tarfiles: `AttributeError: '_Stream' object has no attribute 'exception'` while trying to open tgz file

[balmeida-nokia]

Bug report

If a tar file appears to be a tar.gz file, it can fail in this block of tarfile.py

        while c < size:
            # Skip underlying buffer to avoid unaligned double buffering.
            if self.buf:
                buf = self.buf
                self.buf = b""
            else:
                buf = self.fileobj.read(self.bufsize)
                if not buf:
                    break
            try:
                buf = self.cmp.decompress(buf)
            except self.exception as e:
                raise ReadError("invalid compressed data") from e
            t.append(buf)
            c += len(buf)

at:

buf = self.cmp.decompress(buf)

before self.exception is set.
Because around:

            if comptype == "gz":
                try:
                    import zlib
                except ImportError:
                    raise CompressionError("zlib module is not available") from None
                self.zlib = zlib
                self.crc = zlib.crc32(b"")
                if mode == "r":
                    self._init_read_gz()
                    self.exception = zlib.error
                else:
                    self._init_write_gz()

This: self._init_read_gz()
is called before self.exception = zlib.error

Stack trace:

_read, tarfile.py:548
read, tarfile.py:526
_init_read_gz, tarfile.py:491
__init__, tarfile.py:375
open, tarfile.py:1827
<module>, test.py:2

In my case, buf = self.cmp.decompress(buf) failed to execute but that's a separate issue that may or may not be a bug

Your environment

• CPython versions tested on: 3.11.4
• Operating system and architecture: Windows 10 21H2

Linked PRs

• gh-107396: tarfiles: set self.exception before _init_read_gz() #107485
• [3.12] gh-107396: tarfiles: set self.exception before _init_read_gz() (GH-107485) #108207
• [3.11] gh-107396: tarfiles: set self.exception before _init_read_gz() (GH-107485) #108208

[sunmy2019]

Looks reasonable. Will you open a PR?

[balmeida-nokia]

Sure

[encukou]

Hello,
How does this issue manifest? exception appears like an write-only attribute of an internal class.
I'd like to add a test for the PR, but can't think of a reasonable one. Do you have a use case we currently don't have tests for?

[balmeida-nokia]

@encukou Please see: #107398

Maybe that one will explain better the sequence where this failure can happen.

My interpretation is that gzip supports backwards skipping and it appears to be rarely used. If I open a tgz with r|* (which, by design, doesn't support backwards skipping) and the gzip stream requests a backwards skip, an exception is raised. At the except statement it has except self.exception which is were the attribute is read, which means the attribute is actually read-write.

[encukou]

Oh, I see!
Here's a reproducer:

with open('hm.tar.gz', 'rb') as f:
    f = io.BytesIO(
        b'\x1f\x8b'  # header
        + b'\x08'  # compression method
        + b'\x04'  # flags
        + b'\0\0\0\0\0\0'  # padding?
        + b'\0\x01'  # size
        + bytes(5)  # corrupt data (zeros)
    )
    tarfile.open(fileobj=f, mode='r|gz')

If you can add it as a test using with self.assertRaises(tarfile.ReadError): to Lib/test/test_tarfile.py, it'd speed things up. If not I'll get to it later.