Re: [9fans] novel userspace paradigms introduced by plan 9
Which distro's version of "text utils" ? hehe .. Let's not start an anti-linux flame war now ... On Sun, Jul 3, 2011 at 1:31 AM, Lyndon Nerenberg (VE6BBM/VE7TFX) < lyn...@orthanc.ca> wrote: > > I don't see any meaning in Linux "adopting" some set of plan 9 > > commands... > > Have you read the source code for their cat(1) ??? > > >
Re: [9fans] Survey: Current Fossil+venti Filesystem
On Wed Jun 22 16:10:43 BST 2011, fors...@terzarima.net wrote: > I can't remember ever having lost data with fossil+venti, > on two complete networked systems that have been running > on fossil/venti since 2004 and 2005, I am in a very similar situation, two servers since 2004 and have not lost anything. There is a bug in the ephemeral snapshot code in fossil which used to cause my server to freeze about once a month, however I never lost any data; I have since turned off ephemeral snapshots and have had no problems since then - its a pity though. I have had problems with disks dieing and always run mirrored pairs of disks from different manufacturers (in the hope that they might fail on different days :-). -Steve
Re: [9fans] novel userspace paradigms introduced by plan 9
Hey, On 2 July 2011 19:36, dexen deVries wrote: > linux'c `clone()' syscall (the underpinnings of fork()) actually do accept > CLONE_NEWNS, CLONE_NEWNET, CLONE_VM and other flags, pretty close to p9's. Yeah, clone() is afaik compatible with rfork(), so long as you have CAP_SYS_ADMIN. Similarly mount() and bind(). > afaik, x11 is considered an afterthought, bolted onto POSIX systems, and thus > not integrated all that well. I think what I'd say is the most "novel userspace paradigm" in Plan 9 is its pervasive synthetic filesystems. You have FTP filesystems and so on with FUSE now, but writing something as flexible (technically) as Rio still requires something other than FUSE. But more importantly, since Plan 9 *started* with those synthetic filesystems they're used everywhere, whereas they're pretty uncommon in Linux etc. It would be nice if web browsers used a kind of webfs, and so on. It's unfortunate that clients for dedicated filesystems, like Rio and Acme, need to understand the layout of the directory tree, but that's difficult to work around. Still, FUSE has extended attributes, so you could e.g. configure a window manager just by setting attributes on the 'window manager filesystem' root directory. I know bloated GNU projects are generally frowned upon, but I think it's quite interesting that GNOME's GVFS allows, afaict, per-process synthetic filesystems. But clearly that's extremely ugly compared to Plan 9. On 3 July 2011 00:31, Lyndon Nerenberg (VE6BBM/VE7TFX) wrote: > Have you read the source code for their cat(1) ??? You know Linux != GNU, right? cls
Re: [9fans] novel userspace paradigms introduced by plan 9
> I know bloated GNU projects are generally frowned upon, but I think > it's quite interesting that GNOME's GVFS allows, afaict, per-process > synthetic filesystems. But clearly that's extremely ugly compared to > Plan 9. and yet there's a key difference. this is a private joke amongst gnome processes. i can give "file" references to gnome programs like http://example.com to a gnome proc. cat(1) won't accept the same reference. i don't know how the underpinings work, but i would imagine ls -l http://example.com would result in some hilarity. the worst bit is there's no sense of a global name space. http:// feels all vms-ey. - erik
Re: [9fans] novel userspace paradigms introduced by plan 9
On 3 July 2011 12:55, erik quanstrom wrote: > and yet there's a key difference. this is a private joke amongst gnome > processes. i can give "file" references to gnome programs like > http://example.com > to a gnome proc. cat(1) won't accept the same reference. Well yes, it would only make sense in an OS which only uses GIO, rather than standard Unix IO. cls
Re: [9fans] novel userspace paradigms introduced by plan 9
On Sun Jul 3 08:34:26 EDT 2011, c...@lubutu.com wrote: > On 3 July 2011 12:55, erik quanstrom wrote: > > and yet there's a key difference. this is a private joke amongst gnome > > processes. i can give "file" references to gnome programs like > > http://example.com > > to a gnome proc. cat(1) won't accept the same reference. > > Well yes, it would only make sense in an OS which only uses GIO, > rather than standard Unix IO. what i was trying to say is that even in that case, i think gio is a weak model. it goes back to the vms/dos days where the method of access becomes part of the name. that is, i need to know if it's accessed via http or ftp or local to access a file. further, i can't have a path like /usr/quanstro/remote/http://my.other.site/some/path. i have to attach devices at the root. and i'm pretty sure i can't modify what's accessable without recompiling everything that uses the gnome vfs stuff. in short, it's more a clumsy hack than an i/o model. plan 9 has better answers in all three cases, despite being much older. - erik
Re: [9fans] novel userspace paradigms introduced by plan 9
On 3 July 2011 13:51, erik quanstrom wrote: > what i was trying to say is that even in that case, i think gio is a weak > model. it goes back to the vms/dos days where the method of access > becomes part of the name. that is, i need to know if it's accessed via > http or ftp or local to access a file. further, i can't have a path like > /usr/quanstro/remote/http://my.other.site/some/path. i have to attach > devices at the root. Yeah, that's true. Still, that's not *too* different from Plan 9 binding: you have to know the protocol in order to mount a drive, after all. I know very little about GVFS, admittedly, but it would make sense if you *could* run the equivalent to Plan 9's bind(), so you'd say, % bind http://example.net/some/path ~/example That's pretty much the same as running the appropriate fileserver and then binding the result, only GVFS works out what daemon you need. > and i'm pretty sure i can't modify what's accessable > without recompiling everything that uses the gnome vfs stuff. I think you can add more filesystems without recompiling anything, though I don't know for sure. I think it works quite well conceptually, though I'm really not a fan of linking everything into DBus and so on. Still, in terms of bringing Plan 9 to a "wider audience", GNOME might be a way. Of course, it would be rather a lot nicer if Linux could just work out its issues and stop relying on CAP_SYS_ADMIN, but that doesn't seem likely. On the topic of Plan 9, I was thinking an interesting fileserver would be one which, if you access '/uri/http:/example.net', looks up in a table the fileserver required for 'http:', and hands the request over automatically. That way you get the same as GVFS, only without the DBus snafu. I don't know if anyone's already done that. Thanks, cls
Re: [9fans] novel userspace paradigms introduced by plan 9
On Sun, Jul 3, 2011 at 8:06 AM, Connor Lane Smith wrote: > Still, FUSE has extended attributes, so you > could e.g. configure a window manager just by setting attributes on > the 'window manager filesystem' root directory. > something like extended attributes can be accomplished by layering file servers. iru
Re: [9fans] novel userspace paradigms introduced by plan 9
I think what I'd say is the most "novel userspace paradigm" in Plan 9 is its pervasive synthetic filesystems. You have FTP filesystems and so on with FUSE now, but writing something as flexible (technically) as Rio still requires something other than FUSE. But more importantly, since Plan 9 *started* with those synthetic filesystems they're used everywhere, whereas they're pretty uncommon in Linux etc. It would be nice if web browsers used a kind of webfs, and so on. Actually, what this discussion keep pointing out is the elegance of the Plan9 authentication model vs. UNIX's superuser scheme. It's the lack of a superuser that makes the whole namespace paradigm work in the first place. --lyndon
Re: [9fans] novel userspace paradigms introduced by plan 9
On Sunday 03 July 2011 19:57:16 Lyndon Nerenberg wrote:
> Actually, what this discussion keep pointing out is the elegance of the
> Plan9 authentication model vs. UNIX's superuser scheme. It's the lack
> of a superuser that makes the whole namespace paradigm work in the first
> place.
authentication is a system component by the very nature of the problem it
solves, and keeping it off-line in any way (superuser counts as off-line most
of the time) or high-latency (superuser again) is a significant problem.
in UNIX you have a (non-turing complete) program deployed by the superuser
(contents of /etc/{passwd,group}); any protocol (in the broad sense) that
makes use of authentication has to take that into account.
the current www environment also seems to shifts towards authentication based
on OpenID and similar, which i'd liken to factotum in a broad sense.
--
dexen deVries
> (...) I never use more than 800Mb of RAM. I am running Linux,
> a browser and a terminal.
rjbond3rd in http://news.ycombinator.com/item?id=2692529
Re: [9fans] novel userspace paradigms introduced by plan 9
On Sun Jul 3 13:58:49 EDT 2011, lyn...@orthanc.ca wrote: > > I think what I'd say is the most "novel userspace paradigm" in Plan 9 > > is its pervasive synthetic filesystems. You have FTP filesystems and > > so on with FUSE now, but writing something as flexible (technically) > > as Rio still requires something other than FUSE. But more importantly, > > since Plan 9 *started* with those synthetic filesystems they're used > > everywhere, whereas they're pretty uncommon in Linux etc. It would be > > nice if web browsers used a kind of webfs, and so on. > > Actually, what this discussion keep pointing out is the elegance of the > Plan9 authentication model vs. UNIX's superuser scheme. It's the lack > of a superuser that makes the whole namespace paradigm work in the first > place. why do you think that the lack of a super user make per-process namespaces work? - erik
Re: [9fans] novel userspace paradigms introduced by plan 9
why do you think that the lack of a super user make per-process namespaces work? The fact that you own the hardware you are running on means there's no need to provide enhanced priv's (such as root) to protect things like mount(2). And if you do something stupid, the only damage you can do is to yourself. Just look at all the hoops FUSE must jump through to keep people from being able to bodge the entire system.
Re: [9fans] novel userspace paradigms introduced by plan 9
> > why do you think that the lack of a super user make per-process namespaces > > work? > > The fact that you own the hardware you are running on means there's no > need to provide enhanced priv's (such as root) to protect things like > mount(2). that's a property of per-process namespaces, not the lack of a root user. in this sense plan 9 has a limited root—the hostowner that owns the devices on a machine and is trusted wrt the authentication protocol. > And if you do something stupid, the only damage you can do is > to yourself. Just look at all the hoops FUSE must jump through to keep > people from being able to bodge the entire system. for some reason, the linux guys have convinced themselves that per process namespaces can't be done without security problems. i see no reason that pam couldn't do plan 9 style authentication with a process running on behalf of root with its own namespace. they've changed everything else in unix, why hold so tightly to the clearly unhelpful ideas? - erik
Re: [9fans] novel userspace paradigms introduced by plan 9
> they've changed everything else in unix, why hold so tightly to the clearly > unhelpful ideas? because it's a cult. things don't make sense in cults. i encountered the following quote the other day, which finally convinced me. you can't rationalize things with this sort of thinking: "It is out of question that git is becoming the standard tool to manage code. It is incredible how clever Linus was in inventing it." http://www.dev-articles.com/article/Git-and-Linus-431001 oh, and there's the whole 'user-level filesystems are bad, mmmkay?!' debacle: http://blog.gluster.com/2011/06/linus-torvalds-doesnt-understand-user-space-storage/ their ecosystem won't allow any synthetic filesystems past /sys anytime soon, nevermind the auth bit.
Re: [9fans] novel userspace paradigms introduced by plan 9
following this thread. do something interesting. - build a system with only plan9port binaries - use the cap device in linux to authenticate yourself as a user - have init setuid to that user. - figure out how to make linux work with no root user Anything else is likely to be not that interesting. ron
Re: [9fans] novel userspace paradigms introduced by plan 9
Its a pdf about plan9 authentication in linux by one Ashwin Ganti, sorry for the double post. On Sun, Jul 3, 2011 at 3:38 PM, andrew zerger wrote: > More info for people looking from the same vantage point as me.. > > This document is something I am about to read.. (I wasn't sure what a cap > device in linux was, this was the most relevant google result.) > > > http://static.googleusercontent.com/external_content/untrusted_dlcp/research.google.com/en/us/pubs/archive/34433.pdf > > > -- > ⎼⎺⎺├@┼␊├├≤-␍⎼␊▒␍:/⎺└␊/⎼⎺# > > > -- ⎼⎺⎺├@┼␊├├≤-␍⎼␊▒␍:/⎺└␊/⎼⎺#
Re: [9fans] novel userspace paradigms introduced by plan 9
More info for people looking from the same vantage point as me.. This document is something I am about to read.. (I wasn't sure what a cap device in linux was, this was the most relevant google result.) http://static.googleusercontent.com/external_content/untrusted_dlcp/research.google.com/en/us/pubs/archive/34433.pdf -- ⎼⎺⎺├@┼␊├├≤-␍⎼␊▒␍:/⎺└␊/⎼⎺#
Re: [9fans] novel userspace paradigms introduced by plan 9
as a person who has spent the last three years exclusively in user-level filesystems on Linux, I can safely say this -- my biggest problem during that time has been the root user. from dealing with programs which allow only root-level access (xen tools) to dealing with programs who explicitly disallow root (PBS/torque) much more of my time has been spent twiddling with permissions, sudo config scripts and everything else involving root than actually writing the synthetic user-level file system and getting it running. it appears that every cluster of programs used to do anything systems-y in Linux has a special view of uid 0 -- some revere it, others fear it, but no two treat it the same way. only one piece of software said "chgrp my device file to whoever you want to use it and be free". it felt very Plan9-ey.
Re: [9fans] novel userspace paradigms introduced by plan 9
>something like extended attributes can be accomplished by layering file >servers. or simply make a directory
Re: [9fans] SIP
I'm familiar with MS's definition of SIP, but what's this definition? On Fri, Jul 1, 2011 at 10:16 AM, Charles Forsyth wrote: > i couldn't work out which hardware, usable by Asterix, could be > driven by another system instead. > >
Re: [9fans] this is kind of sad
Again, if it were possible to lose more faith in the gov't and the american lemmings, I would have. On Fri, Jun 24, 2011 at 11:24 PM, hiro <23h...@googlemail.com> wrote: > All these days I've been reading off-topic stuff on 9fans were worth > it because of this thread. > >
Re: [9fans] in ed, how to do i do this?
As a guy researching to possibly re-write the book, torrent? :P Yeah, off line reading gives depth online doesn't offer. On Sun, Jun 19, 2011 at 10:11 AM, Thomas wrote: > > As a guy who first read the book around 40 years ago, I can endorse the > recommendation. > > -tom > > On Jun 17, 2011, at 9:05 PM, Rob Pike wrote: > > > As the guy who literally wrote the book on this, please may I > > recommend some offline reading? > > > > -rob > > > > >
Re: [9fans] novel userspace paradigms introduced by plan 9
one might find http://www.glendix.org/ project interesting 2011/7/2 Robert Seaton : > Hello, 9fans! > ...
