Patchwork block: do_commit() does not pass along error messages for all errors

login
register
mail settings
Submitter Jeff Cody
Date Jan. 16, 2013, 11:04 p.m.
Message ID <9a76a255b7b5e09131dbd8d7503d07b489d7f8d0.1358373940.git.jcody@redhat.com>
Download mbox | patch
Permalink /patch/213089/
State New
Headers show

Comments

Jeff Cody - Jan. 16, 2013, 11:04 p.m.
The non-live bdrv_commit() function may return one of the following
errors: -ENOTSUP, -EBUSY, -EACCES, -EIO.  The only error that is
checked in the HMP handler is -EBUSY, so the monitor command 'commit'
silently fails for all error cases other than 'Device is in use'.

This patch adds the appropriate error messages for the errors
explicitely returned by bdrv_commit().

Signed-off-by: Jeff Cody <jcody@redhat.com>
---
 blockdev.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)
Markus Armbruster - Jan. 17, 2013, 7:49 a.m.
[Cc: Luiz for error stuff]

Jeff Cody <jcody@redhat.com> writes:

> The non-live bdrv_commit() function may return one of the following
> errors: -ENOTSUP, -EBUSY, -EACCES, -EIO.  The only error that is
> checked in the HMP handler is -EBUSY, so the monitor command 'commit'
> silently fails for all error cases other than 'Device is in use'.
>
> This patch adds the appropriate error messages for the errors
> explicitely returned by bdrv_commit().
>
> Signed-off-by: Jeff Cody <jcody@redhat.com>
> ---
>  blockdev.c | 14 ++++++++++++++
>  1 file changed, 14 insertions(+)
>
> diff --git a/blockdev.c b/blockdev.c
> index d724e2d..7db7d8e 100644
> --- a/blockdev.c
> +++ b/blockdev.c
> @@ -657,6 +657,20 @@ void do_commit(Monitor *mon, const QDict *qdict)
>              qerror_report(QERR_DEVICE_IN_USE, device);
>              return;
>          }
> +        if (ret == -EACCES) {
> +            qerror_report(QERR_DEVICE_IS_READ_ONLY, device);
> +            return;
> +        }
> +        if (ret == -EIO) {
> +            qerror_report(QERR_IO_ERROR);
> +            return;
> +        }
> +        if (ret == -ENOTSUP) {
> +            const char *format = bdrv_get_format_name(bs);
> +            qerror_report(QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
> +                          format ? format : "NULL", device, "commit");
> +
> +        }
>      }
>  }

A switch could be more obvious than the if cascade.  Matter of taste.

We're trying to avoid QERR_ in new code.  This case isn't bad, because
it doesn't add new QERR_ defines.  Should we convert the function to
error_setg() instead?  Perhaps a few cases could be collapsed into a
single error_setg_errno() then.
Luiz Capitulino - Jan. 17, 2013, 11:57 a.m.
On Thu, 17 Jan 2013 08:49:38 +0100
Markus Armbruster <armbru@redhat.com> wrote:

> [Cc: Luiz for error stuff]
> 
> Jeff Cody <jcody@redhat.com> writes:
> 
> > The non-live bdrv_commit() function may return one of the following
> > errors: -ENOTSUP, -EBUSY, -EACCES, -EIO.  The only error that is
> > checked in the HMP handler is -EBUSY, so the monitor command 'commit'
> > silently fails for all error cases other than 'Device is in use'.
> >
> > This patch adds the appropriate error messages for the errors
> > explicitely returned by bdrv_commit().
> >
> > Signed-off-by: Jeff Cody <jcody@redhat.com>
> > ---
> >  blockdev.c | 14 ++++++++++++++
> >  1 file changed, 14 insertions(+)
> >
> > diff --git a/blockdev.c b/blockdev.c
> > index d724e2d..7db7d8e 100644
> > --- a/blockdev.c
> > +++ b/blockdev.c
> > @@ -657,6 +657,20 @@ void do_commit(Monitor *mon, const QDict *qdict)
> >              qerror_report(QERR_DEVICE_IN_USE, device);
> >              return;
> >          }
> > +        if (ret == -EACCES) {
> > +            qerror_report(QERR_DEVICE_IS_READ_ONLY, device);
> > +            return;
> > +        }
> > +        if (ret == -EIO) {
> > +            qerror_report(QERR_IO_ERROR);
> > +            return;
> > +        }
> > +        if (ret == -ENOTSUP) {
> > +            const char *format = bdrv_get_format_name(bs);
> > +            qerror_report(QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
> > +                          format ? format : "NULL", device, "commit");
> > +
> > +        }
> >      }
> >  }
> 
> A switch could be more obvious than the if cascade.  Matter of taste.
> 
> We're trying to avoid QERR_ in new code.  This case isn't bad, because
> it doesn't add new QERR_ defines.  Should we convert the function to
> error_setg() instead?  Perhaps a few cases could be collapsed into a
> single error_setg_errno() then.

I'd suggest to convert do_commit() to the qapi. However, we already
have block-commit in QMP. So I'm not sure this is a good idea.

If it isn't, then maybe we could drop all qerro_report() calls and
use monitor_printf() instead (building the error message with
strerror()), as do_commit() seems to be used only from HMP.
Jeff Cody - Jan. 18, 2013, 2:27 p.m.
On Thu, Jan 17, 2013 at 09:57:28AM -0200, Luiz Capitulino wrote:
> On Thu, 17 Jan 2013 08:49:38 +0100
> Markus Armbruster <armbru@redhat.com> wrote:
> 
> > [Cc: Luiz for error stuff]
> > 
> > Jeff Cody <jcody@redhat.com> writes:
> > 
> > > The non-live bdrv_commit() function may return one of the following
> > > errors: -ENOTSUP, -EBUSY, -EACCES, -EIO.  The only error that is
> > > checked in the HMP handler is -EBUSY, so the monitor command 'commit'
> > > silently fails for all error cases other than 'Device is in use'.
> > >
> > > This patch adds the appropriate error messages for the errors
> > > explicitely returned by bdrv_commit().
> > >
> > > Signed-off-by: Jeff Cody <jcody@redhat.com>
> > > ---
> > >  blockdev.c | 14 ++++++++++++++
> > >  1 file changed, 14 insertions(+)
> > >
> > > diff --git a/blockdev.c b/blockdev.c
> > > index d724e2d..7db7d8e 100644
> > > --- a/blockdev.c
> > > +++ b/blockdev.c
> > > @@ -657,6 +657,20 @@ void do_commit(Monitor *mon, const QDict *qdict)
> > >              qerror_report(QERR_DEVICE_IN_USE, device);
> > >              return;
> > >          }
> > > +        if (ret == -EACCES) {
> > > +            qerror_report(QERR_DEVICE_IS_READ_ONLY, device);
> > > +            return;
> > > +        }
> > > +        if (ret == -EIO) {
> > > +            qerror_report(QERR_IO_ERROR);
> > > +            return;
> > > +        }
> > > +        if (ret == -ENOTSUP) {
> > > +            const char *format = bdrv_get_format_name(bs);
> > > +            qerror_report(QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
> > > +                          format ? format : "NULL", device, "commit");
> > > +
> > > +        }
> > >      }
> > >  }
> > 
> > A switch could be more obvious than the if cascade.  Matter of taste.
> > 
> > We're trying to avoid QERR_ in new code.  This case isn't bad, because
> > it doesn't add new QERR_ defines.  Should we convert the function to
> > error_setg() instead?  Perhaps a few cases could be collapsed into a
> > single error_setg_errno() then.
> 
> I'd suggest to convert do_commit() to the qapi. However, we already
> have block-commit in QMP. So I'm not sure this is a good idea.
> 
> If it isn't, then maybe we could drop all qerro_report() calls and
> use monitor_printf() instead (building the error message with
> strerror()), as do_commit() seems to be used only from HMP.

I think I agree in keeping do_commit() HMP only.  I'll submit a v2 to
just use monitor_printf() for all error strings, rather than
qerror_report().

Thanks,
Jeff

Patch

diff --git a/blockdev.c b/blockdev.c
index d724e2d..7db7d8e 100644
--- a/blockdev.c
+++ b/blockdev.c
@@ -657,6 +657,20 @@  void do_commit(Monitor *mon, const QDict *qdict)
             qerror_report(QERR_DEVICE_IN_USE, device);
             return;
         }
+        if (ret == -EACCES) {
+            qerror_report(QERR_DEVICE_IS_READ_ONLY, device);
+            return;
+        }
+        if (ret == -EIO) {
+            qerror_report(QERR_IO_ERROR);
+            return;
+        }
+        if (ret == -ENOTSUP) {
+            const char *format = bdrv_get_format_name(bs);
+            qerror_report(QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
+                          format ? format : "NULL", device, "commit");
+
+        }
     }
 }