utmp and finger (WAS: Re: some fixes.... Re: connections.tdb &messages.tdb getting full of junk)

[Andrew Bartlett]

Toomas Soome wrote:
> 
> On Tue, 9 Jan 2001, Andrew Bartlett wrote:
> 
> > Toomas Soome wrote:
> > >
> > > regarding to junk... there is second posting of some of my work. there
> > > are some issues fixed - look into diff's attached.
> > >
> > > one problem was wrong usage of pointer, while reading session id from
> > > tdb:
> > >
> > > diff -r1.20.4.3 connection.c
> > > 341c341
> > > <       slotnum = (int) dbuf.dptr;    ## this is POINTER, you cant juse
> > > it like that!
> > > ---
> > > >       slotnum = *((int*)dbuf.dptr);  ### this is correct way to do
> > >
> > > my letter from december follows:
> > >
> >
> > --snip quota stuff--
> >
> > > utmp related fixes:
> > > 1. utmp session fails to be closed in utmp_yield(). problem is actually
> > > in utmp_yield_tdb, connection numbed will be saved into utmp.tdb, but
> > > utmp_yield_tdb will return pointer to this number, not the number
> > > itself.
> > > 2. "smb/session_id" in ut_line will confuse finger and other tools (at
> > > least in solaris). the fix is to create dummy /dev/smb/session_id files
> > > - symlinks to /dev/null. current code assumes existing /dev/smb
> > > directory for smb/%d type records, its easy to add call to mkdir() as
> > > well. I left it out because I'm not sure about group for this
> > > directory.... I guess, need for /dev/smb directory should be documented
> > > in smb.conf manpage as well....
> >
> > Does ftp (presuming it uses utmp logging) confuse finger in the same
> > way?  As this appears to be the standard on these issues 'what do other
> > programs do?'. Adding directories to /dev/ seems a bit extreme.
> 
> no, ftp does not. I had no time to dig deep into utmpx world to find the
> differences..
> 
> anyway - I can't see ftp sessions in finger, so definitely there are some
> differences:
> [103] tsoome at kadri:~> ftpcount
> Service class local                -   0 users ( 40 maximum)
> Service class remote               -   5 users ( 40 maximum)
> [104] tsoome at kadri:~> f
> 
> T, 09. jaan  1:28 2001,  püsti 3 tundi, 27 minutit
> SunOS 5.8, sun4u sparc tootja Sun_Microsystems, kadri
> 1 Kasutaja, tööjärjekorra koormus: 0,72, 0,79, 0,96
> 
> Kasutaja     Nimi                  Terminal Eemal   Millal     Kust
> tsoome       Toomas Soome         *pts/1         09.01  1:27  madli.ut.ee
> 
> ftpd is wu-ftpd.
> 
> however, I like to see smb sessions in finger...

So do I, particularly their 'last logon', but I think thats a slightly
seperate issue.

> 
> >
> > Currently the thing that confuses my (linux, RH6.2) finger is the fact
> > that users are not logged in the 'last logons' file, and the 'idle for'
> > times have always been bogus (not even time logged on).  Adding a
> > /dev/smb changes the output from 'w' and 'finger' but the results are
> > still just as meaningless.
> >
> 
> yes, idle time is bogus - smb/%d is quite hard to bind to some session
> specific file.... finger will test file access times to calgulate idle
> time.

So why make them symlinks, why not just touch the file, that way the
idle time is the time since the file was created (it works!).  (You
could even go to the extreme of touching the file after each file
operation...)

> 
> login & logout times are just ok for me - after pointer dereference fix of
> course.
> 
> toomas
> --
> The aim of a joke is not to degrade the human being but to remind him that
> he is already degraded.
>                 -- George Orwell

-- 
Andrew Bartlett
abartlet at pcug.org.au