| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <201003131219.27482.erazor_de@users.sourceforge.net>
Date: Sat, 13 Mar 2010 12:19:27 +0100
From: Stefan Achatz <erazor_de@...rs.sourceforge.net>
To: Dmitry Torokhov <dmitry.torokhov@...il.com>
Cc: Jiri Kosina <jkosina@...e.cz>,
Jussi Kivilinna <jussi.kivilinna@...et.fi>, wylda@...ny.cz,
Pavel Machek <pavel@...e.cz>,
Alessandro Guido <ag@...ssandroguido.name>,
Tomas Hanak <tomas.hanak@...il.com>,
Jason Noble <nobleja@...ezero.com>, simon.windows@...il.com,
Sean Hildebrand <silverwraithii@...il.com>,
Sid Boyce <sboyce@...eyonder.co.uk>,
Henning Glawe <glaweh@...ian.org>, linux-input@...r.kernel.org,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH 4/4] Added locks for sysfs attributes and internal data.
Hello,
I have some more questions regarding the work on my driver:
1. binary vs text sysfs attributes
As I found little information about binary sysfs attributes I used the normal
sysfs attributes to transfer binary data because they are also easier to use.
The transfered data is far less than the 4096 bytes minimum page size.
So, is there a significant difference in these two variants that should force
me to use the binary variant?
2. memory footprint
I'm thinking about a redesign of my sysfs attributes to reduce io and simplify
the use of binary sysfs attributes (if needed, see above). This would
increase the memory footprint of hid_drvdata to around 5kbytes.
Is it allowed for a driver to constantly alloc such an amount of data or
should I stick with more complicated code and more io?
3. grow up of driver
I see my work is not quite ready for kernel inclusion. Maybe I'm wrong in this
list and should bug other people. Would the linuxdriverproject.org be a
better place to raise my work to adulthood?
Thanks in advance
Stefan Achatz
> Hi Stefan,
>
> On Mon, Mar 08, 2010 at 05:04:29PM +0100, Stefan Achatz wrote:
> > From 47678162da8e374d6f132db21d89b718ee5cfbd1 Mon Sep 17 00:00:00 2001
> > From: Stefan Achatz <erazor_de@...rs.sourceforge.net>
> > Date: Mon, 8 Mar 2010 16:54:19 +0100
> > Subject: [PATCH 4/4] Added locks for sysfs attributes and internal data.
> >
> > Added mutex lock to prevent different processes from accessing
> > sysfs attributes at the same time.
> > Added spin lock for internal data.
> > ---
> > drivers/hid/hid-roccat-kone.c | 246
> > ++++++++++++++++++++++++++++------------ drivers/hid/hid-roccat-kone.h |
> > 9 +-
> > 2 files changed, 179 insertions(+), 76 deletions(-)
> >
> > diff --git a/drivers/hid/hid-roccat-kone.c
> > b/drivers/hid/hid-roccat-kone.c index 941f5b3..eded7d4 100644
> > --- a/drivers/hid/hid-roccat-kone.c
> > +++ b/drivers/hid/hid-roccat-kone.c
> > @@ -223,11 +223,6 @@ static int kone_get_profile(struct usb_device
> > *usb_dev, if (number < 1 || number > 5)
> > return -EINVAL;
> >
> > - /*
> > - * The manufacturer uses a control message of type class with interface
> > - * as recipient and provides the profile number as index parameter.
> > - * This is prevented in userspace by function check_ctrlrecip.
> > - */
> > len = usb_control_msg(usb_dev, usb_rcvctrlpipe(usb_dev, 0),
> > USB_REQ_CLEAR_FEATURE, USB_TYPE_CLASS
> >
> > | USB_RECIP_INTERFACE | USB_DIR_IN,
> >
> > @@ -398,16 +393,18 @@ static int kone_get_firmware_version(struct
> > usb_device *usb_dev, int *result) static ssize_t
> > kone_sysfs_set_settings(struct device *dev,
> > struct device_attribute *attr, char const *buf, size_t size)
> > {
> > - struct hid_device *hdev = dev_get_drvdata(dev);
> > - struct kone_device *kone = hid_get_drvdata(hdev);
> > - struct usb_interface *intf = to_usb_interface(dev);
> > - struct usb_device *usb_dev = interface_to_usbdev(intf);
> > + struct kone_device *kone = hid_get_drvdata(dev_get_drvdata(dev));
> > + struct usb_device *usb_dev =
> > interface_to_usbdev(to_usb_interface(dev)); int err;
> > + unsigned long flags;
> >
> > if (size != sizeof(struct kone_settings))
> > return -EINVAL;
> >
> > + mutex_lock(&kone->kone_lock);
> > err = kone_set_settings(usb_dev, (struct kone_settings const *)buf);
>
> Hmm, this kind of casting tells me that this is binary, not text data
> and thus binary sysfs attribute shoudl be used.
>
> > + mutex_unlock(&kone->kone_lock);
> > +
> > if (err)
> > return err;
> >
> > @@ -415,9 +412,10 @@ static ssize_t kone_sysfs_set_settings(struct device
> > *dev, * If we get here, treat buf as okay (apart from checksum) and use
> > value * of startup_profile as its at hand
> > */
> > + spin_lock_irqsave(&kone->actual_lock, flags);
> > kone->act_profile = ((struct kone_settings *)buf)->startup_profile;
> > - kone->act_profile_valid = 1;
> > - kone->act_dpi_valid = 0;
> > + kone->act_dpi = -1;
> > + spin_unlock_irqrestore(&kone->actual_lock, flags);
>
> You don't really need a spinlock to set one integer.
>
> > return sizeof(struct kone_settings);
> > }
> > @@ -425,10 +423,14 @@ static ssize_t kone_sysfs_set_settings(struct
> > device *dev, static ssize_t kone_sysfs_show_settings(struct device *dev,
> > struct device_attribute *attr, char *buf)
> > {
> > - struct usb_interface *intf = to_usb_interface(dev);
> > - struct usb_device *usb_dev = interface_to_usbdev(intf);
> > + struct kone_device *kone = hid_get_drvdata(dev_get_drvdata(dev));
> > + struct usb_device *usb_dev =
> > interface_to_usbdev(to_usb_interface(dev)); int err;
> > +
> > + mutex_lock(&kone->kone_lock);
> > err = kone_get_settings(usb_dev, (struct kone_settings *)buf);
> > + mutex_unlock(&kone->kone_lock);
> > +
> > if (err)
> > return err;
> >
> > @@ -437,10 +439,14 @@ static ssize_t kone_sysfs_show_settings(struct
> > device *dev,
> >
> > static ssize_t kone_sysfs_get_profile(struct device *dev, char *buf, int
> > number) {
> > - struct usb_interface *intf = to_usb_interface(dev);
> > - struct usb_device *usb_dev = interface_to_usbdev(intf);
> > + struct kone_device *kone = hid_get_drvdata(dev_get_drvdata(dev));
> > + struct usb_device *usb_dev =
> > interface_to_usbdev(to_usb_interface(dev)); int err;
> > +
> > + mutex_lock(&kone->kone_lock);
> > err = kone_get_profile(usb_dev, (struct kone_profile *)buf, number);
> > + mutex_unlock(&kone->kone_lock);
> > +
> > if (err)
> > return err;
> > return sizeof(struct kone_profile);
> > @@ -449,15 +455,17 @@ static ssize_t kone_sysfs_get_profile(struct device
> > *dev, char *buf, int number) static ssize_t kone_sysfs_set_profile(struct
> > device *dev, char const *buf, size_t size, int number)
> > {
> > - struct usb_interface *intf = to_usb_interface(dev);
> > - struct usb_device *usb_dev = interface_to_usbdev(intf);
> > -
> > + struct kone_device *kone = hid_get_drvdata(dev_get_drvdata(dev));
> > + struct usb_device *usb_dev =
> > interface_to_usbdev(to_usb_interface(dev)); int err;
> >
> > if (size != sizeof(struct kone_profile))
> > return -EINVAL;
> >
> > + mutex_lock(&kone->kone_lock);
> > err = kone_set_profile(usb_dev, buf, number);
> > + mutex_unlock(&kone->kone_lock);
> > +
> > if (err)
> > return err;
> >
> > @@ -525,32 +533,68 @@ static ssize_t kone_sysfs_set_profile_5(struct
> > device *dev, }
> >
> > /*
> > - * Helper to fill kone_device structure with actual values
> > - * On success returns 0
> > - * On failure returns errno
> > + * Fills act_profile in kone_device.
> > + * Also writes result in @result.
> > + * Once act_profile is valid, its not getting invalid any more.
> > + * Returns on success 0, on failure errno
> > */
> > -static int kone_device_set_actual_values(struct kone_device *kone,
> > - struct device *dev, int both)
> > +static int kone_device_set_actual_profile(struct kone_device *kone,
> > + struct device *dev, int *result)
> > {
> > - struct usb_interface *intf = to_usb_interface(dev);
> > - struct usb_device *usb_dev = interface_to_usbdev(intf);
> > - int err;
> > + struct usb_device *usb_dev =
> > interface_to_usbdev(to_usb_interface(dev)); + int err, temp;
> > + unsigned long flags;
> >
> > - /* first make sure profile is valid */
> > - if (!kone->act_profile_valid) {
> > - err = kone_get_startup_profile(usb_dev, &kone->act_profile);
> > + spin_lock_irqsave(&kone->actual_lock, flags);
> > + if (kone->act_profile == -1) {
> > + spin_unlock_irqrestore(&kone->actual_lock, flags);
> > + err = kone_get_startup_profile(usb_dev, &temp);
> > if (err)
> > return err;
> > - kone->act_profile_valid = 1;
> > + spin_lock_irq(&kone->actual_lock);
> > + kone->act_profile = temp;
> > + spin_unlock_irqrestore(&kone->actual_lock, flags);
>
> You shoudl not be mixing _irq/_irqrestore. Also it is probably not
> needed at all. If it is needed then the common style to acquire and
> release locks only once in a function - this makes checking whether
> lock/unlock is balanced easier.
>
> > + if (result)
> > + *result = temp;
> > + } else {
> > + if (result)
> > + *result = kone->act_profile;
> > + spin_unlock_irqrestore(&kone->actual_lock, flags);
> > }
> > + return 0;
> > +}
> > +
> > +/*
> > + * Fills act_dpi in kone_device.
> > + * Also writes result in @result.
> > + * On success returns 0
> > + * On failure returns errno
> > + */
> > +static int kone_device_set_actual_dpi(struct kone_device *kone,
> > + struct device *dev, int *result)
> > +{
> > + struct usb_device *usb_dev =
> > interface_to_usbdev(to_usb_interface(dev)); + int err, temp;
> > + unsigned long flags;
> > +
> > + kone_device_set_actual_profile(kone, dev, NULL);
> >
> > - /* then get startup dpi value */
> > - if (!kone->act_dpi_valid && both) {
> > + spin_lock_irqsave(&kone->actual_lock, flags);
> > + if (kone->act_dpi == -1) {
> > + spin_unlock_irqrestore(&kone->actual_lock, flags);
>
> So what are we protecting exactly? What if act_dpi will become -1 here
> (after you released the spinlock?
>
> > err = kone_get_profile_startup_dpi(usb_dev, kone->act_profile,
> > - &kone->act_dpi);
> > + &temp);
>
> Stopped looking - locking obviously is bogus and needs to be redone. You
> need to decide what exactly needs protection and what critical sections
> are.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists