x/vulndb: potential Go vuln in github.com/datacharmer/dbdeployer: GHSA-47wr-426j-fr82

[GoVulnBot]

In GitHub Security Advisory GHSA-47wr-426j-fr82, there is a vulnerability in the following Go packages or modules:

Unit	Fixed	Vulnerable Ranges
github.com/datacharmer/dbdeployer	1.58.2	< 1.58.2

Cross references:

• CVE-2020-26277 appears in issue x/vulndb: potential Go vuln in github.com/datacharmer/dbdeployer: GHSA-47wr-426j-fr82 #787 NOT_IMPORTABLE
• GHSA-47wr-426j-fr82 appears in issue x/vulndb: potential Go vuln in github.com/datacharmer/dbdeployer: GHSA-47wr-426j-fr82 #787 NOT_IMPORTABLE

See doc/triage.md for instructions on how to triage this report.

modules:
  - module: TODO
    versions:
      - fixed: 1.58.2
    packages:
      - package: github.com/datacharmer/dbdeployer
description: |-
    ### Impact
    _Users unpacking a tarball through dbdeployer may use a maliciously packaged tarball that contains symlinks to files external to the target. In such scenario, an attacker could induce dbdeployer to write into a system file, thus altering the computer defences._

    ### Mitigating factors
    For the attach to succeed, the following factors need to contribute:

    * The user is logged in as root. While dbdeployer is usable as root, it was designed to run as unprivileged user.
    *  The user has taken a tarball from a non secure source, without testing the checksum. When the tarball is retrieved through dbdeployer, the checksum is compared before attempting to unpack.

    ### Analysis

    An attacker could inject a symbolic link into the tarball, so that a file could result into `fake_file -> /etc/passwd` or some equally  important file.
    As it is now, dbdeployer would create the symlink as defined, with a local file `fake_file` linked to `/etc/passwd`. The danger here is that any process with the privileges to write to both `fake_file` and `/etc/passwd` could overwrite the system file. Even without malicious intent, this could result in the system to become unusable.
    As noted above, the user must have write privileges to the target file to do the damage.

    ### Remedies

    It has been suggested that the extract procedure use `filepath.EvalSymlinks` to determine whether the target is within the extraction directory. Unfortunately, this approach is unavailable in this context, because it would prevent legitimate patterns from being carried out.
    A simple case is a file `mysql-8.0.22-macos10.15-x86_64/bin/libprotobuf-lite.3.11.4.dylib` with a linkName `../lib/libprotobuf-lite.3.11.4.dylib`, if the linked file has not been created yet, `filepath.EvalSymlinks` would fail, as it acts on existing files only.

    An alternative method is comparing the depth (how many directories) of the file name with the depth of the link name. If the link name has a higher depth than the local file, we block the operation with an appropriate error:

    ```
    Unpacking tarball exploit/mysql-8.0.22-macos10.15-x86_64.tar.gz to $HOME/opt/mysql/test8.0.22
    ......
    link '../../../../../../../../../../etc' points outside target directory

    exit status 1
    ```
    As an additional fortifier, we can check whether the link points to an existing file, calculate its absolute name, and compare it with the absolute name of the extraction directory. A link to a full path (such as `/etc/passwd`) would fail this test, and trigger an error.

    The same check can be applied to a link to a non existing file with absolute path.

    ### Patches

    Patched in release 1.58.2

    ### For more information
    If you have any questions or comments about this advisory:
    * Open an issue in [dbdeployer](https://github.com/datacharmer/dbdeployer)
cves:
  - CVE-2020-26277
ghsas:
  - GHSA-47wr-426j-fr82

[tatianab]

Duplicate of #787