Extract file metadata for DiscUtils-based disc image formats

[Copilot]

Extends the file metadata work from #196 to the DiscUtils-based disc image formats, and finishes the parts of this PR that were left incomplete.

What's new

• FileEntryMetadata.FileAttributes — Windows/DOS file attributes.
• FileEntryMetadata.SecurityDescriptorSddl — Windows security descriptor in SDDL form.
• DiscCommon.TryGetFileMetadata() reads metadata from any file system implementing IUnixFileSystem (mode/uid/gid), IDosFileSystem (attributes), or IWindowsFileSystem (SDDL), so VHD/VHDX/VMDK/DMG, ISO 9660, UDF, and WIM all pick up whatever the underlying file system exposes.
• DiscCommon.CollectMetadata() pre-collects metadata for the ISO/UDF extractors, which dispose the reader before entries are consumed.

Support by format, verified by reflecting over the DiscUtils file system types:

Source	Populated
Ext, XFS, Btrfs, HFS+	Mode, Uid, Gid
ISO 9660 with RockRidge	Mode, Uid, Gid, FileAttributes
ISO 9660 without RockRidge	FileAttributes
FAT	FileAttributes
NTFS	FileAttributes, SecurityDescriptorSddl
WIM	FileAttributes, plus SecurityDescriptorSddl when recorded
UDF	None; Metadata stays null

Why this PR was stalled

IsoEntries_MetadataIsNullWithoutRockRidge and its _Sync twin were written in the first commit, when only IUnixFileSystem was handled. The follow-up commit added IDosFileSystem support, and CDReader implements it — so plain ISO entries started reporting FileAttributes = ReadOnly and those two tests began failing on an assertion of null metadata. They were never updated, so the branch was left red and the PR stayed a draft.

Fixes in this pass

• Rebuilt the two ISO tests to assert actual behavior, and tightened the RockRidge and NTFS tests to check concrete values (mode 0755/0644, uid/gid 1001, FileAttributes.Archive, owner and DACL present in the SDDL) instead of just non-null.
• Added UDF and WIM metadata tests, covering the UdfExtractor and WimExtractor paths that previously had no metadata coverage.
• Avoid one thrown exception per file on non-RockRidge ISOs. CDReader implements IUnixFileSystem unconditionally but throws InvalidOperationException for every entry when the active variant isn't RockRidge. TryGetFileMetadata now checks ActiveVariant up front. Since CDReader is constructed with Joliet priority (Joliet > RockRidge > Iso9660) and SUSP records are only parsed for the RockRidge variant, this is an exact check, not a heuristic.
• Normalized an empty security descriptor to null, so SecurityDescriptorSddl is either absent or meaningful. WIM images report an empty SDDL rather than no descriptor.
• Applied the reviewer's XML doc feedback: the interface lists now mention ISO 9660/CDReader, note the RockRidge requirement, and state that the lists aren't exhaustive.
• Documented FileEntryMetadata and the support matrix in the README, and added Metadata to the documented FileEntry surface.
• Rebased onto current main (SharpCompress 0.49.1 and the OpenReader/OpenArchive migration landed after this branch was cut). nuget.config is byte-identical to main; the earlier BOM/newline drift is gone.

Validation

662 tests × net8.0/net9.0/net10.0/net48 and 37 CLI tests × net8.0/net9.0/net10.0, all passing.

Fixes #197

[Copilot]

Pull request overview

Adds DiscUtils-backed filesystem metadata extraction so disc image formats (ISO/UDF/VHDX/WIM) can populate FileEntry.Metadata similarly to existing archive extractors, including Windows-specific NTFS/FAT metadata.

Changes:

• Extend FileEntryMetadata with Windows FileAttributes and NTFS SecurityDescriptorSddl.
• Add DiscCommon.TryGetFileMetadata() / CollectMetadata() and wire metadata into DiscUtils-based extractors.
• Add/expand tests and fixtures for ISO RockRidge Unix metadata and VHDX (NTFS) Windows metadata.

Reviewed changes

Copilot reviewed 9 out of 10 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
nuget.config	Normalizes XML header/format (restored feed configuration).
RecursiveExtractor/FileEntryMetadata.cs	Adds Windows file attributes + SDDL fields to metadata model.
RecursiveExtractor/Extractors/DiscCommon.cs	Implements DiscUtils metadata extraction and pre-collection helper.
RecursiveExtractor/Extractors/WimExtractor.cs	Sets metadata per extracted WIM entry via DiscCommon.
RecursiveExtractor/Extractors/IsoExtractor.cs	Pre-collects metadata while ISO FS is open and assigns to entries.
RecursiveExtractor/Extractors/UdfExtractor.cs	Pre-collects metadata while UDF FS is open and assigns to entries.
RecursiveExtractor.Tests/RecursiveExtractor.Tests.csproj	Includes new RockRidge ISO test fixture in outputs.
RecursiveExtractor.Tests/ExtractorTests/FileMetadataTests.cs	Adds coverage for ISO (with/without RockRidge) and VHDX NTFS metadata.
RecursiveExtractor.Tests/ExtractorTests/ExpectedNumFilesTests.cs	Adds expected file counts for the new RockRidge ISO fixture.

[Copilot]

Review details

Suppressed comments (3)

RecursiveExtractor/Extractor.cs:574

• Same as the non-parallel path: on Windows the output-directory containment check should be case-insensitive to avoid false positives when path casing differs after normalization.

                            var targetPath = Path.Combine(outputDirectory, entry.GetSanitizedPath());
                            var fullTargetPath = Path.GetFullPath(targetPath);
                            if (!fullTargetPath.StartsWith(parallelFullOutputDir, StringComparison.Ordinal))
                            {
                                Logger.Warn("Skipping entry that would escape output directory: {0}", entry.FullPath);
                                return;

RecursiveExtractor/Extractor.cs:672

• Same containment check issue in the async extraction path: use OrdinalIgnoreCase on Windows so in-directory targets aren't accidentally skipped due to casing differences.

                    var targetPath = Path.Combine(outputDirectory, entry.GetSanitizedPath());
                    var fullTargetPath = Path.GetFullPath(targetPath);
                    if (!fullTargetPath.StartsWith(asyncFullOutputDir, StringComparison.Ordinal))
                    {
                        Logger.Warn("Skipping entry that would escape output directory: {0}", entry.FullPath);
                        continue;

RecursiveExtractor/Extractor.cs:505

• On Windows, using a case-sensitive StartsWith comparison can incorrectly treat valid in-directory paths as escaping the output directory if the drive letter or path casing differs (common with symlinks/UNC normalization). Use a Windows-specific case-insensitive comparison for this containment check.

This issue also appears in the following locations of the same file:

• line 569
• line 667

                    var targetPath = Path.Combine(outputDirectory, entry.GetSanitizedPath());
                    var fullTargetPath = Path.GetFullPath(targetPath);
                    if (!fullTargetPath.StartsWith(fullOutputDir, StringComparison.Ordinal))
                    {
                        Logger.Warn("Skipping entry that would escape output directory: {0}", entry.FullPath);

• Files reviewed: 27/28 changed files
• Comments generated: 0 new
• Review effort level: Lite

[gfs]

Reviewed this and pushed a follow-up commit (f61c96d) rather than sending it back for another round. Three things:

Hybrid Joliet + RockRidge ISOs were silently losing Unix metadata. CDReader is opened with Joliet given priority, and DiscUtils only parses SUSP records for the variant it activates, so on an image built with both (mkisofs -J -R — most Linux distro ISOs) ActiveVariant comes back as Joliet and GetUnixFileInfo throws for every file. The ActiveVariant guard added in this PR is exactly right, it just turned a per-file exception into a silent skip. New DiscCommon.CollectIsoMetadata opens a second Joliet-skipping reader when the active variant isn't RockRidge and merges Mode/Uid/Gid back in by path — I checked that both trees produce identical path strings, so the match is reliable. Cost is nothing for RockRidge-only images, one extra volume descriptor scan for plain/Joliet ones (it bails on the variant check before touching any file), and one extra tree walk only for genuine hybrids. FileAttributes still comes from the tree the entries were enumerated from, so nothing else moves.

Added TestDataJolietRockRidge.iso for this, plus sync and async coverage. Confirmed both new tests fail without the fallback.

CollectMetadata was inside the try that guards reader construction and entry enumeration. Everything inside TryGetFileMetadata is caught, but fi.FullName isn't, so a malformed record during the metadata pass would have set failed = true and reported the whole image as a failed archive even though enumeration succeeded. It now reads each entry under its own try/catch and is documented as non-throwing, which also means one bad entry no longer costs metadata for every other entry.

The README support table was missing formats that already populate metadata. ZIP (Mode from the Unix external attributes), RAR and 7z (Mode from the raw attribute field) all set Metadata today, so the table read as exhaustive and implied they carried none. Added rows for them.

Things I looked at and deliberately left alone: plain ISOs now attach a non-null Metadata carrying nothing but a synthesized FileAttributes.ReadOnly, which isn't really recorded data — not worth churning here. The metadata pass also isn't covered by the resource governor, but the governor is a zip-bomb guard rather than an absolute byte meter, so that's fine. And the WIM tests pass vacuously off-Windows instead of skipping, which there's no clean fix for on xunit 2.x.

676 lib tests and 37 CLI tests pass locally on net9.0, and the library still builds on every TFM including netstandard2.0.

Nice work on the write-up, by the way — documenting exactly why the branch went red made this much faster to pick up.