2015-04-13 23:36:47

by NeilBrown

[permalink] [raw]
Subject: [PATCH] hso: fix refcnt leak in recent patch.



Prior to
commit 29bd3bc1194c624ce863cab2a7da9bc1f0c3b47b
hso: fix crash when device disappears while serial port is open

hso_serial_open would always kref_get(&serial->parent->ref) before
returning zero.
Since that commit, it only calls kref_get when returning 0 if
serial->port.count was zero.

This results in calls to
kref_put(&serial->parent->ref, hso_serial_ref_free);

after hso_serial_ref_free has been called, which dereferences a freed
pointer.

This patch adds the missing kref_get().

Fixes: commit 29bd3bc1194c624ce863cab2a7da9bc1f0c3b47b
Cc: [email protected] (v4.0)
Cc: Olivier Sobrie <[email protected]>
Signed-off-by: NeilBrown <[email protected]>

diff --git a/drivers/net/usb/hso.c b/drivers/net/usb/hso.c
index 75befc1bd816..6848fc903340 100644
--- a/drivers/net/usb/hso.c
+++ b/drivers/net/usb/hso.c
@@ -1299,6 +1299,7 @@ static int hso_serial_open(struct tty_struct *tty, struct file *filp)
}
} else {
D1("Port was already open");
+ kref_get(&serial->parent->ref);
}

usb_autopm_put_interface(serial->parent->interface);


Attachments:
(No filename) (811.00 B)
OpenPGP digital signature

2015-04-14 01:03:23

by NeilBrown

[permalink] [raw]
Subject: Re: [PATCH] hso: fix refcnt leak in recent patch.

On Tue, 14 Apr 2015 09:36:34 +1000 NeilBrown <[email protected]> wrote:

>
>
> Prior to
> commit 29bd3bc1194c624ce863cab2a7da9bc1f0c3b47b
> hso: fix crash when device disappears while serial port is open
>
> hso_serial_open would always kref_get(&serial->parent->ref) before
> returning zero.
> Since that commit, it only calls kref_get when returning 0 if
> serial->port.count was zero.
>
> This results in calls to
> kref_put(&serial->parent->ref, hso_serial_ref_free);
>
> after hso_serial_ref_free has been called, which dereferences a freed
> pointer.
>
> This patch adds the missing kref_get().
>
> Fixes: commit 29bd3bc1194c624ce863cab2a7da9bc1f0c3b47b
> Cc: [email protected] (v4.0)
> Cc: Olivier Sobrie <[email protected]>
> Signed-off-by: NeilBrown <[email protected]>
>
> diff --git a/drivers/net/usb/hso.c b/drivers/net/usb/hso.c
> index 75befc1bd816..6848fc903340 100644
> --- a/drivers/net/usb/hso.c
> +++ b/drivers/net/usb/hso.c
> @@ -1299,6 +1299,7 @@ static int hso_serial_open(struct tty_struct *tty, struct file *filp)
> }
> } else {
> D1("Port was already open");
> + kref_get(&serial->parent->ref);
> }
>
> usb_autopm_put_interface(serial->parent->interface);


Sorry - that was wrong.
I'm getting crashes which strongly suggest the kref_put is being called extra
times, but I misunderstood the code and was hasty.

Maybe this instead?

Thanks,
NeilBrown

From: NeilBrown <[email protected]>
Date: Tue, 14 Apr 2015 09:33:03 +1000
Subject: [PATCH] hso: fix refcnt leak in recent patch.

Prior to
commit 29bd3bc1194c624ce863cab2a7da9bc1f0c3b47b
hso: fix crash when device disappears while serial port is open

a kref_get on serial->parent->ref would be taken on each open,
and it would be kref_put on each close.

Now the kref_put happens when the tty_struct is finally put (via
the 'cleanup') providing tty->driver_data has been set.
So the kref_get must be called exact once when tty->driver_data is
set.

With the current code, if the first open fails the kref_get() is never
called, but the kref_put() is called, leaving to a crash.

So change the kref_get call to happen exactly when ->driver_data is
changed from NULL to non-NULL.

Fixes: commit 29bd3bc1194c624ce863cab2a7da9bc1f0c3b47b
Cc: [email protected] (v4.0)
Cc: Olivier Sobrie <[email protected]>
Signed-off-by: NeilBrown <[email protected]>

diff --git a/drivers/net/usb/hso.c b/drivers/net/usb/hso.c
index 75befc1bd816..17fd3820263a 100644
--- a/drivers/net/usb/hso.c
+++ b/drivers/net/usb/hso.c
@@ -1278,6 +1278,8 @@ static int hso_serial_open(struct tty_struct *tty, struct file *filp)
D1("Opening %d", serial->minor);

/* setup */
+ if (tty->driver_data == NULL)
+ kref_get(&serial->parent->ref);
tty->driver_data = serial;
tty_port_tty_set(&serial->port, tty);

@@ -1294,8 +1296,6 @@ static int hso_serial_open(struct tty_struct *tty, struct file *filp)
if (result) {
hso_stop_serial_device(serial->parent);
serial->port.count--;
- } else {
- kref_get(&serial->parent->ref);
}
} else {
D1("Port was already open");


Attachments:
(No filename) (811.00 B)
OpenPGP digital signature

2015-04-14 06:49:34

by Olivier Sobrie

[permalink] [raw]
Subject: Re: [PATCH] hso: fix refcnt leak in recent patch.

Hello Neil,

On Tue, Apr 14, 2015 at 11:03:03AM +1000, NeilBrown wrote:
> On Tue, 14 Apr 2015 09:36:34 +1000 NeilBrown <[email protected]> wrote:
>
> >
> >
> > Prior to
> > commit 29bd3bc1194c624ce863cab2a7da9bc1f0c3b47b
> > hso: fix crash when device disappears while serial port is open
> >
> > hso_serial_open would always kref_get(&serial->parent->ref) before
> > returning zero.
> > Since that commit, it only calls kref_get when returning 0 if
> > serial->port.count was zero.
> >
> > This results in calls to
> > kref_put(&serial->parent->ref, hso_serial_ref_free);
> >
> > after hso_serial_ref_free has been called, which dereferences a freed
> > pointer.
> >
> > This patch adds the missing kref_get().
> >
> > Fixes: commit 29bd3bc1194c624ce863cab2a7da9bc1f0c3b47b
> > Cc: [email protected] (v4.0)
> > Cc: Olivier Sobrie <[email protected]>
> > Signed-off-by: NeilBrown <[email protected]>
> >
> > diff --git a/drivers/net/usb/hso.c b/drivers/net/usb/hso.c
> > index 75befc1bd816..6848fc903340 100644
> > --- a/drivers/net/usb/hso.c
> > +++ b/drivers/net/usb/hso.c
> > @@ -1299,6 +1299,7 @@ static int hso_serial_open(struct tty_struct *tty, struct file *filp)
> > }
> > } else {
> > D1("Port was already open");
> > + kref_get(&serial->parent->ref);
> > }
> >
> > usb_autopm_put_interface(serial->parent->interface);
>
>
> Sorry - that was wrong.
> I'm getting crashes which strongly suggest the kref_put is being called extra
> times, but I misunderstood the code and was hasty.
>
> Maybe this instead?

Indeed, if I undestand correctly the code in tty_io.c, cleanup() method
is also called when the open fails while kref_get is only done if the
open succeeds. Sorry for that mess.
I assume you get that crash when hso_start_serial_device() returns
an error?

At first sight, the patch below looks good to me.
I'll test it in the next days.

Thank you,

Olivier

>
> Thanks,
> NeilBrown
>
> From: NeilBrown <[email protected]>
> Date: Tue, 14 Apr 2015 09:33:03 +1000
> Subject: [PATCH] hso: fix refcnt leak in recent patch.
>
> Prior to
> commit 29bd3bc1194c624ce863cab2a7da9bc1f0c3b47b
> hso: fix crash when device disappears while serial port is open
>
> a kref_get on serial->parent->ref would be taken on each open,
> and it would be kref_put on each close.
>
> Now the kref_put happens when the tty_struct is finally put (via
> the 'cleanup') providing tty->driver_data has been set.
> So the kref_get must be called exact once when tty->driver_data is
> set.
>
> With the current code, if the first open fails the kref_get() is never
> called, but the kref_put() is called, leaving to a crash.
>
> So change the kref_get call to happen exactly when ->driver_data is
> changed from NULL to non-NULL.
>
> Fixes: commit 29bd3bc1194c624ce863cab2a7da9bc1f0c3b47b
> Cc: [email protected] (v4.0)
> Cc: Olivier Sobrie <[email protected]>
> Signed-off-by: NeilBrown <[email protected]>
>
> diff --git a/drivers/net/usb/hso.c b/drivers/net/usb/hso.c
> index 75befc1bd816..17fd3820263a 100644
> --- a/drivers/net/usb/hso.c
> +++ b/drivers/net/usb/hso.c
> @@ -1278,6 +1278,8 @@ static int hso_serial_open(struct tty_struct *tty, struct file *filp)
> D1("Opening %d", serial->minor);
>
> /* setup */
> + if (tty->driver_data == NULL)
> + kref_get(&serial->parent->ref);
> tty->driver_data = serial;
> tty_port_tty_set(&serial->port, tty);
>
> @@ -1294,8 +1296,6 @@ static int hso_serial_open(struct tty_struct *tty, struct file *filp)
> if (result) {
> hso_stop_serial_device(serial->parent);
> serial->port.count--;
> - } else {
> - kref_get(&serial->parent->ref);
> }
> } else {
> D1("Port was already open");


2015-04-14 07:35:48

by NeilBrown

[permalink] [raw]
Subject: Re: [PATCH] hso: fix refcnt leak in recent patch.

On Tue, 14 Apr 2015 08:50:01 +0200 Olivier Sobrie <[email protected]> wrote:

> Hello Neil,
>
> On Tue, Apr 14, 2015 at 11:03:03AM +1000, NeilBrown wrote:
> > On Tue, 14 Apr 2015 09:36:34 +1000 NeilBrown <[email protected]> wrote:
> >
> > >
> > >
> > > Prior to
> > > commit 29bd3bc1194c624ce863cab2a7da9bc1f0c3b47b
> > > hso: fix crash when device disappears while serial port is open
> > >
> > > hso_serial_open would always kref_get(&serial->parent->ref) before
> > > returning zero.
> > > Since that commit, it only calls kref_get when returning 0 if
> > > serial->port.count was zero.
> > >
> > > This results in calls to
> > > kref_put(&serial->parent->ref, hso_serial_ref_free);
> > >
> > > after hso_serial_ref_free has been called, which dereferences a freed
> > > pointer.
> > >
> > > This patch adds the missing kref_get().
> > >
> > > Fixes: commit 29bd3bc1194c624ce863cab2a7da9bc1f0c3b47b
> > > Cc: [email protected] (v4.0)
> > > Cc: Olivier Sobrie <[email protected]>
> > > Signed-off-by: NeilBrown <[email protected]>
> > >
> > > diff --git a/drivers/net/usb/hso.c b/drivers/net/usb/hso.c
> > > index 75befc1bd816..6848fc903340 100644
> > > --- a/drivers/net/usb/hso.c
> > > +++ b/drivers/net/usb/hso.c
> > > @@ -1299,6 +1299,7 @@ static int hso_serial_open(struct tty_struct *tty, struct file *filp)
> > > }
> > > } else {
> > > D1("Port was already open");
> > > + kref_get(&serial->parent->ref);
> > > }
> > >
> > > usb_autopm_put_interface(serial->parent->interface);
> >
> >
> > Sorry - that was wrong.
> > I'm getting crashes which strongly suggest the kref_put is being called extra
> > times, but I misunderstood the code and was hasty.
> >
> > Maybe this instead?
>
> Indeed, if I undestand correctly the code in tty_io.c, cleanup() method
> is also called when the open fails while kref_get is only done if the
> open succeeds. Sorry for that mess.
> I assume you get that crash when hso_start_serial_device() returns
> an error?

Yes, that's correct.

>
> At first sight, the patch below looks good to me.
> I'll test it in the next days.

Thanks.

NeilBrown

>
> Thank you,
>
> Olivier
>
> >
> > Thanks,
> > NeilBrown
> >
> > From: NeilBrown <[email protected]>
> > Date: Tue, 14 Apr 2015 09:33:03 +1000
> > Subject: [PATCH] hso: fix refcnt leak in recent patch.
> >
> > Prior to
> > commit 29bd3bc1194c624ce863cab2a7da9bc1f0c3b47b
> > hso: fix crash when device disappears while serial port is open
> >
> > a kref_get on serial->parent->ref would be taken on each open,
> > and it would be kref_put on each close.
> >
> > Now the kref_put happens when the tty_struct is finally put (via
> > the 'cleanup') providing tty->driver_data has been set.
> > So the kref_get must be called exact once when tty->driver_data is
> > set.
> >
> > With the current code, if the first open fails the kref_get() is never
> > called, but the kref_put() is called, leaving to a crash.
> >
> > So change the kref_get call to happen exactly when ->driver_data is
> > changed from NULL to non-NULL.
> >
> > Fixes: commit 29bd3bc1194c624ce863cab2a7da9bc1f0c3b47b
> > Cc: [email protected] (v4.0)
> > Cc: Olivier Sobrie <[email protected]>
> > Signed-off-by: NeilBrown <[email protected]>
> >
> > diff --git a/drivers/net/usb/hso.c b/drivers/net/usb/hso.c
> > index 75befc1bd816..17fd3820263a 100644
> > --- a/drivers/net/usb/hso.c
> > +++ b/drivers/net/usb/hso.c
> > @@ -1278,6 +1278,8 @@ static int hso_serial_open(struct tty_struct *tty, struct file *filp)
> > D1("Opening %d", serial->minor);
> >
> > /* setup */
> > + if (tty->driver_data == NULL)
> > + kref_get(&serial->parent->ref);
> > tty->driver_data = serial;
> > tty_port_tty_set(&serial->port, tty);
> >
> > @@ -1294,8 +1296,6 @@ static int hso_serial_open(struct tty_struct *tty, struct file *filp)
> > if (result) {
> > hso_stop_serial_device(serial->parent);
> > serial->port.count--;
> > - } else {
> > - kref_get(&serial->parent->ref);
> > }
> > } else {
> > D1("Port was already open");
>
>
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/


Attachments:
(No filename) (811.00 B)
OpenPGP digital signature

2015-04-16 13:21:06

by Olivier Sobrie

[permalink] [raw]
Subject: Re: [PATCH] hso: fix refcnt leak in recent patch.

On Tue, Apr 14, 2015 at 11:03:03AM +1000, NeilBrown wrote:
> On Tue, 14 Apr 2015 09:36:34 +1000 NeilBrown <[email protected]> wrote:
>
> >
> >
> > Prior to
> > commit 29bd3bc1194c624ce863cab2a7da9bc1f0c3b47b
> > hso: fix crash when device disappears while serial port is open
> >
> > hso_serial_open would always kref_get(&serial->parent->ref) before
> > returning zero.
> > Since that commit, it only calls kref_get when returning 0 if
> > serial->port.count was zero.
> >
> > This results in calls to
> > kref_put(&serial->parent->ref, hso_serial_ref_free);
> >
> > after hso_serial_ref_free has been called, which dereferences a freed
> > pointer.
> >
> > This patch adds the missing kref_get().
> >
> > Fixes: commit 29bd3bc1194c624ce863cab2a7da9bc1f0c3b47b
> > Cc: [email protected] (v4.0)
> > Cc: Olivier Sobrie <[email protected]>
> > Signed-off-by: NeilBrown <[email protected]>
> >
> > diff --git a/drivers/net/usb/hso.c b/drivers/net/usb/hso.c
> > index 75befc1bd816..6848fc903340 100644
> > --- a/drivers/net/usb/hso.c
> > +++ b/drivers/net/usb/hso.c
> > @@ -1299,6 +1299,7 @@ static int hso_serial_open(struct tty_struct *tty, struct file *filp)
> > }
> > } else {
> > D1("Port was already open");
> > + kref_get(&serial->parent->ref);
> > }
> >
> > usb_autopm_put_interface(serial->parent->interface);
>
>
> Sorry - that was wrong.
> I'm getting crashes which strongly suggest the kref_put is being called extra
> times, but I misunderstood the code and was hasty.
>
> Maybe this instead?

I tested the patch and it looks fine :-)
Thank you,
Olivier

>
> Thanks,
> NeilBrown
>
> From: NeilBrown <[email protected]>
> Date: Tue, 14 Apr 2015 09:33:03 +1000
> Subject: [PATCH] hso: fix refcnt leak in recent patch.
>
> Prior to
> commit 29bd3bc1194c624ce863cab2a7da9bc1f0c3b47b
> hso: fix crash when device disappears while serial port is open
>
> a kref_get on serial->parent->ref would be taken on each open,
> and it would be kref_put on each close.
>
> Now the kref_put happens when the tty_struct is finally put (via
> the 'cleanup') providing tty->driver_data has been set.
> So the kref_get must be called exact once when tty->driver_data is
> set.
>
> With the current code, if the first open fails the kref_get() is never
> called, but the kref_put() is called, leaving to a crash.
>
> So change the kref_get call to happen exactly when ->driver_data is
> changed from NULL to non-NULL.
>
> Fixes: commit 29bd3bc1194c624ce863cab2a7da9bc1f0c3b47b
> Cc: [email protected] (v4.0)
> Cc: Olivier Sobrie <[email protected]>

Tested-by: Olivier Sobrie <[email protected]>

> Signed-off-by: NeilBrown <[email protected]>
>
> diff --git a/drivers/net/usb/hso.c b/drivers/net/usb/hso.c
> index 75befc1bd816..17fd3820263a 100644
> --- a/drivers/net/usb/hso.c
> +++ b/drivers/net/usb/hso.c
> @@ -1278,6 +1278,8 @@ static int hso_serial_open(struct tty_struct *tty, struct file *filp)
> D1("Opening %d", serial->minor);
>
> /* setup */
> + if (tty->driver_data == NULL)
> + kref_get(&serial->parent->ref);
> tty->driver_data = serial;
> tty_port_tty_set(&serial->port, tty);
>
> @@ -1294,8 +1296,6 @@ static int hso_serial_open(struct tty_struct *tty, struct file *filp)
> if (result) {
> hso_stop_serial_device(serial->parent);
> serial->port.count--;
> - } else {
> - kref_get(&serial->parent->ref);
> }
> } else {
> D1("Port was already open");



--
Olivier