[Gabrielcarvfer]

VLC, NorthStar web browser, GTK4 patches to DCE.
Plus X11 and PulseAudio forwarding, so most GUI applications can run with audio too.

Requires DCE patch #150
Requires NorthStar patch nordstjernen-web/northstar-web-browser#17
Even better with extra streaming patches for NorthStar nordstjernen-web/northstar-web-browser#18

[thehajime]

@cubic-dev-ai ultrareview

[cubic-dev-ai]

@cubic-dev-ai ultrareview

@thehajime cubic can't start this ultrareview because you've reached your trial's review limit. Trial plans have lower review limits than paid plans. Upgrade now to resume reviews.

To help optimise your usage, you can tune cubic to get the most out of your usage limits:

• Stop reviewing test files
• Turn off incremental reviews
• Use the cubic CLI locally
• Upgrade to a pro plan

Learn more →

[Gabrielcarvfer]

Now also works with Chrome (sandbox disabled)

[cubic-dev-ai]

60 issues found across 194 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="model/kernel-socket-fd-factory.cc">

<violation number="1" location="model/kernel-socket-fd-factory.cc:486">
P0: `lkl_netdev_add` runs after `lkl_start_kernel`, but LKL requires every host netdev to be added before the kernel starts. Move netdev registration before `m_kernel->Boot()`, then perform the post-boot ioctl configuration separately.</violation>
</file>

<file name="model/utils.cc">

<violation number="1" location="model/utils.cc:241">
P1: Default signal termination calls `dce_exit()`, so it runs exit handlers, flushes stdio, and reports a normal exit code. Use the signal-specific termination path.</violation>
</file>

<file name="model/dce-signal.cc">

<violation number="1" location="model/dce-signal.cc:202">
P1: `raise()` currently kills the process with SIGKILL for every signal because `dce_kill()` hard-codes SIGKILL. Deliver the requested signal to the calling thread instead.</violation>
</file>

<file name="model/local-datagram-socket-fd.cc">

<violation number="1" location="model/local-datagram-socket-fd.cc:475">
P1: This socketpair implementation makes AF_UNIX/SOCK_DGRAM pairs unreadable: both ends are CONNECTED, but datagram reads only accept BINDED and Recvmsg rejects CONNECTED. Update the receive paths or pair state handling.</violation>
</file>

<file name="model/dce-futex.cc">

<violation number="1" location="model/dce-futex.cc:121">
P1: A thread moved by FUTEX_REQUEUE that later times out is still removed from the wrong list. `RemoveWaiter(uaddr, current)` searches only the original address, so the timeout is reported as a wake-up and the stale entry stays in the `uaddr2` queue. Track the address each waiter is currently queued on, and update it in `FutexRequeue`, so `FutexWait` removes the entry from the correct list.</violation>

<violation number="2" location="model/dce-futex.cc:182">
P2: `FUTEX_WAKE_BITSET` currently wakes all waiters on the address and ignores `val3`, allowing unrelated bitset waits to return. Store each waiter's bitset and filter wakeups by the requested mask.</violation>

<violation number="3" location="model/dce-futex.cc:216">
P1: `dce_syscall` reads six variadic arguments even for syscalls that pass fewer or none, so ordinary calls such as `syscall(SYS_getpid)` have undefined behavior. Decode only the arguments required by each syscall before dispatching.</violation>
</file>

<file name="model/dce-fd.cc">

<violation number="1" location="model/dce-fd.cc:425">
P2: `SOCK_CLOEXEC` is discarded here but never applied to the DCE descriptor. Preserve the flag and set `FD_CLOEXEC`, otherwise callers silently lose close-on-exec semantics.</violation>

<violation number="2" location="model/dce-fd.cc:561">
P2: This parser accepts an oversized AF_UNIX length and can read past `sun_path`. Reject lengths greater than `sizeof(struct sockaddr_un)` before deriving `maxLen`.</violation>

<violation number="3" location="model/dce-fd.cc:842">
P2: A negative `F_DUPFD` lower bound reaches `dce_dup2` and creates descriptor `-1`. Reject negative bounds with `EINVAL` before searching for a free descriptor.</violation>

<violation number="4" location="model/dce-fd.cc:1039">
P2: `dce_fcntl64` unconditionally reads a variadic argument, but valid `F_GETFD`/`F_GETFL` calls provide none. Handle no-argument commands without calling `va_arg`.</violation>

<violation number="5" location="model/dce-fd.cc:1065">
P2: `pipe2` accepts `O_CLOEXEC` but never sets `FD_CLOEXEC` on either pipe end. Apply the descriptor flag after creating both ends.</violation>

<violation number="6" location="model/dce-fd.cc:1088">
P1: `__readlink_chk` ignores `buflen`, allowing `dce_readlink` to write beyond a fortified caller's buffer when `bufsize > buflen`. Reject that case before forwarding the request.</violation>

<violation number="7" location="model/dce-fd.cc:1160">
P1: `__realpath_chk` drops `resolvedlen` before `dce_realpath` uses `strcpy`, so a short destination can overflow. Check the destination capacity before copying, or pass the bound through the realpath implementation.</violation>
</file>

<file name="model/dce-alloc.cc">

<violation number="1" location="model/dce-alloc.cc:33">
P1: `Allocate` can wrap the requested size before passing it to the `uint32_t` Kingsley allocator, returning an undersized block for large requests. Reject arithmetic overflow and totals above `UINT32_MAX` before the addition.</violation>

<violation number="2" location="model/dce-alloc.cc:187">
P2: `dce_posix_memalign` accepts powers of two smaller than `sizeof(void*)`, although POSIX requires `EINVAL` for those values. Validate the pointer-size multiple while keeping `dce_memalign`'s separate alignment contract.</violation>
</file>

<file name="model/libc-ns3.h">

<violation number="1" location="model/libc-ns3.h:745">
P1: These newly exported fortified wrappers bypass their destination-size checks. Validate each object-size argument before delegating, rather than exposing the unchecked compatibility implementations.</violation>
</file>

<file name="model/dce-stdio.cc">

<violation number="1" location="model/dce-stdio.cc:174">
P2: Successful exec leaves stale entries in `streamFds`, so a reused `FILE *` can return the wrong descriptor from `dce_fileno`. Erase these entries when exec closes the old streams or clear the map before creating the new image.</violation>

<violation number="2" location="model/dce-stdio.cc:807">
P1: `dce___fgets_chk` ignores the destination size and can overflow fortified callers. Reject `n > size` before calling `dce_fgets`.</violation>

<violation number="3" location="model/dce-stdio.cc:811">
P1: `dce___fread_chk` ignores `ptrlen` and can overflow fortified callers. Reject element counts whose total size exceeds `ptrlen`.</violation>
</file>

<file name="model/dce-manager.cc">

<violation number="1" location="model/dce-manager.cc:705">
P1: `Clone` returns `uint16_t`, so this `return -1` becomes 65535. `dce_fork` passes that to the guest as a pid, so `fork()` never reports -1 and the `ENOSYS` in `thread->err` is never seen. Change the return type to `int` (in dce-manager.h too) so the error can be returned.</violation>

<violation number="2" location="model/dce-manager.cc:1438">
P1: `execve` loads the new executable's allocator and TLS image into `pTemp`, but the successful replacement never transfers either field to `process`; subsequent allocations and newly-created threads therefore retain the old executable state. Copy both fields during the `Execve` state replacement and release the previous allocator.</violation>
</file>

<file name="model/epoll-fd.cc">

<violation number="1" location="model/epoll-fd.cc:147">
P1: Polling an epoll descriptor consumes `EPOLLONESHOT` entries because `EpollFd::Poll` reuses `Scan`, which disables ready entries. Keep readiness probing separate from `epoll_wait` event consumption so `poll(epfd, ...)` cannot lose one-shot events.</violation>
</file>

<file name="model/local-socket-fd.cc">

<violation number="1" location="model/local-socket-fd.cc:259">
P2: The new oversized-message path sets EMSGSIZE, but Write() overwrites it with ENOMEM. Preserve the errno from DoRecvPacket() so oversized SOCK_SEQPACKET sends report EMSGSIZE.</violation>

<violation number="2" location="model/local-socket-fd.cc:348">
P1: Plain reads consume ancillary-data offsets without discarding their SCM_RIGHTS entries. A later recvmsg can deliver an old descriptor with unrelated data; remove ancillary records when their associated bytes are consumed by any read.</violation>

<violation number="3" location="model/local-socket-fd.cc:352">
P2: Short SOCK_SEQPACKET receives discard the rest of the record but never report MSG_TRUNC. Propagate the truncation result from ReadData() and set msg->msg_flags before returning.</violation>
</file>

<file name="model/local-stream-socket-fd.cc">

<violation number="1" location="model/local-stream-socket-fd.cc:388">
P2: MSG_PEEK bypasses the socket's blocking and nonblocking receive logic. An empty blocking peek returns EOF-like 0 instead of waiting; route peeks through a receive helper that preserves those semantics.</violation>

<violation number="2" location="model/local-stream-socket-fd.cc:489">
P1: SCM_RIGHTS parsing accepts a malformed cmsg_len and can underflow before reading fd values. Validate each header's length and bounds before subtracting CMSG_LEN(0) or accessing CMSG_DATA.</violation>

<violation number="3" location="model/local-stream-socket-fd.cc:526">
P1: Sendmsg queues SCM_RIGHTS before knowing that any payload was delivered. Roll back the queued entry and its references whenever Write() fails without sending data.</violation>
</file>

<file name="model/dce-compat.cc">

<violation number="1" location="model/dce-compat.cc:57">
P2: `dce_accept4` drops `SOCK_CLOEXEC`, so accepted descriptors leak across simulated exec. Set `FD_CLOEXEC` when that flag is requested and validate unsupported flags.</violation>

<violation number="2" location="model/dce-compat.cc:341">
P2: `dce___getcwd_chk` ignores `buflen`, so a request larger than the destination object bypasses the fortified bounds check. Validate `size <= buflen` before calling `dce_getcwd`.</violation>

<violation number="3" location="model/dce-compat.cc:345">
P2: `dce___fgets_unlocked_chk` ignores the destination size, so `n > size` can write past the fortified buffer. Enforce the object-size check before calling `dce_fgets`.</violation>

<violation number="4" location="model/dce-compat.cc:877">
P1: A simulated application can change the simulator's host-wide resource limits through `dce_setrlimit`, affecting other DCE applications and the simulator. Keep limits per simulated process or reject the setter instead of calling the host `setrlimit`.</violation>
</file>

<file name="model/libc.cc">

<violation number="1" location="model/libc.cc:38">
P1: The x86 forwarding path puts extended asm with a memory operand inside a naked function, which GCC does not support reliably. Use a basic-asm/out-of-line assembly trampoline or retain the supported forwarding implementation.</violation>
</file>

<file name="test/test-eventfd.cc">

<violation number="1" location="test/test-eventfd.cc:100">
P2: This infinite wait can hang the entire test suite when the wake-up path regresses. Use a finite timeout (the following check already uses 5 seconds) while still asserting that the writer wakes the poll.</violation>
</file>

<file name="model/dce-pthread.cc">

<violation number="1" location="model/dce-pthread.cc:461">
P2: A terminated joinable thread remains searchable with `task == 0`, but this line passes that null task to `GetStackBottom()` and can crash the simulator. Guard `target->task` before querying its stack.</violation>
</file>

<file name="model/dce-dl.cc">

<violation number="1" location="model/dce-dl.cc:55">
P2: The fallback error path dereferences `filename` even when `dlopen(NULL, ...)` selected `exeFullPath`. Use the selected load name (`filename != 0 ? filename : exeFullPath.c_str()`) when constructing the message.</violation>

<violation number="2" location="model/dce-dl.cc:81">
P2: `dlerror()` stores its pending error on `Process`, so DCE threads can consume one another’s lookup errors. Store the error string and pending flag per `Thread` instead.</violation>
</file>

<file name="build-and-run.sh">

<violation number="1" location="build-and-run.sh:8">
P2: The script only works when launched from the repository root because it uses the caller's working directory as `DCE_DIR`. Derive `DCE_DIR` from the script location so `/path/to/build-and-run.sh build` works from any directory.</violation>

<violation number="2" location="build-and-run.sh:85">
P2: The `env` command is re-entered through distrobox, and this banner goes to stdout. That output is meant to be sourced, so the banner breaks `eval "$(./build-and-run.sh env)"`. Send the banner to stderr.</violation>

<violation number="3" location="build-and-run.sh:126">
P2: `setup-container` omits `unzip`, but the Chrome demo executes `unzip` to extract its download. Add `unzip` to the container package list.</violation>
</file>

<file name="utils/dce-fibers-gdb.py">

<violation number="1" location="utils/dce-fibers-gdb.py:21">
P1: This saves the registers of the currently selected frame, not the innermost frame. If the user ran `up` before `dce-fibers`, the restore at the end writes the caller's rsp/rip/rbp into the real registers. Use `gdb.newest_frame()` so the saved values are the real ones.</violation>
</file>

<file name="model/dce.cc">

<violation number="1" location="model/dce.cc:465">
P2: `ResetGetoptIfOtherProcess` leaves glibc's hidden getopt state untouched when a process resumes with `optind > 1`, so it can inherit the previous process's cluster or permutation state. Preserve the complete parser state per process or use a reentrant/per-process getopt implementation instead of resetting only at the start of argv.</violation>

<violation number="2" location="model/dce.cc:1014">
P2: The new `/proc/self/exe` handling reads `exePath`, but that field is not propagated on exec or fork. After `execv`, `Execve` never copies `pTemp.exePath` into `process->exePath`, so readlink returns the old program's path. A forked child also starts with an empty `exePath` because `Clone` does not copy it, so the branch is skipped. Set `process->exePath = pTemp.exePath;` in `Execve` and copy `exePath` in `Clone`.</violation>
</file>

<file name="model/local-socket-fd-factory.cc">

<violation number="1" location="model/local-socket-fd-factory.cc:77">
P2: The new SOCK_SEQPACKET sockets report SOCK_STREAM from SO_TYPE. Preserve the requested socket type in LocalStreamSocketFd and return SOCK_SEQPACKET for these descriptors.</violation>

<violation number="2" location="model/local-socket-fd-factory.cc:80">
P2: Enabling SOCK_SEQPACKET exposes a zero-length-record deadlock: a zero-sized front buffer is never consumed, blocking later messages. Handle zero-length records explicitly in enqueue and receive logic.</violation>

<violation number="3" location="model/local-socket-fd-factory.cc:80">
P2: The SOCK_SEQPACKET socket only keeps message boundaries on the socket created by `socket()`. `LocalStreamSocketFd::Accept` builds the accepted socket with the peer constructor, which never copies `m_messages`, so connections accepted from a SEQPACKET listener behave as plain streams. Propagate the message-boundary flag to the accepted socket in `Accept`, for example by setting `socket->KeepMessageBoundaries ()` when the listener has it set.</violation>
</file>

<file name="model/unix-datagram-socket-fd.cc">

<violation number="1" location="model/unix-datagram-socket-fd.cc:204">
P2: `MSG_PEEK` bypasses `ICMP6_FILTER`, so callers can observe a blocked ICMPv6 type with `recvmsg`. Apply the filter before exposing peeked packets without consuming the accepted packet.</violation>
</file>

<file name="utils/dce-tap-setup.sh">

<violation number="1" location="utils/dce-tap-setup.sh:79">
P2: `down` unconditionally removes masquerading from firewalld’s default zone, including a rule that existed before this script ran. Track whether `up` added it or isolate the demo in its own firewalld zone.</violation>
</file>

<file name="model/ucontext-fiber-manager.cc">

<violation number="1" location="model/ucontext-fiber-manager.cc:198">
P2: Each newly allocated fiber TLS permanently consumes a detached host thread and stack because `ParkedThread` cannot exit and `Delete` never joins it. Reclaim or explicitly cap these helper threads so simulations with many concurrent tasks cannot exhaust host thread or memory limits.</violation>
</file>

<file name="model/dce-epoll.cc">

<violation number="1" location="model/dce-epoll.cc:103">
P2: `dce_epoll_pwait` ignores its temporary signal mask, and `dce_epoll_pwait2` does the same. Apply the mask for the duration of the wait and restore the previous mask on every return path.</violation>
</file>

<file name="doc/source/getting-started.rst">

<violation number="1" location="doc/source/getting-started.rst:233">
P2: Copying `liblkl.so` into `build/bin_dce` is not enough: LKL is located only through the `DCE_PATH` environment variable. Export `DCE_PATH` to include `build/bin_dce` after the copy, as the bake sections already do.</violation>
</file>

<file name="example/numbers-server.cc">

<violation number="1" location="example/numbers-server.cc:94">
P2: The browser polls `/next` twice per second, but this 64-bit Fibonacci state wraps after about 94 requests. Use arbitrary-precision values or explicitly reset/label the series before the default 600-second demo stops.</violation>
</file>

<file name="model/cooja-loader-factory.cc">

<violation number="1" location="model/cooja-loader-factory.cc:385">
P2: `RTLD_NEXT` is not equivalent to `RTLD_DEFAULT`; this branch searches `m_modules` from the beginning for both. Preserve next-object lookup semantics instead of returning the first matching DCE symbol.</violation>
</file>

<file name="example/dce-wifi-video.cc">

<violation number="1" location="example/dce-wifi-video.cc:557">
P2: `waitpid (playerPid, ...)` blocks indefinitely if the player never gets a writer. The player's shell sits in `open()` on the FIFO when ffmpeg exits before it opens the output. Stop the player with a signal after the simulation ends, or poll with `WNOHANG` and a deadline.</violation>
</file>

<file name="utils/build_lkl.sh">

<violation number="1" location="utils/build_lkl.sh:74">
P2: `rm -rf "${OUT_DIR}/include"` will delete the kernel source's top-level `include/` if `out_dir` equals the LKL tree. Reject that case before the removal.</violation>
</file>

<file name="model/host-socket-fd.cc">

<violation number="1" location="model/host-socket-fd.cc:160">
P2: `CheckReadiness` reschedules itself every millisecond with no exit other than `m_closed`. A socket that is polled and never closed keeps the event queue non-empty, so the simulation cannot end on its own. Consider rescheduling only while a poller is registered, or stopping the check when no waiters remain.</violation>
</file>

<file name="model/unix-socket-fd.cc">

<violation number="1" location="model/unix-socket-fd.cc:309">
P2: `SO_SNDBUF` now succeeds even when the socket has no `SndBufSize` attribute, but `getsockopt(SO_SNDBUF)` still returns `EINVAL` and the requested value is lost. Store a per-socket value or implement matching get/set behavior.</violation>

<violation number="2" location="model/unix-socket-fd.cc:561">
P2: `getsockopt(IPPROTO_ICMPV6, ICMP6_FILTER)` returns success on non-raw sockets, while `setsockopt` rejects those sockets. Add the same raw-socket validation to the get path.</violation>
</file>

<file name="model/epoll-fd.h">

<violation number="1" location="model/epoll-fd.h:73">
P2: Registrations are keyed by fd number and only purged lazily in Scan(), so closing an fd without EPOLL_CTL_DEL and reusing its number leaves a stale entry. Re-adding that number then returns EEXIST, and a wait reports the old user data for the new file. Key entries by the underlying open file (not the fd number), or purge them when the last reference to the file is closed.</violation>
</file>

<sub>Reply with feedback, questions, or to request a fix.

View guided diff | Re-trigger cubic</sub>

[cubic-dev-ai]

P0: lkl_netdev_add runs after lkl_start_kernel, but LKL requires every host netdev to be added before the kernel starts. Move netdev registration before m_kernel->Boot(), then perform the post-boot ioctl configuration separately.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At model/kernel-socket-fd-factory.cc, line 486:

<comment>`lkl_netdev_add` runs after `lkl_start_kernel`, but LKL requires every host netdev to be added before the kernel starts. Move netdev registration before `m_kernel->Boot()`, then perform the post-boot ioctl configuration separately.</comment>

<file context>
@@ -1,964 +1,653 @@
+
+  typedef int (*NetdevAdd)(struct lkl_netdev *, struct lkl_netdev_args *);
+  NetdevAdd netdevAdd = (NetdevAdd) m_kernel->Lookup ("lkl_netdev_add");
+  int id = netdevAdd (&dev->nd, &args);
+  NS_ASSERT_MSG (id >= 0, "lkl_netdev_add failed: " << id);
+
</file context>

[cubic-dev-ai]

P1: Default signal termination calls dce_exit(), so it runs exit handlers, flushes stdio, and reports a normal exit code. Use the signal-specific termination path.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At model/utils.cc, line 241:

<comment>Default signal termination calls `dce_exit()`, so it runs exit handlers, flushes stdio, and reports a normal exit code. Use the signal-specific termination path.</comment>

<file context>
@@ -214,6 +220,42 @@ UtilsSendSignal (Process *process, int signum)
+        default:
+          // the default action of the others ends the process
+          NS_LOG_DEBUG ("signal " << handler.signal << " ends the process");
+          dce_exit (128 + handler.signal);
+          return;
+        }
</file context>

Suggested change

	dce_exit (128 + handler.signal);
	dce_exit_signal (handler.signal);

[cubic-dev-ai]

P1: raise() currently kills the process with SIGKILL for every signal because dce_kill() hard-codes SIGKILL. Deliver the requested signal to the calling thread instead.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At model/dce-signal.cc, line 202:

<comment>`raise()` currently kills the process with SIGKILL for every signal because `dce_kill()` hard-codes SIGKILL. Deliver the requested signal to the calling thread instead.</comment>

<file context>
@@ -165,3 +186,25 @@ int dce_sigprocmask (int how, const sigset_t *set, sigset_t *oldset)
+{
+  Thread *current = Current ();
+  NS_ASSERT (current != 0);
+  return dce_kill (current->process->pid, sig);
+}
+
</file context>

Suggested change

	return dce_kill (current->process->pid, sig);
	sigaddset (&current->pendingSignals, sig);
	UtilsDoSignal ();
	return 0;

[cubic-dev-ai]

P1: This socketpair implementation makes AF_UNIX/SOCK_DGRAM pairs unreadable: both ends are CONNECTED, but datagram reads only accept BINDED and Recvmsg rejects CONNECTED. Update the receive paths or pair state handling.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At model/local-datagram-socket-fd.cc, line 475:

<comment>This socketpair implementation makes AF_UNIX/SOCK_DGRAM pairs unreadable: both ends are CONNECTED, but datagram reads only accept BINDED and Recvmsg rejects CONNECTED. Update the receive paths or pair state handling.</comment>

<file context>
@@ -469,6 +469,17 @@ LocalDatagramSocketFd::Bind (const struct sockaddr *my_addr, socklen_t addrlen)
+void
+LocalDatagramSocketFd::Pair (LocalDatagramSocketFd *a, LocalDatagramSocketFd *b)
+{
+  a->m_state = CONNECTED;
+  a->m_peer = b;
+  b->AddConnected (a);
</file context>

[cubic-dev-ai]

P1: dce_syscall reads six variadic arguments even for syscalls that pass fewer or none, so ordinary calls such as syscall(SYS_getpid) have undefined behavior. Decode only the arguments required by each syscall before dispatching.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At model/dce-futex.cc, line 216:

<comment>`dce_syscall` reads six variadic arguments even for syscalls that pass fewer or none, so ordinary calls such as `syscall(SYS_getpid)` have undefined behavior. Decode only the arguments required by each syscall before dispatching.</comment>

<file context>
@@ -0,0 +1,246 @@
+  va_list ap;
+  va_start (ap, number);
+  long a[6];
+  for (int i = 0; i < 6; i++)
+    {
+      a[i] = va_arg (ap, long);
</file context>

[cubic-dev-ai]

P2: rm -rf "${OUT_DIR}/include" will delete the kernel source's top-level include/ if out_dir equals the LKL tree. Reject that case before the removal.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At utils/build_lkl.sh, line 74:

<comment>`rm -rf "${OUT_DIR}/include"` will delete the kernel source's top-level `include/` if `out_dir` equals the LKL tree. Reject that case before the removal.</comment>

<file context>
@@ -0,0 +1,78 @@
+)
+
+# The headers DCE is built with (some are generated by the build).
+rm -rf "${OUT_DIR}/include"
+cp -a "${LKL_DIR}/tools/lkl/include" "${OUT_DIR}/include"
+
</file context>

Suggested change

	rm -rf "${OUT_DIR}/include"
	if [ "${OUT_DIR}" = "${LKL_DIR}" ]; then
	echo "out_dir must not be the LKL source tree" >&2
	exit 1
	fi
	rm -rf "${OUT_DIR}/include"

[cubic-dev-ai]

P2: CheckReadiness reschedules itself every millisecond with no exit other than m_closed. A socket that is polled and never closed keeps the event queue non-empty, so the simulation cannot end on its own. Consider rescheduling only while a poller is registered, or stopping the check when no waiters remain.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At model/host-socket-fd.cc, line 160:

<comment>`CheckReadiness` reschedules itself every millisecond with no exit other than `m_closed`. A socket that is polled and never closed keeps the event queue non-empty, so the simulation cannot end on its own. Consider rescheduling only while a poller is registered, or stopping the check when no waiters remain.</comment>

<file context>
@@ -0,0 +1,378 @@
+    }
+  // The pollers cannot be observed from here, so keep checking until the
+  // socket is closed: one event per millisecond.
+  m_check = Simulator::Schedule (MilliSeconds (1), &HostSocketFd::CheckReadiness, this);
+}
+
</file context>

[cubic-dev-ai]

P2: getsockopt(IPPROTO_ICMPV6, ICMP6_FILTER) returns success on non-raw sockets, while setsockopt rejects those sockets. Add the same raw-socket validation to the get path.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At model/unix-socket-fd.cc, line 561:

<comment>`getsockopt(IPPROTO_ICMPV6, ICMP6_FILTER)` returns success on non-raw sockets, while `setsockopt` rejects those sockets. Add the same raw-socket validation to the get path.</comment>

<file context>
@@ -483,6 +557,22 @@ UnixSocketFd::Getsockopt (int level, int optname,
   switch (level)
     {
+    case IPPROTO_ICMPV6:
+      if (optname != ICMP6_FILTER || *optlen < (socklen_t) sizeof (struct icmp6_filter))
+        {
+          current->err = optname != ICMP6_FILTER ? ENOPROTOOPT : EINVAL;
</file context>

Suggested change

	if (optname != ICMP6_FILTER || *optlen < (socklen_t) sizeof (struct icmp6_filter))
	if (m_socket->GetInstanceTypeId ().GetName () != "ns3::Ipv6RawSocketImpl")
	{
	current->err = ENOPROTOOPT;
	return -1;
	}
	if (optname != ICMP6_FILTER || *optlen < (socklen_t) sizeof (struct icmp6_filter))

[cubic-dev-ai]

P2: SO_SNDBUF now succeeds even when the socket has no SndBufSize attribute, but getsockopt(SO_SNDBUF) still returns EINVAL and the requested value is lost. Store a per-socket value or implement matching get/set behavior.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At model/unix-socket-fd.cc, line 309:

<comment>`SO_SNDBUF` now succeeds even when the socket has no `SndBufSize` attribute, but `getsockopt(SO_SNDBUF)` still returns `EINVAL` and the requested value is lost. Store a per-socket value or implement matching get/set behavior.</comment>

<file context>
@@ -300,11 +304,9 @@ UnixSocketFd::Setsockopt (int level, int optname,
-              }
+            // Only ns-3 TCP sockets have a send buffer attribute; like the
+            // kernel, accept the option on the other socket types.
+            m_socket->SetAttributeFailSafe ("SndBufSize", UintegerValue (*val));
           } break;
         case SO_RCVBUF:
</file context>

[cubic-dev-ai]

P2: Registrations are keyed by fd number and only purged lazily in Scan(), so closing an fd without EPOLL_CTL_DEL and reusing its number leaves a stale entry. Re-adding that number then returns EEXIST, and a wait reports the old user data for the new file. Key entries by the underlying open file (not the fd number), or purge them when the last reference to the file is closed.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At model/epoll-fd.h, line 73:

<comment>Registrations are keyed by fd number and only purged lazily in Scan(), so closing an fd without EPOLL_CTL_DEL and reusing its number leaves a stale entry. Re-adding that number then returns EEXIST, and a wait reports the old user data for the new file. Key entries by the underlying open file (not the fd number), or purge them when the last reference to the file is closed.</comment>

<file context>
@@ -0,0 +1,78 @@
+  int Scan (struct epoll_event *events, int maxevents, PollTable *table,
+            std::map<int, FileUsage *> &referenced);
+
+  std::map<int, Entry> m_entries;
+};
+
</file context>