- Split cq_lock into two smaller locks: cq_prod_lock and
cq_cached_prod_lock
- Avoid disabling/enabling interrupts in the hot xmit path
In either xsk_cq_cancel_locked() or xsk_cq_reserve_locked() function,
the race condition is only between multiple xsks sharing the same
pool. They are all in the process context rather than interrupt context,
so now the small lock named cq_cached_prod_lock can be used without
handling interrupts.
While cq_cached_prod_lock ensures the exclusive modification of
@cached_prod, cq_prod_lock in xsk_cq_submit_addr_locked() only cares
about @producer and corresponding @desc. Both of them don't necessarily
be consistent with @cached_prod protected by cq_cached_prod_lock.
That's the reason why the previous big lock can be split into two
smaller ones. Please note that SPSC rule is all about the global state
of producer and consumer that can affect both layers instead of local
or cached ones.
Frequently disabling and enabling interrupt are very time consuming
in some cases, especially in a per-descriptor granularity, which now
can be avoided after this optimization, even when the pool is shared by
multiple xsks.
With this patch, the performance number[1] could go from 1,872,565 pps
to 1,961,009 pps. It's a minor rise of around 5%.
[1]: taskset -c 1 ./xdpsock -i enp2s0f1 -q 0 -t -S -s 64
Signed-off-by: Jason Xing <kernelxing@tencent.com>
Acked-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Link: https://patch.msgid.link/20251030000646.18859-3-kerneljasonxing@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Handle the NIC hardware link state events received from the HW
channel, then set the proper link state accordingly.
And, add a feature bit, GDMA_DRV_CAP_FLAG_1_HW_VPORT_LINK_AWARE,
to inform the NIC hardware this handler exists.
Our MANA NIC only sends out the link state down/up messages
when we need to let the VM rerun DHCP client and change IP
address. So, add netif_carrier_on() in the probe(), let the NIC
show the right initial state in /sys/class/net/ethX/operstate.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Link: https://patch.msgid.link/1761770601-16920-1-git-send-email-haiyangz@linux.microsoft.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Florian Westphal says:
====================
netfilter: updates for net-next
1) Convert nf_tables 'nft_set_iter' usage to use C99 struct
initialization, from Fernando Fernandez Mancera.
2) Disallow nf_conntrack_max=0. This was an (undocumented)
historic inheritance from ip_conntrack (ipv4 only nf_conntrack
predecessor). Doing so will simplify future changes to make
this pernet-tuneable.
3) Fix a typo in conntrack.h comment, from Weibiao Tu.
* tag 'nf-next-25-10-30' of https://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf-next:
netfilter: fix typo in nf_conntrack_l4proto.h comment
netfilter: conntrack: disable 0 value for conntrack_max setting
netfilter: nf_tables: use C99 struct initializer for nft_set_iter
====================
Link: https://patch.msgid.link/20251030121954.29175-1-fw@strlen.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Think SMC_WR_BUF_CNT_SEND := SMC_WR_BUF_CNT used in send context and
SMC_WR_BUF_CNT_RECV := 3 * SMC_WR_BUF_CNT used in recv context. Those
get replaced with lgr->max_send_wr and lgr->max_recv_wr respective.
Please note that although with the default sysctl values
qp_attr.cap.max_send_wr == qp_attr.cap.max_recv_wr is maintained but
can not be assumed to be generally true any more. I see no downside to
that, but my confidence level is rather modest.
Signed-off-by: Halil Pasic <pasic@linux.ibm.com>
Reviewed-by: Sidraya Jayagond <sidraya@linux.ibm.com>
Reviewed-by: Dust Li <dust.li@linux.alibaba.com>
Tested-by: Mahanta Jambigi <mjambigi@linux.ibm.com>
Link: https://patch.msgid.link/20251027224856.2970019-2-pasic@linux.ibm.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
In the comment for nf_conntrack_l4proto.h, the word "nfnetink" was
incorrectly spelled. It has been corrected to "nfnetlink".
Fixes a typo to enhance readability and ensure consistency.
Signed-off-by: caivive (Weibiao Tu) <cavivie@gmail.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
mac80211 already reports some basic information in the radiotap header
with the known fields declared by the driver. However, drivers may want
to report more accurate information and in that case the full VHT
radiotap structure needs to be provided.
Add a new RX_FLAG_RADIOTAP_VHT which is set when the VHT information
should be pulled from the skb. Update the code to fill in the VHT fields
to only do so when requested by the driver or if the information has not
yet been set. This way the driver can fully control the information if
it chooses so.
Signed-off-by: Benjamin Berg <benjamin.berg@intel.com>
Reviewed-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
Link: https://patch.msgid.link/20251027142118.0bad1c307a21.I2cf285c20a822698039603f2af00ed9c548f2ee0@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
When a netdev issues a RX async resync request for a TLS connection,
the TLS module handles it by logging record headers and attempting to
match them to the tcp_sn provided by the device. If a match is found,
the TLS module approves the tcp_sn for resynchronization.
While waiting for a device response, the TLS module also increments
rcd_delta each time a new TLS record is received, tracking the distance
from the original resync request.
However, if the device response is delayed or fails (e.g due to
unstable connection and device getting out of tracking, hardware
errors, resource exhaustion etc.), the TLS module keeps logging and
incrementing, which can lead to a WARN() when rcd_delta exceeds the
threshold.
To address this, introduce tls_offload_rx_resync_async_request_cancel()
to explicitly cancel resync requests when a device response failure is
detected. Call this helper also as a final safeguard when rcd_delta
crosses its threshold, as reaching this point implies that earlier
cancellation did not occur.
Signed-off-by: Shahar Shitrit <shshitrit@nvidia.com>
Reviewed-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/1761508983-937977-3-git-send-email-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Update tls_offload_rx_resync_async_request_start() and
tls_offload_rx_resync_async_request_end() to get a struct
tls_offload_resync_async parameter directly, rather than
extracting it from struct sock.
This change aligns the function signatures with the upcoming
tls_offload_rx_resync_async_request_cancel() helper, which
will be introduced in a subsequent patch.
Signed-off-by: Shahar Shitrit <shshitrit@nvidia.com>
Reviewed-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/1761508983-937977-2-git-send-email-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add the ability to append the incoming IP interface information to
ICMPv6 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.ipv6.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 icmp6_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).
Since commit 20e1954fe2 ("ipv6: RFC 4884 partial support for SIT/GRE
tunnels") it is possible for icmp6_send() to be called with an skb that
already contains ICMP extensions. This can happen when we receive an
ICMPv4 message with extensions from a tunnel and translate it to an
ICMPv6 message towards an IPv6 host in the overlay network. I could not
find an RFC that supports this behavior, but it makes sense to not
overwrite the original extensions that were appended to the packet.
Therefore, avoid appending extensions if the length field in the
provided ICMPv6 header is already filled.
Export netdev_copy_name() using EXPORT_IPV6_MOD_GPL() to make it
available to IPv6 when it is built as a module.
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/20251027082232.232571-3-idosch@nvidia.com
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>
'struct sctp_sched_ops' is not modified in these drivers.
Constifying this structure moves some data to a read-only section, so
increases overall security, especially when the structure holds some
function pointers.
On a x86_64, with allmodconfig, as an example:
Before:
======
text data bss dec hex filename
8019 568 0 8587 218b net/sctp/stream_sched_fc.o
After:
=====
text data bss dec hex filename
8275 312 0 8587 218b net/sctp/stream_sched_fc.o
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Link: https://patch.msgid.link/dce03527eb7b7cc8a3c26d5cdac12bafe3350135.1761377890.git.christophe.jaillet@wanadoo.fr
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Remove the NET_IOV_MAX workaround from the net_iov_type enum. This entry
was previously added to force the enum size to unsigned long to satisfy
the NET_IOV_ASSERT_OFFSET static assertions.
After commit f3d85c9ee5 ("netmem: introduce struct netmem_desc
mirroring struct page") this approach became unnecessary by placing the
net_iov_type after the netmem_desc. Placing the net_iov_type after
netmem_desc results in the net_iov_type size having no effect on the
position or layout of the fields that mirror the struct page.
The layout before this patch:
struct net_iov {
union {
struct netmem_desc desc; /* 0 48 */
struct {
long unsigned int _flags; /* 0 8 */
long unsigned int pp_magic; /* 8 8 */
struct page_pool * pp; /* 16 8 */
long unsigned int _pp_mapping_pad; /* 24 8 */
long unsigned int dma_addr; /* 32 8 */
atomic_long_t pp_ref_count; /* 40 8 */
}; /* 0 48 */
}; /* 0 48 */
struct net_iov_area * owner; /* 48 8 */
enum net_iov_type type; /* 56 8 */
/* size: 64, cachelines: 1, members: 3 */
};
The layout after this patch:
struct net_iov {
union {
struct netmem_desc desc; /* 0 48 */
struct {
long unsigned int _flags; /* 0 8 */
long unsigned int pp_magic; /* 8 8 */
struct page_pool * pp; /* 16 8 */
long unsigned int _pp_mapping_pad; /* 24 8 */
long unsigned int dma_addr; /* 32 8 */
atomic_long_t pp_ref_count; /* 40 8 */
}; /* 0 48 */
}; /* 0 48 */
struct net_iov_area * owner; /* 48 8 */
enum net_iov_type type; /* 56 4 */
/* size: 64, cachelines: 1, members: 3 */
/* padding: 4 */
};
Signed-off-by: Bobby Eshleman <bobbyeshleman@meta.com>
Reviewed-by: Mina Almasry <almasrymina@google.com>
Link: https://patch.msgid.link/20251024-b4-devmem-remove-niov-max-v1-1-ba72c68bc869@meta.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
sctp_accept() will use sk_clone_lock(), but it will be called
with the parent socket locked, and sctp_migrate() acquires the
child lock later.
Let's add no lock version of sk_clone_lock().
Note that lockdep complains if we simply use bh_lock_sock_nested().
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Xin Long <lucien.xin@gmail.com>
Link: https://patch.msgid.link/20251023231751.4168390-5-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
During a handshake, an endpoint may specify a maximum record size limit.
Currently, the kernel defaults to TLS_MAX_PAYLOAD_SIZE (16KB) for the
maximum record size. Meaning that, the outgoing records from the kernel
can exceed a lower size negotiated during the handshake. In such a case,
the TLS endpoint must send a fatal "record_overflow" alert [1], and
thus the record is discarded.
Upcoming Western Digital NVMe-TCP hardware controllers implement TLS
support. For these devices, supporting TLS record size negotiation is
necessary because the maximum TLS record size supported by the controller
is less than the default 16KB currently used by the kernel.
Currently, there is no way to inform the kernel of such a limit. This patch
adds support to a new setsockopt() option `TLS_TX_MAX_PAYLOAD_LEN` that
allows for setting the maximum plaintext fragment size. Once set, outgoing
records are no larger than the size specified. This option can be used to
specify the record size limit.
[1] https://www.rfc-editor.org/rfc/rfc8449
Signed-off-by: Wilfred Mallawa <wilfred.mallawa@wdc.com>
Reviewed-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://patch.msgid.link/20251022001937.20155-1-wilfred.opensource@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In multi-radio wiphy architecture, where a single wiphy can have
multiple radios tied to it, radio specific configuration parameters
and global wiphy parameters are maintained for the entire physical
device and common to all radios. But, each radio in a wiphy can have
different values for each radio configuration parameter, like RTS
threshold. With the current debugfs directory structure, the values
of global wiphy configuration parameters can be viewed, but, values
of individual radio configuration parameters cannot be viewed, as
radio specific configuration parameters are not maintained, separately.
To address this, in addition to maintaining global wiphy configuration
parameters common to all radios, create separate debugfs directories
for each radio in a wiphy to maintain parameters corresponding to that
radio in this directory.
In implementation, maintain a dentry structure in wiphy_radio_cfg, a
structure containing radio configurations of a wiphy. This struct is
maintained to denote per-radio configurations of a wiphy. Create
separate directories representing each radio within phy#X directory in
debugfs during wiphy registration.
Sample directory structure with this change:
ls /sys/kernel/debug/ieee80211/phy0/radio
radio0/ radio1/ radio2/
Signed-off-by: Roopni Devanathan <quic_rdevanat@quicinc.com>
Link: https://patch.msgid.link/20251024044649.483557-2-quic_rdevanat@quicinc.com
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
In multi-radio devices, it is possible to have an MLD AP and a monitor
interface active at the same time. In such cases, monitor mode may not
be able to specify a fixed channel and could end up capturing frames
from all radios, including those outside the intended frequency bands.
This patch adds frequency validation for monitor mode. Received frames
are now only processed if their frequency fall within the allowed ranges
of the radios specified by the interface's radio_mask.
This prevents monitor mode from capturing frames outside the supported radio.
Signed-off-by: Ryder Lee <ryder.lee@mediatek.com>
Link: https://patch.msgid.link/700b8284e845d96654eb98431f8eeb5a81503862.1758647858.git.ryder.lee@mediatek.com
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
NEIGH_VAR() is read locklessly in the fast path, and IPv6 ndisc uses
NEIGH_VAR_SET() locklessly.
The next patch will convert neightbl_dump_info() to RCU.
Let's annotate accesses to neigh_param with READ_ONCE() and WRITE_ONCE().
Note that ndisc_ifinfo_sysctl_change() uses &NEIGH_VAR() and we cannot
use '&' with READ_ONCE(), so NEIGH_VAR_PTR() is introduced.
Note also that NEIGH_VAR_INIT() does not need WRITE_ONCE() as it is before
parms is published. Also, the only user hippi_neigh_setup_dev() is no
longer called since commit e3804cbebb ("net: remove COMPAT_NET_DEV_OPS"),
which looks wrong, but probably no one uses HIPPI and RoadRunner.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20251022054004.2514876-3-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Periodic advertising enabled flag cannot be tracked by the enabled
flag since advertising and periodic advertising each can be
enabled/disabled separately from one another causing the states to be
inconsistent when for example an advertising set is disabled its
enabled flag is set to false which is then used for periodic which has
not being disabled.
Fixes: eca0ae4aea ("Bluetooth: Add initial implementation of BIS connections")
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
This reverts commit c9d84da18d. It
replaces in L2CAP calls to msecs_to_jiffies() to secs_to_jiffies()
and updates the constants accordingly. But the constants are also
used in LCAP Configure Request and L2CAP Configure Response which
expect values in milliseconds.
This may prevent correct usage of L2CAP channel.
To fix it, keep those constants in milliseconds and so revert this
change.
Fixes: c9d84da18d ("Bluetooth: L2CAP: convert timeouts to secs_to_jiffies()")
Signed-off-by: Frédéric Danis <frederic.danis@collabora.com>
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
There is a BUG: KASAN: stack-out-of-bounds in set_mesh_sync due to
memcpy from badly declared on-stack flexible array.
Another crash is in set_mesh_complete() due to double list_del via
mgmt_pending_valid + mgmt_pending_remove.
Use DEFINE_FLEX to declare the flexible array right, and don't memcpy
outside bounds.
As mgmt_pending_valid removes the cmd from list, use mgmt_pending_free,
and also report status on error.
Fixes: 302a1f674c ("Bluetooth: MGMT: Fix possible UAFs")
Signed-off-by: Pauli Virtanen <pav@iki.fi>
Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
This fixes the state tracking of advertisement set/instance 0x00 which
is considered a legacy instance and is not tracked individually by
adv_instances list, previously it was assumed that hci_dev itself would
track it via HCI_LE_ADV but that is a global state not specifc to
instance 0x00, so to fix it a new flag is introduced that only tracks the
state of instance 0x00.
Fixes: 1488af7b8b ("Bluetooth: hci_sync: Fix hci_resume_advertising_sync")
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
UDP TX packets destructor is sock_wfree().
It suffers from a cache line bouncing in sock_def_write_space_wfree().
Instead of reading sk->sk_wmem_alloc after we just did an atomic RMW
on it, use __refcount_sub_and_test() to get the old value for free,
and pass the new value to sock_def_write_space_wfree().
Add __sock_writeable() helper.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20251017133712.2842665-1-edumazet@google.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Correct multiple kernel-doc warnings in nl802154.h:
- Fix a typo on one enum name to avoid a kernel-doc warning.
- Drop 2 enum descriptions that are no longer needed.
- Mark 2 internal enums as "private:" so that kernel-doc is not needed
for them.
Warning: nl802154.h:239 Enum value 'NL802154_CAP_ATTR_MAX_MAXBE' not described in enum 'nl802154_wpan_phy_capability_attr'
Warning: nl802154.h:239 Excess enum value '%NL802154_CAP_ATTR_MIN_CCA_ED_LEVEL' description in 'nl802154_wpan_phy_capability_attr'
Warning: nl802154.h:239 Excess enum value '%NL802154_CAP_ATTR_MAX_CCA_ED_LEVEL' description in 'nl802154_wpan_phy_capability_attr'
Warning: nl802154.h:369 Enum value '__NL802154_CCA_OPT_ATTR_AFTER_LAST' not described in enum 'nl802154_cca_opts'
Warning: nl802154.h:369 Enum value 'NL802154_CCA_OPT_ATTR_MAX' not described in enum 'nl802154_cca_opts'
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20251016035917.1148012-1-rdunlap@infradead.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 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>
In {tcp6,udp6,raw6}_sock, struct ipv6_pinfo is always placed at
the beginning of a new cache line because
1. __alignof__(struct tcp_sock) is 64 due to ____cacheline_aligned
of __cacheline_group_begin(tcp_sock_write_tx)
2. __alignof__(struct udp_sock) is 64 due to ____cacheline_aligned
of struct numa_drop_counters
3. in raw6_sock, struct numa_drop_counters is placed before
struct ipv6_pinfo
. struct ipv6_pinfo is 136 bytes, but the last cache line is
only used by ipv6_fl_list:
$ pahole -C ipv6_pinfo vmlinux
struct ipv6_pinfo {
...
/* --- cacheline 2 boundary (128 bytes) --- */
struct ipv6_fl_socklist * ipv6_fl_list; /* 128 8 */
/* size: 136, cachelines: 3, members: 23 */
Let's move ipv6_fl_list from struct ipv6_pinfo to struct inet_sock
to save a full cache line for {tcp6,udp6,raw6}_sock.
Now, struct ipv6_pinfo is 128 bytes, and {tcp6,udp6,raw6}_sock have
64 bytes less, while {tcp,udp,raw}_sock retain the same size.
Before:
# grep -E "^(RAW|UDP[^L\-]|TCP)" /proc/slabinfo | awk '{print $1, "\t", $4}'
RAWv6 1408
UDPv6 1472
TCPv6 2560
RAW 1152
UDP 1280
TCP 2368
After:
# grep -E "^(RAW|UDP[^L\-]|TCP)" /proc/slabinfo | awk '{print $1, "\t", $4}'
RAWv6 1344
UDPv6 1408
TCPv6 2496
RAW 1152
UDP 1280
TCP 2368
Also, ipv6_fl_list and inet_flags (SNDFLOW bit) are placed in the
same cache line.
$ pahole -C inet_sock vmlinux
...
/* --- cacheline 11 boundary (704 bytes) was 56 bytes ago --- */
struct ipv6_pinfo * pinet6; /* 760 8 */
/* --- cacheline 12 boundary (768 bytes) --- */
struct ipv6_fl_socklist * ipv6_fl_list; /* 768 8 */
unsigned long inet_flags; /* 776 8 */
Doc churn is due to the insufficient Type column (only 1 space short).
Suggested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20251014224210.2964778-1-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Remove busylock spinlock and use a lockless list (llist)
to reduce spinlock contention to the minimum.
Idea is that only one cpu might spin on the qdisc spinlock,
while others simply add their skb in the llist.
After this patch, we get a 300 % improvement on heavy TX workloads.
- Sending twice the number of packets per second.
- While consuming 50 % less cycles.
Note that this also allows in the future to submit batches
to various qdisc->enqueue() methods.
Tested:
- Dual Intel(R) Xeon(R) 6985P-C (480 hyper threads).
- 100Gbit NIC, 30 TX queues with FQ packet scheduler.
- echo 64 >/sys/kernel/slab/skbuff_small_head/cpu_partial (avoid contention in mm)
- 240 concurrent "netperf -t UDP_STREAM -- -m 120 -n"
Before:
16 Mpps (41 Mpps if each thread is pinned to a different cpu)
vmstat 2 5
procs -----------memory---------- ---swap-- -----io---- -system-- ------cpu-----
r b swpd free buff cache si so bi bo in cs us sy id wa st
243 0 0 2368988672 51036 1100852 0 0 146 1 242 60 0 9 91 0 0
244 0 0 2368988672 51036 1100852 0 0 536 10 487745 14718 0 52 48 0 0
244 0 0 2368988672 51036 1100852 0 0 512 0 503067 46033 0 52 48 0 0
244 0 0 2368988672 51036 1100852 0 0 512 0 494807 12107 0 52 48 0 0
244 0 0 2368988672 51036 1100852 0 0 702 26 492845 10110 0 52 48 0 0
Lock contention (1 second sample taken on 8 cores)
perf lock record -C0-7 sleep 1; perf lock contention
contended total wait max wait avg wait type caller
442111 6.79 s 162.47 ms 15.35 us spinlock dev_hard_start_xmit+0xcd
5961 9.57 ms 8.12 us 1.60 us spinlock __dev_queue_xmit+0x3a0
244 560.63 us 7.63 us 2.30 us spinlock do_softirq+0x5b
13 25.09 us 3.21 us 1.93 us spinlock net_tx_action+0xf8
If netperf threads are pinned, spinlock stress is very high.
perf lock record -C0-7 sleep 1; perf lock contention
contended total wait max wait avg wait type caller
964508 7.10 s 147.25 ms 7.36 us spinlock dev_hard_start_xmit+0xcd
201 268.05 us 4.65 us 1.33 us spinlock __dev_queue_xmit+0x3a0
12 26.05 us 3.84 us 2.17 us spinlock do_softirq+0x5b
@__dev_queue_xmit_ns:
[256, 512) 21 | |
[512, 1K) 631 | |
[1K, 2K) 27328 |@ |
[2K, 4K) 265392 |@@@@@@@@@@@@@@@@ |
[4K, 8K) 417543 |@@@@@@@@@@@@@@@@@@@@@@@@@@ |
[8K, 16K) 826292 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@|
[16K, 32K) 733822 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ |
[32K, 64K) 19055 |@ |
[64K, 128K) 17240 |@ |
[128K, 256K) 25633 |@ |
[256K, 512K) 4 | |
After:
29 Mpps (57 Mpps if each thread is pinned to a different cpu)
vmstat 2 5
procs -----------memory---------- ---swap-- -----io---- -system-- ------cpu-----
r b swpd free buff cache si so bi bo in cs us sy id wa st
78 0 0 2369573632 32896 1350988 0 0 22 0 331 254 0 8 92 0 0
75 0 0 2369573632 32896 1350988 0 0 22 50 425713 280199 0 23 76 0 0
104 0 0 2369573632 32896 1350988 0 0 290 0 430238 298247 0 23 76 0 0
86 0 0 2369573632 32896 1350988 0 0 132 0 428019 291865 0 24 76 0 0
90 0 0 2369573632 32896 1350988 0 0 502 0 422498 278672 0 23 76 0 0
perf lock record -C0-7 sleep 1; perf lock contention
contended total wait max wait avg wait type caller
2524 116.15 ms 486.61 us 46.02 us spinlock __dev_queue_xmit+0x55b
5821 107.18 ms 371.67 us 18.41 us spinlock dev_hard_start_xmit+0xcd
2377 9.73 ms 35.86 us 4.09 us spinlock ___slab_alloc+0x4e0
923 5.74 ms 20.91 us 6.22 us spinlock ___slab_alloc+0x5c9
121 3.42 ms 193.05 us 28.24 us spinlock net_tx_action+0xf8
6 564.33 us 167.60 us 94.05 us spinlock do_softirq+0x5b
If netperf threads are pinned (~54 Mpps)
perf lock record -C0-7 sleep 1; perf lock contention
32907 316.98 ms 195.98 us 9.63 us spinlock dev_hard_start_xmit+0xcd
4507 61.83 ms 212.73 us 13.72 us spinlock __dev_queue_xmit+0x554
2781 23.53 ms 40.03 us 8.46 us spinlock ___slab_alloc+0x5c9
3554 18.94 ms 34.69 us 5.33 us spinlock ___slab_alloc+0x4e0
233 9.09 ms 215.70 us 38.99 us spinlock do_softirq+0x5b
153 930.66 us 48.67 us 6.08 us spinlock net_tx_action+0xfd
84 331.10 us 14.22 us 3.94 us spinlock ___slab_alloc+0x5c9
140 323.71 us 9.94 us 2.31 us spinlock ___slab_alloc+0x4e0
@__dev_queue_xmit_ns:
[128, 256) 1539830 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ |
[256, 512) 2299558 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@|
[512, 1K) 483936 |@@@@@@@@@@ |
[1K, 2K) 265345 |@@@@@@ |
[2K, 4K) 145463 |@@@ |
[4K, 8K) 54571 |@ |
[8K, 16K) 10270 | |
[16K, 32K) 9385 | |
[32K, 64K) 7749 | |
[64K, 128K) 26799 | |
[128K, 256K) 2665 | |
[256K, 512K) 665 | |
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Toke Høiland-Jørgensen <toke@redhat.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Tested-by: Jamal Hadi Salim <jhs@mojatatu.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Link: https://patch.msgid.link/20251014171907.3554413-7-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
If a socket has sk->sk_bypass_prot_mem flagged, the socket opts out
of the global protocol memory accounting.
Let's control the flag by a new sysctl knob.
The flag is written once during socket(2) and is inherited to child
sockets.
Tested with a script that creates local socket pairs and send()s a
bunch of data without recv()ing.
Setup:
# mkdir /sys/fs/cgroup/test
# echo $$ >> /sys/fs/cgroup/test/cgroup.procs
# sysctl -q net.ipv4.tcp_mem="1000 1000 1000"
# ulimit -n 524288
Without net.core.bypass_prot_mem, charged to tcp_mem & memcg
# python3 pressure.py &
# cat /sys/fs/cgroup/test/memory.stat | grep sock
sock 22642688 <-------------------------------------- charged to memcg
# cat /proc/net/sockstat| grep TCP
TCP: inuse 2006 orphan 0 tw 0 alloc 2008 mem 5376 <-- charged to tcp_mem
# ss -tn | head -n 5
State Recv-Q Send-Q Local Address:Port Peer Address:Port
ESTAB 2000 0 127.0.0.1:34479 127.0.0.1:53188
ESTAB 2000 0 127.0.0.1:34479 127.0.0.1:49972
ESTAB 2000 0 127.0.0.1:34479 127.0.0.1:53868
ESTAB 2000 0 127.0.0.1:34479 127.0.0.1:53554
# nstat | grep Pressure || echo no pressure
TcpExtTCPMemoryPressures 1 0.0
With net.core.bypass_prot_mem=1, charged to memcg only:
# sysctl -q net.core.bypass_prot_mem=1
# python3 pressure.py &
# cat /sys/fs/cgroup/test/memory.stat | grep sock
sock 2757468160 <------------------------------------ charged to memcg
# cat /proc/net/sockstat | grep TCP
TCP: inuse 2006 orphan 0 tw 0 alloc 2008 mem 0 <- NOT charged to tcp_mem
# ss -tn | head -n 5
State Recv-Q Send-Q Local Address:Port Peer Address:Port
ESTAB 111000 0 127.0.0.1:36019 127.0.0.1:49026
ESTAB 110000 0 127.0.0.1:36019 127.0.0.1:45630
ESTAB 110000 0 127.0.0.1:36019 127.0.0.1:44870
ESTAB 111000 0 127.0.0.1:36019 127.0.0.1:45274
# nstat | grep Pressure || echo no pressure
no pressure
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/20251014235604.3057003-4-kuniyu@google.com
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
This is a followup of commit 726e9e8b94 ("tcp: refine
skb->ooo_okay setting") and of prior commit in this series
("net: control skb->ooo_okay from skb_set_owner_w()")
skb->ooo_okay might never be set for bulk flows that always
have at least one skb in a qdisc queue of NIC queue,
especially if TX completion is delayed because of a stressed cpu.
The so-called "strange attractors" has caused many performance
issues (see for instance 9b462d02d6 ("tcp: TCP Small Queues
and strange attractors")), we need to do better.
We have tried very hard to avoid reorders because TCP was
not dealing with them nicely a decade ago.
Use the new net.core.txq_reselection_ms sysctl to let
flows follow XPS and select a more efficient queue.
After this patch, we no longer have to make sure threads
are pinned to cpus, they now can be migrated without
adding too much spinlock/qdisc/TX completion pressure anymore.
TX completion part was problematic, because it added false sharing
on various socket fields, but also added false sharing and spinlock
contention in mm layers. Calling skb_orphan() from ndo_start_xmit()
is not an option unfortunately.
Note for later:
1) move sk->sk_tx_queue_mapping closer
to sk_tx_queue_mapping_jiffies for better cache locality.
2) Study if 9b462d02d6 ("tcp: TCP Small Queues
and strange attractors") could be revised.
Tested:
Used a host with 32 TX queues, shared by groups of 8 cores.
XPS setup :
echo ff >/sys/class/net/eth1/queue/tx-0/xps_cpus
echo ff00 >/sys/class/net/eth1/queue/tx-1/xps_cpus
echo ff0000 >/sys/class/net/eth1/queue/tx-2/xps_cpus
echo ff000000 >/sys/class/net/eth1/queue/tx-3/xps_cpus
echo ff,00000000 >/sys/class/net/eth1/queue/tx-4/xps_cpus
echo ff00,00000000 >/sys/class/net/eth1/queue/tx-5/xps_cpus
echo ff0000,00000000 >/sys/class/net/eth1/queue/tx-6/xps_cpus
echo ff000000,00000000 >/sys/class/net/eth1/queue/tx-7/xps_cpus
...
Launched a tcp_stream with 15 threads and 1000 flows, initially affined to core 0-15
taskset -c 0-15 tcp_stream -T15 -F1000 -l1000 -c -H target_host
Checked that only queues 0 and 1 are used as instructed by XPS :
tc -s qdisc show dev eth1|grep backlog|grep -v "backlog 0b 0p"
backlog 123489410b 1890p
backlog 69809026b 1064p
backlog 52401054b 805p
Then force each thread to run on cpu 1,9,17,25,33,41,49,57,65,73,81,89,97,105,113,121
C=1;PID=`pidof tcp_stream`;for P in `ls /proc/$PID/task`; do taskset -pc $C $P; C=$(($C + 8));done
Set txq_reselection_ms to 1000
echo 1000 > /proc/sys/net/core/txq_reselection_ms
Check that the flows have migrated nicely:
tc -s qdisc show dev eth1|grep backlog|grep -v "backlog 0b 0p"
backlog 130508314b 1916p
backlog 8584380b 126p
backlog 8584380b 126p
backlog 8379990b 123p
backlog 8584380b 126p
backlog 8487484b 125p
backlog 8584380b 126p
backlog 8448120b 124p
backlog 8584380b 126p
backlog 8720640b 128p
backlog 8856900b 130p
backlog 8584380b 126p
backlog 8652510b 127p
backlog 8448120b 124p
backlog 8516250b 125p
backlog 7834950b 115p
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20251013152234.842065-5-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
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>
Now that we have struct netmem_desc, it'd better access the pp fields
via struct netmem_desc rather than struct net_iov.
Introduce netmem_to_nmdesc() for safely converting netmem_ref to
netmem_desc regardless of the type underneath e.i. netmem_desc, net_iov.
While at it, remove __netmem_clear_lsb() and make netmem_to_nmdesc()
used instead.
Suggested-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Byungchul Park <byungchul@sk.com>
Reviewed-by: Mina Almasry <almasrymina@google.com>
Link: https://patch.msgid.link/20251013044133.69472-1-byungchul@sk.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
