[ImadC9]

Pattern

Closes #108.

Adds patterns/fs/fat16.hexpat for FAT12 and FAT16 volumes, the two FAT variants that were still missing next to fat32.hexpat and exfat.hexpat.

• Boot sector with BPB and extended BPB (0x28 and 0x29 boot signatures)
• Every FAT, with FAT12's packed 12-bit entries decoded and each entry labelled: free, next cluster, end of chain, bad, reserved
• The fixed root directory, then every subdirectory and file followed through its FAT chain, including fragmented files and directories spanning several cluster runs
• VFAT long file name entries, deleted entries, volume labels and the Windows NT lowercase flags for 8.3 names
• The FAT type is chosen from the cluster count, as fatgen103 requires, not from the informational FAT12 string. FAT32 volumes are rejected with a pointer to fat32.hexpat.
• Cluster chains are walked iteratively, one array element per run of consecutive clusters, so a long or fragmented chain cannot exhaust the native stack. A chain is never followed for more clusters than the volume has, so cycles in the FAT end with a warning.
• Guards for corrupted images:
  • a directory already open further up the tree is not entered again
  • nesting deeper than 64 levels and more than 10000 directories are not followed
  • a BPB claiming more clusters than its FAT can hold is clamped
• Images cut off part way (partial dumps) parse up to where they end.

fs/pattern.hexpat now dispatches MBR partition types 0x01, 0x04, 0x06 and 0x0E (and their hidden variants) to the new pattern. Previously a disk with a FAT16 partition stopped with Unknown MBR Partition Type MBRPartitionType::FAT16_32MBPlus. Output for the existing fs.hexpat.img is byte-identical.

Verification

Every file the pattern finds was rebuilt from its data runs and compared byte-for-byte with what mcopy -s (mtools) extracts:

Image	Type	Result
64 KiB volume	FAT12, 512 B clusters	4 files, 2 dirs, identical
2.2 MB volume	FAT16, 512 B clusters	4 files, 2 dirs, identical
360 KiB, random churn	FAT12	44 files, 3 dirs, identical
1.44 MB, 2 KiB clusters, churn	FAT12	44 files, 3 dirs, identical
4 MB, churn	FAT16	46 files, 3 dirs, identical
16 MB, 4 KiB clusters, churn	FAT16	35 files, 3 dirs, identical
Standard 1.44 MB floppy	FAT12	identical
One file in 3,000 fragments	FAT16	identical
65,439 clusters (near max)	FAT16	parses in 0.09 s

• Fragmentation: besides the 3,000-fragment file, the churn images exercise 16 fragmented files and 5 multi-run directories.
• Type boundary: at the FAT12/FAT16 limit, the detected type matches mkfs.fat on both sides.
• MBR disk: an image with a FAT16 and a FAT12 partition parses through fs/pattern.hexpat, and file bytes inside both partitions match.
• Corruption: 500 corrupted images all parse, in under 0.1 s each, with every formatted value valid. They include broken and cyclic FAT chains, cycles combined with 4 GiB file sizes, and random bytes. Specific cases that also pass: a file chain looping on itself with size 0xFFFFFFFF, a looping directory chain, and 80 nested directories (stopped at 64 with a warning).
• Truncation: 3,888 truncated prefixes of the test images all parse.

Test data

tests/patterns/test_data/fat16.hexpat/ holds two images made with mkfs.fat and mtools from generated data:

• fat12.img, 64 KiB
• fat16.img, a FAT16 volume cut off after its last allocated cluster to keep it at 25 KiB; it parses identically to the full 2.2 MB image

Note: tests/patterns/CMakeLists.txt only globs patterns/*.hexpat, so nothing under patterns/fs/ runs in CI today, including the existing ext4 and fs test files. The images follow the naming convention so they will be picked up if subfolders are added. A recursive glob currently collides on 8 duplicate test names, so that needs its own change; happy to send one.

Checklist

• A pattern for this format doesn't exist yet
• The new pattern has been added to the relevant table in the Readme
• The new pattern has a description pragma
• MIME types: libmagic reports bare FAT volumes as application/octet-stream, so detection uses #pragma magic on the FAT12 / FAT16 signature instead, like fat32.hexpat and exfat.hexpat
• A test file for this pattern has been added to /tests/patterns/test_data
• Unit tests pass (311/311 locally, PatternLanguage 300d738)