erofs-freebsd-out-tree/tests/TC143-deflate-zstd-partial-reference.md

Test Case: DEFLATE and ZSTD Partial References

Test ID: TC143-deflate-zstd-partial-reference

G5 fixture contract: Use G5-MANUAL-SETUP.md. Its generated paths, source comparisons, structured corruption offsets, and cleanup rules supersede placeholder examples in this file. Category: Compression Priority: Critical Regression: Non-LZ4 partial-reference decoding

Objective

Verify full and partial-reference decoding for raw DEFLATE and ZSTD streams, including exact full-file SHA256, cross-extent reads, random reads, and corrupted-stream EIO behavior.

Preconditions

• FreeBSD 15 guest.
• ZSTDIO-enabled repo22 module for the ZSTD cases.
• Deterministic full-index fixtures whose layout parser proves:
  • the full file points at a complete compressed stream;
  • the partial file reuses that physical stream;
  • the reused HEAD record has Z_EROFS_LI_PARTIAL_REF (0x8000) set;
  • the reused first NONHEAD record has the complete source pcluster's D0_CBLKCNT, not the partial file's original compressed-block count;
  • the partial file's logical length is shorter than the source stream's decompressed length.
• Corrupted copies produced only after locating the target compressed extent; do not use a blind image offset.

Layout Proof

dump.erofs -s deflate-partial-ref.erofs
dump.erofs --path=/a.dat -e deflate-partial-ref.erofs
dump.erofs --path=/b.dat -e deflate-partial-ref.erofs
dump.erofs -s zstd-partial-ref.erofs
dump.erofs --path=/a.dat -e zstd-partial-ref.erofs
dump.erofs --path=/b.dat -e zstd-partial-ref.erofs

The byte-level fixture record must assert the original NID, map-header offset, HEAD record, pblk, and checksum before setting the partial-reference bit or redirecting a pblk.

Test Steps

1. Mount the DEFLATE image and hash /a.dat and /b.dat completely.
2. Read a range crossing a known DEFLATE logical extent boundary.
3. Read another non-aligned random range and compare it byte-for-byte with the source.
4. Repeat steps 1-3 for ZSTD.
5. Mount each corrupted copy and read the targeted file through a small C errno probe.
6. Compare pre/post dmesg and active vmstat -m EROFS allocations.

The G5 DEFLATE partial is 100000 bytes because erofs-utils 1.8.6 splits the 1 MiB DEFLATE source according to the 32 KiB DEFLATE window. It reuses the first 17-block source pcluster and uses offsets 65500/2048 and 90000/4096. The ZSTD partial is 700000 bytes and uses these checks:

dd if=/mnt/repo22/b.dat of=/tmp/guest bs=1 skip=122900 count=512 status=none
dd if=source-b.dat of=/tmp/source bs=1 skip=122900 count=512 status=none
cmp /tmp/guest /tmp/source

dd if=/mnt/repo22/b.dat of=/tmp/guest bs=1 skip=524287 count=4097 status=none
dd if=source-prefix-b.dat of=/tmp/source bs=1 skip=524287 count=4097 status=none
cmp /tmp/guest /tmp/source

Expected Results

• Full DEFLATE and ZSTD streams require complete output and stream completion.
• Partial references succeed after producing the requested logical prefix; they do not require the reused source frame to finish.
• Complete SHA256 and every boundary/random byte comparison match.
• Targeted corruption returns read errno 5 (EIO), with no leaked active allocation, panic, or trap.
• control.bin in each corrupted image remains byte-identical to its source.

Cleanup

Unmount each image, detach every md unit, remove temporary range outputs, and unload the module by its kldstat ID.