Date: Wed, 10 Jul 96 17:46:15 +12000 From: <E00114@vnet.atea.be> (Rob Schofield) To: <questions@freebsd.org> Subject: V2.1 Walnut Creek release Install probs Message-ID: <vines.mmb7%2Bn0xslA@vnet.atea.be>
next in thread | raw e-mail | index | archive | help
FreeBSD 2.1 - Problems with Disk Partioner and Labeller during Install. Hi there, my name's Rob Schofield. I have a problem with the install of FreeBSD 2.1 from the Walnut Creek 2-CD installation set. During the install, although the install can be completed successfully, on re-boot of the system, the disk's partition tables have been corrupted. I have read the documentation and I am still a bit mystified, so here's the call for help. I am a programmer with UNIX SysAdmin experience on SUNs, so you can assume a reasonable level of competence & and Jargon Understanding Threshold (JUT)! First, the kit: I have an Intel architecture, single processor, DEC Application DEC 433MP file server system using EISA bus, 486DX and 16M of RAM. As a disk controller, I'm using the Adaptec AHA1742A EISA/SCSI host bus adaptor in enhanced mode. This has been properly configured via the system's configuration utility, and appears to work OK under Win NT (spit, spit) and DOS (hurrrgh). There's two SCSI disks internally mounted in the system and a 7-tray CD-ROM changer mounted externally; disk at SCSI Address 0 is a Fujitsu 1.032 Gb (SCSI-2 Standard), disk at address 1 is a DEC RZ205 205 Mb (SCSI-1 standard), changer at address 2 is a Nakamichi MBR-7 2X CD-Rom changer (multiple LUN, SCSI-2 standard). The host adaptor is at address 7, unterminated. The Fujitsu is at the end of an internal 2-hop daisychain passively terminated on the drive. The external CD-Rom changer is also terminated passively at the end of a single-hop external cable. The Fujitsu is greater than 1 Gb, and does NOT have sector translation enabled on the controller for use with DOS (ie. I lose 32M under DOS as it can only address volumes up to 1Gb). There is no other disk controller in the system, and the system does not have disk 0 or 1 installed in the CMOS config. At boot time, the Adaptec installs the Fujitsu as disk 80h and the DEC as disk 81 using it's onboard BIOS. All three drives are successfully detected during boot/BIOS load. The controller BIOS loads at C800h, with the VGA BIOS loading at C000-C7FF. ROM shadowing is disabled. Second, the disk layout and what I did: Originally I had only DR-DOS 6 installed on the Fujitsu, in a single slice containing a 100M Big DOS Primary partition (DOS V3.3 format), and a 500M DOS Extended partition containing 5 logical drives of 100M. The MBR was the standard DR-DOS 6 boot loader + tables. The remaining 432 Mb was not used. The drives were sequenced C: - H:, with 7 CD-Rom drive letters being assigned from I: - O: using MSCDEX extensions and an enhanced-mode ASPI driver set. The DEC drive was not installed, the Fuj at add. 0, the Nakamichi at add. 1. I installed BSD 2.0.5 successfully (?) using the Walnut Creek CD-ROM distribution. The install partioner successfully utilised the additional (unreachable by DOS) 32M to give a second slice of 432 Mb for FreeBSD. This was then correctly labelled and partitioned internally to support root, swap, usr, home, and var partitions. The DOS primary disk was mounted as a volume on /dos in root. With the FBSD Boot loader installed in the MBR, I could successfully boot and load both DOS and BSD 2.0.5 (DR-DOS in part. table entry 3, BSD in part. table entry 1). (Note: The XFree86 package would never install correctly[it frequently seemed to hang], and I suspect that this caused the system install sequence to prematurely abort, leaving an incomplete but operational 2.0.5 system on the machine). After buying the 2.1 2-CD distribution, I attempted to update the 2.0.5 installation by a re-install (I had no real data to retain, so it was OK to just blast it). The from-CD install option disagreed with the DR-DOS memory manager, so I formatted a boot floppy and installed from floppy and CD-ROM. The installer floopy mini-kernel correctly found the 1742A, and identified the drives and the CD-ROM correctly. This is where I saw my first problems; when the partitioner came up, it spotted the second 432M slice as unused (it was no longer marked in the partition table as I restored the original MBR) and placed the prompt bar over it, marked as "Unused". Using the (C) command, I tried to create a BSD slice. At this point, I got the "Size" dialogue box, containing the number of sectors shown on the first screen. <OK> created the BSD slice, but *smaller* than the amount specified by some 4,000-odd sectors, plus a *new*, 4,000 sector area marked "unused". This small area cannot be deleted (it is already marked as unused, obviously!). If you then try to create a BSD slice within it, it will succeed and again cause the creation of a second, smaller free area. The second, new BSD slice has zero size. This can be carried on until the partiton table fills. If you delete the BSD slice created this way, the deleted slice's space gets added back to the "unused" area. All of this behaviour is different from that of the 2.0.5 partitioner, which correctly recognised and used all the available space on the disk. The labeller now correctly operates in the newly defined BSD slice to create the required OS partitions (I initially used the (A)utomatic option to size them), but on exit, a warning message box is put up stating: Warning: sd0sX does not end on track boundary Warning: sd0sX does not end on cylinder boundary .. .. .. .. or some such. This "warning message pair" seems to appear (by co- incidence) for each of the predefined DOS drives (6) in slice 1, as if the labeller is operating on the first slice, not the new second slice. The install proceeds correctly to completion, and the machine is re-booted. At this point, with the new boot loader installed, I got the expected: f1: FreeBSD f3: DOS ? .. but selecting either one results in the return of the ? prompt, with no boot sequence. At this point, re-booting with a DOS boot floppy, it is no longer possible to access the primary partition DOS drive (C:) or extended partition logical drives (D: - H:) in the first slice - it is as if the partiton table has been corrupted. After restoring the MBR (boot record and partition table data) from a rescue disk, the system re-boots directly to DOS again, with all drives accessable and no data loss. This, to my mind, implied partition table corruption during the Partioner/labeller run or when the new layout data is committed to disk. Reading the documentation did not really yield a clear explanation or cause, other than a hint that this may be a geometry-related problem (on a SCSI disk???) Just to make things difficult, I then added the second, DEC drive at add. 1, moving the Nakamichi to add. 2 (taking care to sort out the termination properly). Windows NT (spit, spit) was installed to occupy the entire disk. The NT boot loader files installed correctly within the Fuj Slice 1, DOS primary partition. With the now-restored standard MBR, the NT multi-boot loader inside slice 1 correctly gave a choice of a DOS boot or an NT boot. The setup was now disk 80h with 600 M in a single slice (containing DR-DOS and NT boot files), 432 M unused, and disk 81 with 205 M in a single slice with no unused space. A second rescue disk was created as this point. Repeating the FreeBSD install, selecting drive SD0 to create the BSD slice on the Fuji disk, the same thing happened as before - weird partitioner behaviour, followed by warnings, and on re-boot, the ? prompt now had: f1: FreeBSD f3: DOS f5: Unknown ..as choices, none of which worked. A restore with the rescue disk restored the original setup with no data loss, both NT and DOS available for boot. Points to note: I have not been especially scrupulous in my investigation so far, as there are a number of things I could still do: 1) Each time I have installed, I have chosen to mount the DOS primary partition on the root at mount point /dos when the labeller creates partitions using the (A) option. I have not checked whether the "Warning:......." message pairs persist if I do *not* choose to mount it in the labeller menu. 2) I have not yet created a BSD boot floppy with a mini-kernel on board that I could use to attempt to mount the partitions created in the new slice used for the install. 3) I have not yet used a utility to inspect the contents of the partition tables on the Fuji before and after the install sequence, or to manually edit the partition table contents. 4) I have not explored the possibility of using the 2.0.5 Partitioner/Labeller to try and pre-designate the free area as a FreeBSD slice, leaving it empty, then boot and perform an install into the slice from 2.1 I'm writing to see if you have had any experience of this type of problem being reported since the release was made. It is my intention to proceed further to see if I can resolve this myself, but it would of course save a lot of time and effort if you already had an instant solution! I realise you guys are probably up to your neck in work (if it's anything like my environment), so I appreciate greatly in advance any help you could give. It is my intention to have a go at writing a device driver for the Novell/Eagle 3200 32-bit EISA ethernet card, so I may even get an opportunity to feed something back in gratitude! A couple of extra questions for you: - have there been any posted updates or patches to 2.1 since it's initial release? - do you have a standing bug list available anywhere? - are there any plans to improve EISA support? Anyone have anything in the works? How about the AHA274X series? - is anyone working on a JAVA VM and compiler port for 2.1 + X? Due to a rather weird mailing setup (I have to work off-site), can you address any replies to: schofiel@xs4all.nl - and it will be forwarded to me. Thanks again, Rob Schofield
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?vines.mmb7%2Bn0xslA>