--- Topic for #linux is Next talk: "The Hurd: a GNU approach to OS design Friday Dec 13, at 21:00 GMT Wolfgang Jaehrling | Comments #qc; Spanish #redes; Dutch #taee
sarnold	wolfgang is going to be giving his presentation about GNU HURD; he first discoveredthe HURD in 1999 and has given several talks on it since then
sarnold	please direct your questions to #qc, and find dutch translation in #taee; maybe spanish in #redes
sarnold	thanks
wolfgang	may i? :)
wolfgang	i am going to talk about the Hurd, and will explain why GNU develops it, even if we already have a very mature and popular kernel for the GNU system, which of course is Linux
wolfgang	the Hurd, however, is not a kernel.  together with the GNU C library (glibc) and the GNU Mach microkernel, it provides the functionality that traditionally is provided by a Unix kernel.
wolfgang	these three components together with the rest of the GNU system (which includes gcc, GNOME, etc. and even non-GNU software like X11) they provide a POSIX-conformant operating system.
wolfgang	at this point, a few notes on terminology:
wolfgang	the GNU Hurd (or simply the Hurd) is the name for one part of the system.  GNU/Hurd is the name for the whole system.  you can think of "GNU/Hurd" as "The Hurd-based GNU-system"
--> Luiso42 (Luis@199.243.77.79) has joined #linux
wolfgang	basically, this system works as follows:
<-- E0x has quit (Ping timeout for E0x[182stb59.codetel.net.do])
--> manaha (manaha@213.37.56.168) has joined #linux
wolfgang	applications from the Unix-world (i will call them POSIX-applications) use functions in the C library where they would use Unix kernel-calls on a traditional Unix.  these are basic functions like read(), write(), fork() etc.
<-- MiTo has quit (Ping timeout for MiTo[cm074-pop12.bbt.net.ar])
wolfgang	the C library communicates with the Hurd-servers to implement these functions
<-- Telemako (jpardo@231-CORU-X6.libre.retevision.es) has left #linux
--> kheb (kheb@200.78.66.187) has joined #linux
wolfgang	but the Hurd-servers are not special in any way, they are just normal programs.  so they can also use the C-library.
wolfgang	you can think of these Hurd-servers as Daemons.
wolfgang	we have a server that implements the ext2 file system
--> _bvc (~b@M465P001.adsl.highway.telekom.at) has joined #linux
wolfgang	another one implements the TCP/IP protocol
wolfgang	we even have a server that just implements /dev/null
wolfgang	a short note about POSIX conformance:
wolfgang	some people think that even Windows NT is POSIX conformant.
wolfgang	that isn't really the case.  it only implements some _very_ basic parts of POSIX
wolfgang	where GNU/Hurd is completely POSIX-conforming.  well, module bugs of course :)
wolfgang	this is a very important point: we are still compatible, so you can use all the free programs you like on GNU/Hurd.  at least if they are written somewhat portably
wolfgang	so, GNU/Hurd is a combination of complete compatiblity and new concepts
wolfgang	but know about the new concepts:
wolfgang	oh, wait i skipped something ;)
wolfgang	i skipped a lot.  i first wanted to talk about the history and motivation behind the Hurd
wolfgang	as you will probably know, Richard Matthew Stallman started to develop the GNU system in 1984
wolfgang	the primary goal was to be free; to be superior to proprietary systems morally, socially and ethically
wolfgang	but it was also a goal to be technically better
wolfgang	if you compare GNU software with Unix software, you are likely to notice that.  Unix software often has arbitrary limits.  for example, in many Unix tools, input lines longer than 1024 characters are silently truncated.
wolfgang	ok, we have a question here:      <riel> how about running non-free software on the hurd ?
wolfgang	it is possible to run non-free software on GNU/Hurd, but i hope nobody actually wants that.
wolfgang	we are not completely binary-compatble with GNU/Linux, so you cannot use most non-free programs currently.  there are some technical reasons why it might be problematic to be ABI (Application Binary Interface)-compatible with it, i can say more about that later if there is interest in it.
wolfgang	we have another question:    <sarnold> well, being able to, and being legally allowed to, are different things.. is the lgpl-ness of glibc sufficient to shield all programs from GPL's "linking" clause (for using the hurd servers)?
wolfgang	yes, it is enough.  you can even communicate with the Hurd servers directly, since you use Remote Procedure Calls to communicate wit them.  so there is no linking involved.
wolfgang	ok, back to the original topic: the arbitrary limitations of Unix programs and how GNU wants to do better
wolfgang	in fact, the whole design of the Hurd is based on the idea of not imposing arbitrary limitations on users
wolfgang	so if you have an ISO-image of a cd-rom, you can "loopback-mount" the cd as the user without needing special permission.  that is no problem, since the iso9660fs-server will run with your user-id.  it is just a program that you started.
wolfgang	but you can do far more than just that.
wolfgang	in fact, you can change all aspects of the system as you like.  you can create your own world, using POSIX-components where you like and your own components otherwise.
wolfgang	we have a question:   <sarnold> are there security problems with allowing users to modify the filesystem namespace? or is the user performing the mount the only user who can see the newly added filesystem mount?
wolfgang	there is a lot i could say about that :)
wolfgang	and he goes on:   <sarnold> if so, what happens if two users both try to mount to the same mount-point? (well, that sounds like trouble anyway...)
wolfgang	i will answer the second question first, as that is easier :)
wolfgang	you can only "mount" something on a node that you own.  so two users trying to use the same node is impossible
wolfgang	that partially answers the first question even.
wolfgang	you can only use that in your home directory (or in places like /tmp or /var/tmp of course).  but that is not a restriction, since you can also start your own copy of the complete Hurd, so you have your own root file system (which others would not see) which you can modify as you like
wolfgang	we've got a short interruption here:   <Arador> and can you mount something (a iso image in loop device for example) for N users?
wolfgang	i am not aware of any simple way to do that currently.  conceptionally, it is definitely possible if you setup your own authentication server.  but that is some work.  but it is possible.
wolfgang	simplifying everything that anyone might ever want to do is a task we will start to work on at some point in the future ;)
wolfgang	back to the question about security issues.
wolfgang	since you can put arbitrary stuff in the part of the file system that you own, you might be able to get someone into a trap if he accesses your part of the file system.  for example, if you put a file system in your homedir that is purely virtual (think of /proc on GNU/Linux) and creates an endless directory tree dynamically, it might be hard for the admin to remove your home diretory when he deletes your account. ;)
wolfgang	so he would have to remove all your own file systems first.
wolfgang	if he does that, all is fine.
wolfgang	that is a simple fact that you have to pay when you allow users to change their environment.  people just need to be aware of the fact that it is possible to change the environment.
wolfgang	but i guess a person who does not such a basic thing about the system better should not administrate a server with multiple users anyway. :)
wolfgang	a question:   <sarnold> is there an easy way for the sysadmin to prevent users from mounting new filesystems, to prevent someone from doing while true ; do mount ; done ?
wolfgang	if you start a file system server, it runs as a process of you.  so this question is basically the same as "can the admin stop the users from starting a lot of processes".  and of course that is possible.  it is even possible on Unix to do that.  i am not sure inhowfar it is implemented on GNU/Hurd yet, though.
wolfgang	if want to _generally_ prevent your users from using their own servers, that is very easy.
wolfgang	use a file system for /home that does not support this.  you can create such a file system very easiely.  in fact, some people do so by accident when installing Debian GNU/Hurd and then wonder why some stuff does not work ;)
wolfgang	so, GNU/Hurd tries to not impose limits on the user if that can be avoided.  it is worse enough that the hardware is limited, why should we add additional restrictions in the software? :)
wolfgang	(except for the cases where we _want_ to limit things on purpose for security reasons)
wolfgang	and there is another very cute aspect of the system: the Unix philosophy says that one should write small and simple programs which can be combines flexibly.  that is how it was done in the Hurd; the Unix kernel in fact violates this principle, as it does scheduling, hardware support, file systems, network protocols and lots of other stuff
wolfgang	that does not mean that the designers of Unix were stupid, of course :)  they were actually very clever
wolfgang	they had to design Unix for computers that were less powerful than a typical calculator is today.  so it was simply necessary to do it that way
wolfgang	in the Hurd, we think that now it is time to rethink these old decisions, since so much has changed since then.
wolfgang	back than, user-defined file systems were not feasable at all, so nobody asked for it.  but today, everywone wants them, and their are various approaches to providing them.
wolfgang	the GNOME-vfs is one example. KDE's ioslaves are another.  libferris is yet another
wolfgang	all these have something in common: they are doing things in a way that limits their usefulness: you have to write your programs to make use of them.  existing programs won't benefit
wolfgang	and that is the major advantage of the Hurd: we do it at the file system level, so you do not need to modify any POSIX application to enable it to benefit from the features of the Hurd!
* wolfgang finds this so exciting
wolfgang	ok, now some more concrete examples
wolfgang	sure, we do have file systems like ext2fs, ufs and fatfs.
wolfgang	and additionally to these stored file systems, we also have network file systems like nfs and ftpfs
wolfgang	the latter is very nice.  you can just do "tail -n 2 /ftp/ftp.debian.org/pub/welcome.msg" and stuff like that.
wolfgang	some people seem to object to that, arguing "some operations like appending to a file will be terribly inefficient over ftp".  that is true, but if you consider that a problem, don't use this feature. :) often, it is useful to have it, though.
wolfgang	we have a question:    <Arador> how does ftpfs handles things such as "password incorrect" or "timeout" to the POSIX applications?
wolfgang	well, additionally to these file systems, we can also have virtual ones like GNU/Linux has /proc and devfs
wolfgang	but there is far more than that.
wolfgang	a server can provide just a single node, like /dev/null or /dev/random.
wolfgang	and we even have one particular cute one: the "run" translator.  (oh, i almost forgot to mention that servers that reside in the file system are also called translators)
wolfgang	when starting "run", you provide it with a shell command.  whenever you open the file where "run" is sitting, it will execute that command, and when you then read from it you will get the output of that command.
wolfgang	so you can tell it to execute "fortune", every "cat foo" will display a new fortune cookie.
wolfgang	(that can be used to generate random signatures for emails - or for whatever you can think of.)
wolfgang	<vegai> wolfgang: in that example, 'foo' is a translator, right?    <vegai> the run-translator, that is
wolfgang	yes, it is the node where the run translator sits, i.e. the "mountpoint"
wolfgang	all that may be nice, but there are still far more possibilities!
wolfgang	a translator does not need to implement the file system protocol.  it can implement any protocol.
wolfgang	for example, the IP stack implements the socket interface, obviously
wolfgang	<sarnold> does the translator run with the privs of whoever ran "run" or whoever performs the read(2)?
wolfgang	a translator always runs with the priviledges of the owner of the node which it translates (the "mountpoint")
wolfgang	so the permissions need to be right, as always.
wolfgang	there is a simple reason for that.
wolfgang	if iso9660fs translates /cdrom (or /mnt/cdrom if you prefer), then it needs to access /dev/cd0.  same for any other file system: if you have an ext2fs on /var, that server needs to be able to write to e.g. /dev/hd0s3.  and of course users should not be able to use this directly.
wolfgang	ok, the last thing i mentioned was the IP stack, which provides the socket interface (and is available at /servers/socket/inet and - if you want to go by protocol number instead of name - /servers/socket/2).
wolfgang	of course you are not limited to the interfaces that the Hurd uses.  you can use your own interfaces, and just use the file system as namespace for the initial handshake between a server and its users.
wolfgang	just think for yourself a bit about what you can do with all that.  i am sure you will be able come up with some nifty ideas.  no matter whether they are virtual file systems, single nodes or something completely different
wolfgang	a question:     <vegai> what about language independence? Is C the only choice for coding translators?
wolfgang	long ago, someone has written a module for writing translators in perl (and implemented one with it).  later, i wrote a module that enables you to write translators in Ruby (and i implemented a translator with it, but not a useful one ;-)).
wolfgang	but in general, a lot of work is still needed to make it convenient to write arbitrary translators in any language.
wolfgang	i am very interested in that myself, but currently other things are certainly more urgent by far
wolfgang	so if anyone is interested in working on that, don't hesitate to contact me.
wolfgang	<sarnold> if you don't mind (heh heh, you must be tired of me :) what tops the list of what needs doing?
wolfgang	i am not sure if that list fits into memory :)
wolfgang	well, i think we have some lists online even, wait a second
wolfgang	a few very urgent things are listed at http://hurd.gnufans.org/bin/view/Hurd/GNUHurdStatus
wolfgang	we have here a very interesting question:    <mentor> AFAI can see: essentially HURD translators/servers provide a service in a different context (processor context, security context, and any other contexts the microkernel maintains); would it ever be considered useful to implement services in such a way that specfic contexts can be chosen for segregation?
wolfgang	well, translators are far more than that :) then enable you to change your environment as you like (and experienced users usually want to change their working environment to fit their needs better), but sure, the seperation certainly is also a point.  however, i am not sure a) how would you propose could be implemented at all, no matter whether in the Hurd or elsewhere, and b) what would be the advantage of that approach.
wolfgang	ok, then i will just give on more example of a cool feature.  and a very new feature in fact
wolfgang	the new console implementation
wolfgang	now, we have a console server (a translator) that provides a directory for every virtual console.
wolfgang	you can attach a console client to it which sends input to the "input" file in the directory of a virtual console, and gets the contents of the display from the file "display" in the same directory
wolfgang	the client does not need to poll the data from the display: it just requests change notifications, a feature that is provided also by normal stored file systems like ext2fs.  that might be quite interesting for implementing an Intrusion Detection System, BTW.
wolfgang	the nice thing about this console implementation is that you can attack multiple clients to the same server
wolfgang	so you can work at a computer, then go to another computer, ssh over the network to the first one and be able get your original consoles back - all of them.
wolfgang	i don't think i need to mention that the console supports unicode everywhere etc.
wolfgang	i think that the new console is a very nice example for how the Hurd design can be used to implement new powerful stuff.
wolfgang	ok, now i will conclude by referring to http://hurd.gnu.org/ , http://www.debian.org/ports/hurd , http://hurd.gnufans.org/ and of course http://hurd.es.gnu.org/ for further information.  any more questions?
* wolfgang could keep talking about the Hurd for several days :)
wolfgang	<vegai> I'm under the impression, that the current microkernel is limited in several ways, and it's being replaced by L4. What will this cause to the current Hurd?
wolfgang	Mach was a first-generation microkernel, so it is certainly not perfect.
wolfgang	indeed, there are plans to switch to L4 at some point in the future, but we are far away from that.  the Hurd is cutting-edge even when running on Mach ;)
wolfgang	for the Hurd, it will mean that it will become better, hopefully.  and we will have to change some Mach-specific parts.  fortunatly, we depend on Mach-semantics in very few places.
* jmgv thinks wolfgang need a breathle ;-)
bunny	so...
bunny	GNU/Hurd be part of it ;)
bunny	#hurd the channel, #hurd-bunny the underground channel ;)
sarnold	ok, i think it is probably time to move onto our next speaker, xtingray, who will talk about php and flash :)
sarnold	I'd like to thank wolfgang for his extreme patience in answering my questions :)
bunny	and me for bringin him here ;D
wolfgang	we're used to being patient in the Hurd =)
jmgv	so we will consider this talk finish
sarnold	I'd also like to that tiri, garoeda, and arador, for trnaslating this talk into dutch and spanish :)
jmgv	clap clap clap clap clap
jmgv	clap clap clap clap clap
wolfgang	:)
bunny	and me?
bunny	=(
sarnold	bunny: yes, thanks to you too :)
jmgv	plas clap plas clap plas clap
wolfgang	bunny: thank you for bringing me here

Generated by irclog2html.pl 2.1 by Jeff Waugh - find it at freshmeat.net!