port(MachO): Align __LINKEDIT sections

[joanna-jo]

Fixes #2063.

According to dyld (alignments, alignment check) and lld (alignments, code signature alignment), the current __LINKEDIT blobs/“sections” we have are expected to have a minimum alignment of:


STRTAB: 1
CHAINED_FIXUPS: 8
SYMTAB: 8
CODE_SIGNATURE: 16

STRTAB and CODE_SIGNATURE are already correctly aligned, so the alignments were only changed for CHAINED_FIXUPS and SYMTAB. CHAINED_FIXUPS’ size was also aligned upwards to USIZE since OffsetVerifier needs this “section” to have an aligned size (wild: error: Unexpected memory offsets: Part #36 (section DYLD_CHAINED_FIXUPS_TABLE alignment: 8) has non aligned size: 0x4e). This follows what lld does for padding CHAINED_FIXUPS.

I manually checked with dyld_info and otool using misaligned files, and while this fixes the errors for misaligned __LINKEDIT content, there’s another dyld_info error (chained fixups, segment_offset does not match vmaddr from LC_SEGMENT in segment #2) that seems like an off-by-one in how we calculate the index for __DATA_CONST when writing the chained fixup table. I’ve opened a separate issue for that here: #2185.

[davidlattimore]

Thanks! Do you think it's practical to add a test that would catch this bug?

[davidlattimore]

One possible way in which this might be tested is to look at making linker-diff detect that the alignment of these sections is different from what lld emits.

[joanna-jo]

Thanks for looking this over and for the suggestion! :) I was also wondering what the best way to add a regression test here would be - I briefly looked into assertions but I’ll try supporting this directly with linker-diff then!

[marxin]

Thank you @joanna-jo for the very first contribution! Really appreciate your research part where you investigated what's the expected alignment of the various __LINKEDIT sections.

[joanna-jo]

This is now the linker-diff output before vs. after our fix on a file with misaligned SYMTAB, DYLD_CHAINED_FIXUPS, and CODE_SIGNATURE:

Before:

cargo run --bin linker-diff -- --wild-defaults --only 'linkedit.*' --ref /tmp/t.lld /tmp/t.wild
   Compiling linker-diff v0.9.0 (/Users/joannajo/wild/linker-diff)
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 1.18s
     Running `target/debug/linker-diff --wild-defaults --colour never --only 'linkedit.*' --ref /tmp/t.lld /tmp/t.wild`
wild: /tmp/t.wild
lld: /tmp/t.lld
linkedit.CODE_SIGNATURE.dataoff.alignment
  wild offset 0xc11f is not 16-byte aligned
  lld OK

linkedit.DYLD_CHAINED_FIXUPS.dataoff.alignment
  wild offset 0xc027 is not 8-byte aligned
  lld OK

linkedit.SYMTAB.symoff.alignment
  wild offset 0xc0cf is not 8-byte aligned
  lld OK

After:

cargo run --bin linker-diff -- --wild-defaults --only 'linkedit.*' --ref /tmp/t.lld /tmp/t.wild
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.25s
     Running `target/debug/linker-diff --wild-defaults --colour never --only 'linkedit.*' --ref /tmp/t.lld /tmp/t.wild`
No differences or validation failures detected