[pve-devel] [RFC PATCH common] sysfstools: file_write: properly catch errors

[pve-devel] [RFC PATCH common] sysfstools: file_write: properly catch errors

[Dominik Csapak]

since `print` is doing buffered IO, we don't always get an error there,
even if the underlying write does not work.

To properly catch that, do an unbuffered `syswrite` which circumvents
all buffers and writes directly to the file handle.

We aren't actually interested in the specific error here, but only if
the write was successful or not.

Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
---

Note: this is heavily used when doing PCI passthrough, and it seems we
did not catch as many errors as we thought. Maybe we should introduce an
'noerr' parameter and pass that on all current code paths, since i fear
that there are many situations where the sysfs write had failed but it
still works because we ignored it and qemu/the kernel do the right thing
anyway.

 src/PVE/SysFSTools.pm | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/PVE/SysFSTools.pm b/src/PVE/SysFSTools.pm
index 57f0ac8..8eb9f2e 100644
--- a/src/PVE/SysFSTools.pm
+++ b/src/PVE/SysFSTools.pm
@@ -198,7 +198,7 @@ sub file_write {
     my $fh = IO::File->new($filename, "w");
     return undef if !$fh;
 
-    my $res = print $fh $buf;
+    my $res = defined(syswrite($fh, $buf)) ? 1 : 0;
 
     $fh->close();
 
-- 
2.39.2



_______________________________________________
pve-devel mailing list
pve-devel@lists.proxmox.com
https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel

Re: [pve-devel] [RFC PATCH common] sysfstools: file_write: properly catch errors

[Christoph Heiss]

Tested this together with the Nvidia vGPU series [0], so please consider
this also:

Tested-by: Christoph Heiss <c.heiss@proxmox.com>

[0] https://lore.proxmox.com/pve-devel/ul3faxwxsi45hk4e44cl2fw4jomoncy53owx3you2wuvo7pymf@dchu53bqfa3p/T/#m2526c3a031aacc2f52080977febd432b4ab2efcf

On Tue, Jul 23, 2024 at 10:29:25AM GMT, Dominik Csapak wrote:
> since `print` is doing buffered IO, we don't always get an error there,
> even if the underlying write does not work.
>
> To properly catch that, do an unbuffered `syswrite` which circumvents
> all buffers and writes directly to the file handle.
>
> We aren't actually interested in the specific error here, but only if
> the write was successful or not.
>
> Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
> ---
>
> Note: this is heavily used when doing PCI passthrough, and it seems we
> did not catch as many errors as we thought. Maybe we should introduce an
> 'noerr' parameter and pass that on all current code paths, since i fear
> that there are many situations where the sysfs write had failed but it
> still works because we ignored it and qemu/the kernel do the right thing
> anyway.

Actually catching errors here is definitely the right thing to do, IMHO.
But as for not breaking existing callers, maybe a `warn` statement would
be sensible?

That way broken call-sites can be caught in the future and fixed as
necessary. So a `noerr` parameter and `warn` if that is unset.

>
>  src/PVE/SysFSTools.pm | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/src/PVE/SysFSTools.pm b/src/PVE/SysFSTools.pm
> index 57f0ac8..8eb9f2e 100644
> --- a/src/PVE/SysFSTools.pm
> +++ b/src/PVE/SysFSTools.pm
> @@ -198,7 +198,7 @@ sub file_write {
>      my $fh = IO::File->new($filename, "w");
>      return undef if !$fh;
>
> -    my $res = print $fh $buf;
> +    my $res = defined(syswrite($fh, $buf)) ? 1 : 0;
>
>      $fh->close();
>
> --
> 2.39.2
>
>
>
> _______________________________________________
> pve-devel mailing list
> pve-devel@lists.proxmox.com
> https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
>
>


_______________________________________________
pve-devel mailing list
pve-devel@lists.proxmox.com
https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel

[pve-devel] applied: [RFC PATCH common] sysfstools: file_write: properly catch errors

[Thomas Lamprecht]

Am 23/07/2024 um 10:29 schrieb Dominik Csapak:
> since `print` is doing buffered IO, we don't always get an error there,
> even if the underlying write does not work.
> 
> To properly catch that, do an unbuffered `syswrite` which circumvents
> all buffers and writes directly to the file handle.
> 
> We aren't actually interested in the specific error here, but only if
> the write was successful or not.
> 
> Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
> ---
> 
> Note: this is heavily used when doing PCI passthrough, and it seems we
> did not catch as many errors as we thought. Maybe we should introduce an
> 'noerr' parameter and pass that on all current code paths, since i fear
> that there are many situations where the sysfs write had failed but it
> still works because we ignored it and qemu/the kernel do the right thing
> anyway.
> 

I also see some regression potential, but IMO it's something that would
be very good to know, and so erroring out explicitly for now is IMO worth
it. We can then adapt this as needed on actual feedback.

>  src/PVE/SysFSTools.pm | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
>

applied, with Christoph's T-b thanks!


_______________________________________________
pve-devel mailing list
pve-devel@lists.proxmox.com
https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel