TUN-10385: Add connectivity checks foundation
Check / check (1.22.x, macos-latest) (push) Has been cancelled
Check / check (1.22.x, ubuntu-latest) (push) Has been cancelled
Check / check (1.22.x, windows-latest) (push) Has been cancelled
Semgrep config / semgrep/ci (push) Has been cancelled

This adds the foundation for the new cloudflared pre-checks by creating a new prechecks package. This adds the following:

* types.go: Status, CheckResult, Report, Config (add IPVersion allregions.ConfigIPVersion field to Config)
* interfaces.go: DNSResolver, TCPDialer, QUICDialer, ManagementDialer
This commit is contained in:
Miguel da Costa Martins Marcelino
2026-04-15 22:40:23 +00:00
parent b0b898c235
commit df54d27710
2 changed files with 176 additions and 0 deletions
+82
View File
@@ -0,0 +1,82 @@
package prechecks
import (
"context"
"crypto/tls"
"net"
"net/netip"
"time"
"github.com/quic-go/quic-go"
"github.com/rs/zerolog"
"github.com/cloudflare/cloudflared/edgediscovery/allregions"
)
// DNSResolver abstracts edge DNS discovery used by DNS probes.
//
// The production implementation wraps allregions.EdgeDiscovery
// (edgediscovery/allregions/discovery.go), which performs an SRV lookup for
// _v2-origintunneld._tcp.argotunnel.com, falls back to DNS-over-TLS when the
// system resolver fails, and resolves each discovered hostname via
// net.LookupIP. The returned slice already has each address tagged with
// .IPVersion = V4 or V6.
//
// Note: allregions.EdgeDiscovery must be exported (currently unexported as
// edgeDiscovery) before a production adapter can be wired up.
type DNSResolver interface {
// Resolve performs edge discovery for the given region string (empty for
// global, "us" / "fed" for regional endpoints) and returns the resolved
// addresses grouped by CNAME target, mirroring the structure returned by
// allregions.EdgeDiscovery.
Resolve(region string) ([][]*allregions.EdgeAddr, error)
}
// TCPDialer abstracts the TCP + TLS handshake used by HTTP/2 connectivity probes.
//
// The production implementation wraps edgediscovery.DialEdge
// (edgediscovery/dial.go), which is the same function supervisor/tunnel.go
// uses for production HTTP/2 connections. Reusing it ensures the pre-check
// validates the identical dial path the tunnel will take.
type TCPDialer interface {
// DialEdge dials the given edge TCP address with TLS, respecting the
// provided timeout, and returns the established connection. The caller is
// responsible for closing the connection.
DialEdge(ctx context.Context, timeout time.Duration, tlsConfig *tls.Config, addr *net.TCPAddr, localIP net.IP) (net.Conn, error)
}
// QUICDialer abstracts the UDP + QUIC handshake used by QUIC connectivity probes.
//
// The production implementation wraps connection.DialQuic
// (connection/quic.go), which is the same function supervisor/tunnel.go uses
// for production QUIC connections. The pre-check performs a handshake only —
// no streams are opened and no RPC frames are sent — to avoid triggering the
// OTD registration timeout described in TUN-6732.
type QUICDialer interface {
// DialQuic performs a QUIC handshake to the given edge address and returns
// the established connection. The caller is responsible for closing the
// connection. connIndex is used for UDP port reuse bookkeeping consistent
// with the production dial path.
DialQuic(
ctx context.Context,
quicConfig *quic.Config,
tlsConfig *tls.Config,
addr netip.AddrPort,
localAddr net.IP,
connIndex uint8,
logger *zerolog.Logger,
) (quic.Connection, error)
}
// ManagementDialer abstracts the TCP dial to api.cloudflare.com:443 used by
// the Management API probe.
//
// A successful TCP connection (no TLS handshake required) is sufficient to
// confirm that port 443 is reachable. This probe is always a soft failure:
// the tunnel can run without it, but automatic software updates will be
// unavailable.
type ManagementDialer interface {
// DialContext opens a TCP connection to the given network address. The
// caller is responsible for closing the connection.
DialContext(ctx context.Context, network, addr string) (net.Conn, error)
}
+94
View File
@@ -0,0 +1,94 @@
package prechecks
import (
"time"
"github.com/cloudflare/cloudflared/connection"
"github.com/cloudflare/cloudflared/edgediscovery/allregions"
)
// Status represents the outcome of a single connectivity pre-check.
type Status int
const (
// Pass indicates the check completed successfully.
Pass Status = iota
// Warn indicates a soft failure: cloudflared can still run but in a
// degraded state (e.g. one transport blocked, API unreachable).
Warn
// Fail indicates a check failure that the user should act on (e.g.
// DNS unresolvable, both transports blocked). cloudflared still starts;
// this status is purely informational.
Fail
// Skip indicates the check was not executed because a prerequisite
// check (typically DNS) failed first.
Skip
)
// String returns the canonical display name for a Status value.
func (s Status) String() string {
switch s {
case Pass:
return "PASS"
case Warn:
return "WARN"
case Fail:
return "FAIL"
case Skip:
return "SKIP"
default:
return "UNKNOWN"
}
}
// CheckResult holds the outcome of one individual connectivity probe.
type CheckResult struct {
// Component is the human-readable probe category shown in the table header
// column, e.g. "DNS Resolution", "QUIC Connectivity".
Component string
// Target is the address or resource that was probed, e.g.
// "region1.v2.argotunnel.com" or "Port 7844 (QUIC)".
Target string
// ProbeStatus is the outcome of the probe.
ProbeStatus Status
// Details is a short description of the result shown in the table, e.g.
// "Resolved successfully" or "Handshake failed".
Details string
// Action is non-empty when ProbeStatus is Warn or Fail and contains
// a human-readable remediation instruction, e.g.
// "Allow outbound QUIC on port 7844."
Action string
}
// Report aggregates all CheckResults produced by a single Run() invocation.
// Pre-checks run in parallel with tunnel initialization and are purely
// diagnostic: the Report is displayed to the user but never gates startup.
type Report struct {
// Results contains one entry per executed probe, in the order they were
// collected.
Results []CheckResult
// SuggestedProtocol is the connection protocol the pre-checks recommend
// based on transport probe results.
SuggestedProtocol connection.Protocol
}
// Config controls the behavior of a pre-check Run().
type Config struct {
// Region is the optional cloudflared --region flag value. When non-empty
// the pre-check probes the regional edge hostnames instead of the global ones.
Region string
// Timeout is the maximum wall-clock duration allowed for the entire
// pre-check suite to complete.
Timeout time.Duration
// IPVersion controls which address families are probed for transport
// checks. It mirrors the --edge-ip-version CLI flag so that the pre-check
// exercises the same code paths the tunnel itself will use.
IPVersion allregions.ConfigIPVersion
}