[Devel] [PATCH] ms/tcp: fix potential huge kmalloc() calls in TCP_REPAIR

[Konstantin Khorenko]

On 04/09/2019 01:23 PM, Andrey Ryabinin wrote:
>
>
> On 4/9/19 1:19 PM, Andrey Ryabinin wrote:
>>
>>
>> On 4/9/19 12:34 PM, Konstantin Khorenko wrote:
>>> From: Eric Dumazet <edumazet at google.com>
>>>
>>> tcp_send_rcvq() is used for re-injecting data into tcp receive queue.
>>>
>>> Problems :
>>>
>>> - No check against size is performed, allowed user to fool kernel in
>>>   attempting very large memory allocations, eventually triggering
>>>   OOM when memory is fragmented.
>>>
>>> - In case of fault during the copy we do not return correct errno.
>>>
>>> Lets use alloc_skb_with_frags() to cook optimal skbs.
>>>
>>> Fixes: 292e8d8c8538 ("tcp: Move rcvq sending to tcp_input.c")
>>> Fixes: c0e88ff0f256 ("tcp: Repair socket queues")
>>> Signed-off-by: Eric Dumazet <edumazet at google.com>
>>> Cc: Pavel Emelyanov <xemul at parallels.com>
>>> Acked-by: Pavel Emelyanov <xemul at parallels.com>
>>> Signed-off-by: David S. Miller <davem at davemloft.net>
>>>
>>> CRIU often triggers 8 order page allocation while restoring TCP sockets
>>> without this patch.
>>> https://jira.sw.ru/browse/PSBM-93672
>>>
>>> (cherry picked from commit 5d4c9bfbabdb1d497f21afd81501e5c54b0c85d9)
>>> Signed-off-by: Konstantin Khorenko <khorenko at virtuozzo.com>
>>> ---
>>>  net/ipv4/tcp_input.c | 22 +++++++++++++++++++---
>>>  1 file changed, 19 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
>>> index 015c6fd7ec83..d0a6f767189d 100644
>>> --- a/net/ipv4/tcp_input.c
>>> +++ b/net/ipv4/tcp_input.c
>>> @@ -4470,19 +4470,34 @@ static int __must_check tcp_queue_rcv(struct sock *sk, struct sk_buff *skb, int
>>>  int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size)
>>>  {
>>>  	struct sk_buff *skb;
>>> +	int err = -ENOMEM;
>>> +	int data_len = 0;
>>>  	bool fragstolen;
>>>
>>>  	if (size == 0)
>>>  		return 0;
>>>
>>> -	skb = alloc_skb(size, sk->sk_allocation);
>>> +	if (size > PAGE_SIZE) {
>>> +		int npages = min_t(size_t, size >> PAGE_SHIFT, MAX_SKB_FRAGS);
>>> +
>>> +		data_len = npages << PAGE_SHIFT;
>>> +		size = data_len + (size & ~PAGE_MASK);
>>> +	}
>>> +	skb = alloc_skb_with_frags(size - data_len, data_len,
>>> +				   PAGE_ALLOC_COSTLY_ORDER,
>>> +				   &err, sk->sk_allocation);
>>>  	if (!skb)
>>>  		goto err;
>>>
>>> +	skb_put(skb, size - data_len);
>>> +	skb->data_len = data_len;
>>> +	skb->len = size;
>>> +
>>>  	if (tcp_try_rmem_schedule(sk, skb, skb->truesize))
>>>  		goto err_free;
>>>
>>> -	if (memcpy_from_msg(skb_put(skb, size), msg, size))
>>> +	err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
>>
>>              skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size) ?
>
>          Correction: skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov, 0, size)

Andrey, thank you!
Haste would make waste...

>>
>>
>>> +	if (err)
>>>  		goto err_free;
>>>
>>>  	TCP_SKB_CB(skb)->seq = tcp_sk(sk)->rcv_nxt;
>>> @@ -4498,7 +4513,8 @@ int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size)
>>>  err_free:
>>>  	kfree_skb(skb);
>>>  err:
>>> -	return -ENOMEM;
>>> +	return err;
>>> +
>>>  }
>>>
>>>  static void tcp_data_queue(struct sock *sk, struct sk_buff *skb)
>>>
> .
>