TarFile fails to extract even when symbolic link references into the output directory

[tmds]

I already have an open issue about TarFile not wanting to create symbolic links that point outside the output directory: #74140.

Triggered by #77096, I was curious to see if podman would have the same issue as docker.

I did:

$ podman pull hello-world
$ podman save hello-world -o hello-world.tar # uncompressed when saving a local image

The resulting archive fails to extract:

Unhandled exception. System.IO.IOException: Extracting the Tar entry '../8e128b102ca8161cb297e3aa73240b8e537af67422fd4c22bf43c296c91d13ad.tar' would have resulted in a link target outside the specified destination directory: '/tmp/bca4a72dc384466ab6abd59cc585bd5a'
   at System.Formats.Tar.TarEntry.GetDestinationAndLinkPaths(String destinationDirectoryPath)
   at System.Formats.Tar.TarEntry.ExtractRelativeToDirectoryAsync(String destinationDirectoryPath, Boolean overwrite, SortedDictionary`2 pendingModes, CancellationToken cancellationToken)
   at System.Formats.Tar.TarFile.ExtractToDirectoryInternalAsync(Stream source, String destinationDirectoryPath, Boolean overwriteFiles, Boolean leaveOpen, CancellationToken cancellationToken)
   at System.Formats.Tar.TarFile.ExtractToDirectoryInternalAsync(Stream source, String destinationDirectoryPath, Boolean overwriteFiles, Boolean leaveOpen, CancellationToken cancellationToken)
   at Program.<Main>$(String[] args) in /tmp/console/Program.cs:line 6
   at Program.<Main>(String[] args)

The symbolic link it considers problematic is actually still pointing inside the output directory.
These are the archive entries:

8e128b102ca8161cb297e3aa73240b8e537af67422fd4c22bf43c296c91d13ad.tar(RegularFile) -> 
c0e8b50899a56dcd91634bfbb7a8336f2c0f6f6a9bbb5ad9b747dc1547aaa9f4.json(RegularFile) -> 
00335d90e4f471aae0fdbae54765dd7a5a1cb276c55c2fc75bf64ab20c5c97d1/layer.tar(SymbolicLink) -> ../8e128b102ca8161cb297e3aa73240b8e537af67422fd4c22bf43c296c91d13ad.tar
00335d90e4f471aae0fdbae54765dd7a5a1cb276c55c2fc75bf64ab20c5c97d1/VERSION(RegularFile) -> 
00335d90e4f471aae0fdbae54765dd7a5a1cb276c55c2fc75bf64ab20c5c97d1/json(RegularFile) -> 
manifest.json(RegularFile) -> 
repositories(RegularFile) -> 

cc @carlossanlop @jozkee @am11

[ghost]

Tagging subscribers to this area: @dotnet/area-system-io
See info in area-owners.md if you want to be subscribed.

Issue Details

---

I already have an open issue about TarFile not wanting to create symbolic links that point outside the output directory: #74140.

Triggered by #77096, I was curious to see if podman would have the same issue as docker.

I did:

$ podman pull hello-world
$ podman save hello-world -o hello-world.tar # uncompressed when saving a local image

The resulting archive fails to extract:

Unhandled exception. System.IO.IOException: Extracting the Tar entry '../8e128b102ca8161cb297e3aa73240b8e537af67422fd4c22bf43c296c91d13ad.tar' would have resulted in a link target outside the specified destination directory: '/tmp/bca4a72dc384466ab6abd59cc585bd5a'
   at System.Formats.Tar.TarEntry.GetDestinationAndLinkPaths(String destinationDirectoryPath)
   at System.Formats.Tar.TarEntry.ExtractRelativeToDirectoryAsync(String destinationDirectoryPath, Boolean overwrite, SortedDictionary`2 pendingModes, CancellationToken cancellationToken)
   at System.Formats.Tar.TarFile.ExtractToDirectoryInternalAsync(Stream source, String destinationDirectoryPath, Boolean overwriteFiles, Boolean leaveOpen, CancellationToken cancellationToken)
   at System.Formats.Tar.TarFile.ExtractToDirectoryInternalAsync(Stream source, String destinationDirectoryPath, Boolean overwriteFiles, Boolean leaveOpen, CancellationToken cancellationToken)
   at Program.<Main>$(String[] args) in /tmp/console/Program.cs:line 6
   at Program.<Main>(String[] args)

The symbolic link it considers problematic is actually still pointing inside the output directory.
These are the archive entries:

8e128b102ca8161cb297e3aa73240b8e537af67422fd4c22bf43c296c91d13ad.tar(RegularFile) -> 
c0e8b50899a56dcd91634bfbb7a8336f2c0f6f6a9bbb5ad9b747dc1547aaa9f4.json(RegularFile) -> 
00335d90e4f471aae0fdbae54765dd7a5a1cb276c55c2fc75bf64ab20c5c97d1/layer.tar(SymbolicLink) -> ../8e128b102ca8161cb297e3aa73240b8e537af67422fd4c22bf43c296c91d13ad.tar
00335d90e4f471aae0fdbae54765dd7a5a1cb276c55c2fc75bf64ab20c5c97d1/VERSION(RegularFile) -> 
00335d90e4f471aae0fdbae54765dd7a5a1cb276c55c2fc75bf64ab20c5c97d1/json(RegularFile) -> 
manifest.json(RegularFile) -> 
repositories(RegularFile) -> 

cc @carlossanlop @jozkee @am11

Author:	tmds
Assignees:	-
Labels:	area-System.IO, untriaged
Milestone:	-

[am11]

This is a duplicate of #74140, right? podman-hello-world.tar can be used as a test asset when we have the fix.

[tmds]

This is a duplicate of #74140, right?

In that case, the symbolic link points outside the output directory.
Here, the symbolic target is still in the output directory.

[am11]

Ok, I see. It is a bug where we fail to recognize that symlink is in fact pointing to path inside the destination directory (../8e128b102ca8161cb297e3aa73240b8e537af67422fd4c22bf43c296c91d13ad.tar) and throw IOException.

[am11]

Untested, but I think:

runtime/src/libraries/System.Formats.Tar/src/System/Formats/Tar/TarEntry.cs

Line 352 in ddebaf5

linkTargetPath = GetSanitizedFullPath(destinationDirectoryPath, LinkName);

should be: linkTargetPath = GetSanitizedFullPath(fileDestinationPath, LinkName); or linkTargetPath = GetSanitizedFullPath(Path.GetDirectoryName(fileDestinationPath), LinkName);