When a connect fails, if it's an important peer, we set a timer. If
we have a manual connect command, this means we do this again, leading
to another timer.
For a manual command, free any existing timer; the normal fail logic
will start another if necessary.
Reported-by: @ZmnSCPxj
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
At least say whether we failed to connect at all, or failed cryptographic
handshake, or failed reading/writing init messages.
The errno can be "Operation now in progress" if the other end closes the
socket on us: this happens when we handshake with the wrong key and it
hangs up on us. Fixing this would require work on ccan/io though.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
When we get a reconnection, kill the current remote peer, and wait for the
master to tell us it's dead. Then we hand it the new peer.
Previously, we would end up with gossipd holding multiple peers, and
the logging was really hard to interpret; I'm not completely convinced
that we did the right thing when one terminated, either.
Note that this now means we can have peers with neither ->local nor ->remote
populated, so we check that more carefully.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Currently we intuit it from the fd being closed, but that may happen out
of order with when the master thinks it's dead.
So now if the gossip fd closes we just ignore it, and we'll get a
notification from the master when the peer is disconnected.
The notification is slightly ugly in that we have to disable it for
a channel when we manually hand the channel back to gossipd.
Note: as stands, this is racy with reconnects. See the next patch.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This means that openingd and closingd now forward our gossip. But the real
reason we want to do this is that it gives an easy way for gossipd to kill
any active daemon, by closing its fd: previously closingd and openingd didn't
read the fd, so tended not to notice.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
(This was sitting in my gossip-enchancement patch queue, but it simplifies
this set too, so I moved it here).
In 94711969f we added an explicit gossip_index so when gossipd gets
peers back from other daemons, it knows what gossip it has sent (since
gossipd can send gossip after the other daemon is already complete).
This solution is insufficient for the more general case where gossipd
wants to send other messages reliably, so replace it with the other
solution: have gossipd drain the "gossip fd" which the daemon returns.
This turns out to be quite simple, and is probably how I should have
done it originally :(
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
1. Lifetime of 'struct reaching' now only while we're actively doing connect.
2. Always free after a single attempt: if it's an important peer, retry
on a timer.
3. Have a single response message to master, rather than relying on
peer_connected on success and other msgs on failure.
4. If we are actively connecting and we get another command for the same
id, just increment the counter
The result is much simpler in the master daemon, and much nicer for
reconnection: if they say to connect they get an immediate response,
rather than waiting for 10 retries. Even if it's an important peer,
it fires off another reconnect attempt, unless it's actively
connecting now.
This removes exponential backoff: that's restored in next patch. It
also doesn't handle multiple addresses for a single peer.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Rather than using a flag in reaching/peer; we make it self-contained
as the next patch puts it straight into a timer callback.
Also remove unused 'succeeded' field from struct peer.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
And on channel_fail_permanent and closing (the two places we drop to
chain), we tell gossipd it's no longer important.
Fixes: #1316
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
These don't have a maximum number of reconnect attempts, and ensure
that we try to reconnect when the peer dies.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Got some intermittant failures, mainly caused by the tests being slow
enough that the peer reconnected. We should always suppress
reconnection if we can, and not stress too much in the !DEVELOPER case
where we can't.
We should turn off dev-no-reconnect *always* unless told we will
reconnect, and since we can't if !DEVELOPER, don't do the connection
check there.
Instead of adding an option to line_graph, we remove it in favor
of connect (since we only use it with n=2 anyway).
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
The no-rescan change requires us to rescan one last time from the first_blocknum
of our channels (if we have any). The migrations just drop blocks that are
higher, then insert a dummy with the first_blocknum, and then clean up after
us. If we don't have any channels we don't go back at all.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
This shaves off about 15% of our integration testing suite on my machine. It
assumes we never reorg below the first block the node starts with, which is true
for all tests, so it's safe.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
Simplification of the offset calculation to use the rescan parameter, and rename
of `wallet_first_blocknum`. We now use either relative rescan from our last
known location, or absolute if a negative rescan was given. It's all handled in
a single location (except the case in which the blockcount is below our
precomputed offset), so this should reduce surprises.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
This is intended to recover from an inconsistent state, involving
`onchaind`. Should we for some reason not restore the `onchaind` process
correctly we can instruct `lightningd` to go back in time and just replay
everything.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
This is a big simplification, we just report the DBs current blockchain height
as the point to continue scanning, or the passed in default. No more guessing
where to continue from or whether the wallet was used and when it first saw the
light of day.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
Since we reference the channel ID to allow cascades in the database we also need
the ability to look up a channel by its database ID.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
This will allow us in the next commit to store the transactions that triggered
this event in the DB and thus allowing us to replay them later on.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
This will be used to replay transactions that were witnessed in the blockchain
during startup, so that onchaind can be recreate its state.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
These transactions being seen on the blockchain triggered some action in
onchaind so we need to replay them when we restore the onchaind.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
We used to queue the preimages to be sent to onchaind only after receiving the
onchaind_init_reply. Once we start replaying we might end up in a situation in
which we queue the tx that onchaind should react to before providing it with the
preimages. This commit just moves the preimages being sent, making it atomic
with the init, and without changing the order.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
These were so far only used for bolt11 construction, but we'll need them for the
DNS seed as well, so here we just pull them out into their own unit and prefix
them.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
Retrying gives spurious failures, since we see transactions from previous
runs. That makes it near impossible to diagnose the actual problem.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Careful log examination revealed that we were generating a block before one
of the mutual close txs had entered the mempool. This is rare because it
means that both peers have to be too slow.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
d822ba1ee accidentally removed this case, which is important: if the
other side didn't get our final matching closing_signed, it will
reconnect and try again. We consider the channel no longer "active"
and thus ignore it, and get upset when it send the
`channel_reestablish` message.
We could just consider CLOSINGD_COMPLETE to be active, but then we'd
have to wait for the closing transaction to be mined before we'd allow
another connection.
We can't special case it when the peer reconnects, because there
could be (in theory) multiple channels for that peer in CLOSINGD_COMPLETE,
and we don't know which one to reestablish.
So, we need to catch this when they send the reestablish, and hand
that msg to closingd to do negotiation again. We already have code
to note that we're in CLOSINGD_COMPLETE and thus ignore any result
it gives us.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We're about to remove automatic retrying of connect, and that uncovered
that we actually print out our "Server started" message before we create
the listening socket.
Move the init higher (outside the db transaction) and make it a
request/response, the loop until it's done.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
The new connect code revealed an existing race: we tell gossipd to
release the peer, but at the same time it connects in. gossipd fails
the release because the peer is remote, and json_fundchannel fails.
Instead, we catch this race when we get peer_connected() and we were
trying to open a channel. It means keeping a list of fundchannels which
are awaiting a gossipd response though.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We missed it in some corner cases where we crashed/were killed between
being told of the lockin and sending the channel_normal_operation message.
When we were restarted, we were told both sides were locked in already,
so we never updated the state.
Pull the entire "tell channeld" logic into channel_control.c, and make
it clear that we need to keep waching if we cant't tell channeld. I think
we did get this correct in practice, since funding_announce_cb has the
same test, but it's better to be clear.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We'd usually commit to the db soon, but there's a window where it
could be missed.
Also moves loc into the block it's used and make it tmpctx to avoid
an explicit free.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Without this, we can get errors on shutdown:
Valgrind error file: valgrind-errors.27444
==27444== Invalid read of size 8
==27444== at 0x1950E2: secp256k1_pubkey_load (secp256k1.c:127)
==27444== by 0x19CF87: secp256k1_ec_pubkey_serialize (secp256k1.c:189)
==27444== by 0x14FED9: towire_pubkey (towire.c:59)
==27444== by 0x15AAFB: towire_gossipctl_peer_disconnected (gen_gossip_wire.c:969)
==27444== by 0x1253EF: opening_channel_errmsg (opening_control.c:526)
==27444== by 0x1386A3: destroy_subd (subd.c:589)
==27444== by 0x18222C: notify (tal.c:240)
==27444== by 0x1826E1: del_tree (tal.c:400)
==27444== by 0x182733: del_tree (tal.c:410)
==27444== by 0x182733: del_tree (tal.c:410)
==27444== by 0x182B1F: tal_free (tal.c:511)
==27444== by 0x11FC53: main (lightningd.c:410)
==27444== Address 0x6c3af98 is 72 bytes inside a block of size 216 free'd
==27444== at 0x4C30D3B: free (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==27444== by 0x1827BC: del_tree (tal.c:421)
==27444== by 0x182B1F: tal_free (tal.c:511)
==27444== by 0x11F3C7: shutdown_subdaemons (lightningd.c:211)
==27444== by 0x11FC27: main (lightningd.c:406)
==27444== Block was alloc'd at
==27444== at 0x4C2FB0F: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==27444== by 0x182296: allocate (tal.c:250)
==27444== by 0x182863: tal_alloc_ (tal.c:448)
==27444== by 0x12F2DF: new_peer (peer_control.c:74)
==27444== by 0x125600: new_uncommitted_channel (opening_control.c:576)
==27444== by 0x125870: peer_accept_channel (opening_control.c:668)
==27444== by 0x13032A: peer_sent_nongossip (peer_control.c:427)
==27444== by 0x116B9E: peer_nongossip (gossip_control.c:60)
==27444== by 0x116F2B: gossip_msg (gossip_control.c:172)
==27444== by 0x138323: sd_msg_read (subd.c:503)
==27444== by 0x137C02: read_fds (subd.c:330)
==27444== by 0x175550: next_plan (io.c:59)
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
I had a weird failure which was caused by an unexpected disconnect and
reconnecct. Since we are prersistend and recover from these, they can
slip through our tests; most tests don't involve reconnection, so we
need to catch this explicitly.
For the connect() helper, we always suppress reconnection; tests which
want it all want other options so don't use this helper anyway. (Actually,
after I said that, test_closing_while_disconnected was added when I
rebased, which did require it, so I had to open-code that one).
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Rusty Russell
ZmnSCPxj
Christian Decker