I stole the printk_once() macro.

I always wanted to be able to print some error directly if there is a
buffer to dump, however we can't use error_report() where the code path
can be triggered by DDOS attack.  To avoid that, we can introduce a
print-once-like function for it.  Meanwhile, we also introduce the
corresponding helper for warn_report().

CC: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
---
 include/qemu/error-report.h | 26 ++++++++++++++++++++++++++
 1 file changed, 26 insertions(+)

diff --git a/include/qemu/error-report.h b/include/qemu/error-report.h
index e1c8ae1a52..3e6e84801f 100644
--- a/include/qemu/error-report.h
+++ b/include/qemu/error-report.h
@@ -44,6 +44,32 @@ void error_report(const char *fmt, ...) GCC_FMT_ATTR(1, 2);
 void warn_report(const char *fmt, ...) GCC_FMT_ATTR(1, 2);
 void info_report(const char *fmt, ...) GCC_FMT_ATTR(1, 2);
 
+/* Similar to error_report(), but it only prints the message once. */
+#define error_report_once(fmt, ...)             \
+    ({                                          \
+        static bool __print_once;               \
+        bool __ret_print_once = !__print_once;  \
+                                                \
+        if (!__print_once) {                    \
+            __print_once = true;                \
+            error_report(fmt, ##__VA_ARGS__);   \
+        }                                       \
+        unlikely(__ret_print_once);             \
+    })
+
+/* Similar to warn_report(), but it only prints the message once. */
+#define warn_report_once(fmt, ...)              \
+    ({                                          \
+        static bool __print_once;               \
+        bool __ret_print_once = !__print_once;  \
+                                                \
+        if (!__print_once) {                    \
+            __print_once = true;                \
+            warn_report(fmt, ##__VA_ARGS__);   \
+        }                                       \
+        unlikely(__ret_print_once);             \
+    })
+
 const char *error_get_progname(void);
 extern bool enable_timestamp_msg;
 
-- 
2.17.0

On 05/21/2018 10:56 PM, Peter Xu wrote:
> I stole the printk_once() macro.
> 
> I always wanted to be able to print some error directly if there is a
> buffer to dump, however we can't use error_report() where the code path
> can be triggered by DDOS attack.  To avoid that, we can introduce a
> print-once-like function for it.  Meanwhile, we also introduce the
> corresponding helper for warn_report().
> 
> CC: Markus Armbruster <armbru@redhat.com>
> Signed-off-by: Peter Xu <peterx@redhat.com>
> ---
>   include/qemu/error-report.h | 26 ++++++++++++++++++++++++++
>   1 file changed, 26 insertions(+)
> 
> diff --git a/include/qemu/error-report.h b/include/qemu/error-report.h
> index e1c8ae1a52..3e6e84801f 100644
> --- a/include/qemu/error-report.h
> +++ b/include/qemu/error-report.h
> @@ -44,6 +44,32 @@ void error_report(const char *fmt, ...) GCC_FMT_ATTR(1, 2);
>   void warn_report(const char *fmt, ...) GCC_FMT_ATTR(1, 2);
>   void info_report(const char *fmt, ...) GCC_FMT_ATTR(1, 2);
>   
> +/* Similar to error_report(), but it only prints the message once. */
> +#define error_report_once(fmt, ...)             \
> +    ({                                          \
> +        static bool __print_once;               \

You're still using the reserved namespace.  I asked you to fix that on 
v1; it's disappointing when a later revision doesn't make changes 
requested by a reviewers, without also offering a counter-argument for 
why the request does not need to be considered.

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org

On Wed, May 23, 2018 at 10:47:29AM -0500, Eric Blake wrote:
> On 05/21/2018 10:56 PM, Peter Xu wrote:
> > I stole the printk_once() macro.
> > 
> > I always wanted to be able to print some error directly if there is a
> > buffer to dump, however we can't use error_report() where the code path
> > can be triggered by DDOS attack.  To avoid that, we can introduce a
> > print-once-like function for it.  Meanwhile, we also introduce the
> > corresponding helper for warn_report().
> > 
> > CC: Markus Armbruster <armbru@redhat.com>
> > Signed-off-by: Peter Xu <peterx@redhat.com>
> > ---
> >   include/qemu/error-report.h | 26 ++++++++++++++++++++++++++
> >   1 file changed, 26 insertions(+)
> > 
> > diff --git a/include/qemu/error-report.h b/include/qemu/error-report.h
> > index e1c8ae1a52..3e6e84801f 100644
> > --- a/include/qemu/error-report.h
> > +++ b/include/qemu/error-report.h
> > @@ -44,6 +44,32 @@ void error_report(const char *fmt, ...) GCC_FMT_ATTR(1, 2);
> >   void warn_report(const char *fmt, ...) GCC_FMT_ATTR(1, 2);
> >   void info_report(const char *fmt, ...) GCC_FMT_ATTR(1, 2);
> > +/* Similar to error_report(), but it only prints the message once. */
> > +#define error_report_once(fmt, ...)             \
> > +    ({                                          \
> > +        static bool __print_once;               \
> 
> You're still using the reserved namespace.  I asked you to fix that on v1;
> it's disappointing when a later revision doesn't make changes requested by a
> reviewers, without also offering a counter-argument for why the request does
> not need to be considered.

I am sorry.  It's totally because I forgot that one!  (It rarely
happens since I note them down in most cases, but there must be
something that even interrupted the procedure when I took notes...)

I will repost with another one soon.  Sorry again.

-- 
Peter Xu