Possible race condition in usb-serial.c

[J <jhnlmn@yahoo.com>]

Hi,
I read usb-serial.c code (in 2.6.19) and I cannot
figure out how it is
supposed to prevent race condition and premature
deletion of usb_serial
structure. I see that the code attempts to protect
usb_serial by ref
counting, but it does not appear to be correct. I am
not 100% sure in my
findings, so I will appreciate if somebody will double
check.

Suppose:
A:->usb_serial_disconnect
A:  -> usb_serial_put (serial);
A:   -> kref_put
A:    if ((atomic_read(&kref->refcount) == 1)
             Suppose refcount is 1
A:       -> release -> destroy_serial

B: -> serial_open
B:  -> usb_serial_get_by_index
B:     serial = serial_table[index]
B:     -> kref_get(&serial->kref);

A:        -> return_serial(serial);
A:        serial_table[serial->minor + i] = NULL;
A:          -> kfree (serial);

B:   continue to use serial, which was already freed.

So, I am missing something or the USB serial driver is
broken?

As I understand it, the correct use of ref counted
pointers it to increment
ref count of an object for each outstanding pointer to
this object. But
usb-serial.c keeps one or more pointers to usb_serial
in serial_table, and
does not increments the counter for any of them!

Thank you
John




__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/