[PATCH-next 0/3] serial/sysrq: Add MAGIC_SYSRQ_SERIAL_SEQUENCE

[PATCH-next 0/3] serial/sysrq: Add MAGIC_SYSRQ_SERIAL_SEQUENCE

[Dmitry Safonov]

Magic sysrq has proven for Arista usecases to be useful for debugging
issues in field, over serial line when the switch is in such bad state
that it can't accept network connections anymore.

Unfortunately, having sysrq always enabled doesn't work for some
embedded boards that tend to generate garbage on serial line (including
BREAKs). Since commit 732dbf3a6104 ("serial: do not accept sysrq
characters via serial port"), it's possible to keep sysrq enabled, but
over serial line.

Add a way to enable sysrq on a uart, where currently it can be
constantly either on or off (CONFIG_MAGIC_SYSRQ_SERIAL).
While doing so, cleanup __sysrq_enabled and serial_core header file.

Sending against -next tree as it's based on removing SUPPORT_SYSRQ
ifdeffery [1].

[1]: https://lkml.kernel.org/r/20191213000657.931618-1-dima@arista.com

Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Jiri Slaby <jslaby@suse.com>
Cc: Vasiliy Khoruzhick <vasilykh@arista.com>
Cc: linux-serial@vger.kernel.org

Thanks,
             Dmitry

Dmitry Safonov (3):
  serial_core: Move sysrq functions from header file
  sysctl/sysrq: Remove __sysrq_enabled copy
  serial/sysrq: Add MAGIC_SYSRQ_SERIAL_SEQUENCE

 drivers/tty/serial/serial_core.c | 123 +++++++++++++++++++++++++++++++
 drivers/tty/sysrq.c              |   7 ++
 include/linux/serial_core.h      |  86 ++-------------------
 include/linux/sysrq.h            |   1 +
 kernel/sysctl.c                  |  41 ++++++-----
 lib/Kconfig.debug                |   8 ++
 6 files changed, 167 insertions(+), 99 deletions(-)

-- 
2.24.1

[PATCH-next 2/3] sysctl/sysrq: Remove __sysrq_enabled copy

[Dmitry Safonov]

Many embedded boards have a disconnected TTL level serial which can
generate some garbage that can lead to spurious false sysrq detects.

Currently, sysrq can be either completely disabled for serial console
or always disabled (with CONFIG_MAGIC_SYSRQ_SERIAL), since
commit 732dbf3a6104 ("serial: do not accept sysrq characters via serial port")

At Arista, we have such boards that can generate BREAK and random
garbage. While disabling sysrq for serial console would solve
the problem with spurious false sysrq triggers, it's also desirable
to have a way to enable sysrq back.

Having the way to enable sysrq was beneficial to debug lockups with
a manual investigation in field and on the other side preventing false
sysrq detections.

As a preparation to add sysrq_toggle_support() call into uart,
remove a private copy of sysrq_enabled from sysctl - it should reflect
the actual status of sysrq.

Furthermore, the private copy isn't correct already in case
sysrq_always_enabled is true. So, remove __sysrq_enabled and use a
getter-helper for sysrq enabled status.

Cc: Iurii Zaikin <yzaikin@google.com>
Cc: Jiri Slaby <jslaby@suse.com>
Cc: Luis Chamberlain <mcgrof@kernel.org>
Cc: Kees Cook <keescook@chromium.org>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Dmitry Safonov <dima@arista.com>
---
 drivers/tty/sysrq.c   |  7 +++++++
 include/linux/sysrq.h |  1 +
 kernel/sysctl.c       | 41 ++++++++++++++++++++++-------------------
 3 files changed, 30 insertions(+), 19 deletions(-)

diff --git a/drivers/tty/sysrq.c b/drivers/tty/sysrq.c
index f724962a5906..ef3e78967146 100644
--- a/drivers/tty/sysrq.c
+++ b/drivers/tty/sysrq.c
@@ -73,6 +73,13 @@ static bool sysrq_on_mask(int mask)
 	       (sysrq_enabled & mask);
 }
 
+int sysrq_get_mask(void)
+{
+	if (sysrq_always_enabled)
+		return 1;
+	return sysrq_enabled;
+}
+
 static int __init sysrq_always_enabled_setup(char *str)
 {
 	sysrq_always_enabled = true;
diff --git a/include/linux/sysrq.h b/include/linux/sysrq.h
index 8c71874e8485..4a0b351fa2d3 100644
--- a/include/linux/sysrq.h
+++ b/include/linux/sysrq.h
@@ -50,6 +50,7 @@ int unregister_sysrq_key(int key, struct sysrq_key_op *op);
 struct sysrq_key_op *__sysrq_get_key_op(int key);
 
 int sysrq_toggle_support(int enable_mask);
+int sysrq_get_mask(void);
 
 #else
 
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index d396aaaf19a3..6ddb4d7df0e1 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -229,25 +229,8 @@ static int proc_dopipe_max_size(struct ctl_table *table, int write,
 		void __user *buffer, size_t *lenp, loff_t *ppos);
 
 #ifdef CONFIG_MAGIC_SYSRQ
-/* Note: sysrq code uses its own private copy */
-static int __sysrq_enabled = CONFIG_MAGIC_SYSRQ_DEFAULT_ENABLE;
-
 static int sysrq_sysctl_handler(struct ctl_table *table, int write,
-				void __user *buffer, size_t *lenp,
-				loff_t *ppos)
-{
-	int error;
-
-	error = proc_dointvec(table, write, buffer, lenp, ppos);
-	if (error)
-		return error;
-
-	if (write)
-		sysrq_toggle_support(__sysrq_enabled);
-
-	return 0;
-}
-
+			void __user *buffer, size_t *lenp, loff_t *ppos);
 #endif
 
 static struct ctl_table kern_table[];
@@ -747,7 +730,7 @@ static struct ctl_table kern_table[] = {
 #ifdef CONFIG_MAGIC_SYSRQ
 	{
 		.procname	= "sysrq",
-		.data		= &__sysrq_enabled,
+		.data		= NULL,
 		.maxlen		= sizeof (int),
 		.mode		= 0644,
 		.proc_handler	= sysrq_sysctl_handler,
@@ -2844,6 +2827,26 @@ static int proc_dostring_coredump(struct ctl_table *table, int write,
 }
 #endif
 
+#ifdef CONFIG_MAGIC_SYSRQ
+static int sysrq_sysctl_handler(struct ctl_table *table, int write,
+				void __user *buffer, size_t *lenp, loff_t *ppos)
+{
+	int tmp, ret;
+
+	tmp = sysrq_get_mask();
+
+	ret = __do_proc_dointvec(&tmp, table, write, buffer,
+			       lenp, ppos, NULL, NULL);
+	if (ret || !write)
+		return ret;
+
+	if (write)
+		sysrq_toggle_support(tmp);
+
+	return 0;
+}
+#endif
+
 static int __do_proc_doulongvec_minmax(void *data, struct ctl_table *table, int write,
 				     void __user *buffer,
 				     size_t *lenp, loff_t *ppos,
-- 
2.24.1

Re: [PATCH-next 2/3] sysctl/sysrq: Remove __sysrq_enabled copy

[Iurii Zaikin]

> Furthermore, the private copy isn't correct already in case
> sysrq_always_enabled is true. So, remove __sysrq_enabled and use a
> getter-helper for sysrq enabled status.
Since the old way was known to be unreliable, the new way looks like a
great candidate for a KUnit test.

Off topic: I wonder if Magic Sysrq could be extended with reasonable
effort to be triggered by a sequence of keystrokes which would be less
likely to be generated by a noisy serial.

Re: [PATCH-next 2/3] sysctl/sysrq: Remove __sysrq_enabled copy

[Greg Kroah-Hartman]

On Thu, Jan 09, 2020 at 09:54:43PM +0000, Dmitry Safonov wrote:
> Many embedded boards have a disconnected TTL level serial which can
> generate some garbage that can lead to spurious false sysrq detects.
> 
> Currently, sysrq can be either completely disabled for serial console
> or always disabled (with CONFIG_MAGIC_SYSRQ_SERIAL), since
> commit 732dbf3a6104 ("serial: do not accept sysrq characters via serial port")
> 
> At Arista, we have such boards that can generate BREAK and random
> garbage. While disabling sysrq for serial console would solve
> the problem with spurious false sysrq triggers, it's also desirable
> to have a way to enable sysrq back.
> 
> Having the way to enable sysrq was beneficial to debug lockups with
> a manual investigation in field and on the other side preventing false
> sysrq detections.
> 
> As a preparation to add sysrq_toggle_support() call into uart,
> remove a private copy of sysrq_enabled from sysctl - it should reflect
> the actual status of sysrq.
> 
> Furthermore, the private copy isn't correct already in case
> sysrq_always_enabled is true. So, remove __sysrq_enabled and use a
> getter-helper for sysrq enabled status.
> 
> Cc: Iurii Zaikin <yzaikin@google.com>
> Cc: Jiri Slaby <jslaby@suse.com>
> Cc: Luis Chamberlain <mcgrof@kernel.org>
> Cc: Kees Cook <keescook@chromium.org>
> Cc: linux-fsdevel@vger.kernel.org
> Signed-off-by: Dmitry Safonov <dima@arista.com>
> ---
>  drivers/tty/sysrq.c   |  7 +++++++
>  include/linux/sysrq.h |  1 +
>  kernel/sysctl.c       | 41 ++++++++++++++++++++++-------------------
>  3 files changed, 30 insertions(+), 19 deletions(-)
> 
> diff --git a/drivers/tty/sysrq.c b/drivers/tty/sysrq.c
> index f724962a5906..ef3e78967146 100644
> --- a/drivers/tty/sysrq.c
> +++ b/drivers/tty/sysrq.c
> @@ -73,6 +73,13 @@ static bool sysrq_on_mask(int mask)
>  	       (sysrq_enabled & mask);
>  }
>  
> +int sysrq_get_mask(void)
> +{
> +	if (sysrq_always_enabled)
> +		return 1;
> +	return sysrq_enabled;
> +}
> +
>  static int __init sysrq_always_enabled_setup(char *str)
>  {
>  	sysrq_always_enabled = true;
> diff --git a/include/linux/sysrq.h b/include/linux/sysrq.h
> index 8c71874e8485..4a0b351fa2d3 100644
> --- a/include/linux/sysrq.h
> +++ b/include/linux/sysrq.h
> @@ -50,6 +50,7 @@ int unregister_sysrq_key(int key, struct sysrq_key_op *op);
>  struct sysrq_key_op *__sysrq_get_key_op(int key);
>  
>  int sysrq_toggle_support(int enable_mask);
> +int sysrq_get_mask(void);
>  
>  #else
>  
> diff --git a/kernel/sysctl.c b/kernel/sysctl.c
> index d396aaaf19a3..6ddb4d7df0e1 100644
> --- a/kernel/sysctl.c
> +++ b/kernel/sysctl.c
> @@ -229,25 +229,8 @@ static int proc_dopipe_max_size(struct ctl_table *table, int write,
>  		void __user *buffer, size_t *lenp, loff_t *ppos);
>  
>  #ifdef CONFIG_MAGIC_SYSRQ
> -/* Note: sysrq code uses its own private copy */
> -static int __sysrq_enabled = CONFIG_MAGIC_SYSRQ_DEFAULT_ENABLE;
> -
>  static int sysrq_sysctl_handler(struct ctl_table *table, int write,
> -				void __user *buffer, size_t *lenp,
> -				loff_t *ppos)
> -{
> -	int error;
> -
> -	error = proc_dointvec(table, write, buffer, lenp, ppos);
> -	if (error)
> -		return error;
> -
> -	if (write)
> -		sysrq_toggle_support(__sysrq_enabled);
> -
> -	return 0;
> -}
> -
> +			void __user *buffer, size_t *lenp, loff_t *ppos);
>  #endif
>  
>  static struct ctl_table kern_table[];
> @@ -747,7 +730,7 @@ static struct ctl_table kern_table[] = {
>  #ifdef CONFIG_MAGIC_SYSRQ
>  	{
>  		.procname	= "sysrq",
> -		.data		= &__sysrq_enabled,
> +		.data		= NULL,
>  		.maxlen		= sizeof (int),
>  		.mode		= 0644,
>  		.proc_handler	= sysrq_sysctl_handler,
> @@ -2844,6 +2827,26 @@ static int proc_dostring_coredump(struct ctl_table *table, int write,
>  }
>  #endif
>  
> +#ifdef CONFIG_MAGIC_SYSRQ
> +static int sysrq_sysctl_handler(struct ctl_table *table, int write,
> +				void __user *buffer, size_t *lenp, loff_t *ppos)
> +{
> +	int tmp, ret;
> +
> +	tmp = sysrq_get_mask();
> +
> +	ret = __do_proc_dointvec(&tmp, table, write, buffer,
> +			       lenp, ppos, NULL, NULL);
> +	if (ret || !write)
> +		return ret;
> +
> +	if (write)
> +		sysrq_toggle_support(tmp);
> +
> +	return 0;
> +}
> +#endif

Why did you move this function down here?  Can't it stay where it is and
you can just fix the logic there?  Now you have two different #ifdef
blocks intead of just one :(

thanks,

greg k-h

Re: [PATCH-next 2/3] sysctl/sysrq: Remove __sysrq_enabled copy

[Dmitry Safonov]

Hi Greg,

On 1/10/20 4:40 PM, Greg Kroah-Hartman wrote:
> On Thu, Jan 09, 2020 at 09:54:43PM +0000, Dmitry Safonov wrote:
[..]
>> @@ -2844,6 +2827,26 @@ static int proc_dostring_coredump(struct ctl_table *table, int write,
>>  }
>>  #endif
>>  
>> +#ifdef CONFIG_MAGIC_SYSRQ
>> +static int sysrq_sysctl_handler(struct ctl_table *table, int write,
>> +				void __user *buffer, size_t *lenp, loff_t *ppos)
>> +{
>> +	int tmp, ret;
>> +
>> +	tmp = sysrq_get_mask();
>> +
>> +	ret = __do_proc_dointvec(&tmp, table, write, buffer,
>> +			       lenp, ppos, NULL, NULL);
>> +	if (ret || !write)
>> +		return ret;
>> +
>> +	if (write)
>> +		sysrq_toggle_support(tmp);
>> +
>> +	return 0;
>> +}
>> +#endif
> 
> Why did you move this function down here?  Can't it stay where it is and
> you can just fix the logic there?  Now you have two different #ifdef
> blocks intead of just one :(

Yeah, well __do_proc_dointvec() made me do it.

sysrq_sysctl_handler() declaration should be before ctl_table array of
sysctls, so I couldn't remove the forward-declaration.

So, I could forward-declare __do_proc_dointvec() instead, but looking at
the neighborhood, I decided to follow the file-style (there is a couple
of forward-declarations before the sysctl array, some under ifdefs).

I admit that the result is imperfect and can put __do_proc_dointvec()
definition before instead, no hard feelings.

Thanks,
          Dmitry

Re: [PATCH-next 2/3] sysctl/sysrq: Remove __sysrq_enabled copy

[Dmitry Safonov]

Hi Iurii,

On 1/10/20 3:19 AM, Iurii Zaikin wrote:
> Off topic: I wonder if Magic Sysrq could be extended with reasonable
> effort to be triggered by a sequence of keystrokes which would be less
> likely to be generated by a noisy serial.

That's what I do in 3/3 patch from the series:
https://lkml.kernel.org/r/20200109215444.95995-4-dima@arista.com

Or do you mean something else?

Thanks,
          Dmitry

Re: [PATCH-next 2/3] sysctl/sysrq: Remove __sysrq_enabled copy

[Greg Kroah-Hartman]

On Fri, Jan 10, 2020 at 09:45:30PM +0000, Dmitry Safonov wrote:
> Hi Greg,
> 
> On 1/10/20 4:40 PM, Greg Kroah-Hartman wrote:
> > On Thu, Jan 09, 2020 at 09:54:43PM +0000, Dmitry Safonov wrote:
> [..]
> >> @@ -2844,6 +2827,26 @@ static int proc_dostring_coredump(struct ctl_table *table, int write,
> >>  }
> >>  #endif
> >>  
> >> +#ifdef CONFIG_MAGIC_SYSRQ
> >> +static int sysrq_sysctl_handler(struct ctl_table *table, int write,
> >> +				void __user *buffer, size_t *lenp, loff_t *ppos)
> >> +{
> >> +	int tmp, ret;
> >> +
> >> +	tmp = sysrq_get_mask();
> >> +
> >> +	ret = __do_proc_dointvec(&tmp, table, write, buffer,
> >> +			       lenp, ppos, NULL, NULL);
> >> +	if (ret || !write)
> >> +		return ret;
> >> +
> >> +	if (write)
> >> +		sysrq_toggle_support(tmp);
> >> +
> >> +	return 0;
> >> +}
> >> +#endif
> > 
> > Why did you move this function down here?  Can't it stay where it is and
> > you can just fix the logic there?  Now you have two different #ifdef
> > blocks intead of just one :(
> 
> Yeah, well __do_proc_dointvec() made me do it.
> 
> sysrq_sysctl_handler() declaration should be before ctl_table array of
> sysctls, so I couldn't remove the forward-declaration.
> 
> So, I could forward-declare __do_proc_dointvec() instead, but looking at
> the neighborhood, I decided to follow the file-style (there is a couple
> of forward-declarations before the sysctl array, some under ifdefs).
> 
> I admit that the result is imperfect and can put __do_proc_dointvec()
> definition before instead, no hard feelings.

Ah, no, nevermind, I missed that reason, sorry about that.  Moving it is
fine.

greg k-h