net.ipv4.tcp_comp_sack_slack_ns current default value is too high.
When a flow has many drops (1 % or more), and small RTT, adding 100 usec
before sending SACK stalls the sender relying on getting SACK
fast enough to keep the pipe busy.
Decrease the default to 10 usec.
This is orthogonal to Congestion Control heuristics to determine
if drops are caused by congestion or not.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Link: https://patch.msgid.link/20251114135141.3810964-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Cross-merge networking fixes after downstream PR (net-6.18-rc6).
No conflicts, adjacent changes in:
drivers/net/phy/micrel.c
96a9178a29 ("net: phy: micrel: lan8814 fix reset of the QSGMII interface")
61b7ade9ba ("net: phy: micrel: Add support for non PTP SKUs for lan8814")
and a trivial one in tools/testing/selftests/drivers/net/Makefile.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The sit driver's packet transmission path calls: sit_tunnel_xmit() ->
update_or_create_fnhe(), which lead to fnhe_remove_oldest() being called
to delete entries exceeding FNHE_RECLAIM_DEPTH+random.
The race window is between fnhe_remove_oldest() selecting fnheX for
deletion and the subsequent kfree_rcu(). During this time, the
concurrent path's __mkroute_output() -> find_exception() can fetch the
soon-to-be-deleted fnheX, and rt_bind_exception() then binds it with a
new dst using a dst_hold(). When the original fnheX is freed via RCU,
the dst reference remains permanently leaked.
CPU 0 CPU 1
__mkroute_output()
find_exception() [fnheX]
update_or_create_fnhe()
fnhe_remove_oldest() [fnheX]
rt_bind_exception() [bind dst]
RCU callback [fnheX freed, dst leak]
This issue manifests as a device reference count leak and a warning in
dmesg when unregistering the net device:
unregister_netdevice: waiting for sitX to become free. Usage count = N
Ido Schimmel provided the simple test validation method [1].
The fix clears 'oldest->fnhe_daddr' before calling fnhe_flush_routes().
Since rt_bind_exception() checks this field, setting it to zero prevents
the stale fnhe from being reused and bound to a new dst just before it
is freed.
[1]
ip netns add ns1
ip -n ns1 link set dev lo up
ip -n ns1 address add 192.0.2.1/32 dev lo
ip -n ns1 link add name dummy1 up type dummy
ip -n ns1 route add 192.0.2.2/32 dev dummy1
ip -n ns1 link add name gretap1 up arp off type gretap \
local 192.0.2.1 remote 192.0.2.2
ip -n ns1 route add 198.51.0.0/16 dev gretap1
taskset -c 0 ip netns exec ns1 mausezahn gretap1 \
-A 198.51.100.1 -B 198.51.0.0/16 -t udp -p 1000 -c 0 -q &
taskset -c 2 ip netns exec ns1 mausezahn gretap1 \
-A 198.51.100.1 -B 198.51.0.0/16 -t udp -p 1000 -c 0 -q &
sleep 10
ip netns pids ns1 | xargs kill
ip netns del ns1
Cc: stable@vger.kernel.org
Fixes: 67d6d681e1 ("ipv4: make exception cache less predictible")
Signed-off-by: Chuang Wang <nashuiliang@gmail.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20251111064328.24440-1-nashuiliang@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The introduction of IPPROTO_SMC enables eBPF programs to determine
whether to use SMC based on the context of socket creation, such as
network namespaces, PID and comm name, etc.
As a subsequent enhancement, to introduce a new generic hook that
allows decisions on whether to use SMC or not at runtime, including
but not limited to local/remote IP address or ports.
User can write their own implememtion via bpf_struct_ops now to choose
whether to use SMC or not before TCP 3rd handshake to be comleted.
Signed-off-by: D. Wythe <alibuda@linux.alibaba.com>
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Reviewed-by: Dust Li <dust.li@linux.alibaba.com>
Link: https://patch.msgid.link/20251107035632.115950-3-alibuda@linux.alibaba.com
TCP SACK compression has been added in 2018 in commit
5d9f4262b7 ("tcp: add SACK compression").
It is working great for WAN flows (with large RTT).
Wifi in particular gets a significant boost _when_ ACK are suppressed.
Add a new sysctl so that we can tune the very conservative 5 % value
that has been used so far in this formula, so that small RTT flows
can benefit from this feature.
delay = min ( 5 % of RTT, 1 ms)
This patch adds new tcp_comp_sack_rtt_percent sysctl
to ease experiments and tuning.
Given that we cap the delay to 1ms (tcp_comp_sack_delay_ns sysctl),
set the default value to 33 %.
Quoting Neal Cardwell ( https://lore.kernel.org/netdev/CADVnQymZ1tFnEA1Q=vtECs0=Db7zHQ8=+WCQtnhHFVbEOzjVnQ@mail.gmail.com/ )
The rationale for 33% is basically to try to facilitate pipelining,
where there are always at least 3 ACKs and 3 GSO/TSO skbs per SRTT, so
that the path can maintain a budget for 3 full-sized GSO/TSO skbs "in
flight" at all times:
+ 1 skb in the qdisc waiting to be sent by the NIC next
+ 1 skb being sent by the NIC (being serialized by the NIC out onto the wire)
+ 1 skb being received and aggregated by the receiver machine's
aggregation mechanism (some combination of LRO, GRO, and sack
compression)
Note that this is basically the same magic number (3) and the same
rationales as:
(a) tcp_tso_should_defer() ensuring that we defer sending data for no
longer than cwnd/tcp_tso_win_divisor (where tcp_tso_win_divisor = 3),
and
(b) bbr_quantization_budget() ensuring that cwnd is at least 3 GSO/TSO
skbs to maintain pipelining and full throughput at low RTTs
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Link: https://patch.msgid.link/20251106115236.3450026-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit 5903123f66 ("tcp: Use BPF timeout setting for SYN ACK
RTO") introduced req->timeout and initialised it in 3 places:
1. reqsk_alloc() sets 0
2. inet_reqsk_alloc() sets TCP_TIMEOUT_INIT
3. tcp_conn_request() sets tcp_timeout_init()
1. has been always redundant as 2. overwrites it immediately.
2. was necessary for TFO SYN+ACK but is no longer needed after
commit 8ea731d4c2 ("tcp: Make SYN ACK RTO tunable by BPF
programs with TFO").
3. was moved to reqsk_queue_hash_req() in the previous patch.
Now, we always initialise req->timeout just before scheduling
the SYN+ACK timer:
* For non-TFO SYN+ACK : reqsk_queue_hash_req()
* For TFO SYN+ACK : tcp_fastopen_create_child()
Let's remove the redundant initialisation of req->timeout in
reqsk_alloc() and inet_reqsk_alloc().
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20251106003357.273403-4-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Revert struct sockaddr from flexible array to fixed 14-byte "sa_data",
to solve over 36,000 -Wflex-array-member-not-at-end warnings, since
struct sockaddr is embedded within many network structs.
With socket/proto sockaddr-based internal APIs switched to use struct
sockaddr_unsized, there should be no more uses of struct sockaddr that
depend on reading beyond the end of struct sockaddr::sa_data that might
trigger bounds checking.
Comparing an x86_64 "allyesconfig" vmlinux build before and after this
patch showed no new "ud1" instructions from CONFIG_UBSAN_BOUNDS nor any
new "field-spanning" memcpy CONFIG_FORTIFY_SOURCE instrumentations.
Cc: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: Kees Cook <kees@kernel.org>
Link: https://patch.msgid.link/20251104002617.2752303-8-kees@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Convert struct proto pre_connect(), connect(), bind(), and bind_add()
callback function prototypes from struct sockaddr to struct sockaddr_unsized.
This does not change per-implementation use of sockaddr for passing around
an arbitrarily sized sockaddr struct. Those will be addressed in future
patches.
Additionally removes the no longer referenced struct sockaddr from
include/net/inet_common.h.
No binary changes expected.
Signed-off-by: Kees Cook <kees@kernel.org>
Link: https://patch.msgid.link/20251104002617.2752303-5-kees@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Update all struct proto_ops connect() callback function prototypes from
"struct sockaddr *" to "struct sockaddr_unsized *" to avoid lying to the
compiler about object sizes. Calls into struct proto handlers gain casts
that will be removed in the struct proto conversion patch.
No binary changes expected.
Signed-off-by: Kees Cook <kees@kernel.org>
Link: https://patch.msgid.link/20251104002617.2752303-3-kees@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Update all struct proto_ops bind() callback function prototypes from
"struct sockaddr *" to "struct sockaddr_unsized *" to avoid lying to the
compiler about object sizes. Calls into struct proto handlers gain casts
that will be removed in the struct proto conversion patch.
No binary changes expected.
Signed-off-by: Kees Cook <kees@kernel.org>
Link: https://patch.msgid.link/20251104002617.2752303-2-kees@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add the ability to append the incoming IP interface information to
ICMPv4 error messages in accordance with RFC 5837 and RFC 4884. This is
required for more meaningful traceroute results in unnumbered networks.
The feature is disabled by default and controlled via a new sysctl
("net.ipv4.icmp_errors_extension_mask") which accepts a bitmask of ICMP
extensions to append to ICMP error messages. Currently, only a single
value is supported, but the interface and the implementation should be
able to support more extensions, if needed.
Clone the skb and copy the relevant data portions before modifying the
skb as the caller of __icmp_send() still owns the skb after the function
returns. This should be fine since by default ICMP error messages are
rate limited to 1000 per second and no more than 1 per second per
specific host.
Trim or pad the packet to 128 bytes before appending the ICMP extension
structure in order to be compatible with legacy applications that assume
that the ICMP extension structure always starts at this offset (the
minimum length specified by RFC 4884).
Reviewed-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20251027082232.232571-2-idosch@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Function udp_v4_early_demux() was already declared in 'include/net/udp.h',
no need to keep the extern in 'ip_input.c', which may produce the
following checkpatch warning:
WARNING: externs should be avoided in .c files
#45: FILE: net/ipv4/ip_input.c:322:
+enum skb_drop_reason udp_v4_early_demux(struct sk_buff *skb);
Replace it by including 'net/udp.h'. Do the same for tcp_v4_early_demux().
Signed-off-by: Wang Liang <wangliang74@huawei.com>
Link: https://patch.msgid.link/20251025092637.1020960-1-wangliang74@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Fix kernel-doc warnings in tcp_lp.c by adding missing parameter
descriptions for tcp_lp_cong_avoid() and tcp_lp_pkts_acked() when
building with W=1.
Also replace invalid URLs in the file header comment with the currently
valid links to the TCP-LP paper and implementation page.
No functional changes.
Signed-off-by: Rakuram Eswaran <rakuram.e96@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20251025-net_ipv4_tcp_lp_c-v1-1-058cc221499e@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Each time some payload is consumed by user space (recvmsg() and friends),
TCP calls tcp_rcv_space_adjust() to run DRS algorithm to check
if an increase of sk->sk_rcvbuf is needed.
This function is based on time sampling, and currently calls
tcp_mstamp_refresh(tp), which is a wrapper around ktime_get_ns().
ktime_get_ns() has a high cost on some platforms.
100+ cycles for rdtscp on AMD EPYC Turin for instance.
We do not have to refresh tp->tcp_mpstamp, using the last cached value
is enough. We only need to refresh it from __tcp_cleanup_rbuf()
if an ACK must be sent (this is a rare event).
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20251024120707.3516550-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
sctp_do_peeloff() calls sock_create() to allocate and initialise
struct sock, inet_sock, and sctp_sock, but later sctp_copy_sock()
and sctp_sock_migrate() overwrite most fields.
What sctp_do_peeloff() does is more like accept().
Let's use sock_create_lite() and sctp_clone_sock().
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Acked-by: Xin Long <lucien.xin@gmail.com>
Link: https://patch.msgid.link/20251023231751.4168390-8-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
sctp_accept() calls sctp_v[46]_create_accept_sk() to allocate a new
socket and calls sctp_sock_migrate() to copy fields from the parent
socket to the new socket.
sctp_v4_create_accept_sk() allocates sk by sk_alloc(), initialises
it by sock_init_data(), and copy a bunch of fields from the parent
socekt by sctp_copy_sock().
sctp_sock_migrate() calls sctp_copy_descendant() to copy most fields
in sctp_sock from the parent socket by memcpy().
These can be simply replaced by sk_clone().
Let's consolidate sctp_v[46]_create_accept_sk() to sctp_clone_sock()
with sk_clone().
We will reuse sctp_clone_sock() for sctp_do_peeloff() and then remove
sctp_copy_descendant().
Note that sock_reset_flag(newsk, SOCK_ZAPPED) is not copied to
sctp_clone_sock() as sctp does not use SOCK_ZAPPED at all.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Acked-by: Xin Long <lucien.xin@gmail.com>
Link: https://patch.msgid.link/20251023231751.4168390-6-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The 'if (!key && hash_location)' check in tcp_inbound_md5_hash() implies
that hash_location might be null. However, later code in the function
dereferences hash_location anyway, without checking for null first.
Fortunately, there is no real bug, since tcp_inbound_md5_hash() is
called only with non-null values of hash_location.
Therefore, remove the unnecessary and misleading null check of
hash_location. This silences a Smatch static checker warning
(https://lore.kernel.org/netdev/aPi4b6aWBbBR52P1@stanley.mountain/)
Also fix the related comment at the beginning of the function.
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Dmitry Safonov <0x7f454c46@gmail.com>
Link: https://patch.msgid.link/20251022221209.19716-1-ebiggers@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Martin KaFai Lau says:
====================
pull-request: bpf-next 2025-10-16
We've added 6 non-merge commits during the last 1 day(s) which contain
a total of 18 files changed, 577 insertions(+), 38 deletions(-).
The main changes are:
1) Bypass the global per-protocol memory accounting either by setting
a netns sysctl or using bpf_setsockopt in a bpf program,
from Kuniyuki Iwashima.
* tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next:
selftests/bpf: Add test for sk->sk_bypass_prot_mem.
bpf: Introduce SK_BPF_BYPASS_PROT_MEM.
bpf: Support bpf_setsockopt() for BPF_CGROUP_INET_SOCK_CREATE.
net: Introduce net.core.bypass_prot_mem sysctl.
net: Allow opt-out from global protocol memory accounting.
tcp: Save lock_sock() for memcg in inet_csk_accept().
====================
Link: https://patch.msgid.link/20251016204539.773707-1-martin.lau@linux.dev
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Make tcp-md5 use the MD5 library API (added in 6.18) instead of the
crypto_ahash API. This is much simpler and also more efficient:
- The library API just operates on struct md5_ctx. Just allocate this
struct on the stack instead of using a pool of pre-allocated
crypto_ahash and ahash_request objects.
- The library API accepts standard pointers and doesn't require
scatterlists. So, for hashing the headers just use an on-stack buffer
instead of a pool of pre-allocated kmalloc'ed scratch buffers.
- The library API never fails. Therefore, checking for MD5 hashing
errors is no longer necessary. Update tcp_v4_md5_hash_skb(),
tcp_v6_md5_hash_skb(), tcp_v4_md5_hash_hdr(), tcp_v6_md5_hash_hdr(),
tcp_md5_hash_key(), tcp_sock_af_ops::calc_md5_hash, and
tcp_request_sock_ops::calc_md5_hash to return void instead of int.
- The library API provides direct access to the MD5 code, eliminating
unnecessary overhead such as indirect function calls and scatterlist
management. Microbenchmarks of tcp_v4_md5_hash_skb() on x86_64 show a
speedup from 7518 to 7041 cycles (6% fewer) with skb->len == 1440, or
from 1020 to 678 cycles (33% fewer) with skb->len == 140.
Since tcp_sigpool_hash_skb_data() can no longer be used, add a function
tcp_md5_hash_skb_data() which is specialized to MD5. Of course, to the
extent that this duplicates any code, it's well worth it.
To preserve the existing behavior of TCP-MD5 support being disabled when
the kernel is booted with "fips=1", make tcp_md5_do_add() check
fips_enabled itself. Previously it relied on the error from
crypto_alloc_ahash("md5") being bubbled up. I don't know for sure that
this is actually needed, but this preserves the existing behavior.
Tested with bidirectional TCP-MD5, both IPv4 and IPv6, between a kernel
that includes this commit and a kernel that doesn't include this commit.
(Side note: please don't use TCP-MD5! It's cryptographically weak. But
as long as Linux supports it, it might as well be implemented properly.)
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Link: https://patch.msgid.link/20251014215836.115616-1-ebiggers@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Since ehash lookups are lockless, if another CPU is converting sk to tw
concurrently, fetching the newly inserted tw with tw->tw_refcnt == 0 cause
lookup failure.
The call trace map is drawn as follows:
CPU 0 CPU 1
----- -----
inet_twsk_hashdance_schedule()
spin_lock()
inet_twsk_add_node_rcu(tw, ...)
__inet_lookup_established()
(find tw, failure due to tw_refcnt = 0)
__sk_nulls_del_node_init_rcu(sk)
refcount_set(&tw->tw_refcnt, 3)
spin_unlock()
By replacing sk with tw atomically via hlist_nulls_replace_init_rcu() after
setting tw_refcnt, we ensure that tw is either fully initialized or not
visible to other CPUs, eliminating the race.
It's worth noting that we held lock_sock() before the replacement, so
there's no need to check if sk is hashed. Thanks to Kuniyuki Iwashima!
Fixes: 3ab5aee7fe ("net: Convert TCP & DCCP hash tables to use RCU / hlist_nulls")
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Jiayuan Chen <jiayuan.chen@linux.dev>
Signed-off-by: Xuanqiang Luo <luoxuanqiang@kylinos.cn>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20251015020236.431822-4-xuanqiang.luo@linux.dev
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Since ehash lookups are lockless, if one CPU performs a lookup while
another concurrently deletes and inserts (removing reqsk and inserting sk),
the lookup may fail to find the socket, an RST may be sent.
The call trace map is drawn as follows:
CPU 0 CPU 1
----- -----
inet_ehash_insert()
spin_lock()
sk_nulls_del_node_init_rcu(osk)
__inet_lookup_established()
(lookup failed)
__sk_nulls_add_node_rcu(sk, list)
spin_unlock()
As both deletion and insertion operate on the same ehash chain, this patch
introduces a new sk_nulls_replace_node_init_rcu() helper functions to
implement atomic replacement.
Fixes: 5e0724d027 ("tcp/dccp: fix hashdance race for passive sessions")
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Jiayuan Chen <jiayuan.chen@linux.dev>
Signed-off-by: Xuanqiang Luo <luoxuanqiang@kylinos.cn>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20251015020236.431822-3-xuanqiang.luo@linux.dev
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Some protocols (e.g., TCP, UDP) implement memory accounting for socket
buffers and charge memory to per-protocol global counters pointed to by
sk->sk_proto->memory_allocated.
Sometimes, system processes do not want that limitation. For a similar
purpose, there is SO_RESERVE_MEM for sockets under memcg.
Also, by opting out of the per-protocol accounting, sockets under memcg
can avoid paying costs for two orthogonal memory accounting mechanisms.
A microbenchmark result is in the subsequent bpf patch.
Let's allow opt-out from the per-protocol memory accounting if
sk->sk_bypass_prot_mem is true.
sk->sk_bypass_prot_mem and sk->sk_prot are placed in the same cache
line, and sk_has_account() always fetches sk->sk_prot before accessing
sk->sk_bypass_prot_mem, so there is no extra cache miss for this patch.
The following patches will set sk->sk_bypass_prot_mem to true, and
then, the per-protocol memory accounting will be skipped.
Note that this does NOT disable memcg, but rather the per-protocol one.
Another option not to use the hole in struct sock_common is create
sk_prot variants like tcp_prot_bypass, but this would complicate
SOCKMAP logic, tcp_bpf_prots etc.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Reviewed-by: Shakeel Butt <shakeel.butt@linux.dev>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Roman Gushchin <roman.gushchin@linux.dev>
Link: https://patch.msgid.link/20251014235604.3057003-3-kuniyu@google.com
If memcg is enabled, accept() acquires lock_sock() twice for each new
TCP/MPTCP socket in inet_csk_accept() and __inet_accept().
Let's move memcg operations from inet_csk_accept() to __inet_accept().
Note that SCTP somehow allocates a new socket by sk_alloc() in
sk->sk_prot->accept() and clones fields manually, instead of using
sk_clone_lock().
mem_cgroup_sk_alloc() is called for SCTP before __inet_accept(),
so I added the protocol check in __inet_accept(), but this can be
removed once SCTP uses sk_clone_lock().
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Reviewed-by: Shakeel Butt <shakeel.butt@linux.dev>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Roman Gushchin <roman.gushchin@linux.dev>
Link: https://patch.msgid.link/20251014235604.3057003-2-kuniyu@google.com
Some applications uses TCP_TX_DELAY socket option after TCP flow
is established.
Some metrics need to be updated, otherwise TCP might take time to
adapt to the new (emulated) RTT.
This patch adjusts tp->srtt_us, tp->rtt_min, icsk_rto
and sk->sk_pacing_rate.
This is best effort, and for instance icsk_rto is reset
without taking backoff into account.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20251013145926.833198-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Neal reported that using neper tcp_stream with TCP_TX_DELAY
set to 50ms would often lead to flows stuck in a small cwnd mode,
regardless of the congestion control.
While tcp_stream sets TCP_TX_DELAY too late after the connect(),
it highlighted two kernel bugs.
The following heuristic in tcp_tso_should_defer() seems wrong
for large RTT:
delta = tp->tcp_clock_cache - head->tstamp;
/* If next ACK is likely to come too late (half srtt), do not defer */
if ((s64)(delta - (u64)NSEC_PER_USEC * (tp->srtt_us >> 4)) < 0)
goto send_now;
If next ACK is expected to come in more than 1 ms, we should
not defer because we prefer a smooth ACK clocking.
While blamed commit was a step in the good direction, it was not
generic enough.
Another patch fixing TCP_TX_DELAY for established flows
will be proposed when net-next reopens.
Fixes: 50c8339e92 ("tcp: tso: restore IW10 after TSO autosizing")
Reported-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Tested-by: Neal Cardwell <ncardwell@google.com>
Link: https://patch.msgid.link/20251011115742.1245771-1-edumazet@google.com
[pabeni@redhat.com: fixed whitespace issue]
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Similarly to ipv4 tunnel, ipv6 version updates dev->needed_headroom, too.
While ipv4 tunnel headroom adjustment growth was limited in
commit 5ae1e9922b ("net: ip_tunnel: prevent perpetual headroom growth"),
ipv6 tunnel yet increases the headroom without any ceiling.
Reflect ipv4 tunnel headroom adjustment limit on ipv6 version.
Credits to Francesco Ruggeri, who was originally debugging this issue
and wrote local Arista-specific patch and a reproducer.
Fixes: 8eb30be035 ("ipv6: Create ip6_tnl_xmit")
Cc: Florian Westphal <fw@strlen.de>
Cc: Francesco Ruggeri <fruggeri05@gmail.com>
Signed-off-by: Dmitry Safonov <dima@arista.com>
Link: https://patch.msgid.link/20251009-ip6_tunnel-headroom-v2-1-8e4dbd8f7e35@arista.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Some applications (like selftests/net/tcp_mmap.c) call SO_RCVLOWAT
on their listener, before accept().
This has an unfortunate effect on wscale selection in
tcp_select_initial_window() during 3WHS.
For instance, tcp_mmap was negotiating wscale 4, regardless
of tcp_rmem[2] and sysctl_rmem_max.
Do not change tp->window_clamp if it is zero
or bigger than our computed value.
Zero value is special, it allows tcp_select_initial_window()
to enable autotuning.
Note that SO_RCVLOWAT use on listener is probably not wise,
because tp->scaling_ratio has a default value, possibly wrong.
Fixes: d1361840f8 ("tcp: fix SO_RCVLOWAT and RCVBUF autotuning")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Link: https://patch.msgid.link/20251003184119.2526655-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Rx path may be passing around unreferenced sockets, which means
that skb_set_owner_edemux() may not set skb->sk and PSP will crash:
KASAN: null-ptr-deref in range [0x0000000000000010-0x0000000000000017]
RIP: 0010:psp_reply_set_decrypted (./include/net/psp/functions.h:132 net/psp/psp_sock.c:287)
tcp_v6_send_response.constprop.0 (net/ipv6/tcp_ipv6.c:979)
tcp_v6_send_reset (net/ipv6/tcp_ipv6.c:1140 (discriminator 1))
tcp_v6_do_rcv (net/ipv6/tcp_ipv6.c:1683)
tcp_v6_rcv (net/ipv6/tcp_ipv6.c:1912)
Fixes: 659a2899a5 ("tcp: add datapath logic for PSP with inline key exchange")
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20251001022426.2592750-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Pull networking updates from Paolo Abeni:
"Core & protocols:
- Improve drop account scalability on NUMA hosts for RAW and UDP
sockets and the backlog, almost doubling the Pps capacity under DoS
- Optimize the UDP RX performance under stress, reducing contention,
revisiting the binary layout of the involved data structs and
implementing NUMA-aware locking. This improves UDP RX performance
by an additional 50%, even more under extreme conditions
- Add support for PSP encryption of TCP connections; this mechanism
has some similarities with IPsec and TLS, but offers superior HW
offloads capabilities
- Ongoing work to support Accurate ECN for TCP. AccECN allows more
than one congestion notification signal per RTT and is a building
block for Low Latency, Low Loss, and Scalable Throughput (L4S)
- Reorganize the TCP socket binary layout for data locality, reducing
the number of touched cachelines in the fastpath
- Refactor skb deferral free to better scale on large multi-NUMA
hosts, this improves TCP and UDP RX performances significantly on
such HW
- Increase the default socket memory buffer limits from 256K to 4M to
better fit modern link speeds
- Improve handling of setups with a large number of nexthop, making
dump operating scaling linearly and avoiding unneeded
synchronize_rcu() on delete
- Improve bridge handling of VLAN FDB, storing a single entry per
bridge instead of one entry per port; this makes the dump order of
magnitude faster on large switches
- Restore IP ID correctly for encapsulated packets at GSO
segmentation time, allowing GRO to merge packets in more scenarios
- Improve netfilter matching performance on large sets
- Improve MPTCP receive path performance by leveraging recently
introduced core infrastructure (skb deferral free) and adopting
recent TCP autotuning changes
- Allow bridges to redirect to a backup port when the bridge port is
administratively down
- Introduce MPTCP 'laminar' endpoint that con be used only once per
connection and simplify common MPTCP setups
- Add RCU safety to dst->dev, closing a lot of possible races
- A significant crypto library API for SCTP, MPTCP and IPv6 SR,
reducing code duplication
- Supports pulling data from an skb frag into the linear area of an
XDP buffer
Things we sprinkled into general kernel code:
- Generate netlink documentation from YAML using an integrated YAML
parser
Driver API:
- Support using IPv6 Flow Label in Rx hash computation and RSS queue
selection
- Introduce API for fetching the DMA device for a given queue,
allowing TCP zerocopy RX on more H/W setups
- Make XDP helpers compatible with unreadable memory, allowing more
easily building DevMem-enabled drivers with a unified XDP/skbs
datapath
- Add a new dedicated ethtool callback enabling drivers to provide
the number of RX rings directly, improving efficiency and clarity
in RX ring queries and RSS configuration
- Introduce a burst period for the health reporter, allowing better
handling of multiple errors due to the same root cause
- Support for DPLL phase offset exponential moving average,
controlling the average smoothing factor
Device drivers:
- Add a new Huawei driver for 3rd gen NIC (hinic3)
- Add a new SpacemiT driver for K1 ethernet MAC
- Add a generic abstraction for shared memory communication
devices (dibps)
- Ethernet high-speed NICs:
- nVidia/Mellanox:
- Use multiple per-queue doorbell, to avoid MMIO contention
issues
- support adjacent functions, allowing them to delegate their
SR-IOV VFs to sibling PFs
- support RSS for IPSec offload
- support exposing raw cycle counters in PTP and mlx5
- support for disabling host PFs.
- Intel (100G, ice, idpf):
- ice: support for SRIOV VFs over an Active-Active link
aggregate
- ice: support for firmware logging via debugfs
- ice: support for Earliest TxTime First (ETF) hardware offload
- idpf: support basic XDP functionalities and XSk
- Broadcom (bnxt):
- support Hyper-V VF ID
- dynamic SRIOV resource allocations for RoCE
- Meta (fbnic):
- support queue API, zero-copy Rx and Tx
- support basic XDP functionalities
- devlink health support for FW crashes and OTP mem corruptions
- expand hardware stats coverage to FEC, PHY, and Pause
- Wangxun:
- support ethtool coalesce options
- support for multiple RSS contexts
- Ethernet virtual:
- Macsec:
- replace custom netlink attribute checks with policy-level
checks
- Bonding:
- support aggregator selection based on port priority
- Microsoft vNIC:
- use page pool fragments for RX buffers instead of full pages
to improve memory efficiency
- Ethernet NICs consumer, and embedded:
- Qualcomm: support Ethernet function for IPQ9574 SoC
- Airoha: implement wlan offloading via NPU
- Freescale
- enetc: add NETC timer PTP driver and add PTP support
- fec: enable the Jumbo frame support for i.MX8QM
- Renesas (R-Car S4):
- support HW offloading for layer 2 switching
- support for RZ/{T2H, N2H} SoCs
- Cadence (macb): support TAPRIO traffic scheduling
- TI:
- support for Gigabit ICSS ethernet SoC (icssm-prueth)
- Synopsys (stmmac): a lot of cleanups
- Ethernet PHYs:
- Support 10g-qxgmi phy-mode for AQR412C, Felix DSA and Lynx PCS
driver
- Support bcm63268 GPHY power control
- Support for Micrel lan8842 PHY and PTP
- Support for Aquantia AQR412 and AQR115
- CAN:
- a large CAN-XL preparation work
- reorganize raw_sock and uniqframe struct to minimize memory
usage
- rcar_canfd: update the CAN-FD handling
- WiFi:
- extended Neighbor Awareness Networking (NAN) support
- S1G channel representation cleanup
- improve S1G support
- WiFi drivers:
- Intel (iwlwifi):
- major refactor and cleanup
- Broadcom (brcm80211):
- support for AP isolation
- RealTek (rtw88/89) rtw88/89:
- preparation work for RTL8922DE support
- MediaTek (mt76):
- HW restart improvements
- MLO support
- Qualcomm/Atheros (ath10k):
- GTK rekey fixes
- Bluetooth drivers:
- btusb: support for several new IDs for MT7925
- btintel: support for BlazarIW core
- btintel_pcie: support for _suspend() / _resume()
- btintel_pcie: support for Scorpious, Panther Lake-H484 IDs"
* tag 'net-next-6.18' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next: (1536 commits)
net: stmmac: Add support for Allwinner A523 GMAC200
dt-bindings: net: sun8i-emac: Add A523 GMAC200 compatible
Revert "Documentation: net: add flow control guide and document ethtool API"
octeontx2-pf: fix bitmap leak
octeontx2-vf: fix bitmap leak
net/mlx5e: Use extack in set rxfh callback
net/mlx5e: Introduce mlx5e_rss_params for RSS configuration
net/mlx5e: Introduce mlx5e_rss_init_params
net/mlx5e: Remove unused mdev param from RSS indir init
net/mlx5: Improve QoS error messages with actual depth values
net/mlx5e: Prevent entering switchdev mode with inconsistent netns
net/mlx5: HWS, Generalize complex matchers
net/mlx5: Improve write-combining test reliability for ARM64 Grace CPUs
selftests/net: add tcp_port_share to .gitignore
Revert "net/mlx5e: Update and set Xon/Xoff upon MTU set"
net: add NUMA awareness to skb_attempt_defer_free()
net: use llist for sd->defer_list
net: make softnet_data.defer_count an atomic
selftests: drv-net: psp: add tests for destroying devices
selftests: drv-net: psp: add test for auto-adjusting TCP MSS
...
Some applications are stuck to the 20th century and still use
small SO_RCVBUF values.
After the blamed commit, we can drop packets especially
when using LRO/hw-gro enabled NIC and small MSS (1500) values.
LRO/hw-gro NIC pack multiple segments into pages, allowing
tp->scaling_ratio to be set to a high value.
Whenever the receive queue gets full, we can receive a small packet
filling RWIN, but with a high skb->truesize, because most NIC use 4K page
plus sk_buff metadata even when receiving less than 1500 bytes of payload.
Even if we refine how tp->scaling_ratio is estimated,
we could have an issue at the start of the flow, because
the first round of packets (IW10) will be sent based on
the initial tp->scaling_ratio (1/2)
Relax tcp_can_ingest() to use skb->len instead of skb->truesize,
allowing the peer to use final RWIN, assuming a 'perfect'
scaling_ratio of 1.
Fixes: 1d2fbaad7c ("tcp: stronger sk_rcvbuf checks")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20250927092827.2707901-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Pull namespace updates from Christian Brauner:
"This contains a larger set of changes around the generic namespace
infrastructure of the kernel.
Each specific namespace type (net, cgroup, mnt, ...) embedds a struct
ns_common which carries the reference count of the namespace and so
on.
We open-coded and cargo-culted so many quirks for each namespace type
that it just wasn't scalable anymore. So given there's a bunch of new
changes coming in that area I've started cleaning all of this up.
The core change is to make it possible to correctly initialize every
namespace uniformly and derive the correct initialization settings
from the type of the namespace such as namespace operations, namespace
type and so on. This leaves the new ns_common_init() function with a
single parameter which is the specific namespace type which derives
the correct parameters statically. This also means the compiler will
yell as soon as someone does something remotely fishy.
The ns_common_init() addition also allows us to remove ns_alloc_inum()
and drops any special-casing of the initial network namespace in the
network namespace initialization code that Linus complained about.
Another part is reworking the reference counting. The reference
counting was open-coded and copy-pasted for each namespace type even
though they all followed the same rules. This also removes all open
accesses to the reference count and makes it private and only uses a
very small set of dedicated helpers to manipulate them just like we do
for e.g., files.
In addition this generalizes the mount namespace iteration
infrastructure introduced a few cycles ago. As reminder, the vfs makes
it possible to iterate sequentially and bidirectionally through all
mount namespaces on the system or all mount namespaces that the caller
holds privilege over. This allow userspace to iterate over all mounts
in all mount namespaces using the listmount() and statmount() system
call.
Each mount namespace has a unique identifier for the lifetime of the
systems that is exposed to userspace. The network namespace also has a
unique identifier working exactly the same way. This extends the
concept to all other namespace types.
The new nstree type makes it possible to lookup namespaces purely by
their identifier and to walk the namespace list sequentially and
bidirectionally for all namespace types, allowing userspace to iterate
through all namespaces. Looking up namespaces in the namespace tree
works completely locklessly.
This also means we can move the mount namespace onto the generic
infrastructure and remove a bunch of code and members from struct
mnt_namespace itself.
There's a bunch of stuff coming on top of this in the future but for
now this uses the generic namespace tree to extend a concept
introduced first for pidfs a few cycles ago. For a while now we have
supported pidfs file handles for pidfds. This has proven to be very
useful.
This extends the concept to cover namespaces as well. It is possible
to encode and decode namespace file handles using the common
name_to_handle_at() and open_by_handle_at() apis.
As with pidfs file handles, namespace file handles are exhaustive,
meaning it is not required to actually hold a reference to nsfs in
able to decode aka open_by_handle_at() a namespace file handle.
Instead the FD_NSFS_ROOT constant can be passed which will let the
kernel grab a reference to the root of nsfs internally and thus decode
the file handle.
Namespaces file descriptors can already be derived from pidfds which
means they aren't subject to overmount protection bugs. IOW, it's
irrelevant if the caller would not have access to an appropriate
/proc/<pid>/ns/ directory as they could always just derive the
namespace based on a pidfd already.
It has the same advantage as pidfds. It's possible to reliably and for
the lifetime of the system refer to a namespace without pinning any
resources and to compare them trivially.
Permission checking is kept simple. If the caller is located in the
namespace the file handle refers to they are able to open it otherwise
they must hold privilege over the owning namespace of the relevant
namespace.
The namespace file handle layout is exposed as uapi and has a stable
and extensible format. For now it simply contains the namespace
identifier, the namespace type, and the inode number. The stable
format means that userspace may construct its own namespace file
handles without going through name_to_handle_at() as they are already
allowed for pidfs and cgroup file handles"
* tag 'namespace-6.18-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs: (65 commits)
ns: drop assert
ns: move ns type into struct ns_common
nstree: make struct ns_tree private
ns: add ns_debug()
ns: simplify ns_common_init() further
cgroup: add missing ns_common include
ns: use inode initializer for initial namespaces
selftests/namespaces: verify initial namespace inode numbers
ns: rename to __ns_ref
nsfs: port to ns_ref_*() helpers
net: port to ns_ref_*() helpers
uts: port to ns_ref_*() helpers
ipv4: use check_net()
net: use check_net()
net-sysfs: use check_net()
user: port to ns_ref_*() helpers
time: port to ns_ref_*() helpers
pid: port to ns_ref_*() helpers
ipc: port to ns_ref_*() helpers
cgroup: port to ns_ref_*() helpers
...
Currently, packets with fixed IDs will be merged only if their
don't-fragment bit is set. This restriction is unnecessary since
packets without the don't-fragment bit will be forwarded as-is even
if they were merged together. The merged packets will be segmented
into their original forms before being forwarded, either by GSO or
by TSO. The IDs will also remain identical unless NETIF_F_TSO_MANGLEID
is set, in which case the IDs can become incrementing, which is also fine.
Clean up the code by removing the unnecessary don't-fragment checks.
Signed-off-by: Richard Gobert <richardbgobert@gmail.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250923085908.4687-5-richardbgobert@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Only merge encapsulated packets if their outer IDs are either
incrementing or fixed, just like for inner IDs and IDs of non-encapsulated
packets.
Add another ip_fixedid bit for a total of two bits: one for outer IDs (and
for unencapsulated packets) and one for inner IDs.
This commit preserves the current behavior of GSO where only the IDs of the
inner-most headers are restored correctly.
Signed-off-by: Richard Gobert <richardbgobert@gmail.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250923085908.4687-3-richardbgobert@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Remove is_ipv6 from napi_gro_cb and use sk->sk_family instead.
This frees up space for another ip_fixedid bit that will be added
in the next commit.
udp_sock_create always creates either a AF_INET or a AF_INET6 socket,
so using sk->sk_family is reliable. In IPv6-FOU, cfg->ipv6_v6only is
always enabled.
Signed-off-by: Richard Gobert <richardbgobert@gmail.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250923085908.4687-2-richardbgobert@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
