| 1 | // Package unixdial is the one way the edge and the agent dial a unix |
| 2 | // socket. Two checks ride on every dial (isolated-exec design §5.4): |
| 3 | // |
| 4 | // - the reserved control endpoint is refused BEFORE connect(): a registry |
| 5 | // row pointing rt.Socket at the root control channel must never be |
| 6 | // proxied to, however it got there; |
| 7 | // - the listener must answer to the path that was dialled: after |
| 8 | // connect() and before a byte is sent, getpeername() names the path |
| 9 | // the LISTENER bound, which is its real path however it was reached. |
| 10 | // A symlink at /run/carlos/<host>.sock pointing at a sibling's socket |
| 11 | // or at the control socket therefore yields a mismatch, and the |
| 12 | // connection is closed unused. |
| 13 | // |
| 14 | // The first check is purely lexical: it filepath.Cleans the dialled |
| 15 | // string and compares it to the control directory, before any connect. |
| 16 | // A relative path, or a symlinked parent directory that only resolves to |
| 17 | // the control directory at connect time, is not caught here — it is |
| 18 | // still caught by the second check, once getpeername() names the |
| 19 | // control listener's real bound path after connect. A listener bound by |
| 20 | // a relative path is refused by the second check for the same reason: |
| 21 | // the kernel reports the bind string verbatim, and a relative name can |
| 22 | // never filepath.Clean-equal the absolute path Checked was dialled with. |
| 23 | // |
| 24 | // Credentials are deliberately not consulted: a listener systemd created |
| 25 | // and passed to a DynamicUser service reports root in SO_PEERCRED, and |
| 26 | // that shape (carlos-app@.socket) must keep working (codex R4-1). |
| 27 | package unixdial |
| 28 | |
| 29 | import ( |
| 30 | "context" |
| 31 | "errors" |
| 32 | "fmt" |
| 33 | "net" |
| 34 | "path/filepath" |
| 35 | "strings" |
| 36 | |
| 37 | "github.com/carlosframework/platform/internal/controlsock" |
| 38 | ) |
| 39 | |
| 40 | // ErrControlEndpoint is returned, before any connect, for a path under |
| 41 | // provision.ControlSocketDir (internal/controlsock.Dir; provision |
| 42 | // re-exports it — see internal/controlsock's doc comment for why this |
| 43 | // package imports controlsock directly rather than provision). |
| 44 | var ErrControlEndpoint = errors.New("unixdial: refusing to dial the instance control channel as an upstream") |
| 45 | |
| 46 | // ErrPeerMismatch is returned, after connect and before any byte is sent, |
| 47 | // when the listener's bound path is not the path that was dialled. |
| 48 | var ErrPeerMismatch = errors.New("unixdial: listener is bound to a different path than the one dialled") |
| 49 | |
| 50 | // Checked dials path as a unix stream socket with d and applies both |
| 51 | // checks. Connect errors (ENOENT, ECONNREFUSED) are returned unwrapped so |
| 52 | // callers' retry logic keeps seeing them. |
| 53 | func Checked(ctx context.Context, d *net.Dialer, path string) (net.Conn, error) { |
| 54 | if controlsock.IsEndpoint(path) { |
| 55 | return nil, fmt.Errorf("%w: %s", ErrControlEndpoint, path) |
| 56 | } |
| 57 | if d == nil { |
| 58 | d = &net.Dialer{} |
| 59 | } |
| 60 | conn, err := d.DialContext(ctx, "unix", path) |
| 61 | if err != nil { |
| 62 | return nil, err |
| 63 | } |
| 64 | if err := verifyPeer(conn, path); err != nil { |
| 65 | conn.Close() |
| 66 | return nil, err |
| 67 | } |
| 68 | return conn, nil |
| 69 | } |
| 70 | |
| 71 | // verifyPeer compares the peer address of a connected unix socket with the |
| 72 | // dialled path. net's UnixConn fills RemoteAddr from getpeername(2) at |
| 73 | // connect time, which is the listener's bound address. |
| 74 | // |
| 75 | // Only the DIALLED side is cleaned (want, below); the peer name the kernel |
| 76 | // reports must equal want BYTE FOR BYTE. Cleaning got as well would let a |
| 77 | // listener whose bound path merely normalizes to the same string (a |
| 78 | // different-but-equivalent spelling of it) pass as a match, which is more |
| 79 | // than getpeername() actually promised: the kernel reports the bind string |
| 80 | // verbatim, not a normalized one, so an exact match is the only comparison |
| 81 | // that means what this check claims to mean. A peer name containing a ".." |
| 82 | // path component is refused outright for the same reason a relative bind |
| 83 | // path already is: it is never a path this package's callers bind by, so |
| 84 | // its presence means the socket was reached some other way than intended. |
| 85 | func verifyPeer(conn net.Conn, dialled string) error { |
| 86 | ra := conn.RemoteAddr() |
| 87 | ua, ok := ra.(*net.UnixAddr) |
| 88 | if !ok || ua == nil { |
| 89 | return fmt.Errorf("%w: peer address %v is not a unix address", ErrPeerMismatch, ra) |
| 90 | } |
| 91 | want := filepath.Clean(dialled) |
| 92 | got := ua.Name |
| 93 | if got == "" || got[0] == '@' || hasDotDotComponent(got) || got != want { |
| 94 | return fmt.Errorf("%w: dialled %s, listener is bound to %q", ErrPeerMismatch, want, got) |
| 95 | } |
| 96 | return nil |
| 97 | } |
| 98 | |
| 99 | // hasDotDotComponent reports whether name has a ".." path component, |
| 100 | // checked without cleaning it first: cleaning would resolve the very |
| 101 | // traversal this exists to catch before it could ever be seen. |
| 102 | func hasDotDotComponent(name string) bool { |
| 103 | for _, part := range strings.Split(name, string(filepath.Separator)) { |
| 104 | if part == ".." { |
| 105 | return true |
| 106 | } |
| 107 | } |
| 108 | return false |
| 109 | } |
| 110 | |
amadan