Add a daemon-only build path (no Python GUI) that builds the eBPF objects
against the running kernel's headers, deriving the BPF DataLayout from
clang and finalizing with opt; a structural unprivileged check target; an
install path that never clobbers existing units/config/rules; an Arch
PKGBUILD.in + prepare.sh and a test_install.sh harness. Document the
workflow and the event-loss diagnostics in README.md and ebpf_prog/README.
Gitignore the compiled ebpf_prog/*.o artifacts.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add fixed-size, process-lifetime counters (received/enqueued/dropped/
drained/dequeued/decoded/decode-errors/reader-errors/canceled) for the
process and DNS eBPF event paths, exposed as individually-atomic
snapshots. Wire DNSEventCounters and ProcessEventCounters through the
perf reader/enqueue/decode loop in daemon/dns and daemon/procmon/ebpf,
with a matching ebpf_prog/event_stats.h and a structural eBPF test.
Counters hold totals only, never event payloads.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
On armhf, i386 and aarch64 in particular, we don't obtain the path and
cmdline of a process reliably. So on these platforms, just send the
notification to userspace for now.
See: #1521
In order to intercept UDP tunnels traffic, we use iptunnel_xmit for IPv4
connections. The equivalent of IPv6 would be ip6tunnel_xmit, but
unfortunately it's not public.
So instead of ip6tunnel_xmit, we'll use udp_tunnel6_xmit_skb.
This function receives two parameters besides the struct sock*, in6_addr
*saddr and in6_addr *daddr. If these addresses are not set, usually the
connection is a localhost connection.
On the other hand, this function is defined in the module
ip6_udp_tunnel. If this module is not loaded when the daemon starts,
we'll fail to intercept the connections.
Closes: #1250
- better process interception on aarch64 and armhf.
- on armhf getting proc arguments is ignored due to an error, but we
were not marking the event as INCOMPLETE_ARGS, so we were not reading
the compete cmdline.
Change some initializations on inet_dgram_connect, to make the verifier
happy. __builtin_memset() in this particular case was causing errors.
(invalid read from stack off -16+4 size 8)
Up until now we've been using https://github.com/iovisor/gobpf/ for
intercepting events from the kernel, and it has worked very well.
Unfortunately it's deprecated since 2021 (iovisor/gobpf#304), which means
that some of the known issues and limitations won't be fixed.
After evaluating some options, I've migrated the code from gobpf to
github.com/cilium/ebpf.
More info: #1222, #1312
- changed formatting.
- extract source IPs from UDP packets from ancillary messages.
- hook inet_dgram_connect to solve intercepting some apps (parsec).
- remove debugging / unused code.
More info on these changes:
fbdef1673d20a03e11fef44d3e896b93d1cefbc2Closes: #1246.
We were sending to userspace unnecessary exit events, consuming
unnecessary CPU cycles.
We only intercept execve and execveat, but sched_process_exit is invoked
by more functions (sched_process_exit, clone, ...), so we were receiving
on the daemon events that we did nothing with them, apart from consuming
CPU cycles.
On some scenarios like on servers running saltstack (as salt-master),
this caused to consume more CPU than needed.
cherry picked from 15fcf67535
On 68c2c8ae1a we excluded failed execve*
calls from being delivered to userspace, in order to get the binary that
was executed and avoid errors/confusion.
But on aarch64, it seems that we fail to save the exec event to a map,
so the event is never delivered to userspace.
So for the time being, send the exec events as soon as they arrive on
aarch64, without checking if the call failed.
On the previus commit we just disabled dns uprobes for armhf/i386 to
avoid loading errors. A better fix is to initialized the structs used.
On armhf still fails after loading it, when attaching to the uprobes
(offsets?), and on i386 it doesn't seem to send anything to userspace
(more analysis needed).
- Increased the number of IPs associated with a domain that are
delivered to userspace. (getfedora.org returns 30 ipv4+ipv6).
- Fixed getting the aliases of a domain when using gethostbyname().
The opensnitch-dns module was not loading on i386/arm architectures.
With the following changes it loads and some uprobes are attached.
for-loops unrolling doesn't still work though on i386/armhf (help
needed).
And on armhf the perf_output channel fails to load for some uprobes.
We track new processes execution by intercepting the enter and exit
of the functions, but sometimes the exit hook is not called, so the
corresponding entry was not being removed from the map.
In this situation the map becomes full and accepts no new entries.
Now the entry is deleted from the map once the process exits, if it
still exists in the map.
We were not deleting DNS entries from the hash map, so when it reached
the maximum capacity (12k entries), we couldn't allocate new entries,
resulting in events not being sent to userspace.
- Don't rename libbpf's bpf_map_def struct, and distribute the needed bpf
headers.
The bpf_map_def struct has been deprecated for quite some time now,
and it was been removed on >= 6.2 anyway.
We still need it, because we use gobpf.
- Improved compilation behaviour:
- We don't require the kernel sources anymore. We can just use the
kernel headers from the distribution.
- There's no need to copy the sources to the kernel tree, the modules
can be compiled from the ebpf_prog/ dir.
- Compiling against kernels 6.x seems to solve the problem we had with
VPNs, where connections were not intercepted with modules compiled
against 5.8, on kernels >= 5.19.
The modules has been tested on kernels 4.17, 5.4, 5.10, 5.15, 6.1 and
6.2 (kernel connections included).
Closes: #939
BPF_MAP_TYPE_PERCPU_ARRAY was introduced in kernel version 4.6, so with
latest changes to intercept processes we lost support for older kernels
< 4.6.
Now we work again for example on kernels 4.4.
Make use of kernel macros to decide for what architectures we compile
the modules.
On armv7l the connections module was failing due to iptunnel_xmit, so
exclude it from being compiled.
One can export ARCH=arm, ARCH=i386 or ARCH=arm64 to compile the modules
for these architectures, instead of hacing to edit the source files.
- Get cmdline arguments from kernel along with the absolute path to the
binary.
If the cmdline has more than 20 arguments, or one of the arguments is
longer than 256 bytes, get it from ProcFS.
- Improved stopping ebpf monitor method.
Improved process detections by monitoring new processes execution.
It allow us to know the path of a process before a socket is opened.
Closes#617
Other improvements:
- If we fail to retrieve the path of a process, then we'll use the comm
name of the connection/process.
- Better kernel connections detection.
- If debugfs is not loaded, we'll try to mount it, to allow to use
eBPF monitor method.
Future work (help wanted):
- Extract command line arguments from the kernel (sys_execve, or mm
struct).
- Monitor other functions (execveat, clone*, fork, etc).
- Send these events to the server (GUI), and display all the commands
an application has executed.
When using DoT or DoH opensnitch cannot intercept the dns packets.
Therefore the UI always shows IP addresses instead of hostnames. To fix
this issue an ebpf (uprobe) filter was created to hook getaddrinfo and
gethostbyname calls.
In order to be independent of libbcc an additional module was added to
ebpf_prog. Without libbcc the libc function offsets must be resolved
manually. In order to find the loaded glibc version some cgo code was
added.
* Allow to intercept some kernel connections
Some connections are initiated from kernel space, like WireGuard
VPNs (#454), NFS or SMB connections (#502) and ip tunnels (#500).
Note: This feature is complete for x86_64, WIP for aarch64, and not supported for armhf and i386
https://github.com/evilsocket/opensnitch/pull/513#issuecomment-924400824
More information regarding this change: #493
* Use ebpf program to find PID of new connections.
before running the branch you have to compile ebpf_prog/opensnitch.c
opensnitch.c is an eBPF program. Compilation requires getting kernel source.
cd opensnitch
wget https://github.com/torvalds/linux/archive/v5.8.tar.gz
tar -xf v5.8.tar.gz
patch linux-5.8/tools/lib/bpf/bpf_helpers.h < ebpf_prog/file.patch
cp ebpf_prog/opensnitch.c ebpf_prog/Makefile linux-5.8/samples/bpf
cd linux-5.8 && yes "" | make oldconfig && make prepare && make headers_install # (1 min)
cd samples/bpf && make
objdump -h opensnitch.o #you should see many section, number 1 should be called kprobe/tcp_v4_connect
llvm-strip -g opensnitch.o #remove debug info
sudo cp opensnitch.o /etc/opensnitchd
cd ../../../daemon
--opensnitchd expects to find opensnitch.o in /etc/opensnitchd/
--start opensnitchd with:
opensnitchd -rules-path /etc/opensnitchd/rules -process-monitor-method ebpf
Co-authored-by: themighty1 <you@example.com>
Co-authored-by: Gustavo Iñiguez Goia <gooffy1@gmail.com>