rclone/backend/webdav/chunking.go

226 lines
7.2 KiB
Go

package webdav
/*
chunked update for Nextcloud
see https://docs.nextcloud.com/server/20/developer_manual/client_apis/WebDAV/chunking.html
*/
import (
"context"
"crypto/md5"
"encoding/hex"
"errors"
"fmt"
"io"
"net/http"
"path"
"time"
"github.com/rclone/rclone/fs"
"github.com/rclone/rclone/lib/readers"
"github.com/rclone/rclone/lib/rest"
)
func (f *Fs) shouldRetryChunkMerge(ctx context.Context, resp *http.Response, err error, sleepTime *time.Duration, wasLocked *bool) (bool, error) {
// Not found. Can be returned by NextCloud when merging chunks of an upload.
if resp != nil && resp.StatusCode == 404 {
if *wasLocked {
// Assume a 404 error after we've received a 423 error is actually a success
return false, nil
}
return true, err
}
// 423 LOCKED
if resp != nil && resp.StatusCode == 423 {
*wasLocked = true
fs.Logf(f, "Sleeping for %v to wait for chunks to be merged after 423 error", *sleepTime)
time.Sleep(*sleepTime)
*sleepTime *= 2
return true, fmt.Errorf("merging the uploaded chunks failed with 423 LOCKED. This usually happens when the chunks merging is still in progress on NextCloud, but it may also indicate a failed transfer: %w", err)
}
return f.shouldRetry(ctx, resp, err)
}
// set the chunk size for testing
func (f *Fs) setUploadChunkSize(cs fs.SizeSuffix) (old fs.SizeSuffix, err error) {
old, f.opt.ChunkSize = f.opt.ChunkSize, cs
return
}
func (o *Object) getChunksUploadDir() (string, error) {
hasher := md5.New()
_, err := hasher.Write([]byte(o.filePath()))
if err != nil {
return "", fmt.Errorf("chunked upload couldn't hash URL: %w", err)
}
uploadDir := "rclone-chunked-upload-" + hex.EncodeToString(hasher.Sum(nil))
return uploadDir, nil
}
func (f *Fs) getChunksUploadURL() (string, error) {
submatch := nextCloudURLRegex.FindStringSubmatch(f.endpointURL)
if submatch == nil {
return "", errors.New("the remote url looks incorrect. Note that nextcloud chunked uploads require you to use the /dav/files/USER endpoint instead of /webdav. Please check 'rclone config show remotename' to verify that the url field ends in /dav/files/USERNAME")
}
baseURL, user := submatch[1], submatch[2]
chunksUploadURL := fmt.Sprintf("%s/dav/uploads/%s/", baseURL, user)
return chunksUploadURL, nil
}
func (o *Object) shouldUseChunkedUpload(src fs.ObjectInfo) bool {
return o.fs.canChunk && o.fs.opt.ChunkSize > 0 && src.Size() > int64(o.fs.opt.ChunkSize)
}
func (o *Object) updateChunked(ctx context.Context, in0 io.Reader, src fs.ObjectInfo, options ...fs.OpenOption) (err error) {
var uploadDir string
// see https://docs.nextcloud.com/server/24/developer_manual/client_apis/WebDAV/chunking.html#starting-a-chunked-upload
uploadDir, err = o.createChunksUploadDirectory(ctx)
if err != nil {
return err
}
partObj := &Object{
fs: o.fs,
}
// see https://docs.nextcloud.com/server/24/developer_manual/client_apis/WebDAV/chunking.html#uploading-chunks
err = o.uploadChunks(ctx, in0, src.Size(), partObj, uploadDir, options)
if err != nil {
return err
}
// see https://docs.nextcloud.com/server/24/developer_manual/client_apis/WebDAV/chunking.html#assembling-the-chunks
err = o.mergeChunks(ctx, uploadDir, options, src)
if err != nil {
return err
}
return nil
}
func (o *Object) uploadChunks(ctx context.Context, in0 io.Reader, size int64, partObj *Object, uploadDir string, options []fs.OpenOption) error {
chunkSize := int64(partObj.fs.opt.ChunkSize)
// TODO: upload chunks in parallel for faster transfer speeds
for offset := int64(0); offset < size; offset += chunkSize {
if err := ctx.Err(); err != nil {
return err
}
contentLength := chunkSize
// Last chunk may be smaller
if size-offset < contentLength {
contentLength = size - offset
}
endOffset := offset + contentLength - 1
partObj.remote = fmt.Sprintf("%s/%015d-%015d", uploadDir, offset, endOffset)
// Enable low-level HTTP 2 retries.
// 2022-04-28 15:59:06 ERROR : stuff/video.avi: Failed to copy: uploading chunk failed: Put "https://censored.com/remote.php/dav/uploads/Admin/rclone-chunked-upload-censored/000006113198080-000006123683840": http2: Transport: cannot retry err [http2: Transport received Server's graceful shutdown GOAWAY] after Request.Body was written; define Request.GetBody to avoid this error
buf := make([]byte, chunkSize)
in := readers.NewRepeatableLimitReaderBuffer(in0, buf, chunkSize)
getBody := func() (io.ReadCloser, error) {
// RepeatableReader{} plays well with accounting so rewinding doesn't make the progress buggy
if _, err := in.Seek(0, io.SeekStart); err != nil {
return nil, err
}
return io.NopCloser(in), nil
}
err := partObj.updateSimple(ctx, in, getBody, partObj.remote, contentLength, "application/x-www-form-urlencoded", nil, o.fs.chunksUploadURL, options...)
if err != nil {
return fmt.Errorf("uploading chunk failed: %w", err)
}
}
return nil
}
func (o *Object) createChunksUploadDirectory(ctx context.Context) (string, error) {
uploadDir, err := o.getChunksUploadDir()
if err != nil {
return uploadDir, err
}
err = o.purgeUploadedChunks(ctx, uploadDir)
if err != nil {
return "", fmt.Errorf("chunked upload couldn't purge upload directory: %w", err)
}
opts := rest.Opts{
Method: "MKCOL",
Path: uploadDir + "/",
NoResponse: true,
RootURL: o.fs.chunksUploadURL,
}
err = o.fs.pacer.CallNoRetry(func() (bool, error) {
resp, err := o.fs.srv.Call(ctx, &opts)
return o.fs.shouldRetry(ctx, resp, err)
})
if err != nil {
return "", fmt.Errorf("making upload directory failed: %w", err)
}
return uploadDir, err
}
func (o *Object) mergeChunks(ctx context.Context, uploadDir string, options []fs.OpenOption, src fs.ObjectInfo) error {
var resp *http.Response
// see https://docs.nextcloud.com/server/24/developer_manual/client_apis/WebDAV/chunking.html?highlight=chunk#assembling-the-chunks
opts := rest.Opts{
Method: "MOVE",
Path: path.Join(uploadDir, ".file"),
NoResponse: true,
Options: options,
RootURL: o.fs.chunksUploadURL,
}
destinationURL, err := rest.URLJoin(o.fs.endpoint, o.filePath())
if err != nil {
return fmt.Errorf("finalize chunked upload couldn't join URL: %w", err)
}
opts.ExtraHeaders = o.extraHeaders(ctx, src)
opts.ExtraHeaders["Destination"] = destinationURL.String()
sleepTime := 5 * time.Second
wasLocked := false
err = o.fs.pacer.Call(func() (bool, error) {
resp, err = o.fs.srv.Call(ctx, &opts)
return o.fs.shouldRetryChunkMerge(ctx, resp, err, &sleepTime, &wasLocked)
})
if err != nil {
return fmt.Errorf("finalize chunked upload failed, destinationURL: \"%s\": %w", destinationURL, err)
}
return err
}
func (o *Object) purgeUploadedChunks(ctx context.Context, uploadDir string) error {
// clean the upload directory if it exists (this means that a previous try didn't clean up properly).
opts := rest.Opts{
Method: "DELETE",
Path: uploadDir + "/",
NoResponse: true,
RootURL: o.fs.chunksUploadURL,
}
err := o.fs.pacer.Call(func() (bool, error) {
resp, err := o.fs.srv.CallXML(ctx, &opts, nil, nil)
// directory doesn't exist, no need to purge
if resp != nil && resp.StatusCode == http.StatusNotFound {
return false, nil
}
return o.fs.shouldRetry(ctx, resp, err)
})
return err
}