On 11.04.21 09:17, Henrik Riomar wrote:
> On Sun, 11 Apr 2021 07:34:16 +0200
> Juergen Gross <jgross@suse.com> wrote:
> 
>> On 11.04.21 00:02, Henrik Riomar wrote:
>>> Hi,
>>>
>>> In Alpine and Debian (probably elsewhere as well), the -s flag removed in these two commits:
>>>    * https://github.com/xen-project/xen/commit/fa06cb8c38832aafe597d719040ba1d216e367b8
>>>    * https://github.com/xen-project/xen/commit/c65687ed16d2289ec91036ec2862a4b4bd34ea4f
>>> is actually used in the init scripts.
> 
>>> Reverting the two commits mentioned above restores a booting system.
>>>
>>> The -s flag was introduced in 2005 or so, so I guess you can say that dropping it should
>>> at least be mentioned in the release notices, and an alternative be described, or
>>> -s functionally should be brought back.
>>
>> The -s served exactly no purpose.
> 
> yes its used by dists to check that the socket is actually accessible (without falling back to
> direct access to /dev/xen/xenbus).

There are Xen configurations where the socket is never accessible,
as it is not existing.

>> It was meant to force socket usage. A socket will be used anyway in
>> case xenstored is running in dom0, so specifying -s won't change
>> anything in this case.
> 
> yes reverting the to commits above and using -s, brings back the original behavior, returning
> with failure if the socket is not there.

And returning failure when Xenstore is running fine, but not in dom0.

> There are two issues here I think:
>   1. dists actually use -s to check if the daemon is up (and been doing this for a long time)

This should be changed, as it is based on wrong assumptions.

>   2. Reads of /dev/xen/xenbus (via xenstore-read -s /), blocks for ever in 4.15

So you are basically saying that you'd like to have a test "is Xenstore
running", and this test should work with the "-s" flag.

I can look into that, but reverting the "access via socket" removal flag
is not the way to go.


Juergen