qemu-io: fix EOF Ctrl-D handling in qemu-io readline code

Message ID 20171206115726.5237-1-berrange@redhat.com
State New
Headers show
Series
  • qemu-io: fix EOF Ctrl-D handling in qemu-io readline code
Related show

Commit Message

Daniel P. Berrange Dec. 6, 2017, 11:57 a.m.
qemu-io puts the TTY into non-canonical mode, which means no EOF processing is
done and thus getchar() will never return the EOF constant. Instead we have to
check for an explicit Ctrl-D, aka  0x4, to detect EOF and exit the qemu-io
shell. This fixes the regression that prevented Ctrl-D from triggering an exit
of qemu-io that has existed since readline was first added in

  commit 0cf17e181798063c3824c8200ba46f25f54faa1a
  Author: Stefan Hajnoczi <stefanha@redhat.com>
  Date:   Thu Nov 14 11:54:17 2013 +0100

    qemu-io: use readline.c

Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
---
 qemu-io.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

Comments

Eric Blake Dec. 6, 2017, 2:22 p.m. | #1
On 12/06/2017 05:57 AM, Daniel P. Berrange wrote:
> qemu-io puts the TTY into non-canonical mode, which means no EOF processing is
> done and thus getchar() will never return the EOF constant. Instead we have to
> check for an explicit Ctrl-D, aka  0x4, to detect EOF and exit the qemu-io
> shell. This fixes the regression that prevented Ctrl-D from triggering an exit
> of qemu-io that has existed since readline was first added in
> 
>   commit 0cf17e181798063c3824c8200ba46f25f54faa1a
>   Author: Stefan Hajnoczi <stefanha@redhat.com>
>   Date:   Thu Nov 14 11:54:17 2013 +0100
> 
>     qemu-io: use readline.c
> 
> Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
> ---
>  qemu-io.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)

>      while (!line) {
>          int ch = getchar();
> -        if (ch == EOF) {
> +        /* In non-canon tty mode we get 0x4 (Ctrl-D), not the stdio "EOF"
> +         * constant */
> +        if (ch == 0x4) {

Should we instead be looking for a match against the current termios()
c_cc[VEOF] setting, in case the user prefers something other than ^D via
stty?  Does readline provide any functionality for automating this?
Daniel P. Berrange Dec. 6, 2017, 2:25 p.m. | #2
On Wed, Dec 06, 2017 at 08:22:35AM -0600, Eric Blake wrote:
> On 12/06/2017 05:57 AM, Daniel P. Berrange wrote:
> > qemu-io puts the TTY into non-canonical mode, which means no EOF processing is
> > done and thus getchar() will never return the EOF constant. Instead we have to
> > check for an explicit Ctrl-D, aka  0x4, to detect EOF and exit the qemu-io
> > shell. This fixes the regression that prevented Ctrl-D from triggering an exit
> > of qemu-io that has existed since readline was first added in
> > 
> >   commit 0cf17e181798063c3824c8200ba46f25f54faa1a
> >   Author: Stefan Hajnoczi <stefanha@redhat.com>
> >   Date:   Thu Nov 14 11:54:17 2013 +0100
> > 
> >     qemu-io: use readline.c
> > 
> > Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
> > ---
> >  qemu-io.c | 4 +++-
> >  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> >      while (!line) {
> >          int ch = getchar();
> > -        if (ch == EOF) {
> > +        /* In non-canon tty mode we get 0x4 (Ctrl-D), not the stdio "EOF"
> > +         * constant */
> > +        if (ch == 0x4) {
> 
> Should we instead be looking for a match against the current termios()
> c_cc[VEOF] setting, in case the user prefers something other than ^D via
> stty?  Does readline provide any functionality for automating this?

I was afraid someone was going to suggest doing that. I was being lazy by
hardcoding Ctrl-D, but yes the real readline() library will honour the
VEOF value.  QEMU though is using a home-grown reimpl of readline...

Regards,
Daniel
Eric Blake Dec. 6, 2017, 2:30 p.m. | #3
On 12/06/2017 08:25 AM, Daniel P. Berrange wrote:

>> Should we instead be looking for a match against the current termios()
>> c_cc[VEOF] setting, in case the user prefers something other than ^D via
>> stty?  Does readline provide any functionality for automating this?
> 
> I was afraid someone was going to suggest doing that. I was being lazy by
> hardcoding Ctrl-D, but yes the real readline() library will honour the
> VEOF value.  QEMU though is using a home-grown reimpl of readline...

I understand why we can't use modern libreadline (it is licensed GPLv3+,
making it impossible to use with our GPLv2-only code) - but doesn't BSD
have a readline-alike library with BSD licensing that we could use
instead of implementing readline ourselves?  (Was it named libinput?)
Markus Armbruster Dec. 6, 2017, 4:14 p.m. | #4
Eric Blake <eblake@redhat.com> writes:

> On 12/06/2017 08:25 AM, Daniel P. Berrange wrote:
>
>>> Should we instead be looking for a match against the current termios()
>>> c_cc[VEOF] setting, in case the user prefers something other than ^D via
>>> stty?  Does readline provide any functionality for automating this?
>> 
>> I was afraid someone was going to suggest doing that. I was being lazy by
>> hardcoding Ctrl-D, but yes the real readline() library will honour the
>> VEOF value.  QEMU though is using a home-grown reimpl of readline...
>
> I understand why we can't use modern libreadline (it is licensed GPLv3+,
> making it impossible to use with our GPLv2-only code) - but doesn't BSD
> have a readline-alike library with BSD licensing that we could use
> instead of implementing readline ourselves?  (Was it named libinput?)

We've discussed this before.  Short story: we can't use the original due
to our foolish licensing mistake, and the clones don't cut the mustard.
Long story:
http://lists.gnu.org/archive/html/qemu-devel/2016-10/msg02697.html
Dr. David Alan Gilbert Dec. 8, 2017, 12:15 p.m. | #5
* Daniel P. Berrange (berrange@redhat.com) wrote:
> qemu-io puts the TTY into non-canonical mode, which means no EOF processing is
> done and thus getchar() will never return the EOF constant. Instead we have to
> check for an explicit Ctrl-D, aka  0x4, to detect EOF and exit the qemu-io
> shell. This fixes the regression that prevented Ctrl-D from triggering an exit
> of qemu-io that has existed since readline was first added in
> 
>   commit 0cf17e181798063c3824c8200ba46f25f54faa1a
>   Author: Stefan Hajnoczi <stefanha@redhat.com>
>   Date:   Thu Nov 14 11:54:17 2013 +0100
> 
>     qemu-io: use readline.c
> 
> Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
> ---
>  qemu-io.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/qemu-io.c b/qemu-io.c
> index c70bde3eb1..2ea0bfbaf8 100644
> --- a/qemu-io.c
> +++ b/qemu-io.c
> @@ -322,7 +322,9 @@ static char *fetchline_readline(void)
>      readline_start(readline_state, get_prompt(), 0, readline_func, &line);
>      while (!line) {
>          int ch = getchar();
> -        if (ch == EOF) {
> +        /* In non-canon tty mode we get 0x4 (Ctrl-D), not the stdio "EOF"
> +         * constant */
> +        if (ch == 0x4) {

Personally I'd have made that EOF or 0x4  - but that's fine

(I don't see the point of reading the ioctl to figure out which EOF
char we're using; it seems to turn a trivial check into something much
more complex)

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

>              break;
>          }
>          readline_handle_byte(readline_state, ch);
> -- 
> 2.14.3
> 
> 
--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK
Daniel P. Berrange Dec. 8, 2017, 1:35 p.m. | #6
On Fri, Dec 08, 2017 at 12:15:21PM +0000, Dr. David Alan Gilbert wrote:
> * Daniel P. Berrange (berrange@redhat.com) wrote:
> > qemu-io puts the TTY into non-canonical mode, which means no EOF processing is
> > done and thus getchar() will never return the EOF constant. Instead we have to
> > check for an explicit Ctrl-D, aka  0x4, to detect EOF and exit the qemu-io
> > shell. This fixes the regression that prevented Ctrl-D from triggering an exit
> > of qemu-io that has existed since readline was first added in
> > 
> >   commit 0cf17e181798063c3824c8200ba46f25f54faa1a
> >   Author: Stefan Hajnoczi <stefanha@redhat.com>
> >   Date:   Thu Nov 14 11:54:17 2013 +0100
> > 
> >     qemu-io: use readline.c
> > 
> > Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
> > ---
> >  qemu-io.c | 4 +++-
> >  1 file changed, 3 insertions(+), 1 deletion(-)
> > 
> > diff --git a/qemu-io.c b/qemu-io.c
> > index c70bde3eb1..2ea0bfbaf8 100644
> > --- a/qemu-io.c
> > +++ b/qemu-io.c
> > @@ -322,7 +322,9 @@ static char *fetchline_readline(void)
> >      readline_start(readline_state, get_prompt(), 0, readline_func, &line);
> >      while (!line) {
> >          int ch = getchar();
> > -        if (ch == EOF) {
> > +        /* In non-canon tty mode we get 0x4 (Ctrl-D), not the stdio "EOF"
> > +         * constant */
> > +        if (ch == 0x4) {
> 
> Personally I'd have made that EOF or 0x4  - but that's fine

I thought about that, but it is impossible to get 'EOF' when the terminal
is in raw mode, so there's little point.

> (I don't see the point of reading the ioctl to figure out which EOF
> char we're using; it seems to turn a trivial check into something much
> more complex)

I'd already done the work to read termios settings by time I read this
comment, so I've sent a v2 anyway :-)

> Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

Regards,
Daniel

Patch

diff --git a/qemu-io.c b/qemu-io.c
index c70bde3eb1..2ea0bfbaf8 100644
--- a/qemu-io.c
+++ b/qemu-io.c
@@ -322,7 +322,9 @@  static char *fetchline_readline(void)
     readline_start(readline_state, get_prompt(), 0, readline_func, &line);
     while (!line) {
         int ch = getchar();
-        if (ch == EOF) {
+        /* In non-canon tty mode we get 0x4 (Ctrl-D), not the stdio "EOF"
+         * constant */
+        if (ch == 0x4) {
             break;
         }
         readline_handle_byte(readline_state, ch);