Guus Sliepen [Sun, 27 Jun 2021 14:19:37 +0000 (16:19 +0200)]
Don't try to forward packets to a node we don't have a key for.
If we got a packet that's meant to be relayed, don't call
sptps_send_data() if we don't have a valid key yet for the desination
node, but do keep trying to get a working connection to that node. Based
on a patch from thorkill.
Aaron LI [Sun, 5 Apr 2020 11:07:42 +0000 (19:07 +0800)]
Use auto-clone device /dev/{tun,tap} as default on FreeBSD/DragonFly
DragonFly BSD doesn't pre-create `/dev/tunX` or `/dev/tapX` devices
anymore since 2019-Jul-31 [0]. So it's better to use the auto-clone
device `/dev/tun` or `/dev/tap` as the default TUN or TAP device.
The TUN/TAP device has the same behavior on DragonFly BSD and FreeBSD.
See also pull request: https://github.com/DragonFlyBSD/DeltaPorts/pull/925
Because the result of read() was incorrectly stored in an unsigned
variable, an error reading from the random number generator device would
result in an infinite loop that would start writing out of bounds and
eventually corrupt the stack.
pacien [Mon, 20 Jan 2020 12:58:13 +0000 (13:58 +0100)]
fd_device: allow fd to be passed through a unix socket
New restrictions on the Android OS forbid direct leaking of file descriptors.
This patch allows the tinc daemon to have an fd and the associated
permissions transferred to it through a Unix domain socket.
Address caches are associated with nodes, so just use the address cache
in the node_t struct. Also ensure we always have opened an address cache in
setup_outgoing_connection().
inet_checksum() accesses packet data as an array of uint16_t, but the
packet data can be for example of "anonymous struct pseudo" type from
route_ipv6_unreachable().
This type isn't a compatible type with uint16_t so a strict aliasing
violation occurs and causes the checksum to be computed incorrectly.
Fix this by using the memcpy() idiom to read the packet data as an array of
uint16_t in inet_checksum() (this should be understood by compilers and
optimized accordingly, so no actual copy occurs).
Andreas Rammhold [Thu, 28 Feb 2019 19:38:14 +0000 (20:38 +0100)]
fix: use EVP_DecryptUpdate while decrypting
With OpenSSL versions 1.0.2r & 1.1.1b there were changes in regards to
how OpenSSL treats misuse of Encrypt/Decrypt EVP methods in the opposite
case. E.g. using the encrypt methods in a decrypt context. OpenSSL now
returns an error in these situations. [1]
Since tinc used the EVP_EncryptUpdate function in the cipher_decrypt
function the new sanity check was triggered causing tinc to be unusable
with said OpenSSL versions.
Guus Sliepen [Tue, 18 Dec 2018 16:44:08 +0000 (17:44 +0100)]
Prevent large amounts of UDP probes being sent consecutively.
We cannot reset udp_ping_sent to zero when we receive a valid reply to
an UDP probe, because that would cause a new one to be sent immediately
in try_udp(). Instead, add a bit to node_status_t to keep track of whether we
have a UDP probe that's waiting for a reply.
Thanks to Ronny Nilsson for spotting the source of the problem.
Guus Sliepen [Mon, 22 Oct 2018 18:31:37 +0000 (20:31 +0200)]
Attempt to make the test suite work with Windows executables.
The test suite still assumes a POSIX shell to run the tests, but now handles
the case when the executables themselves are (cross-)compiled for Windows,
with a .exe extension. Also, DOS line endings must be converted to UNIX
line endings in some cases.
Some tests now pass on Linux+Wine, but others do not, mainly due to Wine
not handling services very well.
Guus Sliepen [Thu, 18 Oct 2018 15:17:20 +0000 (17:17 +0200)]
Make more variables safe for use in invitations.
When writing one's own invitation files, more variables are now accepted
by the invitee. The goal is to allow anything that doesn't interfere
with the existing network configuration of the invitee and that doesn't
cause any unexpected behaviour, such as starting running commands.
Guus Sliepen [Thu, 18 Oct 2018 14:10:32 +0000 (16:10 +0200)]
Use the onlink flag when adding routes on Linux.
As reported by iczero, adding gateway routes on Linux can fail if the
gateway address is not reachable yet, either because the interface is
down, or if the gateway address is not inside any route that already is
added or is being added to that interface. By adding the onlink flag,
iproute2 will just add the route without questions.
iczero [Wed, 17 Oct 2018 05:58:52 +0000 (22:58 -0700)]
Fix tinc-up generation on windows
- use `%INTERFACE%` instead of `$INTERFACE` on windows
- correct typo in `netsh interface` (was `netsh inetface`)
- remove `static` when setting ipv6 address
Guus Sliepen [Mon, 8 Oct 2018 19:27:08 +0000 (21:27 +0200)]
Fix the compiler attribute test to work with Clang.
Clang doesn't like the __nonnull__ attribute being applied to functions
that don't take pointer arguments, and this causes errors when running
the configure script.
Prevent oracle attacks in the legacy protocol (CVE-2018-16737, CVE-2018-16738)
The legacy authentication protocol allows an oracle attack that could
potentially be exploited. This commit contains several mitigations:
- Connections are no longer closed immediately on error, but put in
a "tarpit".
- The authentication protocol now requires a valid CHAL_REPLY from the
initiator of a connection before sending a CHAL_REPLY of its own.
- Reduce the amount of connections per second accepted.
- Null ciphers or digests are no longer allowed in METAKEYs.
- Connections that claim to have the same name as the local node are
rejected.
Just to be on the safe side:
- The new protocol now requires a valid SIG from the initiator of a
connection before sending a SIG of its own.
Guus Sliepen [Tue, 12 Jun 2018 17:47:02 +0000 (19:47 +0200)]
Remove the wxPython GUI.
This GUI is missing a lot of functionality, and won't be part of an 1.1.0
release. Therefore, it's being removed, and might be brought back after
1.1.0.
Running the "del Port" command when no Port is set returns an error. But
it is not necessary anyway since the later "set Port" command will
overwrite it.
Guus Sliepen [Fri, 30 Mar 2018 09:50:40 +0000 (11:50 +0200)]
Prevent an infinite loop in get_recent_address().
When a node is offline, but we still have edges to it that have the same
address as we already have in our address cache, an infinite loop would
happen in get_recent_address(), because we forgot to advance the pointer
in the list of known addresses, and kept looking at the same one over
and over.
Thanks to Sven-Haegar Koch for spotting the bug and providing
diagnostics.
Todd C. Miller [Tue, 27 Feb 2018 21:20:46 +0000 (14:20 -0700)]
Call WSAWaitForMultipleEvents() in a loop until we have checked all events.
WSAWaitForMultipleEvents() only returns the index of the first event that is read. We need to call WSAWaitForMultipleEvents() repeatedly to check if other events are also ready. Otherwise, a single busy event (such as the TAP device) can starve the other events.
Guus Sliepen [Tue, 27 Feb 2018 20:08:57 +0000 (21:08 +0100)]
Work around a GCC bug that causes inet_checksum() to give wrong results.
Valgrind reports the following bug:
==24877== Conditional jump or move depends on uninitialised value(s)
==24877== at 0x12283E: inet_checksum (route.c:80)
==24877== by 0x12283E: route_ipv6_unreachable (route.c:315)
==24877== by 0x1236AC: route_ipv6 (route.c:751)
==24877== by 0x1236AC: route (route.c:1160)
==24877== by 0x113DE0: receive_tcppacket (net_packet.c:493)
==24877== by 0x1119D4: receive_meta (meta.c:315)
==24877== by 0x113288: handle_meta_connection_data (net.c:287)
==24877== by 0x11A091: handle_meta_io (net_socket.c:491)
==24877== by 0x10FB0C: event_loop (event.c:370)
==24877== by 0x11362E: main_loop (net.c:489)
==24877== by 0x10CACA: main (tincd.c:551)
Clearing the variable pseudo in route_ipv6_unreachable removes this error,
but the resulting checksum is still bad. If one instead adds a dummy
write that depends on checksum, the error goes away and the checksum is
correct.
Guus Sliepen [Mon, 26 Feb 2018 21:19:43 +0000 (22:19 +0100)]
Unconditionally remove timeouts from the queue before calling the callback.
We are going to unlink the timeout from the splay tree anyway, so do it
unconditionally before the callback, instead of waiting until after the
callback to check whether or not to remove it based on its expiration
time.
Todd C. Miller [Thu, 22 Feb 2018 21:27:37 +0000 (14:27 -0700)]
In device_handle_read() we need to reset the read event on error or
it will keep firing. This is easy to reproduce by suspending the
machine while tinc is running.
Todd C. Miller [Wed, 21 Feb 2018 03:18:38 +0000 (20:18 -0700)]
Fix heap corruption on Windows exposed by the use-after free fix.
reset_address_cache() could call free_known_addresses() on a struct
addrinfo * that was returned by getaddrinfo(). It seems safest to just
make a copy of the addresses returned by getaddrinfo() so we can always
use free_known_addresses() instead of trying to determine whether or
not we need to use freeaddrinfo().
Guus Sliepen [Sun, 18 Feb 2018 15:51:06 +0000 (16:51 +0100)]
Reduce memory allocations due to HMAC() and EVP_MD_*().
HMAC() allocates a temporary buffer on the heap each time it is called.
Similarly, we called EVP_MD_CTX_create() every time we wanted to
calculate a hash. Use HMAC_CTX and EVP_MD_CTX variables to store the
state so no (re)allocations are necessary. HMAC() was called for every
legacy packet sent and received.
Guus Sliepen [Sun, 18 Feb 2018 14:38:12 +0000 (15:38 +0100)]
Reduce memory allocations due to zlib's uncompress().
Everytime uncompress() is called, zlib allocates some buffer on the heap
and frees it again. When compression is enabled, this is the biggest source
of memory allocations in tinc. Instead of using this function, use
inflate(), which can store its state in a z_stream variable, which avoids
(re)allocating memory for every packet received.
Guus Sliepen [Sun, 18 Feb 2018 14:33:36 +0000 (15:33 +0100)]
Add code coverage testing support.
Allows configure to be run with the --enable-code-coverage flag, allowing
one to run make check-code-coverage, which runs the test suite and produces
a code coverage report.