Skip to contentSkip to conversation

Single user linux

PATCH [email protected] 12 participants 24 messages Apr 25, 2001 – Apr 27, 2001

lore
Share & tools
lore
#0 lore raw

first, i think i owe you guys apology for didn't make myself clear, which is going harder if you irritated. even my subject went wrong, as the patch isn't really about single user (which confuse some people).

for those who didn't read that patch, i #define capable(), suser(), and fsuser() to 1. the implication is all users will have root capabilities.

then i tried to bring up the single user thing to hear opinions (not flames). and by that, i actually didn't mean to have users share the same uid/gid 0. i know somebody will need to differentiate user.

so when everybody suggested playing with login, getty, etc. i know you have got the wrong idea. if i wanted to play on user space, i'd rather use capset() to set all users capability to "all cap". that's the perfect equivalent.

so the user space solution (capset()) works, but then came the idea to optimize away. that's what blow everybody up. don't get me wrong, i always agree with rik farrow when he wrote in ;login: that we should build software with security in mind.

but i also hate bloat. lets not go to arm devices, how about a notebook. it's a personal thing, naturally to people who doesn't know about computer, personal doesn't go with multi user. by that i mean user with different capabilities, not different persons.

i haven't catch up with all my mails, but my response to some:
- linux is stable not only because security.
- linux was designed for multi-user, dos f.eks. is designed for personal use, so does epoc, palmos, mac, etc.
- i even use plan9 with kfs restrictions disabled sometimes, cause i don't have cpu server, auth server, etc.
- with that patch, people will still have authentication. so ssh for example, will still prevent illegal access, if you had an exploit you're screwed up anyway. sure httpd will give permission to everybody to browse a computer, but i don't think a notebook need to run it.

so i guess i deserve opinions instead of flames. the approach is from personal use, not the usual server use. if you think a server setup is best for all use just say so, i'm listening.

Quoted 1 line

It would be far more interesting to rip out all trace of

security.

Quoted 1 line

That would include the kernel memory access checking,

parts of the

Quoted 1 line

task struct, filesystem and VFS code, and surely much

more.

i did say it clearly that i have other changes which i know won't be a clean patch (too many #ifdefs). f.eks. on my computer i didn't even compile user.c in, i don't have user_struct. filesystem and vfs code are affected by that patch already. memory access is important of course.

Quoted 1 line

Then you can try to show a measurable performance

difference.

nah, performance was never my consideration. i do save about 3kb from my zImage, but i'm not interested.

imel (writing from a
webmail)

----------------------------------------------------
This email was sent using http://webmail.cbn.net.id/

#1 lore raw

Hello [email protected],

Once you wrote about "Re: [PATCH] Single user linux":

Quoted 44 lines

first, i think i owe you guys apology for didn't make myself clear, which is going harder if you irritated. even my subject went wrong, as the patch isn't really about single user (which confuse some people).

for those who didn't read that patch, i #define capable(), suser(), and fsuser() to 1. the implication is all users will have root capabilities.

then i tried to bring up the single user thing to hear opinions (not flames). and by that, i actually didn't mean to have users share the same uid/gid 0. i know somebody will need to differentiate user.

so when everybody suggested playing with login, getty, etc. i know you have got the wrong idea. if i wanted to play on user space, i'd rather use capset() to set all users capability to "all cap". that's the perfect equivalent.

so the user space solution (capset()) works, but then came the idea to optimize away. that's what blow everybody up. don't get me wrong, i always agree with rik farrow when he wrote in ;login: that we should build software with security in mind.

but i also hate bloat. lets not go to arm devices, how about a notebook. it's a personal thing, naturally to people who doesn't know about computer, personal doesn't go with multi user. by that i mean user with different capabilities, not different persons.

i haven't catch up with all my mails, but my response to some:
- linux is stable not only because security.
- linux was designed for multi-user, dos f.eks. is designed for personal use, so does epoc, palmos, mac, etc.
- i even use plan9 with kfs restrictions disabled sometimes, cause i don't have cpu server, auth server, etc.
- with that patch, people will still have authentication. so ssh for example, will still prevent illegal access, if you had an exploit you're screwed up anyway. sure httpd will give permission to everybody to browse a computer, but i don't think a notebook need to run it.

Quoted 4 lines

so i guess i deserve opinions instead of flames. the approach is from personal use, not the usual server use. if you think a server setup is best for all use just say so, i'm listening.

Then, is there any advantage over booting linux with "single" option?
LILO: linux single

Signature · 3 lines
 Best regards,
 Leonid Mamtchenkov
 System Administrator
#2 lore raw
In reply to [email protected]

On Wed, 25 Apr 2001 [email protected] wrote:

[snip]

Quoted 5 lines

so i guess i deserve opinions instead of flames. the approach is from personal use, not the usual server use. if you think a server setup is best for all use just say so, i'm listening.

Heres one.. most of the time I spend cleaning up windows machines is not because of software problems. Usually it's the user acidentally erasing something or installing some program that just modified the boot files by accident.

Protection makes the system easier not harder. You can add SUID aplications to preform administrative tasks such as upgrading / config and be sure that the user won't accidentally erase the system.

I've had users absolutely paranoid of breaking something on my systems it's very reasuring for me to be able to point at the power switch and say "see that? don't touch it and the sustem will be fine"

Gerhard

--
Gerhard Mack

[email protected]

<>< As a computer I find your faith in technology amusing.

#3 lore raw
Quoted 4 lines · On Wed, 25 Apr 2001 [email protected] wrote:

so i guess i deserve opinions instead of flames. the approach is from personal use, not the usual server use. if you think a server setup is best for all use just say so, i'm listening.

Several distributions (Red Hat and Mandrake certainly) offer auto-login tools. In conjunction with those tools, take the approach that Apple used with OS X and setup "sudo" for administrative tasks on the machine. This allows the end user to generally administer the machine without all the need to hack the kernel, modify login, operate as root, etc. You can even restrict their actions with it and log what they do.

In the end though, I really don't see the big deal with having a root user for general home use. Even traditionally stand-alone operating systems have gone to this model (Mac OS X) or are heading that way fast (Windows XP). There are always ways to configure permissions, and even in a stand-alone environment it's always better to protect against accidental deletion of system critical files. In other words, the benefits vastly outweigh the minor inconvenience.

John

#4 lore raw
Quoted 10 lines · John Cavan

Several distributions (Red Hat and Mandrake certainly) offer auto-login tools. In conjunction with those tools, take the approach that Apple used with OS X and setup "sudo" for administrative tasks on the machine. This allows the end user to generally administer the machine without all the need to hack the kernel, modify login, operate as root, etc. You can even restrict their actions with it and log what they do.

In the end though, I really don't see the big deal with having a root user for general home use. Even traditionally stand-alone operating

you're right, we could do it in more than one way. like copying with mcopy without mounting a fat disk. the question is where to put it. why we do it is an important thing.
taking place as a clueless user, i think i should be able to do anything. i'd be happy to accept proof that multi-user is a solution for clueless user, not because it's proven on servers. but because it is a solution by definition.

imel

#5 lore raw
Quoted 3 lines · [email protected] wrote:

i'd be happy to accept proof that multi-user is a solution for clueless user, not because it's proven on servers. but because it is a solution by definition.

Clueless user deletes files critical to running the system. '!@#$% Why can't I boot. Oh my gosh!! Linux sucks!'

Signature · 8 lines
=====================================================================
Mohammad A. Haque                              http://www.haque.net/ 
                                               [email protected]

  "Alcohol and calculus don't mix.             Project Lead
   Don't drink and derive." --Unknown          http://wm.themes.org/
                                               [email protected]
=====================================================================
#6 lore raw
In reply to [email protected]
Quoted 1 line

taking place as a clueless user, i think i should be able to do anything.

Yeah, I thought so when I started using Linux. I stopped thinking so, when I accidentally blew up the FS on my datadrive and lost nearly _everything_ I had written for 2 years...

Quoted 3 lines

i'd be happy to accept proof that multi-user is a solution for clueless user, not because it's proven on servers. but because it is a solution by definition.

Let's turn the question the other way. It's you trying to convince us, that everyone needs root access. What does a clueless user need root access for?

Programming - no.
Writing documents - no.
Surfing the web - no.
Reading email - no.
Installing kernels - yes (but a clueless user won't do this). Running viruses, that blow up the entire system - yes. Installing software - yes. But how often do you do that? And is the 'su'
really so hard to remember?

If you really want to have different uids, why not hack xdm/login to autologin. And when it autologins to a specific user, why do you want different id's?

And if you really want everybody to have access to all files, you can just do a 'chmod 777 /'. Perhaps set it up as a cronjob to run daily?

Besides you write, that a distro shipping single-user is evil. So you want the clueless user to recompile his own kernel to enable single-user mode (why do at all call it 'single-user' when you still have different ID's?)... The clueless user probably does not even know what the kernel is - and then have to recompile it...

Rasmus

Signature · 3 lines
-- [ Rasmus 'Møffe' Bøg Hansen ] --------------------------------------
if (getenv(EDITOR) == "vim") {karma++};
--------------------------------- [ moffe at amagerkollegiet dot dk ] -
#7 lore raw
Quoted 7 lines · On Thu, 26 Apr 2001, [iso-8859-1] Rasmus B�g Hansen wrote:
Quoted 3 lines

i'd be happy to accept proof that multi-user is a solution for clueless user, not because it's proven on servers. but because it is a solution by definition.

Let's turn the question the other way. It's you trying to convince us, that everyone needs root access. What does a clueless user need root access for?

what work around what? right now it's the kernel who thinks that root is special, and applications work around that because there's a division of super-user and plain user. is that a must?
it's trivial to say that in multi-user system, one user shall not mess with other user. in multi-process, a process shall not mess with other process.
but when it comes to a computer which only has one user, why would it stop a user. because the kernel thinks it isn't right? if he felt like killing random process, which is owned by other than the user, is it a wrong thing to do? he owns the computer, he may do anything he wants.

and i'm not even trying to convince anyone. communicating is closer.

Quoted 7 lines

And if you really want everybody to have access to all files, you can just do a 'chmod 777 /'. Perhaps set it up as a cronjob to run daily?

Besides you write, that a distro shipping single-user is evil. So you want the clueless user to recompile his own kernel to enable single-user

iff that distro starts up daemons.

Quoted 1 line

mode (why do at all call it 'single-user' when you still have different

i wrote somewhere that it was my mistake to call it single-user when i mean all user has the same root cap, and reduce "user" (account) to "profile".

imel

#8 lore raw
Quoted 1 line · On Thursday, April 26, 2001, at 07:03 AM, <[email protected]> wrote:

he owns the computer, he may do anything he wants.

This sentence really stood out for me, and implies a profound lack of understanding of multi-user machines. No offense intended.

I've been a Unix admin for over ten years, and I like to think that I know my way around pretty well. But I do not and will NEVER log in to a machine as root to do work. I am the only user of my MacOS X laptop and home Linux boxes, and I still have my own personal login on all of them. What's at issue is not ownership or trust, but one of accountability and safety.

Any OS worth its weight in silicon will make a distinction between blessed and unblessed users. It can be phrased in different ways -- root vs. non-root, admin vs. non-admin. But no one should EVER log in to a machine as root. Period. (1)

Multi-user/modern operating systems exist precisely to destroy the fatal flaw that you are attempting to reintroduce. Users should have reduced privileges during normal use, and conditional privilege on demand. Safe from User Error and no less functional on GUI-based systems.

People keep saying this, but I'll say it again. This can easily be done in user-space. This HAS been done. Many times. Well. It's possible to put a user in privileged mode automatically, but I'm not convinced that an extra prompt to go into privileged mode is a bad thing from a usability standpoint.

So it doesn't need to be in the kernel. And why put it there if it doesn't need to be? Even if it's off by default, it's bloat. And dangerous, conceptually flawed bloat that can't be disabled with 'chkconfig' or 'rpm -e'. And how many people will use it? And should the kernel group allow them to from an out-of-box kernel? As I understand it, part of the responsibility of the maintainers is to maintain a conceptually focused kernel. There's nothing preventing you from distributing your patch, but inserting this into "the" kernel seems unacceptable IMVHO.

I think we understand the "why" of your patch, but I think you need to elucidate further on how the ends justify the means.

Sorry to kick a dead horse,
--
Ken.
[email protected]

(1) Except for gnarly testbed/admin machines, etc. etc.

#9 lore raw
Quoted 4 lines · On Thursday, April 26, 2001, at 07:03 AM, <[email protected]> wrote:
Quoted 1 line

he owns the computer, he may do anything he wants.

<snip>

Quoted 4 lines

Any OS worth its weight in silicon will make a distinction between blessed and unblessed users. It can be phrased in different ways -- root vs. non-root, admin vs. non-admin. But no one should EVER log in to a machine as root. Period. (1)

Also, there is another reason.
If you'r logged in as root, then any exploitable bug in large programs, be it netscape, realplayer, wine, vmware, ... means that the cracker owns your machine.
If they are not, then the cracker has to go through another significant hoop, in order to get access to the machine.
For optimal security, you can do things like running netscape and other apps under unpriveledged users, where they only have access to their own files.

(Note, netscape/.. are just used as examples, I'm not saying they are more buggy than others, just large, and hard to get bug-free)

#10 lore raw
Quoted 10 lines · Ian Stirling

Also, there is another reason.
If you'r logged in as root, then any exploitable bug in large programs, be it netscape, realplayer, wine, vmware, ... means that the cracker owns your machine.
If they are not, then the cracker has to go through another significant hoop, in order to get access to the machine.
For optimal security, you can do things like running netscape and other apps under unpriveledged users, where they only have access to their own files.

Quoted 3 lines

(Note, netscape/.. are just used as examples, I'm not saying they are more buggy than others, just large, and hard to get bug-free)

Heh. You receive all your email on your root account?

Signature · 8 lines
=====================================================================
Mohammad A. Haque                              http://www.haque.net/
                                               [email protected]

  "Alcohol and calculus don't mix.             Project Lead
   Don't drink and derive." --Unknown          http://wm.themes.org/
                                               [email protected]
=====================================================================
#11 lore raw
Quoted 7 lines · Ian Stirling
Quoted 5 lines

Also, there is another reason.
If you'r logged in as root, then any exploitable bug in large programs, be it netscape, realplayer, wine, vmware, ... means that the cracker owns your machine.

<snip>

Quoted 1 line

Heh. You receive all your email on your root account?

Nope.
For historical reasons (I gave out this address before I started using linux) and mail to root here does not actually go to root.

#12 lore raw
In reply to [email protected]
Quoted 6 lines · On Thu, 26 Apr 2001 [email protected] wrote:
Quoted 4 lines · On Thu, 26 Apr 2001, [iso-8859-1] Rasmus Bøg Hansen wrote:
Quoted 3 lines

i'd be happy to accept proof that multi-user is a solution for clueless user, not because it's proven on servers. but because it is a solution by definition.

Quoted 7 lines
Quoted 3 lines

Let's turn the question the other way. It's you trying to convince us, that everyone needs root access. What does a clueless user need root access for?

what work around what? right now it's the kernel who thinks that root is special, and applications work around that because there's a division of super-user and plain user. is that a must?

Basically yes. But if you do not want _any_ security - you can drop it. I started using Linux (and unix in general) in '96 (thanks Linus). And now - feelin like an experienced linux (unix) user I feel more like ever, I do _not_ want to be root

You do not understand the unix security aspects. You do not want unix security and do not want unix. Then stop using it. People from redmond allow you to trash your system without any special effort.

Stop bugging us. Have you noticed you never got response from Linus? He is probably still laughing (or feeling pissed off) - Stop trashing his (good) work, I know he is not the only one (I thank every Linux developer)... Did you ever realize, that the unix security model hasn't changed radically for 30 years? Beacause what? It is (opposite your patch) mostly good.

Quoted 3 lines

it's trivial to say that in multi-user system, one user shall not mess with other user. in multi-process, a process shall not mess with other process.

Ok. If you want to fuck up other people's processes, do it. Kill init and get strange panics. If you want to crash other people's work, do it. But begone from _my_ box!!!! Go to a bar and get drunk (as you do not seem to have anything better to use your time for),.

Quoted 5 lines

but when it comes to a computer which only has one user, why would it stop a user. because the kernel thinks it isn't right? if he felt like killing random process, which is owned by other than the user, is it a wrong thing to do? he owns the computer, he may do anything he wants.

Yeah. If he wants to do that he logs in as root. 'killall -1'? 'dd if=/dev/zero of=/dev/kcore'. Yeah, crash your computer if you want. But the 'clueless user does not want to'!

Quoted 2 lines

and i'm not even trying to convince anyone. communicating is closer.

Who are you not trying to convince? You propose a patch - you try to convince us to drop the unix secuity model...

Quoted 7 lines
Quoted 5 lines

And if you really want everybody to have access to all files, you can just do a 'chmod 777 /'. Perhaps set it up as a cronjob to run daily?

Besides you write, that a distro shipping single-user is evil. So you want the clueless user to recompile his own kernel to enable single-user

iff that distro starts up daemons.

Or the user starts up daemons. He has root privileges after all.

Quoted 5 lines
Quoted 1 line

mode (why do at all call it 'single-user' when you still have different

i wrote somewhere that it was my mistake to call it single-user when i mean all user has the same root cap, and reduce "user" (account) to "profile".

Ok. My mistake. You want to use 'user profiles' but not use the advantages...

You don't have to. You can use Windows if you want to. You can just use root. As long as you do not hack /sbin/login or xdm, you will still have to type login/password - no win, no gain.

If it wasn't for the nips, being so good at bulding ships the yards would still be open in the clyde

get out to a war and get shot!

Rasmus

Signature · 3 lines
-- [ Rasmus 'Møffe' Bøg Hansen ] --------------------------------------
I don't suffer from insanity, i enjoy every minute of it!
--------------------------------- [ moffe at amagerkollegiet dot dk ] -
#13 lore raw
In reply to [email protected]
Quoted 3 lines · [email protected]. writes:

i wrote somewhere that it was my mistake to call it single-user when i mean all user has the same root cap, and reduce "user" (account) to "profile".

Seen this way it makes a tad more sense:

1. you and your spouse share the computer
2. you have different shells, mail folders, etc.
3. both of you are too lazy to use su or sudo

It isn't really bright having UID 0 have properties that can't sanely be granted to other UIDs. Sure, we have the capability bits, but just try using them. On the "would be nice" list goes the ability to grant capabilities to a user, and the Novell-like ability to grant one user complete access to the files of another user without mucking with the permission bits on disk.

#14 lore raw
In reply to [email protected]
Quoted 8 lines · On Thu, 26 Apr 2001 [email protected] wrote:

you're right, we could do it in more than one way. like copying with mcopy without mounting a fat disk. the question is where to put it. why we do it is an important thing.
taking place as a clueless user, i think i should be able to do anything. i'd be happy to accept proof that multi-user is a solution for clueless user, not because it's proven on servers. but because it is a solution by definition.

I think you have it backwards here, given that Linux works one way and you want it to work another. Basically, I would suggest that it is up to you to prove that multi-user is NOT a solution for "clueless" user, especially given that there have been a number of suggestions on how to do it without changing the kernel or even changing software.

If you can't prove the case, I rather suspect that your patch won't make it. Don't feel bad though, I've yet to get one through either. :o)

John

#15 lore raw
Quoted 1 line · John Cavan

I think you have it backwards here, given that Linux works one way and you

yeah, it was a patch for linux, but i wasn't thinking linux. there are quite many os out there. and i don't think they're different just because they have programmers with different intelligence level.

Quoted 2 lines

If you can't prove the case, I rather suspect that your patch won't make it. Don't feel bad though, I've yet to get one through either. :o)

oh no, that patch was useful to explain the idea. i don't even think it's the right way. but it's a good way to exercise the idea. well, thanks anyway.

imel

#16 lore raw
In reply to John Cavan
Quoted 14 lines · On Thu, Apr 26, 2001 at 07:11:24PM +0700, [email protected] wrote:
Quoted 10 lines · John Cavan

Several distributions (Red Hat and Mandrake certainly) offer auto-login tools. In conjunction with those tools, take the approach that Apple used with OS X and setup "sudo" for administrative tasks on the machine. This allows the end user to generally administer the machine without all the need to hack the kernel, modify login, operate as root, etc. You can even restrict their actions with it and log what they do.

In the end though, I really don't see the big deal with having a root user for general home use. Even traditionally stand-alone operating

Quoted 7 lines

you're right, we could do it in more than one way. like copying with mcopy without mounting a fat disk. the question is where to put it. why we do it is an important thing.
taking place as a clueless user, i think i should be able to do anything. i'd be happy to accept proof that multi-user is a solution for clueless user, not because it's proven on servers. but because it is a solution by definition.

Look, all of this is VERY simple. There is only one single person you have to convince to get this into the kernel. And you DO have to convince him, because no matter how many others you try to force this upon, nothing gets into the kernel without the consent of the almighty penguin.

So do us all a favour, send this patch to Linus. I'd give you a 1/10 chance of getting a reply at all, and a 1/100000000000000 that the answer won't be along the terms of "No way in hell, never!" (possibly worded a bit different.) If you don't get any response in say a week or so, just give up.

/David Weinehall
_ _
// David Weinehall <[email protected]> /> Northern lights wander \\

//  Project MCA Linux hacker        //  Dance across the winter sky //
\>  http://www.acc.umu.se/~tao/    </   Full colour fire           </
#17 lore raw
In reply to John Cavan
Quoted 5 lines · David Weinehall

So do us all a favour, send this patch to Linus. I'd give you a 1/10 chance of getting a reply at all, and a 1/100000000000000 that the answer won't be along the terms of "No way in hell, never!" (possibly worded a bit different.) If you don't get any response in say a week or so, just give up.

Amusing thing is that he did CC Linus on the patch and Linus hasn't said a peep. I bet Linus laughed his ass off as he deleted the message bit-by-bit.

Signature · 8 lines
=====================================================================
Mohammad A. Haque                              http://www.haque.net/ 
                                               [email protected]

  "Alcohol and calculus don't mix.             Project Lead
   Don't drink and derive." --Unknown          http://wm.themes.org/
                                               [email protected]
=====================================================================
#18 lore raw
In reply to [email protected]
Quoted 3 lines · [email protected] wrote:

for those who didn't read that patch, i #define capable(), suser(), and fsuser() to 1. the implication is all users will have root capabilities.

And this is better than just having the system auto-login as root because......?

Quoted 17 lines

then i tried to bring up the single user thing to hear opinions (not flames). and by that, i actually didn't mean to have users share the same uid/gid 0. i know somebody will need to differentiate user.

so when everybody suggested playing with login, getty, etc. i know you have got the wrong idea. if i wanted to play on user space, i'd rather use capset() to set all users capability to "all cap". that's the perfect equivalent.

so the user space solution (capset()) works, but then came the idea to optimize away. that's what blow everybody up. don't get me wrong, i always agree with rik farrow when he wrote in ;login: that we should build software with security in mind.

Quoted 6 lines

but i also hate bloat. lets not go to arm devices, how about a notebook. it's a personal thing, naturally to people who doesn't know about computer, personal doesn't go with multi user. by that i mean user with different capabilities, not different persons.

So don't install any services. The security in the kernel is not even bloat compared to some of the cruft that you can just not install.

Quoted 5 lines

- with that patch, people will still have authentication. so ssh for example, will still prevent illegal access, if you had an exploit you're screwed up anyway. sure httpd will give permission to everybody to browse a computer, but i don't think a notebook need to run it.

See above.

Quoted 5 lines

so i guess i deserve opinions instead of flames. the approach is from personal use, not the usual server use. if you think a server setup is best for all use just say so, i'm listening.

I have Linux on my PowerBook. I don't have sendmail, httpd, mysql, and a billion other 'server' processes running. Does that still make it a server?

We're not flaming (well some of us anyways). Just pointing out (loudly) where your thinking is flawed.

Quoted 2 lines

nah, performance was never my consideration. i do save about 3kb from my zImage, but i'm not interested.

But you just said you hate bloat. What other reason do you have for hating bloat?

Signature · 8 lines
=====================================================================
Mohammad A. Haque                              http://www.haque.net/ 
                                               [email protected]

  "Alcohol and calculus don't mix.             Project Lead
   Don't drink and derive." --Unknown          http://wm.themes.org/
                                               [email protected]
=====================================================================
#19 lore raw
In reply to [email protected]
Quoted 5 lines · [email protected] wrote:

so when everybody suggested playing with login, getty, etc. i know you have got the wrong idea. if i wanted to play on user space, i'd rather use capset() to set all users capability to "all cap". that's the perfect equivalent.

The linux kernel ought to be flexible, so most people can use it as-is. It can be used as-is for your purpose, and it have been shown that this offer more security _without_ inconvenience. Your patch however removes multi-user security for the many who needs it - that's why it never will get accepted. Feel free to run your own patched kernels - but your patch will never make it here.

Quoted 6 lines

so the user space solution (capset()) works, but then came the idea to optimize away. that's what blow everybody up. don't get me wrong, i always agree with rik farrow when he wrote in ;login: that we should build software with security in mind.

If you really want optimization, remove all security instead of merely killing a few basic tests.

Quoted 5 lines

but i also hate bloat. lets not go to arm devices, how about a notebook. it's a personal thing, naturally to people who doesn't know about computer, personal doesn't go with multi user. by that i mean user with different capabilities, not different persons.

The notebook user might not care or understand about multi-user security, but it is still useful. The user have several daemons running that he don't know about, they were installed by the distribution.
The security system can protect files from buggy or cracked daemons.

And protecting the
configuration (and essential stuff like the user's GUI) from being deleted by user accident is still a good thing.

The user who don't need password security can still have a "safe" SUID admin program for necessary tasks like changing the dialup phone number even though it resides in a protected file. So you definitely want the protection system, even in a "personal" appliance running linux. Because it protects against stupid mistakes like experimenting with editing files in the /etc directory on the notebook with a word processor. Users don't understand why saving in word processor format might be bad....

A notebook is a particularly bad example. Those with notebooks might not want to use passwords all the time, but it is very convenient if you have to leave a notebook with sensitive data with someone you don't trust. Business secrets or something as simple as a diary. This kind of users can be logged in all the time, mostly avoiding passwords. And log out
in those few cases they need to leave the machine in unsafe places.

Quoted 4 lines

i haven't catch up with all my mails, but my response to some:
- linux is stable not only because security.

Sure, but security definitely adds to its stability. Instead of nuking it all, just remove what bothers you. The security system has plenty to offer even when you skip the password part.

Quoted 7 lines

- linux was designed for multi-user, dos f.eks. is designed for personal use, so does epoc, palmos, mac, etc.
- i even use plan9 with kfs restrictions disabled sometimes, cause i don't have cpu server, auth server, etc.

- with that patch, people will still have authentication. so ssh for example, will still prevent illegal access, if

Nope. Someone ssh'ing into your system still cannot guess someone elses password. They can log in into their own account though, and abuse other users accounts or the machine configuration because there is no protection. Unprotected accounts only means you get your own account _by default_, you have the power to trash all the others. A malicious user could even change the other users passwords and re-enable the security system so they loose.

Quoted 1 line

you had an exploit you're screwed up anyway.

Many exploits are limited. Cracking a damenon running as "nobody" or some daemon user may not be all that satisfying - you might be unable to take over the machine. An exploit doesn't necessarily give root access.

Quoted 1 line

so i guess i deserve opinions instead of flames. the

You get a lot of opinions. Don't mistake them for flames just because they disagree with everything you say.

Quoted 3 lines

approach is from personal use, not the usual server use. if you think a server setup is best for all use just say so, i'm listening.

Multi-user security is useful for much more than server use. A good "personal" setup includes at least 3 users:
* root - for administration
* the user - for running the programs the user himself use. I.e. the word processor on a notebook, the user inteface on a linux phone, and so on.
* a nobody user, for safer daemons. If any kind of daemon is used at all. Surprisingly many appliances might run a daemon - a snmp daemon, or a webserver serving the same purpose (So your can check your home appliance from work perhaps)

Of course passwords can be skipped - maybe you don't worry about guests messing up your phone settings. Still, a buggy phone program shouldn't mess up other things. You don't want the browser on those fancy web-enabled cellphones to accidentally delete the address book due to some oddball bug or exploit.

Quoted 11 lines

i did say it clearly that i have other changes which i know won't be a clean patch (too many #ifdefs). f.eks. on my computer i didn't even compile user.c in, i don't have user_struct. filesystem and vfs code are affected by that patch already. memory access is important of course.

Quoted 1 line

Then you can try to show a measurable performance

difference.

nah, performance was never my consideration. i do save about 3kb from my zImage, but i'm not interested.

You don't want the performance _or_ less memory used. Why then do you want to optimize away the security system instead of merely changing the userspace configuration a bit?

If you optimize away security then you probably want to optimize away things like "login" as they are useless anyway with such a kernel. Much simpler to remove only "login" then.

Helge Hafting

#20 lore raw
Quoted 7 lines · Helge Hafting

The linux kernel ought to be flexible, so most people can use it as-is. It can be used as-is for your purpose, and it have been shown that this offer more security _without_ inconvenience. Your patch however removes multi-user security for the many who needs it - that's why it never will get accepted. Feel free to run your own patched kernels - but your patch will never make it here.

i don't understand, that patch is configurable with 'n' as default, marked "dangerous". so somebody who turned on that option must be know what he's doing, doesn't understand english, or has a broken monitor.

Quoted 2 lines

If you really want optimization, remove all security instead of merely killing a few basic tests.

those tests responsible for almost all EACCESS & EPERM.

Quoted 6 lines

The notebook user might not care or understand about multi-user security, but it is still useful. The user have several daemons running that he don't know about, they were installed by the distribution.
The security system can protect files from buggy or cracked daemons.

must be a devil cursed distro, distributing "single-user" kernel with live daemons. a division of redmon?

Quoted 4 lines

And protecting the
configuration (and essential stuff like the user's GUI) from being deleted by user accident is still a good thing.

Quoted 9 lines

The user who don't need password security can still have a "safe" SUID admin program for necessary tasks like changing the dialup phone number even though it resides in a protected file. So you definitely want the protection system, even in a "personal" appliance running linux. Because it protects against stupid mistakes like experimenting with editing files in the /etc directory on the notebook with a word processor. Users don't understand why saving in word processor format might be bad....

hmm, the other thing i hate is policy. ever consider that you're talking policy? maybe reboot() should sync() first?

Quoted 8 lines

A notebook is a particularly bad example. Those with notebooks might not want to use passwords all the time, but it is very convenient if you have to leave a notebook with sensitive data with someone you don't trust. Business secrets or something as simple as a diary. This kind of users can be logged in all the time, mostly avoiding passwords. And log out
in those few cases they need to leave the machine in unsafe places.

and that someone who had the notebook can't access sensitive data without a passwd?
that's what i'm trying to say. if you carried your server, and leave it in unsafe places, why would anybody try to crack it? just get the harddisks put it in another computer, voila. so much for security.

Quoted 2 lines
Quoted 1 line

- linux is stable not only because security.

Sure, but security definitely adds to its stability.

i don't know what you mean by stability. if you meant linux can run a year without a reboot, what security has anything to do with stability? the kernel is stable, yes, do we here linux server got cracked yes, it's still stable though.

Quoted 11 lines
Quoted 2 lines

- with that patch, people will still have authentication. so ssh for example, will still prevent illegal access, if

Nope. Someone ssh'ing into your system still cannot guess someone elses password. They can log in into their own account though, and abuse other users accounts or the machine configuration because there is no protection. Unprotected accounts only means you get your own account _by default_, you have the power to trash all the others. A malicious user could even change the other users passwords and re-enable the security system so they loose.

i didn't disable password! if someone got into a personal machine through ssh by guessing, most likely that account is the owner's. who else?

Quoted 6 lines
Quoted 2 lines

you had an exploit you're screwed up anyway.

Many exploits are limited. Cracking a damenon running as "nobody" or some daemon user may not be all that satisfying - you might be unable to take over the machine. An exploit doesn't necessarily give root access.

that line was still about ssh. besides, if someone would run a server for the world, then he must had drain bamage.

Quoted 2 lines

You get a lot of opinions. Don't mistake them for flames just because they disagree with everything you say.

you haven't seen my inbox.

Quoted 11 lines

Multi-user security is useful for much more than server use. A good "personal" setup includes at least 3 users:
* root - for administration
* the user - for running the programs the user himself use. I.e. the word processor on a notebook, the user inteface on a linux phone, and so on.
* a nobody user, for safer daemons. If any kind of daemon is used at all. Surprisingly many appliances might run a daemon - a snmp daemon, or a webserver serving the same purpose (So your can check your home appliance from work perhaps)

but think about the idea of multi-user. it means protection for the system and other users. that's a typical server needs.

and how about notebook? i can see that it need authentication to use the system. does the user need to be protected from other users? there's nobody else. well, maybe, like we all used to, that user needed to protect him from himself.

so, system authentication is needed for both single-user and multi-user. (let alone physical access)
user account authentication is certainly not needed for single- user case.

Quoted 6 lines

Of course passwords can be skipped - maybe you don't worry about guests messing up your phone settings. Still, a buggy phone program shouldn't mess up other things. You don't want the browser on those fancy web-enabled cellphones to accidentally delete the address book due to some oddball bug or exploit.

and you're hoping program with root suid will run perfectly?

Quoted 8 lines

You don't want the performance _or_ less memory used. Why then do you want to optimize away the security system instead of merely changing the userspace configuration a bit?

If you optimize away security then you probably want to optimize away things like "login" as they are useless anyway with such a kernel. Much simpler to remove only "login" then.

i wish it was only "a bit". what i want is to have all process flags have PF_SUPERPRIV, but users still own their own uid. doing it in userspace means i had to change this login, my friend had to change that login, maybe this shell, that shell...

that's my setup. i still use login, so only those who i trust can use my machine, yes my trusted user can do anything, but hey it isn't a server. it's a workstation.

imel

#21 lore raw
Quoted 4 lines · [email protected] wrote:

i don't understand, that patch is configurable with 'n' as default, marked "dangerous". so somebody who turned on that option must be know what he's doing, doesn't understand english, or has a broken monitor.

This is a very marginal thing that very few people will want or need. (You may think it is nifty - but we disagree on that)
If everybody get their favourite patch in with a config option then we get a huge amount of config options, and maintainig the kernel will be much harder because there is thousands of ifdefs for all sorts of rare stuff. There will be your 5 ifdefs, and 26000 other people's 5 ifdefs. Someone making a change will have to check if it works, but will it work with all sorts of combinations of config options? What if someone makes a change that works fine, but makes the kernel uncompileable if your option is turned on? This guy didn't check your config option because he never use it himself...

The maintainability issue is why kernel patches usually aren't accepted when the problem can be solved by changing the userspace configuration instead. (In your case by sybstituting "bash" for "getty" in /etc/inittab) This is the case even with very good things - fsck is a userspace program even though it is necessary for any system with
a writeable filesystem.
You have another problem with the way all the leading developers dislike your idea - buteven trying to convince them is useless as you _still_ run up against "this feature is _easily_ done in userspace"

Quoted 4 lines
Quoted 2 lines

If you really want optimization, remove all security instead of merely killing a few basic tests.

those tests responsible for almost all EACCESS & EPERM.

Sure, but now you have a lots of if(1) {something} else {other thing} and a better optimization would be to get rid of the entire test. There is a lot of errors that can't happen with your patch, so you really ought to remove the error handling cases too if optimization is what drives you.

Quoted 10 lines
Quoted 7 lines

The notebook user might not care or understand about multi-user security, but it is still useful. The user have several daemons running that he don't know about, they were installed by the distribution.
The security system can protect files from buggy or cracked daemons.

must be a devil cursed distro, distributing "single-user" kernel with live daemons. a division of redmon?

Is there something you don't understand, or do you really want to run one process at a time?

You were talking about how a notebook is a personal thing, with only one user. Well, the notebook user do of course want to do a bunch of nifty things like read email on the thing. Guess what, you need an email daemon for that! And many users don't want to know the details of setting up an email daemon, so the distribution install one by default. This kind of users would be outraged if the distribution didn't - "what - I have to install more stuff just to get my mail! windows do that out of the box why is this so difficult..."

There are several other examples of things users _expect_ from a notebook, which just happens to include a daemon process running under a different user-id for safety reasons. (For example the print spooler daemon. Users want to print, and unix is nice in that you don't have to wait for the printer - you can go on editing something else while the printer slowly does its work thanks to the print spooler daemon. This one is installed by default too.)

They only ever _log in_ as one user, so the login prompt can safely be eliminated in order to avoid the password hassle. But you still want the multi-user security.

Please try to understand that the kernels concept of a "user" don't mean a "person"! There is only one "person" using his/her very personal device - the unix concept of users is a file security thing. You don't want an error in the mail software to use up all the diskspace or overwrite your word processing files. And you don't want a printer driver problem to mess up your mail or your personal files.

All these little things is included in good distributions, and they don't cause serious trouble because they are all protected against each other. Your machine is multi-user even if it is strictly single-person!

If all this is new to you, please read up on unix before suggesting too much. _Uninformed_ patches easily becomes a nuance, good patches is usually written by people who know very well what they work on. Excellent knowledge of C isn't enough.

Quoted 15 lines
Quoted 5 lines

And protecting the
configuration (and essential stuff like the user's GUI) from being deleted by user accident is still a good thing.

Quoted 9 lines

The user who don't need password security can still have a "safe" SUID admin program for necessary tasks like changing the dialup phone number even though it resides in a protected file. So you definitely want the protection system, even in a "personal" appliance running linux. Because it protects against stupid mistakes like experimenting with editing files in the /etc directory on the notebook with a word processor. Users don't understand why saving in word processor format might be bad....

Quoted 2 lines

hmm, the other thing i hate is policy. ever consider that you're talking policy? maybe reboot() should sync() first?

Reboot does indeed sync on my machine. That isn't the problem here. The problem I described is that a clueless user might try to change a config file with his word processor and save it in word-processor format. This renders the file useless as the programs who depend on config files don't understand word processor format. A "nice synced reboot" won't help here, as the ruined file is then saved properly, but the contents are wrong.
A good security system fixes this though - a user simply isn't allowed to mess up the config files in arbitrary ways. The user is protected against his own lack of knowledge or common sense. Similiar, a car engine don't explode if you press the accelerator too hard at low RPM's. (A racing car might - but racers _know_ how to handle such cars) Unix appliances and personal computers need similiar protection and the security system is used for that.

So what if the user _need_ to change the configuration? Experts do this by using the root password. Non-experts do this by running a safe config program. The config program is SUID (it is privileged and auto-switch to the root user) and it contains all sorts of undo options and safety checks so the user cannot screw things up badly.

Quoted 9 lines
Quoted 8 lines

A notebook is a particularly bad example. Those with notebooks might not want to use passwords all the time, but it is very convenient if you have to leave a notebook with sensitive data with someone you don't trust. Business secrets or something as simple as a diary. This kind of users can be logged in all the time, mostly avoiding passwords. And log out
in those few cases they need to leave the machine in unsafe places.

Quoted 6 lines

and that someone who had the notebook can't access sensitive data without a passwd?
that's what i'm trying to say. if you carried your server, and leave it in unsafe places, why would anybody try to crack it? just get the harddisks put it in another computer, voila. so much for security.

An encrypted filesystem prevents even this - a cracker _can't_ read that disk without the password. Because there is no workaround even with a purpose-built "nasty" kernel.

This is usually overkill though. Normal users worry about things like your sister taking a sneak look at your diary (stored in a word processor file). Or in a business
setting: A visitor taking a look at your business documents while you fetch the coffee for the meeting. He don't have time for disassembling the computer, but he have time
for reading a few documents. A simple password protect against this.

What I do here is providing a few (of many) examples of why passwords sometimes is a good idea even on personal machines.

Quoted 21 lines
Quoted 2 lines
Quoted 1 line

- linux is stable not only because security.

Sure, but security definitely adds to its stability.

i don't know what you mean by stability. if you meant linux can run a year without a reboot, what security has anything to do with stability? the kernel is stable, yes, do we here linux server got cracked yes, it's still stable though.

Quoted 11 lines
Quoted 2 lines

- with that patch, people will still have authentication. so ssh for example, will still prevent illegal access, if

Nope. Someone ssh'ing into your system still cannot guess someone elses password. They can log in into their own account though, and abuse other users accounts or the machine configuration because there is no protection. Unprotected accounts only means you get your own account _by default_, you have the power to trash all the others. A malicious user could even change the other users passwords and re-enable the security system so they loose.

Quoted 3 lines

i didn't disable password! if someone got into a personal machine through ssh by guessing, most likely that account is the owner's. who else?

Yeah, but what if there are several accounts?
I have an account on my machine where strangers can log in in order to play network games. There is a password, but it is handed out to a bunch of strangers I have no particular reason to trust. This is not a problem as my machine has multi-user security. They cannot delete _my_ files, or even read them, even though they can log into the machine. They cannot even run other programs than the games, as they have no permission.

That wouldn't work with your patch. They would still only be able to log into their own game account because of the password security in ssh, but they definitely _would_ be able to delete _my_ personal account without logging into it - because you disabled security. And they could delete the email software so I don't get mail. And they could delete the system software and the kernel, so the machine goes down and remains down. Security prevents all that.

Quoted 11 lines
Quoted 7 lines
Quoted 3 lines

you had an exploit you're screwed up anyway.

Many exploits are limited. Cracking a damenon running as "nobody" or some daemon user may not be all that satisfying - you might be unable to take over the machine. An exploit doesn't necessarily give root access.

that line was still about ssh. besides, if someone would run a server for the world, then he must had drain bamage.

Not at all. There are many game servers out there where people all over the world can log in and run games. But security prevents them for doing anything _but_ running games, so there is no problem. And there are anonymous ftp server where anybode can log in...

Quoted 4 lines
Quoted 2 lines

You get a lot of opinions. Don't mistake them for flames just because they disagree with everything you say.

you haven't seen my inbox.

No, I just see the replies on the linux kernel list. A few flames, and lots of people who merely disagree and try to tell why they thing your patch and the idea behind it is no good.

Quoted 20 lines
Quoted 11 lines

Multi-user security is useful for much more than server use. A good "personal" setup includes at least 3 users:
* root - for administration
* the user - for running the programs the user himself use. I.e. the word processor on a notebook, the user inteface on a linux phone, and so on.
* a nobody user, for safer daemons. If any kind of daemon is used at all. Surprisingly many appliances might run a daemon - a snmp daemon, or a webserver serving the same purpose (So your can check your home appliance from work perhaps)

but think about the idea of multi-user. it means protection for the system and other users. that's a typical server needs.

and how about notebook? i can see that it need authentication to use the system. does the user need to be protected from other users? there's nobody else. well, maybe, like we all used to, that user needed to protect him from himself.

Quoted 4 lines

so, system authentication is needed for both single-user and multi-user. (let alone physical access)
user account authentication is certainly not needed for single- user case.

As I have said already in this mail - a single-user computer need the security system to protect the _system software_ from the (single) user, and to protect the _users_ files from possible bugs in various system software like mail software, printer drivers and various other things the user might want to run. Some people thinks it is cool to run a webserver on their office machine, even if it only servers one page. Security makes things like that safer too.

You may or may not need authentication. But you definitely want security.
A newbie experimenting might stumble upon the command that deletes all files on the system. That won't kill a unix system, the user will merely loose his own personal files, because he has no permission for deleting system files.

Try this on a windows 98 box (which don't have security, exactly the way you tried to patch the kernel) and everything goes, or at least a lot of things until the system crashes due to a missing system file. The system migh be unable to delete everything, but it will certainly manage
to delete enough to amke the machine unuseable and in need of a reinstall.

This protection system ensures that you typically never need to reinstall linux due to user mistakes. (You may want to upgrade often to get the latest software - but you don't need to reinstall to get a upgrade.)

Quoted 18 lines
Quoted 6 lines

Of course passwords can be skipped - maybe you don't worry about guests messing up your phone settings. Still, a buggy phone program shouldn't mess up other things. You don't want the browser on those fancy web-enabled cellphones to accidentally delete the address book due to some oddball bug or exploit.

and you're hoping program with root suid will run perfectly?

Quoted 8 lines

You don't want the performance _or_ less memory used. Why then do you want to optimize away the security system instead of merely changing the userspace configuration a bit?

If you optimize away security then you probably want to optimize away things like "login" as they are useless anyway with such a kernel. Much simpler to remove only "login" then.

Quoted 8 lines

i wish it was only "a bit". what i want is to have all process flags have PF_SUPERPRIV, but users still own their own uid. doing it in userspace means i had to change this login, my friend had to change that login, maybe this shell, that shell...

that's my setup. i still use login, so only those who i trust can use my machine, yes my trusted user can do anything, but hey it isn't a server. it's a workstation.

I wonder what you want different UID's for then. Why not set all the UID's to 0, you'll still have separate usernames, and separate home directories so you can have different setups and so on. I cannot really see what you use the UID's for?

But still, if you want everybody to be able to do anything _and_ have separate UIDS, try userspace options. Possible ways:

1. A single chmod command can make all files available. Something like "chmod -R /* oug+rwx" You may have to do that as root - you won't need root later.

2. Or run a similiar chmod on /bin, /sbin and /usr/bin that makes all executables SUID. power to the users!

3. Or let everybody be members of group 0 and run
"chmod g+rwx /* -R"

There's a multitude of easy ways to achieve the behavior you want without patching the kernel. That alone _ensures_ it won't get into the kernel. They way people dislike your idea don't make it better, but it wouldn't help even if you could convince them.

Many newbie ideas are met with "no, the idea is bad but you can achieve what you want another way, like this:"

Smart people learn from such things. Some other
keep arguing when there is nothing to achieve.
If you're a unix newbie, please consider the option that the other people may have a point. They know things you don't, so asking "why" is a much better idea than insisting on an unpopular idea. They don't all take the time
to explain why they don't like your patch, that don't mean they don't have solid reasons.

Helge Hafting

#22 lore raw
In reply to [email protected]
Quoted 8 lines · Helge Hafting

You were talking about how a notebook is a personal thing, with only one user. Well, the notebook user do of course want to do a bunch of nifty things like read email on the thing. Guess what, you need an email daemon for that! And many users don't want to know the details of setting up an email daemon, so the distribution install one by default. This kind of users would be outraged if the distribution didn't - "what - I have to install more stuff just to get my mail! windows do that out of the box why is this so difficult..."

You don't need to be running an e-mail daemon just to read e-mail.

Signature · 8 lines
=====================================================================
Mohammad A. Haque                              http://www.haque.net/ 
                                               [email protected]

  "Alcohol and calculus don't mix.             Project Lead
   Don't drink and derive." --Unknown          http://wm.themes.org/
                                               [email protected]
=====================================================================
#23 lore raw
In reply to Helge Hafting
Quoted 9 lines · On 2001.04.26 13:31:54 +0200 [email protected] wrote:
Quoted 7 lines · Helge Hafting

The linux kernel ought to be flexible, so most people can use it as-is. It can be used as-is for your purpose, and it have been shown that this offer more security _without_ inconvenience. Your patch however removes multi-user security for the many who needs it - that's why it never will get accepted. Feel free to run your own patched kernels - but your patch will never make it here.

Quoted 4 lines

i don't understand, that patch is configurable with 'n' as default, marked "dangerous". so somebody who turned on that option must be know what he's doing, doesn't understand english, or has a broken monitor.

I can make a virus, patch the kernel and send it in, with a 'N' by default. But what is the use of this? Do you think this will be implemented???

Your thing is as dangerous as a virus, basically. It gives root to everyone, although they have separate UIDs. And whenever there is a way out (i.e. surfing the web, reading mail), there is a way in. So that would make your system a very nice target to hack -> since you basically are root this means they can change anything as soon as they have access. If you're not root, they can't, since they can only do what you as a user can do. The whole goal of your patch is to make computer life easier. This patch doesn't do that - it goes far worse. We gave you a few suggestions on better/easier ways to accomplish this goal - take them as advice and use them instead.

Easy: chmod -R 777 / (same risk, though)
Good: use su for installing software (su -c "make install")

Can't get much easier than that (and if a clueless user needs to do this, let him use redhat's RPM manager, "enter your password" with a nice X-window, and press that button "install" - same effect)...

You don't need to patch the kernel for this...

--
Ronald Bultje