Hi everybody,

I’m writing to propose a set of p2p protocol changes to enable package
relay, soliciting feedback on the design and approach. Here is a link
to the most up-to-date proposal:

https://github.com/bitcoin/bips/pull/1324

If you have concept or approach feedback, *please respond on the
mailing list* to allow everybody to view and participate in the
discussion. If you find a typo or inaccurate wording, please feel free
to leave suggestions on the PR.

I’m also working on an implementation for Bitcoin Core.

The rest of this post will include the same contents as the proposal,
with a bit of reordering and additional context. If you are not 100%
up-to-date on package relay and find the proposal hard to follow, I

hope you find this format more informative and persuasive.

==Background and Motivation==

Users may create and broadcast transactions that depend upon, i.e.
spend outputs of, unconfirmed transactions. A “package” is the
widely-used term for a group of transactions representable by a
connected Directed Acyclic Graph (where a directed edge exists between
a transaction that spends the output of another transaction).

Incentive-compatible mempool and miner policies help create a fair,
fee-based market for block space. While miners maximize transaction
fees in order to earn higher block rewards, non-mining users
participating in transaction relay reap many benefits from employing
policies that result in a mempool with the same contents, including
faster compact block relay and more accurate fee estimation.
Additionally, users may take advantage of mempool and miner policy to
bump the priority of their transactions by attaching high-fee
descendants (Child Pays for Parent or CPFP). Only considering
transactions one at a time for submission to the mempool creates a
limitation in the node's ability to determine which transactions have
the highest feerates, since it cannot take into account descendants
until all the transactions are in the mempool. Similarly, it cannot
use a transaction's descendants when considering which of two
conflicting transactions to keep (Replace by Fee or RBF).

When a user's transaction does not meet a mempool's minimum feerate
and they cannot create a replacement transaction directly, their
transaction will simply be rejected by this mempool. They also cannot
attach a descendant to pay for replacing a conflicting transaction.
This limitation harms users' ability to fee-bump their transactions.
Further, it presents a security issue in contracting protocols which
rely on **presigned**, time-sensitive transactions to prevent cheating
(HTLC-Timeout in LN Penalty [1] [2] [3], Unvault Cancel in Revault
[4], Refund Transaction in Discreet Log Contracts [5], Updates in
eltoo [6]). In other words, a key security assumption of many
contracting protocols is that all parties can propagate and confirm
transactions in a timely manner.

In the past few years, increasing attention [0][1][2][3][6] has been
brought to **pinning attacks**, a type of censorship in which the
attacker uses mempool policy restrictions to prevent a transaction
from being relayed or getting mined. TLDR: revocation transactions
must meet a certain confirmation target to be effective, but their
feerates are negotiated well ahead of broadcast time. If the
forecasted feerate was too low and no fee-bumping options are
available, attackers can steal money from their counterparties. I walk
through a concrete example for stealing Lightning HTLC outputs at
~23:58 in this talk [7][8]. Note that most attacks are only possible
when the market for blockspace at broadcast time demands much higher
feerates than originally anticipated at signing time. Always
overestimating fees may sidestep this issue temporarily (while mempool
traffic is low and predictable), but this solution is not foolproof
and wastes users' money. The feerate market can change due to sudden
spikes in traffic (e.g. huge 12sat/vB dump a few days ago [9]) or
sustained, high volume of Bitcoin payments (e.g. April 2021 and
December 2017).

The best solution is to enable nodes to consider packages of
transactions as a unit, e.g. one or more low-fee parent transactions
with a high-fee child, instead of separately. A package-aware mempool
policy can help determine if it would actually be economically
rational to accept a transaction to the mempool if it doesn't meet fee
requirements individually. Network-wide adoption of these policies
would create a more purely-feerate-based market for block space and
allow contracting protocols to adjust fees (and therefore mining
priority) at broadcast time. Some support for packages has existed in
Bitcoin Core for years. Since v0.13, Bitcoin Core has used ancestor
packages instead of individual transactions to evaluate the incentive
compatibility of transactions in the mempool [10] and select them for
inclusion in blocks [11].

Package Relay, the concept of {announcing, requesting, downloading}
packages between nodes on the p2p network, has also been discussed for
many years. The earliest public mention I can find is from 2015 [12].
The two most common use cases for package relay are fee-bumping
otherwise-too-low-fee transactions and reducing the amount of orphans.
It seems uncontroversial to say that everybody desires package relay
conceptually, with varying degrees of urgency. Lots of work has been
done by others over the past few years, from which I've taken
inspiration from [13][14][15][16].

My approach has been to split the project into two components: (1) Package
Mempool Accept, which includes validation logic and mempool policy.
(3) Package Relay, which includes the p2p protocol changes.

Progress so far:
After discussions with various developers of contracting protocols
(with heavier emphasis towards LN), it was determined that a
package containing a child with all of its unconfirmed parents
(child-with-unconfirmed-parents or 1-child-multi-parent package) would
be sufficient for their use case, i.e. fee-bumping presigned
transactions. A child-with-unconfirmed-parents package has several
properties that make many things easier to reason about.

A few months ago, I proposed a set of policies for safe package
validation and fee assessment for packages of this restricted

topology [17]. A series of PRs implementing this proposal have

been merged into Bitcoin Core [18].

Theoretically, developing a safe and incentive-compatible package
mempool acceptance policy is sufficient to solve this issue. Nodes
could opportunistically accept packages (e.g. by trying combinations
of transactions rejected from their mempools), but this practice would
likely be inefficient at best and open new Denial of Service attacks
at worst. Additional p2p messages may enable nodes to request and
share package validation-related information with one another in a
more communication-efficient way.

Given that only package RBF remains for package mempool accept, and we
can make progress on p2p and mempool in parallel, I think it’s
appropriate to put forward a package relay proposal.

==Proposal==

This proposal contains 2 components: a “generic” package relay
protocol and an extension of it, child-with-unconfirmed-parents
packages, as version 1 package relay. Another version of packages,
“tx-with-unconfirmed-ancestors” can be created to extend package relay
for eliminating orphans.

===Generic Package Relay===

Two main ideas are introduced:

Download and validate packages of transactions together.

Provide information to help peers decide whether to request and/or how
to validate transactions which are part of a package.

====Intended Protocol Flow====

Due to the asynchronous nature of a distributed transaction relay
network, nodes may not receive all of the information needed to
validate a transaction at once. For example, after a node completes
Initial Block Download (IBD) and first starts participating in
transaction relay with an empty mempool, it is common to receive
orphans. In such scenarios where a node is aware that it is missing
information, a ''receiver-initiated'' dialogue is appropriate:

1. Receiver requests package information.

2. The sender provides package information, including the wtxids of
the transactions in the package and anything else that might be
relevant (e.g. total fees and size).

3. The reciever uses the package information to decide how to request
and validate the transactions.

Sometimes, no matter what order transactions are received by a node,
validating them individually is insufficient. When the sender is aware
of additional information that the receiver needs to accept a package,
a proactive ''sender-initiated'' dialogue should be enabled:

1. Sender announces they have package information pertaining to a
transaction that might otherwise be undesired on its own.

2. The receiver requests package information.

3. The sender provides package information, including the wtxids of
the transactions in the package and anything else that might be
relevant (e.g. total fees and size).

4. The reciever uses the package information to decide how to request
and validate the transactions.

Package relay is negotiated between two peers during the version
handshake. Package relay requires both peers to support wtxid-based
relay because package transactions are referenced by their wtxid.

====New Messages====

Three new protocol messages are added for use in any version of
package relay. Additionally, each version of package relay must define
its own inv type and "pckginfo" message version, referred to in this
document as "MSG_PCKG" and "pckginfo" respectively. See
BIP-v1-packages for a concrete example.

=====sendpackages=====

{|
| Field Name || Type || Size || Purpose
|-
|version || uint32_t || 4 || Denotes a package version supported by the node.
|-
|max_count || uint32_t || 4 ||Specifies the maximum number of transactions per package this node is
willing to accept.
|-
|max_weight || uint32_t || 4 ||Specifies the maximum total weight per package this node is willing
to accept.
|-
|}

1. The "sendpackages" message has the structure defined above, with
pchCommand == "sendpackages".

2. During version handshake, nodes should send a "sendpackages"
message indicate they support package relay and may request
packages.

3. The message should contain a version supported by the node. Nodes
should send a "sendpackages" message for each version they support.

4. The "sendpackages" message MUST be sent before sending a "verack"
message. If a "sendpackages" message is received afer "verack", the
sender should be disconnected.

5. If 'fRelay==false' in a peer's version message, the node must not
send "sendpackages" to them. If a "sendpackages" message is
received by a peer after sending `fRelay==false` in their version
message, the sender should be disconnected.

6.. Upon receipt of a "sendpackages" message with a version that is
not supported, a node must treat the peer as if it never received the
message.

7. If both peers send "wtxidrelay" and "sendpackages" with the same
version, the peers should announce, request, and send package
information to each other.

=====getpckgtxns=====

{|
| Field Name || Type || Size || Purpose
|-
|txns_length||CompactSize||1 or 3 bytes|| The number of transactions requested.
|-
|txns||List of wtxids||txns_length * 32|| The wtxids of each transaction in the package.
|}

1. The "getpckgtxns" message has the structure defined above, with
pchCommand == "getpckgtxns".

2. A "getpckgtxns" message should be used to request all or some of
the transactions previously announced in a "pckginfo" message,
specified by witness transactiosome id.

3. Upon receipt of a "getpckgtxns" message, a node must respond with
either a "pckgtxns" containing the requested transactions or a
"notfound" message indicating one or more of the transactions is
unavailable. This allows the receiver to avoid downloading and storing
transactions that cannot be validated immediately.

4. A "getpckgtxns" message should only be sent if both peers agreed to
send packages in the version handshake. If a "getpckgtxns" message
is received from a peer with which package relay was not negotiated,
the sender should be disconnected.

=====pckgtxns=====

{|
| Field Name || Type || Size || Purpose
|-
|txns_length||CompactSize||1 or 3 bytes|| The number of transactions provided.
|-
|txns||List of transactions||variable|| The transactions in the package.
|}

1. The "pckgtxns" message has the structure defined above, with
pchCommand == "pckgtxns".

2. A "pckgtxns" message should contain the transaction data requested
using "getpckgtxns".

3. A "pckgtxns" message should only be sent to a peer that requested
the package using "getpckgtxns". If a node receives an unsolicited
package, the sender should be disconnected.

4. A "pckgtxns" message should only be sent if both peers agreed to
send packages in the version handshake. If a "pckgtxns" message is
received from a peer with which package relay was not negotiated, the
sender should be disconnected.

===Version 1 Packages: child-with-unconfirmed-parents===

This extends package relay for packages consisting of one transaction
and all of its unconfirmed parents,by defining version 1 packages, a
pckginfo1 message, and a MSG_PCKG1 inv type. It enables the use case
in which a child pays for its otherwise-too-low-fee parents and their
mempool conflict(s).

====Intended Protocol Flow====

When relaying a package of low-fee parent(s) and high-fee child, the
sender and receiver do the following:

1. Sender announces they have a child-with-unconfirmed-parents package
for a child that pays for otherwise-too-low-fee parent(s) using
"inv(MSG_PCKG1)".

2. The receiver requests package information using
"getdata(MSG_PCKG1)".

3. The sender provides package information using "pckginfo1",
including the blockhash of the sender's best block, the wtxids of
the transactions in the package, their total fees and total weight.

4. The reciever uses the package information to decide how to request
the transactions. For example, if the receiver already has some of
the transactions in their mempool, they only request the missing ones.
They could also decide not to request the package at all based on the
fee information provided.

5. Upon receiving a "pckgtxns", the receiver submits the transactions
together as a package.

====New Messages====

A new inv type, "MSG_PCKG1", and new protocol message, "PCKGINFO1",
are added.

=====pckginfo1=====

{|
| Field Name || Type || Size || Purpose
|-
|blockhash || uint256 || 32 || The chain tip at which this package is defined.
|-
|pckg_fee||CAmount||4|| The sum total fees paid by all transactions in the package.
|-
|pckg_weight||int64_t||8|| The sum total weight of all transactions in the package.
|-
|txns_length||CompactSize||1 or 3 bytes|| The number of transactions provided.
|-
|txns||List of wtxids||txns_length * 32|| The wtxids of each transaction in the package.
|}

1. The "pckginfo1" message has the structure defined above, with
pchCommand == "pckginfo1".

2. A "pckginfo1" message contains information about a version 1
package (defined below), referenced by the wtxid of the transaction
it pertains to and the current blockhash.

3. Upon receipt of a "pckginfo1" message, the node should decide if it
wants to validate the package, request transaction data if
necessary, etc.

4. Upon receipt of a malformed "pckginfo1" message or package that
does not abide by the max_count, max_weight, or other rules
specified by the version agreed upon in the initial negotiation, the
sender should be disconnected. If a node receives a "pckginfo1"
message for which the "pckg_fee" or "pckg_weight" do not reflect the
true total fees and weight, respectively, or the transactions in the
package, the message is malformed.

5. A node MUST NOT send a "pckginfo1" message that has not been
requested by the recipient. Upon receipt of an unsolicited
"pckginfo1", a node should disconnect the sender.

6. A "pckginfo1" message should only be sent if both peers agreed to
send version 1 packages in the version handshake. If a "pckginfo1"
message is received from a peer with which package relay was not
negotiated, the sender should be disconnected.

=====MSG_PCKG1=====

1. A new inv type (MSG_PCKG1 == 0x6) is added, for use in inv messages
and getdata requests pertaining to version 1 packages.

2. As an inv type, it indicates that both transaction data and version
1 package information are available for the transaction. The
transaction is referenced by its wtxid. As a getdata request type, it
indicates that the sender wants package information for the
transaction.

3. Upon receipt of a "getdata" request for "MSG_PCKG1", the node
should respond with the version 1 package corresponding to the
requested transaction and its current chain tip, or with NOTFOUND.
The node should not assume that the sender is requesting the
transaction data as well.

====Child With Parent Packages Rules====

A child-with-unconfirmed-parents package sent between nodes must abide
by the rules below, otherwise the package is malformed and the sender
should be disconnected.

A version 1 or ''child-with-unconfirmed-parents'' package can be
defined for any transaction that spends unconfirmed inputs. The child
can be thought of as the "representative" of the package. This package
can be uniquely identified by the transaction's wtxid and the current
chain tip block hash.

A ''child-with-unconfirmed-parents'' package MUST be:

1. ''Sorted topologically.'' For every transaction t in the package,
if any of t's parents are present in the package, the parent must
appear somewhere in the list before t. In other words, the
transactions must be sorted in ascending order of the number of
ancestors present in the package.

2. ''Only 1 child with unconfirmed parents.'' The package must consist
of one transaction and its unconfirmed parents. There must not be
any other transactions in the package. Other dependency relationships
may exist within the package (e.g. one parent may spend the output of
another parent) provided that topological order is respected.

3. ''All unconfirmed parents.'' All of the child's unconfirmed parents
must be present.

4. ''No conflicts.'' None of the transactions in the package may
conflict with each other (i.e. spend the same prevout).

5. ''Total fees and weight.'' The 'total_fee' and 'total_weight'
fields must accurately represent the sum total of all transactions'
fees and weights as defined in BIP141, respectively.

Not all of the child's parents must be present; the child transaction
may also spend confirmed inputs. However, if the child has confirmed
parents, they must not be in the package.

While a child-with-unconfirmed-parents package is perhaps most
relevant when the child has a higher feerate than its parents, this
property is not required to construct a valid package.

====Clarifications====

''Q: Under what circumstances should a sender announce a
child-with-unconfirmed-parents package?''

A child-with-unconfirmed-parents package for a transaction should be
announced when it meets the peer's fee filter but one or more of its
parents don't; a "inv(MSG_PCKG1)" instead of "inv(WTX)" should be sent
for the child. Each of the parents which meet the peer's fee filter
should still be announced normally.

''Q: What if a new block arrives in between messages?''

A child-with-unconfirmed-parents package is defined for a transaction
based on the current chain state. As such, a new block extending the
tip may decrease the number of transactions in the package (i.e. if
any of the transaction's parents were included in the block). In a
reorg, the number of transactions in the package may decrease or
increase (i.e. if any of the transaction's parents were included in a
block in the previous chain but not the new one).

If the new block arrives before the "getdata" or "pckginfo1", nothing
needs to change.

If the new block arrives before "getpckgtxns" or before "pckgtxns",
the receiver may need to re-request package information if the block
contained a transaction in the package. If the block doesn't contain
any transactions in the package, whether it extends the previous tip
or causes a reorg, nothing needs to change.

''Q: Can "getpckgtxns" and "pckgtxns" messages contain only one
transaction?''

Yes.

===Further Protocol Extensions===

When introducing a new type of package, assign it a version number "n"
and use an additional "sendpackages" message during version handshake
to negotiate support for it. An additional package information message
"pckginfon" and inv type "MSG_PCKGn" should be defined for the type of
package. However, "getpckgtxns" and "pckgtxns" do not need to be
changed.

Example proposal for tx-with-unconfirmed-ancestors package relay: [19]

===Compatibility===

Older clients remain fully compatible and interoperable after this
change. Clients implementing this protocol will only attempt to send
and request packages if agreed upon during the version handshake.

===Package Erlay===

Clients using BIP330 reconciliation-based transaction relay (Erlay)
are able to use package relay without interference. In fact, a package
of transactions may be announced using both Erlay and package relay.
After reconciliation, if the initiator would have announced a
transaction by wtxid but also has package information for it, they may
send "inv(MSG_PCKG)" instead of "inv(WTX)".

===Rationale===

====P2P Message Design====

These p2p messages are added for communication efficiency and, as
such, one should measure alternative solutions based on the resources
used to communicate (not necessarily trustworthy) information: We
would like to minimize network bandwidth, avoid downloading a
transaction more than once, avoid downloading transactions that are
eventually rejected, and minimize storage allocated for
not-yet-validated transactions.

Consider these (plausible) scenarios in transaction relay:

Alice (the "sender") is relaying transactions to Bob (the "receiver").
Alice's mempool has a minimum feerate of 1sat/vB and Bob's has a
minimum feerate of 3sat/vB. For simplicity, all transactions are
1600Wu in virtual size and 500 bytes in serialized size. Apart from
the spending relationships specified, all other inputs are from
confirmed UTXOs.

1. Package {A, B} where A pays 0 satoshis and B pays 8000 satoshis in
fees.

2. Package {C, D} where C pays 0 satoshis and D pays 1200 satoshis in
fees.

3. Package {E, F, G, H, J} that pays 4000, 8000, 0, 2000, and 4000
satoshis in fees, respectively.

====Alternative Designs Considered====

''Package Information Only:'' Just having "pckginfo" gives enough
information for the receiver to accept the package. Omit the
"getpckgtxns" and "pckgtxns" messages. While this option is a good
fallback if batched transaction download fails for some reason, it
shouldn't be used as the default because it 'always' requires storage
of unvalidated transactions.

''No Package Information Round:'' Instead of having a package
information round, just use the child's wtxid to refer to the package
and always send the entire package together. This would cause nodes to
redownload duplicate transactions.

I have also created a slidedeck exploring various alternative designs
and some examples in which they fall flat [20]. Please feel free to
suggest other alternatives.

====Versioning System====

This protocol should be extensible to support multiple types of
packages based on future desired use cases. Two "flavors" of
versioning were considered:

1. When package mempool acceptance is upgraded to support more types
of packages, increment the version number (similar to Erlay).
During version handshake, peers negotiate which version of package
relay they will use by each sending one "sendpackages" message.

2. When introducing another type of package, assign a version number
to it and announce it as an additional supported version (similar
to Compact Block Relay). During version handshake, peers send one
"sendpackages" message for each version supported.

The second option was favored because it allows different parameters
for different versions. For example, it should be possible to support
both "arbitrary topology but maximum 3-transaction" package as well as
"child-with-unconfirmed-parents with default mempool ancestor limits"
packages simultaneously.

==Acknowledgements==

I hope to have made it abundantly clear that this proposal isn’t
inventing the concept of package relay, and in fact builds upon years
of work by many others, including Suhas Daftuar and Antoine Riard.

Thank you to John Newbery and Martin Zumsande for input on the design.

Thank you to Matt Corallo, Christian Decker, David Harding, Antoine
Poinsot, Antoine Riard, Gregory Sanders, Chris Stewart, Bastien
Teinturier, and others for input on the desired interface for
contracting protocols.

Looking forward to hearing your thoughts!

Best,

Gloria