[pve-devel] [PATCH qemu-server v2 2/3] qmeventd: cancel 'forced cleanup' when normal cleanup succeeds

[Wolfgang Bumiller]

On Thu, Sep 22, 2022 at 04:19:34PM +0200, Dominik Csapak wrote:
> instead of always sending a SIGKILL to the target pid.
> It was not that much of a problem since the timeout previously was 5
> seconds and we used pifds where possible, thus the chance of killing the
> wrong process was rather slim.
> 
> Now we increased the timeout to 60s which makes the race a bit more likely
> (when not using pidfds), so remove it from the 'forced_cleanups' list when
> the normal cleanup succeeds.
> 
> Signed-off-by: Dominik Csapak <d.csapak at proxmox.com>
> ---
>  qmeventd/qmeventd.c | 17 +++++++++++++++++
>  1 file changed, 17 insertions(+)
> 
> diff --git a/qmeventd/qmeventd.c b/qmeventd/qmeventd.c
> index 46bc7eb..eebc19d 100644
> --- a/qmeventd/qmeventd.c
> +++ b/qmeventd/qmeventd.c
> @@ -416,6 +416,22 @@ cleanup_qemu_client(struct Client *client)
>      }
>  }
>  
> +static void
> +remove_cleanup_data(struct CleanupData *data, struct Client *client) {
> +    if (data->pid == client->pid) {
> +	forced_cleanups = g_slist_remove(forced_cleanups, data);
> +	free(data);
> +    }
> +}
> +
> +static void
> +remove_from_forced_cleanup(struct Client *client) {
> +    if (g_slist_length(forced_cleanups) > 0) {
> +	VERBOSE_PRINT("removing %s from forced cleanups\n", client->qemu.vmid);
> +	g_slist_foreach(forced_cleanups, (GFunc)remove_cleanup_data, client);

Foreach + remove feels awkward to me. Sure, it's a linked list and
should be fine™, but I don't like the lack of documentation of
interactions here as a non-glib user. (I mean, eg. for C++ iterator
invalidation is *usually* documented...)

Can't we just give `struct Client` a `struct CleanupData` pointer and
call `g_slist_remove` right here without the iteration?

Or better yet, your previous idea of dropping `CleanupData` sounds
better.
We should be able to just add `struct Client*` to the list, after all,
according to the glib docs `g_slist_remove` should simply leave the list
unchanged if the data is not part of the list, so when we free up the
`Client` we could even call `g_slist_remove` unconditionally (though
we'll know whether it's in there by having a timeout set then...)

(or use `g_slist_find_custom`)

> +    }
> +}
> +
>  void
>  cleanup_client(struct Client *client)
>  {
> @@ -442,6 +458,7 @@ cleanup_client(struct Client *client)
>  	    break;
>      }
>  
> +    remove_from_forced_cleanup(client);
>      free(client);
>  }
>  
> -- 
> 2.30.2