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 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>
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>
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>
Christian points out that we don't get spend notifications for old
channels if we truncate the store. We'd need more work to do this,
either validating the channels are still unspent, or replaying old
blocks from the truncation point.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Since we open with O_APPEND, any write() will append as we want it to.
But we want to distinguish a new store creation from a truncation due
to bad version.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
If something goes (fatally) wrong, we won't add it to the store.
This reveals a latent bug in routing_add_channel_announcement() and
friend which did a take() on msg, which it doesn't own. TAKES means
that it will take ownership IF the caller requests, not an unconditional
ownership transfer (which is an antipattern).
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We enter nodes in the map when we create channels, but those channels
could be local and unannounced. This triggered a failure in
test_gossip_persistence since the store truncated when it saw the
first thing was a node_announce.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Internally both payment and routing use 64-bit, but the interface
between them used 32-bit.
Since both components already support 64-bit we should use that.
In particular, the main daemon and subdaemons share the backtrace code,
with hooks for logging.
The daemon hook inserts the io_poll override, which means we no longer
need io_debug.[ch]. Though most daemons don't need it, they still link
against ccan/io, so it's harmess (suggested by @ZmnSCPxj).
This was tested manually to make sure we get backtraces still.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
If we only remember the actions that added channels then we'd restore them when
re-reading the gossip_store, so put a tombstone in there to remember to delete
it. These will be cleared upon re-writing the store since the announcements wont
be written anymore.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
This was a tricky one to find, it turns out that some nodes are sending
node_announcements even if they don't have a channel announced yet. If they are
a peer and the channel is currently verifying then we'll have a local channel in
the network view, hence accept the node_announcement, but when replaying, the
node_announcement will be replayed and we won't have a channel yet. This just
skips node_announcements, which is always safe.
Reported-by: @laszlohanyecz
Signed-off-by: Christian Decker <decker.christian@gmail.com>
This now works because we no longer call out to masterd or bitcoind to verify
the channels. It's also rather quick and silent so we can just process all
stored messages until we're done.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
Messages from peers and messages from the gossip_store now have completely
different entrypoints, so we don't need to trace their origin around the message
handling code any longer.
This stores and reads the channel_announcements in the wrapping message which
allows us to store associated data with the raw channel_announcements.
The gossip_store applies channel_announcements directly but it also returns it,
and it gets discarded as a duplicate. In the next commit we'll have gossip_store
apply all changes, bypassing verification, so the duplication is only temporary.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
Since we now store additional data along with the original messages they exceed
the length of the peer wire protocol messages.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
If we have a non-empty file and the version doesn't match, then we truncate and
write our own version. If the file is empty we write our version and the
truncate becomes a no-op
Signed-off-by: Christian Decker <decker.christian@gmail.com>
Since we may want to extend the on-disk format by adding custom information we
may as well just go the extra mile and reuse the serialization primitives we
already have.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
Moves any modifications based on an incoming gossip message into its own
function separate from the message verification. This allows us to skip
verification when reading messages from a trusted source, e.g., the
gossip_store, speeding up the gossip replay.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
When we read from the gossip_store we set store=false so that we don't duplicate
messages in the store.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
As proposed by @rustyrussell this makes it a bit easier to truncate and sync on
read errors.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
Ee will be replaying gossip messages from the gossip_store soon. This means that
not all messages originate from a peer, so we move the queuing of error messages
up into the peer message handler.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
If we're going to simply take() a pointer, don't allocate it off a random
object. Using NULL makes our intent clear, particularly with allocating
packets we're going to take() onto a queue.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Now it just returns true if it queued something. This allows it
to queue multiple packets, and lets it share code paths with other code
in future patches.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
As we add more features, the current code is insufficient.
1. Keep an array of single feature bits, for easy switching on and off.
2. Create feature_offered() which checks for both compulsory and optional
variants.
3. Invert requires_unsupported_features() and unsupported_features()
which tend to be double-negative, all_supported_features() and
features_supported().
4. Move single feature definition from wire/peer_wire.h to common/features.h.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We currently keep two copies; one in the broadcast structure to send
in order, and one in the routing information. Since we already keep
the broadcast index in the routing information, use that.
Conveniently, a zero index is the same as the old NULL test.
Rename struct node's announcement_idx to node_announce_msgidx to
make it match the other users.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We tal_dup_arr() it, which does take. Make it const in the structure;
the tal_dup_arr() removes the const, so it compiles without it, but it's
misleading.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We only access via index. We do, however, want to clean up when we
delete nodes and channels, so we tie lifetimes to that. This leads
us to put the index into 'struct queued_message'.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
1. make queue_peer_msg() use both if branches, as both equally likely.
2. Remove redundant *scid = NULL in handle_channel_announcement.
3. Log failing pending channel_updates.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
As per BOLT #7.
We don't do this for channel_update which are queued because the
channel_announcement is pending though.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
If the channel is pending, we queue the node_announcment and if the channel
is OK we re-call process_node_announcement. Make sure that second call
won't fail if the first succeeded.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We already have 'struct node', so rename 'struct routing_channel' to
'struct chan', and 'struct node_connection' to 'struct half_chan'.
Other minor changes:
1. rstate->channels -> rstate->chanmap.
2. 'connections' -> 'half'.
3. connection_to -> half_chan_to
4. connection_from -> half_chan_from
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
The containing `struct routing_channel` contains src and dst, so
remove them. However, the channel_update msgidx does belong int
`struct node_connection` along with the channel_update.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Returning the separate first routing_channel was a weird API: just
return the entire array. Sure, we have to treat the first node a bit
differently (because we don't charge ourselves fees), but it's still
simpler.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
To remove the redundant fields in `struct node_connection` (ie. 'src'
and 'dst' pointers) we need to deal with `struct routing_channel`.
This means we get a series of channels, from which the direction is
implied, so it's a bit more complex to decode. We add a helper
`other_node` to help with this, and since we're the only user of
`connection_to` we change that function to return the index.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Failure and pruning were the two places where a node_connection could
be freed; now they both deal with entire channels, we can remove the
NULL checks, and the destructor.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We discarded it; we should populate it. The comment is wrong, since
local_add_channel() doesn't add public channels, and we test that above.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This is twice the 'update_channel_interval' we get handed.
We delete the non-existent channel_add_connection and delete_connection
declarations from the header too.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We currently give them a free pass. The simplest fix is to give them
an old timestamp on initialization.
We still skip unannounced channels, on the assumption that they're
ours. And we set the last_update_timestamp to -1 when we convert to
gossip_getchannels_entry to indicate no update.
This breaks the DEVELOPER=1 pruning test, since we hardcode the 1
week timeout. That's fixed in the next patch.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We make new_routing_channel() populate both connections
(active=false), so local_add_channel becomes simpler. We also
suppress listchannels output of active=false unannounced channels, to
avoid breaking tests (also, these are unusable, so it makes sense to
omit them)
It also seems the logic in add_channel_direction is legacy: a
channel_announce cannot replace the scid (that would be a different
channel), we don't allow duplicate announcements, and the announcement
is never NULL.
And since we disallow repeated channel_announce already, I believe
'forward' is always true, greatly simplifying the logic in
handle_pending_cannouncement.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This makes 'routing_channel' the primary object in the system; it can have
one or two 'node_connection's attached, and points to two nodes.
The nodes are freed when no more routing_channel refer to them. The
routing_channel are freed when they contain no more 'node_connection'.
This fixes#1072 which I surmise was caused by a dangling
routing_channel after pruning.
Each node contains a single array of 'routing_channel's, not one for
each direction. The 'routing_channel' itself orders nodes in key
order (conveniently the index is equal to the direction flag we use),
and 'node_connection' with source in the same order.
There are helpers to assist with common questions like "which
'node_connection' leads out of this node?".
There are now two ways to find a channel:
1. Direct scid lookup via rstate->channels map.
2. Node key lookup, followed by channel traversal.
Several FIXMEs are inserted for where we can now do things more optimally.
Fixes: #1072
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We're going to make it a first-class citizen, and pending routing_channel
are not real ones (in particular, we don't want to create pending nodes).
We had a linked list called rstate->pending_cannouncement which we didn't
actually use, so put that back for now and add a FIXME to use a faster
data structure.
We need to check that list now in handle_channel_update, but we never
have a real routing_channel and a pending, unless the routing_channel
isn't public.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This hook is called when the queue is empty; we should only send gossip
according to the gossip timer. We're currently dribbling it out after
every message, in violation of the spec.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Now we have them, let's use them. I missed one case deliberately, since
that causes merge conflicts when I replace it in a following patch.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
I'm not completely conviced that we can't end up removing pending things,
so change asserts to simple returns.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
If we make destroy_node() remove itself from the map, then we simply
need to free it.
We can batch the frees (as we need) simply by reparenting all the pruned
nodes onto a single temporary parent, then freeing it, relying on tal's
internal datastructures.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
get_connection_by_scid() and update_to_pending() both do the same
lookup which we did in handle_channel_update().
Do the lookup once, and simplify the others.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We always hand in "NULL" (which means use tal_len on the msg), except
for two places which do that manually for no good reason.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We usually did this, but sometimes they were named after what they did,
rather than what they cleaned up.
There are still a few exceptions:
1. I didn't bother creating destroy_xxx wrappers for htable routines
which already existed.
2. Sometimes destructors really are used for side-effects (eg. to simply
mark that something was freed): these are clearer with boutique names.
3. Generally destructors are static, but they don't need to be: in some
cases we attach a destructor then remove it later, or only attach
to *some* cases. These are best with qualifiers in the destroy_<type>
name.
Suggested-by: @ZmnSCPxj
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
DEBUG:root:lightningd(16333): 2018-02-08T02:12:21.158Z lightningd(8262): lightning_openingd(0382ce59ebf18be7d84677c2e35f23294b9992ceca95491fcf8a56c6cb2d9de199): Failed hdr decrypt with rn=2
We only hand off the peer if we've not started writing, but that was
insufficient: we increment the sn twice on encrypting packet, so there's
a window before we've actually started writing where this is now
wrong.
The simplest fix is only to hand off from master when we've just written,
and have the read-packet path simply wake the write-packet path.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We get intermittant failure: WIRE_UNKNOWN_NEXT_PEER (First peer not ready)
because CHANNELD_NORMAL and actually telling gossipd that the channel
is available are distinct things: we need both.
(For test_closing_different_fees, we were testing CHANNELD_NORMAL on
the peer, not on l1, too).
But we may also directly send the announcement sigs if the height is
sufficient, so the simplest is to unify the messages.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Now we have wirestring, this is much more natural. And with the
24M length limit, we needn't be so concerned about dumping 64k peer
messages in hex.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
These are now logically arrays of pointers. This is much more natural,
and gets rid of the horrible utxo array converters.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Commit a57a2dcb86 introduced a time_t
in routing.h. So also move the time.h include to the header. This
fixes the build on FreeBSD.
Signed-off-by: Wladimir J. van der Laan <laanwj@gmail.com>
We were dropping these on the floor while checking for txout. So now
we add a map that holds announcements while we are checking.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
We are wasting way too much time looking for announcements and updates
in the broadcast. We can just hint where to find the message to be
evicted and safe the traversal.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
Adding channels that we are currently verifying to the map, and
skipping if we already have a channel at that position.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
We use this technique for the other tags, so use it here too.
This was drawn to my attention when I made more than 10 channels in a
block, and the string changed length:
Valgrind error file: valgrind-errors.31415
==31415== Conditional jump or move depends on uninitialised value(s)
==31415== at 0x4C35E20: bcmp (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==31415== by 0x11A624: queue_broadcast (broadcast.c:40)
==31415== by 0x118D93: handle_pending_cannouncement (routing.c:704)
==31415== by 0x1109E3: handle_txout_reply (gossip.c:1796)
==31415== by 0x111177: recv_req (gossip.c:1955)
==31415== by 0x136723: next_plan (io.c:59)
==31415== by 0x137220: do_plan (io.c:387)
==31415== by 0x13725E: io_ready (io.c:397)
==31415== by 0x138B97: io_loop (poll.c:305)
==31415== by 0x111352: main (gossip.c:2022)
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We drop all but the first announcement, so any work that is done for a
channel that we already know is wasted. Pulling this up duplicates
some of the work but allows us to skip the costly txout check.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
`tal_fmt` overallocates the returned string under some circumstances,
meaning that the trailer of the formatted string is unset, but still
considered in `tal_len`. The solution then is to truncate the
formatted string to the real string length. Only necessary here, since
we mix strings and `tal_len`.
Signed-off-by: Christian Decker <decker.christian@gmail.com>
We need to make sure all the updates are known to gossip. Since
one is the local update, we change that message to look the same.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Otherwise, we otherwise end up with out-of-order updates
(ie. preceeding announcements).
I assume that is because of the locally-inserted connections.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
