mirror of
https://github.com/gotenberg/gotenberg.git
synced 2026-10-08 05:23:18 +01:00
fix(api): bound downloadFrom requests by the request deadline
This commit is contained in:
@@ -280,7 +280,12 @@ func newContext(echoCtx echo.Context, logger *slog.Logger, fs *gotenberg.FileSys
|
||||
|
||||
logger.DebugContext(dlCtx, fmt.Sprintf("download file from '%s'", dl.Url))
|
||||
|
||||
req, err := retryablehttp.NewRequest(http.MethodGet, dl.Url, nil)
|
||||
// The request must carry dlCtx: retryablehttp.NewRequest builds
|
||||
// on context.Background(), and its wait between attempts is a
|
||||
// select on the request context, so a contextless request cannot
|
||||
// be interrupted by --api-timeout (env API_TIMEOUT) or by the
|
||||
// caller going away.
|
||||
req, err := retryablehttp.NewRequestWithContext(dlCtx, http.MethodGet, dl.Url, nil)
|
||||
if err != nil {
|
||||
dlSpan.RecordError(err)
|
||||
dlSpan.SetStatus(codes.Error, err.Error())
|
||||
@@ -303,14 +308,28 @@ func newContext(echoCtx echo.Context, logger *slog.Logger, fs *gotenberg.FileSys
|
||||
}
|
||||
}
|
||||
|
||||
// Entries are serialized by the concurrency limit above, so a
|
||||
// late one can start after the deadline has already passed.
|
||||
// Fail closed rather than derive a non-positive timeout, which
|
||||
// [http.Client] reads as no deadline at all.
|
||||
remaining := time.Until(deadline)
|
||||
if remaining <= 0 {
|
||||
dlSpan.RecordError(context.DeadlineExceeded)
|
||||
dlSpan.SetStatus(codes.Error, context.DeadlineExceeded.Error())
|
||||
dlSpan.End()
|
||||
return fmt.Errorf("download file from '%s': %w", dl.Url, context.DeadlineExceeded)
|
||||
}
|
||||
|
||||
client := &retryablehttp.Client{
|
||||
HTTPClient: gotenberg.NewOutboundHttpClient(time.Until(deadline), downloadFromCfg.allowList, downloadFromCfg.denyList, downloadFromCfg.enableEnvironmentProxy, ipOpts...),
|
||||
HTTPClient: gotenberg.NewOutboundHttpClient(remaining, downloadFromCfg.allowList, downloadFromCfg.denyList, downloadFromCfg.enableEnvironmentProxy, ipOpts...),
|
||||
RetryMax: downloadFromCfg.maxRetry,
|
||||
RetryWaitMin: time.Duration(1) * time.Second,
|
||||
RetryWaitMax: time.Until(deadline),
|
||||
RetryWaitMax: remaining,
|
||||
Logger: gotenberg.NewLeveledLogger(logger),
|
||||
CheckRetry: retryablehttp.DefaultRetryPolicy,
|
||||
Backoff: retryablehttp.DefaultBackoff,
|
||||
// Not DefaultBackoff: it hands a hostile origin control of
|
||||
// the wait via Retry-After.
|
||||
Backoff: gotenberg.ClampedBackoff,
|
||||
}
|
||||
|
||||
resp, err := client.Do(req)
|
||||
|
||||
@@ -414,3 +414,117 @@ func TestContext_FileCount(t *testing.T) {
|
||||
t.Errorf("expected 3 files, got %d", got)
|
||||
}
|
||||
}
|
||||
|
||||
// A hostile origin must not choose how long Gotenberg waits.
|
||||
// [retryablehttp.DefaultBackoff] returns a Retry-After header verbatim for 429
|
||||
// and 503, and the wait between attempts is a select on the request context.
|
||||
// Building the request without a context therefore pinned the goroutine, its
|
||||
// connection, and its working directory for the attacker's chosen duration,
|
||||
// well past --api-timeout (env API_TIMEOUT).
|
||||
func TestNewContext_DownloadFromHostileRetryAfterIsBounded(t *testing.T) {
|
||||
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
|
||||
w.Header().Set("Retry-After", "3600")
|
||||
w.WriteHeader(http.StatusTooManyRequests)
|
||||
}))
|
||||
defer server.Close()
|
||||
|
||||
payload, err := json.Marshal([]downloadFrom{{Url: server.URL + "/file"}})
|
||||
if err != nil {
|
||||
t.Fatalf("marshal downloadFrom payload: %v", err)
|
||||
}
|
||||
|
||||
body := new(bytes.Buffer)
|
||||
writer := multipart.NewWriter(body)
|
||||
err = writer.WriteField("downloadFrom", string(payload))
|
||||
if err != nil {
|
||||
t.Fatalf("write downloadFrom field: %v", err)
|
||||
}
|
||||
err = writer.Close()
|
||||
if err != nil {
|
||||
t.Fatalf("close multipart writer: %v", err)
|
||||
}
|
||||
|
||||
req := httptest.NewRequest(http.MethodPost, "/forms/libreoffice/convert", body)
|
||||
req.Header.Set("Content-Type", writer.FormDataContentType())
|
||||
|
||||
echoCtx := echo.New().NewContext(req, httptest.NewRecorder())
|
||||
logger := slog.New(slog.DiscardHandler)
|
||||
fs := gotenberg.NewFileSystem(new(gotenberg.OsMkdirAll))
|
||||
|
||||
const timeout = 500 * time.Millisecond
|
||||
|
||||
start := time.Now()
|
||||
_, cancel, err := newContext(echoCtx, logger, fs, timeout, 0, downloadFromConfig{maxRetry: 2})
|
||||
elapsed := time.Since(start)
|
||||
if cancel != nil {
|
||||
defer cancel()
|
||||
}
|
||||
|
||||
if err == nil {
|
||||
t.Fatal("expected newContext to fail against an origin that only answers 429")
|
||||
}
|
||||
// Generous: the deadline is 500ms and Retry-After asks for an hour. Any
|
||||
// value in seconds means the remote is still in control.
|
||||
if elapsed > 10*time.Second {
|
||||
t.Fatalf("newContext took %s with Retry-After 3600; --api-timeout must bound it", elapsed)
|
||||
}
|
||||
}
|
||||
|
||||
// An entry that starts after the deadline has passed must fail closed. It used
|
||||
// to derive a negative client timeout, which [http.Client] reads as no
|
||||
// deadline at all, leaving the download unbounded.
|
||||
func TestNewContext_DownloadFromExpiredBudgetFailsClosed(t *testing.T) {
|
||||
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
|
||||
<-r.Context().Done()
|
||||
}))
|
||||
defer server.Close()
|
||||
|
||||
// Two entries, serialized by the concurrency limit, so the second one
|
||||
// starts once the first has burned the whole budget.
|
||||
payload, err := json.Marshal([]downloadFrom{
|
||||
{Url: server.URL + "/first"},
|
||||
{Url: server.URL + "/second"},
|
||||
})
|
||||
if err != nil {
|
||||
t.Fatalf("marshal downloadFrom payload: %v", err)
|
||||
}
|
||||
|
||||
body := new(bytes.Buffer)
|
||||
writer := multipart.NewWriter(body)
|
||||
err = writer.WriteField("downloadFrom", string(payload))
|
||||
if err != nil {
|
||||
t.Fatalf("write downloadFrom field: %v", err)
|
||||
}
|
||||
err = writer.Close()
|
||||
if err != nil {
|
||||
t.Fatalf("close multipart writer: %v", err)
|
||||
}
|
||||
|
||||
req := httptest.NewRequest(http.MethodPost, "/forms/libreoffice/convert", body)
|
||||
req.Header.Set("Content-Type", writer.FormDataContentType())
|
||||
|
||||
echoCtx := echo.New().NewContext(req, httptest.NewRecorder())
|
||||
logger := slog.New(slog.DiscardHandler)
|
||||
fs := gotenberg.NewFileSystem(new(gotenberg.OsMkdirAll))
|
||||
|
||||
done := make(chan error, 1)
|
||||
go func() {
|
||||
_, cancel, err := newContext(echoCtx, logger, fs, 400*time.Millisecond, 0, downloadFromConfig{
|
||||
maxRetry: 0,
|
||||
maxConcurrency: 1,
|
||||
})
|
||||
if cancel != nil {
|
||||
cancel()
|
||||
}
|
||||
done <- err
|
||||
}()
|
||||
|
||||
select {
|
||||
case err := <-done:
|
||||
if err == nil {
|
||||
t.Fatal("expected newContext to fail against a stalling origin")
|
||||
}
|
||||
case <-time.After(15 * time.Second):
|
||||
t.Fatal("newContext never returned: an entry starting past the deadline built an unbounded client")
|
||||
}
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user