New: [rhinstaller/anaconda/pulls/181 master] dracut: fix boot with inst.ks and no inst.{repo,stage2} (#1238987)
Okay so: commit 4883b96 moved fetch-kickstart-net.sh from the online hook into initqueue (or initqueue/settled) to fix inst.ks.sendmac.
So when you boot with inst.ks=[URL], the order goes:
1. Network comes up 2. Run `online` hook * Schedule kickstart fetch in initqueue * Other online tasks from boot args (e.g. updates) 3. Run `initqueue` * fetch + run kickstart * add `anaconda-netroot.sh` (etc.) to `online` hook * `udevadm trigger` disk devices * `udevadm trigger` net devices
The problem is, triggering network devices is *not* sufficient to re-run the `online` hook - this only happens when the NIC actually gets _configured_, not when the device is triggered.
The fix is pretty straightforward: re-run the `online` hook for active NICs after we run the kickstart, in order to pick up any new tasks that the kickstart might have scheduled.
Resolves: RHBZ#1238987
From: Will Woods wwoods@redhat.com
Okay so: commit 4883b96 moved fetch-kickstart-net.sh from the online hook into initqueue (or initqueue/settled) to fix inst.ks.sendmac.
So when you boot with inst.ks=[URL], the order goes:
1. Network comes up 2. Run `online` hook * Schedule kickstart fetch in initqueue * Other online tasks from boot args (e.g. updates) 3. Run `initqueue` * fetch + run kickstart * add `anaconda-netroot.sh` (etc.) to `online` hook * `udevadm trigger` disk devices * `udevadm trigger` net devices
The problem is, triggering network devices is *not* sufficient to re-run the `online` hook - this only happens when the NIC actually gets _configured_, not when the device is triggered.
The fix is pretty straightforward: re-run the `online` hook for active NICs after we run the kickstart, in order to pick up any new tasks that the kickstart might have scheduled.
Resolves: RHBZ#1238987 --- dracut/anaconda-lib.sh | 20 ++++++++++++++++++-- 1 file changed, 18 insertions(+), 2 deletions(-)
diff --git a/dracut/anaconda-lib.sh b/dracut/anaconda-lib.sh index c28a28e..0b1679c 100755 --- a/dracut/anaconda-lib.sh +++ b/dracut/anaconda-lib.sh @@ -213,6 +213,15 @@ parse_kickstart() { [ -e "$parsed_kickstart" ] && cp $parsed_kickstart /run/install/ks.cfg }
+# print a list of net devices that dracut says are set up. +online_netdevs() { + local netif="" + for netif in /tmp/net.*.did-setup; do + netif=${netif#*.}; netif=${netif%.*} + [ -d "/sys/class/net/$netif" ] && echo $netif + done +} + # This is where we actually run the kickstart. Whee! # We can't just add udev rules (we'll miss devices that are already active), # and we can't just run the scripts manually (we'll miss devices that aren't @@ -236,7 +245,7 @@ run_kickstart() { # re-parse new cmdline stuff from the kickstart . $hookdir/cmdline/*parse-anaconda-repo.sh . $hookdir/cmdline/*parse-livenet.sh - # TODO: parse for other stuff ks might set (dd? other stuff?) + . $hookdir/cmdline/*parse-anaconda-dd.sh case "$repotype" in http*|ftp|nfs*) do_net=1 ;; cdrom|hd|bd) do_disk=1 ;; @@ -266,14 +275,21 @@ run_kickstart() { rm /tmp/dd_args_ks fi
- # replay udev events to trigger actions + # disk: replay udev events to trigger actions if [ "$do_disk" ]; then + # set up new rules . $hookdir/pre-trigger/*repo-genrules.sh udevadm control --reload + # trigger the rules for all the block devices we see udevadm trigger --action=change --subsystem-match=block fi + + # net: re-run online hook if [ "$do_net" ]; then udevadm trigger --action=change --subsystem-match=net + for netif in $(online_netdevs); do + source_hook initqueue/online $netif + done fi
# and that's it - we're back to the mainloop.
Added label: master.
Added label: rhel7-branch.
anaconda-patches@lists.fedorahosted.org
-
dashea