net: avoid strange behavior with skb_defer_max == 1
Subsystems (2)
Jakub Kicinski
maintainerPATCH 1/1 net: avoid strange behavior with skb_defer_max == 1
When user sets skb_defer_max to 1 the kick threshold is 0 (half of 1). If we increment queue length before the check the kick will never happen, and the skb may get stranded. This is likely harmless but can be avoided by moving the increment after the check. This way skb_defer_max == 1 will always kick. Still a silly config to have, but somehow that feels more correct.
While at it drop a comment which seems to be outdated or confusing, and wrap the defer_count write with a WRITE_ONCE() since it's read on the fast path that avoids taking the lock.
--- net/core/skbuff.c | 13 +++++-------- 1 file changed, 5 insertions(+), 8 deletions(-)
net/core/skbuff.c +5 -8
@@ -6512,18 +6512,15 @@ nodefer: __kfree_skb(skb);if (READ_ONCE (sd->defer_count) >= defer_max) goto nodefer; -/* We do not send an IPI or any signal. -* Remote cpu will eventually call skb_defer_free_flush() -*/ spin_lock_irqsave (&sd->defer_lock, flags); +/* Send an IPI every time queue reaches half capacity. */ + kick = sd->defer_count == (defer_max >>1 ); +/* Paired with the READ_ONCE() few lines above */ +WRITE_ONCE (sd->defer_count, sd->defer_count +1 ); + skb->next = sd->defer_list;/* Paired with READ_ONCE() in skb_defer_free_flush() */ WRITE_ONCE (sd->defer_list, skb); - sd->defer_count++; - -/* Send an IPI every time queue reaches half capacity. */ - kick = sd->defer_count == (defer_max >>1 ); -spin_unlock_irqrestore (&sd->defer_lock, flags);/* Make sure to trigger NET_RX_SOFTIRQ on the remote CPU
Signature · 1 line
2.34.3
Quoted 9 lines · Jakub Kicinski
When user sets
skb_defer_maxto 1 the kick threshold is 0 (half of 1). If we increment queue length before the check the kick will never happen, and the skb may get stranded. This is likely harmless but can be avoided by moving the increment after the check. This wayskb_defer_max== 1 will always kick. Still a silly config to have, but somehow that feels more correct.
Quoted 7 lines
While at it drop a comment which seems to be outdated or confusing, and wrap the
defer_countwrite with aWRITE_ONCE()since it's read on the fast path that avoids taking the lock.Signed-off-by Jakub Kicinski <[email protected]>---
SGTM thanks.
This patch was applied to netdev/net-next.git (master) by Jakub Kicinski <[email protected]>:
Hello:
This patch was applied to netdev/net-next.git (master) by Jakub Kicinski <[email protected]>:
Quoted 9 lines · On Wed, 18 May 2022 11:55:22 -0700 you wrote:
When user sets
skb_defer_maxto 1 the kick threshold is 0 (half of 1). If we increment queue length before the check the kick will never happen, and the skb may get stranded. This is likely harmless but can be avoided by moving the increment after the check. This wayskb_defer_max== 1 will always kick. Still a silly config to have, but somehow that feels more correct.[...]
Here is the summary with links:
- [net-next] net: avoid strange behavior with skb_defer_max == 1
https://git.kernel.org/netdev/net-next/c/c09b0cd2cc6c
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html