Re: [RFC PATCH 0/2 v1] Ioctl for reading block queue information

[Greg KH <gregkh@suse.de>]

On Fri, Dec 10, 2010 at 03:07:20PM +0100, Lukas Czerner wrote:
> On Thu, 9 Dec 2010, Andreas Dilger wrote:
> 
> > On 2010-12-09, at 12:20, Greg KH wrote:
> > > On Thu, Dec 09, 2010 at 04:25:35PM +0100, Lukas Czerner wrote:
> > >> For a long time it has been pretty painful to retrieve informations from
> > >> /sys/block/*/queue for particular block device. Not only it is painful
> > >> to retrieve informations within C tool, parsing strings, etc, but one
> > >> have to run into problem even finding the proper path in sysfs.
> > > 
> > > What's wrong with using libudev?  That should give you all of this
> > > information easily using a .c program without any need to change the
> > > kernel at all.
> 
> What's wrong with using libudev ? Well, fist of all I have never heard
> about it:), one can argue this is kind of my fault, and second of all
> the documentation is kind of non-existent (almost).
> 
> But, despite this I did gave libudev a quick try and I must say, it
> works, however it is not as simple as calling "ioctl(fd,
> BLKGETQUEUEINFO, &val)" as Andreas pointed out.
> 
> So, in my use-case, I have a path to the device provided by the user
> (strictly speaking it may not be device but for example symbolic link
> /dev/mapper/something) and I need to retrieve queue information like
> discard_granularity, discard_alignment etc... usually stored in place
> like /sys/block/sda/queue/*.
> 
> With libudev I need to:
> 
> 1. create the udev obejct:
> 
> 	udev = udev_new();
> 	if (!udev) {
> 		printf("Can't create udev\n");
> 		exit(1);
> 	}
> 
> 2. Check the path for the block device
> 
> 	stat(name, &buf);
> 	if (!S_ISBLK(buf.st_mode)) {
> 		printf("Not a block device\n");
> 		exit(1);
> 	}
> 
> 3. Get udev device object
> 
> 	dev = udev_device_new_from_devnum(udev, 'b', buf.st_rdev);
> 	if (!dev) {
> 		printf("Can not find the device\n");
> 		exit(1);
> 	}
> 
> 4. Construct path for sysfs attribute I need:
> 
> 	snprintf(path, PATH_MAX, "%s/queue/%s",
> 	                udev_device_get_syspath(dev),
> 			"discard_granularity");

Hm, what about just using the libudev functions for attributes instead?
That should save you this step, and the next one.

> 5. Open the sysfs file, get page-sized buffer and parse text :-/ (without
>    checks now):
> 
> 	read(fd, buffer, pagesize);
> 	sscanf(buffer, "%lu", &value);
> 	printf("max_hw_sector_size: %lu\n",value);
> 
> Which is opposed to BLKGETQUEUEINFO steps (define val, invoke ioctl,
> check result) a bit longer. But I can definitely see you point, it is
> feasible and since we have libudev we might want to use this in
> userspace. The fact is I would really want to stand up and defend my
> ioctl approach, but libudev just might provide what I need without
> proceeding the just-another-ioctl-madness on kernel lists :).

Please use libudev.  What happens next week when we add a new sysfs
attribute to block devices?  Then your ioctl just broke and you have to
create a new one.

No, use sysfs for what it was made for, from userspace, don't add custom
ioctls for this.

thanks,

greg k-h