| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
| |
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
| |
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
|
|
|
| |
We do a pretty nasty hack in the allowedips selftest to avoid having to
allocate more memory. Seems to work.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The peer (and keypair and local) lifecycle are managed through EPOCH and
refcounts. Primarily this is used in wg_noise to keep track of active
keypairs, however we can also use it to be sure no more peer references
exist. The structures are linked as such, so noise_remote cannot be
freed until all noise_keypairs are freed, and noise_local cannot be
freed until all noise_remotes are freed.
noise_keypair -> noise_remote -> noise_local
Therefore, if you hold a keypair reference you can be sure that remote
and local will still be around.
There are three main ways peers are referenced:
1) Incoming packets
1.a) Incoming handshake packets are passed to noise_consume_*,
which will (on success) return a refcounted remote which is dropped
at the end of wg_handshake.
1.b) Incoming cookie packets will have their index looked which will (on
success) return a refcounted remote, which is also dropped at the
end of wg_handshake.
1.c) Incoming data packets will have their index looked up which will
(on success) return a refcounted keypair. This keypair will be
dropped after the packet has been passed up the network stack, or
otherwise freed.
2) Outgoing data packets
2.a) Outgoing data packets are first looked up by wg_aip_lookup, which
returns a peer pointer, with an incremented remote refcount. This
is then dropped in wg_transmit after adding the packet to the
staged queue and sending the staged queue.
2.b) Packets in the staged queue do not hold any refcount for the remote
or keypair, because they do not reference the peer in any way, they
are just in the queue.
2.c) Packets finally get a refcoutned keypair in wg_peer_send_staged,
which is dropped after the packet is sent out the UDP socket, or
otherwise freed.
3) wg_timers system
3.a) The wg_timers system holds a reference to the peer whenever a
callout is scheduled. Instead of holding a refcount, we instead
disable the peer's timers, such that no callouts can be scheduled.
Some rationale for changes here:
We move the p_{send,recv} taskqgroup_detach into peer_free_deferred as
they will NULL fields in p_{send,recv}. If there are packets being
processed in wg_{en,de}crypt, then a call tou GROUPTASK_ENQUEUE will
dereference a NULL pointer. In general, we remove all references to the
peer in wg_peer_destroy, and free/deinit all the peer members once no
more references to the remote exist, in wg_peer_free_deferred.
Currently we take a refcount in wg_aip_lookup, which is to be sure that
the peer reference is valid for the entirety of wg_transmit. We do not
care about the refcount in wg_decrypt. It might be worth considering
storing the remote pointer in the allowedip entry, but it could be
argued both ways. For the time being, this is still correct. We don't
have a refcount for the peer stored in the allowedip table, as it is
protected by the table lock. One note here is the NULL p_remote check is
necessary to support selftest/allowedips.c, which does not specify a
p_remote. If we update the tests, then we may remove this check.
There are two added p_enabled checks, in run_retry_handshake and
run_send_keepalive. This is to align them with the other callout_reset
calls. In the case of p_zero_key_material, if we have set p_enabled =
false, then we subsequently clear keypairs and handshakes (on wg_down),
or we free the peer which will clear the keypairs for us.
We want to hold a refcount of remote in wg_{en,de}crypt to ensure that
the peer is still valid in the call to GROUPTASK_ENQUEUE. If we don't
then peer may become invalid after setting p_state. Another thread may
take the packet, put the keypair refcount and free the peer prior to the
call to GROUPTASK_ENQUEUE.
We no longer need to hold (haven't for a while) the EPOCH in
wg_send_initiation and wg_send_response, as we hold valid references for
the duration. This could be either a refcount of a remote or through the
wg_timers system as described above.
We also fix some refcount leaks in wgc_set.
Notes:
We may want to pull NET_EPOCH_WAIT out of wg_timers_disable, to improve
performance. However, we can destroy 20000 peers in less than 20ms so
the performance is not critical for this snapshot and can be addressed
later.
Finally, there is the special case of noise_remote_arg, which stores the
corresponding peer pointer. The peer is not refcounted however it will
have the same scope as the remote. In otherwords it is valid until we
call noise_remote_put on the remote.
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
| |
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
|
|
|
| |
We might add locks and things later. Mainly it doesn't cost much and
makes things easier/safer to reason about.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
|
|
|
| |
Good for hygiene, but also, lock hardening traps on double initialization
if we don't do this.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
|
|
|
| |
This doesn't add any reference counting, opting instead to go right to
the free. This could cause problems, but hopefully not.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Here we remove all M_WAITOK checks, because we don't want to hang while
trying to allocate memory. It is better to return an error so the user
can try again later.
We also make sure to check all the return codes in peer and interface
allocation. The structure of those functions is:
1) Allocate all memory
2) Initialise fields in order of the struct
3) Cleanup gotos
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
| |
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
|
|
| |
The rest of the stack does this.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
|
|
| |
This is already done anyway by if_clone_detach, so let that happen.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
| |
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
|
|
| |
IPs mean different things per-vnet.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
| |
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We can simplify the ratelimit init/deinit calls by allocating the table
statically, that is by not using hashinit_flags. That function ended up
doing some unnecessary calculation and meant that the mask couldn't be
constant.
By increasing the size of struct ratelimit, this also caught a nasty
(but benign) bug, where ratelimit_pool was initialised to allocate
sizeof(struct ratelimit) and not sizeof(struct ratelimit_entry). It has
been this way since FreeBSD tree and I didn't pick up on it while moving
the uma_zcreate call to wg_cookie.
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The two main changes here are:
* Remove cookie_ prefix from static functions. This is a leftover from
OpenBSD where they don't want static functions.
* Rename cm to macs, and cp to cm. Not sure where this came from but it
didn't really make much sense to leave it as is.
The reset are whitespace changes. Overall there is no modification to
functionality here, just appearances.
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Primarily this commit adds a cookie_valid state, to prevent a recently
booted machine from sending a mac2. We also do a little bit of reworking
on locking and a fixup for int to bool.
There is one slight difference to cookie_valid (latest_cookie.is_valid)
on Linux and that is to set cookie_valid to false when the
cookie_birthdate has expired. The purpose of this is to prevent the
expensive timer check after it has expired.
For the locking, we want to hold a write lock in cookie_maker_mac
because we write to mac1_last, mac1_valid and cookie_valid. This
wouldn't cause too much contention as this is a per peer lock and we
only do so when sending handshake packets. This is different from Linux
as Linux writes all it's variables at the start, then downgrades to a
read lock.
We also match cookie_maker_consume_payload locking to Linux, that is to
read lock while checking mac1_valid and decrypting the cookie then take
a write lock to set the cookie.
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
| |
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
| |
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
| |
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
| |
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
| |
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
| |
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
While it was nice to have per peer loop detection, it was not meant to
be. The loop tag has a tag type == 0, which conflicts with other tags.
Therefore we want to at least be a little bit more sure that the tag
cookie is unique to the loop tag. I guess the peer address was also
quite hacky so on the other side, I'm glad to be rid of that.
Now we have a loop of 8 (to any peer) which should be good enough for an
edge case operation.
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
|
|
| |
Also remove the stale entry from the TODO list.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
| |
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
| |
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
|
|
|
| |
And then fix broken allowedips implementation for the static unit tests
to pass.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously we relied on gc being called when adding a new entry, which
could leave us in a gc "blind spot". With this change, we schedule a
callout to run gc whenever we have entries in the table. The callout
will continue to run every ELEMENT_TIMEOUT seconds until the table is
empty.
Access to rl_gc is locked by rl_lock, so we will never have any threads
racing to callout_{pending,stop,reset}.
The alternative (which Linux does currently) is just to run the callout
every ELEMENT_TIMEOUT (1) second even when no entries are in the table.
However, the callout solution proposed here seems simple enough.
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
| |
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
| |
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
| |
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
| |
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
| |
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
| |
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
| |
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
|
|
|
|
|
|
|
| |
Nothing serious here, just use a goto in wg_deliver_{in,out} rather than
another if/else indentation. The code should have no functional change,
just improve readability.
Additionally, use a local `sc` variable rather than `peer->p_sc` in
spots.
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
| |
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
| |
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
|
|
|
|
| |
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
So the last change broke consuming responses, as it may return an
invalid remote pointer. Thanks for the catch zx2c4. We just pass a flag
"lookup_keypair" which will lookup the keypair when we want (for cookie)
and will not when we don't (for consuming responses).
It would be possible to merge both noise_remote_index_lookup and
noise_keypair_lookup, but the result would probably need to return a
void * (for both keypair and remote) or a noise_index * which would need
to be cast to the relevant type somewhere. The trickiest thing here
would be for if_wg to "put" the result of the function, as it may be a
remote or a keypair (which store their refcount in different locations).
Perhaps it would return a noise_index * which could contain the refcount
for both keypair and remote. It all seems easier to leave them separate.
The only argument for combining them would be to reduce duplication of
(similar) functions.
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
|
|
|
|
|
| |
This is needed, to remove the peer from the public key hashtable before
calling noise_remote_destroy. This will prevent any incoming handshakes
from starting in that time. It also cleans up the insert path to make it
more like it was before the wg_noise EPOCH changes.
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
| |
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
| |
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
| |
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
| |
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
| |
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
| |
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|
|
|
|
| |
Signed-off-by: Matt Dunwoodie <ncon@noconroy.net>
|