https://bunny.net/blog/the-stack-overflow-of-death-dns-collapse/

The stack overflow of death. How we lost DNS and what we're doing to prevent this in the future.

The stack overflow of death. How we lost DNS and what we're doing t...

After an almost stellar 2 year uptime, on the 22nd of June 2021, bunny.net experienced a 2+ hour near system-wide outage caused by DNS failure. We are definitely not perfect, but we are doing our best to get as close as possible. We are working hard to learn and improve on our mistakes.

khitiara•9/14/23, 12:10 AM

ive legit had use-after free in c# code like a decade ago when i was a teenager and didnt fully understand finalizers

Aqua•9/14/23, 12:11 AM

every single time i used freetype i was encountering UAF errors

Maxine•9/14/23, 12:11 AM

i didn't mean it literally

Aqua•9/14/23, 12:11 AM

couldn't work it out for years

Aqua•9/14/23, 12:11 AM

so i just ran with it

Aqua•9/14/23, 12:11 AM

UB everywhere

Aqua•9/14/23, 12:11 AM

finally i realized that
it does not take ownership of the memory you pass into it

Aqua•9/14/23, 12:11 AM

that was

Aqua•9/14/23, 12:11 AM

a literal fucking revelation

khitiara•9/14/23, 12:11 AM

oof

Aqua•9/14/23, 12:12 AM

even still it was surprisingly stable using freed memory
which thinking about it
i find hilarious

AAqua finally i realized that it does not take ownership of the memory you pass into i...

Lyris the Kitori•9/14/23, 12:12 AM

its C, there is no real ownership model

khitiara•9/14/23, 12:12 AM

unless youre using ogg

Aqua•9/14/23, 12:12 AM

yeah, although most c libraries i've used tend to take a copy

khitiara•9/14/23, 12:12 AM

in which case its nonsense

Lyris the Kitori•9/14/23, 12:12 AM

"ownership" is just whether it makes a copy or not, which is not ideal for large data structures

Aqua•9/14/23, 12:13 AM

eehhhhhhh
i wrote my rust audio library to take ownership of whatever you pass into it
but not like
the rust way of taking ownership
it copies whatever you pass into it

Aqua•9/14/23, 12:13 AM

although its quite necessary for it to do that

Aqua•9/14/23, 12:14 AM

because it is intentionally designed so that it has full control over the data you pass into it, while still allowing you to retain control of your data

Lyris the Kitori•9/14/23, 12:14 AM

then its not "taking ownership" then, its just making a copy of memory someone else is in control of

Lyris the Kitori•9/14/23, 12:14 AM

you still have to free the memory you pass into it, its still owned by you

Lyris the Kitori•9/14/23, 12:15 AM

its just that theres now 2 copies of the same data, with different owners

Aqua•9/14/23, 12:15 AM

fair point

Lyris the Kitori•9/14/23, 12:16 AM

a better way of phrasing it in documentation is, "this function makes a copy of paramNameparamName, you are able to deallocate paramNameparamName after this method exits"

Aqua•9/14/23, 12:16 AM

tell that to the freetype people
they made it a tiny ass footnote

Aqua•9/14/23, 12:17 AM

my library uses the concept of buffers so its just assumed that it is entirely in control of its own memory

Aqua•9/14/23, 12:17 AM

this design decision still haunts me to this day cause it can make certain things a royal pain

Maxine•9/14/23, 2:10 AM

ReadOnlySpan<int> daysToMonth365 = [0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334, 365];ReadOnlySpan<int> daysToMonth365 = [0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334, 365];
okay, this shit is good

Maxine•9/14/23, 2:11 AM

DOES. NOT. ALLOCATE.

khitiara•9/14/23, 2:38 AM

yup

Maxine•9/14/23, 2:39 AM

The new UnsafeAccessor attribute (introduced in and supported by dotnet/runtime#86932, dotnet/runtime#88626, and dotnet/runtime#88925) lets you define an extern method that bypasses visibility. In this case, I’ve used it to get a ref to the private field, at which point I can just assign null through the ref.

Maxine•9/14/23, 2:39 AM

what the fuck

Maxine•9/14/23, 2:39 AM

you can't just hide this at the bottom

khitiara•9/14/23, 2:39 AM

on little endian anyway, it allocates once globally to convert on other endianness

Kkhitiara on little endian anyway, it allocates once globally to convert on other endianne...

Maxine•9/14/23, 2:39 AM

big endian users can get fucked

MMaxine > The new UnsafeAccessor attribute (introduced in and supported by dotnet/runtim...

khitiara•9/14/23, 2:39 AM

its an implementation detail of the benchmarks lol so they didnt want to draw attention

MMaxine big endian users can get fucked

khitiara•9/14/23, 2:39 AM

true

Kkhitiara its an implementation detail of the benchmarks lol so they didnt want to draw at...

Maxine•9/14/23, 2:40 AM

if i can avoid writing expression emit code for field access, that makes things cleaner

Maxine•9/14/23, 2:41 AM

also this avoids the overhead of a DynamicMethod not scoped to the same module

Maxine•9/14/23, 2:41 AM

which is a stupid thing to care about but FastExpressionCompiler cares about it so

khitiara•9/14/23, 2:42 AM

yeah i think they just want to try and not draw attention to it lmao

khitiara•9/14/23, 2:42 AM

but it is useful in some cases

Maxine•9/14/23, 2:42 AM

khitiara•9/14/23, 2:43 AM

defining an extern method that does private field modification is extremely the sort of thing you dont advertise because its so absurd and unsafe lmao

Maxine•9/14/23, 2:43 AM

IT'S FUN

khitiara•9/14/23, 2:43 AM

oh i agree

khitiara•9/14/23, 2:43 AM

im with you on this one lmao

khitiara•9/14/23, 2:43 AM

its just objectively not the sort of thing thats not getting advertised prominently

Maxine•9/14/23, 2:43 AM

i silly goofy with the obscure runtime errors

khitiara•9/14/23, 2:44 AM

lol

https://bunny.net/blog/the-stack-overflow-of-death-dns-collapse/

Similar Threads

Similar Threads

Similar Threads