On Mon, 2018-07-30 at 07:13 -0700, tj+AEA-kernel.org wrote: +AD4- Also, wouldn't it be better to just expose sysfs+AF8-break/unbreak and +AD4- then do sth like the following from scsi? +AD4- +AD4- kobject+AF8-get()+ADs- +AD4- sysfs+AF8-break+AF8-active+AF8-protection()+ADs- +AD4- do normal sysfs removal+ADs- +AD4- sysfs+AF8-unbreak..()+ADs- +AD4- kobject+AF8-put()+ADs- Hello Tejun, It's not clear to me how the sysfs+AF8-break+AF8-active+AF8-protection() should obtain the struct kernfs+AF8-node pointer to the attribute. Calling that function before device+AF8-remove+AF8-file+AF8-self() causes a double call to kernfs+AF8-break+AF8-active+AF8-protection(), which is wrong. Calling kernfs+AF8-find+AF8-and+AF8-get(kobj-+AD4-sd, attr-+AD4-name) after the attribute has been removed results in a NULL pointer because the attribute that that call tries to look up has already been removed. Should I proceed with the approach proposed in the patches attached to a previous e-mail? Thanks, Bart.
