Re: [PATCH] ubifs: Fix potentially out-of-bounds memory access in ubifs_dump_node

Re: [PATCH] ubifs: Fix potentially out-of-bounds memory access in ubifs_dump_node

[Richard Weinberger]

On Mon, Dec 16, 2019 at 4:45 PM Liu Song <fishland@aliyun.com> wrote:
>
> From: Liu Song <liu.song11@zte.com.cn>
>
> In “ubifs_check_node”, when “node_len + offs> c-> leb_size”, then
> goto label of "out_len". Then, in the following "ubifs_dump_node",
> if inode type is "UBIFS_DATA_NODE", in "print_hex_dump", an
> out-of-bounds access may occur due to the wrong "ch->len".
> We encountered this problem in our environment. If "ch-> len" is
> very large, it may even cause the kernel to crash.
>
> There are three reasons to choose to remove "print_hex_dump".
> 1) As mentioned earlier, the exception "ch-> len" may cause an
> exception in "print_hex_dump";
> 2) Data nodes are often large. When printing data content in
> "print_hex_dump", a large amount of output will bring a high load
> on the system, and may even cause a watchdog reset;
> 3) Even if there is a CRC check error, the stuff of file is difficult
> to identify, and difficult to find the problem from a large amount of
> output. We have already output the LEB and offset of the node. So we
> can take the initiative to view the data of interest, instead of
> printing it directly.

If UBIFS dumps data nodes due an error we are already in deep trouble
and having the content of bad data nodes can be helpful.
This feature helped me more than once to debug issues.

-- 
Thanks,
//richard

Re: [PATCH] ubifs: Fix potentially out-of-bounds memory access in ubifs_dump_node

[Richard Weinberger]

----- Ursprüngliche Mail -----
> Problems with storage devices are indeed a problem, But because the wrong
> "ch->len" causes the kernel to crash, this cost is too heavy. We should
> avoid kernel crashes due to such errors.
> 
> 
> Although a crc error was found in "ubifs_check_node", it is difficult to
> simply judge whether "ch->len" is reasonable, so I think we only need to know
> the _location_ of the error data node, and it is not necessary to present its
> contents together.

What we can try is optionally passing the buffer length to ubifs_dump_node().
If crc is bad but ch->len is within bounds we can still safely dump.

Thanks,
//richard