[packet_interface] Add layer registry and HDLC reliability layer

Layers stack on a link and are configured on its packet_interface entry.
Each layer type registers itself with its kind, so a new type needs only
a new module. The first is a reliability layer in the style of HDLC
asynchronous balanced mode, with go-back-N retransmission.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

esphome/components/packet_interface/__init__.py

@@ -1,3 +1,5 @@
+import importlib
+import pkgutil
 from typing import Any
 
 import esphome.codegen as cg
@@ -7,14 +9,20 @@ from esphome.const import CONF_ID
 from esphome.cpp_generator import MockObj, MockObjClass
 from esphome.types import ConfigType
 
+from . import layers
+from .layers import PacketInterface, build_stack, packet_interface_ns, stack_schema
+
 CODEOWNERS = ["@clydebarrow"]
 DOMAIN = "packet_interface"
 IS_PLATFORM_COMPONENT = True
 
 CONF_PACKET_INTERFACE = "packet_interface"
 
-packet_interface_ns = cg.esphome_ns.namespace("packet_interface")
-PacketInterface = packet_interface_ns.class_("PacketInterface", cg.Component)
+# Each module in layers registers a layer type; import them all before building the schemas from the registry
+for module_info in pkgutil.iter_modules(layers.__path__):
+    importlib.import_module(f".layers.{module_info.name}", package=__package__)
+
+LINK_LAYER_SCHEMAS = stack_schema()
 
 __all__ = [
     "CONF_MAX_PACKET_SIZE",
@@ -31,19 +39,20 @@ def packet_interface_schema(
     default_max_packet_size: int,
     max_packet_size_limit: int = 4096,
 ) -> cv.Schema:
-    """Schema for a link platform."""
+    """Schema for a link platform, with the layers that can be stacked on it."""
     return cv.Schema(
         {
             cv.GenerateID(): cv.declare_id(class_),
             cv.Optional(
                 CONF_MAX_PACKET_SIZE, default=default_max_packet_size
             ): cv.int_range(16, max_packet_size_limit),
+            **LINK_LAYER_SCHEMAS,
         }
     ).extend(cv.COMPONENT_SCHEMA)
 
 
 async def new_packet_interface(config: ConfigType, *args: Any) -> MockObj:
-    """Create the link layer.
+    """Create the link layer and the layers configured on it, from the bottom up.
 
     The link's constructor takes its maximum packet size, usually as a std::span over a buffer that code generation
     sized from the configuration, so the two cannot differ.
@@ -53,4 +62,5 @@ async def new_packet_interface(config: ConfigType, *args: Any) -> MockObj:
     var = cg.new_Pvariable(config[CONF_ID], *args)
     cg.add(var.set_name(config[CONF_ID].id))
     await cg.register_component(var, config)
+    await build_stack(config, var)
     return var

esphome/components/packet_interface/hdlc_reliability.cpp

@@ -0,0 +1,320 @@
+#include "hdlc_reliability.h"
+#ifdef USE_PACKET_INTERFACE_HDLC
+#include "esphome/core/helpers.h"
+#include "esphome/core/log.h"
+
+namespace esphome::packet_interface {
+
+ESPHOME_LOG_TAG(TAG, "packet_interface.hdlc");
+
+static constexpr uint32_t RETRANSMIT_TIMER_ID = 0;
+static constexpr uint32_t ACK_TIMER_ID = 1;
+static constexpr uint32_t SEND_RETRY_TIMER_ID = 2;
+// A frame or acknowledgement the lower layer refused, for example while a radio was receiving, is sent again after this
+static constexpr uint32_t SEND_RETRY_MS = 50;
+// SABM is repeated at this interval, or the timeout if longer, until the peer answers
+static constexpr uint32_t RESET_INTERVAL_MS = 1000;
+
+// Control byte formats, with the poll/final bit (0x10) ignored
+static constexpr uint8_t POLL_FINAL_BIT = 0x10;
+static constexpr uint8_t SUPERVISORY_FRAME = 0x01;
+static constexpr uint8_t SUPERVISORY_RR = 0x00;
+static constexpr uint8_t SUPERVISORY_REJ = 0x02;
+static constexpr uint8_t UNNUMBERED_SABM = 0x2F;
+static constexpr uint8_t UNNUMBERED_UA = 0x63;
+
+static uint16_t frame_crc(uint8_t control, const uint8_t *payload, size_t length) {
+  uint16_t crc = crc16(&control, 1);
+  return crc16(payload, length, crc);
+}
+
+HdlcReliability::HdlcReliability(PacketInterface *lower) : lower_(lower) {
+  this->stack_on_(lower, CAPABILITY_RELIABILITY);
+  this->max_packet_size_ = lower->get_max_packet_size() - FRAME_OVERHEAD;
+}
+
+void HdlcReliability::setup() {
+  this->tx_slots_ = std::make_unique<uint8_t[]>(this->max_queue_length_ * this->max_packet_size_);
+  this->lower_->add_packet_interface_listener(
+      [this](const PacketBuffer &data, const PacketMetaData &meta_data) { this->receive_(data, meta_data); });
+}
+
+void HdlcReliability::dump_config() {
+  ESP_LOGCONFIG(TAG,
+                "HDLC Reliability on '%s':\n"
+                "  Max Packet Size: %zu\n"
+                "  Timeout: %" PRIu32 "ms\n"
+                "  Ack Delay: %" PRIu32 "ms\n"
+                "  Max Queue Length: %u\n"
+                "  Max Retries: %u",
+                this->name_, this->max_packet_size_, this->timeout_ms_, this->ack_delay_ms_, this->max_queue_length_,
+                this->max_retries_);
+}
+
+void HdlcReliability::loop() {
+  // Disabled first, as sending may deliver a reply to this layer and request the loop again
+  this->disable_loop();
+  if (this->state_ == LinkState::LINK_STATE_RESET_PENDING) {
+    this->state_ = LinkState::LINK_STATE_RESETTING;
+    this->start_timer_();
+    this->send_frame_(UNNUMBERED_SABM, nullptr, 0);
+  }
+  if (this->state_ != LinkState::LINK_STATE_CONNECTED)
+    return;
+  // Queued frames go first, as each carries the acknowledgement
+  if (this->send_pending_() && this->ack_due_ && !this->send_supervisory_(SUPERVISORY_RR))
+    this->retry_send_();
+}
+
+void HdlcReliability::retry_send_() {
+  this->set_timeout(SEND_RETRY_TIMER_ID, SEND_RETRY_MS, [this]() { this->request_loop_(); });
+}
+
+void HdlcReliability::owe_ack_(bool now) {
+  if (now || this->ack_delay_ms_ == 0) {
+    this->ack_due_ = true;
+    this->request_loop_();
+  } else if (!this->ack_pending_) {
+    // Waiting lets one acknowledgement cover several frames, or a reply carry it
+    this->set_timeout(ACK_TIMER_ID, this->ack_delay_ms_, [this]() {
+      this->ack_due_ = true;
+      this->request_loop_();
+    });
+  }
+  this->ack_pending_ = true;
+}
+
+void HdlcReliability::ack_sent_() {
+  this->ack_pending_ = false;
+  this->ack_due_ = false;
+  this->cancel_timeout(ACK_TIMER_ID);
+}
+
+void HdlcReliability::request_loop_() { this->enable_loop(); }
+
+bool HdlcReliability::send_to_interface(const PacketBuffer &data, const PacketMetaData &meta_data) {
+  size_t size = data.size();
+  if (size > this->max_packet_size_) {
+    ESP_LOGW(TAG, "%s: Packet of %zu bytes is larger than %zu", this->name_, size, this->max_packet_size_);
+    return false;
+  }
+  uint8_t outstanding = this->outstanding_();
+  if (outstanding >= this->max_queue_length_) {
+    ESP_LOGV(TAG, "%s: Send queue full", this->name_);
+    return false;
+  }
+  uint8_t seq = this->vs_;
+  data.copy_to(this->slot_(seq), size);
+  this->tx_lengths_[this->slot_index_(seq)] = size;
+  this->vs_ = (seq + 1) & 7;
+  // While the link is being reset, the frame waits in the queue and is sent when UA arrives
+  if (this->state_ == LinkState::LINK_STATE_CONNECTED) {
+    this->send_pending_();
+    if (outstanding == 0)
+      this->start_timer_();
+  }
+  return true;
+}
+
+bool HdlcReliability::send_frame_(uint8_t control, const uint8_t *payload, size_t length) {
+  uint16_t crc = frame_crc(control, payload, length);
+  uint8_t trailer[2] = {static_cast<uint8_t>(crc), static_cast<uint8_t>(crc >> 8)};
+  PacketBuffer body(payload, length);
+  PacketBuffer frame(&control, 1, length != 0 ? &body : nullptr, trailer, sizeof(trailer));
+  if (this->lower_->send_to_interface(frame, {}))
+    return true;
+  ESP_LOGV(TAG, "%s: Lower layer did not accept frame 0x%02X", this->name_, control);
+  return false;
+}
+
+bool HdlcReliability::send_supervisory_(uint8_t type) {
+  if (!this->send_frame_(SUPERVISORY_FRAME | (type << 2) | (this->vr_ << 5), nullptr, 0))
+    return false;
+  this->ack_sent_();
+  return true;
+}
+
+// Sends the next queued frame not yet sent, and returns true if none are left. One frame per loop pass lets other
+// components run between frames: on a radio, sending blocks, and the peer's frames are received in between. If the
+// lower layer refuses the frame, it is sent again shortly.
+bool HdlcReliability::send_pending_() {
+  if (this->vn_ == this->vs_)
+    return true;
+  uint8_t control = (this->vn_ << 1) | (this->vr_ << 5);
+  if (!this->send_frame_(control, this->slot_(this->vn_), this->tx_lengths_[this->slot_index_(this->vn_)])) {
+    this->retry_send_();
+    return false;
+  }
+  // Every I-frame carries an acknowledgement, so a pending RR is no longer needed
+  this->ack_sent_();
+  this->vn_ = (this->vn_ + 1) & 7;
+  if (this->vn_ == this->vs_)
+    return true;
+  this->request_loop_();
+  return false;
+}
+
+void HdlcReliability::resend_outstanding_() {
+  this->vn_ = this->va_;
+  this->send_pending_();
+}
+
+void HdlcReliability::start_timer_() {
+  uint32_t interval = this->timeout_ms_;
+  if (this->state_ != LinkState::LINK_STATE_CONNECTED && interval < RESET_INTERVAL_MS)
+    interval = RESET_INTERVAL_MS;
+  // Up to a quarter longer at random, so devices whose frames collided on a shared channel do not resend together
+  interval += random_uint32() % (interval / 4 + 1);
+  this->set_timeout(RETRANSMIT_TIMER_ID, interval, [this]() { this->on_timeout_(); });
+}
+
+void HdlcReliability::on_timeout_() {
+  if (this->state_ != LinkState::LINK_STATE_CONNECTED) {
+    this->send_frame_(UNNUMBERED_SABM, nullptr, 0);
+    this->start_timer_();
+    return;
+  }
+  if (this->outstanding_() == 0)
+    return;
+  if (++this->retries_ > this->max_retries_) {
+    ESP_LOGW(TAG, "%s: No acknowledgement after %u retries, resetting link", this->name_, this->max_retries_);
+    this->start_reset_();
+    return;
+  }
+  ESP_LOGV(TAG, "%s: Timeout, resending %u frames", this->name_, this->outstanding_());
+  this->resend_outstanding_();
+  this->start_timer_();
+}
+
+// Unacknowledged frames are discarded; frames queued from now on are numbered from 0 and sent once the link is up.
+// Resetting a connected link can lose data both ways, which is reported to the layers above.
+void HdlcReliability::start_reset_() {
+  bool was_connected = this->state_ == LinkState::LINK_STATE_CONNECTED;
+  this->state_ = LinkState::LINK_STATE_RESET_PENDING;
+  this->va_ = this->vs_ = this->vn_ = 0;
+  this->head_ = 0;
+  this->vr_ = 0;
+  this->retries_ = 0;
+  this->rej_sent_ = false;
+  this->ack_sent_();
+  this->cancel_timeout(RETRANSMIT_TIMER_ID);
+  this->request_loop_();
+  if (was_connected)
+    this->session_callback_.call(false);
+}
+
+void HdlcReliability::link_up_() {
+  ESP_LOGD(TAG, "%s: Link connected", this->name_);
+  this->state_ = LinkState::LINK_STATE_CONNECTED;
+  this->retries_ = 0;
+  this->cancel_timeout(RETRANSMIT_TIMER_ID);
+  if (this->outstanding_() != 0) {
+    this->resend_outstanding_();
+    this->start_timer_();
+  }
+}
+
+void HdlcReliability::process_ack_(uint8_t nr) {
+  uint8_t acked = (nr - this->va_) & 7;
+  if (acked > this->outstanding_()) {
+    ESP_LOGV(TAG, "%s: Ignoring invalid N(R) %u", this->name_, nr);
+    return;
+  }
+  if (acked == 0)
+    return;
+  // The peer may have had frames that were refused when last resent
+  if (acked > ((this->vn_ - this->va_) & 7))
+    this->vn_ = nr;
+  this->head_ = (this->head_ + acked) % this->max_queue_length_;
+  this->va_ = nr;
+  this->retries_ = 0;
+  if (this->outstanding_() == 0) {
+    this->cancel_timeout(RETRANSMIT_TIMER_ID);
+  } else {
+    this->start_timer_();
+  }
+}
+
+void HdlcReliability::receive_(const PacketBuffer &data, const PacketMetaData &meta_data) {
+  // Lower layers deliver packets in one piece; the payload is passed up without copying
+  if (!data.is_contiguous()) {
+    ESP_LOGW(TAG, "%s: Packet split across buffers is not supported", this->name_);
+    return;
+  }
+  auto frame = data.as_span();
+  if (frame.size() < FRAME_OVERHEAD) {
+    this->note_received_(false);
+    return;
+  }
+  size_t length = frame.size() - FRAME_OVERHEAD;
+  uint8_t control = frame[0];
+  const uint8_t *payload = frame.data() + 1;
+  uint16_t crc = frame[length + 1] | (frame[length + 2] << 8);
+  if (frame_crc(control, payload, length) != crc) {
+    ESP_LOGV(TAG, "%s: CRC error, frame dropped", this->name_);
+    this->note_received_(false);
+    return;
+  }
+  this->note_received_(true);
+
+  uint8_t command = control & ~POLL_FINAL_BIT;
+  if (command == UNNUMBERED_SABM) {
+    if (this->state_ == LinkState::LINK_STATE_CONNECTED) {
+      // The peer has restarted, so frames it has not acknowledged are lost
+      ESP_LOGD(TAG, "%s: Link reset by peer", this->name_);
+      this->start_reset_();
+    }
+    this->vr_ = 0;
+    this->rej_sent_ = false;
+    this->send_frame_(UNNUMBERED_UA, nullptr, 0);
+    // Both sides now start from 0, so the link is up even if our own SABM is unanswered
+    this->link_up_();
+    return;
+  }
+  if (command == UNNUMBERED_UA) {
+    if (this->state_ == LinkState::LINK_STATE_RESETTING)
+      this->link_up_();
+    return;
+  }
+  // Numbered frames are ignored until the link is reset; the peer answers our SABM first
+  if (this->state_ != LinkState::LINK_STATE_CONNECTED)
+    return;
+
+  uint8_t nr = control >> 5;
+  if ((control & 1) == 0) {
+    this->process_ack_(nr);
+    this->receive_info_(control, payload, length, meta_data);
+    return;
+  }
+  if ((control & 3) != SUPERVISORY_FRAME)
+    return;
+  this->process_ack_(nr);
+  if (((control >> 2) & 3) == SUPERVISORY_REJ && this->outstanding_() != 0) {
+    ESP_LOGV(TAG, "%s: Peer rejected from %u", this->name_, nr);
+    this->resend_outstanding_();
+    this->start_timer_();
+  }
+}
+
+void HdlcReliability::receive_info_(uint8_t control, const uint8_t *payload, size_t length,
+                                    const PacketMetaData &meta_data) {
+  uint8_t ns = (control >> 1) & 7;
+  if (ns != this->vr_) {
+    // A gap or a duplicate; ask once for a resend from the expected frame, then just acknowledge at once
+    if (!this->rej_sent_) {
+      this->rej_sent_ = this->send_supervisory_(SUPERVISORY_REJ);
+      if (this->rej_sent_)
+        return;
+    }
+    this->owe_ack_(true);
+    return;
+  }
+  this->vr_ = (this->vr_ + 1) & 7;
+  this->rej_sent_ = false;
+  // The acknowledgement is sent from loop(), so a reply sent from the callback can carry it instead
+  this->owe_ack_(false);
+  this->on_receive_from_interface_(PacketBuffer(payload, length), meta_data);
+}
+
+}  // namespace esphome::packet_interface
+#endif

esphome/components/packet_interface/hdlc_reliability.h

@@ -0,0 +1,93 @@
+#pragma once
+
+#include "esphome/core/defines.h"
+#ifdef USE_PACKET_INTERFACE_HDLC
+#include <array>
+#include <memory>
+
+#include "packet_interface.h"
+
+namespace esphome::packet_interface {
+
+/**
+ * A reliability layer that delivers packets in order, without loss or duplicates, between two peers.
+ *
+ * It follows HDLC asynchronous balanced mode, with sequence numbers modulo 8 and go-back-N retransmission:
+ * - I-frames carry a payload, the sequence number N(S) and an acknowledgement N(R) of the peer's frames.
+ * - RR frames acknowledge without a payload; REJ frames ask the peer to resend from N(R).
+ * - SABM resets both peers' sequence numbers, and UA answers it.
+ *
+ * Each frame is the HDLC control byte, the payload, then a CRC-16 of both in little endian order.
+ * Unacknowledged frames are resent after the timeout; after max_retries the link is reset.
+ */
+class HdlcReliability : public PacketInterface {
+ public:
+  /// Most frames that can be unacknowledged at once, set by the 3 bit sequence numbers.
+  static constexpr uint8_t MAX_WINDOW = 7;
+  /// Control byte plus CRC-16.
+  static constexpr size_t FRAME_OVERHEAD = 3;
+
+  explicit HdlcReliability(PacketInterface *lower);
+
+  void setup() override;
+  void loop() override;
+  void dump_config() override;
+  bool send_to_interface(const PacketBuffer &data, const PacketMetaData &meta_data) override;
+  size_t get_send_capacity() const override { return this->max_queue_length_ - this->outstanding_(); }
+
+  void set_ack_timeout(uint32_t timeout_ms) { this->timeout_ms_ = timeout_ms; }
+  void set_ack_delay(uint32_t ack_delay_ms) { this->ack_delay_ms_ = ack_delay_ms; }
+  void set_max_queue_length(uint8_t length) { this->max_queue_length_ = length; }
+  void set_max_retries(uint8_t retries) { this->max_retries_ = retries; }
+
+  bool is_connected() const { return this->state_ == LinkState::LINK_STATE_CONNECTED; }
+
+ protected:
+  enum class LinkState : uint8_t {
+    LINK_STATE_RESET_PENDING,  // SABM to be sent from loop()
+    LINK_STATE_RESETTING,      // SABM sent, waiting for UA
+    LINK_STATE_CONNECTED,
+  };
+
+  void receive_(const PacketBuffer &data, const PacketMetaData &meta_data);
+  void receive_info_(uint8_t control, const uint8_t *payload, size_t length, const PacketMetaData &meta_data);
+  bool send_frame_(uint8_t control, const uint8_t *payload, size_t length);
+  bool send_supervisory_(uint8_t type);
+  bool send_pending_();
+  void resend_outstanding_();
+  void retry_send_();
+  /// An RR is owed: at once, or after the ack delay unless an I-frame carries it first.
+  void owe_ack_(bool now);
+  void ack_sent_();
+  void process_ack_(uint8_t nr);
+  void on_timeout_();
+  void start_reset_();
+  void link_up_();
+  void start_timer_();
+  void request_loop_();
+  uint8_t outstanding_() const { return (this->vs_ - this->va_) & 7; }
+  uint8_t slot_index_(uint8_t seq) const { return (this->head_ + ((seq - this->va_) & 7)) % this->max_queue_length_; }
+  uint8_t *slot_(uint8_t seq) { return this->tx_slots_.get() + this->slot_index_(seq) * this->max_packet_size_; }
+
+  PacketInterface *lower_;
+  // max_queue_length_ slots of max_packet_size_ bytes, a ring starting at head_ which holds frame va_
+  std::unique_ptr<uint8_t[]> tx_slots_;
+  std::array<uint16_t, MAX_WINDOW> tx_lengths_{};  // indexed like the slots
+  uint32_t timeout_ms_{200};
+  uint32_t ack_delay_ms_{0};
+  uint8_t max_queue_length_{4};
+  uint8_t max_retries_{5};
+  uint8_t retries_{0};
+  uint8_t head_{0};
+  uint8_t vs_{0};  // sequence number of the next new frame
+  uint8_t vn_{0};  // next frame to send; frames from va_ up to it have been sent
+  uint8_t va_{0};  // oldest unacknowledged frame
+  uint8_t vr_{0};  // next frame expected from the peer
+  LinkState state_{LinkState::LINK_STATE_RESET_PENDING};
+  bool ack_pending_{false};  // an RR is owed
+  bool ack_due_{false};      // and should be sent now
+  bool rej_sent_{false};
+};
+
+}  // namespace esphome::packet_interface
+#endif

esphome/components/packet_interface/layers/__init__.py

@@ -0,0 +1,92 @@
+"""The kinds of layer that can be stacked on a packet interface, and the registry of their implementations.
+
+Each module in this package implements one layer type and registers it with its kind, for example:
+
+    @RELIABILITY.register("hdlc", schema=HDLC_SCHEMA)
+    async def build_hdlc(config: ConfigType, lower: MockObj) -> MockObj:
+        ...
+
+The packet_interface component imports every module here before it builds its schemas, so a new layer type needs
+only a new module. The kinds, and the order in which they are stacked, are fixed here.
+"""
+
+from collections.abc import Awaitable, Callable
+from dataclasses import dataclass
+from typing import Any
+
+import esphome.codegen as cg
+import esphome.config_validation as cv
+from esphome.const import CONF_TYPE
+from esphome.cpp_generator import MockObj
+from esphome.types import ConfigType
+
+packet_interface_ns = cg.esphome_ns.namespace("packet_interface")
+PacketInterface = packet_interface_ns.class_("PacketInterface", cg.Component)
+
+CONF_RELIABILITY = "reliability"
+
+# Builds a layer above lower and returns it
+LayerBuilder = Callable[[ConfigType, MockObj], Awaitable[Any]]
+
+
+@dataclass(frozen=True)
+class LayerType:
+    """One implementation of a kind of layer."""
+
+    schema: Any
+    build: LayerBuilder
+
+
+class LayerKind:
+    """A kind of layer, such as reliability, with the types that implement it."""
+
+    def __init__(self, key: str, default_type: str) -> None:
+        self.key = key
+        self.default_type = default_type
+        self.types: dict[str, LayerType] = {}
+
+    def register(
+        self, name: str, *, schema: Any
+    ) -> Callable[[LayerBuilder], LayerBuilder]:
+        """Register a type with its schema."""
+
+        def decorator(build: LayerBuilder) -> LayerBuilder:
+            self.types[name] = LayerType(schema, build)
+            return build
+
+        return decorator
+
+    @staticmethod
+    def _expand(value: Any) -> Any:
+        """Allow a layer to be enabled with no options, as in `reliability:`."""
+        return {} if value is None else value
+
+    def schema(self) -> Callable[[Any], ConfigType]:
+        """The schema for this kind, from the types registered so far."""
+        return cv.All(
+            self._expand,
+            cv.typed_schema(
+                {name: layer.schema for name, layer in self.types.items()},
+                default_type=self.default_type,
+            ),
+        )
+
+    def get_type(self, config: ConfigType) -> LayerType:
+        return self.types[config[CONF_TYPE]]
+
+
+RELIABILITY = LayerKind(CONF_RELIABILITY, "hdlc")
+# The kinds stacked above the link, from the bottom up
+STACK_ORDER = (RELIABILITY,)
+
+
+def stack_schema() -> dict:
+    """Schema entries for the layers of one stack; call only once all layer types are registered."""
+    return {cv.Optional(kind.key): kind.schema() for kind in STACK_ORDER}
+
+
+async def build_stack(config: ConfigType, lower: MockObj) -> None:
+    """Create the layers configured in config, from the bottom up, above lower."""
+    for kind in STACK_ORDER:
+        if (layer := config.get(kind.key)) is not None:
+            lower = await kind.get_type(layer).build(layer, lower)

esphome/components/packet_interface/layers/hdlc_reliability.py

@@ -0,0 +1,57 @@
+"""Reliability layer using HDLC style acknowledgement and go-back-N retransmission."""
+
+import esphome.codegen as cg
+import esphome.config_validation as cv
+from esphome.const import CONF_ID, CONF_TIMEOUT
+from esphome.cpp_generator import MockObj
+from esphome.types import ConfigType
+
+from . import RELIABILITY, PacketInterface, packet_interface_ns
+
+CONF_ACK_DELAY = "ack_delay"
+CONF_MAX_QUEUE_LENGTH = "max_queue_length"
+CONF_MAX_RETRIES = "max_retries"
+
+HdlcReliability = packet_interface_ns.class_("HdlcReliability", PacketInterface)
+
+
+def _ack_delay_within_timeout(config: ConfigType) -> ConfigType:
+    """The peer, configured the same, must hear the acknowledgement before its timeout."""
+    if (
+        config[CONF_ACK_DELAY].total_milliseconds
+        >= config[CONF_TIMEOUT].total_milliseconds
+    ):
+        raise cv.Invalid(
+            f"Must be shorter than '{CONF_TIMEOUT}'", path=[CONF_ACK_DELAY]
+        )
+    return config
+
+
+HDLC_SCHEMA = cv.All(
+    cv.Schema(
+        {
+            cv.GenerateID(): cv.declare_id(HdlcReliability),
+            cv.Optional(
+                CONF_TIMEOUT, default="200ms"
+            ): cv.positive_time_period_milliseconds,
+            cv.Optional(
+                CONF_ACK_DELAY, default="0ms"
+            ): cv.positive_time_period_milliseconds,
+            cv.Optional(CONF_MAX_QUEUE_LENGTH, default=4): cv.int_range(1, 7),
+            cv.Optional(CONF_MAX_RETRIES, default=5): cv.int_range(1, 255),
+        }
+    ).extend(cv.COMPONENT_SCHEMA),
+    _ack_delay_within_timeout,
+)
+
+
+@RELIABILITY.register("hdlc", schema=HDLC_SCHEMA)
+async def build_hdlc(config: ConfigType, lower: MockObj) -> MockObj:
+    cg.add_define("USE_PACKET_INTERFACE_HDLC")
+    var = cg.new_Pvariable(config[CONF_ID], lower)
+    cg.add(var.set_ack_timeout(config[CONF_TIMEOUT]))
+    cg.add(var.set_ack_delay(config[CONF_ACK_DELAY]))
+    cg.add(var.set_max_queue_length(config[CONF_MAX_QUEUE_LENGTH]))
+    cg.add(var.set_max_retries(config[CONF_MAX_RETRIES]))
+    await cg.register_component(var, config)
+    return var

esphome/core/defines.h

@@ -257,6 +257,7 @@
 #define USE_OTA_STATE_LISTENER
 #define USE_OUTPUT
 #define USE_OUTPUT_FLOAT_POWER_SCALING
+#define USE_PACKET_INTERFACE_HDLC
 #define USE_POWER_SUPPLY
 // Only defined by key-lookup preference backends; the slot-based platforms
 // (esp8266, rp2040) never set it in generated builds, and their preferences

tests/components/packet_interface/__init__.py

@@ -0,0 +1,9 @@
+import esphome.codegen as cg
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    async def to_code_testing(config: object) -> None:
+        cg.add_define("USE_PACKET_INTERFACE_HDLC")
+
+    manifest.to_code = to_code_testing

tests/components/packet_interface/common.yaml

@@ -9,9 +9,15 @@ packet_interface:
   - platform: uart
     id: packet_interface_uart
     max_packet_size: 256
+    reliability:
+      timeout: 300ms
+      ack_delay: 50ms
+      max_queue_length: 3
+      max_retries: 4
   - platform: udp
     id: packet_interface_network
     udp_id: packet_interface_udp
+    reliability:
 
 sensor:
   - platform: template

tests/components/packet_interface/hdlc_reliability.cpp

@@ -0,0 +1,308 @@
+#include <gtest/gtest.h>
+#include <cstdint>
+#include <functional>
+#include <memory>
+#include <vector>
+#include "esphome/components/packet_interface/hdlc_reliability.h"
+#include "esphome/core/application.h"
+#include "esphome/core/helpers.h"
+
+namespace esphome::packet_interface::testing {
+
+using Frame = std::vector<uint8_t>;
+
+// A link layer that records what is sent, so a test can deliver it to the peer
+class MockLink : public PacketInterface {
+ public:
+  MockLink() { this->set_max_packet_size(64); }
+  bool send_to_interface(const PacketBuffer &data, const PacketMetaData &) override {
+    if (this->budget == 0) {
+      this->refused++;
+      return false;
+    }
+    this->budget--;
+    this->sent.push_back(data.to_vector());
+    return true;
+  }
+  void inject(const Frame &frame) { this->on_receive_from_interface_(PacketBuffer(frame), {}); }
+
+  std::vector<Frame> sent;
+  size_t budget{SIZE_MAX};  // frames accepted before the link refuses, like a full network stack
+  size_t refused{0};
+};
+
+class TestHdlc : public HdlcReliability {
+ public:
+  using HdlcReliability::HdlcReliability;
+  using HdlcReliability::on_timeout_;
+  using HdlcReliability::outstanding_;
+};
+
+struct Peer {
+  explicit Peer(uint8_t queue_length = 4) {
+    this->hdlc = std::make_unique<TestHdlc>(&this->link);
+    this->hdlc->set_max_queue_length(queue_length);
+    this->hdlc->set_max_retries(3);
+    this->hdlc->setup();
+    this->hdlc->add_packet_interface_listener(
+        [this](const PacketBuffer &data, const PacketMetaData &) { this->received.push_back(data.to_vector()); });
+    this->hdlc->add_on_session_state_callback([this](bool connected) { this->states.push_back(connected); });
+  }
+  bool send(const Frame &data) { return this->hdlc->send_to_interface(PacketBuffer(data), {}); }
+
+  MockLink link;
+  std::unique_ptr<TestHdlc> hdlc;
+  std::vector<Frame> received;
+  std::vector<bool> states;
+};
+
+// Returns true to drop a frame on its way from one peer to the other
+using DropFilter = std::function<bool(const Frame &)>;
+
+static bool is_info(const Frame &frame) { return (frame[0] & 1) == 0; }
+static bool is_ack(const Frame &frame) { return (frame[0] & 0x0F) == 0x01; }
+
+// Move frames between the peers and run their loops until nothing more is sent
+static void pump(Peer &a, Peer &b, const DropFilter &drop_a_to_b = nullptr, const DropFilter &drop_b_to_a = nullptr) {
+  for (int i = 0; i != 50; i++) {
+    a.hdlc->loop();
+    b.hdlc->loop();
+    if (a.link.sent.empty() && b.link.sent.empty())
+      return;
+    auto from_a = std::move(a.link.sent);
+    auto from_b = std::move(b.link.sent);
+    a.link.sent.clear();
+    b.link.sent.clear();
+    for (auto &frame : from_a) {
+      if (!drop_a_to_b || !drop_a_to_b(frame))
+        b.link.inject(frame);
+    }
+    for (auto &frame : from_b) {
+      if (!drop_b_to_a || !drop_b_to_a(frame))
+        a.link.inject(frame);
+    }
+  }
+  FAIL() << "Peers did not go quiet";
+}
+
+static Frame make_frame(uint8_t control, const Frame &payload) {
+  Frame frame{control};
+  frame.insert(frame.end(), payload.begin(), payload.end());
+  uint16_t crc = crc16(frame.data(), frame.size());
+  frame.push_back(crc & 0xFF);
+  frame.push_back(crc >> 8);
+  return frame;
+}
+
+// The test main does not construct App; the retransmit timer needs its scheduler
+class HdlcReliabilityTest : public ::testing::Test {
+ protected:
+  void SetUp() override {
+    new (&App) Application;
+    App.pre_setup("hdlc-tests", 10, "", 0);
+  }
+  void TearDown() override { App.~Application(); }
+};
+
+TEST_F(HdlcReliabilityTest, MaxPacketSizeAllowsForOverhead) {
+  Peer a;
+  EXPECT_EQ(a.hdlc->get_max_packet_size(), 64 - HdlcReliability::FRAME_OVERHEAD);
+  EXPECT_EQ(a.link.top(), a.hdlc.get());
+  EXPECT_EQ(a.hdlc->get_capabilities(), CAPABILITY_RELIABILITY);
+}
+
+TEST_F(HdlcReliabilityTest, ResetHandshake) {
+  Peer a, b;
+  a.hdlc->loop();
+  ASSERT_EQ(a.link.sent.size(), 1);
+  EXPECT_EQ(a.link.sent[0], make_frame(0x2F, {}));  // SABM
+  pump(a, b);
+  EXPECT_TRUE(a.hdlc->is_connected());
+  EXPECT_TRUE(b.hdlc->is_connected());
+}
+
+TEST_F(HdlcReliabilityTest, QueuedBeforeConnectIsSentAfter) {
+  Peer a, b;
+  ASSERT_TRUE(a.send({0x01}));
+  ASSERT_TRUE(a.send({0x02}));
+  pump(a, b);
+  EXPECT_EQ(b.received, (std::vector<Frame>{{0x01}, {0x02}}));
+}
+
+TEST_F(HdlcReliabilityTest, DeliversInOrderBothWays) {
+  Peer a, b;
+  pump(a, b);
+  a.send({0x10});
+  b.send({0x20});
+  a.send({0x11});
+  pump(a, b);
+  EXPECT_EQ(b.received, (std::vector<Frame>{{0x10}, {0x11}}));
+  EXPECT_EQ(a.received, (std::vector<Frame>{{0x20}}));
+}
+
+TEST_F(HdlcReliabilityTest, SequenceNumbersWrap) {
+  Peer a(3), b(3);
+  pump(a, b);
+  std::vector<Frame> expected;
+  for (uint8_t i = 0; i != 20; i++) {
+    expected.push_back({i, static_cast<uint8_t>(i * 3)});
+    ASSERT_TRUE(a.send(expected.back()));
+    if (i % 2 == 1)
+      pump(a, b);
+  }
+  pump(a, b);
+  EXPECT_EQ(b.received, expected);
+}
+
+TEST_F(HdlcReliabilityTest, GapRecoveredByReject) {
+  Peer a, b;
+  pump(a, b);
+  bool dropped = false;
+  a.send({0x01});
+  a.send({0x02});
+  a.send({0x03});
+  pump(a, b, [&dropped](const Frame &frame) {
+    if (dropped || !is_info(frame))
+      return false;
+    dropped = true;
+    return true;
+  });
+  EXPECT_EQ(b.received, (std::vector<Frame>{{0x01}, {0x02}, {0x03}}));
+}
+
+TEST_F(HdlcReliabilityTest, LossRecoveredByTimeout) {
+  Peer a, b;
+  pump(a, b);
+  a.send({0x42});
+  pump(a, b, [](const Frame &frame) { return is_info(frame); });
+  EXPECT_TRUE(b.received.empty());
+  a.hdlc->on_timeout_();
+  pump(a, b);
+  EXPECT_EQ(b.received, (std::vector<Frame>{{0x42}}));
+}
+
+TEST_F(HdlcReliabilityTest, LostAckDoesNotDuplicate) {
+  Peer a, b;
+  pump(a, b);
+  a.send({0x42});
+  pump(a, b, nullptr, [](const Frame &frame) { return is_ack(frame); });
+  ASSERT_EQ(b.received.size(), 1);
+  a.hdlc->on_timeout_();
+  pump(a, b);
+  EXPECT_EQ(b.received.size(), 1);
+  // The resend was acknowledged, so the full queue is available again
+  for (int i = 0; i != 4; i++)
+    EXPECT_TRUE(a.send({0x01}));
+}
+
+TEST_F(HdlcReliabilityTest, QueueFull) {
+  Peer a(2), b(2);
+  pump(a, b);
+  EXPECT_TRUE(a.send({0x01}));
+  EXPECT_TRUE(a.send({0x02}));
+  EXPECT_FALSE(a.send({0x03}));
+}
+
+TEST_F(HdlcReliabilityTest, OversizedPacketRejected) {
+  Peer a;
+  EXPECT_FALSE(a.send(Frame(a.hdlc->get_max_packet_size() + 1)));
+}
+
+TEST_F(HdlcReliabilityTest, CrcErrorDropped) {
+  Peer a, b;
+  pump(a, b);
+  Frame frame = make_frame(0x00, {0x55});  // I-frame N(S)=0 N(R)=0
+  frame[1] ^= 0xFF;
+  b.link.inject(frame);
+  EXPECT_TRUE(b.received.empty());
+  b.link.inject(make_frame(0x00, {0x55}));
+  EXPECT_EQ(b.received, (std::vector<Frame>{{0x55}}));
+}
+
+TEST_F(HdlcReliabilityTest, RetriesExhaustedResetsLink) {
+  Peer a, b;
+  pump(a, b);
+  a.send({0x01});
+  pump(a, b, [](const Frame &frame) { return is_info(frame); });
+  for (int i = 0; i != 4; i++) {
+    a.hdlc->on_timeout_();
+    pump(a, b, [](const Frame &frame) { return is_info(frame); });
+  }
+  // The unacknowledged frame was discarded and the link was reset, which both report as data lost
+  EXPECT_TRUE(b.received.empty());
+  EXPECT_TRUE(a.hdlc->is_connected());
+  EXPECT_EQ(a.states, std::vector<bool>{false});
+  EXPECT_EQ(b.states, std::vector<bool>{false});
+  a.send({0x02});
+  pump(a, b);
+  EXPECT_EQ(b.received, (std::vector<Frame>{{0x02}}));
+}
+
+TEST_F(HdlcReliabilityTest, PeerRestart) {
+  Peer a;
+  auto b = std::make_unique<Peer>();
+  pump(a, *b);
+  a.send({0x01});
+  pump(a, *b);
+  // The peer restarts with fresh sequence numbers
+  b = std::make_unique<Peer>();
+  pump(a, *b);
+  a.send({0x02});
+  b->send({0x03});
+  pump(a, *b);
+  EXPECT_EQ(b->received, (std::vector<Frame>{{0x02}}));
+  EXPECT_EQ(a.received, (std::vector<Frame>{{0x03}}));
+  // The restarted peer's reset discarded what a had not yet had acknowledged
+  EXPECT_EQ(a.states, std::vector<bool>{false});
+  EXPECT_TRUE(b->states.empty());
+}
+
+TEST_F(HdlcReliabilityTest, RefusedFramesSentAgainWithoutTimeout) {
+  Peer a, b;
+  for (uint8_t i = 0; i != 3; i++)
+    ASSERT_TRUE(a.send({i}));
+  a.link.budget = 3;  // SABM, UA to the peer's SABM, and the first queued frame
+  pump(a, b);
+  EXPECT_GE(a.link.refused, 1);
+  EXPECT_EQ(b.received, (std::vector<Frame>{{0}}));
+  a.link.budget = SIZE_MAX;  // the lower layer accepts again, as when a radio stops receiving
+  pump(a, b);
+  EXPECT_EQ(b.received, (std::vector<Frame>{{0}, {1}, {2}}));
+  EXPECT_EQ(a.hdlc->outstanding_(), 0);
+}
+
+TEST_F(HdlcReliabilityTest, RefusedAckSentAgain) {
+  Peer a, b;
+  pump(a, b);
+  a.send({1});
+  for (auto &frame : a.link.sent)
+    b.link.inject(frame);
+  a.link.sent.clear();
+  b.link.budget = 0;
+  b.hdlc->loop();
+  EXPECT_TRUE(b.link.sent.empty());
+  b.link.budget = SIZE_MAX;
+  pump(a, b);
+  EXPECT_EQ(a.hdlc->outstanding_(), 0);
+}
+
+TEST_F(HdlcReliabilityTest, AckDelayLetsReplyCarryAck) {
+  Peer a, b;
+  b.hdlc->set_ack_delay(100);
+  pump(a, b);
+  a.send({1});
+  for (auto &frame : a.link.sent)
+    b.link.inject(frame);
+  a.link.sent.clear();
+  b.hdlc->loop();
+  EXPECT_TRUE(b.link.sent.empty());  // no RR yet
+  b.send({2});
+  ASSERT_EQ(b.link.sent.size(), 1);
+  EXPECT_TRUE(is_info(b.link.sent[0]));
+  EXPECT_EQ(b.link.sent[0][0] >> 5, 1);  // N(R) acknowledges the frame from a
+  b.link.sent.clear();
+  b.hdlc->loop();
+  EXPECT_TRUE(b.link.sent.empty());  // the RR is no longer owed
+}
+
+}  // namespace esphome::packet_interface::testing

tests/unit_tests/test_packet_interface.py

@@ -1,10 +1,20 @@
 """Unit tests for the packet_interface component."""
 
+from typing import Any
+
 import pytest
 
 from esphome import config_validation as cv
 from esphome.components import packet_interface
 from esphome.components.packet_interface import CONF_MAX_PACKET_SIZE
+from esphome.components.packet_interface.layers import (
+    CONF_RELIABILITY,
+    RELIABILITY,
+    STACK_ORDER,
+)
+from esphome.components.packet_interface.layers.hdlc_reliability import (
+    CONF_MAX_QUEUE_LENGTH,
+)
 from esphome.components.uart import packet_interface as uart_packet_interface
 from esphome.components.udp import packet_interface as udp_packet_interface
 
@@ -34,3 +44,55 @@ def test_schema_default(setup_core) -> None:
         packet_interface.PacketInterface, 100
     )
     assert schema({})[CONF_MAX_PACKET_SIZE] == 100
+
+
+def test_reliability_defaults(setup_core) -> None:
+    """An empty reliability key selects the HDLC layer with its defaults."""
+    config = uart_packet_interface.CONFIG_SCHEMA({CONF_RELIABILITY: None})
+    reliability = config[CONF_RELIABILITY]
+    assert reliability["type"] == "hdlc"
+    assert reliability["timeout"].total_milliseconds == 200
+    assert reliability[CONF_MAX_QUEUE_LENGTH] == 4
+    assert reliability["max_retries"] == 5
+
+
+def test_reliability_queue_length_limit(setup_core) -> None:
+    """The queue cannot hold more frames than 3 bit sequence numbers allow."""
+    with pytest.raises(cv.Invalid):
+        uart_packet_interface.CONFIG_SCHEMA(
+            {CONF_RELIABILITY: {CONF_MAX_QUEUE_LENGTH: 8}}
+        )
+
+
+def test_ack_delay_shorter_than_timeout(setup_core) -> None:
+    """The acknowledgement must not wait longer than the peer's timeout."""
+    reliability = uart_packet_interface.CONFIG_SCHEMA(
+        {CONF_RELIABILITY: {"timeout": "500ms", "ack_delay": "100ms"}}
+    )[CONF_RELIABILITY]
+    assert reliability["ack_delay"].total_milliseconds == 100
+    with pytest.raises(cv.Invalid, match="shorter"):
+        uart_packet_interface.CONFIG_SCHEMA(
+            {CONF_RELIABILITY: {"timeout": "500ms", "ack_delay": "500ms"}}
+        )
+
+
+def test_layer_types_registered() -> None:
+    """Importing packet_interface registers the built-in layer types."""
+    assert {kind.key: set(kind.types) for kind in STACK_ORDER} == {
+        CONF_RELIABILITY: {"hdlc"},
+    }
+
+
+def test_registered_type_selectable(setup_core) -> None:
+    """A newly registered layer type can be selected with type."""
+
+    async def build(config: dict, lower: Any) -> Any:
+        return lower
+
+    RELIABILITY.register("test", schema=cv.Schema({}))(build)
+    try:
+        config = RELIABILITY.schema()({"type": "test"})
+        assert config["type"] == "test"
+        assert RELIABILITY.get_type(config).build is build
+    finally:
+        del RELIABILITY.types["test"]