Modernize architectures and generate seccomp policy for all architectures supported by host OS #26

basilgello · 2020-10-12T07:48:13Z

Fixes #19

basilgello · 2020-10-12T07:49:26Z

@happyCoder92 @robertswiecki please review it and merge! This fixes a CVE used in many CTFs and also allows using nsjail as a Linux container replacement in fully rootless mode.

happyCoder92 · 2020-10-12T10:16:00Z

tools/gen_syscalls/extract.py

@@ -19,6 +19,15 @@

 import re

+# "xrange" was renamed to "range" in Python 3


Just use range instead.
It will work also in python2 but will be less efficient (does not matter here though)

Ack, thanks!

happyCoder92 · 2020-10-12T10:34:55Z

src/syscall.c

@@ -52,15 +52,18 @@ const struct syscall_list syscall_lists[] = {
 #endif
 #ifdef AUDIT_ARCH_AARCH64
    SYSCALL_LIST(AUDIT_ARCH_AARCH64, aarch64),
+    SYSCALL_LIST(AUDIT_ARCH_ARM, arm),


will result in duplicate entries
perhaps just verify that defined(AUDIT_ARCH_AARCH64) implies defined(AUDIT_ARCH_ARM) etc. else #error

From what I see, AArch64 syscall list is not duplicating arm one because the extract.py script dumps syscall table and there is separate syscall table for every arch within kernel. This also applies to amd64/i386/x32, I had to expose SYSCALLTABLENAME environment variable to select the table I want to dump.

I've meant duplicated entries in syscall_lists array.
Note above there is:

#ifdef AUDIT_ARCH_ARM SYSCALL_LIST(AUDIT_ARCH_ARM, arm), #endif

I am not sure AUDIT_ARCH_AARCH64 implies AUDIT_ARCH_ARM: https://elixir.bootlin.com/linux/latest/source/include/uapi/linux/audit.h#L389

Or do you want me to do:

#if defined(AUDIT_ARCH_AARCH64) SYSCALL_LIST(AUDIT_ARCH_AARCH64, aarch64), SYSCALL_LIST(AUDIT_ARCH_ARM, arm), #elif defined(AUDIT_ARCH_ARM) SYSCALL_LIST(AUDIT_ARCH_ARM, arm), #elif defined(AUDIT_ARCH_X86_64) … #else #error #endif

here?

I mean if AUDIT_ARCH_ARM is not defined while AUDIT_ARCH_AARCH64 is defined then the above will not compile anyhow.

Uhm, I think I got what I did I do wrong here 👍 My intention was to check if KAFEL_DEFAULT_TARGET_ARCH contains the respective architecture to include what is only needed. Target i386 can not be executed on aarch64 stuff etc. I will checnge to #ifdef __x86-64__ etc here.

happyCoder92 · 2020-10-12T10:38:21Z

include/kafel.h

+ * Sets compilation target architecture for ctxt to target_archs
+ * target_arch must be a supported AUDIT_ARCH_* value (see <linux/audit.h>)
+ */
+void kafel_set_target_architectures(kafel_ctxt_t ctxt, uint32_t* target_archs, uint32_t size);


use int or size_t for size
or better just accept fixed number of KAFEL_MAX_TARGET_ARCHS entries

Makes sense, ack!

happyCoder92 · 2020-10-12T10:39:05Z

src/context.h

@@ -30,6 +30,8 @@
 #include "policy.h"
 #include "syscall.h"

+#define MAX_TARGET_ARCHS 4


prefix with KAFEL_ and expose in include/kafel.h

also makes sense!

happyCoder92 · 2020-10-12T10:41:33Z

src/kafel.c

+  }
+
+  // Read YY_BUF_SIZE from file as string
+  filebuf = calloc(1, YY_BUF_SIZE);


this will not handle files larger than YY_BUF_SIZE

https://github.com/google/kafel/blob/master/src/kafel.c#L46 I was in doubt how many bytes does this buffer have. I deduced that the buffer used internally is exactly YY_BUF_SIZE and followed that assumption. How big buffer do we really need here?

The buffer is YY_BUF_SIZE bytes but it'll refill with more data from the file if needed.

happyCoder92 · 2020-10-12T12:59:48Z

src/syscall.c

@@ -87,6 +87,17 @@ const struct syscall_list* syscalls_lookup(uint32_t arch) {
  return NULL;
 }

+const struct syscall_list* companion_syscalls_lookup(uint32_t arch) {
+  for (size_t i = 0;
+       i < sizeof(companion_syscall_lists) / sizeof(companion_syscall_lists[0]);


This commit is not self-contained.
companion_syscall_lists is undefined.

Ack, will declare companion_syscall_lists there and populate it with x32 using the x32 commit as fixup.

happyCoder92 · 2020-10-12T13:05:23Z

src/context.c

+uint32_t KAFEL_DEFAULT_TARGET_ARCH[4] =
+
+#if defined(__x86_64__)
+  { AUDIT_ARCH_X86_64, AUDIT_ARCH_I386, 0, 0 };


"pay only for what you use".
include just AUDIT_ARCH_X86_64 there to preserve existing behavior (same for other archs).

Include just AUDIT_ARCH_X86_64 there to preserve existing behavior (same for other archs).

And to enable new target architectures in, say, nsjail i will need a new CLI switch invoking kafel_set_target_architectures, right?

Yes, interaction with nsjail will be a bit inconvenient.
I see 2 solutions:

make the generated archs depend on default action (the mentioned bypass applies only to allowlists):

generate all if the action is ALLOW/LOG

otherwise generate just for native arch by default

introduce to kafel a language construct to change options, e.g.

#option ARCH amd64,x32,i386

introduce to kafel a language construct to change option

I think it is the best option! Can you please help me with it as I am still not very familiar with lexer?

happyCoder92 · 2020-10-12T13:10:44Z

src/syscalls/mipso32_syscalls.c

@@ -33,15 +33,15 @@

 const struct syscall_descriptor mipso32_syscall_list[] = {
    {"accept",
-     4168,
+     168,


Seems all syscall numbers here differ by 4000.
Similary for MIPS64 the difference is 5000.
I have no experience with MIPS but I looked up syscall tables online and seems they all list 4xxx and 5xxx syscalls.

This is the most doubtful commit. Do you know by chance which kernel sources were used to build kernels later dumped with extract.py? Debian kernels have syscall numbers changed and I really don't like it. I'd like to dump syscalls following the initial procedure to get the closest possible result.

MIPS support was contributed by another user.
I don't know the procedure used there, but seems the syscall list was compiled partly by hand.

I don't know the procedure used there, but seems the syscall list was compiled partly by hand

If we know whetger the kernel built was vanilla one and toolchains used, I can build all these vanilla kernels using the same toolchains and run gdb extractor. Then making syscall argument names the same - and I hope we get an extended but not repkaced syscall list.

happyCoder92 · 2020-10-12T13:28:30Z

src/codegen.c

@@ -615,15 +626,23 @@ int compile_policy(struct kafel_ctxt *kafel_ctxt, struct sock_fprog *prog) {
  normalize_rules(rules, kafel_ctxt->default_action);


for companion arch just issue another add_policy_rules here (with another kafel_ctxt containing policy parsed using the companion arch).
Companion policy generation should be turned off by default and this should be configurable.

So should expose kafel_set_use_companion_architecture to the public API and use it to control add_policy_rules added, right? And filter companion architectures in nsjail CLI switch, too?

Correct, just perhaps drop the set, kafel_use_companion_architecture(kafel_ctxt_t*, bool) sounds better.
For the nsjail interaction see my other comment.

Ack, will do!

happyCoder92 · 2020-10-12T13:58:09Z

src/codegen.c

@@ -615,15 +626,23 @@ int compile_policy(struct kafel_ctxt *kafel_ctxt, struct sock_fprog *prog) {
  normalize_rules(rules, kafel_ctxt->default_action);
  int begin = CURRENT_LOC;
  int next = generate_rules(ctxt, rules->data, rules->len);
+  int next2 = 0;


This will not work as it will skip the marker and target arch check if the resulting policy is a constant action.
There are cases where a policy is a constant action on one arch and is not a constant action on another arch due to e.g. different argument sizes:

KILL { some_syscall { arg1 > 65535 } } DEFAULT ALLOW

will result in just ALLOW when arg1 is 16-bit and a more complex policy otherwise.

Arch checks should be added after all per arch generation is done.
ctxt should be reused when generating policies for all the archs.

Something like:

struct codegen_ctxt *ctxt = context_create(); for (int i = 0; i < num_target_archs; ++i) { struct kafel_ctxt* target_ctxt = copy_kafel_ctxt(kafel_ctxt); kafel_set_target_arch(target_archs[i]); parse(target_ctxt); // TODO: error handling policy_for_arch[i] = generate_policy_without_arch_check(ctxt, target_ctxt); kafel_ctxt_destroy(target_ctxt); } int begin = CURRENT_LOC; int next = -ACTION_KILL; for (int i = 0; i < num_target_archs; ++i) { next = add_jump(ctxt, BPF_JEQ, target_archs[i], policy_for_arch[i], next); } if (next > begin) { begin = next = ADD_INSTR(BPF_LOAD_ARCH); } if (next < 0) { resolve_location(ctxt, next); }

Yeah, this seems the better approach. It will probably not eliminate the need to cache policy read from file (we simply cannot rewind stdin) but overall the code looks cleaner.

What I really don't understand now is what is wrong with your example.

I tried:

/tmp/kafel$ echo 'KILL { ptrace { addr > 0xFFFFFFFE } } DEFAULT ALLOW' | tools/dump_policy_bpf/dump_policy_bpf BPF program with 32 instructions 0: A := architecture 1: if A == 0xc000003e goto 3 2: jump to 21 3: A := syscall number 4: if A < 0x65 goto 10 5: if A >= 0x66 goto 10 6: A := arg 2 high 7: if A > 0 goto 11 8: A := arg 2 low 9: if A > 0xfffffffe goto 11 10: jump to 12 11: KILL 12: A := syscall number 13: if A < 0x40000209 goto 19 14: if A >= 0x4000020a goto 19 15: A := arg 2 high 16: if A > 0 goto 20 17: A := arg 2 low 18: if A > 0xfffffffe goto 20 19: ALLOW 20: KILL 21: A := architecture 22: if A == 0x40000003 goto 24 23: jump to 31 24: A := syscall number 25: if A < 0x1a goto 29 26: if A >= 0x1b goto 29 27: A := arg 2 low 28: if A > 0xfffffffe goto 30 29: ALLOW 30: KILL 31: KILL

/tmp/kafel$ echo 'KILL { ptrace { addr > 0xFFFFFFFE } } DEFAULT ALLOW' | tools/dump_policy_bpf/dump_policy_bpf -c BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, arch)), BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, 0xc000003eu, 1, 0), BPF_JUMP(BPF_JMP | BPF_JA, 0x12u, 0, 0), BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, nr)), BPF_JUMP(BPF_JMP | BPF_JGE | BPF_K, 0x65u, 0, 5), BPF_JUMP(BPF_JMP | BPF_JGE | BPF_K, 0x66u, 4, 0), BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, args[2]) + sizeof(__u32)), BPF_JUMP(BPF_JMP | BPF_JGT | BPF_K, 0u, 3, 0), BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, args[2])), BPF_JUMP(BPF_JMP | BPF_JGT | BPF_K, 0xfffffffeu, 1, 0), BPF_JUMP(BPF_JMP | BPF_JA, 0x1u, 0, 0), BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_KILL), BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, nr)), BPF_JUMP(BPF_JMP | BPF_JGE | BPF_K, 0x40000209u, 0, 5), BPF_JUMP(BPF_JMP | BPF_JGE | BPF_K, 0x4000020au, 4, 0), BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, args[2]) + sizeof(__u32)), BPF_JUMP(BPF_JMP | BPF_JGT | BPF_K, 0u, 3, 0), BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, args[2])), BPF_JUMP(BPF_JMP | BPF_JGT | BPF_K, 0xfffffffeu, 1, 0), BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_ALLOW), BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_KILL), BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, arch)), BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, 0x40000003u, 1, 0), BPF_JUMP(BPF_JMP | BPF_JA, 0x7u, 0, 0), BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, nr)), BPF_JUMP(BPF_JMP | BPF_JGE | BPF_K, 0x1au, 0, 3), BPF_JUMP(BPF_JMP | BPF_JGE | BPF_K, 0x1bu, 2, 0), BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, args[2])), BPF_JUMP(BPF_JMP | BPF_JGT | BPF_K, 0xfffffffeu, 1, 0), BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_ALLOW), BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_KILL), BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_KILL),

/tmp/kafel$ echo 'KILL { ptrace { addr > 0x1FFFFFFFE } } DEFAULT ALLOW' | tools/dump_policy_bpf/dump_policy_bpf BPF program with 37 instructions 0: A := architecture 1: if A == 0xc000003e goto 3 2: jump to 23 3: A := syscall number 4: if A < 0x65 goto 11 5: if A >= 0x66 goto 11 6: A := arg 2 high 7: if A > 0x1 goto 12 8: if A < 0x1 goto 11 9: A := arg 2 low 10: if A > 0xfffffffe goto 12 11: jump to 13 12: KILL 13: A := syscall number 14: if A < 0x40000209 goto 21 15: if A >= 0x4000020a goto 21 16: A := arg 2 high 17: if A > 0x1 goto 22 18: if A < 0x1 goto 21 19: A := arg 2 low 20: if A > 0xfffffffe goto 22 21: ALLOW 22: KILL 23: A := architecture 24: if A == 0x40000003 goto 26 25: jump to 36 26: A := syscall number 27: if A < 0x1a goto 34 28: if A >= 0x1b goto 34 29: A := arg 2 high 30: if A > 0x1 goto 35 31: if A < 0x1 goto 34 32: A := arg 2 low 33: if A > 0xfffffffe goto 35 34: ALLOW 35: KILL 36: KILL

/tmp/kafel$ echo 'KILL { ptrace { addr > 0x1FFFFFFFE } } DEFAULT ALLOW' | tools/dump_policy_bpf/dump_policy_bpf -c BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, arch)), BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, 0xc000003eu, 1, 0), BPF_JUMP(BPF_JMP | BPF_JA, 0x14u, 0, 0), BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, nr)), BPF_JUMP(BPF_JMP | BPF_JGE | BPF_K, 0x65u, 0, 6), BPF_JUMP(BPF_JMP | BPF_JGE | BPF_K, 0x66u, 5, 0), BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, args[2]) + sizeof(__u32)), BPF_JUMP(BPF_JMP | BPF_JGT | BPF_K, 0x1u, 4, 0), BPF_JUMP(BPF_JMP | BPF_JGE | BPF_K, 0x1u, 0, 2), BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, args[2])), BPF_JUMP(BPF_JMP | BPF_JGT | BPF_K, 0xfffffffeu, 1, 0), BPF_JUMP(BPF_JMP | BPF_JA, 0x1u, 0, 0), BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_KILL), BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, nr)), BPF_JUMP(BPF_JMP | BPF_JGE | BPF_K, 0x40000209u, 0, 6), BPF_JUMP(BPF_JMP | BPF_JGE | BPF_K, 0x4000020au, 5, 0), BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, args[2]) + sizeof(__u32)), BPF_JUMP(BPF_JMP | BPF_JGT | BPF_K, 0x1u, 4, 0), BPF_JUMP(BPF_JMP | BPF_JGE | BPF_K, 0x1u, 0, 2), BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, args[2])), BPF_JUMP(BPF_JMP | BPF_JGT | BPF_K, 0xfffffffeu, 1, 0), BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_ALLOW), BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_KILL), BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, arch)), BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, 0x40000003u, 1, 0), BPF_JUMP(BPF_JMP | BPF_JA, 0xau, 0, 0), BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, nr)), BPF_JUMP(BPF_JMP | BPF_JGE | BPF_K, 0x1au, 0, 6), BPF_JUMP(BPF_JMP | BPF_JGE | BPF_K, 0x1bu, 5, 0), BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, args[2]) + sizeof(__u32)), BPF_JUMP(BPF_JMP | BPF_JGT | BPF_K, 0x1u, 4, 0), BPF_JUMP(BPF_JMP | BPF_JGE | BPF_K, 0x1u, 0, 2), BPF_STMT(BPF_LD | BPF_W | BPF_ABS, offsetof(struct seccomp_data, args[2])), BPF_JUMP(BPF_JMP | BPF_JGT | BPF_K, 0xfffffffeu, 1, 0), BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_ALLOW), BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_KILL), BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_KILL),

kafel_reset should not reset target architectures to default.
Otherwise it is impossible to set different architectures for compilation.

Then after that is fixed if the user sets architectures like this:

uint32_t archs[4] = {AUDIT_ARCH_I386, AUDIT_ARCH_X86_64, 0, 0}; kafel_set_target_architectures(ctxt, archs, 4);

then compilation of KILL { open { filename > 0xFFFFFFFF } } DEFAULT ALLOW' will result in

BPF program with 1 instructions 0: ALLOW

I think I will rewrite this part as you suggested and drop the commits updating syscall lists and the commit introducing file caching. You will add the lexer modifications and the PR will be ready for merge.

I just pushed 46ae6b8
For multi-target_archs it will require changes in validate_references - check that syscall is present in all the syscall tables and validate used arguments for all architectures.

Great! Sorry for the delay, I will rebase the PR and complete what I have started :)

happyCoder92 · 2021-04-06T13:28:44Z

Hi! Are you still going to work on this one?
Seems there are more users requesting it and it would be nice to merge it soon :)

basilgello · 2021-04-06T13:32:57Z

Hi! Yes, I noticed the request! Let me wrap around Android ASB & get back to it this weekend.

happyCoder92 · 2021-12-23T12:26:34Z

Hi! I had a while to work on it and just commited multi-arch support.
It still needs to be integrated with nsjail and perhaps exposed in new language constructs.
Also I did not add x32 syscalls yet :)

mattgodbolt · 2022-12-01T00:44:49Z

Just a note: we're thankful someone's looked into this and it would be a shame if it stalled. Compiler Explorer relies on nsjail and thus kafel and we currently can't enable seccomp stuff without this PR. which would be nice!

basilgello · 2022-12-01T06:48:30Z

oh! OK @mattgodbolt I will resurrect this.

* Use LANG=C everywhere to make readelf happy on non-English systems Signed-off-by: Vasyl Gello <vasek.gello@gmail.com>

Make the extractor configurable to extract {amd64,i386,x32} from amd64 and other companion architecture syscalls if needed. Also introduce "GDB" environment variable allowing invocation of custom GDB, i.e: GDB=gdb-multiarch ./gen_syscalls.sh /tmp/vmlinux-armmp Signed-off-by: Vasyl Gello <vasek.gello@gmail.com>

happyCoder92 · 2024-08-07T10:05:49Z

tools/gen_syscalls/gen_syscalls.sh


-cat output_syscalls.c | sort -k1,1 --unique --stable -t',' >> "$outname"
-rm output_syscalls.c
+	if [ -f "missing/${arch,,}.c" ]; then


would prefer that missing/{arch}.c takes precedence over automatically extracted as it was before

happyCoder92 requested changes Oct 12, 2020

View reviewed changes

basilgello requested a review from happyCoder92 October 12, 2020 20:02

basilgello mentioned this pull request Aug 3, 2023

Arm Support uazo/cromite#146

Closed

4 tasks

gen_syscalls: Modernize scripts

a5cc253

* Use LANG=C everywhere to make readelf happy on non-English systems Signed-off-by: Vasyl Gello <vasek.gello@gmail.com>

basilgello force-pushed the modernize-architectures branch from 5b58e5d to 0ffeda5 Compare July 25, 2024 18:45

basilgello force-pushed the modernize-architectures branch from 0ffeda5 to d40625d Compare July 25, 2024 20:05

happyCoder92 requested changes Aug 7, 2024

View reviewed changes

		@@ -19,6 +19,15 @@

		import re

		# "xrange" was renamed to "range" in Python 3

		@@ -615,15 +626,23 @@ int compile_policy(struct kafel_ctxt kafel_ctxt, struct sock_fprog prog) {
		normalize_rules(rules, kafel_ctxt->default_action);

Modernize architectures and generate seccomp policy for all architectures supported by host OS #26

Are you sure you want to change the base?

Modernize architectures and generate seccomp policy for all architectures supported by host OS #26

Conversation

basilgello commented Oct 12, 2020

basilgello commented Oct 12, 2020

Choose a reason for hiding this comment

basilgello Oct 12, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

basilgello Oct 12, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

happyCoder92 Oct 12, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

basilgello Oct 12, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

happyCoder92 Oct 12, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

happyCoder92 commented Apr 6, 2021

basilgello commented Apr 6, 2021 via email

happyCoder92 commented Dec 23, 2021

mattgodbolt commented Dec 1, 2022

basilgello commented Dec 1, 2022

Choose a reason for hiding this comment

basilgello Oct 12, 2020 •

edited

Loading

basilgello Oct 12, 2020 •

edited

Loading

happyCoder92 Oct 12, 2020 •

edited

Loading

basilgello Oct 12, 2020 •

edited

Loading

happyCoder92 Oct 12, 2020 •

edited

Loading