linux-kernel - Re: [PATCH 2/2] mtd: concat: implement _is

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] [day] [month] [year] [list]
Message-ID: <82de06c4122b4b958b5c840aa2166dd8@svr-chch-ex1.atlnz.lc>
Date:   Wed, 22 May 2019 22:16:44 +0000
From:   Chris Packham <Chris.Packham@...iedtelesis.co.nz>
To:     Richard Weinberger <richard.weinberger@...il.com>
CC:     David Woodhouse <dwmw2@...radead.org>,
        Brian Norris <computersforpeace@...il.com>,
        Marek Vasut <marek.vasut@...il.com>,
        "Miquel Raynal" <miquel.raynal@...tlin.com>,
        Richard Weinberger <richard@....at>,
        Vignesh Raghavendra <vigneshr@...com>,
        "linux-mtd@...ts.infradead.org" <linux-mtd@...ts.infradead.org>,
        LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH 2/2] mtd: concat: implement _is_locked mtd operation

On 23/05/19 9:27 AM, Richard Weinberger wrote:
> On Wed, May 22, 2019 at 11:06 PM Chris Packham
> <Chris.Packham@...iedtelesis.co.nz> wrote:
>>
>> On 23/05/19 8:44 AM, Richard Weinberger wrote:
>>> On Wed, May 22, 2019 at 2:08 AM Chris Packham
>>> <chris.packham@...iedtelesis.co.nz> wrote:
>>>>
>>>> Add an implementation of the _is_locked operation for concatenated mtd
>>>> devices. As with concat_lock/concat_unlock this can simply use the
>>>> common helper and pass mtd_is_locked as the operation.
>>>>
>>>> Signed-off-by: Chris Packham <chris.packham@...iedtelesis.co.nz>
>>>> ---
>>>>    drivers/mtd/mtdconcat.c | 6 ++++++
>>>>    1 file changed, 6 insertions(+)
>>>>
>>>> diff --git a/drivers/mtd/mtdconcat.c b/drivers/mtd/mtdconcat.c
>>>> index 9514cd2db63c..0e919f3423af 100644
>>>> --- a/drivers/mtd/mtdconcat.c
>>>> +++ b/drivers/mtd/mtdconcat.c
>>>> @@ -496,6 +496,11 @@ static int concat_unlock(struct mtd_info *mtd, loff_t ofs, uint64_t len)
>>>>           return __concat_xxlock(mtd, ofs, len, mtd_unlock);
>>>>    }
>>>>
>>>> +static int concat_is_locked(struct mtd_info *mtd, loff_t ofs, uint64_t len)
>>>> +{
>>>> +       return __concat_xxlock(mtd, ofs, len, mtd_is_locked);
>>>> +}
>>>
>>> Hmm, here you start abusing your own new API. :(
>>
>> Abusing because xxlock is a poor choice of name? I initially had a third
>> copy of the logic from lock/unlock which is what lead me to do the
>> cleanup first. mtd_lock(), mtd_unlock() and mtd_is_locked() all work the
>> same way namely given an offset and a length either lock, unlock or
>> return the status of the len/erasesz blocks at ofs.
> 
> Well, for unlock/lock it is just a loop which applies an operation to
> a given range on all submtds.
> But as soon an operation returns non-zero, the loop stops and returns
> that error.
> This makes sense for unlock/lock.
> 
> Now you abuse this as "apply a random mtd operation to a given range".
> So, giving it a proper name is the first step. Step two is figuring
> for what kind
> of mtd operations it makes sense and is correct.

Ah now I understand you concern. I guess the question is what is the 
right thing for MEMISLOCKED to return when consecutive blocks differ in 
lock status.

>>>
>>> Did you verify that the unlock/lock-functions deal correctly with all
>>> semantics from mtd_is_locked?
>>> i.e. mtd_is_locked() with len = 0 returns 1 for spi-nor.
>>>
>>
>> I believe so. I've only got access to a parallel NOR flash system that
>> uses concatenation and that seems sane  (is mtdconcat able to work with
>> spi memories?). The concat_is_locked() should just reflect what the
>> underlying mtd device driver returns.
> 
> mtdconcat *should* work with any mtd. But I never used it much, I see
> it more as legacy
> code.
> 
> What happens if one submtd is locked and another not?
> Does concat_is_locked() return something sane then?
> I'd expect it to return true if at least one submtd is locked and 0
> of no submtd is locked.
> 
> If the loop and return code handling in __concat_xxlock() can take care of that,
> awesome. Then all you need is giving it a better name. :-)

As implemented right now the loop will stop at the first locked block. 
So if the range starts in a unlocked block and spans into a locked one 
the return value will be 1.

Is that correct? Well do_ppb_xxlock and  do_getlockstatus_oneblock seem 
to only care about the first block (they both ignore len)? So they'd 
return 0 in the case of unlocked,locked.

stm_is_locked_sr does about the len and will return 0 if len falls 
outside the locked region or if ofs starts before the locked region.

So here's a quick breakdown

                  ppb_is_locked intelext_is_locked stm_is_locked concat
unlocked,unlocked            0                  0             0      0
locked,locked                1                  1             1      1
locked,unlocked              1                  1             0      1
unlocked,locked              0                  0             0      1

I'll try and make concat_is_locked consistent with the two cfi 
implementations.

Thanks for your feedback on this. I think the v2 series should look a 
lot better as a result.