Re: Linux 6.2-rc1

["Pali Rohár" <pali@kernel.org>]

On Wednesday 04 January 2023 21:56:40 Pali Rohár wrote:
> On Wednesday 04 January 2023 11:25:41 Linus Torvalds wrote:
> > On Wed, Jan 4, 2023 at 11:01 AM Pali Rohár <pali@kernel.org> wrote:
> > >
> > > Driver is still used and userspace tools for it are part of the udftools
> > > project, which is still under active maintenance. More people already
> > > informed me about this "surprise".
> > 
> > Why is that driver used?
> > 
> > It's *literally* pointless. It's just a shell that forwards ioctl's to
> > the real drivers.
> > 
> > > Any comments on this? Because until now nobody answered why this
> > > actively used driver was removed from kernel without informing anybody:
> > 
> > Well, it's been marked as deprecated for five years, so any kernel
> > config should have gotten this notice for the help entry
> > 
> >           Note: This driver is deprecated and will be removed from the
> >           kernel in the near future!
> > 
> > but I guess people didn't notice.
> > 
> > It could be re-instated, but it really is a completely useless driver.
> > Just use the *regular* device nodes, not the pointless pktcd ones.
> > 
> > Is there any real reason why udftools can't just use the normal device node?
> > 
> > The historical reason for this driver being pointless goes back *much*
> > longer than five years - it used to be that the pktcd driver was
> > special, and was the only thing that did raw commands.
> > 
> > But the regular block layer was taught to do that back around 2004, so
> > the "pktcd" driver has literally just been a pointless shell for
> > almost two decades.
> > 
> > And I know it was in 2004, because I actually did most of that "make
> > SCSI commands generic" work myself (but had to go back to the old BK
> > archives to find the exact date - it's been two decades, after all).
> > 
> > I did it because I was fed up with the crazy pktcd driver requiring
> > extra work, when I just wanted to write CD's on my regular IDE CD-ROM
> > the obvious way.
> > 
> > So if there is some reason to actually use the pktcd driver, please
> > tell us what that is.
> > 
> >               Linus
> 
> Last time I did big retest of optical media was two years ago. At that
> time kernel was not able to mount CD-RW disc in full read-write mode
> from the normal node /dev/cdrom. Via pktcdvd driver mapping it was
> possible without any issue. Was there any change in last 5 (or more)
> years in this CD-RW area? Mounting CD-RW media in read-only mode via
> normal /dev/cdrom node always worked fine. Also "burning" CD-R media
> with userspace burning tools on normal /dev/cdrom node also worked.
> But here it is CD-RW media in read-write mode with kernel udf filesystem
> driver without any userspace involved (after proper formatting).

In commit where was pktcdvd dropped is written:
https://git.kernel.org/torvalds/c/f40eb99897af665f11858dd7b56edcb62c3f3c67

  * At the lowest level, there is the standard driver for the CD/DVD device,
  * such as drivers/scsi/sr.c. This driver can handle read and write requests,
  * but it doesn't know anything about the special restrictions that apply to
  * packet writing. One restriction is that write requests must be aligned to
  * packet boundaries on the physical media, and the size of a write request
  * must be equal to the packet size. Another restriction is that a
  * GPCMD_FLUSH_CACHE command has to be issued to the drive before a read
  * command, if the previous command was a write.
  *
  * The purpose of the packet writing driver is to hide these restrictions from
  * higher layers, such as file systems, and present a block device that can be
  * randomly read and written using 2kB-sized blocks.

Were all these write restrictions implemented in sr.c driver? Do you
remember other details?

Because CD-RW support into kernel was really introduced in 2004 in
this historical commit, but it was not for SCSI sr.c driver:
https://git.kernel.org/history/history/c/2f8e2dc86c9876edca632e8ef2ab1f68d1b753f0