Slashdot Mirror
User-Mode Linux Merged Into 2.5 Kernel
An anonymous reader writes "With little fanfare, User Mode-Linux (UML) has been merged into Linus' BitKeeper tree. The merge followed a patch by UML author Jeff Dike, resynching UML with the 2.5.34 development kernel.
From the UML homepage, User-Mode Linux provides you with a virtual machine that offers 'a safe, secure way of running Linux versions and Linux processes. Run buggy software, experiment with new Linux kernels or distributions, and poke around in the internals of Linux, all without risking your main Linux setup.'" There's more UML resources available at the community site.
They are entirely different things. FreeBSD's emulation does translation of the system calls into the corresponding FreeBSD ones; UML is a full Linux kernel running in user space.
were you expecting to see a sig here? perhaps you'd rather see the inside of an ambulance!
In our office (Wholesale/Virtual ISP), we've been experimenting with UML for a while as a decent alternative to trying to virtualize every service and allowing some of our reselling ISPs to have their own username space - something I don't even want to attempt with Apache and most FTP servers.
So far it's been fairly stable, after working out a few quirks. Definitely worth the trouble of getting everything set up. Makes backups on the UML servers stupidly-simple too.
Congrats to the UML developers on clearing this hurdle, and here's to hoping it betters future development on the project!
Comparing UML to FreeBSD's API redirector (usually misnamed "Linux emulation") is like comparing Windows running in VMware to WINE. VMWare runs the true full blown MS-Windows while WINE redirects the calls to approbate native calls/code. Likewise, UML runs the true full blown Linux kernel while FreeBSD just redirects the calls to native calls/code.
UML adds more layers before a system call makes it to the hardware than simple API redirection. For example, for a program running in UML to read from the CD-ROM, the real kernel only provides access to the block device and the UML kernel translates the block device/ISO9660 accordingly for the file access calls. In UML, reading of the structures as following ISO9660 is done is *user space*. FreeBSD's API redirector breaks the block device structures from ISO9660 to approbate formats for file system calls all in the FreeBSD kernel. FreeBSD's Linux "emulator" does not achieve the same redefination of what occures in user space as opposed to kernel space at all. If FreeBSD can't run Linux binaries faster than UML then something is very wrong. However, it would be interesting to see if FreeBSD's API redirector could run UML and see if UML runs faster on top of FreeBSD or on top of Linux.
Sorry, no.
The use of TLA's (Three Letter Acronyms) has become so rampant that it is hard to find things which aren't referred to by their TLA. In many cases, the same TLA has more than one meaning, so the users of the acronym are able to keep the context straight. In this case, where the software sector has a standard definition of UML, reusing the acronyn will only spread confusion
If I were to create a software application called Great New Utility and referred to it as GNU, people would rightly be upset at me for trying to usurp an already common use of an acronym. In this case, I would probably be violating a trademark. The acronym of UML is already trademarked by Object Management Group, and has a common and well known usage.
"Microsoft has made computing accessible to a population who would otherwise not be able to use computers" - B. Kernigha
And yes, you can run many instances of UML.
SIG: HUP
http://davidcoulson.net/writing/pcp/187/masterclas s-linuxexpert.pdf
You forget MVS (even better than OS/400) which is based on running multiple virtual OSes of radically different types if needed all under extremely powerful management tools.
Still I agree 100% with your main point that while this is a big step up for Linux it ain't playing in the same ballpark yet.
Odds of being killed by lightning and winning the lottery in the same day: 1 in 2^55
Honeypots are not for normal businesses to run. They are mostly of interest to security people who want to get insight into the latest tools and exploits.
I've had enough abrasive sigs. Kittens are cute and fuzzy.
Native speed - it basically passes all syscalls directly to the parent kernel (as far as I can tell). Very cool shit.
Actually, it's much slower than running on the "real" kernel, at least on a 233 MHz machine. On a 600MHz, I haven't yet noticed any large difference in speed.
It may have something to do with PMMX vs. P3, instead of clock speed.
strace runs just fine.
/. "junk filter" doesn't let me, no matter how much I try. Sorry, you'll have to verify for yourself.
I wanted to attach a few lines as proof here, but
Assorted stuff I do sometimes: Lemuria.org
But since its only an emulator, it isnt always good enough. You dont get to test how the app uses the processor caches, you dont get true concurrency - which is where you find the nastiest bugs.
Well ... it's free.
Seriously, VMware lets you run a regular OS on emulated hardware. UML lets you run an emulation OS on real hardware. Both run on top of a standard OS (with real hardware).
VMware is not itself an OS - it provides a virtual PC which your regular OS thinks it owns. UML is an OS, which knows full well it does not own any hardware. The processes running under UML do not necessarily know that, though, since UML provides a near-standard environment.
(Related point: for those of you who think you're going to set up a honeypot that the cracker will break into and think he 0wns the bare metal - keep in mind that UML was not really designed to hide its own nature, so it's not hard to check for. And if you crack root in the UML, it's possible to get out of it. So if you're using UML for security reasons, you probably should run it in its own chroot.)
"How can you claim that you are anti-crack, while still writing a window manager?" — Metacity README