erofs-freebsd-out-tree/tests/TC018-48bit-large-fs.md

Test Case: Large Filesystem with 48-bit Addressing

Test ID: TC018-48bit-large-fs Category: Filesystem Metadata Priority: Low

Objective

Verify a real FLAT_PLAIN read whose extended inode has startblk_hi=1, so the provider I/O occurs above 16 TiB rather than only reporting a large statfs total.

Fixture Qualification

work=$(mktemp -d /tmp/repo22-tc018.XXXXXX)
tests/prepare_error_fixtures.sh "$work/fixtures"
python3 tests/erofs_fixture.py inspect \
  "$work/fixtures/48bit-high-file-prefix.erofs" --path /high-offset.txt
grep '^48bit-high-file-prefix ' "$work/fixtures/fixture-evidence.txt"
sha256 -q "$work/fixtures/48bit-high-file-prefix.erofs"

Require an extended nonempty FLAT_PLAIN inode, startblk_hi=1, a physical data_offset >= 17592186044416, a valid CRC, and an end offset within required_provider_length. The mutator zeros the original low-block payload, so a reader that ignores startblk_hi cannot return the expected source bytes. Read required_provider_length, start_block, data_offset, low_decoy_offset, and file_size from the evidence line.

FreeBSD Procedure

provider="$work/48bit-high-file-provider.raw"
cp "$work/fixtures/48bit-high-file-prefix.erofs" "$provider"
truncate -s "$required_provider_length" "$provider"
dd if="$work/fixtures/source/high-offset.txt" of="$provider" bs=4096 \
  seek="$start_block" conv=notrunc
cc -O2 -Wall -Wextra -std=c17 tests/read_probe.c -o "$work/read_probe"
"$work/read_probe" pread "$provider" "$low_decoy_offset" "$file_size" \
  "$work/raw-low-offset.txt"
"$work/read_probe" pread "$provider" "$data_offset" "$file_size" \
  "$work/raw-high-offset.txt"
dd if=/dev/zero of="$work/zero.bin" bs="$file_size" count=1
cmp "$work/zero.bin" "$work/raw-low-offset.txt"
! cmp -s "$work/fixtures/source/high-offset.txt" "$work/raw-low-offset.txt"
cmp "$work/fixtures/source/high-offset.txt" "$work/raw-high-offset.txt"
unit=$(mdconfig -a -t vnode -f "$provider")
diskinfo -v "/dev/$unit"
mkdir "$work/mnt"
mount -t erofs -o ro "/dev/$unit" "$work/mnt"
cmp "$work/fixtures/source/high-offset.txt" "$work/mnt/high-offset.txt"
cmp "$work/fixtures/source/control.txt" "$work/mnt/control.txt"
df -h "$work/mnt"
umount "$work/mnt"
mdconfig -d -u "${unit#md}"
rm -f "$provider"

Expected Results

The raw provider probe and mounted file both return the exact source bytes from above 16 TiB, while a low-offset control remains valid. Record the prefix hash, sparse media size/allocation, high offset, source/data hash, dmesg delta, and cleanup. If sparse high-offset I/O is unavailable, mark SHELVED with all attempts; a large df result alone cannot pass TC018.