Message ID | 1480507521-17766-1-git-send-email-lars@metafoo.de |
---|---|
State | New |
Headers | show |
On 11/30/2016 01:05 PM, Lars-Peter Clausen wrote: > ... > noop_llseek() is primarily meant for > devices that can not support seek, but where there might be a user that > depends on the seek() operation succeeding. For newly added devices that > can not support seek operations it is recommended to use no_llseek(), which > will return an error. For more information see commit 6038f373a3dc > ("llseek: automatically add .llseek fop"). > ... Arnd, on this subject. As far as I can see FMODE_SEEK is only set if the file was created using sys_open() or a equivalent operation. Files that are created using other system calls (e.g. like pipe(), eventfd()) or IOCTLs (like for the GPIO line and event handles) do not have FMODE_SEEK set. If FMODE_SEEK is not set vfs_seek() will ignore the fops llseek() callback and always use no_llseek(). Many of these files that don't have FMODE_SEEK set still use noop_llseek(). Partially introduced by your commit, partially spread by copy and paste. Since the llseek field is ignored for those, what is your recommendation here? Drop the llseek initialization or change it to no_llseek? And if it is the later what is the reason that it should not be removed even if the field is never used? Thanks, - Lars -- To unsubscribe from this list: send the line "unsubscribe linux-gpio" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wed, Nov 30, 2016 at 1:05 PM, Lars-Peter Clausen <lars@metafoo.de> wrote: > The GPIO chardev is used for management tasks (allocating line and event > handles) and does neither support read() nor write() operations. Hence it > does not make much sense to allow seek operations. > > Currently the chardev uses noop_llseek() for its seek implementation. This > function does not move the pointer and simply returns the current position > (always 0 for the GPIO chardev). noop_llseek() is primarily meant for > devices that can not support seek, but where there might be a user that > depends on the seek() operation succeeding. For newly added devices that > can not support seek operations it is recommended to use no_llseek(), which > will return an error. For more information see commit 6038f373a3dc > ("llseek: automatically add .llseek fop"). > > Unfortunately this was overlooked when the GPIO chardev ABI was introduced. > But it is highly unlikely that since then userspace applications have > appeared that rely on being able to perform non-failing seek operations on > a GPIO chardev file descriptor. So it should be safe to change from > noop_llseel() to no_seek(). Also use nonseekable_open() in the chardev > open() callback to clear the FMODE_SEEK, FMODE_PREAD and FMODE_PWRITE flags > from the file. Neither of these should be set on a file that does not > support seek operations. > > Fixes: 3c702e9987e2 ("gpio: add a userspace chardev ABI for GPIOs") > Signed-off-by: Lars-Peter Clausen <lars@metafoo.de> > --- > Ideally this gets added to stable as well to get a consistent ABI > throughout. Thanks Lars-Peter, I applied this and tagged it for stable. Yours, Linus Walleij -- To unsubscribe from this list: send the line "unsubscribe linux-gpio" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c index 71a2dac..f1615f5 100644 --- a/drivers/gpio/gpiolib.c +++ b/drivers/gpio/gpiolib.c @@ -986,7 +986,8 @@ static int gpio_chrdev_open(struct inode *inode, struct file *filp) return -ENODEV; get_device(&gdev->dev); filp->private_data = gdev; - return 0; + + return nonseekable_open(inode, filp); } /** @@ -1011,7 +1012,7 @@ static const struct file_operations gpio_fileops = { .release = gpio_chrdev_release, .open = gpio_chrdev_open, .owner = THIS_MODULE, - .llseek = noop_llseek, + .llseek = no_llseek, .unlocked_ioctl = gpio_ioctl, #ifdef CONFIG_COMPAT .compat_ioctl = gpio_ioctl_compat,
The GPIO chardev is used for management tasks (allocating line and event handles) and does neither support read() nor write() operations. Hence it does not make much sense to allow seek operations. Currently the chardev uses noop_llseek() for its seek implementation. This function does not move the pointer and simply returns the current position (always 0 for the GPIO chardev). noop_llseek() is primarily meant for devices that can not support seek, but where there might be a user that depends on the seek() operation succeeding. For newly added devices that can not support seek operations it is recommended to use no_llseek(), which will return an error. For more information see commit 6038f373a3dc ("llseek: automatically add .llseek fop"). Unfortunately this was overlooked when the GPIO chardev ABI was introduced. But it is highly unlikely that since then userspace applications have appeared that rely on being able to perform non-failing seek operations on a GPIO chardev file descriptor. So it should be safe to change from noop_llseel() to no_seek(). Also use nonseekable_open() in the chardev open() callback to clear the FMODE_SEEK, FMODE_PREAD and FMODE_PWRITE flags from the file. Neither of these should be set on a file that does not support seek operations. Fixes: 3c702e9987e2 ("gpio: add a userspace chardev ABI for GPIOs") Signed-off-by: Lars-Peter Clausen <lars@metafoo.de> --- Ideally this gets added to stable as well to get a consistent ABI throughout. --- drivers/gpio/gpiolib.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)