I'm proposing to address the most obvious issues with dpt_i2o on stable branches. At this stage it may be better to remove it as has been done upstream, but I'd rather limit the regression for anyone still using the hardware.
The changes are:
- "scsi: dpt_i2o: Remove broken pass-through ioctl (I2OUSERCMD)", which closes security flaws including CVE-2023-2007. - "scsi: dpt_i2o: Do not process completions with invalid addresses", which removes the remaining bus_to_virt() call and may slightly improve handling of misbehaving hardware.
These changes have been compiled on all the relevant stable branches, but I don't have hardware to test on.
Ben.
On Sat, May 27, 2023 at 10:42:00PM +0200, Ben Hutchings wrote:
I'm proposing to address the most obvious issues with dpt_i2o on stable branches. At this stage it may be better to remove it as has been done upstream, but I'd rather limit the regression for anyone still using the hardware.
The changes are:
- "scsi: dpt_i2o: Remove broken pass-through ioctl (I2OUSERCMD)", which closes security flaws including CVE-2023-2007.
- "scsi: dpt_i2o: Do not process completions with invalid addresses", which removes the remaining bus_to_virt() call and may slightly improve handling of misbehaving hardware.
These changes have been compiled on all the relevant stable branches, but I don't have hardware to test on.
Why don't we just delete it in the stable trees as well? If no one has the hardware (otherwise the driver would not have been removed), who is going to hit these issues anyway?
thanks,
greg k-h
On Sun, 28 May 2023, Greg Kroah-Hartman wrote:
On Sat, May 27, 2023 at 10:42:00PM +0200, Ben Hutchings wrote:
I'm proposing to address the most obvious issues with dpt_i2o on stable branches. At this stage it may be better to remove it as has been done upstream, but I'd rather limit the regression for anyone still using the hardware.
The changes are:
- "scsi: dpt_i2o: Remove broken pass-through ioctl (I2OUSERCMD)", which closes security flaws including CVE-2023-2007.
- "scsi: dpt_i2o: Do not process completions with invalid addresses", which removes the remaining bus_to_virt() call and may slightly improve handling of misbehaving hardware.
These changes have been compiled on all the relevant stable branches, but I don't have hardware to test on.
Why don't we just delete it in the stable trees as well? If no one has the hardware (otherwise the driver would not have been removed), who is going to hit these issues anyway?
It's already gone from two stable trees. Would you also have it deleted from users' machines, or would you have each distro separately maintain out-of-tree that code which it is presently shipping, or something else?
On Sun, May 28, 2023 at 07:58:11PM +1000, Finn Thain wrote:
On Sun, 28 May 2023, Greg Kroah-Hartman wrote:
On Sat, May 27, 2023 at 10:42:00PM +0200, Ben Hutchings wrote:
I'm proposing to address the most obvious issues with dpt_i2o on stable branches. At this stage it may be better to remove it as has been done upstream, but I'd rather limit the regression for anyone still using the hardware.
The changes are:
- "scsi: dpt_i2o: Remove broken pass-through ioctl (I2OUSERCMD)", which closes security flaws including CVE-2023-2007.
- "scsi: dpt_i2o: Do not process completions with invalid addresses", which removes the remaining bus_to_virt() call and may slightly improve handling of misbehaving hardware.
These changes have been compiled on all the relevant stable branches, but I don't have hardware to test on.
Why don't we just delete it in the stable trees as well? If no one has the hardware (otherwise the driver would not have been removed), who is going to hit these issues anyway?
It's already gone from two stable trees. Would you also have it deleted from users' machines, or would you have each distro separately maintain out-of-tree that code which it is presently shipping, or something else?
Delete it as obviously no one actually has this hardware. Or just leave it alone, as obviously no one has this hardware so any changes made to the code would not actually affect anyone.
Or am I missing something here?
thanks,
greg k-h
On Sun, 28 May 2023, Greg Kroah-Hartman wrote:
On Sun, May 28, 2023 at 07:58:11PM +1000, Finn Thain wrote:
On Sun, 28 May 2023, Greg Kroah-Hartman wrote:
On Sat, May 27, 2023 at 10:42:00PM +0200, Ben Hutchings wrote:
I'm proposing to address the most obvious issues with dpt_i2o on stable branches. At this stage it may be better to remove it as has been done upstream, but I'd rather limit the regression for anyone still using the hardware.
The changes are:
- "scsi: dpt_i2o: Remove broken pass-through ioctl (I2OUSERCMD)", which closes security flaws including CVE-2023-2007.
- "scsi: dpt_i2o: Do not process completions with invalid addresses", which removes the remaining bus_to_virt() call and may slightly improve handling of misbehaving hardware.
These changes have been compiled on all the relevant stable branches, but I don't have hardware to test on.
Why don't we just delete it in the stable trees as well? If no one has the hardware (otherwise the driver would not have been removed), who is going to hit these issues anyway?
It's already gone from two stable trees. Would you also have it deleted from users' machines, or would you have each distro separately maintain out-of-tree that code which it is presently shipping, or something else?
Delete it as obviously no one actually has this hardware. Or just leave it alone, as obviously no one has this hardware so any changes made to the code would not actually affect anyone.
Or am I missing something here?
Under the assumption that the hardware does not exist, surely there's no value in a distro shipping the driver. No argument from me on that point. But the assumption is questionable and impossible to validate.
As b04e75a4a8a8 was never reverted, I infer that users of v6.0 (and later) do not need the driver. How do you infer that users of distro kernels are not using a given driver?
On Sun, 2023-05-28 at 08:02 +0100, Greg Kroah-Hartman wrote:
On Sat, May 27, 2023 at 10:42:00PM +0200, Ben Hutchings wrote:
I'm proposing to address the most obvious issues with dpt_i2o on stable branches. At this stage it may be better to remove it as has been done upstream, but I'd rather limit the regression for anyone still using the hardware.
The changes are:
- "scsi: dpt_i2o: Remove broken pass-through ioctl (I2OUSERCMD)", which closes security flaws including CVE-2023-2007.
- "scsi: dpt_i2o: Do not process completions with invalid addresses", which removes the remaining bus_to_virt() call and may slightly improve handling of misbehaving hardware.
These changes have been compiled on all the relevant stable branches, but I don't have hardware to test on.
Why don't we just delete it in the stable trees as well? If no one has the hardware (otherwise the driver would not have been removed), who is going to hit these issues anyway?
We don't know that no-one is using the hardware, just because no-one among a small group of kernel developers and early adopters has spoken up yet.
Ben.
On Sun, May 28, 2023 at 02:40:52PM +0200, Ben Hutchings wrote:
On Sun, 2023-05-28 at 08:02 +0100, Greg Kroah-Hartman wrote:
On Sat, May 27, 2023 at 10:42:00PM +0200, Ben Hutchings wrote:
I'm proposing to address the most obvious issues with dpt_i2o on stable branches. At this stage it may be better to remove it as has been done upstream, but I'd rather limit the regression for anyone still using the hardware.
The changes are:
- "scsi: dpt_i2o: Remove broken pass-through ioctl (I2OUSERCMD)", which closes security flaws including CVE-2023-2007.
- "scsi: dpt_i2o: Do not process completions with invalid addresses", which removes the remaining bus_to_virt() call and may slightly improve handling of misbehaving hardware.
These changes have been compiled on all the relevant stable branches, but I don't have hardware to test on.
Why don't we just delete it in the stable trees as well? If no one has the hardware (otherwise the driver would not have been removed), who is going to hit these issues anyway?
We don't know that no-one is using the hardware, just because no-one among a small group of kernel developers and early adopters has spoken up yet.
So what are we supposed to do here. Take patches that even if the driver is added back upstream will not get merged there (as it will not be obvious they are needed)? Or just ignore this?
Why did you work on these changes, were there reports of problems? Or complaints from users? Something else?
thanks,
greg k-h
On Sun, 28 May 2023, Greg Kroah-Hartman wrote:
So what are we supposed to do here. Take patches that even if the driver is added back upstream will not get merged there (as it will not be obvious they are needed)?
As long as there's no maintainer, I don't think you can accept the patch. If a maintainer volunteered themselves, I think b04e75a4a8a8 could be reverted. Perhaps Ben will apply for that position -- I'm unable to do so.
On Sat, May 27, 2023 at 10:42:00PM +0200, Ben Hutchings wrote:
I'm proposing to address the most obvious issues with dpt_i2o on stable branches. At this stage it may be better to remove it as has been done upstream, but I'd rather limit the regression for anyone still using the hardware.
The changes are:
- "scsi: dpt_i2o: Remove broken pass-through ioctl (I2OUSERCMD)", which closes security flaws including CVE-2023-2007.
- "scsi: dpt_i2o: Do not process completions with invalid addresses", which removes the remaining bus_to_virt() call and may slightly improve handling of misbehaving hardware.
These changes have been compiled on all the relevant stable branches, but I don't have hardware to test on.
All now queued up, thanks.
greg k-h
linux-stable-mirror@lists.linaro.org
-
Ben Hutchings -
Finn Thain -
Greg Kroah-Hartman