On Sun, Oct 16, 2022 at 05:22:57AM +0000, Klemens Nanni wrote:
> On Sat, Oct 08, 2022 at 07:25:26PM +0000, Klemens Nanni wrote:
> > If /usr is mounted read-only, kernel relinking fails silently without
> > any log trace.
> >
> > # ksh /usr/libexec/reorder_kernel
> > /usr/libexec/reorder_kernel[35]: cannot create
> > /usr/share/relink/kernel/GENERIC.MP/relink.log: Read-only file system
> >
> > This stderr line does not show up during boot because init(8) redirects
> > stdout/err to /dev/null, executes rc(8) which inherits this and executes
> > reorder_kernel with both streams thus discarded.
> >
> > Eventually, I'd like reorder_kernel to temporarily remount $KERNEL_DIR
> > read-write, just like rc's reorder_libs() does. I have a diff for that
> > but would like to approach things in smaller steps.
> >
> > So first, defer the logfile setup after the syslog/console setup such
> > that mkdir's failure triggers the ERR trap and users get notified:
> >
> > # ksh ./reorder_kernel.sh
> > ./reorder_kernel.sh[40]: cannot create
> > /usr/share/relink/kernel/GENERIC.MP/relink.log: Read-only file system
> > # tail -n1 /var/log/daemon
> > Oct 13 17:04:06 pbp reorder_kernel.sh: kernel relinking done
> >
> > As of now, this logs success despite actual failure because of ksh's
> > redirection/trap bug (see mail on tech@): mkdir does not fail because
> > the directory already exists, so `exec 1>$LOGFILE' tries to truncate but
> > fails to trigger the ERR trap, thus hitting the non-default EXIT trap.
> >
> > With ksh fixed this would log
> > Oct 13 17:04:06 pbp reorder_kernel.sh: failed -- see
> > /usr/share/relink/kernel/GENERIC.MP/relink.log
> > which admittedly misleading, given that creating/truncating that vary
> > logfile is the failure reason, but it seems better than silent nothing.
>
> ksh's exec got fixed, so failure is now logged with this diff:
>
> # tail -n0 -f /var/log/messages
> [1] 95776
> # /usr/libexec/reorder_kernel
> /usr/libexec/reorder_kernel[35]: cannot create
> /usr/share/relink/kernel/GENERIC.MP/relink.log: Read-only file system
> no log except stderr
>
> # ksh ./reorder_kernel.sh
> ./reorder_kernel.sh[40]: cannot create
> /usr/share/relink/kernel/GENERIC.MP/relink.log: Read-only file system
> Oct 16 09:16:37 eru reorder_kernel.sh: failed -- see
> /usr/share/relink/kernel/GENERIC.MP/relink.log
> same "failed" message in xconsole(1)
>
> > It could set up an intermediate trap or truly silently bail out on
> > read-only FFS just like it does for NFS now, but it seems best to handle
> > ro /usr as mentioned as a follow-up step.
>
> Anyone?
>
> This diff now works as expected; not perfect but an improvement, imho.
> Feedback? Objection? OK?
No feedback so far.
Supporting /usr on read-only FFS or not, I think reorder_kernel must not
fail silently without relinking anything.
There is definitely room for improvement.
OK?
Index: libexec/reorder_kernel/reorder_kernel.sh
===================================================================
RCS file: /cvs/src/libexec/reorder_kernel/reorder_kernel.sh,v
retrieving revision 1.11
diff -u -p -r1.11 reorder_kernel.sh
--- libexec/reorder_kernel/reorder_kernel.sh 13 May 2022 13:20:16 -0000
1.11
+++ libexec/reorder_kernel/reorder_kernel.sh 10 Oct 2022 00:15:58 -0000
@@ -30,15 +30,15 @@ LOGFILE=$KERNEL_DIR/$KERNEL/relink.log
PROGNAME=${0##*/}
SHA256=/var/db/kernel.SHA256
-# Create kernel compile dir and redirect stdout/stderr to a logfile.
-mkdir -m 700 -p $KERNEL_DIR/$KERNEL
-exec 1>$LOGFILE
-exec 2>&1
-
# Install trap handlers to inform about success or failure via syslog.
trap 'trap - EXIT; logger -st $PROGNAME \
"failed -- see $LOGFILE" >>/dev/console 2>&1' ERR
trap 'logger -t $PROGNAME "kernel relinking done"' EXIT
+
+# Create kernel compile dir and redirect stdout/stderr to a logfile.
+mkdir -m 700 -p $KERNEL_DIR/$KERNEL
+exec 1>$LOGFILE
+exec 2>&1
if [[ -f $KERNEL_DIR.tgz ]]; then
rm -rf $KERNEL_DIR/$KERNEL/*
