Re: [PATCH] usb: gadget: f_mass_storage: break IO operations via configfs

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



В Sat, 9 Apr 2022 16:22:29 -0400
Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote:

> On Sat, Apr 09, 2022 at 05:08:37PM +0300, Maxim Devaev wrote:
> > В Sat, 9 Apr 2022 09:46:32 -0400
> > Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote:
> >   
> > > On Sat, Apr 09, 2022 at 11:57:56AM +0300, Maxim Devaev wrote:  
> > > > В Fri, 8 Apr 2022 10:59:45 -0400
> > > > Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote:    
> > > > > > At least there is one situation where the behavior of f_mass_storage differs
> > > > > > from the behavior of a real drive. What happens when you click on the physical
> > > > > > "eject" button?      
> > > > > 
> > > > > If the host has prevented ejection, nothing happens.  Otherwise the disc 
> > > > > gets ejected.
> > > > >     
> > > > > > Yes, the OS can block this, but the problem is that we don't have
> > > > > > an "eject" here.      
> > > > > 
> > > > > What do you mean?  Writing an empty string to the sysfs "file" attribute 
> > > > > is the virtual analog of pressing the eject button.    
> > > > 
> > > > But I can't eject the disc event it's not mounted on Linux host. It seems to me
> > > > it differs from the real drive behavior.    
> > > 
> > > It sounds like either there's a bug or else you're not doing the right 
> > > thing.  Tell me exactly what you do when this fails.  
> > 
> > I'm using Raspberry Pi with DWC2. So:
> > - Connect RPi-based gadget to the Linux host.
> > - Set image in the "file" attribute.  
> 
> Exactly what is the full pathname you're using for the "file" attribute?

/sys/kernel/config/usb_gadget/kvmd/functions/mass_storage.usb0/lun.0/file

> > - Mount gadget's drive on the Linux host.
> > - Umount it.
> > - Try to eject using emptying the "file" attribute.
> > - Get EBUSY error.  
> 
> This must mean that some program on the host is keeping the device file 
> open, even though it isn't mounted.  (I tried running a similar test on 
> my system and it worked perfectly, with no other programs accessing the 
> device).  You might be able to identify which program is accessing the 
> device by running lsof on the host and searching the output for the 
> device name.

I've been thinking about it too. I tried lsof but it didn't display anything
related with sr0 device. But after execution the "eject" command on the host,
I was able to emptify the "file" attribute of the gadget.

> I also tried sending a USR1 signal to the driver's kernel thread while 
> an image was mounted and being accessed.  It did clear the prevent_allow 
> flag, so I could eject the image.  But it also caused a 30-second delay 
> on the host, as predicted.  Now, maybe you don't care about such delays 
> when you're going to eject the media anyway, but it still seems like a 
> bad thing to do.

It looks like the prevent_medium_removal flag switching really works better in this case.
> 
> > > > I have reflected on the rest of your arguments and changed my mind.
> > > > I think that "forced_eject" for a specific lun without interrupting operations would
> > > > really be the best solution. I wrote a simple patch and tested it, everything seems
> > > > to work. What do you think about something like this?
> > > > 
> > > > 
> > > > static ssize_t fsg_lun_opts_forced_eject_store(struct config_item *item,
> > > >                                                const char *page, size_t len)
> > > > {
> > > >         struct fsg_lun_opts *opts = to_fsg_lun_opts(item);
> > > >         struct fsg_opts *fsg_opts = to_fsg_opts(opts->group.cg_item.ci_parent);
> > > >         int ret;
> > > > 
> > > >         opts->lun->prevent_medium_removal = 0;
> > > >         ret = fsg_store_file(opts->lun, &fsg_opts->common->filesem, "", 0);
> > > >         return ret < 0 ? ret : len;
> > > > }
> > > > 
> > > > CONFIGFS_ATTR_WO(fsg_lun_opts_, forced_eject);    
> > > 
> > > The basic idea is right.  But this should not be a CONFIGFS option; it 
> > > should be an ordinary LUN attribute.  For an example, see the definition of 
> > > file_store() in f_mass_storage.c; your routine should look very similar.  
> > 
> > Okay, but where this attribute is located in sysfs? How can I use it?  
> 
> Well, it's going to be in different places depending on what UDC driver 
> your gadget uses.  On my system I'm using the dummy_udc driver, so the 
> sysfs "file" attribute is located at:
> 
> 	/sys/devices/platform/dummy_ucd.0/gadget/lun0/file
> 
> If instead you're looking at
> 
> 	/sys/module/g_mass_storage/parameters/file
> 
> or in some configfs directory, that's the wrong place.  You can eject 
> the media simply by doing (as root):
> 
> 	echo >/sys/devices/.../gadget/lun0/file
> 
> (fill in the "..." appropriately for your system).
> 
> > Sorry for the stupid question.  
> 
> Not at all.

Thanks! Unfortunately I'm using dwc2 driver and it doesn't have any gadget parameters
outside of the configfs:

[root@pikvm ~]# find /sys -iname lun0
[root@pikvm ~]# find /sys -iname lun.0
/sys/kernel/config/usb_gadget/kvmd/functions/mass_storage.usb0/lun.0
[root@pikvm ~]#

So in my local case configfs is only way to place forced_eject :(
Could we add both device attrs and configfs file?




[Index of Archives]     [Linux Media]     [Linux Input]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [Old Linux USB Devel Archive]

  Powered by Linux