linux-kernel - Re: [PATCH net v1 1/2] netlink: let len field used to parse type-not-care nested attrs

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: <38179c76.f308d.189aed2db99.Coremail.linma@zju.edu.cn>
Date:   Tue, 1 Aug 2023 10:00:01 +0800 (GMT+08:00)
From:   "Lin Ma" <linma@....edu.cn>
To:     "Jakub Kicinski" <kuba@...nel.org>
Cc:     davem@...emloft.net, edumazet@...gle.com, pabeni@...hat.com,
        fw@...len.de, yang.lee@...ux.alibaba.com, jgg@...pe.ca,
        markzhang@...dia.com, phaddad@...dia.com, yuancan@...wei.com,
        ohartoov@...dia.com, chenzhongjin@...wei.com, aharonl@...dia.com,
        leon@...nel.org, netdev@...r.kernel.org,
        linux-kernel@...r.kernel.org, linux-rdma@...r.kernel.org
Subject: Re: [PATCH net v1 1/2] netlink: let len field used to parse
 type-not-care nested attrs

Hello Jakub,

> > 
> > However, this is tedious and just like Leon said: add another layer of
> > cabal knowledge. The better solution should leverage the nla_policy and
> > discard nlattr whose length is invalid when doing parsing. That is, we
> > should defined a nested_policy for the X above like
> 
> Hard no. Putting array index into attr type is an advanced case and the
> parsing code has to be able to deal with low level netlink details.

Well, I just known that the type field for those attributes is used as array
index.
Hence, for this advanced case, could we define another NLA type, maybe 
NLA_NESTED_IDXARRAY enum? That may be much clearer against modifying existing
code.

> Higher level API should remove the nla_for_each_nested() completely
> which is rather hard to achieve here.

By investigating the code uses nla_for_each_nested macro. There are basically
two scenarios:

1. manually parse nested attributes whose type is not cared (the advance case
   use type as index here).
2. manually parse nested attributes for *one* specific type. Such code do
   nla_type check.

From the API side, to completely remove nla_for_each_nested and avoid the
manual  parsing. I think we can choose two solutions.

Solution-1: add a parsing helper that receives a function pointer as an
            argument, it will call this pointer after carefully verify the
            type and length of an attribute.

Solution-2: add a parsing helper that traverses this nested twice, the first
            time  to do counting size for allocating heap buffer (or stack
            buffer from the caller if the max count is known). The second
            time to fill this buffer with attribute pointers.

Which one is preferred? Please enlighten me about this and I can try to propose
a fix. (I personally like the solution-2 as it works like the existing parsers
like nla_parse) 

> 
> Nacked-by: Jakub Kicinski <kuba@...nel.org>

Thanks
Lin