[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <182bc0ee-51e3-b8c4-59f7-dcf702865a95@kernel.dk>
Date: Thu, 5 Jan 2023 13:33:11 -0700
From: Jens Axboe <axboe@...nel.dk>
To: Linus Torvalds <torvalds@...ux-foundation.org>
Cc: Pali Rohár <pali@...nel.org>,
Christoph Hellwig <hch@...radead.org>,
Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
Subject: Re: Linux 6.2-rc1
On 1/5/23 1:03?PM, Linus Torvalds wrote:
> On Thu, Jan 5, 2023 at 11:40 AM Jens Axboe <axboe@...nel.dk> wrote:
>>
>> Or even implement it in UDF itself somehow. But yes, ideally we'd punt all
>> of this data gathering to userspace and just leave the trivial init/stop
>> atapi/scsi commands to cdrom/sr.
>
> I wonder how much of that could be done by just having a different
> elevator algorithm for cdrw devices..
>
> Anyway, realistically I suspect the real answer is that "nobody cares
> enough any more". I suspect most people haven't used RW optical media
> in over a decade, and we're talking about an increasingly dwindling
> niche use.
While there's some overlap with IO scheduling, I don't think that'd be a
good layer to solve it at. And since this isn't exactly up-and-coming
technology that we expect to proliferate, that makes me especially
hesitant to invest any time in that particular direction.
I still think that doing something with ublk would be the best approach,
and push the data gathering and fixed sized write bits in userspace.
That would still allow arbitrary filesystem usage for these kinds of
devices.
> Optical media may still make sense for backup, but probably not the
> "filesystem" kind.
I don't think it ever made sense, except from a convenience point of
view. And that's most likely what drove the adoption there. It is way
easier to mount a cdrw read/write and copy files there, even if it's
slower than burning an iso image...
> So nobody is going to be motivated to do any development in this area,
> and the best we can do is probably to just keep it limping along.
Indeed...
> Now, it's a bit sad how pktcdvd is the only user of that 'struct
> block_device_operations' devnode thing, and I liked how that went away
> after the removal of this driver.
>
> And I'm not sure why pktcdvd needs it, everybody else seems to be
> happy with gendisk->disk_name.
Let me look into that, I actually don't know. Would be nice if we could
fix that up and re-instate that particular patch.
> There's probably other cruft in pktcdvd that could be removed without
> removing the whole driver, but I do get the feeling that it's just
> less pain to keep the status quo, and that there isn't really much
> motivation for anybody to do anything else.
I'm reluctant to touch it outside of changes that are driven by core
changes, and of course the motivation to remove it was driven by not
wanting to do that either. Any kind of re-architecting of how it works I
would not advocate for. It supposedly works well enough that none of the
(few) users are reporting issues with it, best to just let it remain
like that imho.
--
Jens Axboe
Powered by blists - more mailing lists