Re: stty blocks forever when the line is already opened (and the tx buffer can't be flushed)

Re: stty blocks forever when the line is already opened (and the tx buffer can't be flushed)

[Richard Genoud]

[ sorry for the noise, I forgot to Cc the lkml ]

Hi,
I've found a case were calling
stty -F /dev/ttyS1 clocal
blocks forever.
And I don't know if it's a very old bug or if it's meant to be like that.

Here is how to reproduce the lock :
NB: there's NO modem on ttyS1
stty -F /dev/ttyS1 clocal cread crtscts
cat < /dev/ttyS1

#on another terminal :
echo "dummy" > /dev/ttyS1  # This call doesn't block

stty -F /dev/ttyS1 -crtscts # this blocks forever on ioctl(TCSETSW )


looking at tty_port_close_start(), it's pretty clear that nothing is
flushed until the last user, so it explains why the "echo dummy"
returns directly, despite the crtscts flags.
And in tty_mode_ioctl(), there are the lines:
        case TCSETSW:
                return set_termios(real_tty, p, TERMIOS_WAIT | TERMIOS_OLD);
That explain why the stty blocks.

But this behavior seems really strange.
... Or it's meant to be like that ?


Regards,
Richard

NB: This is actually a real life use case with mgetty, a modem losing
its power and another process trying to speak to the modem.

Re: stty blocks forever when the line is already opened (and the tx buffer can't be flushed)

[Peter Hurley]

Hi Richard,

On 01/26/2016 08:19 AM, Richard Genoud wrote:
> [ sorry for the noise, I forgot to Cc the lkml ]
> 
> Hi,
> I've found a case were calling
> stty -F /dev/ttyS1 clocal
> blocks forever.
> And I don't know if it's a very old bug or if it's meant to be like that.
> 
> Here is how to reproduce the lock :
> NB: there's NO modem on ttyS1
> stty -F /dev/ttyS1 clocal cread crtscts
> cat < /dev/ttyS1
> 
> #on another terminal :
> echo "dummy" > /dev/ttyS1  # This call doesn't block
> 
> stty -F /dev/ttyS1 -crtscts # this blocks forever on ioctl(TCSETSW )
> 
> 
> looking at tty_port_close_start(), it's pretty clear that nothing is
> flushed until the last user, so it explains why the "echo dummy"
> returns directly, despite the crtscts flags.
> And in tty_mode_ioctl(), there are the lines:
>         case TCSETSW:
>                 return set_termios(real_tty, p, TERMIOS_WAIT | TERMIOS_OLD);
> That explain why the stty blocks.
> 
> But this behavior seems really strange.
> ... Or it's meant to be like that ?

Yeah, meant to be like that.

When mgetty writes the login prompt but h/w flow control is enabled
and nothing's connected, the output is buffered.

Since stty uses tcsetattr(TCSADRAIN), the attempt to turn off h/w flow
control blocks, waiting for output to empty.

In this situation, stopping mgetty will allow the other process
to unblock and advance.

Hmmm, I could add a -f,--force flag to stty so it uses tcsetattr(TCSANOW)...

Regards,
Peter Hurley


> Regards,
> Richard
> 
> NB: This is actually a real life use case with mgetty, a modem losing
> its power and another process trying to speak to the modem.
> 

Re: stty blocks forever when the line is already opened (and the tx buffer can't be flushed)

[Richard Genoud]

2016-01-26 18:13 GMT+01:00 Peter Hurley <peter@hurleysoftware.com>:
> Hi Richard,
>
> On 01/26/2016 08:19 AM, Richard Genoud wrote:
>> [ sorry for the noise, I forgot to Cc the lkml ]
>>
>> Hi,
>> I've found a case were calling
>> stty -F /dev/ttyS1 clocal
>> blocks forever.
>> And I don't know if it's a very old bug or if it's meant to be like that.
>>
>> Here is how to reproduce the lock :
>> NB: there's NO modem on ttyS1
>> stty -F /dev/ttyS1 clocal cread crtscts
>> cat < /dev/ttyS1
>>
>> #on another terminal :
>> echo "dummy" > /dev/ttyS1  # This call doesn't block
>>
>> stty -F /dev/ttyS1 -crtscts # this blocks forever on ioctl(TCSETSW )
>>
>>
>> looking at tty_port_close_start(), it's pretty clear that nothing is
>> flushed until the last user, so it explains why the "echo dummy"
>> returns directly, despite the crtscts flags.
>> And in tty_mode_ioctl(), there are the lines:
>>         case TCSETSW:
>>                 return set_termios(real_tty, p, TERMIOS_WAIT | TERMIOS_OLD);
>> That explain why the stty blocks.
>>
>> But this behavior seems really strange.
>> ... Or it's meant to be like that ?
>
> Yeah, meant to be like that.
>
> When mgetty writes the login prompt but h/w flow control is enabled
> and nothing's connected, the output is buffered.
>
> Since stty uses tcsetattr(TCSADRAIN), the attempt to turn off h/w flow
> control blocks, waiting for output to empty.
Damn ! I forgot about the TCSADRAIN/FLUSH/NOW flags

> In this situation, stopping mgetty will allow the other process
> to unblock and advance.
>
> Hmmm, I could add a -f,--force flag to stty so it uses tcsetattr(TCSANOW)...
Yes, I think such a flag will make sense.

Thank for your enlightenment !

Richard.

Re: stty blocks forever when the line is already opened (and the tx buffer can't be flushed)

[Richard Genoud]

2016-01-27 10:14 GMT+01:00 Richard Genoud <richard.genoud@gmail.com>:
> 2016-01-26 18:13 GMT+01:00 Peter Hurley <peter@hurleysoftware.com>:
>> Hi Richard,
>>
>> On 01/26/2016 08:19 AM, Richard Genoud wrote:
>>> [ sorry for the noise, I forgot to Cc the lkml ]
>>>
>>> Hi,
>>> I've found a case were calling
>>> stty -F /dev/ttyS1 clocal
>>> blocks forever.
>>> And I don't know if it's a very old bug or if it's meant to be like that.
>>>
>>> Here is how to reproduce the lock :
>>> NB: there's NO modem on ttyS1
>>> stty -F /dev/ttyS1 clocal cread crtscts
>>> cat < /dev/ttyS1
>>>
>>> #on another terminal :
>>> echo "dummy" > /dev/ttyS1  # This call doesn't block
>>>
>>> stty -F /dev/ttyS1 -crtscts # this blocks forever on ioctl(TCSETSW )
>>>
>>>
>>> looking at tty_port_close_start(), it's pretty clear that nothing is
>>> flushed until the last user, so it explains why the "echo dummy"
>>> returns directly, despite the crtscts flags.
>>> And in tty_mode_ioctl(), there are the lines:
>>>         case TCSETSW:
>>>                 return set_termios(real_tty, p, TERMIOS_WAIT | TERMIOS_OLD);
>>> That explain why the stty blocks.
>>>
>>> But this behavior seems really strange.
>>> ... Or it's meant to be like that ?
>>
>> Yeah, meant to be like that.
>>
>> When mgetty writes the login prompt but h/w flow control is enabled
>> and nothing's connected, the output is buffered.
>>
>> Since stty uses tcsetattr(TCSADRAIN), the attempt to turn off h/w flow
>> control blocks, waiting for output to empty.
> Damn ! I forgot about the TCSADRAIN/FLUSH/NOW flags
>
>> In this situation, stopping mgetty will allow the other process
>> to unblock and advance.
>>
>> Hmmm, I could add a -f,--force flag to stty so it uses tcsetattr(TCSANOW)...
> Yes, I think such a flag will make sense.
In fact, a similar flag exist in the very last version of coreutils (8.25)
http://git.savannah.gnu.org/cgit/coreutils.git/commit/?id=cb7d8b90a213c0186a8c8ba66da959e1f5930e78
:)

Re: stty blocks forever when the line is already opened (and the tx buffer can't be flushed)

[Peter Hurley]

On 01/27/2016 02:26 AM, Richard Genoud wrote:
> 2016-01-27 10:14 GMT+01:00 Richard Genoud <richard.genoud@gmail.com>:
>> 2016-01-26 18:13 GMT+01:00 Peter Hurley <peter@hurleysoftware.com>:
>>> Hi Richard,
>>>
>>> On 01/26/2016 08:19 AM, Richard Genoud wrote:
>>>> [ sorry for the noise, I forgot to Cc the lkml ]
>>>>
>>>> Hi,
>>>> I've found a case were calling
>>>> stty -F /dev/ttyS1 clocal
>>>> blocks forever.
>>>> And I don't know if it's a very old bug or if it's meant to be like that.
>>>>
>>>> Here is how to reproduce the lock :
>>>> NB: there's NO modem on ttyS1
>>>> stty -F /dev/ttyS1 clocal cread crtscts
>>>> cat < /dev/ttyS1
>>>>
>>>> #on another terminal :
>>>> echo "dummy" > /dev/ttyS1  # This call doesn't block
>>>>
>>>> stty -F /dev/ttyS1 -crtscts # this blocks forever on ioctl(TCSETSW )
>>>>
>>>>
>>>> looking at tty_port_close_start(), it's pretty clear that nothing is
>>>> flushed until the last user, so it explains why the "echo dummy"
>>>> returns directly, despite the crtscts flags.
>>>> And in tty_mode_ioctl(), there are the lines:
>>>>         case TCSETSW:
>>>>                 return set_termios(real_tty, p, TERMIOS_WAIT | TERMIOS_OLD);
>>>> That explain why the stty blocks.
>>>>
>>>> But this behavior seems really strange.
>>>> ... Or it's meant to be like that ?
>>>
>>> Yeah, meant to be like that.
>>>
>>> When mgetty writes the login prompt but h/w flow control is enabled
>>> and nothing's connected, the output is buffered.
>>>
>>> Since stty uses tcsetattr(TCSADRAIN), the attempt to turn off h/w flow
>>> control blocks, waiting for output to empty.
>> Damn ! I forgot about the TCSADRAIN/FLUSH/NOW flags
>>
>>> In this situation, stopping mgetty will allow the other process
>>> to unblock and advance.
>>>
>>> Hmmm, I could add a -f,--force flag to stty so it uses tcsetattr(TCSANOW)...
>> Yes, I think such a flag will make sense.
> In fact, a similar flag exist in the very last version of coreutils (8.25)
> http://git.savannah.gnu.org/cgit/coreutils.git/commit/?id=cb7d8b90a213c0186a8c8ba66da959e1f5930e78
> :)
> 

Thanks for saving me the time doing it myself.

Regards,
Peter Hurley