On 5/23/07, Jiri Slaby <[email protected]> wrote:
> Simon Arlott napsal(a):
> > On 22/05/07 21:06, Matthias Kaehlcke wrote:
> >> would the following resolve the problem?
> >>
> >> if(mutex_lock_interruptible(&info->write_mtx)) return
> >> -ERESTARTSYS
> >>
> >> thanks for your comments
Hum. I remember suggesting the fix when this came few weeks back,
somehow the older version seems to have made it into Linus' tree.
> > No. At least one user of tty_operations/tty_driver's write function
> > doesn't check the return value so it would never be retried, mutex_lock
> > should be used instead.
Replacing mutex_lock_interruptible() with a mutex_lock() could be a
change in behaviour visible to userspace. Best to continue to use
_interruptible. Returning EINTR would be safe choice because if
Simon's right that this isn't retried, we could end up breaking
userspace that isn't quite ready to see ERESTARTSYS coming out of
a write(2), but ...
> Who? There are some drivers that returns ERESTARTSYS from write function.
Right, several others in drivers/char/ do return ERESTARTSYS. Hope
somebody picks this up -- no email listed in MAINTAINERS for this.
---
Check the return of mutex_lock_interruptible() in drivers/char/rocket.c
and return ERESTARTSYS if we were interrupted.
Signed-off-by: Satyam Sharma <[email protected]>
---
drivers/char/rocket.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
---
diff -ruNp a/drivers/char/rocket.c b/drivers/char/rocket.c
--- a/drivers/char/rocket.c 2007-05-20 04:11:54.000000000 +0530
+++ b/drivers/char/rocket.c 2007-05-23 18:14:35.000000000 +0530
@@ -1702,7 +1702,8 @@ static int rp_write(struct tty_struct *t
if (count <= 0 || rocket_paranoia_check(info, "rp_write"))
return 0;
- mutex_lock_interruptible(&info->write_mtx);
+ if (mutex_lock_interruptible(&info->write_mtx))
+ return -ERESTARTSYS;
#ifdef ROCKET_DEBUG_WRITE
printk(KERN_INFO "rp_write %d chars...", count);
El Wed, May 23, 2007 at 06:25:49PM +0530 Satyam Sharma ha dit:
> On 5/23/07, Jiri Slaby <[email protected]> wrote:
> >Simon Arlott napsal(a):
> >> On 22/05/07 21:06, Matthias Kaehlcke wrote:
> >>> would the following resolve the problem?
> >>>
> >>> if(mutex_lock_interruptible(&info->write_mtx)) return
> >>> -ERESTARTSYS
> >>>
> >>> thanks for your comments
>
> Hum. I remember suggesting the fix when this came few weeks back,
> somehow the older version seems to have made it into Linus' tree.
i also remembered your suggestion but thought it was a similar case in
another driver. i think it's my fault that the older version made it
into the Linus tree cause i sent the patch in a new thread instead of
responding to the initial post :(
> >> No. At least one user of tty_operations/tty_driver's write function
> >> doesn't check the return value so it would never be retried, mutex_lock
> >> should be used instead.
>
> Replacing mutex_lock_interruptible() with a mutex_lock() could be a
> change in behaviour visible to userspace. Best to continue to use
> _interruptible. Returning EINTR would be safe choice because if
> Simon's right that this isn't retried, we could end up breaking
> userspace that isn't quite ready to see ERESTARTSYS coming out of
> a write(2), but ...
thanks for your fix!
--
Matthias Kaehlcke
Linux Application Developer
Barcelona
The yellow ships hung in the air just like bricks dont do
(The Hitch-Hiker's Guide to the Galaxy)
.''`.
using free software / Debian GNU/Linux | http://debian.org : :' :
`. `'`
gpg --keyserver pgp.mit.edu --recv-keys 47D8E5D4 `-