[PATCH] vt: vt_ioctl: Add new ioctl to read vt switch lock state

[PATCH] vt: vt_ioctl: Add new ioctl to read vt switch lock state

[Johannes Hostert]

The existing ioctls VT_LOCKSWITCH and VT_UNLOCKSWITCH can be used to
allow/disallow switching the virtual terminal. However, no mechanism
exists that allows software to read this lock state.

Userspace programs that try to switch to another virtual terminal
like chvt have no mechanism to figure out whether they will be able
to actually switch the terminal. When eg. chvt is run while terminal
switching is disabled, it simply sleeps forever waiting for the target
terminal to become active.

This commit introduces a new ioctl VT_GETLOCKSWITCH that allows
reading the current state of the switch lock flag. Userspace
software can then use that flag and handle not being able to switch
virtual terminals.

Example program using this:

#include <linux/vt.h>
#include <sys/ioctl.h>
#include <fcntl.h>
#include <stdio.h>
#include <errno.h>
#include <error.h>
#include <unistd.h>
const char* console_device = "/dev/tty0";
int main(int argc, char* argv[]) {
	int fd;
	while ((fd = open(console_device, O_RDWR)) == -1 && errno == EINTR);
	if (fd < 0)
		error(1, errno, "Opening %s", console_device);
        int ret;
	while ((ret = ioctl(fd, VT_GETLOCKSWITCH, 1)) == -1 && errno == EINTR);
	if (ret == -1)
		error(1, errno, "%s: VT_GETLOCKSWITCH", console_device);
	printf("VT switching is %s\n", ret == 1 ? "locked" : "unlocked");
	close(fd);
	return 0;
}

Signed-off-by: Johannes Hostert <jojohostert@gmail.com>
---
 drivers/tty/vt/vt_ioctl.c | 5 +++++
 include/uapi/linux/vt.h   | 1 +
 2 files changed, 6 insertions(+)

diff --git a/drivers/tty/vt/vt_ioctl.c b/drivers/tty/vt/vt_ioctl.c
index daf61c28ba76..08b808e1fbf0 100644
--- a/drivers/tty/vt/vt_ioctl.c
+++ b/drivers/tty/vt/vt_ioctl.c
@@ -1028,6 +1028,11 @@ int vt_ioctl(struct tty_struct *tty,
 			return -EPERM;
 		vt_dont_switch = false;
 		break;
+	case VT_GETLOCKSWITCH:
+		if (!capable(CAP_SYS_TTY_CONFIG))
+			return -EPERM;
+		ret = vt_dont_switch ? 1 : 0;
+		break;
 	case VT_GETHIFONTMASK:
 		ret = put_user(vc->vc_hi_font_mask,
 					(unsigned short __user *)arg);
diff --git a/include/uapi/linux/vt.h b/include/uapi/linux/vt.h
index e9d39c48520a..13a1e82dfb14 100644
--- a/include/uapi/linux/vt.h
+++ b/include/uapi/linux/vt.h
@@ -61,6 +61,7 @@ struct vt_consize {
 #define VT_RESIZEX      0x560A  /* set kernel's idea of screensize + more */
 #define VT_LOCKSWITCH   0x560B  /* disallow vt switching */
 #define VT_UNLOCKSWITCH 0x560C  /* allow vt switching */
+#define VT_GETLOCKSWITCH 0x5610  /* return vt switch lock state */
 #define VT_GETHIFONTMASK 0x560D  /* return hi font mask */
 
 struct vt_event {
-- 
2.26.2

Re: [PATCH] vt: vt_ioctl: Add new ioctl to read vt switch lock state

[Jiri Slaby]

On 26. 04. 20, 1:14, Johannes Hostert wrote:
> The existing ioctls VT_LOCKSWITCH and VT_UNLOCKSWITCH can be used to
> allow/disallow switching the virtual terminal. However, no mechanism
> exists that allows software to read this lock state.
> 
> Userspace programs that try to switch to another virtual terminal
> like chvt have no mechanism to figure out whether they will be able
> to actually switch the terminal. When eg. chvt is run while terminal
> switching is disabled, it simply sleeps forever waiting for the target
> terminal to become active.
> 
> This commit introduces a new ioctl VT_GETLOCKSWITCH that allows
> reading the current state of the switch lock flag. Userspace
> software can then use that flag and handle not being able to switch
> virtual terminals.
> 
> Example program using this:
> 
> #include <linux/vt.h>
> #include <sys/ioctl.h>
> #include <fcntl.h>
> #include <stdio.h>
> #include <errno.h>
> #include <error.h>
> #include <unistd.h>
> const char* console_device = "/dev/tty0";
> int main(int argc, char* argv[]) {
> 	int fd;
> 	while ((fd = open(console_device, O_RDWR)) == -1 && errno == EINTR);
> 	if (fd < 0)
> 		error(1, errno, "Opening %s", console_device);
>         int ret;
> 	while ((ret = ioctl(fd, VT_GETLOCKSWITCH, 1)) == -1 && errno == EINTR);
> 	if (ret == -1)
> 		error(1, errno, "%s: VT_GETLOCKSWITCH", console_device);

And now if you do VT_LOCKSWITCH even when VT_GETLOCKSWITCH returns 0,
vt_dont_switch can be 1 already.

So NACK.

Please don't add another racy ioctl. Introduce VT_LOCKSWITCH_NONBLOCK or
something like that instead (like respecting O_NONBLOCK).

thanks,
-- 
js
suse labs