- Fix a file descriptor bug on Minix 3 (all users are affected, unless their Minix 3 version supports >64 fds already)
- Support ACK on Minix 3 by means of a workaround
- Change structure alignment and padding (affects people getting a SIGBUS or sometimes SIGSEGV on strict-alignment arches, e.g. IA64)
Since bsiegert@ has not indicated willingness to take over Freshmeat announcements of mksh releases, and I am almost unable and certainly unwilling to, this ceases our use of said site's services. Please use https://www.mirbsd.org/tag_mksh.rss to keep informed about things related to the MirBSD Korn Shell.
The MirBSD Korn Shell R37b has been released, incorporating a plethora of changes from R37, which was only available from CVS, and a few news by itself.
The new ‘-combine’ Build.sh option (gcc4 only) is about as noteworthy as the Minix 3 support. Users of Emdebian, FreeWRT and Beastiebox however will find that the new memory allocator has less overhead, i.e. the shell shrank in size. On the other hand, it is now much more standards compliant, especially wrt. “set -e”.
As such, mksh R37b is a strongly suggested upgrade.
Note: Freshmeat has a new website. I am utterly confused from it, both in Lynx and Opera, so I don’t know if this announcement has made it to there. I probably will cease to use Freshmeat, similarily as Slashdot, when it became unusable in Lynx. Damn “Web 2.0”!
laffer1@mnbsd and tg@mbsd worked together to create compatibility code for the FreeBSD® ash extensions; the version R37 of mksh(1) used already has been proven as /bin/sh in MirBSD and FreeWRT for some time, and Debian as well, which provided valuable corner case tests leading to eventual bugfixes.
We can only suggest others follow ☺ Apple, pkgsrc®, where are you?
With sincere apologies to Han Boetes, the dynamically linked copies of the system binaries have been removed from the system due to performance and manageability issues. For the sysutils/fakeroot port, there is a documented way to recompile these yourself, dynamically linked; you better be sure to only do that if /usr isn’t a separate filesystem, because otherwise, the programmes will be unable to find the dynamic link libraries or interpreter.
During an upgrade to the next snapshot, the files in these directories will be replaced by symbolic links. However, it is usually not instantly safe to remove them, because, for example, the /etc scripts use them, your user shell will be /usr/dbin/mksh, and the likes. So be careful wrt that directory. – A fresh MirOS installation however will no longer contain it.
“CPAN is the host for hundreds of Perl modules. Creating ports for these modules is often trivial but may still take some time. cpan2port is a new utility available in MirPorts, the MirOS ports framework, designed to facilitate this task. It should be easily adaptable for other platforms, e.g. pkgsrc®.
“The aim of this talk is to present the implementation and practical usage of the utility. Interested developers from other BSD projects are very welcome, some hints for porting the tool will be given.”
The slides for bsiegert@'s talk at FOSDEM 2009 are now available on slideshare. Please note they require a Macrobe Flash player.
Note: the /etc/security daily cronjob has been adjusted to no longer warn about empty passwords for the anoncvs and anonrsync users: “anoncvs”, “_anoncvs”, “rsync”, “_rsync”, and by chance, “__anoncvs” and other versions with more underscores, too.
Due to bug-hiding circumstances, this problem was only identified during FOSDEM Sunday afternoon. The first stage boot loader would overwrite itself trying to load the second stage boot loader, due to them sharing the same 16-bit (64 KiB) segment after the workaround for the Parallels bug. installboot(8/i386) would pass the sectors covered by filesystem blocks, which amounts up to multiples of 8 or 16 KiB, even though the last block was not entirely filled. Fix is to do bounds checking in the assembly code at boot time.
fixes dist set is available for people doing a network
installation anyway, or to extract later with
$ cd /
$ sudo tar xzphvvf /path/to/fixes10.ngz
If you do a CD installation, you have to do the following steps:
Location of sets? (cd disk ftp http shttp nfs or 'done') [done] shttp HTTP/FTP proxy URL? (e.g. 'http://proxy:8080', or 'none') [none] «Enter» Server? (IP address, hostname or 'done') www.mirbsd.org Server directory? [v10/i386] MirOS/current/older/i386 … Set name? (or 'done') […] * [X] bsd [X] fixes10.ngz Set name? (or 'done') […] done Ready to install sets? [yes] «Enter»
This sequence will add the fixes set from network after finishing a disc installation, before the installboot(8/i386) part is run. Of course, you can substitute shttp with http too or specify a proxy to use.
If you have already installed, follow the above mentioned tar
command to unpack the fixes set (in /mnt if
you are still in the installer), then use the command:
$ sudo /usr/mdec/installboot -v /boot /usr/mdec/bootxx wd0
# /mnt/usr/mdec/installboot -v /mnt/boot /mnt/usr/mdec/bootxx sd0
(wd0 or sd0 depending on which is your root disc; the second line is for within the installer)
My (tg@) sincerest apologies for this bug, which was introduced during the Parallels Desktop BIOS bug workaround’s creation. Remember, if you already have an (unbootable) installation, you can do all this by booting from the CD again (into the installer/rescue kernel).
Update 11.04: changed link to fixes10.ngz to new location, now that a new snapshot is up.
Benny, read below before killing me.
AH=41h INT 13h is a function with which one can ask a BIOS if it supports LBA on a specific drive or not. The El Torito bootable CDs only function with LBA (EDD). A gzipped LBA test ISO is available for you to check your BIOS against it; I had not yet come across one which does not report it correctly.
Our new bootloader, instead of having separate copies for network boot, CD-ROMs, floppies and hard discs, asks the BIOS what type drive it is running on (both first and second stage). If we were to remove the check if LBA is supported on the boot drive from the first stage bootloader (the PBR, bootxx), it would no longer work on MFM HDDs, floppies, etc; besides, the second stage boot(8/i386) loader is still using it to distinguish CDs from hard discs.
Anyway: the Parallels BIOS fails the EDD installation check and, as such, does not conform to the El Torito standard. Our new bootloader just happens to expose that problem – try the ISO for yourself…
We hereby cease to support working around the brokenness in order to use one unified second stage boot(8) loader per platform for both local (HDD, CD) and network boot. Parallels should fix their BIOS.
As workaround you can make use of the functionality that MirOS ISOs can be dd(1)d to a hard disc and run from there. Just enter it as an additional hard disc drive instead of CD-ROM drive. Or use netboot (load b_i386.ldr and the bsd.rd kernel).
Update 31.01: tg@ has hacked a workaround – reporting this bug to Parallels is still recommended though.
It has come to our attention that a good share of the available binary packages for the last stable version have actually been accidentally built against a (very old) version of MirOS #10-current, thus demanding slightly newer libraries. We estimate this problem being resolved when the next batch of binary packages for #10-stable is built, although there is none scheduled as of yet, and a run for #10-current will probablt precede it. Sorry about that.
An analysis of the code in question however shows that it is almost certain to be safe – for the purpose of running the aforementioned binary packages only – to rename libc.so.41.0 to libc.so.41.1 to quell the warning encountered (or one of them, possibly, but the most frequent one at that). The addition of functions was almost certainly not relevant for MirPorts use.
MirOS-current has security fixes for xterm (CVE-2006-7236, CVE-2008-2383) and OpenSSL (CVE-2008-5077) and updating from CVS is recommended.
On unrelated news, ports/net/irssi/files/randex.pl, the Irssi plugin for the MirOS RANDEX protocol to exchange entropy over IRC, has been updated as well.
mksh formed new allegiances: the Beastiebox Project (10x hubertf@TNF for mentioning) has added mksh(1) today, and the latest project of CcSsNET, CcSsLIVE (broken link removed 20100916), will include it soonish as well. Nicely, he already links to MirBSD, for he sits in our IRC channel usually and is one of the sparc users.
The next version of grml GNU/Linux might very well come not just with mksh(1) (except grml-small), but also with bootbsd.com née boot(8/i386) and bsd.rd (a slightly “tuned” version with added e3 editor and sans the Install/Upgrade/Shell prompt). On the contrary, I’m toying with the idea of adding a ports/sysutils/pxegrub for local boot along with a slightly tuned (add mksh at the very least) on the DuaLive ISOs if I find we have the space. We’ll have to work out something, licence and GPL-source-requirement wise, but we’re positive this would work out well. Considering just how good of a rescue system our bsd.rd kernel is, already, and what added benefit a minimal pretty standard Live Linux may provide.
The second-stage boot(8/i386) loader has gained several interesting new features:
- When called as DOS .com programme, it defaults to DL=80h because no actual drive is passed from the BIOS
- When named boot.com it can be run as SYSLINUX (ISOLINUX, PXELINUX, EXTLINUX) module and boot any OpenBSD/i386 and MirBSD/i386 kernels from local filesystems supported
- In addition to 4.2FFS and ISO 9660, the code to support the FAT12, FAT16, and FAT28 filesystems works and is enabled now, for both booting and the “ls” command
- If a hard disc has no type 27h partition, the disklabel is searched using the superfloppy approach (i.e., LBA sector #1)
- If no disklabel can be found, one is spoofed with a somewhat reasonable ‘c’ slice, the four MBR primary partitions as slices ‘d’, ‘e’, ‘f’ and ‘g’, and an ‘a’ slice spanning either the partition passed from DS:SI via SYSLINUX, the first non-empty MBR partition, or the entire disc
- The new command “cat” allows you to display files from within the bootloader, lest they be accessible
- The new command “machine label [hd0a]” dumps the disklabel
- You can place more than one command on the same input line, because ‘`’ now acts the same as ‘\n’ in the command parser; this is useful for instance if the serial console is not accessible from the boot loader, but required for the kernel, and using a boot.cfg is no option
The /usr/mdec/mbrldr, /usr/mdec/mbrmgr and the MBR built in fdisk(8/i386) have been upgraded to pass the partition data in DS:SI (or junk if reading from FDD or whole HDD)
*smile* Now we just need to make our bootloader multiboot-compliant…