Skip to contentSkip to conversation

usb: misc: uss720: properly clean up reference in uss720_probe()

PATCH Superseded Greg Kroah-Hartman 2 participants 5 messages Feb 23 USB

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Greg Kroah-Hartman

author maintainer
#0 lore raw

PATCH v2 1/1 usb: misc: uss720: properly clean up reference in uss720_probe()

If get_1284_register() fails, the usb device reference count is incorrect and needs to be properly dropped before returning.

Cc stable <[email protected]>
Assisted-by gkh_clanker_2000

---
v2: properly clean up the previously allocated resources by jumping to
the error path and not just retrning directly

 drivers/usb/misc/uss720.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/usb/misc/uss720.c +4 -2
index ec8bd968c4de..54107bf280df 100644
@@ -735,8 +735,10 @@ static int uss720_probe(struct usb_interface *intf,
 	 * here. */
 	ret = get_1284_register(pp, 0, &reg, GFP_KERNEL);
 	dev_dbg(&intf->dev, "reg: %7ph\n", priv->reg);
-	if (ret < 0)
-		return ret;
+	if (ret < 0) {
+		usb_put_dev(usbdev);
+		goto probe_abort;
+	}
 
 	ret = usb_find_last_int_in_endpoint(interface, &epd);
 	if (!ret) {
Signature · 1 line
2.53.0
#1 lore raw

Hi,

Quoted 18 lines · Greg Kroah-Hartman

If get_1284_register() fails, the usb device reference count is incorrect and needs to be properly dropped before returning.

Cc stable <[email protected]>
Assisted-by gkh_clanker_2000

---
v2: properly clean up the previously allocated resources by jumping to
the error path and not just retrning directly

  drivers/usb/misc/uss720.c | 6 ++++--
  1 file changed, 4 insertions(+), 2 deletions(-)
drivers/usb/misc/uss720.c +0 -0
index ec8bd968c4de..54107bf280df 100644

drivers/usb/misc/uss720.c static int uss720_probe(struct usb_interface *intf, · line 740

Quoted 8 lines
drivers/usb/misc/uss720.c +3 -2
@@ -735,8 +735,10 @@ static int uss720_probe(struct usb_interface *intf,
  	 * here. */
  	ret = get_1284_register(pp, 0, &reg, GFP_KERNEL);
  	dev_dbg(&intf->dev, "reg: %7ph\n", priv->reg);
-	if (ret < 0)
-		return ret;
+	if (ret < 0) {
+		usb_put_dev(usbdev);
+		goto probe_abort;

This jumps to probe_abort, which calls destroy_priv() indirectly via kref_put(). Either that works, which would mean that you must _not_ do a second usb_put_dev() or it does not, in which case the earlier error handling for parport_register_port() is incorrect.

Either way, there is a problem.

Regards
Oliver

author maintainer
#2 lore raw
Quoted 35 lines · Oliver Neukum

Hi,

Quoted 18 lines · Greg Kroah-Hartman

If get_1284_register() fails, the usb device reference count is incorrect and needs to be properly dropped before returning.

Cc stable <[email protected]>
Assisted-by gkh_clanker_2000

---
v2: properly clean up the previously allocated resources by jumping to
the error path and not just retrning directly

  drivers/usb/misc/uss720.c | 6 ++++--
  1 file changed, 4 insertions(+), 2 deletions(-)
drivers/usb/misc/uss720.c +0 -0
index ec8bd968c4de..54107bf280df 100644

drivers/usb/misc/uss720.c static int uss720_probe(struct usb_interface *intf, · line 740

Quoted 8 lines
drivers/usb/misc/uss720.c +3 -2
@@ -735,8 +735,10 @@ static int uss720_probe(struct usb_interface *intf,
  	 * here. */
  	ret = get_1284_register(pp, 0, &reg, GFP_KERNEL);
  	dev_dbg(&intf->dev, "reg: %7ph\n", priv->reg);
-	if (ret < 0)
-		return ret;
+	if (ret < 0) {
+		usb_put_dev(usbdev);
+		goto probe_abort;

This jumps to probe_abort, which calls destroy_priv() indirectly via kref_put(). Either that works, which would mean that you must _not_ do a second usb_put_dev() or it does not, in which case the earlier error handling for parport_register_port() is incorrect.

Quoted 1 line

Either way, there is a problem.

Argh, I missed that part of the error path, you are right. The existing code is wrong for the earlier cleanup logic, I'll go fix that instead, thanks for the review.

It's tricky stuff like this which makes me can't wait for rust to be possible for USB drivers, someday...

thanks,

greg k-h

#3 lore raw
Quoted 2 lines · Greg Kroah-Hartman

It's tricky stuff like this which makes me can't wait for rust to be possible for USB drivers, someday...

Well, as you wish to touch upon that topic ...

I am afraid this is not a problem of language. Rust solves the issue of object life time. Unfortunately that is useless in this case. I know we all wish to solve the issue, but let me explain.

Now, before you either start despairing or get angry, we need to look at concepts. It seems to my you're looking at the problem in terms of life time and basically something that can be solved by reference counting respectively life time rules.
That is unfortunately not true. Our problem with USB drivers is a question of binding. probe() does not mean that a device has been plugged in, nor does disconnect() mean that a device has gone away. It means that the binding between an interface and a driver is requested respectively goes away. Hotplug is merely the most common cause of these requests.

Nevertheless you have to cease using the interface as disconnect() is called. References cannot change that. There are two reasons for that

a) there is no object representing the binding. It is technically a pointer not a data structure. There is nothing to refcount and no object whose lifetime you can specify. There is a variable that is changing
b) there is a state transition, not a life time question. There is an event that changes the state of a binding if you will.

Sorry you don't like this, but this is a design issue, not a language issue.

Regards
Oliver

author maintainer
#4 lore raw
Quoted 36 lines · Oliver Neukum
Quoted 2 lines · Greg Kroah-Hartman

It's tricky stuff like this which makes me can't wait for rust to be possible for USB drivers, someday...

Well, as you wish to touch upon that topic ...

I am afraid this is not a problem of language. Rust solves the issue of object life time. Unfortunately that is useless in this case. I know we all wish to solve the issue, but let me explain.

Now, before you either start despairing or get angry, we need to look at concepts. It seems to my you're looking at the problem in terms of life time and basically something that can be solved by reference counting respectively life time rules.
That is unfortunately not true. Our problem with USB drivers is a question of binding. probe() does not mean that a device has been plugged in, nor does disconnect() mean that a device has gone away. It means that the binding between an interface and a driver is requested respectively goes away. Hotplug is merely the most common cause of these requests.

Nevertheless you have to cease using the interface as disconnect() is called. References cannot change that. There are two reasons for that

a) there is no object representing the binding. It is technically a pointer not a data structure. There is nothing to refcount and no object whose lifetime you can specify. There is a variable that is changing
b) there is a state transition, not a life time question. There is an event that changes the state of a binding if you will.

Quoted 2 lines

Sorry you don't like this, but this is a design issue, not a language issue.

It's not a choice of me liking it or not, but I think if you look at how the driver model binds through rust to drivers, most/all of what you refer to above is already worked out as there the issues here are with the driver handling resources that it thinks it still has access to (or forgets about.) We have working bindings for PCI and other busses already for rust, and this is handled there pretty cleanly, I see no reason why USB would be any different.

thanks,

greg k-h