| 1 | //go:build linux |
| 2 | |
| 3 | package hostexec |
| 4 | |
| 5 | import ( |
| 6 | "context" |
| 7 | "crypto/sha256" |
| 8 | "encoding/hex" |
| 9 | "encoding/json" |
| 10 | "errors" |
| 11 | "fmt" |
| 12 | "os" |
| 13 | "path/filepath" |
| 14 | "strconv" |
| 15 | "strings" |
| 16 | "syscall" |
| 17 | |
| 18 | "github.com/carlosframework/platform/internal/hostfloor" |
| 19 | "github.com/carlosframework/platform/internal/hostguard" |
| 20 | "golang.org/x/sys/unix" |
| 21 | ) |
| 22 | |
| 23 | func controllerProcessStart(pid int) (string, error) { |
| 24 | if pid <= 0 { |
| 25 | return "", ErrState |
| 26 | } |
| 27 | raw, err := os.ReadFile(fmt.Sprintf("/proc/%d/stat", pid)) |
| 28 | if err != nil { |
| 29 | return "", err |
| 30 | } |
| 31 | end := strings.LastIndexByte(string(raw), ')') |
| 32 | if end < 0 { |
| 33 | return "", ErrState |
| 34 | } |
| 35 | fields := strings.Fields(string(raw[end+1:])) |
| 36 | if len(fields) < 20 { |
| 37 | return "", ErrState |
| 38 | } |
| 39 | start := fields[19] |
| 40 | if _, err = strconv.ParseUint(start, 10, 64); err != nil { |
| 41 | return "", ErrState |
| 42 | } |
| 43 | return start, nil |
| 44 | } |
| 45 | func currentControllerProcess() (ControllerProcess, error) { |
| 46 | boot, err := BootID() |
| 47 | if err != nil { |
| 48 | return ControllerProcess{}, err |
| 49 | } |
| 50 | start, err := controllerProcessStart(os.Getpid()) |
| 51 | if err != nil { |
| 52 | return ControllerProcess{}, err |
| 53 | } |
| 54 | return ControllerProcess{KernelBootID: boot, PID: os.Getpid(), Start: start}, nil |
| 55 | } |
| 56 | func proveControllerDead(p ControllerProcess, boot string, probe func(int) (string, error)) error { |
| 57 | if !validID(p.KernelBootID) || p.PID <= 0 || p.Start == "" || !validID(boot) { |
| 58 | return ErrState |
| 59 | } |
| 60 | if n, err := strconv.ParseUint(p.Start, 10, 64); err != nil || n == 0 || strconv.FormatUint(n, 10) != p.Start { |
| 61 | return ErrState |
| 62 | } |
| 63 | if p.KernelBootID != boot { |
| 64 | return nil |
| 65 | } |
| 66 | start, err := probe(p.PID) |
| 67 | if errors.Is(err, os.ErrNotExist) { |
| 68 | return nil |
| 69 | } |
| 70 | if err != nil { |
| 71 | return err |
| 72 | } |
| 73 | if start == "" || start == p.Start { |
| 74 | return ErrControllerBusy |
| 75 | } |
| 76 | return nil |
| 77 | } |
| 78 | |
| 79 | // RecoverPending is root's offline update recovery. It never clears active |
| 80 | // or lost state and never removes remote participation or claim authority. |
| 81 | func RecoverPending(ctx context.Context, dataDir, guardDir string) error { |
| 82 | if os.Geteuid() != 0 { |
| 83 | return os.ErrPermission |
| 84 | } |
| 85 | return recoverPending(ctx, hostguard.Guard{DataDir: dataDir}, "/run/carlos/hostexec", filepath.Join(guardDir, "settled-controllers"), BootID, controllerProcessStart) |
| 86 | } |
| 87 | func recoverPending(ctx context.Context, g hostguard.Guard, lockDir, archive string, boot func() (string, error), probe func(int) (string, error)) error { |
| 88 | d, err := openLocal(g.DataDir) |
| 89 | if err != nil { |
| 90 | return err |
| 91 | } |
| 92 | defer d.close() |
| 93 | fd, err := unix.Dup(d.fd) |
| 94 | if err != nil { |
| 95 | return err |
| 96 | } |
| 97 | f := os.NewFile(uintptr(fd), "controllers") |
| 98 | entries, err := f.ReadDir(-1) |
| 99 | _ = f.Close() |
| 100 | if err != nil { |
| 101 | return err |
| 102 | } |
| 103 | if len(entries) == 0 { |
| 104 | return nil |
| 105 | } |
| 106 | kernelBoot, err := boot() |
| 107 | if err != nil { |
| 108 | return err |
| 109 | } |
| 110 | for _, e := range entries { |
| 111 | if err = ctx.Err(); err != nil { |
| 112 | return err |
| 113 | } |
| 114 | if !strings.HasSuffix(e.Name(), ".json") { |
| 115 | return ErrState |
| 116 | } |
| 117 | host := strings.TrimSuffix(e.Name(), ".json") |
| 118 | err = g.WithLocal(ctx, host, func() error { |
| 119 | unlock, err := tryExistingPendingLock(lockDir, host, ".lock") |
| 120 | if err != nil { |
| 121 | return err |
| 122 | } |
| 123 | defer unlock() |
| 124 | unlive, err := tryExistingPendingLock(lockDir, host, ".live") |
| 125 | if err != nil { |
| 126 | return err |
| 127 | } |
| 128 | defer unlive() |
| 129 | old, err := d.read(host) |
| 130 | if err != nil { |
| 131 | return err |
| 132 | } |
| 133 | if old.State != "pending" { |
| 134 | return ErrState |
| 135 | } |
| 136 | if err = proveControllerDead(old.Process, kernelBoot, probe); err != nil { |
| 137 | return err |
| 138 | } |
| 139 | // Keep the complete original tuple durably before removing its denial. |
| 140 | // A crash after this write replays the same content-addressed evidence. |
| 141 | raw, err := json.Marshal(old) |
| 142 | if err != nil { |
| 143 | return err |
| 144 | } |
| 145 | sum := sha256.Sum256(raw) |
| 146 | if err = os.Mkdir(archive, 0700); err != nil && !errors.Is(err, os.ErrExist) { |
| 147 | return err |
| 148 | } |
| 149 | info, err := os.Lstat(archive) |
| 150 | if err != nil { |
| 151 | return err |
| 152 | } |
| 153 | stat, ok := info.Sys().(*syscall.Stat_t) |
| 154 | if !ok || stat.Uid != uint32(os.Geteuid()) || !info.IsDir() || info.Mode().Perm() != 0700 { |
| 155 | return ErrState |
| 156 | } |
| 157 | parent, err := os.Open(filepath.Dir(archive)) |
| 158 | if err != nil { |
| 159 | return err |
| 160 | } |
| 161 | syncErr := parent.Sync() |
| 162 | closeErr := parent.Close() |
| 163 | if err = errors.Join(syncErr, closeErr); err != nil { |
| 164 | return err |
| 165 | } |
| 166 | if err = hostfloor.Durable(filepath.Join(archive, hex.EncodeToString(sum[:])+".json"), raw, 0600); err != nil { |
| 167 | return err |
| 168 | } |
| 169 | current, err := d.read(host) |
| 170 | if err != nil { |
| 171 | return err |
| 172 | } |
| 173 | currentRaw, err := json.Marshal(current) |
| 174 | if err != nil { |
| 175 | return err |
| 176 | } |
| 177 | if string(currentRaw) != string(raw) { |
| 178 | return ErrState |
| 179 | } |
| 180 | return d.remove(host) |
| 181 | }) |
| 182 | if err != nil { |
| 183 | return err |
| 184 | } |
| 185 | } |
| 186 | return nil |
| 187 | } |
| 188 | |
| 189 | // The enclosing host guard prevents a cooperating Open from creating these |
| 190 | // names until recovery finishes. Root must never create agent lock files. |
| 191 | func tryExistingPendingLock(dir, host, suffix string) (func(), error) { |
| 192 | if _, err := Key(host, "check"); err != nil { |
| 193 | return nil, err |
| 194 | } |
| 195 | fd, err := openDir(dir) |
| 196 | if errors.Is(err, os.ErrNotExist) { |
| 197 | return func() {}, nil |
| 198 | } |
| 199 | if err != nil { |
| 200 | return nil, err |
| 201 | } |
| 202 | defer unix.Close(fd) |
| 203 | lock, err := unix.Openat(fd, host+suffix, unix.O_RDWR|unix.O_NOFOLLOW|unix.O_CLOEXEC|unix.O_NONBLOCK, 0) |
| 204 | if errors.Is(err, os.ErrNotExist) { |
| 205 | return func() {}, nil |
| 206 | } |
| 207 | if err != nil { |
| 208 | return nil, err |
| 209 | } |
| 210 | fail := func(e error) (func(), error) { unix.Close(lock); return nil, e } |
| 211 | var st unix.Stat_t |
| 212 | if err = unix.Fstat(lock, &st); err != nil { |
| 213 | return fail(err) |
| 214 | } |
| 215 | if st.Mode&unix.S_IFMT != unix.S_IFREG || st.Nlink != 1 { |
| 216 | return fail(ErrState) |
| 217 | } |
| 218 | if err = unix.Flock(lock, unix.LOCK_EX|unix.LOCK_NB); err != nil { |
| 219 | if errors.Is(err, unix.EWOULDBLOCK) || errors.Is(err, unix.EAGAIN) { |
| 220 | return fail(ErrControllerBusy) |
| 221 | } |
| 222 | return fail(err) |
| 223 | } |
| 224 | return func() { _ = unix.Flock(lock, unix.LOCK_UN); _ = unix.Close(lock) }, nil |
| 225 | } |
| 226 | |
amadan