read / write / close sync for device files

read / write / close sync for device files

[Amit Nagal]

Hi ,

if  a application is  doing read /write I/O with a  usb device ,  in 2
different threads in userspace ,

and at usb disconnect event ,  read thread returns with error and
application calls close() subsequently without waiting for write
thread to finish .

Do  the VFS upper layers  maintain locks that  take care that until
both read and write methods in usb driver donot finish and return with
error in kernel space

close will pend and usb driver release method will not be called ?

Regards
Amit Nagal

read / write / close sync for device files

[Greg KH]

On Mon, Jul 04, 2011 at 06:13:49PM +0530, Amit Nagal wrote:
> Hi ,
> 
> if  a application is  doing read /write I/O with a  usb device ,  in 2
> different threads in userspace ,

What type of USB device?

> and at usb disconnect event ,  read thread returns with error and
> application calls close() subsequently without waiting for write
> thread to finish .
> 
> Do  the VFS upper layers  maintain locks that  take care that until
> both read and write methods in usb driver donot finish and return with
> error in kernel space
> 
> close will pend and usb driver release method will not be called ?

No, close will succeed, and then all will be cleaned up properly.

Try it yourself and see :)

greg k-h

read / write / close sync for device files

[Amit Nagal]

On Mon, Jul 4, 2011 at 9:40 PM, Greg KH <greg@kroah.com> wrote:
> On Mon, Jul 04, 2011 at 06:13:49PM +0530, Amit Nagal wrote:
>> Hi ,
>>
>> if ?a application is ?doing read /write I/O with a ?usb device , ?in 2
>> different threads in userspace ,
>
> What type of USB device?

Thanx for reply .

i am using USB hid interface and using USBHID  driver with hid-raw
device interface .

Actually i have 2 threads in userspace app  : one for read and another
one for write .

my problem is like this :
In case of usb disconnect  , if read() returns first with error and
application calls close() in read thread itself in userspace   , while
write() is still in progress() in usb driver in a separate thread   ,
is this ok and safe  ?

or shall the application be designed such that both until read and
write return with error , it should not call close in userspace ?

Also to test this  , i applied a 10 seconds delay at the end of
hidraw_read() function in hidraw driver like :

static ssize_t hidraw_read(struct file *file, char __user *buffer,
size_t count, loff_t *ppos)
{
............................................
                 /********************* test block starts */

                printk(KERN_EMERG "[%s] - sleep start\n" , __func__);
                mdelay(10000);  /* 10s delay */
	printk(KERN_EMERG "[%s] - sleep end\n" , __func__);

                /*****************  test block ends */

	mutex_unlock(&list->read_mutex);
	return ret;
}

now while hidraw_read is executing 10s delay  , i disconnects  usb
device  , calls close(fd) .
i observed  , while close(fd) returns with 0 successfully immediately
, usb driver release() function corresponding to close() was called
only when hidraw_read returns after completing delay which is approx
9s later calling to close() in userspace  .

so do the usb-core layers or VFS layers maintains some locks that take
care that if read / write is in progress
and application calls close() , corresponding release() method in usb
driver will be called only upon completion
of all read / writes ?

Regards
Amit Nagal

read / write / close sync for device files

[Greg KH]

On Tue, Jul 05, 2011 at 12:37:26PM +0530, Amit Nagal wrote:
> so do the usb-core layers or VFS layers maintains some locks that take
> care that if read / write is in progress
> and application calls close() , corresponding release() method in usb
> driver will be called only upon completion
> of all read / writes ?

No, release will be called when all of the userspace file handles call
close(), and not before.

hope this helps,

greg k-h