13 "github.com/RoaringBitmap/roaring"
14 "github.com/anacrolix/torrent/common"
15 "github.com/anacrolix/torrent/metainfo"
16 "github.com/anacrolix/torrent/segments"
19 type RequestSpec = segments.Extent
21 type requestPartResult struct {
26 type requestPart struct {
29 result chan requestPartResult
31 // Wrap http response bodies for such things as download rate limiting.
32 responseBodyWrapper ResponseBodyWrapper
37 Result chan RequestResult
40 func (r Request) Cancel() {
45 HttpClient *http.Client
47 fileIndex segments.Index
49 // The pieces we can request with the Url. We're more likely to ban/block at the file-level
50 // given that's how requests are mapped to webseeds, but the torrent.Client works at the piece
51 // level. We can map our file-level adjustments to the pieces here. This probably need to be
52 // private in the future, if Client ever starts removing pieces.
54 ResponseBodyWrapper ResponseBodyWrapper
57 type ResponseBodyWrapper func(io.Reader) io.Reader
59 func (me *Client) SetInfo(info *metainfo.Info) {
60 if !strings.HasSuffix(me.Url, "/") && info.IsDir() {
61 // In my experience, this is a non-conforming webseed. For example the
62 // http://ia600500.us.archive.org/1/items URLs in archive.org torrents.
65 me.fileIndex = segments.NewIndex(common.LengthIterFromUpvertedFiles(info.UpvertedFiles()))
67 me.Pieces.AddRange(0, uint64(info.NumPieces()))
70 type RequestResult struct {
75 func (ws *Client) NewRequest(r RequestSpec) Request {
76 ctx, cancel := context.WithCancel(context.Background())
77 var requestParts []requestPart
78 if !ws.fileIndex.Locate(r, func(i int, e segments.Extent) bool {
79 req, err := NewRequest(ws.Url, i, ws.info, e.Start, e.Length)
83 req = req.WithContext(ctx)
86 result: make(chan requestPartResult, 1),
88 responseBodyWrapper: ws.ResponseBodyWrapper,
92 resp, err := ws.HttpClient.Do(req)
93 part.result <- requestPartResult{
99 requestParts = append(requestParts, part)
102 panic("request out of file bounds")
106 Result: make(chan RequestResult, 1),
109 b, err := readRequestPartResponses(ctx, requestParts)
110 req.Result <- RequestResult{
118 type ErrBadResponse struct {
120 Response *http.Response
123 func (me ErrBadResponse) Error() string {
127 func recvPartResult(ctx context.Context, buf io.Writer, part requestPart) error {
128 result := <-part.result
129 // Make sure there's no further results coming, it should be a one-shot channel.
131 if result.err != nil {
134 defer result.resp.Body.Close()
135 var body io.Reader = result.resp.Body
136 if part.responseBodyWrapper != nil {
137 body = part.responseBodyWrapper(body)
139 // Prevent further accidental use
140 result.resp.Body = nil
141 if ctx.Err() != nil {
144 switch result.resp.StatusCode {
145 case http.StatusPartialContent:
146 copied, err := io.Copy(buf, body)
150 if copied != part.e.Length {
151 return fmt.Errorf("got %v bytes, expected %v", copied, part.e.Length)
155 // This number is based on
156 // https://archive.org/download/BloodyPitOfHorror/BloodyPitOfHorror.asr.srt. It seems that
157 // archive.org might be using a webserver implementation that refuses to do partial
158 // responses to small files.
159 if part.e.Start < 48<<10 {
160 if part.e.Start != 0 {
161 log.Printf("resp status ok but requested range [url=%q, range=%q]",
163 part.req.Header.Get("Range"))
165 // Instead of discarding, we could try receiving all the chunks present in the response
166 // body. I don't know how one would handle multiple chunk requests resulting in an OK
167 // response for the same file. The request algorithm might be need to be smarter for
169 discarded, _ := io.CopyN(io.Discard, body, part.e.Start)
171 log.Printf("discarded %v bytes in webseed request response part", discarded)
173 _, err := io.CopyN(buf, body, part.e.Length)
176 return ErrBadResponse{"resp status ok but requested range", result.resp}
178 case http.StatusServiceUnavailable:
181 return ErrBadResponse{
182 fmt.Sprintf("unhandled response status code (%v)", result.resp.StatusCode),
188 var ErrTooFast = errors.New("making requests too fast")
190 func readRequestPartResponses(ctx context.Context, parts []requestPart) (_ []byte, err error) {
192 for _, part := range parts {
194 err = recvPartResult(ctx, &buf, part)
196 err = fmt.Errorf("reading %q at %q: %w", part.req.URL, part.req.Header.Get("Range"), err)
200 return buf.Bytes(), err