Files
silo-server/internal/diagnostics/bundle.go
Quick104andClaude Fable 5 2d5d4980de fix(diagnostics): address round-2 review findings on PR #445
- settings.go: cap the parsed cleanup interval at 7 days before converting to
  time.Duration so a huge configured value can't overflow int64 nanoseconds and
  wrap into a tiny/negative interval; add boundary tests.
- settings.go: propagate genuine settings read failures from LoadSettings
  (missing/empty -> default, error -> fail) so a transient DB error surfaces
  retryably instead of silently reporting uploads disabled or wrong quotas.
- bundle.go: validate non-manifest bundle entries while streaming with bounded
  memory -- device.json and crash/*.json must be a single JSON object,
  logs.jsonl/breadcrumbs.jsonl must be newline-delimited JSON objects with a
  per-line byte cap (new contract.MaxLogLineBytes); binary members stay opaque.
- diagnostics upload handler: extend the read deadline per-route via
  http.ResponseController.SetReadDeadline (10m) so slow mobile uploads of large
  bundles aren't cut off by the shared 30s server ReadTimeout.
- web admin download: request the ?proxy=1 streaming path directly so downloads
  work when S3Private is only server-reachable and errors can surface in-page.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012e3QjbPo96ed9Mn2qRiUkh
2026-07-21 11:29:02 -04:00

582 lines
16 KiB
Go

package diagnostics
import (
"archive/tar"
"bytes"
"compress/gzip"
"crypto/sha256"
"encoding/hex"
"encoding/json"
"errors"
"fmt"
"hash"
"io"
"math"
"path"
"strings"
"github.com/Silo-Server/silo-server/internal/diagnostics/contract"
)
const (
DefaultMaxBundleEntries = 16
DefaultMaxCompressionRatio = int64(200)
defaultBundleReadBufferSize = 32 * 1024
// Standard tar writers (GNU tar, Python tarfile, Apache Commons Compress)
// pad the archive with zero blocks to a record boundary, 10240 bytes by
// default; 64 KiB also covers writers configured with larger blocking
// factors.
maxTarTrailingPaddingBytes = 64 * 1024
)
var (
ErrCompressedTooLarge = errors.New("diagnostics bundle compressed size exceeds limit")
ErrUncompressedTooLarge = errors.New("diagnostics bundle uncompressed size exceeds limit")
ErrEntryTooLarge = errors.New("diagnostics bundle entry size exceeds limit")
ErrTooManyEntries = errors.New("diagnostics bundle has too many entries")
ErrInvalidBundle = errors.New("invalid diagnostics bundle")
ErrCompressionRatio = errors.New("diagnostics bundle compression ratio exceeds limit")
)
type BundleLimits struct {
MaxCompressedBytes int64
MaxUncompressedBytes int64
MaxEntryBytes int64
MaxEntries int
MaxCompressionRatio int64
}
type BundleInfo struct {
CompressedBytes int64
// UncompressedBytes is the total size of the decompressed tar stream —
// headers, entry payloads, end-of-archive marker, and padding — matching
// what a client counts on the way into its gzip writer (and `gzip -l`).
UncompressedBytes int64
SHA256 string
Entries []string
// EmbeddedManifest is the raw bytes of the archive's first entry
// (manifest.json). Per the bundle contract this is the part-1 manifest with
// the `archive` object removed; Ingest compares the two so a stored archive
// can never carry a manifest that disagrees with the accepted report.
EmbeddedManifest []byte
}
var allowedBundleEntries = allowlistMap(contract.ArchiveEntryAllowlist)
func ValidateBundle(r io.Reader, limits BundleLimits) (BundleInfo, error) {
limits = normalizeBundleLimits(limits)
metered := newCompressedMeter(r, limits.MaxCompressedBytes)
gz, err := gzip.NewReader(metered)
if err != nil {
if isBundleUploadAbortError(err) {
return BundleInfo{}, err
}
if errors.Is(err, ErrCompressedTooLarge) {
return BundleInfo{}, ErrCompressedTooLarge
}
return BundleInfo{}, fmt.Errorf("%w: open gzip: %v", ErrInvalidBundle, err)
}
gz.Multistream(false)
gzipClosed := false
defer func() {
if !gzipClosed {
_ = gz.Close()
}
}()
uncompressed := &uncompressedCounter{r: gz}
tr := tar.NewReader(uncompressed)
info := BundleInfo{
Entries: make([]string, 0, len(allowedBundleEntries)),
}
buffer := make([]byte, defaultBundleReadBufferSize)
var payloadBytes int64
for {
hdr, err := tr.Next()
if errors.Is(err, io.EOF) {
break
}
if err != nil {
return BundleInfo{}, classifyBundleReadError(err)
}
if len(info.Entries) >= limits.MaxEntries {
return BundleInfo{}, ErrTooManyEntries
}
name, err := validateEntryHeader(hdr)
if err != nil {
return BundleInfo{}, err
}
isManifestEntry := len(info.Entries) == 0
if isManifestEntry && name != "manifest.json" {
return BundleInfo{}, fmt.Errorf("%w: first entry must be manifest.json", ErrInvalidBundle)
}
info.Entries = append(info.Entries, name)
if hdr.Size > limits.MaxEntryBytes {
return BundleInfo{}, ErrEntryTooLarge
}
if isManifestEntry && hdr.Size > MaxManifestBytes {
return BundleInfo{}, fmt.Errorf("%w: embedded manifest exceeds %d bytes", ErrInvalidBundle, MaxManifestBytes)
}
if payloadBytes+hdr.Size > limits.MaxUncompressedBytes {
return BundleInfo{}, ErrUncompressedTooLarge
}
entry := &tarEntryReader{tr: tr, limits: limits, metered: metered, total: &payloadBytes}
var manifestCapture *bytes.Buffer
var consumeErr error
switch {
case isManifestEntry:
// Capture the embedded manifest so Ingest can compare it against the
// part-1 manifest; it is the only entry buffered whole.
manifestCapture = &bytes.Buffer{}
manifestCapture.Grow(int(hdr.Size))
_, consumeErr = io.Copy(manifestCapture, entry)
case isNDJSONBundleEntry(name):
// logs.jsonl / breadcrumbs.jsonl: newline-delimited JSON objects,
// validated line-by-line so memory stays bounded to one capped line.
consumeErr = validateNDJSONObjectEntry(entry)
case isJSONObjectBundleEntry(name):
// device.json / crash/*.json: a single JSON object, validated with a
// streaming token decoder (no whole-entry buffering).
consumeErr = validateJSONObjectEntry(entry)
default:
// Opaque members (crash/stack.txt, crash/tombstone.pb) carry no JSON
// contract; drain them so their bytes still count toward the size,
// ratio, and count limits.
_, consumeErr = io.Copy(io.Discard, entry)
}
if consumeErr != nil {
return BundleInfo{}, consumeErr
}
if entry.entryBytes != hdr.Size {
return BundleInfo{}, fmt.Errorf("%w: entry size mismatch for %s", ErrInvalidBundle, name)
}
if manifestCapture != nil {
info.EmbeddedManifest = manifestCapture.Bytes()
}
}
if len(info.Entries) == 0 {
return BundleInfo{}, fmt.Errorf("%w: empty archive", ErrInvalidBundle)
}
if err := rejectPostTarData(uncompressed, buffer); err != nil {
return BundleInfo{}, err
}
if err := gz.Close(); err != nil {
return BundleInfo{}, classifyBundleReadError(err)
}
gzipClosed = true
if err := rejectTrailingCompressedData(metered); err != nil {
return BundleInfo{}, err
}
if uncompressed.count > limits.MaxUncompressedBytes {
return BundleInfo{}, ErrUncompressedTooLarge
}
if ratioExceeded(uncompressed.count, metered.Count(), limits.MaxCompressionRatio) {
return BundleInfo{}, ErrCompressionRatio
}
info.UncompressedBytes = uncompressed.count
info.CompressedBytes = metered.Count()
info.SHA256 = hex.EncodeToString(metered.Sum())
return info, nil
}
func normalizeBundleLimits(limits BundleLimits) BundleLimits {
if limits.MaxCompressedBytes <= 0 {
limits.MaxCompressedBytes = DefaultMaxBundleBytes
}
if limits.MaxUncompressedBytes <= 0 {
limits.MaxUncompressedBytes = DefaultMaxUncompressed
}
if limits.MaxEntryBytes <= 0 || limits.MaxEntryBytes > limits.MaxUncompressedBytes {
limits.MaxEntryBytes = limits.MaxUncompressedBytes
}
if limits.MaxEntries <= 0 {
limits.MaxEntries = DefaultMaxBundleEntries
}
if limits.MaxCompressionRatio <= 0 {
limits.MaxCompressionRatio = DefaultMaxCompressionRatio
}
return limits
}
func validateEntryHeader(hdr *tar.Header) (string, error) {
if hdr == nil {
return "", fmt.Errorf("%w: missing tar header", ErrInvalidBundle)
}
// PAX/GNU extension records are consumed by archive/tar before the next
// header surfaces, so bytes carried in them never reach the allowlist and
// count checks. The contract fixtures are plain USTAR; reject anything else
// rather than let extension records smuggle data past validation.
if hdr.Format == tar.FormatPAX || hdr.Format == tar.FormatGNU {
return "", fmt.Errorf("%w: unsupported tar format", ErrInvalidBundle)
}
if len(hdr.PAXRecords) > 0 {
return "", fmt.Errorf("%w: unsupported tar extension records", ErrInvalidBundle)
}
if hdr.Typeflag != tar.TypeReg && hdr.Typeflag != tar.TypeRegA {
return "", fmt.Errorf("%w: unsupported tar entry type", ErrInvalidBundle)
}
if hdr.Size < 0 {
return "", fmt.Errorf("%w: negative tar entry size", ErrInvalidBundle)
}
// Reject names that are not already trimmed instead of normalizing them:
// otherwise a padded name like "manifest.json " would be recorded as the
// allowlisted entry while the archive stores a different literal member.
name := hdr.Name
if name != strings.TrimSpace(name) {
return "", fmt.Errorf("%w: unsafe entry name", ErrInvalidBundle)
}
if name == "" || strings.Contains(name, "\\") || strings.HasPrefix(name, "/") || path.IsAbs(name) {
return "", fmt.Errorf("%w: unsafe entry name", ErrInvalidBundle)
}
clean := path.Clean(name)
if clean == "." || clean != name || strings.HasPrefix(clean, "../") || strings.Contains(clean, "/../") {
return "", fmt.Errorf("%w: unsafe entry name", ErrInvalidBundle)
}
if _, ok := allowedBundleEntries[name]; !ok {
return "", fmt.Errorf("%w: disallowed entry %q", ErrInvalidBundle, name)
}
return name, nil
}
// tarEntryReader streams one tar entry's payload while charging every read
// against the bundle's entry-size, total-size, and compression-ratio limits —
// the same accounting the opaque drain path applied — so structural validators
// can pull the payload without letting a member slip past the guards. It also
// tracks entryBytes so the caller can compare it against the declared header
// size.
type tarEntryReader struct {
tr *tar.Reader
limits BundleLimits
metered *compressedMeter
total *int64
entryBytes int64
}
func (er *tarEntryReader) Read(p []byte) (int, error) {
n, err := er.tr.Read(p)
if n > 0 {
er.entryBytes += int64(n)
if er.entryBytes > er.limits.MaxEntryBytes {
return n, ErrEntryTooLarge
}
*er.total += int64(n)
if *er.total > er.limits.MaxUncompressedBytes {
return n, ErrUncompressedTooLarge
}
if ratioExceeded(*er.total, er.metered.Count(), er.limits.MaxCompressionRatio) {
return n, ErrCompressionRatio
}
}
if err != nil && !errors.Is(err, io.EOF) {
return n, classifyBundleReadError(err)
}
return n, err
}
func isNDJSONBundleEntry(name string) bool {
return name == "logs.jsonl" || name == "breadcrumbs.jsonl"
}
func isJSONObjectBundleEntry(name string) bool {
return name == "device.json" ||
(strings.HasPrefix(name, "crash/") && strings.HasSuffix(name, ".json"))
}
// validateJSONObjectEntry streams a single JSON object from r, rejecting
// anything that is not exactly one top-level object. It uses the token decoder
// so memory stays bounded to the nesting depth rather than the entry size.
func validateJSONObjectEntry(r io.Reader) error {
dec := json.NewDecoder(r)
tok, err := dec.Token()
if err != nil {
return classifyEntryContentError(err)
}
if delim, ok := tok.(json.Delim); !ok || delim != '{' {
return fmt.Errorf("%w: entry is not a JSON object", ErrInvalidBundle)
}
depth := 1
for depth > 0 {
tok, err := dec.Token()
if err != nil {
return classifyEntryContentError(err)
}
if delim, ok := tok.(json.Delim); ok {
switch delim {
case '{', '[':
depth++
case '}', ']':
depth--
}
}
}
// A well-formed entry ends here; anything after the top-level object is
// smuggled content.
if _, err := dec.Token(); !errors.Is(err, io.EOF) {
if err == nil {
return fmt.Errorf("%w: trailing data after JSON object", ErrInvalidBundle)
}
return classifyEntryContentError(err)
}
return nil
}
// validateNDJSONObjectEntry streams newline-delimited JSON, requiring each
// non-blank line to be a single JSON object. Lines are validated one at a time
// with a byte cap so a pathological entry can never buffer unbounded memory.
func validateNDJSONObjectEntry(r io.Reader) error {
buffer := make([]byte, defaultBundleReadBufferSize)
line := make([]byte, 0, 512)
for {
n, err := r.Read(buffer)
if err != nil && !errors.Is(err, io.EOF) {
return classifyEntryContentError(err)
}
chunk := buffer[:n]
for len(chunk) > 0 {
idx := bytes.IndexByte(chunk, '\n')
if idx < 0 {
line = append(line, chunk...)
if len(line) > contract.MaxLogLineBytes {
return errLogLineTooLarge()
}
break
}
line = append(line, chunk[:idx]...)
if len(line) > contract.MaxLogLineBytes {
return errLogLineTooLarge()
}
if verr := validateJSONObjectLine(line); verr != nil {
return verr
}
line = line[:0]
chunk = chunk[idx+1:]
}
if errors.Is(err, io.EOF) {
break
}
}
// Handle a trailing line without a closing newline.
if len(bytes.TrimSpace(line)) > 0 {
if len(line) > contract.MaxLogLineBytes {
return errLogLineTooLarge()
}
return validateJSONObjectLine(line)
}
return nil
}
func validateJSONObjectLine(line []byte) error {
trimmed := bytes.TrimSpace(line)
if len(trimmed) == 0 {
return nil
}
if trimmed[0] != '{' || !json.Valid(trimmed) {
return fmt.Errorf("%w: entry contains a line that is not a JSON object", ErrInvalidBundle)
}
return nil
}
func errLogLineTooLarge() error {
return fmt.Errorf("%w: entry line exceeds %d bytes", ErrInvalidBundle, contract.MaxLogLineBytes)
}
// classifyEntryContentError maps a decode-time error to the right bundle
// sentinel: the streaming limit sentinels and existing bundle errors pass
// through unchanged; a truncated/empty stream or malformed JSON becomes
// ErrInvalidBundle.
func classifyEntryContentError(err error) error {
switch {
case errors.Is(err, ErrEntryTooLarge),
errors.Is(err, ErrUncompressedTooLarge),
errors.Is(err, ErrCompressionRatio),
errors.Is(err, ErrCompressedTooLarge),
errors.Is(err, ErrInvalidBundle):
return err
case isBundleUploadAbortError(err):
return err
case errors.Is(err, io.EOF), errors.Is(err, io.ErrUnexpectedEOF):
return fmt.Errorf("%w: truncated or empty JSON entry", ErrInvalidBundle)
default:
return fmt.Errorf("%w: %v", ErrInvalidBundle, err)
}
}
func rejectPostTarData(r io.Reader, buffer []byte) error {
var padding int64
for {
n, err := r.Read(buffer)
if errors.Is(err, ErrCompressedTooLarge) {
return ErrCompressedTooLarge
}
if n > 0 {
for _, b := range buffer[:n] {
if b != 0 {
return fmt.Errorf("%w: trailing data after tar archive", ErrInvalidBundle)
}
}
padding += int64(n)
if padding > maxTarTrailingPaddingBytes {
return fmt.Errorf("%w: excessive padding after tar archive", ErrInvalidBundle)
}
}
if errors.Is(err, io.EOF) {
return nil
}
if err != nil {
return classifyBundleReadError(err)
}
}
}
func rejectTrailingCompressedData(metered *compressedMeter) error {
var one [1]byte
n, err := metered.Read(one[:])
if errors.Is(err, ErrCompressedTooLarge) {
return ErrCompressedTooLarge
}
if n > 0 {
return fmt.Errorf("%w: trailing data after gzip stream", ErrInvalidBundle)
}
if errors.Is(err, io.EOF) {
return nil
}
if err != nil {
return classifyBundleReadError(err)
}
return nil
}
func ratioExceeded(uncompressed, compressed, maxRatio int64) bool {
if compressed <= 0 || maxRatio <= 0 {
return false
}
if compressed > math.MaxInt64/maxRatio {
return false
}
return uncompressed > compressed*maxRatio
}
func classifyBundleReadError(err error) error {
if isBundleUploadAbortError(err) {
return err
}
if errors.Is(err, ErrCompressedTooLarge) {
return ErrCompressedTooLarge
}
if errors.Is(err, io.ErrUnexpectedEOF) {
return fmt.Errorf("%w: truncated archive", ErrInvalidBundle)
}
return fmt.Errorf("%w: %v", ErrInvalidBundle, err)
}
type bundleUploadAbortError struct {
err error
}
func (e *bundleUploadAbortError) Error() string {
if e == nil || e.err == nil {
return "diagnostics bundle upload aborted"
}
return e.err.Error()
}
func (e *bundleUploadAbortError) Unwrap() error {
if e == nil {
return nil
}
return e.err
}
func isBundleUploadAbortError(err error) bool {
var abortErr *bundleUploadAbortError
return errors.As(err, &abortErr)
}
func allowlistMap(entries []string) map[string]struct{} {
allowed := make(map[string]struct{}, len(entries))
for _, entry := range entries {
allowed[entry] = struct{}{}
}
return allowed
}
// uncompressedCounter tracks the total number of decompressed bytes read out
// of the gzip stream, including tar headers, end-of-archive blocks, and
// record padding.
type uncompressedCounter struct {
r io.Reader
count int64
}
func (c *uncompressedCounter) Read(p []byte) (int, error) {
n, err := c.r.Read(p)
c.count += int64(n)
return n, err
}
type compressedMeter struct {
r io.Reader
max int64
n int64
hash hash.Hash
limit bool
}
func newCompressedMeter(r io.Reader, max int64) *compressedMeter {
return &compressedMeter{
r: r,
max: max,
hash: sha256.New(),
limit: max > 0,
}
}
func (m *compressedMeter) Read(p []byte) (int, error) {
if m.limit {
remaining := m.max - m.n
if remaining < 0 {
return 0, ErrCompressedTooLarge
}
if int64(len(p)) > remaining+1 {
p = p[:int(remaining+1)]
}
}
n, err := m.r.Read(p)
if n > 0 {
m.n += int64(n)
_, _ = m.hash.Write(p[:n])
if m.limit && m.n > m.max {
return n, ErrCompressedTooLarge
}
}
return n, err
}
func (m *compressedMeter) ReadByte() (byte, error) {
var one [1]byte
n, err := m.Read(one[:])
if n == 1 {
if errors.Is(err, io.EOF) {
err = nil
}
return one[0], err
}
return 0, err
}
func (m *compressedMeter) Count() int64 {
return m.n
}
func (m *compressedMeter) Sum() []byte {
return m.hash.Sum(nil)
}