Again: fsck_ffs memory requirements

[Kris Kennaway]

Polytropon wrote:
> Hi again!
> 
> I had problems running fsck_ffs of FreeBSD 5. In my particular
> problem, version 5 seemed to help me more than version 7, which
> stopped with this:
> 
> 	fsck_ffs: bad inode number 306176 to nextinode
> 
> Version 5 didn't seem to have this problem, but it did complain:
> 
> 	fsck_ffs: cannot alloc 1073796864 bytes for inoinfo
> 
> So I asked for help and was adviced to enable swap space before
> running fsck_ffs in SUM. My main problem seemed to be that I hadn't
> enough RAM (512 + 256 MB SRD-SDRAM), exactly 1 GB seemed to be
> required.
> 
> In the recovery setting, ad0 is a FreeBSD 5 installation and ad2
> is a FreeBSD 7 installation with a /home partition holding a dd
> image of the defective partition (which's original is on a harddisk
> on the shelf for security reasons).
> 
> I did the following after successful system startup in SUM:
> 
> 	# swapon /dev/ad2s1b
> 
> 	# swapctl -l
> 	Device:       1024-blocks     Used:
> 	/dev/ad2s1b     2097152         0
> 
> 	# mount /dev/ad2s1g /mnt
> 
> 	# mdconfig -a -u 0 -t vnode -f /mnt/poly/rescue/ad1s1f.dd
> 
> 	# fsck_ffs -yf /dev/md0
> 	** /dev/md0
> 	** Last Mounted on 
> 	** Phase 1 - Check Blocks and Sizes
> 	1035979 BAD I=259127
> 	UNEXPECTED SOFT UPDATE INCONSISTENCY
> 
> 	[...]
> 
> 	9187201950452580480 BAD I=262821
> 	UNEXPECTED SOFT UPDATE INCONSISTENCY
> 
> 	9114861777597661055 BAD I=262821
> 	UNEXPECTED SOFT UPDATE INCONSISTENCY
> 
> 	EXCESSIVE BAD BLKS I=262821
> 	CONTINUE? yes
> 
> 	1779904 DUP I=262822
> 	UNEXPECTED SOFT UPDATE INCONSISTENCY
> 
> 	1779905 DUP I=262822
> 	UNEXPECTED SOFT UPDATE INCONSISTENCY
> 
> 	[...]
> 
> 	4007775 DUP I=262845
> 	UNEXPECTED SOFT UPDATE INCONSISTENCY
> 
> 	fsck_ffs: cannot alloc 1073796864 bytes for inoinfo
> 
> 	#
> 
> So I examined /usr/src/sbin/fsck_ffs/pass1.c and found out
> that a calloc() call caused the error; 1 GB was needed, but
> 2 GB were present in swap. Not enough?
> 
> Any suggestions what I could do to help fsck_ffs calloc()ing
> the needed memory?

Increase kern.maxdsiz.  The attempt to malloc this much may be wrong 
anyway though, if the filesystem is so badly corrupted that fsck is 
confused.

Kris

Kris