The install from LiveCD approach

[Jeronimus Linuxius]

There's a point about the installer "model" that I think could be useful for ROS right now.

The current paradigm of some Linux distributions (e.g. Ubuntu) of installing from a normal LiveCD (by "normal LiveCD" I mean a LiveCD with the components needed to run the regular OS... something like our current livecd), by clicking on a desktop link is adequate for any modern OSs, including one that is still in alpha-stage, like ReactOS... I think that installer paradigm could help development right now, for the reasons:

- It would ease testing of LiveCD (i.e. while testing the installcd, one would also be doing a basic boot test on the livecd, and perhaps others). More testing means more bugs reported. More bugs reported means potentially less difficult-to-solve regressions (because they are detected sooner, they are easier to track and solve)

- It would bring the stability of the LiveCD somewhat more to first plane, which is important for people who wants to try ROS without installing it.

- IMHO, it's a good paradigm for a working OS, too. After all, if we have to build a system that is bootable from CD to make an install CD, why not put there the needed tools to troubleshoot the system and/or get troubleshooting info from the net?

Of course, the disadvantages are that it's not a windows-compatible idea (from the user point of view). I don't know what's the strength of this argument for you, ROS developers, though.

JJ

[GoBusto]

Something like this?

[Jeronimus Linuxius]

As a matter of fact, looking at http://www.reactos.org/wiki/index.php/Bootstate, we can see that most people don't test the LiveCD...

JJ

[Jeronimus Linuxius]

Something like this?

I didn't know that page. Yes, joining the livecd and the installcd was what I meant.
Ouch! That page was created just last Wednesday!

(Later LiveCD and GUI-Setup mode might be combined to one entry.)

I would unify them from the start, though.
The classical usetup is another story, yes. However, I could also imagine a recover console (text-only), from which users could start the menu-driven, text-mode installer if they wished to, or not, if they did not wish to.

JJ

[oiaohm]

Ok Reactos is not exactly copying the Linux Distribution model.

Closest you would get is a way of reactos downloader working with a new live/install cd production tool to produce a disk user wants.

So basically not a Distribution. Its here the core and some tools to roll your own model. Working to that is valid.

There will be things inside NT design that have to be looked at. Remember OS's like Linux and Solaras are quite virus resistant without a anti-virus installed. These are things the NT design has to be looked at being extended so anti-virus and other tools like it are not the be all and end all of OS protection.

Model of Anti-viruses is flawed. You have to know about the virus to find it or at least part of a virus. The reverse model solaris and Linux model you know what applications should be doing. If a application does something its not meant to be doing kill it. Kinda makes virus development for the platform little bit more tricky. One wrong step and viruses operation is stopped.

This is more OS design upgrading to be looked at before version 1.0 and after ROS is kind operational.

[Jeronimus Linuxius]

Ok Reactos is not exactly copying the Linux Distribution model.

Seriously? No!

Closest you would get is a way of reactos downloader working with a new live/install cd production tool to produce a disk user wants.

I don't think so... Have you (even) read the wiki page GoBusto indicated? Don't think so either...

So basically not a Distribution. Its here the core and some tools to roll your own model. Working to that is valid.

I was talking about the *official* model, and presented some arguments.

There will be things inside NT design that have to be looked at.

I don't understand what NT OS design has to do with the installer model.

Model of Anti-viruses is flawed. You have to know about the virus to find it or at least part of a virus.

Of course. According to Rice's Theorem, there's no general solution for deciding whether some program has a particular property (in this case, replicating itself). But the discussion about anti-virus vs. resource protection is not relevant here (I was talking about a CD, not about virus).

JJ

[oiaohm]

http://www.reactos.org/forum/viewtopic.php?f=14&t=2120

Sorry Jeronimus Linuxius I was reading two topics at once and cross mixed the answered. In the wrong one. The merge between live and boot has been on the cards for a long time.

Of course. According to Rice's Theorem, there's no general solution for deciding whether some program has a particular property (in this case, replicating itself). But the discussion about anti-virus vs. resource protection is not relevant here (I was talking about a CD, not about virus).

Rice's Theorem is wrong and right at the same time. Rice theorem is used to say 100 percent prediction model cannot be done. Sorry to say that is wrong its doable http://llvm.org/ does it. It works out every possible code path. Turns out you need the complete program and everything it depends on and a fair bit of processing time and what is in the source code limits the paths removing the imposable status. Catch is doing in a virus scanner on a asm code major overhead so its right that its not practical. Its also the key reason why Anti-Virus software is flawed. Rice theorem is wrong because its trying to look at a small section of a program too small of section. Its bit like reading a single word out a book. You cannot tell what the book is about. Yet if you read the complete book that answer is simple even reading the complete sentence ie all the code it depends on and interacts with also answers what its upto. llvm is read the complete book and work out the sentences.

Rice's Theorem is dragged out to justify Anti-Virus failure. Really the theory is invalid for that usage. Reason Anti-Viruses do have the option to look at the complete book. I am 100 percent sick of hearing it used for that. The answer is simple look at bigger picture.

Mandatory Access Controls(MAC) monitor the functions and locations a program uses. So if a program does not use the functions virus needs to replicate and cannot create files that can be run the Mandatory Access Controls blocks virus from being able to use that section of the OS even if that section has flaws. Mandatory Access Controls simply reduce the numbers of system parts usable to a virus. Lots of attack able places become nothing more than infection of ram alone so as soon as the system is reboots or process restarted virus is killed. IE did not find the virus. Blocked its long term life.

Also you have systems under development like http://web.yl.is.s.u-tokyo.ac.jp/~tosh/kml/ that wrap protective code around program preventing from being altered. This is also processing the complete program. If we remove every flaw from the program its not a risk anymore.

Then there is another defense used in combination with Mandatory Access Controls. Its HIDS Host Intrusion Detection System. That detects any system core alterations. On systems where you can have audit able sources virus chance of living the core of the OS is reduced to Almost Zero with MAC and HIDS combined unless there is a flaw in the MAC and HIDS. So biggest threat to system become being rootkitted by passing protections not viruses. Reason why OS's like Linux have more rootkits than viruses.

Virus scanners are reduced basically to user files and service created files only. A Mac can even tag service created files for scanning. Reason getting deeper in the OS is taking on heavy defense. Instead of having to just trick a Antivirus you have to trick a Mac and HIDS as well.

Lot of this tech needs to start in the Installer particularly the HIDS and MAC settings out box. So as soon as the OS comes up it has the most defense able. Issue is windows design is going to make doing this hard its missing the MAC built into its base design.

The Mac/HIDS is also needed to prevent something. If you can run a installer from a Live disk and nothing stops it being run inside a Live OS cloning it into a new directory but with your own alterations then swapping over to booting from that big problems. Deeply embedded rootkit. Done well enough user might not even notice. Major reason why Linux is moving to more more a least privileged option. This has been done to Linux Kernels on Servers many years ago and lots of the Linux world don't forget it.

Microsoft Signed parts is really a form of HIDS. We need to get at least 1 of Mac or HIDS into ros from the install stage. Due to ros being open source most likely both are required.