netdev - Re: [PATCH v1 bpf-next 03/10] bpf: tcp: Get rid of st_bucket

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <495201b0-36b9-4a97-8eb3-aedd57e039a9@linux.dev>
Date: Fri, 23 May 2025 15:07:32 -0700
From: Martin KaFai Lau <martin.lau@...ux.dev>
To: Kuniyuki Iwashima <kuniyu@...zon.com>, jordan@...fe.io
Cc: alexei.starovoitov@...il.com, bpf@...r.kernel.org, daniel@...earbox.net,
 netdev@...r.kernel.org, willemdebruijn.kernel@...il.com
Subject: Re: [PATCH v1 bpf-next 03/10] bpf: tcp: Get rid of st_bucket_done

On 5/22/25 1:42 PM, Kuniyuki Iwashima wrote:
> From: Jordan Rife <jordan@...fe.io>
> Date: Thu, 22 May 2025 11:16:13 -0700
>>>>>   static void bpf_iter_tcp_put_batch(struct bpf_tcp_iter_state *iter)
>>>>>   {
>>>>> -	while (iter->cur_sk < iter->end_sk)
>>>>> -		sock_gen_put(iter->batch[iter->cur_sk++]);
>>>>> +	unsigned int cur_sk = iter->cur_sk;
>>>>> +
>>>>> +	while (cur_sk < iter->end_sk)
>>>>> +		sock_gen_put(iter->batch[cur_sk++]);
>>>>
>>>> Why is this chunk included in this patch ?
>>>
>>> This should be in patch 5 to keep cur_sk for find_cookie
>>
>> Without this, iter->cur_sk is mutated when iteration stops, and we lose
>> our place. When iteration resumes and we call bpf_iter_tcp_batch the
>> iter->cur_sk == iter->end_sk condition will always be true, so we will
>> skip to the next bucket without seeking to the offset.
>>
>> Before, we relied on st_bucket_done to tell us if we had remaining items
>> in the current bucket to process but now need to preserve iter->cur_sk
>> through iterations to make the behavior equivalent to what we had before.
> 
> Thanks for explanation, I was confused by calling tcp_seek_last_pos()
> multiple times, and I think we need to preserve/restore st->offset too
> in patch 2 and need this change.
> 
> diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
> index ac00015d5e7a..0816f20bfdff 100644
> --- a/net/ipv4/tcp_ipv4.c
> +++ b/net/ipv4/tcp_ipv4.c
> @@ -2791,6 +2791,7 @@ static void *tcp_seek_last_pos(struct seq_file *seq)
>   			break;
>   		st->bucket = 0;
>   		st->state = TCP_SEQ_STATE_ESTABLISHED;
> +		offset = 0;

This seems like an existing bug not necessarily related to this set.

The patch 5 has also removed the tcp_seek_last_pos() dependency, so I think it 
can be a standalone fix on its own.


>   		fallthrough;
>   	case TCP_SEQ_STATE_ESTABLISHED:
>   		if (st->bucket > hinfo->ehash_mask)> 
> 
> Let's say we are resuming at an offset (10) in the last lhash bucket
> but a few sockets (3) disappeared, then we go to the ehash part with
> a non-zero offset (3), which will overwrite st->offset (3).
> 
> If the ehash does not fit into the batch size, we need to allocate
> a new batch and retry, but the offset (3) is different from the
> first try (10).