[llvm-readobj] Add GNU style dumper for .gnu.version section

[Higuoxing]

Summary

Currently, llvm-readobj do not support GNU style dumper for symbol versioning sections. In this patch, I would like to implement dumper for .gnu.version section

[Higuoxing]

• Add test file

[rupprecht]

Very cool. It looks like -V also prints a definition/needed section that this patch doesn't support yet. Is that coming in a later patch?

i.e.,

$ readelf -V ~/src/llvm-project/llvm/test/tools/llvm-readobj/Inputs/verdef.elf-x86-64

Version symbols section '.gnu.version' contains 8 entries:
 Addr: 000000000000024c  Offset: 0x00024c  Link: 1 (.dynsym)
  000:   0 (*local*)       1 (*global*)      1 (*global*)      3 (VERSION2)
  004:   1 (*global*)      2 (VERSION1)      2 (VERSION1)      3 (VERSION2)

# This block is missing from this patch
Version definition section '.gnu.version_d' contains 3 entries:
  Addr: 0x000000000000025c  Offset: 0x00025c  Link: 2 (.dynstr)
  000000: Rev: 1  Flags: BASE  Index: 1  Cnt: 1  Name: blah
  0x001c: Rev: 1  Flags: none  Index: 2  Cnt: 1  Name: VERSION1
  0x0038: Rev: 1  Flags: none  Index: 3  Cnt: 2  Name: VERSION2
  0x0054: Parent 1: VERSION1

$ readelf -V ~/src/llvm-project/llvm/test/tools/llvm-readobj/Inputs/verneed.elf-x86-64 

Version symbols section '.gnu.version' contains 4 entries:
 Addr: 0000000000010228  Offset: 0x000228  Link: 1 (.dynsym)
  000:   0 (*local*)       2 (v3)            3 (v2)            4 (v1)         

# This block is missing from this patch
Version needs section '.gnu.version_r' contains 2 entries:
 Addr: 0x0000000000010230  Offset: 0x000230  Link: 5 (.dynstr)
  000000: Version: 1  File: verneed1.so.0  Cnt: 2
  0x0020:   Name: v2  Flags: none  Version: 3
  0x0030:   Name: v3  Flags: none  Version: 2
  0x0010: Version: 1  File: verneed2.so.0  Cnt: 1
  0x0040:   Name: v1  Flags: none  Version: 4

[Higuoxing]

In D59877#1445270, @rupprecht wrote:

Very cool. It looks like -V also prints a definition/needed section that this patch doesn't support yet. Is that coming in a later patch?

Thanks for reviewing! I have implemented prototype dumpers for them, and I would like to introduce them in next series of patches.

[jhenderson]

llvm/tools/llvm-readobj/ELFDumper.cpp

3358–3408

Use uint64_t (or possibly size_t) here. ELF64 sections have section sizes of 64-bits.

Various variables below will similarly need adjusting.

3373

Does this behaviour hold if using --wide in GNU readelf?

3391

Test case for hidden symbols?

3401

What happens if the version name is longer than this? (Is that even possible?)

[Higuoxing]

Addressed @jhenderson 's comments

• Add test case for invalid versioning symbol.

[Higuoxing]

llvm/tools/llvm-readobj/ELFDumper.cpp

3373

No, in GNU readelf, --wide option has no effect on -V option.

3391

Done.

3401

Yes, it's possible. But GNU readelf does not do any improvement here, shall we?

[Higuoxing]

.

[jhenderson]

llvm/test/tools/llvm-readobj/elf-hidden-versym.test

32–41 ↗

(On Diff #193420)

Could you add less padding between the Name and other fields, and the values for those fields, please? Ditto in the other new test.

llvm/tools/llvm-readobj/ELFDumper.cpp

3375

Entries is now a uin64_t, so VersymRow and VersymIndex below should be too.

3401

I think it's okay to leave for now, but you might want to consider improving it in a future version, by using a column width that varies depending on name widths.

[Higuoxing]

• Using uint64_t instead of uint32_t

[Higuoxing]

llvm/tools/llvm-readobj/ELFDumper.cpp

3401

Thanks, I will consider improve this behavior

[jhenderson]

LGTM.

[Higuoxing]

In D59877#1453108, @jhenderson wrote:

LGTM.

Thanks for reviewing!

[Higuoxing]

Seems that this patch breaks some tests, I will take a look

[grimar]

In D59877#1453320, @Higuoxing wrote:

Seems that this patch breaks some tests, I will take a look

It was me. Fixed in https://reviews.llvm.org/rL357598

[Higuoxing]

In D59877#1453327, @grimar wrote:

In D59877#1453320, @Higuoxing wrote:

Seems that this patch breaks some tests, I will take a look

It was me. Fixed in https://reviews.llvm.org/rL357598

Thank you very much :)

llvm/test/tools/llvm-readobj/elf-invalid-versioning.test

+# RUN: yaml2obj %s -o %t
+# RUN: llvm-readelf -V %t | FileCheck %s --check-prefix=INVALID
+
+# INVALID:      Version symbols section '.gnu.version' contains 2 entries:
+# INVALID-NEXT:  Addr: 0000000000200210  Offset: 0x000240  Link: 6 (.dynsym)
+# INVALID-NEXT:   000:   0 (*local*)       3 (*invalid*)
+
+--- !ELF
+FileHeader:
+  Class:           ELFCLASS64
+  Data:            ELFDATA2LSB
+  Type:            ET_EXEC
+  Machine:         EM_X86_64
+  Entry:           0x0000000000201000
+Sections:
+  - Name:            .gnu.version
+    Type:            SHT_GNU_versym
+    Flags:           [ SHF_ALLOC ]
+    Address:         0x0000000000200210
+    Link:            .dynsym
+    AddressAlign:    0x0000000000000002
+    EntSize:         0x0000000000000002
+    Entries:         [ 0, 3 ]
+  - Name:            .gnu.version_r
+    Type:            SHT_GNU_verneed
+    Flags:           [ SHF_ALLOC ]
+    Address:         0x0000000000200250
+    Link:            .dynstr
+    AddressAlign:    0x0000000000000004
+    Info:            0x0000000000000001
+    Dependencies:
+      - Version:         1
+        File:            somefile
+        Entries:
+          - Name:            ''         # invalid name
+            Hash:            1937
+            Flags:           233
+            Other:           3
+DynamicSymbols:
+  Global:
+    - Name:            f
+...

llvm/test/tools/llvm-readobj/elf-versioninfo.test

 LLVM-VERDEF-NEXT:     Predecessor: VERSION1
 LLVM-VERDEF-NEXT:   }
 LLVM-VERDEF-NEXT: }
 
 GNU-VERDEF: 0x000000006ffffff0 VERSYM               0x24c
 GNU-VERDEF: 0x000000006ffffffc VERDEF               0x25c
 GNU-VERDEF: 0x000000006ffffffd VERDEFNUM            3
 
-GNU-VERDEF: Dumper for .gnu.version is not implemented
+GNU-VERDEF:      Version symbols section '.gnu.version' contains 8 entries:
+GNU-VERDEF-NEXT:  Addr: 000000000000024c  Offset: 0x00024c  Link: 1 (.dynsym)
+GNU-VERDEF-NEXT:   000:   0 (*local*)       1 (*global*)      1 (*global*)      3 (VERSION2)
+GNU-VERDEF-NEXT:   004:   1 (*global*)      2 (VERSION1)      2 (VERSION1)      3 (VERSION2)
+
 GNU-VERDEF: Dumper for .gnu.version_d is not implemented
 
 RUN: llvm-readobj -V %p/Inputs/verneed.elf-x86-64 | FileCheck %s --check-prefix=LLVM-VERNEED
 RUN: llvm-readelf -V %p/Inputs/verneed.elf-x86-64 | FileCheck %s --check-prefix=GNU-VERNEED
 
 LLVM-VERNEED:       SHT_GNU_verneed {
 LLVM-VERNEED-NEXT:   Dependency {
 LLVM-VERNEED-NEXT:     Version: 1

 LLVM-VERNEED-NEXT:       Hash: 1937
 LLVM-VERNEED-NEXT:       Flags: 0x0
 LLVM-VERNEED-NEXT:       Index: 4
 LLVM-VERNEED-NEXT:       Name: v1
 LLVM-VERNEED-NEXT:     }
 LLVM-VERNEED-NEXT:   }
 LLVM-VERNEED-NEXT: }
 
-GNU-VERNEED: Dumper for .gnu.version is not implemented
+GNU-VERNEED:      Version symbols section '.gnu.version' contains 4 entries:
+GNU-VERNEED-NEXT:  Addr: 0000000000010228  Offset: 0x000228  Link: 1 (.dynsym)
+GNU-VERNEED-NEXT:   000:   0 (*local*)       2 (v3)            3 (v2)            4 (v1)
+
 GNU-VERNEED: Dumper for .gnu.version_r is not implemented
 

llvm/tools/llvm-readobj/ELFDumper.cpp

 
 template <class ELFT>
 void GNUStyle<ELFT>::printVersionSymbolSection(const ELFFile<ELFT> *Obj,
                                                const Elf_Shdr *Sec) {
   if (!Sec)
     return;
 
   StringRef SecName = unwrapOrError(Obj->getSectionName(Sec));
-  OS << "Dumper for " << SecName << " is not implemented\n";
+  uint64_t Entries = Sec->sh_size / sizeof(Elf_Versym);
+
+  OS << "Version symbols section '" << SecName << "' "
+     << "contains " << Entries << " entries:\n";
+
+  const Elf_Shdr *SymTab = unwrapOrError(Obj->getSection(Sec->sh_link));
+  StringRef SymTabName = unwrapOrError(Obj->getSectionName(SymTab));
+  OS << " Addr: " << format_hex_no_prefix(Sec->sh_addr, 16)
+     << "  Offset: " << format_hex(Sec->sh_offset, 8)
+     << "  Link: " << Sec->sh_link << " (" << SymTabName << ")\n";
+
+  const uint8_t *VersymBuf =
+      reinterpret_cast<const uint8_t *>(Obj->base() + Sec->sh_offset);
+  const ELFDumper<ELFT> *Dumper = this->dumper();
+  StringRef StrTable = Dumper->getDynamicStringTable();
+
+  // readelf prints 4 entries per line.
+  for (uint32_t VersymRow = 0; VersymRow < Entries; VersymRow += 4) {
+    OS << "  " << format_hex_no_prefix(VersymRow, 3) << ":";
+
+    for (uint32_t VersymIndex = 0;
+         (VersymIndex < 4) && (VersymIndex + VersymRow) < Entries;
+         ++VersymIndex) {
+      const Elf_Versym *Versym =
+          reinterpret_cast<const Elf_Versym *>(VersymBuf);
+      switch (Versym->vs_index) {
+      case 0:
+        OS << "   0 (*local*)    ";
+        break;
+      case 1:
+        OS << "   1 (*global*)   ";
+        break;
+      default:
+        OS << format("%4x%c", Versym->vs_index & VERSYM_VERSION,
+                     Versym->vs_index & VERSYM_HIDDEN ? 'h' : ' ');
+
+        bool IsDefault = true;
+        std::string VersionName = Dumper->getSymbolVersionByIndex(
+            StrTable, Versym->vs_index, IsDefault);
+
+        if (!VersionName.empty())
+          VersionName = "(" + VersionName + ")";
+        else
+          VersionName = "(*invalid*)";
+        OS << left_justify(VersionName, 13);
+      }
+      VersymBuf += sizeof(Elf_Versym);
+    }
+    OS << '\n';
+  }
+  OS << '\n';
 }
 
 template <class ELFT>
 void GNUStyle<ELFT>::printVersionDefinitionSection(const ELFFile<ELFT> *Obj,
                                                    const Elf_Shdr *Sec) {
   if (!Sec)
     return;