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
55 PathEscaper PathEscaper
58 type ResponseBodyWrapper func(io.Reader) io.Reader
60 func (me *Client) SetInfo(info *metainfo.Info) {
61 if !strings.HasSuffix(me.Url, "/") && info.IsDir() {
62 // In my experience, this is a non-conforming webseed. For example the
63 // http://ia600500.us.archive.org/1/items URLs in archive.org torrents.
66 me.fileIndex = segments.NewIndex(common.LengthIterFromUpvertedFiles(info.UpvertedFiles()))
68 me.Pieces.AddRange(0, uint64(info.NumPieces()))
71 type RequestResult struct {
76 func (ws *Client) NewRequest(r RequestSpec) Request {
77 ctx, cancel := context.WithCancel(context.Background())
78 var requestParts []requestPart
79 if !ws.fileIndex.Locate(r, func(i int, e segments.Extent) bool {
80 req, err := newRequest(
81 ws.Url, i, ws.info, e.Start, e.Length,
87 req = req.WithContext(ctx)
90 result: make(chan requestPartResult, 1),
92 responseBodyWrapper: ws.ResponseBodyWrapper,
96 resp, err := ws.HttpClient.Do(req)
97 part.result <- requestPartResult{
103 requestParts = append(requestParts, part)
106 panic("request out of file bounds")
110 Result: make(chan RequestResult, 1),
113 b, err := readRequestPartResponses(ctx, requestParts)
114 req.Result <- RequestResult{
122 type ErrBadResponse struct {
124 Response *http.Response
127 func (me ErrBadResponse) Error() string {
131 func recvPartResult(ctx context.Context, buf io.Writer, part requestPart) error {
132 result := <-part.result
133 // Make sure there's no further results coming, it should be a one-shot channel.
135 if result.err != nil {
138 defer result.resp.Body.Close()
139 var body io.Reader = result.resp.Body
140 if part.responseBodyWrapper != nil {
141 body = part.responseBodyWrapper(body)
143 // Prevent further accidental use
144 result.resp.Body = nil
145 if ctx.Err() != nil {
148 switch result.resp.StatusCode {
149 case http.StatusPartialContent:
150 copied, err := io.Copy(buf, body)
154 if copied != part.e.Length {
155 return fmt.Errorf("got %v bytes, expected %v", copied, part.e.Length)
159 // This number is based on
160 // https://archive.org/download/BloodyPitOfHorror/BloodyPitOfHorror.asr.srt. It seems that
161 // archive.org might be using a webserver implementation that refuses to do partial
162 // responses to small files.
163 if part.e.Start < 48<<10 {
164 if part.e.Start != 0 {
165 log.Printf("resp status ok but requested range [url=%q, range=%q]",
167 part.req.Header.Get("Range"))
169 // Instead of discarding, we could try receiving all the chunks present in the response
170 // body. I don't know how one would handle multiple chunk requests resulting in an OK
171 // response for the same file. The request algorithm might be need to be smarter for
173 discarded, _ := io.CopyN(io.Discard, body, part.e.Start)
175 log.Printf("discarded %v bytes in webseed request response part", discarded)
177 _, err := io.CopyN(buf, body, part.e.Length)
180 return ErrBadResponse{"resp status ok but requested range", result.resp}
182 case http.StatusServiceUnavailable:
185 return ErrBadResponse{
186 fmt.Sprintf("unhandled response status code (%v)", result.resp.StatusCode),
192 var ErrTooFast = errors.New("making requests too fast")
194 func readRequestPartResponses(ctx context.Context, parts []requestPart) (_ []byte, err error) {
196 for _, part := range parts {
198 err = recvPartResult(ctx, &buf, part)
200 err = fmt.Errorf("reading %q at %q: %w", part.req.URL, part.req.Header.Get("Range"), err)
204 return buf.Bytes(), err