| 1 | //go:build linux |
| 2 | |
| 3 | package instancectl |
| 4 | |
| 5 | import ( |
| 6 | "context" |
| 7 | "errors" |
| 8 | "fmt" |
| 9 | "log" |
| 10 | "net" |
| 11 | "os" |
| 12 | "os/exec" |
| 13 | "os/user" |
| 14 | "strconv" |
| 15 | "strings" |
| 16 | "sync" |
| 17 | "syscall" |
| 18 | "time" |
| 19 | |
| 20 | "golang.org/x/sys/unix" |
| 21 | ) |
| 22 | |
| 23 | // peerCred reads SO_PEERCRED from a connected unix socket. |
| 24 | func peerCred(conn net.Conn) (*unix.Ucred, error) { |
| 25 | uc, ok := conn.(*net.UnixConn) |
| 26 | if !ok { |
| 27 | return nil, errors.New("not a unix connection") |
| 28 | } |
| 29 | raw, err := uc.SyscallConn() |
| 30 | if err != nil { |
| 31 | return nil, err |
| 32 | } |
| 33 | var cred *unix.Ucred |
| 34 | var cerr error |
| 35 | if err := raw.Control(func(fd uintptr) { |
| 36 | cred, cerr = unix.GetsockoptUcred(int(fd), unix.SOL_SOCKET, unix.SO_PEERCRED) |
| 37 | }); err != nil { |
| 38 | return nil, err |
| 39 | } |
| 40 | return cred, cerr |
| 41 | } |
| 42 | |
| 43 | // productionPeerOK requires the peer to be uid carlos AND the main process |
| 44 | // of carlos-edge.service as systemd reports it (design §5.4): a flat tenant |
| 45 | // is a child of that pid, never the pid itself, and a nested tenant cannot |
| 46 | // reach the socket at all. MainPID is asked of systemctl and cached for one |
| 47 | // second, because the agent restarts and the pid changes. |
| 48 | func productionPeerOK() func(net.Conn) error { |
| 49 | // A missing carlos user or an unparsable uid is fatal at startup, not a |
| 50 | // permanent refusal cached into the returned closure: socket activation |
| 51 | // restarts this service on the very next connection attempt, so a fix |
| 52 | // to the box (the user provisioned, say) takes effect on the next |
| 53 | // accept instead of requiring every already-running process to be |
| 54 | // found and killed by hand (final review, plan 2, item 6). |
| 55 | u, err := user.Lookup("carlos") |
| 56 | if err != nil { |
| 57 | log.Fatalf("instance-control: no carlos user: %v", err) |
| 58 | } |
| 59 | carlosUID, err := strconv.Atoi(u.Uid) |
| 60 | if err != nil { |
| 61 | log.Fatalf("instance-control: carlos uid %q: %v", u.Uid, err) |
| 62 | } |
| 63 | var mu sync.Mutex |
| 64 | var mainPID int |
| 65 | var lastErr error |
| 66 | var at time.Time |
| 67 | edgePID := func() (int, error) { |
| 68 | mu.Lock() |
| 69 | defer mu.Unlock() |
| 70 | // Cache both a success and a failure for one second: while the |
| 71 | // edge has no MainPID yet (a restart window), every refused |
| 72 | // connect would otherwise re-exec systemctl under this mutex — |
| 73 | // the negative case needs the same cache the positive one always |
| 74 | // had (codex review fix round 2 item 5). |
| 75 | if time.Since(at) < time.Second { |
| 76 | if mainPID > 0 { |
| 77 | return mainPID, nil |
| 78 | } |
| 79 | if lastErr != nil { |
| 80 | return 0, lastErr |
| 81 | } |
| 82 | } |
| 83 | ctx, cancel := context.WithTimeout(context.Background(), 3*time.Second) |
| 84 | defer cancel() |
| 85 | out, err := exec.CommandContext(ctx, "systemctl", "show", "-p", "MainPID", "--value", edgeUnit).Output() |
| 86 | if err != nil { |
| 87 | mainPID, lastErr, at = 0, err, time.Now() |
| 88 | return 0, err |
| 89 | } |
| 90 | pid, err := strconv.Atoi(strings.TrimSpace(string(out))) |
| 91 | if err != nil || pid <= 0 { |
| 92 | err = fmt.Errorf("carlos-edge.service has no MainPID") |
| 93 | mainPID, lastErr, at = 0, err, time.Now() |
| 94 | return 0, err |
| 95 | } |
| 96 | mainPID, lastErr, at = pid, nil, time.Now() |
| 97 | return pid, nil |
| 98 | } |
| 99 | return func(conn net.Conn) error { |
| 100 | cred, err := peerCred(conn) |
| 101 | if err != nil { |
| 102 | return err |
| 103 | } |
| 104 | if int(cred.Uid) != carlosUID { |
| 105 | return fmt.Errorf("peer uid %d is not carlos", cred.Uid) |
| 106 | } |
| 107 | pid, err := edgePID() |
| 108 | if err != nil { |
| 109 | return err |
| 110 | } |
| 111 | if int(cred.Pid) != pid { |
| 112 | return fmt.Errorf("peer pid %d is not the edge's main pid %d", cred.Pid, pid) |
| 113 | } |
| 114 | return nil |
| 115 | } |
| 116 | } |
| 117 | |
| 118 | // activatedListener returns the socket-activated listener when systemd |
| 119 | // passed one (LISTEN_PID is us, LISTEN_FDS is 1, the fd is 3). |
| 120 | func activatedListener() (net.Listener, bool, error) { |
| 121 | if os.Getenv("LISTEN_PID") != strconv.Itoa(os.Getpid()) { |
| 122 | return nil, false, nil |
| 123 | } |
| 124 | if os.Getenv("LISTEN_FDS") != "1" { |
| 125 | return nil, false, fmt.Errorf("LISTEN_FDS=%q, want 1", os.Getenv("LISTEN_FDS")) |
| 126 | } |
| 127 | f := os.NewFile(3, "control.sock") |
| 128 | defer f.Close() |
| 129 | ln, err := net.FileListener(f) |
| 130 | if err != nil { |
| 131 | return nil, false, err |
| 132 | } |
| 133 | return ln, true, nil |
| 134 | } |
| 135 | |
| 136 | // bindControlSocket is the hand-run fallback: bind the path with the |
| 137 | // ownership the socket unit would give it (carlos, 0600). |
| 138 | func bindControlSocket(path string) (net.Listener, error) { |
| 139 | _ = os.Remove(path) |
| 140 | ln, err := net.Listen("unix", path) |
| 141 | if err != nil { |
| 142 | return nil, err |
| 143 | } |
| 144 | u, err := user.Lookup("carlos") |
| 145 | if err != nil { |
| 146 | ln.Close() |
| 147 | return nil, err |
| 148 | } |
| 149 | uid, _ := strconv.Atoi(u.Uid) |
| 150 | gid, _ := strconv.Atoi(u.Gid) |
| 151 | if err := os.Chown(path, uid, gid); err != nil { |
| 152 | ln.Close() |
| 153 | return nil, err |
| 154 | } |
| 155 | if err := os.Chmod(path, 0o600); err != nil { |
| 156 | ln.Close() |
| 157 | return nil, err |
| 158 | } |
| 159 | return ln, nil |
| 160 | } |
| 161 | |
| 162 | // probeFixturesOK checks the two probe fixtures tmpfiles stages: the |
| 163 | // secret the probe must not read, and the host /dev/shm file it must not |
| 164 | // see. Either one missing would make its check pass vacuously. |
| 165 | func probeFixturesOK() error { |
| 166 | shm, err := os.Lstat("/dev/shm/carlos-probe") |
| 167 | if err != nil { |
| 168 | return fmt.Errorf("probe shm fixture absent: %v", err) |
| 169 | } |
| 170 | if sst, ok := shm.Sys().(*syscall.Stat_t); !ok || sst.Uid != 0 || !shm.Mode().IsRegular() { |
| 171 | return errors.New("probe shm fixture: not a root-owned regular file") |
| 172 | } |
| 173 | fi, err := os.Stat("/etc/carlos/probe-secret") |
| 174 | if err != nil { |
| 175 | return fmt.Errorf("probe fixture absent: %v", err) |
| 176 | } |
| 177 | st, ok := fi.Sys().(*syscall.Stat_t) |
| 178 | if !ok { |
| 179 | return errors.New("probe fixture: no stat") |
| 180 | } |
| 181 | g, err := user.LookupGroup("carlos") |
| 182 | if err != nil { |
| 183 | return err |
| 184 | } |
| 185 | gid, _ := strconv.Atoi(g.Gid) |
| 186 | if st.Uid != 0 || int(st.Gid) != gid || fi.Mode().Perm() != 0o640 { |
| 187 | return fmt.Errorf("probe fixture is uid %d gid %d mode %o, want root:carlos 0640", st.Uid, st.Gid, fi.Mode().Perm()) |
| 188 | } |
| 189 | return nil |
| 190 | } |
| 191 | |
| 192 | // statfsBytes is the free space an unprivileged writer can use (f_bavail, |
| 193 | // so root's reserved blocks do not count) and the filesystem's size. |
| 194 | func statfsBytes(path string) (avail, total uint64, err error) { |
| 195 | var st unix.Statfs_t |
| 196 | if err := unix.Statfs(path, &st); err != nil { |
| 197 | return 0, 0, err |
| 198 | } |
| 199 | bs := uint64(st.Bsize) |
| 200 | return st.Bavail * bs, st.Blocks * bs, nil |
| 201 | } |
| 202 | |
amadan