kernel/linux.git/net/core/sysctl_net_core.c, branch v6.19.11

net: add net.core.qdisc_max_burst

2026-01-13T09:12:11+00:00

In blamed commit, I added a check against the temporary queue built in __dev_xmit_skb(). Idea was to drop packets early, before any spinlock was acquired. if (unlikely(defer_count > READ_ONCE(q->limit))) { kfree_skb_reason(skb, SKB_DROP_REASON_QDISC_DROP); return NET_XMIT_DROP; } It turned out that HTB Qdisc has a zero q->limit. HTB limits packets on a per-class basis. Some of our tests became flaky. Add a new sysctl : net.core.qdisc_max_burst to control how many packets can be stored in the temporary lockless queue. Also add a new QDISC_BURST_DROP drop reason to better diagnose future issues. Thanks Neal ! Fixes: 100dfa74cad9 ("net: dev_queue_xmit() llist adoption") Reported-and-bisected-by: Neal Cardwell Signed-off-by: Eric Dumazet Reviewed-by: Neal Cardwell Link: https://patch.msgid.link/20260107104159.3669285-1-edumazet@google.com Signed-off-by: Paolo Abeni

net: Introduce net.core.bypass_prot_mem sysctl.

2025-10-16T19:04:47+00:00

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 Signed-off-by: Martin KaFai Lau Reviewed-by: Shakeel Butt Reviewed-by: Eric Dumazet Acked-by: Roman Gushchin Link: https://patch.msgid.link/20251014235604.3057003-4-kuniyu@google.com

net: add /proc/sys/net/core/txq_reselection_ms control

2025-10-15T16:04:21+00:00

Add a new sysctl to control how often a queue reselection can happen even if a flow has a persistent queue of skbs in a Qdisc or NIC queue. A value of zero means the feature is disabled. Default is 1000 (1 second). This sysctl is used in the following patch. Signed-off-by: Eric Dumazet Reviewed-by: Neal Cardwell Reviewed-by: Kuniyuki Iwashima Link: https://patch.msgid.link/20251013152234.842065-4-edumazet@google.com Signed-off-by: Jakub Kicinski

net: remove RTNL use for /proc/sys/net/core/rps_default_mask

2025-07-08T01:42:12+00:00

Use a dedicated mutex instead. Signed-off-by: Eric Dumazet Reviewed-by: Kuniyuki Iwashima Link: https://patch.msgid.link/20250702061558.1585870-1-edumazet@google.com Signed-off-by: Jakub Kicinski

net: rps: remove kfree_rcu_mightsleep() use

2025-04-08T19:30:55+00:00

Add an rcu_head to sd_flow_limit and rps_sock_flow_table structs to use the more conventional and predictable k[v]free_rcu(). Signed-off-by: Eric Dumazet Reviewed-by: Willem de Bruijn Link: https://patch.msgid.link/20250407163602.170356-5-edumazet@google.com Signed-off-by: Jakub Kicinski

net: rps: change skb_flow_limit() hash function

2025-04-08T19:30:55+00:00

As explained in commit f3483c8e1da6 ("net: rfs: hash function change"), masking low order bits of skb_get_hash(skb) has low entropy. A NIC with 32 RX queues uses the 5 low order bits of rss key to select a queue. This means all packets landing to a given queue share the same 5 low order bits. Switch to hash_32() to reduce hash collisions. Signed-off-by: Eric Dumazet Reviewed-by: Willem de Bruijn Link: https://patch.msgid.link/20250407163602.170356-2-edumazet@google.com Signed-off-by: Jakub Kicinski

net: set the minimum for net_hotdata.netdev_budget_usecs

2025-02-22T00:27:54+00:00

Commit 7acf8a1e8a28 ("Replace 2 jiffies with sysctl netdev_budget_usecs to enable softirq tuning") added a possibility to set net_hotdata.netdev_budget_usecs, but added no lower bound checking. Commit a4837980fd9f ("net: revert default NAPI poll timeout to 2 jiffies") made the *initial* value HZ-dependent, so the initial value is at least 2 jiffies even for lower HZ values (2 ms for 1000 Hz, 8ms for 250 Hz, 20 ms for 100 Hz). But a user still can set improper values by a sysctl. Set .extra1 (the lower bound) for net_hotdata.netdev_budget_usecs to the same value as in the latter commit. That is to 2 jiffies. Fixes: a4837980fd9f ("net: revert default NAPI poll timeout to 2 jiffies") Fixes: 7acf8a1e8a28 ("Replace 2 jiffies with sysctl netdev_budget_usecs to enable softirq tuning") Signed-off-by: Jiri Slaby (SUSE) Cc: Dmitry Yakunin Cc: Konstantin Khlebnikov Link: https://patch.msgid.link/20250220110752.137639-1-jirislaby@kernel.org Signed-off-by: Jakub Kicinski

net: let net.core.dev_weight always be non-zero

2025-01-18T03:00:11+00:00

The following problem was encountered during stability test: (NULL net_device): NAPI poll function process_backlog+0x0/0x530 \ returned 1, exceeding its budget of 0. ------------[ cut here ]------------ list_add double add: new=ffff88905f746f48, prev=ffff88905f746f48, \ next=ffff88905f746e40. WARNING: CPU: 18 PID: 5462 at lib/list_debug.c:35 \ __list_add_valid_or_report+0xf3/0x130 CPU: 18 UID: 0 PID: 5462 Comm: ping Kdump: loaded Not tainted 6.13.0-rc7+ RIP: 0010:__list_add_valid_or_report+0xf3/0x130 Call Trace: ? __warn+0xcd/0x250 ? __list_add_valid_or_report+0xf3/0x130 enqueue_to_backlog+0x923/0x1070 netif_rx_internal+0x92/0x2b0 __netif_rx+0x15/0x170 loopback_xmit+0x2ef/0x450 dev_hard_start_xmit+0x103/0x490 __dev_queue_xmit+0xeac/0x1950 ip_finish_output2+0x6cc/0x1620 ip_output+0x161/0x270 ip_push_pending_frames+0x155/0x1a0 raw_sendmsg+0xe13/0x1550 __sys_sendto+0x3bf/0x4e0 __x64_sys_sendto+0xdc/0x1b0 do_syscall_64+0x5b/0x170 entry_SYSCALL_64_after_hwframe+0x76/0x7e The reproduction command is as follows: sysctl -w net.core.dev_weight=0 ping 127.0.0.1 This is because when the napi's weight is set to 0, process_backlog() may return 0 and clear the NAPI_STATE_SCHED bit of napi->state, causing this napi to be re-polled in net_rx_action() until __do_softirq() times out. Since the NAPI_STATE_SCHED bit has been cleared, napi_schedule_rps() can be retriggered in enqueue_to_backlog(), causing this issue. Making the napi's weight always non-zero solves this problem. Triggering this issue requires system-wide admin (setting is not namespaced). Fixes: e38766054509 ("[NET]: Fix sysctl net.core.dev_weight") Fixes: 3d48b53fb2ae ("net: dev_weight: TX/RX orthogonality") Signed-off-by: Liu Jian Link: https://patch.msgid.link/20250116143053.4146855-1-liujian56@huawei.com Signed-off-by: Jakub Kicinski

net: sysctl: allow dump_cpumask to handle higher numbers of CPUs

2024-10-23T08:28:26+00:00

This fixes the output of rps_default_mask and flow_limit_cpu_bitmap when the CPU count is > 448, as it was truncated. The underlying values are actually stored correctly when writing to these sysctl but displaying them uses a fixed length temporary buffer in dump_cpumask. This buffer can be too small if the CPU count is > 448. Fix this by dynamically allocating the buffer in dump_cpumask, using a guesstimate of what we need. Signed-off-by: Antoine Tenart Reviewed-by: Simon Horman Signed-off-by: Paolo Abeni

net: sysctl: do not reserve an extra char in dump_cpumask temporary buffer

2024-10-23T08:28:19+00:00

When computing the length we'll be able to use out of the buffers, one char is removed from the temporary one to make room for a newline. It should be removed from the output buffer length too, but in reality this is not needed as the later call to scnprintf makes sure a null char is written at the end of the buffer which we override with the newline. Signed-off-by: Antoine Tenart Signed-off-by: Paolo Abeni