lukechilds
/
lightning
mirror of https://github.com/lukechilds/lightning.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
6008 Commits
55 Branches
107 Tags
38 MiB
Tree: b3215a866b
Branches Tags
${ item.name }
Create tag ${ searchTerm }
Create branch ${ searchTerm }
from 'b3215a866b'
${ noResults }
lightning/connectd/connect_wire.csv

68 lines
2.3 KiB
Raw Normal View History

connectd: wean off gossip_wire message types, use our own. This involves much renaming of gossip -> channel. Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
7 years ago
#include <common/cryptomsg.h>
common/per_per_state: generalize lightningd/peer_comm Part 1 Encapsulating the peer state was a win for lightningd; not surprisingly, it&#39;s even more of a win for the other daemons, especially as we want to add a little gossip information. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
6 years ago
#include <common/per_peer_state.h>
connectd: do socket binding during initial setup. gossipd does a two-step initialization: it only tries to create the listening sockets when it&#39;s activated. This means it doesn&#39;t know the addresses to announce until this point. Now connectd is doing this, this would mean we&#39;d have to tell gossipd later (&#34;oh, BTW here are your addresses&#34;) since we need to start gossipd before connectd activation. So make connectd do all the setup, but only actually listen on the fds once we activate it. We clone the gossip_init message into connect_wire.csv. The master daemon still waits for a reply from connectd for the activate message, since it wants to be listening before it prints &#34;Server started&#34;. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
7 years ago
#include <common/wireaddr.h>
connectd: wean off gossip_wire message types, use our own. This involves much renaming of gossip -&gt; channel. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
7 years ago
#include <lightningd/gossip_msg.h>
connectd: do socket binding during initial setup. gossipd does a two-step initialization: it only tries to create the listening sockets when it&#39;s activated. This means it doesn&#39;t know the addresses to announce until this point. Now connectd is doing this, this would mean we&#39;d have to tell gossipd later (&#34;oh, BTW here are your addresses&#34;) since we need to start gossipd before connectd activation. So make connectd do all the setup, but only actually listen on the fds once we activate it. We clone the gossip_init message into connect_wire.csv. The master daemon still waits for a reply from connectd for the activate message, since it wants to be listening before it prints &#34;Server started&#34;. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
7 years ago
connectctl_init,2000
Use node_id everywhere for nodes. I tried to just do gossipd, but it was uncontainable, so this ended up being a complete sweep. We didn&#39;t get much space saving in gossipd, even though we should save 24 bytes per node. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
6 years ago
connectctl_init,,id,struct node_id
connectd: do socket binding during initial setup. gossipd does a two-step initialization: it only tries to create the listening sockets when it&#39;s activated. This means it doesn&#39;t know the addresses to announce until this point. Now connectd is doing this, this would mean we&#39;d have to tell gossipd later (&#34;oh, BTW here are your addresses&#34;) since we need to start gossipd before connectd activation. So make connectd do all the setup, but only actually listen on the fds once we activate it. We clone the gossip_init message into connect_wire.csv. The master daemon still waits for a reply from connectd for the activate message, since it wants to be listening before it prints &#34;Server started&#34;. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
7 years ago
connectctl_init,,num_wireaddrs,u16
connectctl_init,,wireaddrs,num_wireaddrs*struct wireaddr_internal
connectctl_init,,listen_announce,num_wireaddrs*enum addr_listen_announce
connectctl_init,,tor_proxyaddr,?struct wireaddr
connectctl_init,,use_tor_proxy_always,bool
connectctl_init,,dev_allow_localhost,bool
connectctl_init,,use_dns,bool
connectctl_init,,tor_password,wirestring
# Connectd->master, here are the addresses I bound, can announce.
connectctl_init_reply,2100
connectctl_init_reply,,num_bindings,u16
connectctl_init_reply,,bindings,num_bindings*struct wireaddr_internal
connectctl_init_reply,,num_announcable,u16
connectctl_init_reply,,announcable,num_announcable*struct wireaddr
# Activate the connect daemon, so others can connect.
connectctl_activate,2025
# Do we listen?
connectctl_activate,,listen,bool
# Connectd->master, I am ready.
connectctl_activate_reply,2125
connectd: new daemon to handle connections. This is just copying most of gossipd/gossip.c into connectd/connect.c. It shares the same wire format as gossipd during transition, and changes are deliberately minimal. It also has an additional message &#39;connect_reconnected&#39; which it sends to the master daemon to tell it to kill a peer; gossipd relied on closing the gossipfd to do this, but connectd doesn&#39;t maintain an fd with remote peers. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
7 years ago
# connectd->master: disconnect this peer please (due to reconnect).
connect_reconnected,2112
Use node_id everywhere for nodes. I tried to just do gossipd, but it was uncontainable, so this ended up being a complete sweep. We didn&#39;t get much space saving in gossipd, even though we should save 24 bytes per node. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
6 years ago
connect_reconnected,,id,struct node_id
connectd: wean off gossip_wire message types, use our own. This involves much renaming of gossip -&gt; channel. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
7 years ago
# Master -> connectd: connect to a peer.
connectctl_connect_to_peer,2001
Use node_id everywhere for nodes. I tried to just do gossipd, but it was uncontainable, so this ended up being a complete sweep. We didn&#39;t get much space saving in gossipd, even though we should save 24 bytes per node. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
6 years ago
connectctl_connect_to_peer,,id,struct node_id
connectd: rely on the master to tell us to reconnect. connectd tells master about every disconnection, and master knows whether it&#39;s important to reconnect. Just get the master to invoke a new connect command if it considers the peer important! The only twist is timeouts: we don&#39;t want to immediately reconnect if we&#39;ve failed to connect. To solve this, connectd passes a &#39;delaytime&#39; to the master when a connection fails, and the master passes it back when it asks for a connection. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
6 years ago
connectctl_connect_to_peer,,seconds_waited,u32
connectd: remove separate address hint message. Include it as an optional field in the connect_to_peer message (it was added before we had optional fields). The only issue is that reconnects want it too, so again connectd hands it back to master in connectctl_connect_failed. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
6 years ago
connectctl_connect_to_peer,,addrhint,?struct wireaddr_internal
connectd: wean off gossip_wire message types, use our own. This involves much renaming of gossip -&gt; channel. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
7 years ago
connectd: always tell master when connection fails/succeeded. We used to separate implicit connection requests (ie. timed retries for important peers) and explicit ones, and send a WIRE_CONNECTCTL_CONNECT_TO_PEER_RESULT for the latter. In the success case, that&#39;s now redundant, since we hand the connected peer to the master using WIRE_CONNECT_PEER_CONNECTED; we just need a message for the failure case. And we might as well tell the master every failure, so we don&#39;t have to distinguish internally. This also solves a race we had before: connectd would send WIRE_CONNECTCTL_CONNECT_TO_PEER_RESULT which completes the incoming JSON connect command, then send WIRE_CONNECT_PEER_CONNECTED. So there&#39;s a window where the JSON command can return, but the peer isn&#39;t known to lightningd yet. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
6 years ago
# Connectd->master: connect failed.
connectctl_connect_failed,2020
Use node_id everywhere for nodes. I tried to just do gossipd, but it was uncontainable, so this ended up being a complete sweep. We didn&#39;t get much space saving in gossipd, even though we should save 24 bytes per node. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
6 years ago
connectctl_connect_failed,,id,struct node_id
connectd: always tell master when connection fails/succeeded. We used to separate implicit connection requests (ie. timed retries for important peers) and explicit ones, and send a WIRE_CONNECTCTL_CONNECT_TO_PEER_RESULT for the latter. In the success case, that&#39;s now redundant, since we hand the connected peer to the master using WIRE_CONNECT_PEER_CONNECTED; we just need a message for the failure case. And we might as well tell the master every failure, so we don&#39;t have to distinguish internally. This also solves a race we had before: connectd would send WIRE_CONNECTCTL_CONNECT_TO_PEER_RESULT which completes the incoming JSON connect command, then send WIRE_CONNECT_PEER_CONNECTED. So there&#39;s a window where the JSON command can return, but the peer isn&#39;t known to lightningd yet. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
6 years ago
connectctl_connect_failed,,failreason,wirestring
connectd: rely on the master to tell us to reconnect. connectd tells master about every disconnection, and master knows whether it&#39;s important to reconnect. Just get the master to invoke a new connect command if it considers the peer important! The only twist is timeouts: we don&#39;t want to immediately reconnect if we&#39;ve failed to connect. To solve this, connectd passes a &#39;delaytime&#39; to the master when a connection fails, and the master passes it back when it asks for a connection. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
6 years ago
connectctl_connect_failed,,seconds_to_delay,u32
connectd: remove separate address hint message. Include it as an optional field in the connect_to_peer message (it was added before we had optional fields). The only issue is that reconnects want it too, so again connectd hands it back to master in connectctl_connect_failed. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
6 years ago
connectctl_connect_failed,,addrhint,?struct wireaddr_internal
connectd: wean off gossip_wire message types, use our own. This involves much renaming of gossip -&gt; channel. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
7 years ago
gossipd: hand a gossip_store_fd to all subdaemons. This will let them read from the gossip store directly. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
6 years ago
# Connectd -> master: we got a peer. Three fds: peer, gossip and gossip_store
connectd: wean off gossip_wire message types, use our own. This involves much renaming of gossip -&gt; channel. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
7 years ago
connect_peer_connected,2002
Use node_id everywhere for nodes. I tried to just do gossipd, but it was uncontainable, so this ended up being a complete sweep. We didn&#39;t get much space saving in gossipd, even though we should save 24 bytes per node. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
6 years ago
connect_peer_connected,,id,struct node_id
connectd: wean off gossip_wire message types, use our own. This involves much renaming of gossip -&gt; channel. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
7 years ago
connect_peer_connected,,addr,struct wireaddr_internal
gossipd: don&#39;t send gossip stream, let per-peer daemons read it themselves. Keeping the uintmap ordering all the broadcastable messages is expensive: 130MB for the million-channels project. But now we delete obsolete entries from the store, we can have the per-peer daemons simply read that sequentially and stream the gossip itself. This is the most primitive version, where all gossip is streamed; successive patches will bring back proper handling of timestamp filtering and initial_routing_sync. We add a gossip_state field to track what&#39;s happening with our gossip streaming: it&#39;s initialized in gossipd, and currently always set, but once we handle timestamps the per-peer daemon may do it when the first filter is sent. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
6 years ago
connect_peer_connected,,pps,struct per_peer_state
connectd: wean off gossip_wire message types, use our own. This involves much renaming of gossip -&gt; channel. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
7 years ago
connect_peer_connected,,gflen,u16
Use localfeatures and globalfeatures consistently. That&#39;s what BOLT #1 calls them; make it easier for people to grep. Reported-by: @niftynei Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
6 years ago
connect_peer_connected,,globalfeatures,gflen*u8
connectd: wean off gossip_wire message types, use our own. This involves much renaming of gossip -&gt; channel. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
7 years ago
connect_peer_connected,,lflen,u16
Use localfeatures and globalfeatures consistently. That&#39;s what BOLT #1 calls them; make it easier for people to grep. Reported-by: @niftynei Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
6 years ago
connect_peer_connected,,localfeatures,lflen*u8
connectd: wean off gossip_wire message types, use our own. This involves much renaming of gossip -&gt; channel. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
7 years ago
# master -> connectd: peer has disconnected.
connectctl_peer_disconnected,2015
Use node_id everywhere for nodes. I tried to just do gossipd, but it was uncontainable, so this ended up being a complete sweep. We didn&#39;t get much space saving in gossipd, even though we should save 24 bytes per node. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
6 years ago
connectctl_peer_disconnected,,id,struct node_id
connectd: wire up dev_memleak. We need several notleak() annotations here: 1. The temporary structure which is handed to retry_peer_connected(). It&#39;s waiting for the master to respond to our connect_reconnected message. 2. We don&#39;t keep a pointer to the io_conn for a peer, so we need to mark those as not being a leak. Signed-off-by: Rusty Russell &lt;rusty@rustcorp.com.au&gt;
6 years ago
# master -> connectd: do you have a memleak?
connect_dev_memleak,2033
connect_dev_memleak_reply,2133
connect_dev_memleak_reply,,leak,bool