OBOS (not vibecoded) | Operating System Development | Page 38

flint idol Feb 15, 2025, 11:38 PM

#

fixed it

#

was passing some bogus stuff to reallocate

#

I was able to connect to the gdbstub

#

but memory reading isn't quite working

#

yeah so

#

uh

#

the received packet from GDB that is querying the memory

#

is corrupted

flint idol Feb 16, 2025, 1:08 AM

#

bro how the fuck

#

is the gdbstub corrupting shit

#

it didn't do this before

flint idol Feb 16, 2025, 2:05 PM

#

this gdbstub seems to be dogshit

#

it can't even receive packets properly anymore

#

without completely trashing the kernel's state, of course

short mortar Feb 16, 2025, 2:08 PM

#

Why not just use qemu gdbstub?

flint idol Feb 16, 2025, 2:08 PM

#

flint idol Feb 16, 2025, 2:36 PM

#

okay I fixed that bug (I removed a printf and it magically started working again)

#

and I also fixed a bug with memory reading/writing (I think)

#

now breakpoints aren't working

#

    uintptr_t phys = 0;
    MmS_QueryPageInfo(Mm_KernelContext.pt, bp->addr, nullptr, &phys);
    if (!phys)
    {
        response = "E.Page fault";
        goto respond;
    }
    
    bp->at = *(uint8_t*)Arch_MapToHHDM(phys);
    *(uint8_t*)Arch_MapToHHDM(phys) = X86_INT3;```

#

something tells me something like this might be a problem

#

because like instruction cache or something

#

but this is x86-64 where the cpu does all the caching work for me, right?

#

ah wait I'm dumb

#

    uintptr_t phys = 0;
    MmS_QueryPageInfo(Mm_KernelContext.pt, bp->addr, nullptr, &phys);
    if (!phys)
    {
        response = "E.Page fault";
        goto respond;
    }

+    phys += (bp->addr & 0xfff);

    bp->at = *(uint8_t*)Arch_MapToHHDM(phys);
    *(uint8_t*)Arch_MapToHHDM(phys) = X86_INT3;```

#

iirc Arch_MapToHHDM doesn't align the passed physical address, so I should be fine there...

#

okay it works

#

and it's still just as slow as it was before, but alas

#

anyway this is only over serial

#

I am now going to implement the UDP backend for it

flint idol Feb 16, 2025, 3:07 PM

#

done

#

time to test it

#

bruh

#

no route to host

#

(gdb) target remote 192.168.100.2:1534
`/home/oberrow/Code/obos/out/oboskrnl' has changed; re-reading symbols.
could not connect: No route to host.```

#

oh

#

yeah it needs to be udp

#

oopsie

#

ok the remote receives the UDP packet, notifies the UDP backend that it got a packet, then nothing

#

rn_image_picker_lib_temp_911d5f2d-87dd-4fda-94b2-501815b51dd6.jpg

#

wireshark shows no response from the remote

#

so it's probably hanging

#

and if the port were unbound it would've printed destination unreachable on the debug logs...

#

ok so this is going terribly

#

so for some reason

#

it only ever reads $

#

it does receive a full UDP packet

#

oopsies

#

    const size_t nRead = OBOS_MIN(blkCount, hnd->curr_rx->sz);
-   memcpy(buf, hnd->curr_rx->buff, nRead);
+   memcpy(buf, hnd->curr_rx->buff+hnd->rx_off, nRead);
    printf("Read: %.*s\n", nRead, buf);
    hnd->rx_off += blkCount;```

#

should be the solution

#

holy shit

#

I connected

#

to a gdbstub

#

over the internet

#

on OBOS

#

#

and also, it's faster than GDB stub over serial

#

for some reason

white mulch Feb 16, 2025, 3:51 PM

#

flint idol and also, it's faster than GDB stub over serial

on real hw?

flint idol Feb 16, 2025, 3:51 PM

#

yes

#

no

white mulch Feb 16, 2025, 3:51 PM

#

or like faster than qemu serial on real hw

flint idol Feb 16, 2025, 3:51 PM

#

faster than qemu serial on real hw with udp

white mulch Feb 16, 2025, 3:52 PM

#

yeah that makes sense considering literally every serial write is a vmexit

flint idol Feb 16, 2025, 3:52 PM

#

💀

#

white mulch Feb 16, 2025, 3:54 PM

#

nice

flint idol Feb 16, 2025, 3:54 PM

#

I also want to see if I could use this interface for user input/output for bash

#

on real hw

white mulch Feb 16, 2025, 3:55 PM

#

also a gdbstub would only follow a specific thread could be useful

flint idol Feb 16, 2025, 3:55 PM

#

flint idol I also want to see if I could use this interface for user input/output for bash

but that's gonna be an experiment for another time

#

okay so the gdb monitor command crashes the kernel

white mulch Feb 16, 2025, 4:00 PM

#

why

flint idol Feb 16, 2025, 4:00 PM

#

page fault

#

currently debugging

#

made a typo while memcpying something

#

I think

#

nvm

#

    size_t commandLen = strchr(arguments, ' ');
    if (commandLen != argumentsLen)
        commandLen -= 1;
    char* command = 
        (char*)memcpy(Kdbg_Calloc(commandLen+1, sizeof(char)), arguments, commandLen);```

#

commandLen = -1

#

arguments in this case should be "ping\0"

real pecan Feb 16, 2025, 4:04 PM

#

flint idol ```c size_t commandLen = strchr(arguments, ' '); if (commandLen != argum...

im starting to think oberrow is the reason they made memory safe languages KEKW

flint idol Feb 16, 2025, 4:05 PM

#

flint idol arguments in this case *should* be `"ping\0"`

argumentsLen = 4 in this case

#

I think my strchr has regressed

#

that, or arguments isn't null-terminated

#

    char* arguments = Kdbg_Calloc(argumentsLen + 1, sizeof(char));

nvm

#

it's really stupid how the people at GNU thought "ah yes let's make the gdb server protocol send hex for the monitor command"

#

what if I

#

use my gdbstub

#

to debug my gdbstub

#

inb4 triple fault trl

#

oh yeah btw the bug in particular is a buffer overflow

#

hmm why is commandLen = -1

#

thonk

#

yeah so basically

#

it's not reading the arguments properly

#

I fixed the bug

#

I forgot to initialize argumentsOffset in the packet dispatcher when I was fixing it

#

I was able to connect to the gdbstub on a different machine too

#

I just needed to forward a port

#

I should forward it on my router, what could possibly go wrong trl

flint idol Feb 16, 2025, 4:55 PM

#

anyway so having a gdbstub on real hardware is pretty damn cool

#

it'd be interesting to see if this works after wake-from-suspend

#

although it wouldn't work until wake-from-suspend has completely woken the computer

#

cuz like

#

the r8169 driver would be dead

#

and drivers are waken up last

#

// in OBOS_Suspend
    for (driver_node* node = Drv_LoadedDrivers.head; node; )
    {
        if (node->data->header.ftable.on_wake)
            node->data->header.ftable.on_wake();

        node = node->next;
    }
    Core_MutexRelease(&suspend_lock);
    OBOS_Log("oboskrnl: Woke up from suspend.\n");
    return OBOS_STATUS_SUCCESS;
}

short mortar Feb 16, 2025, 6:14 PM

#

flint idol it'd be interesting to see if this works after wake-from-suspend

Be careful since UDP does not guarantee packet delivery

flint idol Feb 16, 2025, 6:14 PM

#

ye

#

I did try

#

it hangs

#

and stops responding to ARP requests

#

I love it when the kernel randomly stops responding to ARP requests

#

and basically any requests

short mortar Feb 16, 2025, 7:18 PM

#

Just port Mesa, be the uDRM test kernel, and let it restore the framebuffer after sleep

timid elk Feb 16, 2025, 7:18 PM

#

OBOS: our entire thing is that we support sleep

#

suspend*

torpid wigeon Feb 16, 2025, 7:21 PM

#

i mean linux is already the uDRM test kernel

flint idol Feb 16, 2025, 7:50 PM

#

timid elk OBOS: our entire thing is that we support sleep

bruh

flint idol Feb 16, 2025, 8:48 PM

#

ok so for some reason

#

when a memory read in the gdbstub

#

passes a page boundary

#

it hangs

white mulch Feb 16, 2025, 10:48 PM

#

btw there is a nice thread safety analysis in clang (that korona originally mentioned about), https://clang.llvm.org/docs/ThreadSafetyAnalysis.html, basically it can warn you when you eg. acquire the same lock multiple times, when you call a function requiring a specific lock to be held or when you forget to release it (and there is also an attribute to make it warn if you access a specific field/variable without holding a lock though its kinda broken and doesn't report if you lock the lock of an another object and access the field in an other one, hopefully that gets fixed)

#

idk how much do you compile obos with clang but I though it's worth mentioning in case you are interested

flint idol Feb 16, 2025, 10:49 PM

#

I can compile the kernel+drivers with clang

#

maybe I'll do that one day

flint idol Feb 16, 2025, 11:33 PM

#

flint idol passes a page boundary

#

gdbstub was last seen here

sweet compass Feb 16, 2025, 11:36 PM

#

@haughty abyss since OBOS seems to have at least semi-functional networking, a kernel pwn chall against OBOS would maybe be fun coolmeme

flint idol Feb 16, 2025, 11:36 PM

#

should be easy

#

on my home network, it dies after ~5 minutes

sweet compass Feb 16, 2025, 11:37 PM

#

well it's a reversing challenge too (reading OBOS source code) KEKW

flint idol Feb 16, 2025, 11:37 PM

#

bruh

#

I'll just git clang-format on the next commit trl

sweet compass Feb 16, 2025, 11:38 PM

#

no but like, function names and such

#

I tried reading the OBOS suspend code, and I genuinely couldn't tell at what line the actual suspend happens

#

(me skill issue)

flint idol Feb 16, 2025, 11:39 PM

#

line 92

#

of src/oboskrnl/power/suspend.c

#

if you started reading at suspend.h

#

it would lead you to OBOS_Suspend

#

in suspend.c

#

which starts a mysterious worker thread that starts in static void suspend_impl() (wonder what that could be doing)

#

static void suspend_impl()
{
    if (OBOS_WokeFromSuspend)
    {
        // do shit
    }
    // NOTE: It is up to the arch to unsuspend the scheduler.
    Core_SuspendScheduler(true);
    Core_WaitForSchedulerSuspend();
    // printf("good night\n");
    OBOS_ECSave();
    OBOSS_SuspendSavePlatformState();
    uacpi_prepare_for_sleep_state(UACPI_SLEEP_STATE_S3);
    UACPI_ARCH_DISABLE_INTERRUPTS();
    // good night computer.
    uacpi_enter_sleep_state(UACPI_SLEEP_STATE_S3);
    while(1)
        asm volatile("");
}```

flint idol Feb 16, 2025, 11:45 PM

#

flint idol gdbstub was last seen here

huh what

#

it works until it tries to transmit the packet

#

the packet is 5022 bytes long...

#

I wonder if that has anything to do with it

#

    /*
     * From RFC791 (IPv4 specification)
     * It is recommended that hosts only send datagrams
     * larger than 576 octets if they have assurance that the destination
     * is prepared to accept the larger datagrams.
    */
    uint16_t packet_length;```

#

bruh

#

OBOS_NO_UBSAN obos_status Net_FormatIPv4Packet(ip_header** phdr, void* data, uint16_t sz, uint8_t precedence, const ip_addr* restrict source, const ip_addr* restrict destination, uint8_t lifetime_seconds, uint8_t protocol, uint8_t service_type, bool override_preferred_size)
{
    if (!phdr || !data || !sz || !source || !destination)
        return OBOS_STATUS_INVALID_ARGUMENT;
    if (!override_preferred_size && sz > IPv4_PREFERRED_PACKET_LENGTH)
        return OBOS_STATUS_INVALID_ARGUMENT;```

#

and IPv4_PREFERRED_PACKET_LENGTH = 576,

#

last time I checked, 5022 > 576

#

that still doesn't explain why it doesn't respond to arp requests

flint idol Feb 17, 2025, 12:19 AM

#

yeah so debugging this is boring as shit

#

so I'm just going to push the code

#

I think it's because for some reason DPCs don't really get called...

#

which is bizzare

#

theoretically on each timer IRQ it should be lowering irql to < IRQL_DISPATCH

#

triggering DPCs to be dispatched

#

but idk

#

yeah it's contantly being lowered to < IRQL_DISPATCH

#

on timer IRQs

main girder Feb 17, 2025, 12:38 AM

#

there was some problem i ran into with using the actual lower irql function while leaving an interrupt handler

#

but i fixed it so long ago i dont remember what it was

flint idol Feb 17, 2025, 12:38 AM

#

do you maybe have an idea of when so I could find a commit

#

in mintia(2?)

main girder Feb 17, 2025, 12:39 AM

#

maybe early 2021 idk lol

#

send me links to the relevant source files in obos

flint idol Feb 17, 2025, 12:41 AM

#

irq/dpc.c
irq/irql.c
has everything relavant to dispatching DPCs n'stuff
irq/irq.c
has the code for my IRQ interface (main function in there is Core_IRQDispatcher)

#

my current hypothesis is the r8169 driver receives IRQs for incoming data, but the DPC it registers for actually processing the data never gets called

#

perhaps the kernel is spinning at >= IRQL_DISPATCH

#

somewhere

#

#

the 2nd packet is after a breakpoint is hit

#

but no further packets are received in the kernel after

#

at all

flint idol Feb 17, 2025, 12:44 AM

#

flint idol the 2nd packet is after a breakpoint is hit

sent by the function Kdbg_NotifyGDB

#

which then calls Kdbg_GeneralDebugExceptionHandler

#

maybe a breakpoint being triggered at >IRQL_DISPATCH is the culprit, because then the entire GDB stub would be running at > IRQL_DISPATCH

#

I just noticed that acpi GPEs aren't being called

#

meaning that it's hanging at > IRQL_GPE (8)

#

but the only IRQL that is used above that is IRQL_MASKED (0xf)

#

which is only used in one place

#

    irql oldIrql = Core_SpinlockAcquireExplicit(&target->dpc_queue_lock, IRQL_MASKED, false);
    LIST_PREPEND(dpc_queue, &target->dpcs, dpc);
    Core_SpinlockRelease(&target->dpc_queue_lock, oldIrql);```

#

in dpc.c

#

but that wouldn't be able to hang, to my knowledge

#

unless there is a deadlock

#

but the kernel would've said something about it by now, if that were the case

    size_t spin = 0;
    while (atomic_flag_test_and_set_explicit(&lock->val, memory_order_acq_rel))
    {
        OBOSS_SpinlockHint();
        if (++spin >= 10000000)
            OBOS_Panic(OBOS_PANIC_FATAL_ERROR, "lock hanging!\n");
    }```

flint idol Feb 17, 2025, 12:54 AM

#

flint idol which is only used in one place

wait it's also used in the ec driver...

#

it'll probably be easier to debug this in qemu

short mortar Feb 17, 2025, 12:58 AM

#

sweet compass <@696410219663851552> since OBOS seems to have at least semi-functional networki...

I'm thinking about writing network subsystem in Rust for that reason

flint idol Feb 17, 2025, 12:59 AM

#

flint idol it'll probably be easier to debug this in qemu

oh wait the bug only reproduces with UDP nooo

short mortar Feb 17, 2025, 1:02 AM

#

Why not TCP

flint idol Feb 17, 2025, 1:02 AM

#

because I don't have TCP?

#

anyway I feel like it'd be cool

#

to make the --enable-kdbg command line option better

#

so that it can specify stuff like serial:com1 or smth

#

udp:192.168.100.2:1534

#

although maybe that's out of scope for the kernel initialization function

short mortar Feb 17, 2025, 1:08 AM

#

Idk how sensitive GDB is, but UDP is like difficult to deal with

flint idol Feb 17, 2025, 1:08 AM

#

the bug itself is not because of UDP, but is a bug elsewhere

#

because at one point the kernel decides to stop receiving packets, for whatever reason that may be

#

anyway just pushed this code

#

since the most complaints about obos code readability seem to say that the naming convention is ass

#

I will be changing that after merging net-stack with master (which I will do after implementing TCP)

sweet compass Feb 17, 2025, 1:19 AM

#

flint idol since the most complaints about obos code readability seem to say that the namin...

as long as it is consistently ass it is fine

flint idol Feb 17, 2025, 1:19 AM

#

flint idol I will be changing that after merging net-stack with master (which I will do aft...

or maybe not, depends on how lazy I am

flint idol Feb 17, 2025, 1:19 AM

#

sweet compass as long as it is _consistently_ ass it is fine

it is consistent

#

static symbols are snake_case

#

structs are snake_case

#

any global symbols adhere to Namespace_NameOfFunction

#

usually it goes Namespace_NounVerb (noun is the object the function operates on, verb is the operation that should be done)

#

but it sometimes strays from that

#

and I do stuff like Core_ExitCurrentThread

#

even though other thread apis use CoreH_Thread*

#

note that namespaces also have suffixes, like S, and H

#

S standing for architecture Specific

#

H standing for Helper

flint idol Feb 17, 2025, 5:58 PM

#

marker

white mulch Feb 17, 2025, 6:01 PM

#

one note for that is that you don't actually need to disable interrupts at all, basically its enough to do smth like ```cpp
cpu_local->software_irql = new_irql
if (new_irql < cpu_local->hardware_irql) {
cpu_local->hardware_irql = new_irql
cr8 = new_irql
}

if it is interrupted before writing the new software irql then the if gets entered
if it is interrupted after writing the software irql then the interrupt will set cr8 to the new irql (which is fine if it happens either before that if or when inside it, though if its inside it then its done twice for no reason but still better than unconditionally doing it)

flint idol Feb 17, 2025, 6:36 PM

#

idk what I did and when I did it

#

but obos only gets 400k on master

#

on my PC

#

global Core_RaiseIrqlNoThread: function weak
Core_RaiseIrqlNoThread:
    mov rax, [Arch_cpu_local_currentIrql_offset]
    xchg [gs:rax], rdi
    ret```
and it also seems as if my raise irql function has broken lol

#

#

oh wait I'm stupid

#

hmm I have an idea to optimize the allocator

flint idol Feb 17, 2025, 7:05 PM

#

that worked

#

changing uacpi to use the non paged allocator also speeds stuff significantly

#

[uACPI][INFO]: successfully loaded 1 AML blob, 1705 ops in 1ms (avg 1009939/s)

was able to get this score on my fastest machine...

white mulch Feb 17, 2025, 8:13 PM

#

do you support nested interrupts in obos?

#

and if yes how do you manage the stack use of them, wouldn't you need huge per thread kernel stacks for it?

flint idol Feb 17, 2025, 8:18 PM

#

a) yes
b) I don't manage that

#

although nested interrupts only happen with IRQs that have a greater IRQL than the IRQ before it

#

so most nested IRQs run at > DISPATCH level

#

so they won't end up doing much anyway

#

and currently all drivers almost immediately register a DPC to handle their IRQs

white mulch Feb 17, 2025, 8:21 PM

#

ah, ig that makes sense so then they don't really need that much space

#

probably the interrupt frame itself is the biggest thing at that point

flint idol Feb 17, 2025, 8:22 PM

#

ye

#

https://github.com/OBOS-dev/obos/blob/master/src%2Fdrivers%2Fgeneric%2Fr8169%2Fdevice.c#L68-L82

GitHub

obos/src/drivers/generic/r8169/device.c at master · OBOS-dev/obos

I was bored... Contribute to OBOS-dev/obos development by creating an account on GitHub.

#

an example

#

it clears the IRQ status, starts a DPC, then returns

#

the uart driver does the exact same

white mulch Feb 17, 2025, 8:26 PM

#

nice

flint idol Feb 17, 2025, 8:27 PM

#

anyway, I plan to use attribute malloc to help track down memory leaks

#

and use after frees

#

statically

#

ok time to merge a PR that definitely does not change the allocator interface drastically just to optimize the uacpi score

flint idol Feb 19, 2025, 4:07 AM

#

oh fuck

#

I just fucked my github repo

#

#

thank god for git reflog

uncut narwhal Feb 19, 2025, 4:11 AM

#

the heck did you do? lmao

#

thank god for reflog indeed

flint idol Feb 19, 2025, 4:11 AM

#

git filter-branch

uncut narwhal Feb 19, 2025, 4:12 AM

#

oh classic

#

ive done that command before

torpid wigeon Feb 19, 2025, 10:12 AM

#

flint idol

force push

frank flare Feb 19, 2025, 3:18 PM

#

flint idol `git filter-branch`

what does that do

#

tbh I only use the basic subset of git

flint idol Feb 19, 2025, 3:23 PM

#

frank flare what does that do

I forgor

frank flare Feb 19, 2025, 3:24 PM

#

flint idol

yeah seems like

#

meme

uncut narwhal Feb 19, 2025, 4:52 PM

#

frank flare what does that do

its probably the most powerful history rewriting tool git has lol

#

i used it once to retroactively change my name in commits on an entire repo in one command

frank flare Feb 19, 2025, 4:57 PM

#

might learn about it then

uncut narwhal Feb 19, 2025, 4:58 PM

#

dont

#

its basically never useful

#

and that level of rewriting history should be avoided where possible anyway

#

it can be nice to now ig but id at least wait until you end up with a problem that needs it lol

frank flare Feb 19, 2025, 5:29 PM

#

usually I make some changes which I then separate in 3-4 commits (because I don't like commiting unrelated changes at once, although sometimes it's fine to do a big commit)
then I push all commits
then I realize that I've introduced a bug in my first commit of this series, now I do git reset --soft <hash>, fix the bug, rewrite all the commit messages and force push

#

so yeah I should learn to properly rewrite history

#

cause that ^ works fine when you need to rewrite your last pushed commit

#

otherwise it's a pain and it's completely my fault for being lazy and not looking it up

#

hijacking obos thread cause yeah

uncut narwhal Feb 19, 2025, 5:40 PM

#

yeah filter-branch isnt the tool for that

#

idk what is but it aint that one

#

filter-branch is for bulk rewrites

frank flare Feb 19, 2025, 5:45 PM

#

oh okay

flint idol Feb 20, 2025, 12:45 AM

#

I want to implement ICMP in obos

#

specifically RFC792

#

it shouldn't be too difficult

#

I will be doing some basic stuff for reception

#

mainly just ICMP echo replies

#

and notifying a user if a destination is unreachable

#

like if it tries to access a remote UDP port but the host returns destination unreachable or smth

#

I want to receive that (currently it's discarded), and abort any reads/unbind the port maybe

flint idol Feb 20, 2025, 2:05 AM

#

so currently ICMP is destroying the network stack

#

causes a hang

#

while sending a status message

#

in the UDP part of the stack

flint idol Feb 20, 2025, 2:07 AM

#

flint idol while sending a status message

it's last seen sending an ARP request

#

ah wait that makes sense

flint idol Feb 20, 2025, 2:24 AM

#

ok so now I send an ICMP port unreachable reply

#

when the port is unreachable

#

but the only problem

#

is that linux doesn't give a shit

#

what I mean is that when I connect to other things and get a port unreachable reply back, nc exits

#

it does not do the same with obos

#

oh wait

#

oops

#

I was setting the wrong destination mac address meme

#

okay now I am getting checksum problems

#

and I think wireshark is having a stroke

#

it's first trying to respond to this with port unreachable

#

hollow geyser Feb 20, 2025, 2:44 AM

#

Pretty cool dude

flint idol Feb 20, 2025, 2:45 AM

#

ah shi I forgot you were in this server

#

I think linux is discriminating against obos

#

update

#

when the ICMP packet's length is even it causes nc to stop

#

which means linux only wants to recognize packets with even lengths?

flint idol Feb 23, 2025, 12:38 AM

#

I want to make a change to obos-strap so that it rebuilds a package if it sees that a file dependency (and by that I mean a recipe or patch) gets modified

#

me when obos regression

#

#

wtf is happening

#

me when:

diff --git a/src/oboskrnl/mm/alloc.c b/src/oboskrnl/mm/alloc.c
index 383ea9d..1079f1b 100644
--- a/src/oboskrnl/mm/alloc.c
+++ b/src/oboskrnl/mm/alloc.c
@@ -592,6 +592,7 @@ obos_status Mm_VirtualMemoryFree(context* ctx, void* base_, size_t size)
             page what = {.phys=info.phys};
             page* pg = RB_FIND(phys_page_tree, &Mm_PhysicalPages, &what);
             // printf("derefing physical page %p representing %p\n", info.phys, addr);
+            pg->pagedCount--;
             MmH_DerefPage(pg);
         }
         MmS_SetPageMapping(ctx->pt, &pg, 0, true);

#

lmao

#

what the fuck

#

I was sure bash worked on this commit

#

fuck compiler optimizations and fuck gcc

frigid tide Feb 23, 2025, 12:58 AM

#

Heavy on the fuck GCC

frigid tide Feb 23, 2025, 12:58 AM

#

flint idol me when: ```diff diff --git a/src/oboskrnl/mm/alloc.c b/src/oboskrnl/mm/alloc.c ...

There’s alot going on here 😭

flint idol Feb 23, 2025, 12:58 AM

#

+            pg->pagedCount--;
             MmH_DerefPage(pg);```

#

although I accidentally had that git diff command the wrong way

vale nymph Feb 23, 2025, 1:00 AM

#

flint idol me when: ```diff diff --git a/src/oboskrnl/mm/alloc.c b/src/oboskrnl/mm/alloc.c ...

what would happen if multiple cores unref a page at once

flint idol Feb 23, 2025, 1:00 AM

#

destruction.

vale nymph Feb 23, 2025, 1:00 AM

#

and the pg->pagedCount gets freaky

flint idol Feb 23, 2025, 1:01 AM

#

I'm just hoping _Atomic is normal

#

and does atomic shit

#

me: wants to take a small break from stuff and work on obos
meanwhile obos random regression #6547456435

#

there is quite fucking literally no reason

#

for this

#

#

I decided to change the crash to not be an assert anymore

#

it seems to be a bug with VfsH_PageCacheGetEntry

#

ok now fucking kvm

#

crashed

#

wtf is this

#

I can't fucking deal with this fucking shit right now

flint idol Feb 23, 2025, 2:22 PM

#

ok I figured out why

#

I think it's stack corruption meme

#

#

or maybe a stack overflow meme

#

there I go, fixed it

main girder Feb 23, 2025, 2:25 PM

#

obos reminds me of like early 2021 mintia

weak kestrel Feb 23, 2025, 2:29 PM

#

flint idol

konsole?

flint idol Feb 23, 2025, 2:29 PM

#

yuews

#

yes

#

honestly I think I now want to just make the pagecache rework thing I wanted to do a few months ago

#

but was busy with getting bash to work

flint idol Feb 23, 2025, 3:10 PM

#

yeah the more I think the more standby/dirty page lists make sense

#

and I'm here thinking, "wow my current pagecache system is absolute dogshit"

#

anyway

#

the more I read my old code the more I want to die

real pecan Feb 23, 2025, 3:39 PM

#

Read it less

flint idol Feb 23, 2025, 3:40 PM

#

galaxybrain

#

why didn't I think that

#

so I finally got the kernel to build again

#

inb4 triple fault after mm initialization

#

surprisingly not

#

it hangs

#

sometime after loading the kernel symbol table

#

deadlocks on the kernel context lock

#

while makes perfect semse

#

*sense

#

since the dirty page writer was woken while the kernel context was locked

#

which is a bug

#

fixed that hang

#

now it crashes a bit later

#

    pg = nullptr;
    memcpy(buffer, MmS_MapVirtFromPhys(in->phys), pg->sz);```

#

I just love this code

flint idol Feb 23, 2025, 4:30 PM

#

I am getting closer to getting it to boot again

#

of course it hangs again

#

now somewhere after the COM driver finishes init, it seems

#

ah wait nvm

#

it was an infinite loop I had for debugging purposes

#

(it now crashes)

#

[ LIBC  ] Unexpected PHDR type 0x464c457f in DSO libdl.so```

#

bruh

weary hound Feb 23, 2025, 4:35 PM

#

flint idol ``` [ LIBC ] Unexpected PHDR type 0x464c457f in DSO libdl.so```

nice read from offset 0 in the file

flint idol Feb 23, 2025, 4:36 PM

#

yeah I was about to say that those bytes looked familiar

flint idol Feb 23, 2025, 4:52 PM

#

I think I have improved io performance a lot by making this page cache change

#

at the expense of the uacpi score

#

ok it boots

#

into bash

#

except for:

#

this is a bug in map view of user memory

#

it stopped page faulting

#

instead execve returns ENOEXEC

#

the behaviour that happens is mixed

#

#

ok yeah I'm just going to change my execve syscall

#

to take in a filename

flint idol Feb 23, 2025, 5:35 PM

#

yay I can ls again

#

just committed everything trl

sweet compass Feb 23, 2025, 5:42 PM

#

😢

flint idol Feb 23, 2025, 5:42 PM

#

install libcjson-dev

#

on debian/ubuntu

#

idk what the arch package is

#

probably something similar

sweet compass Feb 23, 2025, 5:43 PM

#

flint idol idk what the arch package is

only cjson

flint idol Feb 23, 2025, 5:43 PM

#

install that ig

sweet compass Feb 23, 2025, 5:43 PM

#

now it gives me this.

#

it would probably be a good idea to mention that in the docs

flint idol Feb 23, 2025, 5:43 PM

#

oops yeah

sweet compass Feb 23, 2025, 5:44 PM

#

:p

flint idol Feb 23, 2025, 5:44 PM

#

cp settings-x86_64.json settings.json

#

lol

sweet compass Feb 23, 2025, 5:44 PM

#

oh fuck no it wants to build a cross-compiler

flint idol Feb 23, 2025, 5:45 PM

#

ye

#

lol

sweet compass Feb 23, 2025, 5:45 PM

#

well I'm training AI in the background

#

so that's not going to work

#

would you happen to have a .iso?

flint idol Feb 23, 2025, 5:46 PM

#

yes

sweet compass Feb 23, 2025, 5:46 PM

#

pepepray

flint idol Feb 23, 2025, 5:46 PM

#

also that's for building a full obos distro, if you jsut want the kernel, you should use method 2 of building

#

📎 obos.iso

sweet compass Feb 23, 2025, 5:46 PM

#

yeah but I want to test the full distro ~~to see how badly I can break it~~

flint idol Feb 23, 2025, 5:46 PM

#

sure

weak kestrel Feb 23, 2025, 5:46 PM

#

flint idol yes

obos built with clang when?

flint idol Feb 23, 2025, 5:46 PM

#

it takes a few commands until bash mysteriously crashes

flint idol Feb 23, 2025, 5:47 PM

#

weak kestrel obos built with clang when?

the kernel can be built with clang

#

also that wouldn't change the fact that it'd need to build a cross compiler

#

since clang+llvm obviously won't accept obos into mainstream

weak kestrel Feb 23, 2025, 5:47 PM

#

ah I thought you needed a cross-compiler for the kernel too

flint idol Feb 23, 2025, 5:47 PM

#

nah

#

it can be built using clang

#

the cross compiler is mainly used for userspace

sweet compass Feb 23, 2025, 5:48 PM

#

#

meme

flint idol Feb 23, 2025, 5:48 PM

#

it still builds the cross compiler

#

what you could do is

#

copy these two files into repo root/pkginfo

📎 pkginfo_bootstrap-binutils.bin 📎 pkginfo_bootstrap-gcc.bin

flint idol Feb 23, 2025, 5:49 PM

#

flint idol the cross compiler is mainly used for userspace

but the kernel has a configure option to build it using the x86_64-obos-gcc compiler

flint idol Feb 23, 2025, 5:49 PM

#

flint idol copy these two files into repo root/pkginfo

then do what you just did, but with x86_64-obos as the target triplet

#

whether it works or not, probably not

#

unless you change the sysroot

#

with that

sweet compass Feb 23, 2025, 5:49 PM

#

what the fuck?

#

oh

#

I forgot -cpu host

flint idol Feb 23, 2025, 5:50 PM

#

wait what the fuck since when did that happen

#

oh yeah

#

I once fixed it but never upstreamed the changes

sweet compass Feb 23, 2025, 5:50 PM

#

regression #TREE(3)

flint idol Feb 23, 2025, 5:50 PM

#

if your cpu doesn't have that, it's not an obos regression meme

#

anyway

#

bash works over serial btw

#

use nc or smth and -serial whatever

sweet compass Feb 23, 2025, 5:51 PM

#

oh lol

#

it doesn't echo

#

(with -serial stdio)

flint idol Feb 23, 2025, 5:52 PM

#

yeah the (host) tty would need to be in cooked mode

#

I just need to implement

#

ttys one day

sweet compass Feb 23, 2025, 5:53 PM

#

I don't even have to try to break it to break it ultrameme

flint idol Feb 23, 2025, 5:53 PM

#

it's not my fault you misconfigured it meme

#

this is more of "unimplemented" than "broken"

#

anyway PRs for a ps2 keyboard driver and TTY support are welcome meme

#

also that build does not have the network stack

#

@sweet compass did it work? (or did you give up?)

sweet compass Feb 23, 2025, 5:58 PM

#

I gave up

#

too much effort :p

#

get a graphical tty working

flint idol Feb 23, 2025, 10:26 PM

#

oops

real pecan Feb 23, 2025, 10:26 PM

#

obos for quantum computers

flint idol Feb 23, 2025, 10:27 PM

#

yeah idk what's up with the bug in accounting there

#

but I will fix it one day

weak kestrel Feb 23, 2025, 10:28 PM

#

nootlikethis

flint idol Feb 23, 2025, 10:29 PM

#

same thing happens after wake from suspend thinkong

#

yeah but anyway

#

obos now has sys_sync

#

it probably wouldn't have been possible with the old page cache system

#

obos has a bug with the initrd driver where if the path starts with./ it kind of shits itself

#

which I found out when I wrote the oboskrnl recipe for obos-pkgs

#

it'd be interesting to see if I could get the obos package repo to work

#

for linux

#

sometimes I look at my code and wonder

#

what brought me to do all this

#

spend all this time

#

that could've been used to do anything else

weak kestrel Feb 23, 2025, 10:46 PM

#

flint idol that could've been used to do anything else

there's nothing else better

#

obos is a great achievement

flint idol Feb 23, 2025, 10:47 PM

#

ye-```
)
( /( (
)()) ( ( ( )\ ) (
(()\ ))\ )( ( ))(() ) ( /( ( )\ ( (_ ((_)/((_)(()\ )\ ) /((_)_ /(/( )(_)) )\ )((_) )\ | |/ /(_)) ((_p) _(_/( (_)) | | ((_)_\ ((_)_ _(_/( (_) ((_) | ' < / -_) | '_|| ' \))/ -_)| | | '_ \)/ _ || ' ))| |/ |
||_\| || |||| _||_| | ./ _,||||| ||_|
|_|

Kernel Panic in OBOS b3f451c8534626837090ad27ee7446e2867016e9-dirty built on Feb 19 2025 at 20:17:24.
Crash is on CPU 0 in thread 6, owned by process 1. Reason: OBOS_PANIC_ASSERTION_FAILED.
Currently running on a Intel(R) Core(TM) i5-4590T CPU @ 2.00GHz. We are currently running on a hypervisor
Information on the crash is below:
Assertion failed in function Free. File: /home/oberrow/Code/obos/src/oboskrnl/allocators/basic_allocator.c, line 342. reg->alloc_size == nBytes

    Address                 Symbol

ffffffff80015030 oboskrnl!Free+4d0
ffffffff80033749 oboskrnl!Sys_FdOpen+139
ffffffff8000c332 Unresolved/External
0000000000000000 End

    Address                  Main TID     Driver Name
    ffff9000000ec000               -1     Initial Ramdisk (InitRD) Driver
    ffff9000003a8000               -1     Bochs VBE Driver
    ffff9000003d2000               -1     COM Driver
    ffff90000022a000               -1     FAT Driver

weak kestrel Feb 23, 2025, 10:47 PM

#

I can't even read a doc to write something and your os is the first to implement suspend for one

flint idol Feb 23, 2025, 10:47 PM

#

*first hobby os on this server

#

to be exact

weak kestrel Feb 23, 2025, 10:48 PM

#

yes

flint idol Feb 23, 2025, 10:48 PM

#

like it is pretty cool

#

it's just

#

yes

#

anyway brb

#

I need to make a FIXME list for obos soon

#

some drivers have been starting to regress

#

and I was "fixing" that by not loading them

#

meme

weak kestrel Feb 23, 2025, 10:49 PM

#

flint idol it's just

I look at hobby osdev not as a thing to make something useful but as a learning process

flint idol Feb 23, 2025, 10:49 PM

#

indeed

#

it's just a wonder I have

#

like a wonder: what else could I have been doing

flint idol Feb 23, 2025, 10:58 PM

#

flint idol I need to make a FIXME list for obos soon

I think I will do that

#

I'll use github issues

#

for easier tracking

#

anyway, I feel like implementing some sort of init program for obos

#

that just does (add something useful here) then starts a shell

#

or what it hands control to will be configurable

sweet compass Feb 23, 2025, 11:31 PM

#

flint idol that just does (add something useful here) then starts a shell

(add something useful here):

fork-and-exec a graphical tty host

flint idol Feb 24, 2025, 12:12 AM

#

sweet compass (add something useful here): - fork-and-exec a graphical tty host

Fine. I will finally implement TTYs

flint idol Feb 24, 2025, 12:33 AM

#

flint idol at the expense of the uacpi score

nvm that was because the thing was being compiled in debug mode

flint idol Feb 24, 2025, 12:50 AM

#

fuck you gcc

#

stop optimizing out

#

my goddamn

#

variables

weak kestrel Feb 24, 2025, 12:54 AM

#

-O0 troll2

uncut narwhal Feb 24, 2025, 12:55 AM

#

flint idol variables

pass a pointer to it as an operand to an inline asm block with no instructions and it cant optimize it out

flint idol Feb 24, 2025, 12:56 AM

#

it was for debugging purposes

#

anyway I solved it

ornate ginkgo Feb 24, 2025, 1:17 AM

#

uncut narwhal pass a pointer to it as an operand to an inline asm block with no instructions a...

Lol

#

That's the proper fix for that

flint idol Feb 24, 2025, 1:21 AM

#

me: does nothing
meanwhile obos:

[ LIBC  ] __cxa_guard_acquire contention
[ DEBUG ] (thread 7) syscall Sys_LibCLog returned 0x0000000000000000
[  LOG  ] User thread 7 SIGILL
[ DEBUG ] Exitting process 1 after receiving signal 4

#

ah it only happens when I change one of the arguments passed to /init

#

fuck this shit

#

uh oh

#

the Mm_AnonPage page is kind of not all zeroes

#

either it was corrupted

#

or I forgot to zero it

#

it was corrupted nooo

flint idol Feb 24, 2025, 9:35 PM

#

so today I discovered a bug

flint idol Feb 24, 2025, 9:36 PM

#

flint idol the `Mm_AnonPage` page is kind of not all zeroes

I kinda accidentally added this to the standby list

weak kestrel Feb 24, 2025, 9:44 PM

#

flint idol so today I discovered a bug

did you fix it?

flint idol Feb 24, 2025, 9:44 PM

#

no

weak kestrel Feb 24, 2025, 9:45 PM

#

Sadge

flint idol Feb 24, 2025, 9:45 PM

#

but that was still a bug

#

which would come back to bite me soon

proper trellis Feb 24, 2025, 10:06 PM

#

could it be this memzero? The kbase has been aligned down (from Mm_MapViewOfUserMemory) and when you add the phdrs[i].p_filesz to it you are missing the removed padding, essentially zeroing the wrong range. https://github.com/OBOS-dev/obos/blob/26aebb546fd794872c2f23dff1287bf53ae7cdcd/src/oboskrnl/elf/load.c#L224

flint idol Feb 24, 2025, 10:07 PM

#

perhaps...

#

let me check

proper trellis Feb 24, 2025, 10:09 PM

#

The padding removed:

        uintptr_t real_base = phdrs[i].p_vaddr;
        real_base -= (real_base%OBOS_PAGE_SIZE);

so add it back

memzero((void*)((uintptr_t)kbase+phdrs[i].p_filesz+(phdrs[i].p_vaddr%OBOS_PAGE_SIZE)), phdrs[i].p_memsz-phdrs[i].p_filesz);

flint idol Feb 24, 2025, 10:10 PM

#

        kbase = Mm_MapViewOfUserMemory(ctx,
                                       (void*)real_base,
                                       nullptr,
                                       real_limit-real_base,
                                       0,
                                       false, // disrespect user protection flags.
                                       &status) + (phdrs[i].p_vaddr % OBOS_PAGE_SIZE);```

#

the padding is indeed readded

proper trellis Feb 24, 2025, 10:11 PM

#

ah ok nvm

flint idol Feb 24, 2025, 10:11 PM

#

BUT this will help finding the actual bug

#

yeah it's sometime after a Mm_MapViewOfUserMemory

#

probably happens at the memcpy above it

#

because of a bug in Mm_MapViewOfUserMemory

#

well well well...

#

it indeed is a bug in Mm_MapViewOfUserMemory

#

holy...

#

lol

#

although in no case is said memory writable...

#

it's somewhere in the elf loader

#

and that's all I know

proper trellis Feb 24, 2025, 10:23 PM

#

flint idol well well well...

what exactly is wrong here?

flint idol Feb 24, 2025, 10:24 PM

#

0x7a800000 is the physical address of the anon page

#

that was in map view of user memory

#

ok so it isn't in the elf loader

#

it happens at precisely line 117 of arch/x86_64/execve.c

#

in memcpy_k_to_usr

#

oh

flint idol Feb 24, 2025, 11:57 PM

#

anyway

#

so that bug was fixed

#

but fork seems to be broken

#

or rather execve

#

#

maybe this is the same bug

#

that crashes bash

#

holy

#

there is no way that bug in bash

#

this entire time

#

was caused because

#

of the rip of the entry being zero

#

when it shouldn't

#

what the fudge

#

I think the elf loader is failing...

#

I will test that theory in a bit

flint idol Feb 25, 2025, 1:05 AM

#

#

Status: OBOS_STATUS_IN_USE

flint idol Feb 25, 2025, 2:28 AM

#

the bug is that

#

the user stack is the only thing that isn't nuked by execve

#

and then the program is trying to be loaded somewhere where the stack exists

#

therefore: OBOS_STATUS_IN_USE from Mm_VirtualMemoryAlloc

#

uh oh

#

for some reason

#

the kernel accidentally reallocates a handle

flint idol Feb 25, 2025, 2:59 AM

#

I fucking hate userspace

#

why can't it just be fucking normal

#

and not fucking crash

#

I hate this

#

I hate this.

#

[ DEBUG ] (thread 7) syscall Sys_FdSeek(0x2, 0x0, 0x1, 0x0, 0x0)
[ DEBUG ] (thread 7) syscall Sys_FdSeek returned 0x0000000000000000
[ DEBUG ] (thread 7) syscall Sys_FdTellOff(0x2, 0x0, 0x0, 0x0, 0x0)
[ DEBUG ] (thread 7) syscall Sys_FdTellOff returned 0x0000000000000000
[ DEBUG ] (thread 7) syscall Sys_FdRead(0x0, 0x915c5f, 0x1, 0x915c28, 0x0)
[ DEBUG ] (thread 7) syscall Sys_FdRead returned 0x0000000000000000
[  LOG  ] User thread 7 SIGSEGV (rip 0x0, cr2 0x0, error code 0x00000014)

#

anyway

#

I just closed https://github.com/OBOS-dev/obos/issues/34

GitHub

Bash crash after running a few commands · Issue #34 · OBOS-dev/obos

Bash has been observed to easily SEGFAULT after running a few "external" commands (i.e., commands that require a fork+execve) Configure command: cmake -Bbuild -DCMAKE_BUILD_TYPE=Debug -DO...

#

so nice to be able to run bash without constant crashes

#

#

ran kill -sKILL 3

#

trl

#

I want to implement whatever is needed for getrusage one day

flint idol Feb 25, 2025, 5:49 PM

#

my friend said he is going to implement "minigames" for obos

#

so I am going to see where that goes

#

cc @candid cape

#

btw this code is still untested

#

lollll

#

I have come such a long way

#

idk why but I have an urge to add a c++ interface to the kernel for drivers

#

so I can make c++ drivers and abstract kernel objects behind OOP

weak kestrel Feb 25, 2025, 7:29 PM

#

based

frigid tide Feb 25, 2025, 7:30 PM

#

flint idol so I can make c++ drivers and abstract kernel objects behind OOP

This is the way

#

High level abstractions>

#

WTF IS AN EXO KERNEL

real pecan Feb 25, 2025, 7:31 PM

#

flint idol so nice to be able to run bash without constant crashes

what was the bug

flint idol Feb 25, 2025, 8:06 PM

#

real pecan what was the bug

basically, sometimes stacks were allocated at the same place as where the ELF wants to be loaded

#

and I didn't unmap user stacks in execve

#

for some reason

real pecan Feb 25, 2025, 8:07 PM

#

what the fuck

flint idol Feb 25, 2025, 8:07 PM

#

therefore the elf loader failed

#

but I never checked for that

#

since I did a dry run beforehand

real pecan Feb 25, 2025, 8:07 PM

#

bruh

flint idol Feb 25, 2025, 8:07 PM

#

and then the aux values including the entry

#

were nullptr

#

so when control was handed off it jumped to nullptr

flint idol Feb 26, 2025, 12:04 AM

#

hmm for some reason

#

signals never get ran

#

when signalling the current thread

#

unless it's a bug with fork

flint idol Feb 26, 2025, 12:24 AM

#

it isn't any of that

#

basically OBOS_SyncPendingSignal is never getting called

#

for that specific user thread

#

because it's never actually leaving an IRQ handler, which is where that is usually called

#

void CoreS_ExitIRQHandler(interrupt_frame* frame)
{
    if (~frame->cs & 0x3 && CoreS_GetCPULocalPtr()->currentThread)
        CoreS_GetCPULocalPtr()->currentContext = CoreS_GetCPULocalPtr()->currentThread->proc ? CoreS_GetCPULocalPtr()->currentThread->proc->ctx : &Mm_KernelContext;
    else
        OBOS_SyncPendingSignal(frame);
    cli();
}```

timid elk Feb 26, 2025, 9:47 PM

#

when DNS in OBOS

#

so you can blame issues on DNS

#

it is always DNS

short mortar Feb 26, 2025, 10:18 PM

#

What if I do networking after the sleep stuff?

flint idol Feb 26, 2025, 10:23 PM

#

short mortar What if I do networking after the sleep stuff?

wrong thread?

flint idol Feb 26, 2025, 10:24 PM

#

timid elk when DNS in OBOS

where is DNS even usually done, kernel mode or userspace?

white mulch Feb 26, 2025, 10:24 PM

#

usually its done in the libc

trim bay Feb 26, 2025, 10:24 PM

#

pretty sure linux does dns in userspace (libc)

short mortar Feb 26, 2025, 10:25 PM

#

But you still need a bunch of stuff for it?

white mulch Feb 26, 2025, 10:25 PM

#

udp sockets

short mortar Feb 26, 2025, 10:26 PM

#

to localhost:5353?

#

(I forgot what was the file with the DNS servers)

white mulch Feb 26, 2025, 10:27 PM

#

to localhost:53 or to a proper dns server yes

#

/etc/resolv.conf is where they are

short mortar Feb 26, 2025, 10:27 PM

#

Like you probably want dnsmasq or something

flint idol Feb 26, 2025, 10:27 PM

#

does mlibc do it?

white mulch Feb 26, 2025, 10:27 PM

#

yes

flint idol Feb 26, 2025, 10:27 PM

#

ok so no need to worry about that

white mulch Feb 26, 2025, 10:27 PM

#

short mortar Like you probably want dnsmasq or something

that's only required for server

#

usually for client the dns queries are sent to whatever nameserver, either the router or some public one like 1.1.1.1

flint idol Feb 26, 2025, 10:29 PM

#

when managarm dns server??

white mulch Feb 26, 2025, 10:30 PM

#

flint idol ok so no need to worry about that

unless you want something like kgdb be able to connect via a hostname meme

flint idol Feb 26, 2025, 10:31 PM

#

isn't kgdb usually the other way around

#

the kernel just has the gdb stub

#

and someone connects to it

#

what does the hostname have to do

white mulch Feb 26, 2025, 10:31 PM

#

ah yeah, klog would be a better candidate for that

#

though it could also be initialized by the listener side if it knows how to access the sender side

flint idol Feb 27, 2025, 12:28 AM

#

ok now I am pink

opaque pulsar Feb 27, 2025, 12:34 AM

#

this thread will be locked now, as the OP has been banned (reason, see here: #modmail message)

inland radish Jan 4, 2026, 7:14 PM

#

#modmail message welcome back i guess

vale nymph Jan 4, 2026, 7:14 PM

#

https://tenor.com/view/the-return-of-the-king-lord-of-the-rings-ro-tk-title-card-gif-22191928

Tenor

flint idol Jan 4, 2026, 7:22 PM

#

hi guys im back

sterile sparrow Jan 4, 2026, 7:22 PM

#

wb

real pecan Jan 4, 2026, 7:29 PM

#

welcome back

flint idol Jan 4, 2026, 7:38 PM

#

thx yall

#

anyway

#

flint idol Jan 4, 2026, 7:39 PM

#

flint idol

^ Xorg and xeyes running on obos natively

#

that is my last progress report i posted on my obos server

#

last month

#

but what is happening now:

#

i am fixing ext2 driver bugs

#

and also getting mad at qemu because it's segfaulting

#

https://nvd.nist.gov/vuln/detail/CVE-2019-12067

flint idol Jan 4, 2026, 7:40 PM

#

flint idol https://nvd.nist.gov/vuln/detail/CVE-2019-12067

because of this

#

although luckily my ahci driver manages to relatively consistently reproduce the bug

#

which someone else (with my help) managed to narrow it down to a race condition in the ahci virtual device in qemu

#

I also wrote an audio server for obos

#

it aint the greatest but it's honest work

#

https://github.com/OBOS-dev/obos-aud

GitHub

GitHub - OBOS-dev/obos-aud: OBOS Audio Server

OBOS Audio Server. Contribute to OBOS-dev/obos-aud development by creating an account on GitHub.

flint idol Jan 4, 2026, 7:41 PM

#

flint idol https://github.com/OBOS-dev/obos-aud

(it can be found here)

flint idol Jan 4, 2026, 7:42 PM

#

flint idol basically `OBOS_SyncPendingSignal` is never getting called

I did fix this bug pretty recently

#

void CoreS_ForceYieldOnSyscallReturn()
{
    ipi_vector_info vector = {
        .deliveryMode = LAPIC_DELIVERY_MODE_FIXED,
        .info.vector = Core_SchedulerIRQ->vector->id + 0x20
    };
    asm volatile ("cli");
    static ipi_lapic_info self = {
        .isShorthand=true,
        .info.shorthand=LAPIC_DESTINATION_SHORTHAND_SELF
    };
    Arch_LAPICSendIPI(self, vector);    
}```

#

with this

#

and then this is called if it exists at the end of blocking syscalls

#

i also fixed a lot of posix compliance issues with pipes, etc.

#

and i also have a network stack and posix sockets (including AF_UNIX, SOCK_STREAM sockets, as well as AF_INET sockets)

flint idol Jan 4, 2026, 7:44 PM

#

flint idol i am fixing ext2 driver bugs

I have also added file creation and hard link creation to this driver

#

but alas, symlink creation does not work

#

i have also recently ported ffmpeg

#

and it doesnt completely work yet which i will debug some other time

#

who just reacted with 🫃

#

💀

real pecan Jan 4, 2026, 7:47 PM

#

lol

flint idol Jan 4, 2026, 7:48 PM

#

infy, have u any idea how fast symlinks work in ext2

real pecan Jan 4, 2026, 7:48 PM

#

nope

flint idol Jan 4, 2026, 7:48 PM

#

damn

#

guess im gonna actually have to research

devout niche Jan 4, 2026, 7:49 PM

#

the text is stashed where the data block pointers would usually be

flint idol Jan 4, 2026, 7:49 PM

#

hm

#

    bool dirty = false;
    int ea_blocks = le32_to_host(inode->file_acl) ? (hnd->cache->block_size>>9) : 0;
    bool fast_symlink = !(le32_to_host(inode->blocks) - ea_blocks);
    size_t path_len = strlen(to);

    if (fast_symlink)
        memcpy((char*)&inode->direct_blocks, to, path_len);
    else
    {
        obos_status status = ext_ino_resize(hnd->cache, hnd->ino, path_len, false);
        if (obos_is_error(status))
        {
            MmH_DerefPage(pg);
            return status;
        }

        status = ext_ino_commit_blocks(hnd->cache, hnd->ino, 0, path_len);
        if (obos_is_error(status))
        {
            MmH_DerefPage(pg);
            return status;
        }
        
        status = ext_ino_write_blocks(hnd->cache, hnd->ino, 0, path_len, to, nullptr);
        if (obos_is_error(status))
        {
            MmH_DerefPage(pg);
            return status;
        }
    }```

#

i currently have this code to set the linked path but alas, it does not work

#

fsck.ext2 reports that the symlink on that inode is corrupted

#

ill just look at an inode for a fast symlink

#

and see what file_acl should be

#

i do see that 'size' is the path length

#

yeah that makes sense

#

yeah i set i_size equal to path_len and that fixes it

vale nymph Jan 4, 2026, 7:54 PM

#

flint idol infy, have u any idea how fast symlinks work in ext2

u just store it in the actual inode

#

    // if the length of a symlink is larger than 60 bytes, its stored normally
    // otherwise, its stored in the inode strucutre itself
    if (linksize >= 60)
        err = rwbytes(fs, node, buf, linksize, 0, false, true);
    else
        memcpy(buf, node->inode.directpointer, linksize);

#

actual astral code

#

💀

flint idol Jan 4, 2026, 8:00 PM

#

    size_t path_len = strlen(to);
    bool fast_symlink = path_len <= 60;

    if (fast_symlink)
        memcpy((char*)&inode->direct_blocks, to, path_len);
    else
    {
        obos_status status = ext_ino_resize(hnd->cache, hnd->ino, path_len, false);
        if (obos_is_error(status))
        {
            MmH_DerefPage(pg);
            return status;
        }

        status = ext_ino_commit_blocks(hnd->cache, hnd->ino, 0, path_len);
        if (obos_is_error(status))
        {
            MmH_DerefPage(pg);
            return status;
        }
        
        status = ext_ino_write_blocks(hnd->cache, hnd->ino, 0, path_len, to, nullptr);
        if (obos_is_error(status))
        {
            MmH_DerefPage(pg);
            return status;
        }
    }
    
    inode->size = path_len;```

#

dw this is what i ended up settling on

#

except when i have a long symlink

#

everything (including debugfs and obos and fsck) starts tripping

#

obos thinks the link is there but also doesnt at the same time

#

(stat and ls do not show it, but an attempt to make the link will say it already exists on obos)

#

(debugfs shows it existing but doesnt allow me to stat it)

#

(fsck shows the same amount of files existing)

#

#

nvm it works

#

i mean like non fast links

torpid wigeon Jan 4, 2026, 8:28 PM

#

welcome back

flint idol Jan 4, 2026, 9:06 PM

#

so stupid

#

why does sys_rmdir exist

#

is AT_REMOVEDIR linux specific or something

#

no it isnt

flint idol Jan 4, 2026, 9:07 PM

#

flint idol why does sys_rmdir exist

i personally expected rmdir just to use unlinkat with that flag

white mulch Jan 4, 2026, 9:07 PM

#

you can compile mlibc as ansi-only, then it can't use unlinkat

flint idol Jan 4, 2026, 9:08 PM

#

sir

#

i did just end up making sys_rmdir return sys_unlinkat(AT_FDCWD, path, AT_REMOVEDIR)

white mulch Jan 4, 2026, 9:10 PM

#

ah, that's kinda stupid it uses sys_rmdir but then it also needs unlinkat for the full functionality anyway

#

though I guess given that it doesn't pass any flags to unlinkat and only cares about the path part it's possible to just implement that as a sys_remove_file or smth

leaden carbon Jan 4, 2026, 10:32 PM

#

welcome back

flint idol Jan 4, 2026, 11:24 PM

#

leaden carbon welcome back

thanks

flint idol Jan 4, 2026, 11:29 PM

#

flint idol ran `kill -sKILL 3`

it would be funny

#

if that still happens

empty kernel Jan 4, 2026, 11:46 PM

#

Welcome back

ornate ginkgo Jan 5, 2026, 12:20 AM

#

^

flint idol Jan 5, 2026, 12:23 AM

#

flint idol me: *does nothing* meanwhile obos: ``` [ LIBC ] __cxa_guard_acquire contention ...

i remember this this would always piss me off so fucking much

#

iirc the bug was in my scheduling code after all (i had come to the premature conclusion that it wasnt)

flint idol Jan 5, 2026, 12:24 AM

#

flint idol anyway, I feel like implementing some sort of init program for obos

i did end up doing that

#

oh yeah gcc+binutils works now

#

and obos also now has a file server

#

http://obos-dev.ddns.net/

#

(update on this: i literally ended up rewriting pipes like twice because my implementation was so fuckin shit)

#

oh dear i still havent gotten formatting done

#

oh dear god what will the diff be now

#

the ms style though is the closest to obos' preexisting style

#

i remember the day i added this, t'was a nice day, went on a walk at some point

#

man i really want to make an xhci or some sort of usb driver

#

but ive no idea how usb works

#

yeah since february my kernel has changed a shitton, too much to cover easily

#

back then i had a good 15 packages (in obos-pkgs, does not mean 15 ports) at most

#

now there are 121 packages registered in the index

#

and i have a lot more ports (not every port works though because posix or something)

#

and man was i incompetent back when i was writing the base of the kernel

#

the amount of bug fixes that i have had to do because "why the fuck was that code here in the first place??"

#

is way too much

#

real pecan Jan 5, 2026, 1:00 AM

#

Damn

#

When drm

flint idol Jan 5, 2026, 2:21 AM

#

real pecan When drm

prs welcome

#

trl

flint idol Jan 5, 2026, 2:22 AM

#

flint idol

lol i still have the example code comments in my html doc

#

on a separate note, obos' io speed is so terrible it hurts

#

500 bytes a goddamn second

#

or 300 in that screenshot

#

omg obos it's a .5 megabyte file

#

the bottleneck seems to be in writing back the file

#

weird... timeouts do not seem to be working properly

#

curl is setting a 1s timeout for poll, and hte kernel is returning immediately

#

this is like 90s era download speed

flint idol Jan 5, 2026, 2:37 AM

#

flint idol the bottleneck seems to be in writing back the file

or also perhaps because i set the rx window to 4096 bytes

#

(i should probably increase that)

stark sigil Jan 5, 2026, 2:38 AM

#

operating system

flint idol Jan 5, 2026, 2:41 AM

#

uh yeah that's what this is

flint idol Jan 5, 2026, 2:47 AM

#

flint idol (i should probably increase that)

ok making the rx window 16KiB improves download speed a ton

#

like by literal orders of magnitude

#

increasing the rx buffer size in the e1000 driver also improves network speeds

#

i will say that xbps-fetch is a lot faster for some reason

sterile sparrow Jan 5, 2026, 2:58 AM

#

Thanks for reminding me that I need to write a network stack for my own os

flint idol Jan 5, 2026, 2:59 AM

#

np

#

oh tip

#

when u write the tcp part of the stack

#

do not be overwhelmed by the rfc document

#

at the bottom

#

there is an entire guide on how to process an incoming segment

#

in rfc793

#

section 3.9, "SEGMENT ARRIVES"

#

it describes in detail what u should do in any case

#

it's what saved my tcp implementation from being absolute garbage

flint idol Jan 5, 2026, 3:01 AM

#

flint idol it describes in detail what u should do in any case

(to the point where it tells u shit like "increase rcv.ack by seg.len bytes")

#

although tcp has greatly changed since... september of 1981, rfc 793 still works with modern tcp implementations

#

maybe one day ill implement RFC9293

#

but until then my tcp stack is gonna stay 44 years in the past

flint idol Jan 5, 2026, 3:27 PM

#

icl ts pmo

#

I wanna go home and obosdev

stark sigil Jan 5, 2026, 6:55 PM

#

flint idol I wanna go home and obosdev

more operating system

stark sigil Jan 5, 2026, 6:55 PM

#

flint idol uh yeah that's what this is

i wanted to comment something but like idk

#

this looks cool

flint idol Jan 5, 2026, 10:07 PM

#

thx

real pecan Jan 5, 2026, 11:58 PM

#

obos updates waiting room

flint idol Jan 5, 2026, 11:58 PM

#

oh yeah

#

currently about to debug this

#

ok that was simple

#

turns out i forgot to initialize the timer event in pselect

flint idol Jan 6, 2026, 12:37 AM

#

i fixed some dumb memory leaks

#

in the userspace

#

because not all user memory views were being unmapped fully

flint idol Jan 6, 2026, 1:18 AM

#

i feel like it could be worth it

#

to make it so IRPs can get a scatter gather list

#

instead of a virtual address

#

because rn the page cache does:

needs data from disk driver
has physical address
before reading, it needs that address in hhdm since read_sync takes a virtual address
in the disk driver, it needs a physical address
it converts the virtual address back to a scatter gather list with physical addresses
we're back where we started with the physical address

#

man did i forget how much i got distracted from this server

flint idol Jan 6, 2026, 1:50 AM

#

real pecan obos updates waiting room

infy can u give me a random thing to do in my kernel

#

im bored and i dont wanna debug ext2 or xorg

#

"make a usb driver" does not count btw

#

i know

#

ill debug real hardware

#

because that always does not lose me the will to do osdev at all!

flint idol Jan 6, 2026, 2:11 AM

#

so for some reason

#

on real hardware

#

with OBOS_DEBUG_FREE_SIZE enabled

#

it will hang in teh allocator

#

during pci init

#

only if that's enabled

inland radish Jan 6, 2026, 6:44 AM

#

flint idol to make it so IRPs can get a scatter gather list

that's just about the MDL's purpose in NT, fwiw

flint idol Jan 6, 2026, 12:50 PM

#

inland radish that's just about the MDL's purpose in NT, fwiw

do you think it would be worth my time

#

to implement such a thing

inland radish Jan 6, 2026, 1:45 PM

#

flint idol do you think it would be worth my time

Yeah, I think it would

#

They boil down to "list of pinned physical pages"

flint idol Jan 7, 2026, 4:42 AM

#

#

someone is working on software entropy for obos

#

(or well they are porting their software etnropy thing to obos)

flint idol Jan 7, 2026, 5:46 PM

#

they added some aes ndndd sja thing for the rng

#

torpid wigeon Jan 7, 2026, 5:47 PM

#

did bro have a stroke

flint idol Jan 7, 2026, 5:48 PM

#

aes ctr drbg

#

or something

#

is what the rng is called

flint idol Jan 7, 2026, 5:48 PM

#

torpid wigeon did bro have a stroke

lol

flint idol Jan 8, 2026, 11:46 AM

#

https://github.com/OBOS-dev/obos/pull/52

GitHub

TJEC and AES128-CTR-DRBG random source by MarcasRealAccount · Pull...

This Pull Request replaces the Mersenne Twister random source for a better combination of Time Jitter Entropy Collector in conjunction with AES128 in Counter mode as a Deterministic Random Bit Gene...

#

this is what I was talking about

astral mason Jan 8, 2026, 7:33 PM

#

Oh hey, oberrow's back! poggers

flint idol Jan 8, 2026, 7:48 PM

#

astral mason Oh hey, oberrow's back! <:poggers:1337923490329792552>

Indeed I am

real pecan Jan 8, 2026, 8:34 PM

#

where updates

flint idol Jan 8, 2026, 9:03 PM

#

flint idol https://github.com/OBOS-dev/obos/pull/52

here

#

im also working on ipv6

real pecan Jan 8, 2026, 9:07 PM

#

this is literally marca KEKW

flint idol Jan 8, 2026, 9:10 PM

#

what no im not marca

#

why would u say that....

real pecan Jan 8, 2026, 9:11 PM

#

i mean the pr u linked is literally by marca

flint idol Jan 8, 2026, 9:11 PM

#

yes it is

#

a few days ago he was asking me when im gonna get real rng for obos

#

and I said prs accepted

#

and well

#

here we are

real pecan Jan 8, 2026, 9:12 PM

#

lol

flint idol Jan 8, 2026, 9:12 PM

#

infy when will u contribute code to obos trl

#

Emily contributed a tcc patch

real pecan Jan 8, 2026, 9:12 PM

#

I only contribute to minecraft trl

flint idol Jan 8, 2026, 9:12 PM

#

flint idol Emily contributed a tcc patch

Marca is going to have his rng pr merged when he fixes the shit i said

flint idol Jan 9, 2026, 1:40 AM

#

    char** ret = ZeroAllocate(OBOS_KernelAllocator, count+1, sizeof(char*), statusp);
    // ...
        obos_status status = OBOSH_ReadUserString(vec[i], nullptr, &str_len);
        if (obos_is_error(status))
        {
            Free(OBOS_KernelAllocator, ret, count*sizeof(char*)); // <- here

just found a vulnerability

#

if the user passes 5 arguments in argv to execve, but argv[4] is an invalid pointer

#

the kernel will free 32 bytes, when it should be freeing 40 bytes

#

and like

#

i forgot how this is a problem

#

brain fart, freeing less won't ever cause a problem except for memory leaked

flint idol Jan 9, 2026, 3:42 AM

#

fvwm3 is getting ENOSYS on a file open

#

and i have no idea

#

where

#

it is from an ioctl

#

(then SIGSEGV)

flint idol Jan 9, 2026, 3:56 AM

#

flint idol it is from an ioctl

it was failing on FIOCLEX

#

for some time after finals: get fvwm3 to work

flint idol Jan 9, 2026, 3:03 PM

#

oh yeah yesterday

#

I implemented the interpreter thing in execve

flint idol Jan 9, 2026, 3:30 PM

#

https://github.com/OBOS-dev/obos/blob/master/src%2Foboskrnl%2Fscheduler%2Fsched_sys.c#L353

GitHub

obos/src/oboskrnl/scheduler/sched_sys.c at master · OBOS-dev/obos

I was bored... Contribute to OBOS-dev/obos development by creating an account on GitHub.

flint idol Jan 9, 2026, 3:58 PM

#

todo fix bug there

#

memory leaks here

#

https://github.com/OBOS-dev/obos/blob/master/src%2Foboskrnl%2Fscheduler%2Fsched_sys.c#L230

GitHub

obos/src/oboskrnl/scheduler/sched_sys.c at master · OBOS-dev/obos

I was bored... Contribute to OBOS-dev/obos development by creating an account on GitHub.

flint idol Jan 9, 2026, 11:08 PM

#

i might implement virtual page locking finally

#

what ill do is have a flag in the struct page saying the thing is locked

#

as well as a bit in the PTE signifying the page is locked

#

in anything to do with paging out, it will check the PTE bit first

#

then if it's true, it will not page it out

#

otherwise it checks the struct page and if that bit is set, it will not page it out, else it will remap the page with the PTE "page locked" bit set

#

else it goes on with consideration for paging out

#

either i will do that

#

or put the pages in the working set on lock, and remove them on unlock

#

but i dont like that because removing a page from the working set causes it to be paged out which could be bad

inland radish Jan 10, 2026, 7:37 AM

#

flint idol i might implement virtual page locking finally

What would you use this for?

#

I would just use the page's reference count for that purpose

#

As in, if the page is free of references, it goes to the standby list, modified list, or I haven't decided for anonymous sections

flint idol Jan 10, 2026, 12:09 PM

#

I forgot why I needed it

#

I do know I will at least want it for posix

#

and iirc my kernel allows pages with a positive refcount to be put on standby

#

and then i do some special handling to make sure that if a standby page is taken off standby, that the referencers will know to search for the page in swap

flint idol Jan 11, 2026, 12:05 AM

#

fucking

#

great

#

my pc's wifi adapter no longer exists

#

wtf

#

like, the device doesnt exist on the pci bus

#

???

#

it reappeared

leaden carbon Jan 11, 2026, 12:38 AM

#

https://tenor.com/view/homer-simpson-appears-unhide-peek-gif-16884583

Tenor

flint idol Jan 11, 2026, 12:39 AM

#

fack

#

the memory leak bug is back

#

again

#

or wait

#

it's not a leak

#

it's just a bug with the memory stats

flint idol Jan 11, 2026, 1:32 AM

#

ok i think that's fixed

flint idol Jan 11, 2026, 1:51 AM

#

i just implemented virtual page locking

#

no clue if it works, all i know is that it doesnt crash anything

flint idol Jan 11, 2026, 2:46 AM

#

ext2 is putting me on the edge of my sanity

flint idol Jan 11, 2026, 4:08 AM

#

oopsie

#

i forgot to check permissions in Sys_OpenDir

#

nvm i didnt forget

#

fixed le bug

#

oh yeah

#

i implemented this coolish capability thing for obos a few months ago

#

that uses the filesystem to get permissions for certain syscalls

#

e.g., shutdown

#

it's pretty simple

#

essentailly, it checks if accessing the capability file would not fail, and if that is the case, then it allows it

#

if the file doesnt exist it defaults to root UID allow, root GID allow, and other disallow (unless otherwise specified by the syscall)

#

and every file is under /sys/perm

#

so if one wanted to disallow the net/tcp cap for all non root users (idk why), they would make /sys/perm/net/tcp, and chmod that to 110

#

it also works to disable a capability entirely

#

by making the mode zero

#

i am yet to actually document what capabilities actually exist, though

#

(ill do it eventually, right?)

flint idol Jan 12, 2026, 12:13 AM

#

i do wonder what will happen to the kernel if i try to read /dev/power_button

#

also todo: fix uniprocessor obos on real hardware

#

also todo: fix ahci driver on real hardware

flint idol Jan 12, 2026, 12:16 AM

#

flint idol i do wonder what will happen to the kernel if i try to read /dev/power_button

it hangs

#

wut

#

only on real hardware

#

it has nothing to do with /dev/power_button itself, but instead with obos' stability (or rather, its lack thereof)

real pecan Jan 12, 2026, 12:19 AM

#

Lol

flint idol Jan 12, 2026, 12:20 AM

#

oh yeah an http server works on obos

vale nymph Jan 12, 2026, 12:22 AM

#

flint idol oh yeah an http server works on obos

PUT /dev/sda HTTP/1.1
Host: obos-dev.ddns.net
Content-type: text/html
Content-length: 15

<p>trolled</p>

flint idol Jan 12, 2026, 12:23 AM

#

hold the fuck on do i have PUT requests disabled in apache

#

FUCK

#

oh wait the root isn't / for apache

#

instead of fucking around with obos i should PROBABLY finish that final project due in two days that i also am presenting in two days...

#

that i was supposed to spend all day working on instead of bed rotting and procrastinating

#

also i do gotta say that obos' tcp performance is surprisingly not bad as long as obos is the server

#

ok wish me luck yall im gonna go make the presentation

#

born to use .txt files, forced to use ms powerpoint nooo

ornate ginkgo Jan 12, 2026, 1:11 AM

#

flint idol instead of fucking around with obos i should PROBABLY finish that final project ...

Just present obos instead, smh

flint idol Jan 12, 2026, 1:11 AM

#

if only

#

i mean i dont technically have to do it, as long as i show up to the exam im passing with at least 70

#

but im not tryna get a 70

#

from a ~90

#

yknow

ornate ginkgo Jan 12, 2026, 1:13 AM

#

Yeah ofc, get the most of out of your education.

flint idol Jan 12, 2026, 1:13 AM

#

im also pretty much done the majority of everything i just need do the presentation

#

which im doing rn

flint idol Jan 12, 2026, 2:31 PM

#

TEST

#

ok the vpn works

flint idol Jan 12, 2026, 3:20 PM

#

how does this thread have 51 stars

#

a year ago I had like absolutely nothing to show for myself

#

(dont unstar)

#

((thx))

frigid tide Jan 12, 2026, 8:44 PM

#

flint idol how does this thread have 51 stars

52 now 😝

flint idol Jan 12, 2026, 8:47 PM

#

I might want to port libpcap to obos

#

for packet capturing purposes

#

I think it'd be useful at the very least for debugging the loopback interface

flint idol Jan 12, 2026, 11:01 PM

#

wow 53 stars guys im flattered

#

obos work in two days

#

after I am done my final projects for my courses

#

(one due tomorrow, the other due after tomorrow, the other is for music and is simply a playing test i am doing tomorrow, for the curious)

#

the other course is math and my final mark is just the final exam as well as the standardized test we do in Ontario for people my age

flint idol Jan 13, 2026, 1:37 PM

#

Just presented my shit

#

I think it was fine

flint idol Jan 13, 2026, 2:02 PM

#

I think i will be fixing the dogshit which is the obos timer interface

#

specifically the backend

#

currently it uses the shit timer known as the HPET

#

I will be migrating it to use the invariant tsc

#

and fallback to hpet if that does not exist

#OBOS (not vibecoded)