linux-kernel - Re: [PATCH] drivers/ata: print trim features at device initialization

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]

Date:   Mon, 10 Jun 2019 10:49:01 +0300
From:   Konstantin Khlebnikov <khlebnikov@...dex-team.ru>
To:     James Bottomley <James.Bottomley@...senPartnership.com>,
        "Martin K. Petersen" <martin.petersen@...cle.com>,
        Christoph Hellwig <hch@...radead.org>
Cc:     Jens Axboe <axboe@...nel.dk>, linux-ide@...r.kernel.org,
        linux-kernel@...r.kernel.org,
        Dmitry Monakhov <dmtrmonakhov@...dex-team.ru>
Subject: Re: [PATCH] drivers/ata: print trim features at device initialization

On 10.06.2019 0:37, James Bottomley wrote:
> On Sat, 2019-06-08 at 17:13 +0300, Konstantin Khlebnikov wrote:
>>> On 08.06.2019 11:25, Christoph Hellwig wrote:> On Fri, Jun 07, 2019
>>> at 10:34:39AM +0300, Konstantin Khlebnikov wrote:
>>>   >
>>>   > Do we really need to spam dmesg with even more ATA crap?  What
>>> about
>>>   > a sysfs file that can be read on demand instead?
>>>   >
>>>
>>> Makes sense.
>>>
>>> Trim state is exposed for ata_device:
>>> /sys/class/ata_device/devX.Y/trim
>>> but there is no link from scsi device to ata device so they hard to
>>> match.
>>>
>>> I'll think about it.
>>
>> Nope. There is no obvious way to link scsi device with ata_device.
>> ata_device is built on top of "transport_class" and
>> "attribute_container".
>> This some extremely over engineered sysfs framework used only in
>> ata/scsi. I don't want to touch this.
> 
> You don't need to know any of that.  The problem is actually when the
> ata transport classes were first created, the devices weren't properly
> parented.  What should have happened, like every other transport class,
> is that the devices should have descended down to the scsi device as
> the leaf in an integrated fashion.  Instead, what we seem to have is
> three completely separate trees.
> 
> So if you look at a SAS device, you see from the pci device:
> 
> host2/port-2:0/end_device-2:0/target2:0:0/2:0:0:0/block/sdb/sdb1
> 
> But if you look at a SATA device, you see three separate paths:
> 
> ata3/host3/target3\:0\:0/3\:0\:0\:0/block/sda/sda1
> ata3/link3/dev3.0/ata_device/dev3.0
> ata3/ata_port/ata3
> 
> Instead of an integrated tree
> 
> Unfortunately, this whole thing is unfixable now.  If I integrate the
> tree properly, the separate port and link directories will get subsumed
> and we won't be able to recover them with judicious linking so scripts
> relying on them will break.  The best we can probably do is add
> additional links with what we have.
> 
> To follow the way we usually do it, there should be a link from the ata
> device to the scsi target, but that wouldn't help you find the "trim"
> files, so it sounds like you want a link from the scsi device to the ata device, which would?

Yes, I'm talking about link from scsi device to leaf ata_device node.

In libata scsi_device has one to one relation with ata_device.
So making link like /sys/class/block/sda/device/ata_device should be possible easy.
But I haven't found implicit reference from struct ata_device to ata_device in sysfs.

In simplest ahci case whole path looks like:
/sys/devices/pci0000:00/0000:00:1f.2/ata1/link1/dev1.0/ata_device/dev1.0
|______________________|__ata_host__|port|link_|_tdev_|___ata_device___|


/sys/class/ata_device/dev1.0 points directly to leaf ata_device
While in struct ata_device tdev is different intermediate node.

It would be nice merge tdev and ata_device into one node, or at least embed leaf
struct device into struct ata_device too.

As I see ata registers only "transport" device while scsi transport template
magically matches it and creates actual ata device of ata_dev_class.
I see no reason for this complexity. Why ata host couldn't enumerate and
register all these devices explicitly?