Re: hfs

[Martin Costabel <costabel@wanadoo.fr>]

jingai wrote:

> I haven't tried mounting HFS cds, but I can most certainly read and write
> to HFS partiitons just fine.  I don't know exactly what could be different,
> but I haven't had any troubles with it since 2.3.99x..

I confirm Scott's observations:

1. No mounting of HFS CDs. The error message (for the LinuxPPC2000 CD,
for instance) is

root[2]#mount -t hfs /dev/cdrom /mnt/cdrom/
sr.c:Bad 2K block number requested (2 1) I/O error: dev 0b:00, sector 2
hfs_fs: unable to read block 0x00000002 from dev 0b:00
hfs_fs: Unable to read superblock
sr.c:Bad 2K block number requested (0 1) I/O error: dev 0b:00, sector 0
hfs_fs: unable to read block 0x00000000 from dev 0b:00
hfs_fs: Unable to read block 0.
mount: wrong fs type, bad option, bad superblock on /dev/cdrom,
       or too many mounted file systems

Note that this sam CD mounts perfectly with a 2.2.15-pre20 kernel.

2. Hard freeze after mounting of a HFS partition. Sometimes (I did a
couple of tests with a hfs ZIP; it's a PITA to wait 10 minutes for the
reboot with fsck each time) I can copy one or two files on the HFS
partitions, but sometimes it freezes immediately, and it freezes for
sure when I cd there and compile some little C program.
With xmon enabled, the following OOPS is captured:

bad magic 0 (should be c50c7740, creator 0), wq bug, forcing oops.

Then there follow addresses, with NIP decoded as <hfs_bnode_relse+110>
(this is defined in fs/hfs/bnode.c). All other addresses are in the
C[3-5]xxxxxx range which is not in kernel space, so I cannot decode
them. "Task" is 'cd' or 'ld' or whatever I was doing when the freeze
occurred. The oops message is generated by include/linux/wait.h.

I can see that Paul applied some changes to fs/hfs/inode.c recently, but
this doesn't change anything for me. I noted (and reported) the
brokenness of hfs at least as early as 2.3.18. I am willing to help, but
my own understanding of file systems in general, hfs in particular, and
wait queues and such stuff is too limited to get me anywhere on my own.

--
Martin

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/