[PATCH] tcp: another fix of uncloning packets before mangling them

[PATCH] tcp: another fix of uncloning packets before mangling them

[Wen Yang]

The following warning was caught：

[937151.638394] Call Trace:
[937151.638401]  [<ffffffff8163f2f6>] dump_stack+0x19/0x1b
[937151.638405]  [<ffffffff8107dd70>] warn_slowpath_common+0x70/0xb0
[937151.638407]  [<ffffffff8107deba>] warn_slowpath_null+0x1a/0x20
[937151.638410]  [<ffffffff8158bb7b>] tcp_set_skb_tso_segs+0xeb/0x100
[937151.638412]  [<ffffffff8158bbc7>] tcp_init_tso_segs+0x37/0x50
[937151.638414]  [<ffffffff8158d7b9>] tcp_write_xmit+0x1d9/0xce0
[937151.638417]  [<ffffffff8158e53e>] __tcp_push_pending_frames+0x2e/0xc0
[937151.638419]  [<ffffffff8157cf3c>] tcp_push+0xec/0x120
[937151.638421]  [<ffffffff81580728>] tcp_sendmsg+0xc8/0xc20
[937151.638424]  [<ffffffff815aae24>] inet_sendmsg+0x64/0xb0
[937151.638428]  [<ffffffff810b9565>] ? check_preempt_curr+0x75/0xa0
[937151.638434]  [<ffffffff81519917>] sock_aio_write+0x157/0x180
[937151.638437]  [<ffffffff811e267d>] do_sync_write+0x8d/0xd0
[937151.638440]  [<ffffffff811e2f95>] vfs_write+0x1b5/0x1e0
[937151.638442]  [<ffffffff811e393f>] SyS_write+0x7f/0xe0
[937151.638445]  [<ffffffff816513fd>] system_call_fastpath+0x16/0x1b

According commit c52e2421f736 ("tcp: must unclone packets before
mangling them"), TCP stack should make sure it owns skbs before
mangling them.
And there is another place where skb_unclone() is needed. This patch
fix that.

Signed-off-by: Wen Yang <wen.yang99@zte.com.cn>
Tested-by: Liu Bo <liu.bo9@zte.com.cn>
Reviewed-by: Jiang Biao <jiang.biao2@zte.com.cn>
---
 net/ipv4/tcp_output.c | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 597dbd7..fbe8140 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -1793,6 +1793,9 @@ static int tcp_init_tso_segs(struct sk_buff *skb, unsigned int mss_now)
 	int tso_segs = tcp_skb_pcount(skb);
 
 	if (!tso_segs || (tso_segs > 1 && tcp_skb_mss(skb) != mss_now)) {
+		if (skb_unclone(skb, GFP_ATOMIC))
+			return -ENOMEM;
+
 		tcp_set_skb_tso_segs(skb, mss_now);
 		tso_segs = tcp_skb_pcount(skb);
 	}
@@ -2045,6 +2048,7 @@ static int tcp_mtu_probe(struct sock *sk)
 	int copy, len;
 	int mss_now;
 	int interval;
+	int err;
 
 	/* Not currently probing/verifying,
 	 * not in recovery,
@@ -2151,7 +2155,9 @@ static int tcp_mtu_probe(struct sock *sk)
 		if (len >= probe_size)
 			break;
 	}
-	tcp_init_tso_segs(nskb, nskb->len);
+	err = tcp_init_tso_segs(nskb, nskb->len);
+	if (unlikely(err < 0))
+		return err;
 
 	/* We're ready to send.  If this fails, the probe will
 	 * be resegmented into mss-sized pieces by tcp_write_xmit().
@@ -2309,6 +2315,8 @@ static bool tcp_write_xmit(struct sock *sk, unsigned int mss_now, int nonagle,
 
 		tso_segs = tcp_init_tso_segs(skb, mss_now);
 		BUG_ON(!tso_segs);
+		if (unlikely(tso_segs < 0))
+			break;
 
 		if (unlikely(tp->repair) && tp->repair_queue == TCP_SEND_QUEUE) {
 			/* "skb_mstamp" is used as a start point for the retransmit timer */
-- 
1.8.3.1

Re: [PATCH] tcp: another fix of uncloning packets before mangling them

[Eric Dumazet]

On Tue, Aug 28, 2018 at 12:32 AM Wen Yang <wen.yang99@zte.com.cn> wrote:
>
> The following warning was caught：
>
> [937151.638394] Call Trace:
> [937151.638401]  [<ffffffff8163f2f6>] dump_stack+0x19/0x1b
> [937151.638405]  [<ffffffff8107dd70>] warn_slowpath_common+0x70/0xb0
> [937151.638407]  [<ffffffff8107deba>] warn_slowpath_null+0x1a/0x20
> [937151.638410]  [<ffffffff8158bb7b>] tcp_set_skb_tso_segs+0xeb/0x100
> [937151.638412]  [<ffffffff8158bbc7>] tcp_init_tso_segs+0x37/0x50
> [937151.638414]  [<ffffffff8158d7b9>] tcp_write_xmit+0x1d9/0xce0
> [937151.638417]  [<ffffffff8158e53e>] __tcp_push_pending_frames+0x2e/0xc0
> [937151.638419]  [<ffffffff8157cf3c>] tcp_push+0xec/0x120
> [937151.638421]  [<ffffffff81580728>] tcp_sendmsg+0xc8/0xc20
> [937151.638424]  [<ffffffff815aae24>] inet_sendmsg+0x64/0xb0
> [937151.638428]  [<ffffffff810b9565>] ? check_preempt_curr+0x75/0xa0
> [937151.638434]  [<ffffffff81519917>] sock_aio_write+0x157/0x180
> [937151.638437]  [<ffffffff811e267d>] do_sync_write+0x8d/0xd0
> [937151.638440]  [<ffffffff811e2f95>] vfs_write+0x1b5/0x1e0
> [937151.638442]  [<ffffffff811e393f>] SyS_write+0x7f/0xe0
> [937151.638445]  [<ffffffff816513fd>] system_call_fastpath+0x16/0x1b
>
> According commit c52e2421f736 ("tcp: must unclone packets before
> mangling them"), TCP stack should make sure it owns skbs before
> mangling them.
> And there is another place where skb_unclone() is needed. This patch
> fix that.
>
> Signed-off-by: Wen Yang <wen.yang99@zte.com.cn>
> Tested-by: Liu Bo <liu.bo9@zte.com.cn>
> Reviewed-by: Jiang Biao <jiang.biao2@zte.com.cn>
> ---
>  net/ipv4/tcp_output.c | 10 +++++++++-
>  1 file changed, 9 insertions(+), 1 deletion(-)
>
> diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
> index 597dbd7..fbe8140 100644
> --- a/net/ipv4/tcp_output.c
> +++ b/net/ipv4/tcp_output.c
> @@ -1793,6 +1793,9 @@ static int tcp_init_tso_segs(struct sk_buff *skb, unsigned int mss_now)
>         int tso_segs = tcp_skb_pcount(skb);
>

Certainly not needed.

TCP stack owns its packets, as long they were never sent (not yet in
retransmit queue)

You probably are using an old kernel, missing some backport...