Sort dynamic symbols #507

Clcanny · 2020-12-06T15:25:10Z

// foo.cpp
#include <iostream>
namespace {
int var = 1;

void bar() {
  std::cout << "bar" << std::endl;
}
}  // namespace
void foo() {
}

// merge.cpp
#include <LIEF/ELF.hpp>
#include <memory>
#include <vector>
int main() {
  std::unique_ptr<LIEF::ELF::Binary> binary_(
      LIEF::ELF::Parser::parse("libfoo.so"));
  std::vector<LIEF::ELF::Symbol> dynamic_symbols{
      binary_->dynamic_symbols().begin(), binary_->dynamic_symbols().end()};
  uint32_t added_symbol_num = 0;
  for (const auto& symbol : dynamic_symbols) {
    binary_->add_dynamic_symbol(symbol);
    added_symbol_num += 1;
  }
  uint64_t entry_size = binary_->get_section(".dynsym").entry_size();
  assert(entry_size == 24);
  binary_->extend(binary_->get_section(".dynsym"),
                  dynamic_symbols.size() * entry_size);
  binary_->write("libfoo-modified.so");
}

all : clean compile run

clean :
        rm -f libfoo.so merge libfoo-modified.so

compile :
        g++ -std=c++11 foo.cpp -O0 -ggdb -shared -fPIC -o libfoo.so
        g++ -std=c++11 -O0 -ggdb merge.cpp -I/usr/include/LIEF-0.11.0-Linux/include -L/usr/lib/LIEF-0.11.0-Linux/lib -lLIEF -o merge

run :
        ./merge
        chmod u+x libfoo-modified.so
        readelf --dyn-syms libfoo-modified.so

Dynamic symbol table of libfoo.so is:

# readelf --dyn-syms libfoo.so
Symbol table '.dynsym' contains 19 entries:
   Num:    Value          Size Type    Bind   Vis      Ndx Name
     0: 0000000000000000     0 NOTYPE  LOCAL  DEFAULT  UND
     1: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND __gmon_start__
     2: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND _Jv_RegisterClasses
     3: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZNSt8ios_base4InitC1Ev@GLIBCXX_3.4 (2)
     4: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND __cxa_atexit@GLIBC_2.2.5 (3)
     5: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZNSt8ios_base4InitD1Ev@GLIBCXX_3.4 (2)
     6: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND _ITM_deregisterTMCloneTab
     7: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZStlsISt11char_traitsIcE@GLIBCXX_3.4 (2)
     8: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND _ITM_registerTMCloneTable
     9: 0000000000000000     0 FUNC    WEAK   DEFAULT  UND __cxa_finalize@GLIBC_2.2.5 (3)
    10: 0000000000000000     0 OBJECT  GLOBAL DEFAULT  UND _ZSt4cout@GLIBCXX_3.4 (2)
    11: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZNSolsEPFRSoS_E@GLIBCXX_3.4 (2)
    12: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZSt4endlIcSt11char_trait@GLIBCXX_3.4 (2)
    13: 0000000000201048     0 NOTYPE  GLOBAL DEFAULT   24 _end
    14: 0000000000201044     0 NOTYPE  GLOBAL DEFAULT   23 _edata
    15: 0000000000000932     7 FUNC    GLOBAL DEFAULT   12 _Z3foov
    16: 0000000000201044     0 NOTYPE  GLOBAL DEFAULT   24 __bss_start
    17: 0000000000000788     0 FUNC    GLOBAL DEFAULT    9 _init
    18: 0000000000000998     0 FUNC    GLOBAL DEFAULT   13 _fini

Without this change:

# make
readelf --dyn-syms libfoo-modified.so

Symbol table '.dynsym' contains 38 entries:
   Num:    Value          Size Type    Bind   Vis      Ndx Name
     0: 0000000000000000     0 NOTYPE  LOCAL  DEFAULT  UND
     1: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND __gmon_start__
     2: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND _Jv_RegisterClasses
     3: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZNSt8ios_base4InitC1Ev@GLIBCXX_3.4 (2)
     4: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND __cxa_atexit@GLIBC_2.2.5 (3)
     5: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZNSt8ios_base4InitD1Ev@GLIBCXX_3.4 (2)
     6: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND _ITM_deregisterTMCloneTab
     7: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZStlsISt11char_traitsIcE@GLIBCXX_3.4 (2)
     8: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND _ITM_registerTMCloneTable
     9: 0000000000000000     0 FUNC    WEAK   DEFAULT  UND __cxa_finalize@GLIBC_2.2.5 (3)
    10: 0000000000000000     0 OBJECT  GLOBAL DEFAULT  UND _ZSt4cout@GLIBCXX_3.4 (2)
    11: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZNSolsEPFRSoS_E@GLIBCXX_3.4 (2)
    12: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZSt4endlIcSt11char_trait@GLIBCXX_3.4 (2)
    13: 0000000000204210     0 NOTYPE  GLOBAL DEFAULT   24 _end
    14: 000000000020420c     0 NOTYPE  GLOBAL DEFAULT   23 _edata
    15: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZStlsISt11char_traitsIcE
    16: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND _ITM_registerTMCloneTable
    17: 0000000000000000     0 FUNC    WEAK   DEFAULT  UND __cxa_finalize
    18: 0000000000204210     0 NOTYPE  GLOBAL DEFAULT   24 _end
    19: 000000000020420c     0 NOTYPE  GLOBAL DEFAULT   23 _edata
    20: 0000000000003afa     7 FUNC    GLOBAL DEFAULT   12 _Z3foov
    21: 000000000020420c     0 NOTYPE  GLOBAL DEFAULT   24 __bss_start
    22: 0000000000003950     0 FUNC    GLOBAL DEFAULT    9 _init
    23: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND __cxa_atexit
    24: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZSt4endlIcSt11char_trait
    25: 0000000000003afa     7 FUNC    GLOBAL DEFAULT   12 _Z3foov
    26: 000000000020420c     0 NOTYPE  GLOBAL DEFAULT   24 __bss_start
    27: 0000000000003950     0 FUNC    GLOBAL DEFAULT    9 _init
    28: 0000000000003b60     0 FUNC    GLOBAL DEFAULT   13 _fini
    29: 0000000000000000     0 NOTYPE  LOCAL  DEFAULT  UND
readelf: Warning: local symbol 29 found at index >= .dynsym's sh_info value of 1
    30: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND __gmon_start__
    31: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND _Jv_RegisterClasses
    32: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZNSt8ios_base4InitC1Ev
    33: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZNSt8ios_base4InitD1Ev
    34: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND _ITM_deregisterTMCloneTab
    35: 0000000000000000     0 OBJECT  GLOBAL DEFAULT  UND _ZSt4cout
    36: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZNSolsEPFRSoS_E
    37: 0000000000003b60     0 FUNC    GLOBAL DEFAULT   13 _fini

With this change:

# make
information of .dynsym section changes from 1 to 2
symndx of .gnu.hash section changes from 13 to 26
symndx of .gnu.hash section changes from 13 to 26
symndx of .gnu.hash section changes from 13 to 26
chmod u+x libfoo-modified.so
readelf --dyn-syms libfoo-modified.so

Symbol table '.dynsym' contains 38 entries:
   Num:    Value          Size Type    Bind   Vis      Ndx Name
     0: 0000000000000000     0 NOTYPE  LOCAL  DEFAULT  UND
     1: 0000000000000000     0 NOTYPE  LOCAL  DEFAULT  UND
     2: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND __gmon_start__
     3: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND _Jv_RegisterClasses
     4: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZNSt8ios_base4InitC1Ev@GLIBCXX_3.4 (2)
     5: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND __cxa_atexit@GLIBC_2.2.5 (3)
     6: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZNSt8ios_base4InitD1Ev@GLIBCXX_3.4 (2)
     7: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND _ITM_deregisterTMCloneTab
     8: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZStlsISt11char_traitsIcE@GLIBCXX_3.4 (2)
     9: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND _ITM_registerTMCloneTable
    10: 0000000000000000     0 FUNC    WEAK   DEFAULT  UND __cxa_finalize@GLIBC_2.2.5 (3)
    11: 0000000000000000     0 OBJECT  GLOBAL DEFAULT  UND _ZSt4cout@GLIBCXX_3.4 (2)
    12: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZNSolsEPFRSoS_E@GLIBCXX_3.4 (2)
    13: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZSt4endlIcSt11char_trait@GLIBCXX_3.4 (2)
    14: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND __gmon_start__
    15: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND _Jv_RegisterClasses
    16: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZNSt8ios_base4InitC1Ev
    17: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND __cxa_atexit
    18: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZNSt8ios_base4InitD1Ev
    19: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND _ITM_deregisterTMCloneTab
    20: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZStlsISt11char_traitsIcE
    21: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND _ITM_registerTMCloneTable
    22: 0000000000000000     0 FUNC    WEAK   DEFAULT  UND __cxa_finalize
    23: 0000000000000000     0 OBJECT  GLOBAL DEFAULT  UND _ZSt4cout
    24: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZNSolsEPFRSoS_E
    25: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND _ZSt4endlIcSt11char_trait
    26: 0000000000204210     0 NOTYPE  GLOBAL DEFAULT   24 _end
    27: 000000000020420c     0 NOTYPE  GLOBAL DEFAULT   23 _edata
    28: 0000000000204210     0 NOTYPE  GLOBAL DEFAULT   24 _end
    29: 000000000020420c     0 NOTYPE  GLOBAL DEFAULT   23 _edata
    30: 0000000000003afa     7 FUNC    GLOBAL DEFAULT   12 _Z3foov
    31: 000000000020420c     0 NOTYPE  GLOBAL DEFAULT   24 __bss_start
    32: 0000000000003950     0 FUNC    GLOBAL DEFAULT    9 _init
    33: 0000000000003afa     7 FUNC    GLOBAL DEFAULT   12 _Z3foov
    34: 000000000020420c     0 NOTYPE  GLOBAL DEFAULT   24 __bss_start
    35: 0000000000003950     0 FUNC    GLOBAL DEFAULT    9 _init
    36: 0000000000003b60     0 FUNC    GLOBAL DEFAULT   13 _fini
    37: 0000000000003b60     0 FUNC    GLOBAL DEFAULT   13 _fini

Clcanny · 2020-12-06T15:42:35Z

@romainthomas Hello, please review this change. Thank you!

romainthomas · 2020-12-06T17:51:19Z

Hi @Clcanny
Thanks for the PR. Is there any reference of such sorting in LLVM linker / binutils or did you implement the sort with your own method ?

Clcanny · 2020-12-07T02:55:57Z

Hi @romainthomas ,

Imported symbols should be sorted before exported symbols: I have investigated .gnu.hash section and read do_lookup_x function in ld-linux-x86-64.so.2.
Local symbols should be sorted before global symbols: I just want to fix warning "local symbol 29 found at index >= .dynsym's sh_info value of 1" of readelf, I am not so confident of this sorting.

Clcanny · 2020-12-07T02:58:29Z

2. Local symbols should be sorted before global symbols

FYI. https://sourceware.org/bugzilla/show_bug.cgi?id=24857

romainthomas · 2020-12-07T06:54:45Z

What about: https://github.com/llvm/llvm-project/blob/2fc704a0a529dd7eba7566a293f981a86bfa5c3e/lld/ELF/SyntheticSections.cpp#L2081-L2086 ? which seems to avoid two stable_sort ?

Clcanny · 2020-12-07T07:13:42Z

According to llvm, we need to sort dynamic symbol table 2 times:

Let undefined symbols come before defined symbols: https://github.com/llvm/llvm-project/blob/2fc704a0a529dd7eba7566a293f981a86bfa5c3e/lld/ELF/SyntheticSections.cpp#L2415

std::vector<SymbolTableEntry>::iterator mid =
    std::stable_partition(v.begin(), v.end(), [&](const SymbolTableEntry &s) {
      return !s.sym->isDefined() || s.sym->partition != partition;
    });

Sort by bucket ids: https://github.com/llvm/llvm-project/blob/2fc704a0a529dd7eba7566a293f981a86bfa5c3e/lld/ELF/SyntheticSections.cpp#L2441

llvm::stable_sort(symbols, [](const Entry &l, const Entry &r) {
  return l.bucketIdx < r.bucketIdx;
});

Besides the above two sorting, we need another sort (3) to find local symbols (null entries).

I am going to combine the first sort and the third sort, and do other things to improve my code.

Thanks for your guide!

Clcanny · 2020-12-10T17:19:07Z

@romainthomas Hi, I have changed my code. I think calling stable_partition twice is needed, it makes code easy to understand.

And I add a unit test for this change. There is one thing worth noting: size of libadd.dynamic_symbols is bigger than the real size.

romainthomas · 2020-12-11T06:29:24Z

Perfect @Clcanny 👍
I merged the pr through the commit 3613f47 and I updated the changelog as well

Clcanny · 2020-12-11T06:34:04Z

Perfect @Clcanny 👍
I merged the pr through the commit 3613f47 and I updated the changelog as well

Thanks! But the ut still has some problems when running in Travis CI while it is okay when running on my local machine. I will fix it and commit it.

romainthomas · 2020-12-11T06:59:16Z

Oh yes I mess that. Open a PR with your fixe when it's done

romainthomas · 2020-12-21T15:08:30Z

To be reopened

Clcanny force-pushed the charles-sort-dynamic-symbols branch from 64ea440 to 1d3a1d9 Compare December 10, 2020 16:56

Clcanny force-pushed the charles-sort-dynamic-symbols branch from 1d3a1d9 to 2da3560 Compare December 10, 2020 17:44

Sort dynamic symbols

50a348e

Clcanny force-pushed the charles-sort-dynamic-symbols branch from 2da3560 to 50a348e Compare December 11, 2020 03:23

romainthomas merged commit 50a348e into lief-project:master Dec 11, 2020

romainthomas added a commit that referenced this pull request Dec 11, 2020

Comment until the CI is fixed (related to #507)

fe6f2a0

Clcanny added a commit to Clcanny/LIEF that referenced this pull request Dec 12, 2020

Fix bug of unittest (related to lief-project#507)

f4aad63

Clcanny added a commit to Clcanny/LIEF that referenced this pull request Dec 16, 2020

Fix bug of unittest (related to lief-project#507)

71941f8

Clcanny added a commit to Clcanny/LIEF that referenced this pull request Dec 16, 2020

Fix bug of unittest (related to lief-project#507)

c7d43bd

Clcanny added a commit to Clcanny/LIEF that referenced this pull request Dec 17, 2020

Fix bug of unittest (related to lief-project#507)

4bc7865

Clcanny added a commit to Clcanny/LIEF that referenced this pull request Dec 17, 2020

Fix bug of unittest (related to lief-project#507)

1977479

Clcanny added a commit to Clcanny/LIEF that referenced this pull request Dec 21, 2020

Fix bug of unittest (related to lief-project#507)

7b36d86

Clcanny added a commit to Clcanny/LIEF that referenced this pull request Dec 23, 2020

Fix bug of unittest (related to lief-project#507)

4ddd109

romainthomas pushed a commit that referenced this pull request Jan 31, 2021

Fix bug of unittest (related to #507)

3d0da01

jheydebrand mentioned this pull request May 27, 2021

.dynsym not sorted by symbol visibility when using sysv hash style #582

Closed

romainthomas added a commit that referenced this pull request Jan 17, 2022

Comment until the CI is fixed (related to #507)

1b50429

Sort dynamic symbols #507

Sort dynamic symbols #507

Clcanny commented Dec 6, 2020 •

edited

Clcanny commented Dec 6, 2020

romainthomas commented Dec 6, 2020

Clcanny commented Dec 7, 2020 •

edited

Clcanny commented Dec 7, 2020

romainthomas commented Dec 7, 2020

Clcanny commented Dec 7, 2020 •

edited

Clcanny commented Dec 10, 2020 •

edited

romainthomas commented Dec 11, 2020

Clcanny commented Dec 11, 2020 •

edited

romainthomas commented Dec 11, 2020

romainthomas commented Dec 21, 2020

Sort dynamic symbols #507

Sort dynamic symbols #507

Conversation

Clcanny commented Dec 6, 2020 • edited

Clcanny commented Dec 6, 2020

romainthomas commented Dec 6, 2020

Clcanny commented Dec 7, 2020 • edited

Clcanny commented Dec 7, 2020

romainthomas commented Dec 7, 2020

Clcanny commented Dec 7, 2020 • edited

Clcanny commented Dec 10, 2020 • edited

romainthomas commented Dec 11, 2020

Clcanny commented Dec 11, 2020 • edited

romainthomas commented Dec 11, 2020

romainthomas commented Dec 21, 2020

Clcanny commented Dec 6, 2020 •

edited

Clcanny commented Dec 7, 2020 •

edited

Clcanny commented Dec 7, 2020 •

edited

Clcanny commented Dec 10, 2020 •

edited

Clcanny commented Dec 11, 2020 •

edited