| Message ID | 20250212141021.1663666-1-edumazet@google.com (mailing list archive) |
|---|---|
| State | Accepted |
| Commit | a527750d877fd334de87eef81f1cb5f0f0ca3373 |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | [net] ipv6: mcast: add RCU protection to mld_newpack() | expand |
On 2/12/25 7:10 AM, Eric Dumazet wrote: > mld_newpack() can be called without RTNL or RCU being held. > > Note that we no longer can use sock_alloc_send_skb() because > ipv6.igmp_sk uses GFP_KERNEL allocations which can sleep. > > Instead use alloc_skb() and charge the net->ipv6.igmp_sk > socket under RCU protection. > > Fixes: b8ad0cbc58f7 ("[NETNS][IPV6] mcast - handle several network namespace") > Signed-off-by: Eric Dumazet <edumazet@google.com> > --- > net/ipv6/mcast.c | 14 ++++++++++---- > 1 file changed, 10 insertions(+), 4 deletions(-) > Reviewed-by: David Ahern <dsahern@kernel.org>
Hello: This patch was applied to netdev/net.git (main) by Jakub Kicinski <kuba@kernel.org>: On Wed, 12 Feb 2025 14:10:21 +0000 you wrote: > mld_newpack() can be called without RTNL or RCU being held. > > Note that we no longer can use sock_alloc_send_skb() because > ipv6.igmp_sk uses GFP_KERNEL allocations which can sleep. > > Instead use alloc_skb() and charge the net->ipv6.igmp_sk > socket under RCU protection. > > [...] Here is the summary with links: - [net] ipv6: mcast: add RCU protection to mld_newpack() https://git.kernel.org/netdev/net/c/a527750d877f You are awesome, thank you!
diff --git a/net/ipv6/mcast.c b/net/ipv6/mcast.c index 81a739ebf7094694a6f5de5020cd4c4d1c9642d1..65831b4fee1fda1be110cf3f7d5abebaf076694d 100644 --- a/net/ipv6/mcast.c +++ b/net/ipv6/mcast.c @@ -1773,21 +1773,19 @@ static struct sk_buff *mld_newpack(struct inet6_dev *idev, unsigned int mtu) struct net_device *dev = idev->dev; int hlen = LL_RESERVED_SPACE(dev); int tlen = dev->needed_tailroom; - struct net *net = dev_net(dev); const struct in6_addr *saddr; struct in6_addr addr_buf; struct mld2_report *pmr; struct sk_buff *skb; unsigned int size; struct sock *sk; - int err; + struct net *net; - sk = net->ipv6.igmp_sk; /* we assume size > sizeof(ra) here * Also try to not allocate high-order pages for big MTU */ size = min_t(int, mtu, PAGE_SIZE / 2) + hlen + tlen; - skb = sock_alloc_send_skb(sk, size, 1, &err); + skb = alloc_skb(size, GFP_KERNEL); if (!skb) return NULL; @@ -1795,6 +1793,12 @@ static struct sk_buff *mld_newpack(struct inet6_dev *idev, unsigned int mtu) skb_reserve(skb, hlen); skb_tailroom_reserve(skb, mtu, tlen); + rcu_read_lock(); + + net = dev_net_rcu(dev); + sk = net->ipv6.igmp_sk; + skb_set_owner_w(skb, sk); + if (ipv6_get_lladdr(dev, &addr_buf, IFA_F_TENTATIVE)) { /* <draft-ietf-magma-mld-source-05.txt>: * use unspecified address as the source address @@ -1806,6 +1810,8 @@ static struct sk_buff *mld_newpack(struct inet6_dev *idev, unsigned int mtu) ip6_mc_hdr(sk, skb, dev, saddr, &mld2_all_mcr, NEXTHDR_HOP, 0); + rcu_read_unlock(); + skb_put_data(skb, ra, sizeof(ra)); skb_set_transport_header(skb, skb_tail_pointer(skb) - skb->data);
mld_newpack() can be called without RTNL or RCU being held. Note that we no longer can use sock_alloc_send_skb() because ipv6.igmp_sk uses GFP_KERNEL allocations which can sleep. Instead use alloc_skb() and charge the net->ipv6.igmp_sk socket under RCU protection. Fixes: b8ad0cbc58f7 ("[NETNS][IPV6] mcast - handle several network namespace") Signed-off-by: Eric Dumazet <edumazet@google.com> --- net/ipv6/mcast.c | 14 ++++++++++---- 1 file changed, 10 insertions(+), 4 deletions(-)