mindi kernel fs detection is wrong

[victor gattegno]

kernel fs detection by mindi (grep -q gzip, then strings etc.) is wrong, the tests are made on vmlinuz, but it works only with initrd.

In /usr/sbin/mindi :

file $fname | grep -q gzip

I attached tests on RHEL 4 and SLES 11 SP1.

Also the tests are based on Linux kernel 2.*, but in newer GNU/LInux distribs are used Linux kernel 3.* (exemple Ubuntu 12.04, etc.)

Then at boot can occur : kernel panic : VFS : Unable to mount root fs

[victor gattegno]

RHEL 4 fs detection tests

[victor gattegno]

SLES 11 SP1 fs detection tests

[victor gattegno]

More precisely, in mindi 2.1.4 REV=3091

# grep -n "file $fname \| grep -q gzip" /usr/sbin/mindi
766:    file $lvKernelImage 2>&1 | grep -q gzip
2330:           file $fname | grep -q gzip
2373:           file $fname | grep -q gzip

[victor gattegno]

I recommend the following solution :

In order to find the fs, the same tests (grep -q gzip on current vmlinuz) should be done on current initrd before this (Line 771):

	else
		# In that case, we are after 2.6.30 and use the supported initramfs 
		lvUseFilesystem="initramfs"

[victor gattegno]

On a RHEL 4 server I compared the logs of an old mindi version which detected well the kernel and fs (mindi 2.0.6-r2151 ok) with the last mindi which fails to detect it (mindi 2.1.3-r3026 not ok)

mindi 2.0.6-r2151 ok - ext2fs is detected

Call GetInitrdFilesystemToUse() with parameter /boot/vmlinuz-2.6.9-42.0.10.ELsmp to get filesystem to use for initrd.
  GetInitrdFilesystemToUse(): called with parameter: /boot/vmlinuz-2.6.9-42.0.10.ELsmp.\n
  GetInitrdFilesystemToUse(): gzip magic found at lvOffset 17956.\n
  GetInitrdFilesystemToUse(): Filesytem to use for initrd is ext2fs.\n

mindi 2.1.3-r3026 not ok

INFO: Call GetInitrdFilesystemToUse() with parameter /boot/vmlinuz-2.6.9-42.0.10.ELsmp to get filesystem to use for initrd.
INFO:   GetInitrdFilesystemToUse(): called with parameter: /boot/vmlinuz-2.6.9-42.0.10.ELsmp
INFO:   GetInitrdFilesystemToUse(): Filesytem to use for initial ram disk is initramfs.
INFO: Creating a gzip'ed cpio (AKA initramfs) initrd image...

[victor gattegno]

I compared the mindi 2.0.6 and 2.1.3 sources and found what differs concerning the fs detection.

Check "diff-mindi2.1.3-mindi2.0.6.png" screenshot attached.

I created a little shell-script from these differences ("test.sh" attached), and I attached (in "test-results.htm" attached) the results for RHEL 4, RHEL 6, SLES 10 SP4 and SLES 11 SP1.

[victor gattegno]

Diff between mindi 2.0.6 and mindi 2.1.3 concerning fs detection

[victor gattegno]

Sorry, little error in my script, I'll modify it and run it again.

[victor gattegno]

results of test.sh (old detection mode) on RHEL 4 and 6 and SLES 10 and 11

[victor gattegno]

results (html file compressed) of test.sh (old detection mode) on RHEL 4 and 6 and SLES 10 and 11

[victor gattegno]

I modified my shell-script, results are good now, fs detection works.

Just a remark, sometimes lvScanRes get two results, and then ext2fs is choiced because it's the first of the if/elif of the shell-script.

Example :

lvScanRes=<6>checking if image is initramfs...
<3>EXT2-fs: blocksize too small for device.

I attached the shell-script modified which replaces my old one.

Remark : test-results.htm, test-results2.htm, and test-results.zip are the same file, with the results of my shell-script modified.

[victor gattegno]

I modified again the test.sh shell-script, to better reflect what could be the mindi tests.

Results are in test-results.htm.

[victor gattegno]

shell-script that use the old detection mode

[victor gattegno]

results of test.sh (old detection mode) on RHEL 4 and 6 and SLES 10 and 11

[Bruno Cornec]

Many thanks Victori for all these tests.

I've now re-introduced the old code for gzip detection inside the kernel at the offset as this is really mandatory.

I'm rebuilding a new mindi version for all distros based on rev [3103]. Could you check that it's fixing the issue for good on the various one you have ? (under the test subdir).

I'll publish that version as a new stable for mindi this week-end then.

[Bruno Cornec]

Seems to have been fixed with mindi-2.1.5