<zid>
okay so it loads 801 bytes of bss to 40200, then 7 bytes of data to 402801, overlapping by 1 byte?
* zid
throws the missing 0 in the trash
<mrvn>
801 bytes?
<klange>
heat: there is no proper mmap in toaru, no; I had little doubt my naïve loader would get this right
<mrvn>
klange: if you just read the segments into memory without needing to page align and map then you don't get the bug
<heat>
zid, so it maps .bss
<heat>
no need to even zero anything, this section has no filesz
<heat>
then, when loading .data (which I put after .bss), it has to mmap on 0x402801. since this is not page aligned, the loader has to align the address and the corresponding offset
<heat>
so it maps <offset> on top of 0x402000
<heat>
so it undoes what the bss loading/clearing did
<mrvn>
"If drive right at the screen I will crash into those indians". yes, you will. :)
<klange>
It really shouldn't be mapping an unaligned portion of the file like that, it should realize there's a potential conflict in the aligned area before it and do a partial read; there is absolutely no requirement that every page be mapped directly to the file under any circumstances and these are exactly the circumstances where it's just straight up wrong to do so
<klange>
simpsons comic guy moment - boy i really hope somebody got fired for that blunder!
<mrvn>
klange: if the page only holds one segment then there is no danger
<mrvn>
heat: why does mapping the file over the .bss work? What happens to the page that is already there? Does it leak or does it get freed?
<heat>
obviously gets freed
<heat>
(actually, it never even allocates or gets mapped. virtual memory eh)
<mrvn>
heat: why? The page table entry gets replaced but the page could get lost
<mrvn>
heat: zeroing the bss allocates a page, right?
<heat>
depends
<heat>
erm, linux isn't that naiive
<clever>
you could also just CoW map the zero page
<heat>
elf loading works in terms of mmap
<clever>
and then its already zeroed
<clever>
which makes me wonder, does /dev/zero have a special mmap handler?
<heat>
loading is just calling mmap(p_vaddr, p_filesz, MAP_FIXED | MAP_PRIVATE, ...)
<heat>
yes it does
<mrvn>
clever: you could. Didn't look like linux does that. But if it does then does mapping the file correct the COW mapping correctly?
<heat>
sortie, yo wanna test something out
<sortie>
hey it's me
<mrvn>
clever: mmaping /dev/zero is how you get zero initialized memory. :)
<heat>
and patch that __asm__ __volatile__("mov $") shtick to invoke SYS_exit on sortix
<sortie>
You wanna run that on Sortix?
Celelibi has quit [Ping timeout: 255 seconds]
<sortie>
Kinda doubt that might run
<sortie>
Since it has a custom linker script
<heat>
why would that matter
<klange>
you really should do ("syscall" :: "a"(231)) anyway; I replaced it with ("int $0x7F" :: "a"(0), "b"(0)) for toaru which still doesn't do `syscall` because I'm lazy and "ain't broke, don't fix it"
<mrvn>
didn't sortix have a linker without script support?
<sortie>
Just unclear whether that script happens to be compatible with whatever my vintage of gcc and binutils with Sortix support
<klange>
it's probably fine
<sortie>
heat, but yeah, you can just spin up a Sortix volatile VM and try it out :)
<heat>
if you can pastebin the readelf I can confirm whether or not it is
<heat>
if this is a correct invocation of exit that is
lanodan has quit [Read error: Connection reset by peer]
lanodan has joined #osdev
<klange>
hm, before I fixed the syscall I tried to run the thing in my debugger and it didn't get to the prompt and now I need to investigate why..
<sortie>
heat, I mean that's not how Sortix syscalls work
<heat>
what
elastic_dog has quit [Killed (silver.libera.chat (Nickname regained by services))]
elastic_dog has joined #osdev
<klange>
hopefully it's something dumb like the debugger looping trying to do the traceback and not an actual ptrace issue... `dbg test-sigsegv` still works...
<sortie>
heat, here's a nickle, go read an ABI
heat has quit [Remote host closed the connection]
<klange>
so busy reading the abi, he quit!
<klange>
I am also reminded that I really need to fix my x86-64 kernel setup to not put the kernel in low memory
heat has joined #osdev
<heat>
sortie, how is your ABI weird
nyah has quit [Ping timeout: 246 seconds]
<sortie>
heat, well first of all the exit system call is not always 1 although it does happen to be 1 on Sortix lol
<sortie>
But the syscalls are done with interrupt 0x80
<heat>
i know, i checked the header
<heat>
even in x86_64?
<sortie>
Yeah
<klange>
I use 0 but I'm weird.
<sortie>
#define SYSCALL_BAD_SYSCALL 0
<sortie>
For shaming
<heat>
linux uses 200 and something in x86_64
<klange>
I figured the first thing I'd want to have work is exiting.
<klange>
231 isn't exit, it's exit-group
<heat>
which is _exit
<heat>
or exit(2)
<klange>
that's a glibc decision
<heat>
SYS_exit just exits your thread
xenos1984 has joined #osdev
<heat>
sortie: it still crashes with the correct invocation
<sortie>
Ah
<sortie>
Fix your program
<heat>
fix your elf loader
<heat>
and you call yourself POSIX compliant smh
<sortie>
Go complain to santa
<heat>
zid: hey santa
<heat>
fix sortix
<heat>
i heard they take patches
<klange>
POSIX says nothing about any of this
<klange>
POSIX doesn't know what a _binary_ is much less an elf.
<heat>
sorry
<heat>
and you call yourself a UNIX smh
<klange>
*cough*
<klange>
excuse me, I mean *COFF*
<heat>
SCO rolling in its grave
<sortie>
If I can interject it's actually GNU+NotUnix
<mrvn>
klange: a coff is bad for your elf.
<klange>
I'd just like the interject for a moment.
<heat>
geist, I'd ask you to try this out in fuchsia but you can't statically link can you?
<klange>
What you're referring to as ToaruOS, is in fact, ToaruOS, or as I've taken to calling it, just ToaruOS. ToaruOS IS an operationg system unto itself, and not all just another component of a GNU system.
<heat>
i miss the days when toaru was python/toaruos
<klange>
Good thing I have my own Python these days.
Celelibi has joined #osdev
<klange>
The current file browser has more functionality than the Python one. The only thing I really miss from the Python days is the text layout engine.
<klange>
Which I could absolutely port to Kuroko, but I'm trying to build something better and closer to a real HTML engine.
<mrvn>
klange: port QT?
<heat>
html engine heh
<heat>
are you also shifting to web browser development like serenity
<klange>
I had planned on writing a browser engine before Serenity existed.
<klange>
kling stole everything from me, my shtick, my project goals, my audience, my name
<mrvn>
including script?
<heat>
lol
<klange>
I wrote a Python, I can write a JS!
<mrvn>
klange: and a DOM?
<mrvn>
and css
<heat>
never underestimate an osdever
<heat>
they'll NIH you wrong
<mrvn>
till the manhours exceed the lifetime
<klange>
Like Kuroko, my browser engine was intended to be a more open-to-contributions project than what I've arbitrarily limited the kernel / other parts of the OS to for the last several years.
<klange>
but you are right that I have way too much stuff I want to do and not the time to do it
ElementW has quit [Read error: Software caused connection abort]
<mrvn>
os-dev challenge: write a kernel that outputs it's own source.
<klange>
Screenshot of Misaka, my 64-bit kernel, from 2015. A project I started and then didn't have time to actually work on until six years later.
<Mutabah>
Eh, pretty easy - just embed the source in a tarball
<mrvn>
bonus points if you make a quint that goes through several archs before closing the loop.
<Mutabah>
klange: Ooh, so the really early builds. I thought it was just the modules that were interesting
<klange>
The ToaruOS CDs do come with a /src directory. They everything except Kuroko because I haven't fixed the script that builds them.
spikeheron has quit [Quit: WeeChat 3.7.1]
[itchyjunk] has quit [Ping timeout: 248 seconds]
dutch has joined #osdev
[itchyjunk] has joined #osdev
ss4 has quit [Read error: Connection reset by peer]
dutch has quit [Quit: WeeChat 3.7]
dutch has joined #osdev
dude12312414 has quit [Quit: THE RAM IS TOO DAMN HIGH]
heat has quit [Remote host closed the connection]
heat has joined #osdev
Gooberpatrol66 has joined #osdev
sonny has joined #osdev
[itchyjunk] has quit [Remote host closed the connection]
sonny has left #osdev [#osdev]
smeso has quit [Quit: smeso]
smeso has joined #osdev
gruetzkopf has quit [Read error: Software caused connection abort]
gruetzkopf has joined #osdev
<heat>
fun fact: not zeroing the last bss page will make ld-linux crash
<heat>
as in, the whole page. you can't only zero the .bss bytes.
<heat>
their malloc uses the page's free space after _end, and I suspect they forgot to memset an allocation :V
<clever>
heat: i had similar problems on the rpi, there was a magic number to indicate if the heap was initialized or not, and it was surviving a reset
<clever>
so the heap was partially intact, baring some bit flips due to a few skipped dram refreshes
<clever>
and it either ran out of space (allocations from another boot) or corrupted itself to death
<clever>
big patch, not deeply familiar with that area of the linux kernel
<clever>
not sure i could add anything constructive
bradd has joined #osdev
<heat>
the 80 column limit is deeply annoying
<geist>
yeah that was a big fight at fuchsia, we managed to compromise at 100
<geist>
i think it turns out that are surprisingly large number of people use tiling WMs with tons and tons of little 80 column wide shells everywhere
<clever>
just turn on word-wrap :P
<heat>
the worst part is that linux has 8 width tabs
<heat>
a loop and an if () and you're already in line width trouble
<clever>
thats why i use 2 spaces for indent
<heat>
i don't like 2 spaces, too short. i use 4
<Griwes>
I "compromised" with myself on 110 for my projects because that nicely fits 4 side by side files on my screen in vim. but I'm afraid that now that I've got a bunch of LSP-powered inlays showing, I may need to reevaluate what that limit should be exactly
<geist>
yah i dont like 2 as well, but that was the compromise. some of us gave up 4 space tabs but we at least got 100 columns
<Griwes>
have I mentioned how nice all the lsp-based thingies in neovim are? because dang
<geist>
and yeah 8 space hard tabs + 80 is rough
<geist>
one of the other compromises i think we reached was calling x86-64 x64 (which offends me greatly) but the compromise was aarch64 -> arm64
<heat>
lmao
<geist>
so i'll take x64 over aarch64
<heat>
I think i'd give up aarch64 for x86_64
<Griwes>
was the person insisting on x64 ex-microsoft
* Griwes
is firmly in the amd64 camp
<heat>
aarch64 at least fits well with GNU and GNU target naming
* Griwes
is as of yet unsure if he will go with arm64 for the extra confusion with amd64 or just with aarch64 like a normal person
<geist>
Griwes: possibly. it i think was more forced out of GN and some other tooling, which wa born in chromium world
* Griwes
is aware he is making a lot of loaded statements
<geist>
and chrome/chromium uses x64 because they deal with windows a lot
<heat>
aaaaaaaaaaaaaaaaaaaaaaaaaarch64
<geist>
but... arm64 + amd64 is hard to deal with, so i'll gladly take something distinct
<geist>
and yes, aarch64 is officially the actual name of the architecture, as is x86-64
<heat>
if aarch64 is 64 bits, will aaarch be aaarch128
<geist>
well okay x86-64 is not so officialy
<heat>
or... aaarch96
<geist>
well, to be fair it is 'arm architecture<N>'
<klys>
iapx86-64?
<Griwes>
ie32-e, was it?
<Griwes>
err, ia
<heat>
IA32-e or intel 64
<geist>
oh gosh there are like 8 different ones
<heat>
or amd64
<heat>
x86_64, x86-64, x64
<heat>
am I missing one?
<Griwes>
see, because intel can't get its shit together, going with amd64 is only logical
<geist>
intel 64, ia32e, x86-64, amd64, x64, uh think there's at least another one in there
<Griwes>
lol
<geist>
sure but *obviously* intel is not going to call it amd64
<geist>
so they call it intel64 now
<geist>
despite there was a thing called ia64, but now i thik they j ust dont talka bout tht
<heat>
real ones know what the real intel64 is
* geist
winks
<heat>
they had the guts to delete itanium from tianocore back in 2019
<klys>
then you can end up with a duplicate x as in iapx86-x32
<Griwes>
well, they *finally* killed ia64, what, last year?
<CompanionCube>
does x86_64 count as x86-64
<heat>
they are really trying to rewrite history
<geist>
honestly it annoys me enough that lots of folks call x86 'intel'
<geist>
but that is a battle i fight slowly, always reminding people that AMD exists
<Griwes>
maybe now that ia64 is dead, they could rename ia32-e to ia64
<Griwes>
just to make things more confusing for everyone
<heat>
geist, you should try and talk about x86 with intel people
<heat>
as intelcentric as they come
<heat>
particularly the old heads
<geist>
eh, i think they're generally okay with x86
<geist>
if anything theyre *more* okay with x86 because 8086, etc
<geist>
lest new people forget there were old cpus!
<CompanionCube>
has anyone ever used 'i64'
<klys>
ooh that sounds like an apple
<heat>
rust does
<geist>
right next to u64
<heat>
we could shorten intel to int
<heat>
int64
<geist>
but yeah that's true apple now calls it 'apple' and 'intel' cpus
<geist>
but thats at least sort of at the superficial level, at least they're consistent there
<heat>
tbf they always used intel didn't they
<geist>
ie, naming it based on who uses it
<geist>
er who it came from
<geist>
at the uname level they curiously use arm64 instead of aarch64
<heat>
i'll never understand why we went from x86 -> x64
<heat>
we had 86 bits!
<geist>
windows XP 64. i think that may have been the first use of it
<geist>
someone thought it was clever and then it stuck
<Griwes>
heat: it's an archiTecture, so clearly int64_t would be the correct name
<klys>
iapx is Intel Advanced Performance Architecture
<\Test_User>
86 bits was too awkward to use, so we dropped it to 64 so it's more easily managable
<geist>
but honestly in my heirarchy of things you should call the arch a) it should be easily readable b) short as possible c) include the least number of company names i the name
<heat>
ah so that's why aarch64 is better than arm64
<heat>
less company name
<geist>
arm64 and amd64 fail C, but it's also shorter and easily readable
<geist>
i think i've seen a64 somewhere too
<Griwes>
ah yes, very unambiguous
<klys>
haha that's what we were just complaining about \test_user, 22:36:53< geist> one of the other compromises i think we reached was calling x86-64 x64 (which offends me greatly) but the compromise was aarch64 -> arm64
<heat>
looooooooooooongarch best
<CompanionCube>
geist: i mean, the 'a' in aarch64 is also arm, no?
<heat>
shhhhhhhhh
<geist>
tis true. the troble with ARM is it's also the name of the company and the architecture
<geist>
so you can sort of argue if you want to be pedantic that the 'arm' in arm64 is not the name of the company
<klange>
never should have stopped going by acorn
<geist>
but yeah it's confusing
<\Test_User>
how about we make an architexture called , it will fit a, b, and c very well
<klys>
and if you can abbreviate Architecture as `x', then ax64 !
<geist>
right riscv i'm fairly fond of 'rvNN' but i'm not sure that's so official
<\Test_User>
a simply bc you don't even have to read it
<geist>
but i think it satisfies all the constraints
<geist>
short, unique, no companies, etc
<CompanionCube>
superior names like 'superh' 'riscv' but not 'alpha'
<heat>
'sigma'
<geist>
and superh can be shorted to just sh (as it usually was)
<geist>
sh32, sh64, etc
<klange>
i'm going to make own architecture and call it toaru, and then i'll build a full hardware platform around it and call that toaru, and then my target triplet will be toaru-toaru-toaru
* geist
groans
<kazinsal>
the only issue with the risc-v architecture naming convention is that you have a whole alphabet's worth of extensions packed onto what should be a fairly straightforward architecture name
<CompanionCube>
axp is a weird abbreivation for alpha though
<klys>
that is the most confusing part of decompressing an old kernel, arch/sh/...what is this?
<Griwes>
klange, don't forget the abi!
<heat>
old unices had /bin/<archname> to check the arch
<Griwes>
toaru-toaru-toaru-toaru
<klange>
and people will forever confuse it with tora-tora-tora
<heat>
does that mean sh would be /bin/sh
* geist
torpedo bombs your OS
* Griwes
groans
<klange>
Griwes: that ruins the joke about 1970 war film
<CompanionCube>
heat: which unices?
<heat>
CompanionCube, SVR4 at least
<Griwes>
klange, thus making it an anti-joke, which makes it superior
<kazinsal>
CompanionCube: AXP never stood for anything, it was solely because DEC's legal team wanted something unencumbered to trademark
<geist>
though oddly enough intel was also calling things axp before DEC did
<heat>
it was basically like /bin/i386 -> "#/bin/true"
<geist>
ie, axp432, etc
<CompanionCube>
iirc ~solaris~ does the normal multilib thing
<kazinsal>
geist: close, iAPX
<geist>
oh oh. never mind. close see!
<klys>
yeah that's what I was saying
<klange>
people already mess up toaru, I get lots of torus (donut) or taurus (astrological sign) or tauros (bull pokémon)
<geist>
honestly i can never remember how to spell it myself
<geist>
i always end up typing tauros
<Mutabah>
to-awoo?
<klange>
i should retire ponyos and do TaurOS
<klys>
=)
<geist>
yah good quesiton: how do you pronounce it?
<kazinsal>
tOwOS
<heat>
UWU
<klange>
I could also do a PonyOS that is actually Ponyo-OS...
<geist>
i always thinkf of it as 'two-war-ru'
<geist>
or something like that
<kazinsal>
*nuzzles your kernel*
<klange>
toe ah roo
<geist>
'two-are-ooh'
<geist>
ah
<klange>
I'm also totally happy with a translation: _A Certain OS_ is fine.
<klange>
heat: it's full of typos and broken english in the other parts, too
<heat>
yeah but
<heat>
Control and Status Resgers(CSRs) are support.
<heat>
where the fuck were the supposed editors?
<CompanionCube>
now the spec is at the same level as the impl
<klange>
< klange> i should retire ponyos and do TaurOS ← I say this, but Nintendo would beat the shit out of me, unlike Hasbro
<klange>
hasbro are bros and have let so much unlicensed mlp fan shit slide
<CompanionCube>
heat: hence 'supposed'
<heat>
CompanionCube: I've had to fix major broken english from freaking *americans* submitting patches
<klange>
typing are hard
<CompanionCube>
heat: yes, and?
srjek|home has quit [Ping timeout: 260 seconds]
<klange>
90% of my messages on discord are marked 'edited' because I almost always fuck something up the first time around
<klange>
this is why I don't irc much anymore, cvan't fux my tytpos
<heat>
no one in uefi knows how to type
<klange>
what I really love are the typos in the 'pseudo-code'
<klange>
including entirely incorrect spellings of struct names or macros
<heat>
lmao
<klys>
typething : public metatypething
<klange>
> cvan't fux my tytpos
<klange>
i absolutely swear this was legitimate and not a joke
<klys>
=)
<klange>
but I'm also on multiple grams of painkillers due to a toothache right now, so it' snot necessarily reflective of my general typing ability
<klange>
that detached 's' probably is, tho
<Mutabah>
klange is high as a kite, got it
<CompanionCube>
high as fluttershy
<klys>
oh I met someone with a DEC rx01 8" floppy drive in another chan
<klys>
speaking of DEC
<klys>
geist, when was DEC doing that?
<geist>
oh probably most of the 70s? depends on when rx01 was made
<klys>
'Alpha (original name Alpha AXP) is a 64-bit reduced instruction set computer (RISC) instruction set architecture (ISA)"
<geist>
lets see, started with PDP-8 so probably about 1970
<geist>
oh you mean alpha? that was very late 80s to early 90s
<klys>
AXP
<geist>
they were already starting to work on it under a thing called PRISM i think, canned it, dave cutler left, and then they picked it back up again and made alpha
<geist>
and AFAIK the alpha was th oenly thing called with AXP
<geist>
none of the vaxen were AXP
<geist>
and looks like first alpha was introduced in 1992
<klys>
ah so that is after the intel iapx dealio
<geist>
yeah, by about 10 years
<klys>
"A persistent report attributed to DEC insiders suggests the choice of the AXP tag for the processor was made by DEC's legal department, which was still smarting from the VAX trademark fiasco.[22] After a lengthy search the tag "AXP" was found to be entirely unencumbered. Within the computer industry, a joke got started that the acronym AXP meant "Almost eXactly PRISM".[citation needed]"
<kazinsal>
DEC decided that a new RISC architecture with good tooling behind it would be a better use of money than to try to do RISC backend op translation stuff for VAX like Intel and AMD did for x86
bgs has joined #osdev
<klys>
that's cool, transpeta decided to do pretty much the opposite.
<klys>
that's cool, transmeta decided to do pretty much the opposite.
<klys>
that is, when risc backend op translation became feasible enough
<klys>
so would you blow up if someone started using ax64 instead of aarch64
<klys>
probably about as much as you would with: amd64 -> amdarch64
<heat>
amd64 -> aarch64
<klys>
advanced
<heat>
x86 can into aarch64
<klys>
presumably intel can join forces with altera, so...
<klys>
not soo clear that's what actually happened, looked more like a takeover from my persective.
<heat>
i just had my early reminder of how shitty gmail is at plain text
<heat>
gooooooooooogle psl
<heat>
pls
<heat>
s/early/yearly/
<klys>
yeah that roman font never was easy on the eye
<heat>
it can't handle fucking lines
<klys>
got to preface your emails with <pre> i guess
<klys>
oh dear, nroff may be required as a debugging measure
<heat>
srsly what do google people use for the lkml
<heat>
lore.kernel.org suggests replying with git send-email but i'm not too convinced lol
heat has quit [Ping timeout: 248 seconds]
<geist>
i thin you use git send-email yeah
<geist>
i used it to send some bits to the qemu mailing list. once you set it up its pretty slick
<geist>
and yeah gmail used to deal with plain text a bit better. you can still view the raw incoming email but AFAIK you can't respond in plain text
<klys>
prolly not with gmail smtp tho?
janemba has quit [Ping timeout: 272 seconds]
janemba has joined #osdev
janemba has quit [Ping timeout: 248 seconds]
<kazinsal>
thanks to the magic of BGP looking glasses, I am currently watching my ISP's NOC frantically try various things to fix connectivity to cloudflare
<kazinsal>
(so far, it's not working so well)
janemba has joined #osdev
bgs has quit [Remote host closed the connection]
<geist>
eep
<geist>
i'm digging the proxmox thing. i added a second node to a cluster and then live migrated an instance across it
<geist>
even transferred the handles to the same NFS file apparently transparently
<geist>
i should try migrating something that is using a local disk to see what it does
<Mutabah>
I think it'll copy the disk, then freeze the VM to copy the changes?
<Mutabah>
(Never tried migrating a local disk VM, and have been using some older versions of proxmox)
<geist>
perhaps. since this was pointing at a disk on a NFS share it just handed it over apparently transparently
<Mutabah>
Wait until you try ceph - it's even more magical/transparent... but has been known to explode
<geist>
i hadn't even set up NFS on the second machine, but it apparently either automatically did it when i added it to the cluster, or it copied the setup when i migrated it
<geist>
obviously my network is not set up right. it says pretty specifically you really should have a dedicated network for the inter-VM comms, 5ms max latency, etc
<geist>
presumably for the quorom voting
<geist>
perhaps a dedicated VLAN + higher priority on the switches would solve that
<kazinsal>
looks like the Telus NOC figured it out
<kazinsal>
haha the AS path changed. someone broke somethiiiing
<geist>
so far i'm also digging that seems the backend config files it fiddles with are all plaintext simple to edit files
<geist>
and seems like in general fiddling with them with an editor it'll just pick it up
<geist>
so many of these things dump it all into some sort of centraliized db and then just take over all the config of the system
LostFrog has joined #osdev
PapaFrog has quit [Ping timeout: 276 seconds]
[itchyjunk] has joined #osdev
gog has quit [Ping timeout: 252 seconds]
janemba has quit [Ping timeout: 248 seconds]
janemba has joined #osdev
epony has quit [Ping timeout: 272 seconds]
gog has joined #osdev
gog has quit [Client Quit]
[itchyjunk] has quit [Remote host closed the connection]
potash has joined #osdev
elastic_dog is now known as Guest566
Guest566 has quit [Killed (zirconium.libera.chat (Nickname regained by services))]
elastic_dog has joined #osdev
<mrvn>
I like the glowing pickle when eletrocuted thing for alpha much funnier, was it EVO? Something like that.
<dminuoso>
kazinsal: Since I work for an ISP, I have a particular love-hate relationship with BGP. Some of the extensions make it non-deterministic and potentially never-converging - traffic engineering is extremely indirect and difficult, since most of BGP setups are very adhoc and free of semantics.
<kazinsal>
yeah, I work for a multi-vendor VAR that occasionally has to deal with gigantic funky BGP setups for universities and smaller (yet usually still full-tables) ISPs and the like
<dminuoso>
The first sign of overcomplexity is when you start using BGP MED
<kazinsal>
usually every six years or so our customers in that sort of field call us and go "yeah we need another new set of BGP routers, the tables got too complex"
<kazinsal>
especially for the poor saps redistributing eBGP into iBGP
<zid>
BGP is amazing
<dminuoso>
It's like this seemingly harmless setting that lets you invert control in a nice fashion. At the cost of losing determinism and convergence guarantees.
<zid>
It's like, superdns in terms of issues it constantly causes
<dminuoso>
BGP is, by the way, fully turing complete!
<zid>
internet not working? Blame BGP, then check if it's just a DNS issue.
<dminuoso>
It's also absurdly amusing how in 2022 RPKI has not found wide adoption, and how many AS do not have RR generated router filters...
<dminuoso>
BGP hijacking is still as simple as ever. It's somewhat beyond me how this is not massively abused on a daily rate.
<dminuoso>
Its a poor routing protocol for any scaled network, internal, external or global.
<kazinsal>
simple. no one wants to end up being the main character on NANOG
<zid>
It is, just nobody broadcasts that it happened :P
<zid>
at least once a year a major isp has all their traffic routed to china for 4 hours 'accidentally'
<zid>
and those are just the big ones we hear about
epony has joined #osdev
GeDaMo has joined #osdev
zaquest has quit [Remote host closed the connection]
zaquest has joined #osdev
<j`ey>
heat where are youuuu
heat has joined #osdev
<heat>
j`ey, sup
<heat>
geist: yeah I use git send-email to send patches. but lore.kernel.org actively suggests it for replying to emails, which is a bit more uncomfortable (particularly when quoting people I guess)
genpaku has quit [Read error: Connection reset by peer]
<bslsk05>
github.com: EFI firmware has bugs which may mean that calling EFI runtime service… · openbsd/src@f3c5c95 · GitHub
<heat>
huh
<heat>
"let's patch this because linux said so" but linux didn't really have any concrete sources on there being major breakage on stuff (unless there's something else in the ml but I can't bother to look)
<j`ey>
"Ultimately, we might end up having to revert d3549a938b73 ("efi/arm64:
<j`ey>
be rather unfortunate: that puts us in the same situation as x86,
<j`ey>
where some systems need SetVirtualAddressMap() to be called, and some
<j`ey>
libstub: avoid SetVirtualAddressMap() when possible"), but this would
<j`ey>
crash when you call it (the Snapdragon WoA laptops)"
<gog>
yeh they were using one of those to put down new hot water conduit nearby
<gog>
p neat
<geist>
anyway in this case it's because there are apparently serious power outages upstream from here, so probably it's grid wise power blips
<geist>
this is why i have battery UPSes for all of my servers and network infra. can at least ride out these sort of things long enough to get onto a generator (about 30 seconds)
<geist>
but it also rides out the many little power blips i have here
<gog>
and keeps those blips from damaging sensitive, aging power supplies
<gog>
power conditioning ftw
<zid>
if your psu is damaged by power loss you bought it from the wrong part of china
<gog>
less power loss and more spikes and dips when the grid is balancing itself
<gog>
or rather failing to balance itself
<geist>
oh also fun fact i got a 25U server rack yesterday delivered
<gog>
o:
<zid>
for all your vax?
<geist>
still have to assemble it and dang it's a lot heavier than i thought it'd be
<geist>
nah but i'm going to try to consolodate networking hardware and whatnot into a single rack
<geist>
right now i have an ikea table with a bunch of crap laid out on it, etc
<geist>
half of which is rackmount, but no rack
<gog>
ikea server rack
<gog>
made of MDF and held together with pins and cams
<gog>
i'd buy one
<geist>
heh, no it's a StarTech
<geist>
they seem to make reasonably good but fairly cheap gear
<geist>
actually out of canada it seems
<zid>
startech is an interesting company
<zid>
they mainly seem to sell "obscure chip on a pci-e card"
<geist>
yah that seems to be their modus operandi for their electronic stuff. i dunno if they got into the rack hardware biz later or what but they also sell fairly cheap rackmount gear
<zid>
It was the only company selling a VT6216L
<zid>
which I needed for my desktop
<geist>
i always assumed they were some chinese brand, but nope, out of London, ON
<gog>
was a big fan of their cheap accessories
<gog>
need a sata cable? startech
<geist>
this is also why i'm looking at proxmox a bunch. thinking of consolodting even more hardware into a single machine, or a single cluster
<geist>
like maybe even moving my firewall into a VM instance. probably wont, but seems like a nice idea
<heat>
you should move your firewall to lk
<gog>
yes
<heat>
make everything run lk or fuchsia
<gog>
also moving your firewall into a VM instance is not a good idea
<geist>
yah probably wont, but my main concern is my firewall box is getting long in the tooth
<gog>
use the G5
<geist>
gog: what would be your primary concern re: firewall in VM instance?
<heat>
evolve and surpass the needs for operating systems not written by you
<gog>
my primary concern is the host being compromised in some way
<gog>
but i guess that's the same on hardware
<geist>
ah, fair point
<geist>
i was thinking it'd be more of a 'single point of failure' issue
<gog>
speed would be the other one
<geist>
which is my main concern
<gog>
it's already a single point of failure
<gog>
considering it a little more deeply, i think it could work
<geist>
speed i'm actually thinking might be better honestly
<zid>
huh they don't even seem to have a webpage mentioning the VT6216L anymore
<geist>
iirc i think kazinsal said they were doing the same thing?
<geist>
i thik with modern machines where i think networking speed under VM would start kicking is is probably at the 10gbe/40gbe range. i haven't personally seen any real issues with a machine at 1gbe under virt
<geist>
OTOH yeah i'd probably ideally want to dedicate at least one eth port to the WAN side of things if not the LANE side
<geist>
LANE
<geist>
LAN (keep typing e for some reason)
<geist>
that's why iw as fiddling with trying to punch an eth port directly into the VM via pcie virtualization and coming up sad
<zid>
I think if I was going to do fpga messing around, it'd be to reimplement that chip
<zid>
usb ports are easy to find and solder
<gog>
what's your current firewall running on?
<geist>
lets see. it's a netgate uh.. SG something
<geist>
not for sale anymore, so hard to look up
<geist>
basically it's a Rangeley atom + a bunch of e1000 ports
<gog>
ah yeah i made a similar custom box for my old client
<geist>
trouble is the rangeley atoms are apparently failing left and right now
<geist>
ah netgate SG-4860
<gog>
my old client finally retired their janky old POS in favor of ipads running square so my gateway device became obsolete :(
<geist>
that being said it has more than enogh horsepower to route my current stuff, so there's no reason to replace it per se. i just dont want it to crap out on me some day
<gog>
yeh
<geist>
without at least a solution in the wings
<geist>
though it does annoy me it doesn't have 2.5 or 10gbe and no good way to update it
<geist>
whereas a VM in a big box...
<gog>
can your CPE put your incoming traffic on a tagged vlan?
<gog>
no need for the separate wan port then
<gog>
and possible to set up a failover sytem in a snap
<geist>
yah i was actually hinking of doing precisely that: run the incoming wan into a switch on its own vlan
<geist>
but then i need a bigger switch, which i'll probably do at some point
<geist>
bigger as in > 1gbe
<geist>
also would be kinda neat to be able to turn mirroring on to inspect WAN traffic
<geist>
hypothetically i could also say run the 2.5gbe cable modem into the switch on its own vlan and then run a bonded 1gbe pair of eth out to the router
<geist>
adn then get cheezy >1gbe into the router
<geist>
but i know that the standard eth bonding doesn't really perfectly distribute stuff
<geist>
though may be that wan traffic does hash out enough that it ends up fairly smeared out across both ports. might be worth trying for lulz if nothing else
<heat>
how do you guys know so much about network topologies
<gog>
i learned it for a job
<gog>
had to make a place using an ancient cash register software PCI DSS compliant
<gog>
which meant totally remaking their network
<gog>
but their aforementioned transition to square took away that gig
ss4 has quit [Read error: Connection reset by peer]
<geist>
yeah i learned a lot more in the last 5-10 years when i started getting managed switches w/vlans and whatnot
<gog>
you can do cool-ass shit with managed switches
<mrvn>
geist: why not just plug the cable modem into the router?
<geist>
that's what i currently do
<geist>
and have forever
<gog>
that's the pleb way
<gog>
not very enterprise grade
<geist>
at the very moment there'd be little reason to run the WAN into the switch on its own vlan, since there'd be no particular speed upgrade or anything (my switch is 1gbe, my router is 1gbe, cable modem is 2.5gbe)
<mrvn>
gog: yeah, lets add components to increase the failure rate and attck surfaces
<geist>
but if i had a 2.5 or 10gbe switch then maybe could start to do some interesting things
<gog>
that's right
<mrvn>
do you actually have more than 1GB traffic?
<geist>
also i guess one tiny advantage is the router can sense the link go down if the modem reboots or crashes, but then it generally doesn't. when the modem loses the network most of the time the link stays up, but it simply stops providing packets
<geist>
mrvn: shush
<heat>
mjg, interesting, google still uses SLAB and not SLUB
<mrvn>
geist: unless the modem resets it shouldn't even drop it's ip. just routes into nirvana
<geist>
huh. i'm sure there's some gigantic pile of benchmarks that prove it internally
<geist>
mrvn: when the modem loses the link upstream it usually just stops getting external packets and outgoing ones just go no where
<mrvn>
like any bridge
<geist>
shortly afterwards the DHCP/DHCPv6 leases expire. i think they're set to something very short
<geist>
so it's a little frustrating to be honest. would be kinda nicer if the modem actually dropped the link
<geist>
sicne the router can't really tell if the upstream is down or not
<mrvn>
that sounds like upstream doesn't have enough IPs so they randomly kick people to get some leases back.
<geist>
but of course can do something like a heartbeat ping i suppose
<geist>
no it alwys gets the same thing. this is in case there's some upstream routing failure, or whatnot
<geist>
which sometimes just happens. some switch is getting rebooted somewhere
<geist>
'scheduled network maintenance'
<geist>
fwiw the dhcp v4 and v6 link has always given me the exact same addresses and a full /60 on v6 for the last few years
<geist>
but yeah i do frequently max out the 1gbe link whereas the cable is actually providing 1.2
<geist>
does it make sense to do anything to get the .2? not really, but i am a nerd and it bothers me that i can't get to the last .2
<geist>
OTOH i've heard some folks in other places o the east coast are now getting 2gbe or 5 i think as an option from their fiber provider
<geist>
<drool>
<heat>
is that google fiber
<geist>
no i think some sort of AT&T thing
<geist>
in seattle, centurylink fiber is 940mbps symmetric
<geist>
but since i moved into the woods all i have is an overpriced cable modem service, but at least it gives 1.2 down (but 45 mbps up)
<heat>
woah google fiber is $63/mo in SF
<heat>
this is fucking pricey
<geist>
hahaha. i was just about to say 'yeah how is it that cheap!'
<geist>
i was literally typing that out. as in centurylink in seattle is $65 which is great
<geist>
i'm paying like $125 for cable modem here n the woods
<heat>
1.2 what? gbps?
<geist>
yeah
<\Test_User>
bps
<geist>
and i dont even get a dialup backup modem bank!
<geist>
(not that i have a land line)
<heat>
is all US internet that expensive?
<geist>
it varies *highly* because it's all just local monopolies usually
<geist>
to be fair i am in the fucking woods. running cable all the way to my house is kinda a pain
<geist>
but not that i am sympathetic to comcast at all
<heat>
60 here gets you good internet (500mbps down, 200mbps up lets say), cable and even landline
<geist>
to be fair i'm paying for absolutely the highest tier
<geist>
there's a far more reasonable tier at 60 or 70 but it's only like 100mbit down, and more importantly like 15 up. and my experience is very small up pipes like that can lead to trouble
<clever>
i recently upgraded my internet, to ge the costs down
<heat>
you're also in the woods
<clever>
i cant explain why, but trippling my download speed, halved the cost, lol
<geist>
heat: yeah when i was looking at houses, it was one of the top three bullet items: must have goo dinternet
<geist>
lilke at least 50% of them didn't. finding a house with good cable in the woods is actually kinda hard
<geist>
this house is interestingly wired to the hilt with cable, but turns out the original owner worked at comcast
<heat>
wait, are these metered connections?
<heat>
do you have actual data caps?
<geist>
not at 125
<clever>
i dont believe i have a data cap
<geist>
that's precisely why i pay for the highest tier. no cap
<clever>
1.5gig down, as much as i can move
<geist>
but the lower tiers, yep.
<heat>
lmfao
<geist>
but again, that's comcast being dicks. i told you it was overpriced
<gog>
spectrum was talking about capping their customers on top of making legacy time warner customers buy a worse package. that's when i dumped them
<geist>
centurylink in seattle for example has no caps. full 1gpbs symmetric for $65
<geist>
which is what i had before i moved
<heat>
i mean, i've heard they're all dicks
<heat>
at&t sounds like the worst
<geist>
theyre not all dicks, comcast is just bad
<gog>
the big cable companies are generally terrible
<geist>
also i can shave like $20 off the price if i use my own modem
<gog>
the companies descended from old phone companies are ok
<gog>
not great
<gog>
but better
<geist>
but there are lots of suggestions online that that's a bad idea, because that puts you in a different bucket of support. ie, as soon as you use your own modem and you have a problem they just throw up their hands
GeDaMo has quit [Quit: I'm going to show these people something you don't want them to see. I'm going to show them a world without you.]
<geist>
so i think $20 of that is the modem rental from them basically
<geist>
OTOH work pays for most of it so *shrug*
<heat>
gog, you mean like att
<heat>
:v
<geist>
heat: where are you getting att from this? I've actually heard fairly good stories about att as a internet provider
<geist>
they're just not the local monopoly here
<gog>
it's not actually at&t
<gog>
i mean it is in a sense
orthoplex64 has joined #osdev
<heat>
just loose opinions i've heard
<geist>
yah also keep in mind these megacorps usually have local regions that run their biz differently
<heat>
also john oliver seems to have a personal vendetta against them
<gog>
it originated with one of the companies that was spun off from the original at&t
<geist>
centurylink is huge too, but supposedly their fiber biz is a different unit than their DSL/etc biz
<geist>
and i say comcast i really mean 'xfinity' which may or may not be that closely related to the media conglomerate
<gog>
xfinity is the sucessor to the original comcast, which is now a holding company for its various media things
<geist>
yah, and i think as a result is *not* of the AT&T nobility
<geist>
since comcast started off as a cable provider on its own i think
<heat>
UNIX family tree but instead of UNIX it's AT&T
<gog>
it's not related to any of the bell family
<gog>
afaik
<gog>
it's always been a cable company
<geist>
you get similar stuff with like tmobile usa being of voicestream which was iirc a microwave provider for a while, and thus also not of att
<gog>
i still maintain that at&t was better as a monopoly and congress should have nationalized it and made it part of the postal service
<gog>
but america can't have nice things
<heat>
nah
<heat>
monopolies are poopy
<geist>
side note, on curiousmarc's youtube he visited a local telecom history museum here in seattle that i didn't know about
<geist>
has some neat stuff, nicluding some late 70s early 80s AT&T routing computers that they have resurrected
<bslsk05>
'Tech Monopolies: Last Week Tonight with John Oliver (HBO)' by LastWeekTonight (00:26:50)
<clever>
gog: speaking of bell, its in the reverse dns for my pubip! heh
<gog>
nice
<geist>
yah i think a lot of AT&T split companies are the general monopolies in canada i think
<geist>
rogers is i think one of them?
<geist>
ah actually no
xenos1984 has quit [Read error: Connection reset by peer]
<heat>
hey, if you multiple companies, it's not a monopoly, it's a cartel
<gog>
nope they were a broadcaster
<geist>
i'm thinking rogers wireless which has ties to att
<geist>
probably via buyouts, not necessarily spinoffs
<gog>
yes
<gog>
reading on it
<gog>
at&t bought a stake in 1986
<gog>
they don't own it anymore tho
<geist>
i only remember any of this because at some point when i worked at Danger we were talking to rogers wireless, probably some deal to launch in canada
<heat>
move to europe
<heat>
enjoy vodafone
<geist>
oh as much as i'm complaining, in general internet access has gotten better over the years, at least where i have lived, so i can't complain that much
<zid>
your internet actually sounds really good, compared to the average I hear about US internet
<zid>
lots of monopolies ignoring chronic issues, bad speeds, high prices, limited availability, etc
<heat>
cartels ftw
<mrvn>
geist: they should improve the access in rural areas instead of toally overpowering cities.
<zid>
FREE MARKET
<heat>
praise sky
<heat>
(division of comcast)
<zid>
Sky daddy or Virgin daddy are my choices, weird
xenos1984 has joined #osdev
<mrvn>
geist: think of the internet price as distance / users and the price in the woods makes sense.
awita has quit [Quit: Leaving]
<kazinsal>
aloha, missed a ping, was staging some nexuses (nexii?). please hold
<kazinsal>
yeah my firewall is a VM on my ESXi/storage box
gildasio has joined #osdev
<zid>
nessen
<mrvn>
essen? hunger
bgs has quit [Remote host closed the connection]
<kazinsal>
one of these days I will figure out how to reheat a burrito in a manner that does not turn its contents from delicious meat and cheese into molten pain
<kazinsal>
but for now this burrito will sit here cooling off and I will resist the temptation to burn my mouth on it
<mats1>
use a toaster oven / air fryer
<mrvn>
so it's hot on the outside and cold inside?
<mats1>
yes
rorx has quit [Ping timeout: 246 seconds]
smach has joined #osdev
Dyskos has joined #osdev
ZipCPU_ has joined #osdev
ZipCPU has quit [Ping timeout: 276 seconds]
ZipCPU_ is now known as ZipCPU
<gog>
mmm molten pain
sortie has quit [Remote host closed the connection]
sortie has joined #osdev
catern has quit [Remote host closed the connection]
<mrvn>
Nothing better than the burning hot tomatoe slices on a pizza that stick to the roof of your mouth.