On Wed, Jun 17, 2015 at 07:13:17PM +0200, Michal Židek wrote:
> Hi,
>
> this patch partially solves the ticket
>
https://fedorahosted.org/sssd/ticket/2590
>
> It does not subscribe to the systemd dbus interface and does only
> the polling (this would have to be done anyway for the case
> the interface is not available).
>
> I will investigate how to add the systemd/dbus check as well,
> but the attached patch should decrease priority of the ticket.
>
>
> Thanks,
> Michal
> From bb57b2bcba891cd395d795a7706cf2ea77f8bca2 Mon Sep 17 00:00:00 2001
> From: =?UTF-8?q?Michal=20=C5=BDidek?= <mzidek(a)redhat.com>
> Date: Wed, 17 Jun 2015 18:54:14 +0200
> Subject: [PATCH] MONITOR: Poll for resolv.conf if not available during boot
>
> If resolv.conf is not available when SSSD is starting,
> check for its existence later.
>
> Ticket:
>
https://fedorahosted.org/sssd/ticket/2590
> ---
> src/monitor/monitor.c | 38 ++++++++++++++++++++++++++++++++++++--
> 1 file changed, 36 insertions(+), 2 deletions(-)
>
> diff --git a/src/monitor/monitor.c b/src/monitor/monitor.c
> index 01c882a..24fca28 100644
> --- a/src/monitor/monitor.c
> +++ b/src/monitor/monitor.c
> @@ -2259,11 +2259,39 @@ static errno_t monitor_config_file_fallback(TALLOC_CTX
*mem_ctx,
> return EOK;
> }
>
> +#define MISSING_RESOLV_CONF_POLL_TIME 10
> +
> +static void missing_resolv_conf(struct tevent_context *ev,
> + struct tevent_timer *te,
> + struct timeval tv, void *data)
> +{
> + int ret;
> + struct mt_ctx *ctx = talloc_get_type(data, struct mt_ctx);
> +
> + ret = monitor_config_file(ctx, ctx, RESOLV_CONF_PATH,
> + monitor_update_resolv, false);
> + if (ret == ENOENT) {
> + tv = tevent_timeval_current_ofs(MISSING_RESOLV_CONF_POLL_TIME, 0);
> + te = tevent_add_timer(ctx->ev, ctx, tv, missing_resolv_conf, ctx);
> + if (te == NULL) {
> + DEBUG(SSSDBG_FATAL_FAILURE,
> + "tevent_add_timer failed. resolv.conf will be
ignored.\n");
> + }
> + } else if (ret != EOK) {
> + DEBUG(SSSDBG_FATAL_FAILURE,
> + "Monitor_config_file failed. resolv.conf will be
ignored.\n");
> + }
> +
I think we should add signal_res_init() here, because in my testing the
inotify watch is established if I mv a resolv.conf file into place, but
the backends remain offline.
Yes, sorry for that. It worked during my testing because I had something
in /etc/hosts, so the missing resolv.conf was shadowed.
We should also (in a different patch) change the DEBUG levels in
monitor_config_file, because currently we print a 0x0010-level message.
Ok, but I think we should special case resolv.conf here. Otherwise
the missing (and not ignored) config file should always be reported.