5.0-CURRENT diskless boot recognition
Hello. As I posted prior to this question, I have problems in getting diskless started with 5.0-Current as cvsupdated today. The problem is still the same and after a night of hairloosing work I think I got closer to the problem. We use PXE as bootstrap environment, isc-dhcp and a read-only disk partition /usr/diskless conatining for each architecture we boot diskless its appropriate directory. The diskless kernels are compiled with individual 'ident Marker', have options NFS_ROOT, MD_ROOT, UNIONFS, PSEUDOFS and device md. The whole setup is as described in /etc/rc.d/initdiskless with special /conf-dir entries and this scheme works perfect under FreeBSD 4.7. FreeBSD 5.0-CURRENT seems to have problems within its bootstrap process to recognize that it is a diskless system. After the diskless station got its IP, loaded and booted the kernel I see this on screen: Mounting root from nfs: NF ROOT: MY.IP : /usr/diskless/xterm Loading configuration files. Starting file system checks: mount: / : unknown special file or filesystem Mounting NFS file system:. eval: /etc/rc.d/cleanvar: Permission denied. . . . After the last message a lot of deny error occur. I modified all the diskless-scripts in rc.d with my own echo commands to check which one gets involved, but none of them get touched! The above process looks identical of what a normal standalone machine does when booting. No wonder when diskless does not work when the init process does not recognize that it is booting diskless. We do not use BOOTP and I do not know whether FBSD 5.X does only support this scheme. We would like to stay with the NFS process. But I think technically this can not be the problem, because after the station has already booted the kernel it doesnt care what mechanism it booted from. NFS is the dominating facility and I could see, the root partition got mounted as expected. Can anyone help? Do I mark each kernel with 'ident DISKLESS' to give the init process any idea what it should do? If you need more informations about my configuration, please contact me. If someone could provide me with further informations how a init process or init scripts figures out whether it configures a diskless kernel or not, please let me know it. thanks in advance, Oliver -- MfG O. Hartmann [EMAIL PROTECTED] -- Systemadministration des Institutes fuer Physik der Atmosphaere (IPA) -- Johannes Gutenberg Universitaet Mainz Becherweg 21 55099 Mainz Tel: +496131/3924662 (Maschinenraum) Tel: +496131/3924144 (Buero) FAX: +496131/3923532 To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: 5.0-CURRENT diskless boot recognition
As I posted prior to this question, I have problems in getting diskless started with 5.0-Current as cvsupdated today. The problem is still the same and after a night of hairloosing work I think I got closer to the problem. We use PXE as bootstrap environment, isc-dhcp and a read-only disk partition /usr/diskless conatining for each architecture we boot diskless its appropriate directory. The diskless kernels are compiled with individual 'ident Marker', have options NFS_ROOT, MD_ROOT, UNIONFS, PSEUDOFS and device md. The whole setup is as described in /etc/rc.d/initdiskless with special /conf-dir entries and this scheme works perfect under FreeBSD 4.7. I don't think you need MD_ROOT or UNIONFS. I also have BOOTP, BOOTP_NFSROOT and BOOTP_NFSV3. FreeBSD 5.0-CURRENT seems to have problems within its bootstrap process to recognize that it is a diskless system. After the diskless station got its IP, loaded and booted the kernel I see this on screen: Mounting root from nfs: NF ROOT: MY.IP : /usr/diskless/xterm Loading configuration files. Starting file system checks: mount: / : unknown special file or filesystem Do you have a mount point for / in fstab? I have something like this: my.nfs.ip.number:/export/current / nfs ro 0 0 Mounting NFS file system:. eval: /etc/rc.d/cleanvar: Permission denied. . . . After the last message a lot of deny error occur. I modified all the diskless-scripts in rc.d with my own echo commands to check which one gets involved, but none of them get touched! The above process looks identical of what a normal standalone machine does when booting. No wonder when diskless does not work when the init process does not recognize that it is booting diskless. The only mods that I made was to mount /var over NFS and not as a RAM disk. It would be nice if there was a knob to select that. :-) We do not use BOOTP and I do not know whether FBSD 5.X does only support this scheme. We would like to stay with the NFS process. But I think technically this can not be the problem, because after the station has already booted the kernel it doesnt care what mechanism it booted from. NFS is the dominating facility and I could see, the root partition got mounted as expected. Remember BOOTP is just a subset of DHCP. Actually the BOOTP in the kernel will first try dhcp requests before fallng back, IIRC. And it seems to need it in -current while it didn't when I last did a -stable diskless setup. Can anyone help? Do I mark each kernel with 'ident DISKLESS' to give the init process any idea what it should do? I use DLESS, so it shouldn't be necesary. :-) Well I learned a lot by watching tcpdump while it is all happening. :-) John -- John Hay -- [EMAIL PROTECTED] / [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
Re: 5.0-CURRENT diskless boot recognition
On Tue, 11 Mar 2003, John Hay wrote: : : As I posted prior to this question, I have problems in getting diskless : started with 5.0-Current as cvsupdated today. The problem is still the same : and after a night of hairloosing work I think I got closer to the problem. : : We use PXE as bootstrap environment, isc-dhcp and a read-only disk partition : /usr/diskless conatining for each architecture we boot diskless its appropriate : directory. The diskless kernels are compiled with individual 'ident Marker', : have options NFS_ROOT, MD_ROOT, UNIONFS, PSEUDOFS and device md. : The whole setup is as described in /etc/rc.d/initdiskless with special : /conf-dir entries and this scheme works perfect under FreeBSD 4.7. : :I don't think you need MD_ROOT or UNIONFS. I also have BOOTP, :BOOTP_NFSROOT and BOOTP_NFSV3. I do not have BOOTP and the other BOOTP_XXX options. When enabling this, I fall into trouble and the kernel panics. As you can see, the BOOTP-stage is left behind when the problems I have occurs. BOOTP should be necessary for loading and bootstrapping, but the kernel was already running when the fault happens. : : FreeBSD 5.0-CURRENT seems to have problems within its bootstrap process : to recognize that it is a diskless system. : : After the diskless station got its IP, loaded and booted the kernel I see this : on screen: : : Mounting root from nfs: : NF ROOT: MY.IP : /usr/diskless/xterm : Loading configuration files. : Starting file system checks: : mount: / : unknown special file or filesystem : :Do you have a mount point for / in fstab? I have something like this: : :my.nfs.ip.number:/export/current / nfs ro 0 0 No, I do not. I tried this, but without success. The problem is, as far as I can follow up, that the booted kernl/system does not know to be a diskless machine, so it goes on in initializing itself with the normal scripts and not with those especially made for diskless booting. : : Mounting NFS file system:. : eval: /etc/rc.d/cleanvar: Permission denied. : . : . : . : After the last message a lot of deny error occur. : I modified all the diskless-scripts in rc.d with my own echo : commands to check which one gets involved, but none of them : get touched! The above process looks identical of what a : normal standalone machine does when booting. No wonder when : diskless does not work when the init process does not recognize : that it is booting diskless. : :The only mods that I made was to mount /var over NFS and not as a RAM :disk. It would be nice if there was a knob to select that. :-) : : : We do not use BOOTP and I do not know whether FBSD 5.X does only support this : scheme. We would like to stay with the NFS process. But I think technically : this can not be the problem, because after the station has already booted : the kernel it doesnt care what mechanism it booted from. NFS is the dominating : facility and I could see, the root partition got mounted as expected. : :Remember BOOTP is just a subset of DHCP. Actually the BOOTP in the kernel :will first try dhcp requests before fallng back, IIRC. And it seems to :need it in -current while it didn't when I last did a -stable diskless :setup. If this is really the case, how prevent the kernel from panicing? What is the benefit of doing a bootp-bootstrap the new way compared against the way it was done in 4.X? If bootp is used, I need to reconfigure the whole stuff, especially the DHCP-server and this is not a nice work to do ... :-( : : Can anyone help? Do I mark each kernel with 'ident DISKLESS' to give the init : process any idea what it should do? : :I use DLESS, so it shouldn't be necesary. :-) : :Well I learned a lot by watching tcpdump while it is all happening. :-) : :John :-- :John Hay -- [EMAIL PROTECTED] / [EMAIL PROTECTED] : Thanks. Oliver -- MfG O. Hartmann [EMAIL PROTECTED] -- Systemadministration des Institutes fuer Physik der Atmosphaere (IPA) -- Johannes Gutenberg Universitaet Mainz Becherweg 21 55099 Mainz Tel: +496131/3924662 (Maschinenraum) Tel: +496131/3924144 (Buero) FAX: +496131/3923532 To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-current in the body of the message
