Sorry for my laziness and the low effort to understand a thread I gave
a link :(.
To prove I can give a more effort to this thing I sit down to change
os-prober code to comply with grub-mount feature (symlink behaviour).
A first level of problem is duplicating output of the same kernel/initrd
filename pairs. Duplicates appears because the patterns tried are not
disjunctive at the line 15 of /usr/lib/linux-boot-probes/mounted/90fallback.
for kernpat in /vmlinuz /vmlinux /boot/vmlinuz /boot/vmlinux "/boot/vmlinuz*" \
"/boot/vmlinux*" "/vmlinuz*" "/vmlinux*" "/kernel-*" "/boot/kernel-*"; do
I don't know why there are pattern pairs /vmlinuz and "/vmlinuz*" and so
on. Shell wild-card "/vmlinuz*" is expanded in the alphabetical order,
so potential /vmlinuz alone is expanded always first if exists. I'm not
certain the reasoning... aha in between is /boot/vmlinuz, so if both
exists /vmlinuz and /boot/vmlinuz, they appear in that order
(/boot/vmlinuz appears before /vmlinuz<something>).
...initrd patterns at the line 27...:
initrdname=$(echo "$kernfile" | sed "s/vmlinu[zx]/initrd\*/")
# Yellow Dog Linux appends .img to it.
initrdname1="${initrdname}.img"
The following patch removes duplicates from expanded patterns of kernels and
initrd images:
--- /usr/lib/linux-boot-probes/mounted/90fallback~ 2012-11-18 20:40:05.050334822 +0100
+++ /usr/lib/linux-boot-probes/mounted/90fallback 2012-11-18 21:16:50.846268406 +0100
@@ -21,6 +21,8 @@
fi
for kernfile in $(eval ls "$mpoint$kernpat" 2>/dev/null); do
kernbasefile=$(echo "$kernfile" | sed "s!^$mpoint!!")
+ echo "$kernfile_seen" | fgrep -q " $kernbasefile " && continue
+ kernfile_seen="$kernfile_seen $kernbasefile "
if [ -f "$kernfile" ] && [ ! -L "$kernfile" ]; then
initrdname=$(echo "$kernfile" | sed "s/vmlinu[zx]/initrd\*/")
# Yellow Dog Linux appends .img to it.
@@ -35,7 +37,7 @@
# And Gentoo's also
initrdname4=$(echo "$kernfile" | sed "s/kernel/initramfs\*/")
foundinitrd=0
- for initrd in $(eval ls "$initrdname" "$initrdname1" "$initrdname2" "$initrdname3" "$initrdname4" 2>/dev/null); do
+ for initrd in $(eval ls "$initrdname" "$initrdname1" "$initrdname2" "$initrdname3" "$initrdname4" 2>/dev/null | sort | uniq); do
if [ "$initrd" != "$kernfile" ] && [ -f "$initrd" ] && [ ! -L "$initrd" ]; then
initrd=$(echo "$initrd" | sed "s!^$mpoint!!")
result "$partition:$kernbootpart::$kernbasefile:$initrd:root=$mappedpartition"
Before the patch:
bobek:~# /usr/lib/linux-boot-probes/50mounted-tests /dev/mapper/vg-deboot
/dev/mapper/vg-deboot:/dev/mapper/vg-deboot::/vmlinuz:/initrd.img:root=/dev/dm-23
/dev/mapper/vg-deboot:/dev/mapper/vg-deboot::/vmlinuz:/initrd.img:root=/dev/dm-23
/dev/mapper/vg-deboot:/dev/mapper/vg-deboot::/boot/vmlinuz-3.2.0-4-amd64:/boot/initrd.img-3.2.0-4-amd64:root=/dev/dm-23
/dev/mapper/vg-deboot:/dev/mapper/vg-deboot::/vmlinuz:/initrd.img:root=/dev/dm-23
/dev/mapper/vg-deboot:/dev/mapper/vg-deboot::/vmlinuz:/initrd.img:root=/dev/dm-23
After the patch:
bobek:~# /usr/lib/linux-boot-probes/50mounted-tests /dev/mapper/vg-deboot
/dev/mapper/vg-deboot:/dev/mapper/vg-deboot::/vmlinuz:/initrd.img:root=/dev/dm-23
/dev/mapper/vg-deboot:/dev/mapper/vg-deboot::/boot/vmlinuz-3.2.0-4-amd64:/boot/initrd.img-3.2.0-4-amd64:root=/dev/dm-23
I hope the above change is worth doing anyway.
We can go further and try to identify the identical file's with
different filenames. (This problem is introduced by grub-mount behaviour
on symlinks.)
--- /usr/lib/linux-boot-probes/mounted/90fallback~zito1 2012-11-19 01:04:58.817863969 +0100
+++ /usr/lib/linux-boot-probes/mounted/90fallback 2012-11-19 01:49:23.941785223 +0100
@@ -9,6 +9,11 @@
mpoint="$3"
type="$4"
+getmd5()
+{
+ md5sum "$1" | cut -f1 -d ' '
+}
+
mappedpartition=$(mapdevfs "$partition" 2>/dev/null) || mappedpartition="$partition"
exitcode=1
@@ -24,6 +29,7 @@
echo "$kernfile_seen" | fgrep -q " $kernbasefile " && continue
kernfile_seen="$kernfile_seen $kernbasefile "
if [ -f "$kernfile" ] && [ ! -L "$kernfile" ]; then
+ kernmd5=$(getmd5 "$kernfile")
initrdname=$(echo "$kernfile" | sed "s/vmlinu[zx]/initrd\*/")
# Yellow Dog Linux appends .img to it.
initrdname1="${initrdname}.img"
@@ -39,6 +45,13 @@
foundinitrd=0
for initrd in $(eval ls "$initrdname" "$initrdname1" "$initrdname2" "$initrdname3" "$initrdname4" 2>/dev/null | sort | uniq); do
if [ "$initrd" != "$kernfile" ] && [ -f "$initrd" ] && [ ! -L "$initrd" ]; then
+ initrdmd5=$(getmd5 "$initrd")
+ pairmd5="$kernmd5:$initrdmd5"
+ if echo "$pairmd5_seen" | fgrep -q " $pairmd5 "; then
+ foundinitrd=1
+ continue
+ fi
+ pairmd5_seen="$pairmd5_seen $pairmd5 "
initrd=$(echo "$initrd" | sed "s!^$mpoint!!")
result "$partition:$kernbootpart::$kernbasefile:$initrd:root=$mappedpartition"
exitcode=0
Adding above change causes:
bobek:~# /usr/lib/linux-boot-probes/50mounted-tests /dev/mapper/vg-deboot
/dev/mapper/vg-deboot:/dev/mapper/vg-deboot::/vmlinuz:/initrd.img:root=/dev/dm-23
Maybe better is to get the result with files under /boot directory using
reordered items:
--- 90fallback~zito2 2012-11-19 01:54:11.357776732 +0100
+++ 90fallback 2012-11-19 01:57:53.389770171 +0100
@@ -17,8 +17,8 @@
mappedpartition=$(mapdevfs "$partition" 2>/dev/null) || mappedpartition="$partition"
exitcode=1
-for kernpat in /vmlinuz /vmlinux /boot/vmlinuz /boot/vmlinux "/boot/vmlinuz*" \
- "/boot/vmlinux*" "/vmlinuz*" "/vmlinux*" "/kernel-*" "/boot/kernel-*"; do
+for kernpat in /boot/vmlinuz /boot/vmlinux "/boot/vmlinuz*" "/boot/vmlinux*" "/boot/kernel-*" \
+ /vmlinuz /vmlinux "/vmlinuz*" "/vmlinux*" "/kernel-*"; do
if echo "$kernpat" | grep -q boot/; then
kernbootpart="$bootpart"
else
Now the result is:
bobek:~# /usr/lib/linux-boot-probes/50mounted-tests /dev/mapper/vg-deboot
/dev/mapper/vg-deboot:/dev/mapper/vg-deboot::/boot/vmlinuz-3.2.0-4-amd64:/boot/initrd.img-3.2.0-4-amd64:root=/dev/dm-23
This change seems to me better for Debian. I'm not certain about intent
of original writer about ordering items (other probed systems), so maybe
this is not acceptable.
The root file-system root=/dev/dm-23 is resolved inappropriately. I must
change kernel argument root=/dev/dm-23 to root=/dev/mapper/vg-deboot in
the grub by hand to boot. This is another issue, witch should have
different bug number so I'm not solving it there.
Hope the attached complete diff is helpful a bit.
Sorry for my English :(.
Kindly regards
--
Zito