patchelf doesn't update PT_NOTE segment when relocating SHT_NOTE sections

[delroth]

Describe the bug

SHT_NOTE sections (e.g. .note.gnu.build-id) can sometimes be aliased by a PT_NOTE segment. When this is the case, patchelf might move the SHT_NOTE section without also updating the PT_NOTE segment. The PT_NOTE segment ends up pointing to the middle of random stuff that might have been overwritten.

This is as far as I know the root cause of NixOS/nixpkgs#91145 breaking Mesa in NixOS with patchelf 0.11.

Steps To Reproduce

$ echo 'int foo(int x) { return x; }' > test.c
$ gcc -s -shared -Wl,-build-id -o test.so test.c
$ readelf -n test.so
Displaying notes found in: .note.gnu.build-id
  Owner                 Data size	Description
  GNU                  0x00000014	NT_GNU_BUILD_ID (unique build ID bitstring)
    Build ID: 767c996573519ca3392910e472102d68d87342e7

$ patchelf --set-rpath $(perl -e 'print "A"x1000') test.so
$ readelf -n test.so
Displaying notes found in: .note.gnu.build-id
  Owner                 Data size	Description
readelf: Warning: note with invalid namesz and/or descsz found at offset 0x0
readelf: Warning:  type: 0x3, namesize: 0x00000004, descsize: 0x00000014, alignment: 8

Expected behavior

Not completely sure yet. PT_NOTE should be updated, but that means we might need to ensure we keep multiple note sections adjacent if the PT_NOTE contains multiple SHT_NOTE sections.

patchelf --version output

master

[domenkozar]

cc @rpurdie as this is part of stripping fix.

[delroth]

I implemented a fix in #218. I tested it with Mesa and confirmed that now Mesa is happily finding its build-id and not segfaulting anymore. readelf is also happy, though that was actually a separate problem from the section/segment mismatch! (alignment issue, see PR)