[libvirt] [PATCH] libvirtd: fix potential deadlock when starting vm

owen.si at ucloud.cn owen.si at ucloud.cn
Wed Oct 17 02:19:52 UTC 2018


Yes, this is CentOS 6 only, The monitor socket is by default set to NONBLOCK with CentOS 7.

I compare the udev code on centos6 and centos7,

in udev-147 on centos6:
udev_monitor->sock = socket(PF_NETLINK, SOCK_DGRAM|SOCK_CLOEXEC, NETLINK_KOBJECT_UEVENT);

in systemd-219 on centos7:
udev_monitor->sock = socket(PF_NETLINK, SOCK_RAW|SOCK_CLOEXEC|SOCK_NONBLOCK, NETLINK_KOBJECT_UEVENT);


owen.si at ucloud.cn
 
From: John Ferlan
Date: 2018-10-17 05:57
To: Bingsong Si; libvir-list
CC: Erik Skultety
Subject: Re: [libvirt] [PATCH] libvirtd: fix potential deadlock when starting vm
 
 
On 10/11/18 4:13 AM, Bingsong Si wrote:
> On CentOS 6, udev_monitor_receive_device will block until the socket becomes
 
Is this really CentOS6 only or just where you've seen it?
 
> readable, udevEventHandleThread will hold the lock all the time and
> udevEventHandleCallback hard to get the lock, will block the event poll.
> To fix this, set dataReady to false after receive an udev event.
> 
> Signed-off-by: Bingsong Si <owen.si at ucloud.cn>
> ---
>  src/node_device/node_device_udev.c | 5 +----
>  1 file changed, 1 insertion(+), 4 deletions(-)
> 
 
I've CC'd Erik since he wrote and perhaps remembers all the "gotchas" he
discovered in the udev callback code.
 
I wonder if this has to do with the EAGAIN and EWOULDBLOCK @errno checks
done in the !device loop that are different in "older" (much older) code.
 
Although I have this very vague recollection that there was some problem
with centos6 that was fixed by some OS patch.  Hopefully Erik remembers
(and maybe we should log it in the code at this point ;-)) - I did do
some searching, but came up empty.
 
John
 
> diff --git a/src/node_device/node_device_udev.c b/src/node_device/node_device_udev.c
> index 22897591de..ce1101d7cc 100644
> --- a/src/node_device/node_device_udev.c
> +++ b/src/node_device/node_device_udev.c
> @@ -1616,6 +1616,7 @@ udevEventHandleThread(void *opaque ATTRIBUTE_UNUSED)
>  
>          errno = 0;
>          device = udev_monitor_receive_device(priv->udev_monitor);
> +        priv->dataReady = false;
>          virObjectUnlock(priv);
>  
>          if (!device) {
> @@ -1637,10 +1638,6 @@ udevEventHandleThread(void *opaque ATTRIBUTE_UNUSED)
>                  return;
>              }
>  
> -            virObjectLock(priv);
> -            priv->dataReady = false;
> -            virObjectUnlock(priv);
> -
>              continue;
>          }
>  
> 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/libvir-list/attachments/20181017/d3c20aec/attachment-0001.htm>


More information about the libvir-list mailing list