Re: [Bugme-new] [Bug 19932] New: Getting lots of name_count maxed, losing inode data messages in /var/log/messages

Re: [Bugme-new] [Bug 19932] New: Getting lots of name_count maxed, losing inode data messages in /var/log/messages

[Andrew Morton]

(switched to email.  Please respond via emailed reply-to-all, not via the
bugzilla web interface).

On Sat, 9 Oct 2010 06:58:14 GMT
bugzilla-daemon@bugzilla.kernel.org wrote:

> https://bugzilla.kernel.org/show_bug.cgi?id=19932
> 
>            Summary: Getting lots of   name_count maxed, losing inode data
>                     messages in /var/log/messages
>            Product: Other
>            Version: 2.5
>     Kernel Version: 2.6.35.7
>           Platform: All
>         OS/Version: Linux
>               Tree: Mainline
>             Status: NEW
>           Severity: normal
>           Priority: P1
>          Component: Other
>         AssignedTo: other_other@kernel-bugs.osdl.org
>         ReportedBy: jd1008@gmail.com
>         Regression: No
> 
> 
> This old problem was also reported in redhat in redhat bugzilla:
> https://bugzilla.redhat.com/show_bug.cgi?id=445757
> in mid 2008.
> 
> 
> The message is coming from kernel/auditsc.c in audit_inc_name_count():
> 
> 
> static int audit_inc_name_count(struct audit_context *context,
>                                 const struct inode *inode)
> {
>         if (context->name_count >= AUDIT_NAMES) {
>                 if (inode)
>                         printk(KERN_DEBUG "audit: name_count maxed, losing
> inode data: "
>                                "dev=%02x:%02x, inode=%lu\n",
>                                MAJOR(inode->i_sb->s_dev),
>                                MINOR(inode->i_sb->s_dev),
>                                inode->i_ino);
> 
>                 else
>                         printk(KERN_DEBUG "name_count maxed, losing inode
> data\n");
>                 return 1;
>         }
>         context->name_count++;
> #if AUDIT_DEBUG
>         context->ino_count++;
> #endif
>         return 0;
> }
> 
> I wish the kernel would also dump the stack, and funcs param values into
> /var/log/messages
> so a user can take a look at the source and do some searching
> for the cause of this.
> 
> Also, if you can advise on how I can rebuild the kernel so that 
> it will drop into the debugger when this happens, would be a great
> help.
> 

Re: [Bugme-new] [Bug 19932] New: Getting lots of name_count maxed, losing inode data messages in /var/log/messages

[Eric Paris]

On Mon, 2010-10-11 at 14:34 -0700, Andrew Morton wrote:
> (switched to email.  Please respond via emailed reply-to-all, not via the
> bugzilla web interface).
> 
> On Sat, 9 Oct 2010 06:58:14 GMT
> bugzilla-daemon@bugzilla.kernel.org wrote:
> 
> > https://bugzilla.kernel.org/show_bug.cgi?id=19932
> > 
> >            Summary: Getting lots of   name_count maxed, losing inode data
> >                     messages in /var/log/messages
> >            Product: Other
> >            Version: 2.5
> >     Kernel Version: 2.6.35.7
> >           Platform: All
> >         OS/Version: Linux
> >               Tree: Mainline
> >             Status: NEW
> >           Severity: normal
> >           Priority: P1
> >          Component: Other
> >         AssignedTo: other_other@kernel-bugs.osdl.org
> >         ReportedBy: jd1008@gmail.com
> >         Regression: No
> > 
> > 
> > This old problem was also reported in redhat in redhat bugzilla:
> > https://bugzilla.redhat.com/show_bug.cgi?id=445757
> > in mid 2008.
> > 
> > 
> > The message is coming from kernel/auditsc.c in audit_inc_name_count():
> > 
> > 
> > static int audit_inc_name_count(struct audit_context *context,
> >                                 const struct inode *inode)
> > {
> >         if (context->name_count >= AUDIT_NAMES) {
> >                 if (inode)
> >                         printk(KERN_DEBUG "audit: name_count maxed, losing
> > inode data: "
> >                                "dev=%02x:%02x, inode=%lu\n",
> >                                MAJOR(inode->i_sb->s_dev),
> >                                MINOR(inode->i_sb->s_dev),
> >                                inode->i_ino);
> > 
> >                 else
> >                         printk(KERN_DEBUG "name_count maxed, losing inode
> > data\n");
> >                 return 1;
> >         }
> >         context->name_count++;
> > #if AUDIT_DEBUG
> >         context->ino_count++;
> > #endif
> >         return 0;
> > }
> > 
> > I wish the kernel would also dump the stack, and funcs param values into
> > /var/log/messages
> > so a user can take a look at the source and do some searching
> > for the cause of this.
> > 
> > Also, if you can advise on how I can rebuild the kernel so that 
> > it will drop into the debugger when this happens, would be a great
> > help.

The audit systems has a hard coded value of 20 for the number of
inodes/pathnames it will collect information about in a single syscall.
Loading modules (such as kvm) are known to create all sorts of entries
in debugfs which overflows audit's tracking.  I also know that mounting
nfs the first time can do it setting up rpcfs.

I fixed it in RHEL by just changing the static 20 to a dynamic list but
I don't think Al liked that solution for upstream and instead wanted to
find all of the places we were overrunning 20 (hell overrunning 5 is a
problem) and figure out how to stop them from overrunning (typically by
disabling audit was his thought)

I just haven't gotten back to it as they are harmless (if scary
sounding)

-Eric