[syslinux] pxelinux takes too long trying all the config file options

[Steve Rikli]

On Mon, Apr 03, 2023 at 07:03:28PM -0400, Rob Roschewsk wrote:
> Hi thanks for replying .... forgot I should have mentioned the version
> of pxelinux is 6.03
> ubuntu netkit-tftp tftpd 0.17-22ubuntu2
> 
> Checked all of the items you listed ... all is in order
> 
> I can't see how it would be the tftp server because when it finds the
> files to load it has no issue .... it's PXELINUX looking for stuff
> that is NOT there .... per the boot sequence in the PXELINUX wiki ...

If you suspect PXELINUX version, perhaps try 6.04-pre1 rather than
the 6.03 you have now.

I've been using 6.04-pre1 for both EFI & BIOS clients for years at this
point; I switched when 6.03 didn't cope with an EFI server we got, but
the later version would work well enough.

It's unfortunate, of course, that 6.04-pre1 or thereabouts seems to be
the end of the road for PXELINUX.

> What is taking so long, it tries each one of those options, multiple
> times, backing off after each failure.
> 
> It loads pxelinux.0 and ldlinux.c32 instantly without issue

Hmm, your first email included syslog lines showing multiple get attempts
for pxelinux.0 too at the start of the log snippet, which led me to suspect
tftpd, i.e. since pxelinux.0 hadn't been loaded yet.

But perhaps those were from legitimate separate boot attempts -- can't
tell for certain because the tftp client info isn't in the log, hence my
suggestion for more "-v" on tftpd.

> If I go in explicitly and create a config file for the GUID for
> example ....  It finds its first try and proceeds. (but that's not
> what I want)
> 
> I have not seen this DELAY in earlier versions of pxelinux ... I've
> seen it rapidly try each one of the above bootfile names and quickly
> fall through to default.

Agreed, that's the expected behavior.

Cheers,
sr.


> On Mon, Apr 3, 2023 at 4:19 PM Steve Rikli <sr at genyosha.net> wrote:
> > On Mon, Apr 03, 2023 at 12:52:41PM -0400, Rob Roschewsk via Syslinux wrote:
> > > Hi All,
> > >
> > > I have a situation where I want to netboot and ONLY use pxelinux.cfg/default.
> > >
> > > It seems to take FOREVER for pxelinux to try all of the config file
> > > file options before falling through to default .... trying each
> > > possible combination multiple times ... increasing the wait time
> > > everytime it doesn't get a response.
> > >
> > > From my log:
> > > Apr  2 19:53:17 helium tftpd[3372124]: tftpd: trying to get file: /tftpboot/pxelinux.0
> > > Apr  2 19:54:54 helium tftpd[3372133]: tftpd: trying to get file: /tftpboot/pxelinux.0
> > > Apr  2 19:54:54 helium tftpd[3372135]: tftpd: trying to get file: /tftpboot/pxelinux.0
> >
> > Agreed, this seems slow. I also wouldn't expect to see it try & re-try
> > multiple times for /tftpboot/pxelinux.0 either.
> >
> > > Apr  2 19:54:54 helium tftpd[3372139]: tftpd: trying to get file: /tftpboot/pxelinux.cfg/44454c4c-4700-104d-805a-cac04f5a4c32
> > > ......
> > > Apr  2 19:56:54 helium tftpd[3372496]: tftpd: trying to get file: /tftpboot/pxelinux.cfg/44454c4c-4700-104d-805a-cac04f5a4c32
> > > Apr  2 19:57:08 helium tftpd[3372505]: tftpd: trying to get file: /tftpboot/pxelinux.cfg/01-d8-9e-f3-13-22-18
> > > .....
> > > Apr  2 19:59:08 helium tftpd[3372623]: tftpd: trying to get file: /tftpboot/pxelinux.cfg/01-d8-9e-f3-13-22-18
> > > Apr  2 19:59:22 helium tftpd[3372626]: tftpd: trying to get file: /tftpboot/pxelinux.cfg/C0A81674
> > > .....
> > > Apr  2 20:01:22 helium tftpd[3372784]: tftpd: trying to get file: /tftpboot/pxelinux.cfg/C0A81674
> > >
> > > That's 16 minutes of waiting and it's still not finished!
> > >
> > > It's been a while since I played with this stuff but it never took
> > > that long to fall through to default before.
> > >
> > > How can I fix this?? Is there some option I can use to speed this up??
> >
> > The multiple retries as tftpd scans for ipv6(?), mac/ethernet address
> > patterns and so on, makes me wonder if some or all of the tftpd directory
> > or contents isn't readable to the service.
> >
> > Some things to check:
> >
> > - perms & owners on /tftpboot/ and subdirs, e.g. /tftpboot/pxelinux.cfg/
> >   typically world-read, write isn't needed
> >
> > - perms & owners for the files and images you're trying to load;
> >   typically world-read, write shouldn't be needed for this
> >
> > - double-check your pxelinux.cfg/default file exists, readable, etc.
> >
> > - look for a "tftpboot -> ." symlink in your tftpboot root dir, e.g.
> >
> >   $ ls -la /tftpboot/tftpboot
> >   lrwxrwxrwx 1 root root 1 Oct 27  2010 /tftpboot/tftpboot -> .
> >
> > - daemon/service args and directory paths for tftpd in your config
> >
> > - tftp daemon startup & config is different circa RHEL7 and prior with
> >   xinetd, vs. e.g. RHEL8 and newer, modern Debian et al with systemD
> >
> > For more troubleshooting, maybe add "-v" or "-vv" to your tftpd service,
> > see if a little more syslog verbosity shows anything.
> >
> > I suspect a problem with tftpd (config?) rather than a PXELINUX issue;
> > are you using tftpd from your bootserver's OS, or perhaps some other
> > package or src?
> >
> > Cheers,
> > sr.