| 1 | package moves |
| 2 | |
| 3 | import ( |
| 4 | "bytes" |
| 5 | "encoding/json" |
| 6 | "errors" |
| 7 | "fmt" |
| 8 | "io/fs" |
| 9 | "os" |
| 10 | "path/filepath" |
| 11 | "strings" |
| 12 | "sync" |
| 13 | "time" |
| 14 | ) |
| 15 | |
| 16 | // journalVersion is the only journal format this code reads or writes. A |
| 17 | // file with any other version is refused, never reinterpreted. |
| 18 | const journalVersion = 1 |
| 19 | |
| 20 | const journalFile = "journal.json" |
| 21 | |
| 22 | // Journal is the box-local record of how far one move operation has got on |
| 23 | // this box (§9.1, §9.2). A step that is Done is never redone after a |
| 24 | // restart; the value a step was marked with is whatever that step needs to |
| 25 | // resume (a digest, an id, a timestamp). Values are internal state, not |
| 26 | // operator copy. |
| 27 | type Journal struct { |
| 28 | V int `json:"v"` |
| 29 | Host string `json:"host"` |
| 30 | Op string `json:"op"` |
| 31 | Phase string `json:"phase,omitempty"` |
| 32 | Steps map[string]string `json:"steps,omitempty"` |
| 33 | Attempt int `json:"attempt,omitempty"` |
| 34 | UpdatedAt time.Time `json:"updated_at"` |
| 35 | |
| 36 | // changes are what this value changed since it was read or last |
| 37 | // committed (Mark, Unmark), in order, and loadedPhase and loadedAttempt |
| 38 | // the Phase and Attempt it was read with: what Commit applies. |
| 39 | changes []journalChange |
| 40 | loadedPhase string |
| 41 | loadedAttempt int |
| 42 | } |
| 43 | |
| 44 | // journalChange is one Mark (del false) or Unmark of a step. |
| 45 | type journalChange struct { |
| 46 | step, value string |
| 47 | del bool |
| 48 | } |
| 49 | |
| 50 | // A move's journal has more than one writer on a box: the pass that runs the |
| 51 | // phase, the probe owners and attempts (Plan 6 fix round 1, ruling 3). Every |
| 52 | // read-modify-write of one journal goes through UpdateJournal, which holds |
| 53 | // that journal's lock, the one lock every writer in this process shares |
| 54 | // (journalLock), across the read, the change and the write (fix round 2, |
| 55 | // ruling 2). There is no plain write: a writer that read the journal earlier |
| 56 | // commits its own changes (Commit), which UpdateJournal applies to the |
| 57 | // journal as it is then, so nothing the writer read and did not change is |
| 58 | // ever written back over another writer's. |
| 59 | var journalLocks sync.Map // the journal's directory -> *sync.Mutex |
| 60 | |
| 61 | // journalLock is the lock of the journal of (host, op) under dataDir. |
| 62 | func journalLock(dataDir, host, op string) *sync.Mutex { |
| 63 | l, _ := journalLocks.LoadOrStore(filepath.Clean(JournalDir(dataDir, host, op)), &sync.Mutex{}) |
| 64 | return l.(*sync.Mutex) |
| 65 | } |
| 66 | |
| 67 | // UpdateJournal reads the journal of (host, op), calls fn with it and writes |
| 68 | // it, all under the journal's lock. It makes the journal when there is none. |
| 69 | // An error from fn writes nothing and is returned. fn must not call Commit. |
| 70 | func UpdateJournal(dataDir, host, op string, fn func(*Journal) error) error { |
| 71 | if err := journalNames(host, op); err != nil { |
| 72 | return err |
| 73 | } |
| 74 | l := journalLock(dataDir, host, op) |
| 75 | l.Lock() |
| 76 | defer l.Unlock() |
| 77 | j, err := ReadJournal(dataDir, host, op) |
| 78 | if err != nil { |
| 79 | return err |
| 80 | } |
| 81 | if err := fn(j); err != nil { |
| 82 | return err |
| 83 | } |
| 84 | return j.write(dataDir) |
| 85 | } |
| 86 | |
| 87 | // Commit makes j's own changes durable: through UpdateJournal, on the |
| 88 | // journal as it is on disk now, it applies what j changed since it was read |
| 89 | // or last committed (its Marks and Unmarks in order, and Phase and Attempt |
| 90 | // if j changed them) and nothing else, and j becomes the journal as written. |
| 91 | // A step another writer advanced or deleted after j was read stays as that |
| 92 | // writer left it unless j itself marks it. |
| 93 | func (j *Journal) Commit(dataDir string) error { |
| 94 | var now *Journal |
| 95 | err := UpdateJournal(dataDir, j.Host, j.Op, func(d *Journal) error { |
| 96 | for _, c := range j.changes { |
| 97 | if c.del { |
| 98 | delete(d.Steps, c.step) |
| 99 | } else { |
| 100 | d.Steps[c.step] = c.value |
| 101 | } |
| 102 | } |
| 103 | if j.Phase != j.loadedPhase { |
| 104 | d.Phase = j.Phase |
| 105 | } |
| 106 | if j.Attempt != j.loadedAttempt { |
| 107 | d.Attempt = j.Attempt |
| 108 | } |
| 109 | now = d |
| 110 | return nil |
| 111 | }) |
| 112 | if err != nil { |
| 113 | return err |
| 114 | } |
| 115 | j.V, j.Phase, j.Attempt, j.UpdatedAt = now.V, now.Phase, now.Attempt, now.UpdatedAt |
| 116 | j.Steps = make(map[string]string, len(now.Steps)) |
| 117 | for k, v := range now.Steps { |
| 118 | j.Steps[k] = v |
| 119 | } |
| 120 | j.changes, j.loadedPhase, j.loadedAttempt = nil, j.Phase, j.Attempt |
| 121 | return nil |
| 122 | } |
| 123 | |
| 124 | // JournalDir is where one operation's journal lives on the box: |
| 125 | // <dataDir>/.moves/<host>/<op>/. |
| 126 | func JournalDir(dataDir, host, op string) string { |
| 127 | return filepath.Join(dataDir, ".moves", host, op) |
| 128 | } |
| 129 | |
| 130 | // journalName refuses a host or operation id that is empty or could leave |
| 131 | // its directory. Both come from console-issued records, but the journal |
| 132 | // path is built from them, so they are checked here as well. |
| 133 | func journalName(kind, s string) error { |
| 134 | if s == "" || s == "." || s == ".." || strings.ContainsAny(s, "/\\\x00") { |
| 135 | return fmt.Errorf("moves journal: unusable %s %q", kind, s) |
| 136 | } |
| 137 | return nil |
| 138 | } |
| 139 | |
| 140 | func journalNames(host, op string) error { |
| 141 | if err := journalName("host", host); err != nil { |
| 142 | return err |
| 143 | } |
| 144 | return journalName("op", op) |
| 145 | } |
| 146 | |
| 147 | // JournalFile is the journal's file for (host, op): |
| 148 | // <JournalDir>/journal.json. Root reads it only through an anchored, |
| 149 | // no-symlink open of a regular file (internal/host's held release), never |
| 150 | // by name with ReadJournal. |
| 151 | func JournalFile(dataDir, host, op string) string { |
| 152 | return filepath.Join(JournalDir(dataDir, host, op), journalFile) |
| 153 | } |
| 154 | |
| 155 | // ReadJournal loads the journal for (host, op). A journal that does not |
| 156 | // exist yet is an empty one. A file that exists but cannot be trusted |
| 157 | // (unreadable, not valid JSON, an unknown version, or naming a different |
| 158 | // host or op) is an error and never an empty journal: starting over could |
| 159 | // redo a step that already ran. |
| 160 | func ReadJournal(dataDir, host, op string) (*Journal, error) { |
| 161 | if err := journalNames(host, op); err != nil { |
| 162 | return nil, err |
| 163 | } |
| 164 | b, err := os.ReadFile(JournalFile(dataDir, host, op)) |
| 165 | if errors.Is(err, fs.ErrNotExist) { |
| 166 | return &Journal{V: journalVersion, Host: host, Op: op, Steps: map[string]string{}}, nil |
| 167 | } |
| 168 | if err != nil { |
| 169 | return nil, fmt.Errorf("moves journal %s/%s: %w", host, op, err) |
| 170 | } |
| 171 | return DecodeJournal(b, host, op) |
| 172 | } |
| 173 | |
| 174 | // DecodeJournal is ReadJournal's judgement of a journal's bytes b, read by |
| 175 | // the caller: one JSON object of this version naming exactly (host, op). |
| 176 | func DecodeJournal(b []byte, host, op string) (*Journal, error) { |
| 177 | if err := journalNames(host, op); err != nil { |
| 178 | return nil, err |
| 179 | } |
| 180 | var j Journal |
| 181 | dec := json.NewDecoder(bytes.NewReader(b)) |
| 182 | if err := dec.Decode(&j); err != nil { |
| 183 | return nil, fmt.Errorf("moves journal %s/%s is corrupt: %w", host, op, err) |
| 184 | } |
| 185 | if dec.More() { |
| 186 | return nil, fmt.Errorf("moves journal %s/%s is corrupt: trailing data", host, op) |
| 187 | } |
| 188 | if j.V != journalVersion { |
| 189 | return nil, fmt.Errorf("moves journal %s/%s: unknown version %d", host, op, j.V) |
| 190 | } |
| 191 | if j.Host != host || j.Op != op { |
| 192 | return nil, fmt.Errorf("moves journal %s/%s names %s/%s", host, op, j.Host, j.Op) |
| 193 | } |
| 194 | if j.Steps == nil { |
| 195 | j.Steps = map[string]string{} |
| 196 | } |
| 197 | j.loadedPhase, j.loadedAttempt = j.Phase, j.Attempt |
| 198 | return &j, nil |
| 199 | } |
| 200 | |
| 201 | // write stamps UpdatedAt and saves the journal whole: a temp file in the |
| 202 | // same directory, fsync, rename over the old file, then fsync of the |
| 203 | // directory, so a crash leaves either the old journal or the new one. Only |
| 204 | // UpdateJournal calls it, with the journal's lock held since its read |
| 205 | // (TestJournalHasNoPlainWrite). |
| 206 | func (j *Journal) write(dataDir string) error { |
| 207 | if err := journalNames(j.Host, j.Op); err != nil { |
| 208 | return err |
| 209 | } |
| 210 | if j.V == 0 { |
| 211 | j.V = journalVersion |
| 212 | } |
| 213 | if j.V != journalVersion { |
| 214 | return fmt.Errorf("moves journal %s/%s: unknown version %d", j.Host, j.Op, j.V) |
| 215 | } |
| 216 | j.UpdatedAt = time.Now().UTC() |
| 217 | b, err := json.Marshal(j) |
| 218 | if err != nil { |
| 219 | return err |
| 220 | } |
| 221 | dir := JournalDir(dataDir, j.Host, j.Op) |
| 222 | created, err := missingDirs(dir) |
| 223 | if err != nil { |
| 224 | return err |
| 225 | } |
| 226 | if err := os.MkdirAll(dir, 0o700); err != nil { |
| 227 | return err |
| 228 | } |
| 229 | // A directory this call created is only durable once its parent's |
| 230 | // entry is, so sync the parent of each one, or a crash could lose the |
| 231 | // whole chain with the journal inside it. |
| 232 | for _, c := range created { |
| 233 | if err := syncDir(filepath.Dir(c)); err != nil { |
| 234 | return err |
| 235 | } |
| 236 | } |
| 237 | tmp, err := os.CreateTemp(dir, journalFile+".tmp-*") |
| 238 | if err != nil { |
| 239 | return err |
| 240 | } |
| 241 | tmpName := tmp.Name() |
| 242 | fail := func(err error) error { |
| 243 | tmp.Close() |
| 244 | os.Remove(tmpName) |
| 245 | return err |
| 246 | } |
| 247 | if _, err := tmp.Write(b); err != nil { |
| 248 | return fail(err) |
| 249 | } |
| 250 | if err := tmp.Sync(); err != nil { |
| 251 | return fail(err) |
| 252 | } |
| 253 | if err := tmp.Close(); err != nil { |
| 254 | os.Remove(tmpName) |
| 255 | return err |
| 256 | } |
| 257 | if err := os.Rename(tmpName, filepath.Join(dir, journalFile)); err != nil { |
| 258 | os.Remove(tmpName) |
| 259 | return err |
| 260 | } |
| 261 | return syncDir(dir) |
| 262 | } |
| 263 | |
| 264 | // missingDirs lists the directories at and above dir that do not exist yet, |
| 265 | // deepest first: the ones MkdirAll is about to create. |
| 266 | func missingDirs(dir string) ([]string, error) { |
| 267 | var out []string |
| 268 | for p := filepath.Clean(dir); ; p = filepath.Dir(p) { |
| 269 | _, err := os.Lstat(p) |
| 270 | if err == nil { |
| 271 | return out, nil |
| 272 | } |
| 273 | if !errors.Is(err, fs.ErrNotExist) { |
| 274 | return nil, err |
| 275 | } |
| 276 | out = append(out, p) |
| 277 | if filepath.Dir(p) == p { |
| 278 | return out, nil |
| 279 | } |
| 280 | } |
| 281 | } |
| 282 | |
| 283 | func syncDir(dir string) error { |
| 284 | d, err := os.Open(dir) |
| 285 | if err != nil { |
| 286 | return err |
| 287 | } |
| 288 | defer d.Close() |
| 289 | return d.Sync() |
| 290 | } |
| 291 | |
| 292 | // Done reports whether step has been marked. |
| 293 | func (j *Journal) Done(step string) bool { |
| 294 | _, ok := j.Steps[step] |
| 295 | return ok |
| 296 | } |
| 297 | |
| 298 | // Mark records step as done with value. It changes the in-memory journal |
| 299 | // only; Commit makes it durable. |
| 300 | func (j *Journal) Mark(step, value string) { |
| 301 | if j.Steps == nil { |
| 302 | j.Steps = map[string]string{} |
| 303 | } |
| 304 | j.Steps[step] = value |
| 305 | j.changes = append(j.changes, journalChange{step: step, value: value}) |
| 306 | } |
| 307 | |
| 308 | // Unmark removes step. It changes the in-memory journal only; Commit makes |
| 309 | // it durable. |
| 310 | func (j *Journal) Unmark(step string) { |
| 311 | delete(j.Steps, step) |
| 312 | j.changes = append(j.changes, journalChange{step: step, del: true}) |
| 313 | } |
| 314 | |
amadan