zlatin (03/10/83)
Has anyone else run into this kind of problem? We're running 4.1BSD on a
2Mb Vax 750 with RM80 on DEC MBA0 and CDC 9762 on Emulex SC750 MBA1.
Here is what we know about the bug: Its symptoms are that the count
in the incore inode table entry gets set to 255. Since the count is
stored as a char, it looks suspiciously as if the count got decremented
to -1. When in this state, the inode is still linked in the hash chains
of active inodes. (Correctly, as far as we can determine ... we haven't
yet made certain that it doesn't appear in the free inode list, but ...)
While thus still appearing to be active, the inode table entry does
not appear in the open file table, nor does it appear in the text table.
We have yet to check the mount table, but so far, the mounts look OK.
We haven't figured out how to dump the user structure for each process
to check the current directory.
The files which have been affected by this anomoly are nothing special.
They were: /tmp/., /usr/lib/crontab, and one which was subsequently
deleted (I.e., there was no directory entry for it ... ncheck)
Also, when we tried to do backups this morning, /usr would not
unmount (As we expected it would not). The file with the incore
inode count = 255 had a busy inode and allocated disk blocks on
the filesystem. All this in spite of the fact that it was not
referenced in any directory.
Any help or ideas would be appreciated. Reply to:
...utah-cs!sask!kimnovax!bocking
or
...utcsrgv!sask!kimnovax!bocking
Thanks.