Re: [RFC PATCH 1/3] ACPI: scan: add userland notification while handling eject events

[Chester Lin <clin@suse.com>]

On Mon, Mar 30, 2020 at 04:50:20PM +0800, Chester Lin wrote:
> Hi Greg,
> 
> On Fri, Mar 27, 2020 at 12:38:42PM +0100, Greg KH wrote:
> > On Fri, Mar 27, 2020 at 07:22:45PM +0800, Chester Lin wrote:
> > > Add a request_offline attribute in order to tell the kernel if it's
> > > required to send notifications to userland first while handling an eject
> > > event. Userland will have to put the target device offline when this
> > > attribute is set.
> > > 
> > > Signed-off-by: Chester Lin <clin@suse.com>
> > > ---
> > >  Documentation/ABI/testing/sysfs-bus-acpi | 16 ++++++++++
> > >  drivers/acpi/device_sysfs.c              | 40 +++++++++++++++++++++++-
> > >  drivers/acpi/scan.c                      | 39 +++++++++++++++++++----
> > >  include/acpi/acpi_bus.h                  |  1 +
> > >  4 files changed, 89 insertions(+), 7 deletions(-)
> > > 
> > > diff --git a/Documentation/ABI/testing/sysfs-bus-acpi b/Documentation/ABI/testing/sysfs-bus-acpi
> > > index e7898cfe5fb1..b9c467704889 100644
> > > --- a/Documentation/ABI/testing/sysfs-bus-acpi
> > > +++ b/Documentation/ABI/testing/sysfs-bus-acpi
> > > @@ -93,3 +93,19 @@ Description:
> > >  		hardware, if the _HRV control method is present.  It is mostly
> > >  		useful for non-PCI devices because lspci can list the hardware
> > >  		version for PCI devices.
> > > +
> > > +What:		/sys/bus/acpi/devices/.../request_offline
> > > +Date:		Mar, 2020
> > > +Contact:	Chester Lin <clin@suse.com>
> > > +Description:
> > > +		(RW) Allows the userland to receive offline requests when
> > > +		devices are planning to be ejected.
> > > +
> > > +		If bit [0] is clear, the kernel will automatically try putting
> > > +		the target offline before the target can be ejected.
> > > +
> > > +		If bit [0] is set, a uevent will be sent to userland as an
> > > +		offline request and userland is responsible for handling offline
> > > +		operations before the target can be ejected. This approach
> > > +		provides flexibility while some applications could need more
> > > +		time to release resources.
> > 
> > Don't use "bit", use 1/0/y/n/Y/N as the kernel will parse all of that
> > for you with the kstrtobool() which was created just for this type of
> > sysfs file.
> > 
> 
> I'm sorry for this mistake. Based on my code they should be ASCII char '1' and
> '0' but not bitwise ops. I will fix this description.
> 
> > > diff --git a/drivers/acpi/device_sysfs.c b/drivers/acpi/device_sysfs.c
> > > index 96869f1538b9..453bd1b9edf5 100644
> > > --- a/drivers/acpi/device_sysfs.c
> > > +++ b/drivers/acpi/device_sysfs.c
> > > @@ -506,6 +506,37 @@ static ssize_t status_show(struct device *dev, struct device_attribute *attr,
> > >  }
> > >  static DEVICE_ATTR_RO(status);
> > >  
> > > +static ssize_t request_offline_show(struct device *dev,
> > > +		struct device_attribute *attr, char *buf)
> > > +{
> > > +	struct acpi_device *acpi_dev = to_acpi_device(dev);
> > > +
> > > +	return sprintf(buf, "%u\n", acpi_dev->request_offline?1:0);
> > > +}
> > > +
> > > +static ssize_t request_offline_store(struct device *dev,
> > > +		struct device_attribute *attr, const char *buf, size_t count)
> > > +{
> > > +	struct acpi_device *acpi_dev = to_acpi_device(dev);
> > > +
> > > +	if (!count)
> > > +		return -EINVAL;
> > > +
> > > +	switch (buf[0]) {
> > > +	case '0':
> > > +		acpi_dev->request_offline = false;
> > > +		break;
> > > +	case '1':
> > > +		acpi_dev->request_offline = true;
> > > +		break;
> > > +	default:
> > > +		return -EINVAL;
> > > +	}
> > > +
> > > +	return count;
> > > +}
> > > +static DEVICE_ATTR_RW(request_offline);
> > > +
> > >  /**
> > >   * acpi_device_setup_files - Create sysfs attributes of an ACPI device.
> > >   * @dev: ACPI device object.
> > > @@ -580,6 +611,11 @@ int acpi_device_setup_files(struct acpi_device *dev)
> > >  		result = device_create_file(&dev->dev, &dev_attr_eject);
> > >  		if (result)
> > >  			return result;
> > > +
> > > +		result = device_create_file(&dev->dev,
> > > +					    &dev_attr_request_offline);
> > > +		if (result)
> > > +			return result;
> > >  	}
> > >  
> > >  	if (dev->flags.power_manageable) {
> > > @@ -623,8 +659,10 @@ void acpi_device_remove_files(struct acpi_device *dev)
> > >  	/*
> > >  	 * If device has _EJ0, remove 'eject' file.
> > >  	 */
> > > -	if (acpi_has_method(dev->handle, "_EJ0"))
> > > +	if (acpi_has_method(dev->handle, "_EJ0")) {
> > >  		device_remove_file(&dev->dev, &dev_attr_eject);
> > > +		device_remove_file(&dev->dev, &dev_attr_request_offline);
> > 
> > You all really should be using an attribute group and the is_visible()
> > callback to handle all of this for you automatically.
> > 
> > But that's a separate issue than this specific patch.
> > 
> 
> That sounds good to me. I will refine this part by declaring an attribute_group
> with a is_visible() callback.
> 
> > > +	}
> > >  
> > >  	if (acpi_has_method(dev->handle, "_SUN"))
> > >  		device_remove_file(&dev->dev, &dev_attr_sun);
> > > diff --git a/drivers/acpi/scan.c b/drivers/acpi/scan.c
> > > index 6d3448895382..1cb39c5360cf 100644
> > > --- a/drivers/acpi/scan.c
> > > +++ b/drivers/acpi/scan.c
> > > @@ -145,6 +145,7 @@ static acpi_status acpi_bus_offline(acpi_handle handle, u32 lvl, void *data,
> > >  	struct acpi_device_physical_node *pn;
> > >  	bool second_pass = (bool)data;
> > >  	acpi_status status = AE_OK;
> > > +	char *envp[] = { "EVENT=offline", NULL };
> > >  
> > >  	if (acpi_bus_get_device(handle, &device))
> > >  		return AE_OK;
> > > @@ -166,7 +167,18 @@ static acpi_status acpi_bus_offline(acpi_handle handle, u32 lvl, void *data,
> > >  		} else {
> > >  			pn->put_online = false;
> > >  		}
> > > -		ret = device_offline(pn->dev);
> > > +
> > > +		/* Don't offline directly but need to notify userland first */
> > > +		if (device->request_offline) {
> > > +			if (pn->dev->offline)
> > > +				ret = 0;
> > > +			else
> > > +				ret = kobject_uevent_env(&pn->dev->kobj,
> > > +							KOBJ_CHANGE, envp);
> > 
> > So this is a userspace visable change with regards to kobject events?
> > 
> > Are you sure that is ok?
> > 
> 
> Since udev can see kobject events when devices are added, I haven't seen any
> risk if we make offline events visible too. Besides, normally online/eject
> attributes can only be written by root in userspace.
> 
Correct my explanation here: So far udev can see several device events already,
such as add, online, offline and remove. So I think it should not be risky if
we send additional change events to userspace as notification.

> Thanks,
> Chester Lin