Re: [PATCH] UBI: dereference after kfree in create_vtbl

[Artem Bityutskiy <dedekind@infradead.org>]

Hi,

thanks for finding bugs in this patch. Although this path will likely
never happen, this is good to have it bug-free.

On Sat, 2007-05-05 at 09:25 +0530, Satyam Sharma wrote:
> Artem would have to step in here to verify if there really is a good
> reason why we kmalloc a fresh ubi_scan_leb every time we want to add
> one to a list. 
Particularly in vtbl.c there is no good reason. Leftover of itsy-bitsy
units. I'll make ubi_scan_add_to_list static, as well as
ubi_scan_add_used(). And I'll rename them to something shorter. They are
only useful in scan.c.

And it is fine to use list_add_tail() directly in vtbl.c. Will be fixed.

> If possible, the best solution would be to change
> ubi_scan_add_to_list() to take in a valid struct ubi_scan_leb and just
> add that to the specified list (using list_add_tail or whatever) --
> and leave allocation up to callers, 
In scan.c it is useful because _all_ callers have to allocate it. vtbl.c
is the only place which does not need it. I'll fix this.

> >though this likely requires a
> major cleanup of this driver w.r.t. ubi_scan_leb lifetime semantics.
What is wrong with the semantics, please be more specific.

I'll fix this shortly.

-- 
Best regards,
Artem Bityutskiy (Битюцкий Артём)