| 1 | //go:build linux |
| 2 | |
| 3 | package host |
| 4 | |
| 5 | import ( |
| 6 | "context" |
| 7 | "errors" |
| 8 | "os" |
| 9 | "os/exec" |
| 10 | "path/filepath" |
| 11 | "strconv" |
| 12 | "strings" |
| 13 | |
| 14 | "github.com/carlosframework/platform/internal/hostbox" |
| 15 | "github.com/carlosframework/platform/internal/instancectl" |
| 16 | "golang.org/x/sys/unix" |
| 17 | ) |
| 18 | |
| 19 | // linuxClaimLaunchKernel is root's production view of a claim launch. Owner |
| 20 | // is the uid the release tree belongs to: root in production. |
| 21 | type linuxClaimLaunchKernel struct { |
| 22 | OptRoot, CgroupRoot, ProcRoot string |
| 23 | Owner uint32 |
| 24 | // BoundApp is the build the unit binds, relative to the unit's root; |
| 25 | // empty is opt/carlos/app. |
| 26 | BoundApp string |
| 27 | } |
| 28 | |
| 29 | // rootOnly is a directory or file only Owner can change. |
| 30 | func (k linuxClaimLaunchKernel) rootOnly(st unix.Stat_t, kind uint32) bool { |
| 31 | return st.Mode&unix.S_IFMT == kind && st.Uid == k.Owner && st.Mode&0o022 == 0 |
| 32 | } |
| 33 | |
| 34 | // Verify accepts a build only inside the release tree, outside its |
| 35 | // retargetable live links and the platform binary, reached with no symlink, |
| 36 | // in directories and a file only Owner can change, with the pinned identity. |
| 37 | func (k linuxClaimLaunchKernel) Verify(path string, want hostbox.LaunchFile) error { |
| 38 | if !claimLaunchAbs(k.OptRoot) || !claimLaunchAbs(path) || !strings.HasPrefix(path, k.OptRoot+"/") || |
| 39 | strings.HasPrefix(path, filepath.Join(k.OptRoot, "live")+"/") || path == filepath.Join(k.OptRoot, "carlos") || want.Device == 0 || want.Inode == 0 { |
| 40 | return hostbox.ErrRoleLaunchShape |
| 41 | } |
| 42 | rel, err := filepath.Rel(k.OptRoot, filepath.Dir(path)) |
| 43 | if err != nil || !filepath.IsLocal(rel) { |
| 44 | return hostbox.ErrRoleLaunchShape |
| 45 | } |
| 46 | dir := k.OptRoot |
| 47 | parts := []string{} |
| 48 | if rel != "." { |
| 49 | parts = strings.Split(rel, "/") |
| 50 | } |
| 51 | for i := -1; i < len(parts); i++ { |
| 52 | if i >= 0 { |
| 53 | dir = filepath.Join(dir, parts[i]) |
| 54 | } |
| 55 | var st unix.Stat_t |
| 56 | if err := unix.Lstat(dir, &st); err != nil { |
| 57 | return err |
| 58 | } |
| 59 | if !k.rootOnly(st, unix.S_IFDIR) { |
| 60 | return hostbox.ErrChanged |
| 61 | } |
| 62 | } |
| 63 | fd, err := unix.Openat2(unix.AT_FDCWD, path, &unix.OpenHow{Flags: unix.O_RDONLY | unix.O_CLOEXEC | unix.O_NOFOLLOW, Resolve: unix.RESOLVE_NO_SYMLINKS | unix.RESOLVE_NO_MAGICLINKS}) |
| 64 | if err != nil { |
| 65 | return err |
| 66 | } |
| 67 | defer unix.Close(fd) |
| 68 | var st unix.Stat_t |
| 69 | if err := unix.Fstat(fd, &st); err != nil { |
| 70 | return err |
| 71 | } |
| 72 | if !k.rootOnly(st, unix.S_IFREG) || st.Mode&0o111 == 0 || uint64(st.Dev) != want.Device || st.Ino != want.Inode { |
| 73 | return hostbox.ErrChanged |
| 74 | } |
| 75 | return nil |
| 76 | } |
| 77 | |
| 78 | func claimLaunchProcessIdentity(procRoot string, pid int) (string, int, error) { |
| 79 | raw, err := os.ReadFile(filepath.Join(procRoot, strconv.Itoa(pid), "stat")) |
| 80 | if err != nil { |
| 81 | return "", 0, err |
| 82 | } |
| 83 | i := strings.LastIndexByte(string(raw), ')') |
| 84 | if i < 0 { |
| 85 | return "", 0, hostbox.ErrChanged |
| 86 | } |
| 87 | fields := strings.Fields(string(raw)[i+1:]) |
| 88 | if len(fields) < 20 || fields[0] == "Z" || fields[0] == "X" { |
| 89 | return "", 0, hostbox.ErrChanged |
| 90 | } |
| 91 | parent, err := strconv.Atoi(fields[1]) |
| 92 | if err != nil { |
| 93 | return "", 0, hostbox.ErrChanged |
| 94 | } |
| 95 | return fields[19], parent, nil |
| 96 | } |
| 97 | |
| 98 | func claimLaunchProcessStart(procRoot string, pid int) (string, error) { |
| 99 | start, _, err := claimLaunchProcessIdentity(procRoot, pid) |
| 100 | return start, err |
| 101 | } |
| 102 | |
| 103 | func (k linuxClaimLaunchKernel) member(pid int, group string) error { |
| 104 | raw, err := os.ReadFile(filepath.Join(k.ProcRoot, strconv.Itoa(pid), "cgroup")) |
| 105 | if err != nil { |
| 106 | return err |
| 107 | } |
| 108 | for _, line := range strings.Split(string(raw), "\n") { |
| 109 | if line == "0::/"+group { |
| 110 | return nil |
| 111 | } |
| 112 | } |
| 113 | return hostbox.ErrChanged |
| 114 | } |
| 115 | |
| 116 | func (k linuxClaimLaunchKernel) tenant(main int) (int, error) { |
| 117 | tasks := filepath.Join(k.ProcRoot, strconv.Itoa(main), "task") |
| 118 | entries, err := os.ReadDir(tasks) |
| 119 | if err != nil { |
| 120 | return 0, err |
| 121 | } |
| 122 | children := map[int]bool{} |
| 123 | for _, entry := range entries { |
| 124 | raw, err := os.ReadFile(filepath.Join(tasks, entry.Name(), "children")) |
| 125 | if errors.Is(err, os.ErrNotExist) { |
| 126 | continue |
| 127 | } |
| 128 | if err != nil { |
| 129 | return 0, err |
| 130 | } |
| 131 | for _, word := range strings.Fields(string(raw)) { |
| 132 | pid, err := strconv.Atoi(word) |
| 133 | if err != nil || pid <= 0 || pid == main { |
| 134 | return 0, hostbox.ErrChanged |
| 135 | } |
| 136 | children[pid] = true |
| 137 | } |
| 138 | } |
| 139 | if len(children) == 0 { |
| 140 | return 0, errClaimLaunchPending |
| 141 | } |
| 142 | if len(children) != 1 { |
| 143 | return 0, hostbox.ErrChanged |
| 144 | } |
| 145 | for pid := range children { |
| 146 | return pid, nil |
| 147 | } |
| 148 | return 0, hostbox.ErrChanged |
| 149 | } |
| 150 | |
| 151 | // Observe keeps the main process as the unit identity and verifies its one |
| 152 | // direct tenant child. runInit forks on a locked thread, so every task's |
| 153 | // children file participates; the leader's children file alone is incomplete. |
| 154 | func (k linuxClaimLaunchKernel) Observe(s instancectl.ClaimUnitStarted) (ClaimLaunchObserved, error) { |
| 155 | if s.MainPID <= 0 || !filepath.IsLocal(s.ControlGroup) || !claimLaunchAbs(k.CgroupRoot) || !claimLaunchAbs(k.ProcRoot) { |
| 156 | return ClaimLaunchObserved{}, hostbox.ErrChanged |
| 157 | } |
| 158 | start, err := claimLaunchProcessStart(k.ProcRoot, s.MainPID) |
| 159 | if err != nil { |
| 160 | return ClaimLaunchObserved{}, err |
| 161 | } |
| 162 | var group unix.Stat_t |
| 163 | if err = unix.Lstat(filepath.Join(k.CgroupRoot, s.ControlGroup), &group); err != nil { |
| 164 | return ClaimLaunchObserved{}, err |
| 165 | } |
| 166 | if group.Mode&unix.S_IFMT != unix.S_IFDIR { |
| 167 | return ClaimLaunchObserved{}, hostbox.ErrChanged |
| 168 | } |
| 169 | var launcher unix.Stat_t |
| 170 | if err := unix.Stat(filepath.Join(k.ProcRoot, strconv.Itoa(s.MainPID), "exe"), &launcher); err != nil { |
| 171 | return ClaimLaunchObserved{}, err |
| 172 | } |
| 173 | observed := ClaimLaunchObserved{Launcher: hostbox.LaunchFile{Device: uint64(launcher.Dev), Inode: launcher.Ino}, ProcessStart: start, Group: hostbox.LaunchFile{Device: uint64(group.Dev), Inode: group.Ino}} |
| 174 | if err := k.member(s.MainPID, s.ControlGroup); err != nil { |
| 175 | return observed, err |
| 176 | } |
| 177 | child, childErr := k.tenant(s.MainPID) |
| 178 | if childErr != nil { |
| 179 | again, err := claimLaunchProcessStart(k.ProcRoot, s.MainPID) |
| 180 | if err != nil || again != start { |
| 181 | return observed, hostbox.ErrChanged |
| 182 | } |
| 183 | return observed, childErr |
| 184 | } |
| 185 | childStart, parent, err := claimLaunchProcessIdentity(k.ProcRoot, child) |
| 186 | if err != nil || parent != s.MainPID { |
| 187 | return observed, hostbox.ErrChanged |
| 188 | } |
| 189 | observed.TenantPID, observed.TenantStart = child, childStart |
| 190 | if err := k.member(child, s.ControlGroup); err != nil { |
| 191 | return observed, err |
| 192 | } |
| 193 | bound := k.BoundApp |
| 194 | if bound == "" { |
| 195 | bound = "opt/carlos/app" |
| 196 | } |
| 197 | var app, mainApp, running unix.Stat_t |
| 198 | boundErr := unix.Stat(filepath.Join(k.ProcRoot, strconv.Itoa(child), "root", bound), &app) |
| 199 | mainBoundErr := unix.Stat(filepath.Join(k.ProcRoot, strconv.Itoa(s.MainPID), "root", bound), &mainApp) |
| 200 | execErr := unix.Stat(filepath.Join(k.ProcRoot, strconv.Itoa(child), "exe"), &running) |
| 201 | againChild, parent, err := claimLaunchProcessIdentity(k.ProcRoot, child) |
| 202 | if err != nil || againChild != childStart || parent != s.MainPID { |
| 203 | return observed, hostbox.ErrChanged |
| 204 | } |
| 205 | again, err := claimLaunchProcessStart(k.ProcRoot, s.MainPID) |
| 206 | if err != nil || again != start { |
| 207 | return observed, hostbox.ErrChanged |
| 208 | } |
| 209 | if err := k.member(s.MainPID, s.ControlGroup); err != nil { |
| 210 | return observed, err |
| 211 | } |
| 212 | if err := k.member(child, s.ControlGroup); err != nil { |
| 213 | return observed, err |
| 214 | } |
| 215 | againPID, err := k.tenant(s.MainPID) |
| 216 | if err != nil || againPID != child { |
| 217 | return observed, hostbox.ErrChanged |
| 218 | } |
| 219 | for _, err := range []error{boundErr, mainBoundErr} { |
| 220 | if errors.Is(err, os.ErrNotExist) { |
| 221 | return observed, errClaimLaunchPending |
| 222 | } |
| 223 | if err != nil { |
| 224 | return observed, err |
| 225 | } |
| 226 | } |
| 227 | if app.Dev != mainApp.Dev || app.Ino != mainApp.Ino { |
| 228 | return observed, hostbox.ErrChanged |
| 229 | } |
| 230 | if execErr != nil { |
| 231 | return observed, execErr |
| 232 | } |
| 233 | observed.Executable = hostbox.LaunchFile{Device: uint64(app.Dev), Inode: app.Ino} |
| 234 | observed.Running = hostbox.LaunchFile{Device: uint64(running.Dev), Inode: running.Ino} |
| 235 | return observed, nil |
| 236 | } |
| 237 | |
| 238 | func (k linuxClaimLaunchKernel) VerifyDigest(ctx context.Context, path string, want hostbox.LaunchFile, digest string) error { |
| 239 | if err := k.Verify(path, want); err != nil { |
| 240 | return err |
| 241 | } |
| 242 | f, err := verifiedClaimExecutable(ctx, exec.Command(path), path, digest) |
| 243 | if err != nil { |
| 244 | return err |
| 245 | } |
| 246 | defer f.Close() |
| 247 | var st unix.Stat_t |
| 248 | if err := unix.Fstat(int(f.Fd()), &st); err != nil { |
| 249 | return err |
| 250 | } |
| 251 | if uint64(st.Dev) != want.Device || st.Ino != want.Inode { |
| 252 | return hostbox.ErrChanged |
| 253 | } |
| 254 | return nil |
| 255 | } |
| 256 | |