alexshendi has quit [Read error: Connection reset by peer]
<beneroth>
rick42, o/
<beneroth>
rick42, well, this was to be expected.
<beneroth>
it was clear that Spectre (and there are like, what, 5+ variants of Spectre) was certainly not the last vulnerability. this whole side-channel CPU attack vector is just such a powerful idea - and it was overlooked so long.
<beneroth>
the whole topic leads to: control over the hard- and software of a computer is necessary (RMS style - or others can/will do the controlling). Too many turtles (technologies, layers) stapled on another are just not manageable. KISS. For secure computing, there is no alternative for developers to have an (at least an rough) understanding of the whole vertical stack.
<beneroth>
The whole shared hosting/computing - running untrustable code from multiple people/companies on the same physical machine - so the whole cloud business model cannot be done in a secure way, impossible. (except dedicated private cloud, which is simply outsourced datacenter operation)
<beneroth>
Too many turtles how malicious code can attack other code running on the same physical machine.
<beneroth>
And as the code that an attacker can run is in principle turing complete - no chance to fully defend against it.
rob_w has joined #picolisp
_whitelogger has joined #picolisp
freemint has joined #picolisp
<beneroth>
hi Regenaxer :)
freemint has quit [Ping timeout: 252 seconds]
<beneroth>
we could also discuss here :P
<beneroth>
I don't think you would gain much by studying web.l
<beneroth>
it's cyborgars work, and I believe you helped him much during initial development
<Regenaxer>
Hi beneroth!
<beneroth>
I happened to have written my first picolisp applications with it, so I know it a bit.
<Regenaxer>
Yes, I know, but why would I gain something?
<Regenaxer>
I feel it does too much
<beneroth>
you would not - if you could gain some good insight from studying, I would recommend you to study it. but you would not :P
<beneroth>
too much? how so?
<Regenaxer>
You see in the mail dscssion
<beneroth>
you mean, because it parses everything before the application code is even informed about the request?
<Regenaxer>
right
<Regenaxer>
builds data structures first
<Regenaxer>
Like readJson which I also never use
<beneroth>
haha, I think the same, but there were are just a weird radical minority. kinda all webservers work this way :)
<Regenaxer>
Indeed
<beneroth>
mostly because when you have multiple layers of different technologies on each other, this is the easiest way to get a clean interface and hide all internals on every layer.
<Regenaxer>
true
<beneroth>
good for lazy programmers, bad for efficiency
<Regenaxer>
In fact what I *do* use is lib/xm.l
<Regenaxer>
XML is a bit more harder to parse directly
<beneroth>
the initial motivation for cyborgar was to have a web-framework that can be used with multiple connection-frontends, be it HTTP browser connections or to handle websockets or whatnot - for that, it's not a bad software design, all in all.
<Regenaxer>
Yes, I think so too
<beneroth>
XML is an abomination. even in theory. but in practice it's usually just so much horror, often practical real XML is not even valid XML in XML theory...
<Regenaxer>
haha, yeah
<Regenaxer>
Still some people *love* it ;)
<beneroth>
there are kinks for everything..
<Regenaxer>
T
<beneroth>
and IT people are usually on the masochistic spectrum :P or they become managers :P
<Regenaxer>
☺
<beneroth>
the "parse everything in webserver thrown at you before even talking to the application layer" is an interesting thing. beside doing maybe-unnecessary work, it is also a nice attack vector for DoS
<Regenaxer>
Because it is costly on the server?
<beneroth>
e.g. most webservers/webframeworks put all the post arguments in a hash map - if you construct a POST request in a way which provokes hash collisions for every POST variable (needs tryout/knowledge of the hash algo in use), you get a pretty elegant DoS
<beneroth>
the application cannot defend against it :)
<beneroth>
nor the parsing layer before, as it cannot know if the form data might be valid or not - only the application layer knows.
<Regenaxer>
I see
<beneroth>
well you could do a watchdog process, killing the request when it's processing clearly takes too long. that is anyway often a standard tool in production use. but yeah, kinda stupid :)
<beneroth>
Regenaxer, did you look into https://mdsattacks.com/ ? I can't wait till cloud business dies and gets replaced with running everything on tiny ARM hardware (one piece of hardware per software that needs to run isolated) using pilOS :D
<beneroth>
(not that this will happen, but amusing to be snarky about it)
<Regenaxer>
Checking ...
<beneroth>
the Spectre guys figured out how to do side-channel attacks on the CPU memory load buffers instead CPU Cache (= Spectre)
<Regenaxer>
Yeah, endless ... :(
<beneroth>
fascinating tech and amusing implications