Re: [Linux-kernel-mentees] [PATCH] scsi/megaraid: Prevent kernel-infoleak in kioc_to_mimd()

[Peilin Ye]

On Tue, Jul 28, 2020 at 11:41:37AM +0300, Dan Carpenter wrote:
> On Mon, Jul 27, 2020 at 05:02:35PM -0400, Peilin Ye wrote:
> > hinfo_to_cinfo() does no operation on `cinfo` when `hinfo` is NULL,
> > causing kioc_to_mimd() to copy uninitialized stack memory to userspace.
> > Fix it by initializing `cinfo` with memset().
> 
> But "hinfo" can't be NULL so this patch isn't required.  It's a bit
> hard for Smatch to follow the code.
> 
> We know that "opcode" is 82 so the buffer is allocated by mimd_to_kioc()
> -> mraid_mm_attach_buf().

You are right. mraid_mm_ioctl() returns -ENOMEM and never reaches
kioc_to_mimd() if mraid_mm_attach_buf() failed to get a buffer, so
`hinfo` can never be NULL for kioc_to_mimd().

Next time I will trace the data flow more carefully. Thank you for
pointing this out!

Peilin Ye

> Generally, don't silence static checker warnings unless it makes the
> code more readable.  It's the checker writer's job to fix their own code.
> In this case, that's me, but parsing the code is quite complicated and I
> don't have a plan for how to fix it.
> 
> regards,
> dan carpenter
>