On 2022/06/06 22:16, Bart Van Assche wrote: > On 6/5/22 19:48, Damien Le Moal wrote: >> On 6/5/22 12:55, Bart Van Assche wrote: >>> On 6/1/22 10:05, Avri Altman wrote: >>>> As a design rule, sysfs attribute files should not be used to make >>>> persistent modifications to a device configuration. This rule applies >>>> to all subsystems and ufs is no different. >>> >>> Hmm ... where does that rule come from? I can't find it in >>> Documentation/admin-guide/sysfs-rules.rst. Did I perhaps overlook something? >> >> I am not aware of any writable sysfs attribute file that can be used to >> make persistent device configuration changes, at least in storage area. >> I know of plenty that do change a device setting, but without saving this >> setting to maintain it across power cycles. Do you know of any such >> attribute ? I was under the impression that sysfs should not be used to >> persistently reconfigure a device... > > I don't think the above is sufficient as an argument to reject a new > patch that introduces a sysfs attribute that changes the device > configuration. It depends if we can guarantee that the write access to the sysfs file is done with the same security checks as for a passthrough command issued from user space. I have not checked. I would also argue that this particular feature is related to the boot device management, which is not something we do in the kernel. There is no sysfs interface to set the bootable flag of a partition on a disk, right ? That is very similar to me. The kernel should not bother about that kind of interface. User application tools can deal with that. > > Thanks, > > Bart. > -- Damien Le Moal Western Digital Research
