Skip to content

Regression in storage url signing #1346

Description

@mickeyreiss

@ostronom @mziccard It seems #1277 has caused a regression in my code. I use / characters in the blob names, and they are now being unexpectedly encoded as %2F in the signed path.

It seems that object naming requirements are not equivalent to URL encoding.

When I round-trip a signed url, Storage#get now fails to find the successfully uploaded file (confirmed via gsutil and Cloud Console).

My understanding is that /s are valid characters for blob names. From the same page quoted above:

The canonical resource is everything that follows the host name. For example, if the Cloud Storage URL is https://storage.googleapis.com/example-bucket/cat-pics/tabby.jpeg, then the canonical resource is /example-bucket/cat-pics/tabby.jpeg.

It seems like this library needs to use a different escape function for object names. (Also, I noticed that the new test testSignUrlForBlobWithSpecialChars, does not cover slashes.)

Does this make sense? What is the intended behavior for blob names containing slashes?


repro test:

    BlobInfo blobInfo1 = BlobInfo.builder(bucketName, "foo/bar/baz #%20other cool stuff.txt").build();
    storage.create(blobInfo1, "content".getBytes());
    URL url = storage.signUrl(
        blobInfo1,
        1,
        TimeUnit.HOURS,
        httpMethod(HttpMethod.PUT),
        httpMethod(HttpMethod.GET)
    );
    final String[] split = url.getPath().split("/");
    Blob notFound = storage.get(split[1], split[2]); // unexpectedly null!
    Blob found = storage.get(split[1], URLDecoder.decode(split[2])); // found, surprisingly

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions