udp: re-score reuseport groups when connected sockets are present")
Signed-off-by: Gabriel Krisman Bertazi <krisman@suse.de>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
RTO_ONLINK was a flag used in ->flowi4_tos that allowed to alter the
scope of an IPv4 route lookup. Setting this flag was equivalent to
specifying RT_SCOPE_LINK in ->flowi4_scope.
With commit ec20b2830093 ("ipv4: Set scope explicitly in
ip_route_output()."), the last users of RTO_ONLINK have been removed.
Therefore, we can now drop the code that checked this bit and stop
modifying ->flowi4_scope in ip_route_output_key_hash().
Signed-off-by: Guillaume Nault <gnault@redhat.com>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/57de760565cab55df7b129f523530ac6475865b2.1712754146.git.gnault@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When reading received messages from a socket with MSG_PEEK, we may want
to read the contents with an offset, like we can do with pread/preadv()
when reading files. Currently, it is not possible to do that.
In this commit, we add support for the SO_PEEK_OFF socket option for TCP,
in a similar way it is done for Unix Domain sockets.
In the iperf3 log examples shown below, we can observe a throughput
improvement of 15-20 % in the direction host->namespace when using the
protocol splicer 'pasta' (https://passt.top).
This is a consistent result.
pasta(1) and passt(1) implement user-mode networking for network
namespaces (containers) and virtual machines by means of a translation
layer between Layer-2 network interface and native Layer-4 sockets
(TCP, UDP, ICMP/ICMPv6 echo).
Received, pending TCP data to the container/guest is kept in kernel
buffers until acknowledged, so the tool routinely needs to fetch new
data from socket, skipping data that was already sent.
At the moment this is implemented using a dummy buffer passed to
recvmsg(). With this change, we don't need a dummy buffer and the
related buffer copy (copy_to_user()) anymore.
passt and pasta are supported in KubeVirt and libvirt/qemu.
jmaloy@freyr:~/passt$ perf record -g ./pasta --config-net -f
SO_PEEK_OFF not supported by kernel.
jmaloy@freyr:~/passt# iperf3 -s
-----------------------------------------------------------
Server listening on 5201 (test #1)
-----------------------------------------------------------
Accepted connection from 192.168.122.1, port 44822
[ 5] local 192.168.122.180 port 5201 connected to 192.168.122.1 port 44832
[ ID] Interval Transfer Bitrate
[ 5] 0.00-1.00 sec 1.02 GBytes 8.78 Gbits/sec
[ 5] 1.00-2.00 sec 1.06 GBytes 9.08 Gbits/sec
[ 5] 2.00-3.00 sec 1.07 GBytes 9.15 Gbits/sec
[ 5] 3.00-4.00 sec 1.10 GBytes 9.46 Gbits/sec
[ 5] 4.00-5.00 sec 1.03 GBytes 8.85 Gbits/sec
[ 5] 5.00-6.00 sec 1.10 GBytes 9.44 Gbits/sec
[ 5] 6.00-7.00 sec 1.11 GBytes 9.56 Gbits/sec
[ 5] 7.00-8.00 sec 1.07 GBytes 9.20 Gbits/sec
[ 5] 8.00-9.00 sec 667 MBytes 5.59 Gbits/sec
[ 5] 9.00-10.00 sec 1.03 GBytes 8.83 Gbits/sec
[ 5] 10.00-10.04 sec 30.1 MBytes 6.36 Gbits/sec
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bitrate
[ 5] 0.00-10.04 sec 10.3 GBytes 8.78 Gbits/sec receiver
-----------------------------------------------------------
Server listening on 5201 (test #2)
-----------------------------------------------------------
^Ciperf3: interrupt - the server has terminated
jmaloy@freyr:~/passt#
logout
[ perf record: Woken up 23 times to write data ]
[ perf record: Captured and wrote 5.696 MB perf.data (35580 samples) ]
jmaloy@freyr:~/passt$
jmaloy@freyr:~/passt$ perf record -g ./pasta --config-net -f
SO_PEEK_OFF supported by kernel.
jmaloy@freyr:~/passt# iperf3 -s
-----------------------------------------------------------
Server listening on 5201 (test #1)
-----------------------------------------------------------
Accepted connection from 192.168.122.1, port 52084
[ 5] local 192.168.122.180 port 5201 connected to 192.168.122.1 port 52098
[ ID] Interval Transfer Bitrate
[ 5] 0.00-1.00 sec 1.32 GBytes 11.3 Gbits/sec
[ 5] 1.00-2.00 sec 1.19 GBytes 10.2 Gbits/sec
[ 5] 2.00-3.00 sec 1.26 GBytes 10.8 Gbits/sec
[ 5] 3.00-4.00 sec 1.36 GBytes 11.7 Gbits/sec
[ 5] 4.00-5.00 sec 1.33 GBytes 11.4 Gbits/sec
[ 5] 5.00-6.00 sec 1.21 GBytes 10.4 Gbits/sec
[ 5] 6.00-7.00 sec 1.31 GBytes 11.2 Gbits/sec
[ 5] 7.00-8.00 sec 1.25 GBytes 10.7 Gbits/sec
[ 5] 8.00-9.00 sec 1.33 GBytes 11.5 Gbits/sec
[ 5] 9.00-10.00 sec 1.24 GBytes 10.7 Gbits/sec
[ 5] 10.00-10.04 sec 56.0 MBytes 12.1 Gbits/sec
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bitrate
[ 5] 0.00-10.04 sec 12.9 GBytes 11.0 Gbits/sec receiver
-----------------------------------------------------------
Server listening on 5201 (test #2)
-----------------------------------------------------------
^Ciperf3: interrupt - the server has terminated
logout
[ perf record: Woken up 20 times to write data ]
[ perf record: Captured and wrote 5.040 MB perf.data (33411 samples) ]
jmaloy@freyr:~/passt$
The perf record confirms this result. Below, we can observe that the
CPU spends significantly less time in the function ____sys_recvmsg()
when we have offset support.
Without offset support:
----------------------
jmaloy@freyr:~/passt$ perf report -q --symbol-filter=do_syscall_64 \
-p ____sys_recvmsg -x --stdio -i perf.data | head -1
46.32% 0.00% passt.avx2 [kernel.vmlinux] [k] do_syscall_64 ____sys_recvmsg
With offset support:
----------------------
jmaloy@freyr:~/passt$ perf report -q --symbol-filter=do_syscall_64 \
-p ____sys_recvmsg -x --stdio -i perf.data | head -1
28.12% 0.00% passt.avx2 [kernel.vmlinux] [k] do_syscall_64 ____sys_recvmsg
Suggested-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Jon Maloy <jmaloy@redhat.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20240409152805.913891-1-jmaloy@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add a new header, linux/skbuff_ref.h, which contains all the skb_*_ref()
helpers. Many of the consumers of skbuff.h do not actually use any of
the skb ref helpers, and we can speed up compilation a bit by minimizing
this header file.
Additionally in the later patch in the series we add page_pool support
to skb_frag_ref(), which requires some page_pool dependencies. We can
now add these dependencies to skbuff_ref.h instead of a very ubiquitous
skbuff.h
Signed-off-by: Mina Almasry <almasrymina@google.com>
Link: https://lore.kernel.org/r/20240410190505.1225848-2-almasrymina@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In my recent commit, I missed that do_replace() handlers
use copy_from_sockptr() (which I fixed), followed
by unsafe copy_from_sockptr_offset() calls.
In all functions, we can perform the @optlen validation
before even calling xt_alloc_table_info() with the following
check:
if ((u64)optlen < (u64)tmp.size + sizeof(tmp))
return -EINVAL;
Fixes: 0c83842df40f ("netfilter: validate user input for expected length")
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Pablo Neira Ayuso <pablo@netfilter.org>
Link: https://lore.kernel.org/r/20240409120741.3538135-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
I forgot 32bit arches might have 64bit alignment for u64
fields.
tcp_sock_write_txrx group does not contain pointers,
but two u64 fields. It is possible that on 32bit kernel,
a 32bit hole is before tp->tcp_clock_cache.
I will try to remember a group can be bigger on 32bit
kernels in the future.
With help from Vladimir Oltean.
Fixes: d2c3a7eb1afa ("tcp: more struct tcp_sock adjustments")
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202404082207.HCEdQhUO-lkp@intel.com/
Signed-off-by: Eric Dumazet <edumazet@google.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://lore.kernel.org/r/20240409140914.4105429-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The implementations of these 2 functions are almost identical. Remove
the implementation of napi_frag_unref, and make it a call into
skb_page_unref so we don't duplicate the implementation.
Signed-off-by: Mina Almasry <almasrymina@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20240408153000.2152844-2-almasrymina@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
