Hello,
get_next_ino would allocate a number between 0...2^32 - 1 to be used as
an inode number. The implementation of this mechanism relies on an
unsigned int which is 32 bits. On one server I'm observing that every
couple of months grsec complains that the percpu variable last_ino
overflows (due to shared_last_ino) being incremented to the limit of a
32 bit value and then then the machine becomes unstable due to grsec
starting to kill processes. Now, I understand this isssue stems from the
fact how grsec detects the overflow.
My question is what are the repercussions of get_next_ino overflowing
and at some point having possibly multiple inodes on my system with the
same i_ino id? And why is it safe to have the inode id's overflow and
wrap around?
Would simply changing the inode numbering code work with 64 bit value
remedy the situation or would it require a more involved fix?
Regards,
Nikolay
Hello,
Nikolay Borisov:
> My question is what are the repercussions of get_next_ino overflowing
> and at some point having possibly multiple inodes on my system with the
> same i_ino id? And why is it safe to have the inode id's overflow and
> wrap around?
I am afraid some applications won't work correctly.
As far as I know, ls(1) and find(1) don't show the file whose inum is
zero.
See also
Subject: [PATCH v2] vfs: get_next_ino(), never inum=0
Date: 2014-05-28 14:06:32
http://marc.info/?l=linux-fsdevel&m=140128600801771&w=2
and their thread.
For tmpfs, I have another patch. Just FYI, here attached.
J. R. Okajima