2.4.20 /proc/partitions corruption with many partitions

[Nick Burrett]

Hi,

I'm finding it difficult to create more than, say 80, logical partitions 
on a server.  The symptoms are that lvcreate keeps segfaulting.  The 
reason is that /proc/partitions contains invalid data.

I initially wondered whether it might be the device naming convention 
since when we go above 26 logical paritions we start writing out lvm{, 
lvm}, lvm[ etc. so I added the following patch to write out the devices 
as digits but this didn't have any effect.

--- fs/partitions/check.c 2002-11-28 23:53:15.000000000 +0000
+++ fs/partitions/check.c  2003-06-03 14:43:19.000000000 +0100
@@ -148,6 +148,7 @@ char *disk_name (struct gendisk *hd, int
                         maj = "hd";
                         break;
                 case MD_MAJOR:
+               case LVM_BLK_MAJOR:
                         sprintf(buf, "%s%d", maj, unit);
                         return buf;
         }

This is quite easy to replicate, just create many logical volumes until 
lvcreate fails e.g.

$ lvcreate -L 10M -n foo01 vol01
etc.

The attached file is simply the output of:

$ cat /proc/partitions >/root/foo.txt

Any thoughts ?

Regards,

Nick.