[llvm-objcopy][ELF] Implement --only-keep-debug

--only-keep-debug produces a debug file as the output that only
preserves contents of sections useful for debugging purposes (the
binutils implementation preserves SHT_NOTE and non-SHF_ALLOC sections),
by changing their section types to SHT_NOBITS and rewritting file
offsets.

See https://sourceware.org/gdb/onlinedocs/gdb/Separate-Debug-Files.html

The intended use case is:

```
llvm-objcopy --only-keep-debug a a.dbg
llvm-objcopy --strip-debug a b
llvm-objcopy --add-gnu-debuglink=a.dbg b
```

The current layout algorithm is incapable of deleting contents and
shrinking segments, so it is not suitable for implementing the
functionality.

This patch adds a new algorithm which assigns sh_offset to sections
first, then modifies p_offset/p_filesz of program headers. It bears a
resemblance to lld/ELF/Writer.cpp.

Reviewed By: jhenderson, jakehehrlich

Differential Revision: https://reviews.llvm.org/D67137

1 files changed, 224 insertions, 0 deletions

diff --git a/llvm/test/tools/llvm-objcopy/ELF/only-keep-debug.test b/llvm/test/tools/llvm-objcopy/ELF/only-keep-debug.test
new file mode 100644
index 00000000000..1dc29c97365
--- /dev/null
+++ b/llvm/test/tools/llvm-objcopy/ELF/only-keep-debug.test
@@ -0,0 +1,224 @@
+# RUN: yaml2obj --docnum=1 %s -o %t1
+# RUN: llvm-objcopy --only-keep-debug %t1 %t1.dbg
+# RUN: llvm-readelf -S -l -x .note1 -x .note2 -x .debug_abbrev -x .debug_frame -x .debug_info %t1.dbg | FileCheck %s
+
+## Check that SHT_NOTE and .debug* are kept, but others are changed to SHT_NOBITS.
+## SHT_NOBITS sections do not occupy space in the output.
+
+# CHECK:      [Nr] Name          Type     Address          Off    Size   ES Flg Lk Inf Al
+# CHECK:      [ 1] .note1        NOTE     0000000000000400 000400 000001 00   A  0   0 1024
+# CHECK-NEXT: [ 2] .note2        NOTE     0000000000000401 000401 000001 00   A  0   0  0
+# CHECK-NEXT: [ 3] .text         NOBITS   0000000000000402 000402 000001 00  AX  0   0  0
+# CHECK-NEXT: [ 4] .tdata        NOBITS   0000000000001480 000480 000007 00 WAT  0   0 128
+# CHECK-NEXT: [ 5] .tbss         NOBITS   0000000000001487 000480 000005 00 WAT  0   0  0
+# CHECK-NEXT: [ 6] .bss          NOBITS   00000000000014a0 000480 00003f 00  WA  0   0 32
+## objcopy sets sh_offset to 0x402. We don't do this to keep sh_offset non-decreasing.
+# CHECK-NEXT: [ 7] .debug_abbrev PROGBITS 0000000000000000 000480 000001 00      0   0  0
+# CHECK-NEXT: [ 8] .debug_frame  PROGBITS 0000000000000000 000488 000001 00      0   0  8
+# CHECK-NEXT: [ 9] .debug_info   PROGBITS 0000000000000000 000489 000001 00      0   0  0
+# CHECK-NEXT: [10] .strtab       STRTAB   0000000000000000 00048a 000001 00      0   0  1
+# CHECK-NEXT: [11] .shstrtab     STRTAB   0000000000000000 00048b 000060 00      0   0  1
+
+# CHECK:      Type Offset   VirtAddr           PhysAddr           FileSiz  MemSiz   Flg Align
+# CHECK-NEXT: LOAD 0x000000 0x0000000000000000 0x0000000000000000 0x000402 0x000403 R E 0x1000
+# CHECK-NEXT: LOAD 0x000480 0x0000000000001480 0x0000000000000000 0x000000 0x00005f RW  0x1000
+# CHECK-NEXT: TLS  0x000480 0x0000000000001480 0x0000000000000000 0x000000 0x00000c RW  0x80
+# CHECK-NEXT: NOTE 0x000400 0x0000000000000400 0x0000000000000000 0x000002 0x000002     0x400
+
+## Contents of SHT_NOTE and .debug* are kept.
+
+# CHECK:      Hex dump of section '.note1':
+# CHECK-NEXT: 0x00000400 01
+# CHECK:      Hex dump of section '.note2':
+# CHECK-NEXT: 0x00000401 02
+# CHECK:      Hex dump of section '.debug_abbrev':
+# CHECK-NEXT: 0x00000000 03
+# CHECK:      Hex dump of section '.debug_frame':
+# CHECK-NEXT: 0x00000000 04
+# CHECK:      Hex dump of section '.debug_info':
+# CHECK-NEXT: 0x00000000 05
+
+--- !ELF
+FileHeader:
+  Class:     ELFCLASS64
+  Data:      ELFDATA2LSB
+  Type:      ET_DYN
+  Machine:   EM_X86_64
+Sections:
+  - Name:         .note1
+    Type:         SHT_NOTE
+    Flags:        [ SHF_ALLOC ]
+    Address:      0x400
+    AddressAlign: 0x400
+    Content:      01
+  - Name:         .note2
+    Type:         SHT_NOTE
+    Flags:        [ SHF_ALLOC ]
+    Address:      0x401
+    Content:      02
+  - Name:         .text
+    Type:         SHT_PROGBITS
+    Flags:        [ SHF_ALLOC, SHF_EXECINSTR ]
+    Address:      0x402
+    Content:      c3
+  - Name:         .tdata
+    Type:         SHT_PROGBITS
+    Flags:        [ SHF_ALLOC, SHF_WRITE, SHF_TLS ]
+    Address:      0x1480          # Ensure Address=0x1000+Offset
+    AddressAlign: 0x80
+    # An arbitrary non-zero Size tests that .tdata does not occupy space
+    # and we can rewrite p_filesz of PT_TLS.
+    Size:         7
+  - Name:         .tbss
+    Type:         SHT_NOBITS
+    Flags:        [ SHF_ALLOC, SHF_WRITE, SHF_TLS ]
+    Address:      0x1487          # Ensure Address=0x1000+Offset
+    Size:         5
+  - Name:         .bss
+    Type:         SHT_NOBITS
+    Flags:        [ SHF_ALLOC, SHF_WRITE ]
+    Address:      0x14a0          # Ensure Address=0x1000+Offset
+    AddressAlign: 0x20
+    # An arbitrary non-zero Size tests that .bss does not occupy space.
+    Size:         63
+  - Name:         .debug_abbrev
+    Type:         SHT_PROGBITS
+    Content:      03
+  - Name:         .debug_frame
+    Type:         SHT_PROGBITS
+    # AddressAlign tests the file offset assignment leaves a gap.
+    AddressAlign: 0x8
+    Content:      04
+  - Name:         .debug_info
+    Type:         SHT_PROGBITS
+    Content:      05
+ProgramHeaders:
+  - Type:     PT_LOAD
+    Flags:    [ PF_R, PF_X ]
+    Offset:   0
+    Align:    0x1000
+    Sections:
+      - Section: .note1
+      - Section: .note2
+      - Section: .text
+  - Type:     PT_LOAD
+    Flags:    [ PF_R, PF_W ]
+    VAddr:    0x1480    # Ensure Offset=VAddr (mod Align) if Offset changes
+    Align:    0x1000
+    Sections:
+      - Section: .tdata
+      - Section: .bss
+  - Type:     PT_TLS
+    Flags:    [ PF_R, PF_W ]
+    VAddr:    0x1480    # Ensure Offset=VAddr (mod Align) if Offset changes
+    Sections:
+      - Section: .tdata
+      - Section: .tbss
+  - Type:     PT_NOTE
+    VAddr:    0x400
+    Sections:
+      - Section: .note1
+      - Section: .note2
+...
+
+# RUN: yaml2obj --docnum=2 %s -o %t2
+# RUN: llvm-objcopy --only-keep-debug %t2 %t2.dbg
+# RUN: llvm-readelf -S -l %t2.dbg | FileCheck --check-prefix=CHECK2 %s
+
+## Only the tail of a segment can be trimmed. .text still occupies space because
+## it is followed by .note which is not SHT_NOBITS.
+# CHECK2:      [Nr] Name        Type     Address          Off    Size   ES Flg Lk Inf Al
+# CHECK2:      [ 1] .text       NOBITS   0000000000000200 000200 000001 00  AX  0   0 512
+# CHECK2-NEXT: [ 2] .note       NOTE     0000000000000201 000201 000001 00   A  0   0  0
+# CHECK2-NEXT: [ 3] .debug_info PROGBITS 0000000000000000 000220 000001 00      0   0 32
+# CHECK2-NEXT: [ 4] .strtab     STRTAB   0000000000000000 000221 000001 00      0   0  1
+# CHECK2-NEXT: [ 5] .shstrtab   STRTAB   0000000000000000 000222 00002b 00      0   0  1
+
+## Check that p_offset or p_filesz of empty segments or PT_PHDR are not modified.
+# CHECK2:      Type Offset   VirtAddr           PhysAddr           FileSiz  MemSiz   Flg Align
+# CHECK2-NEXT: PHDR 0x000040 0x0000000000000040 0x0000000000000000 0x0000a8 0x0000a8 R   0x8
+# CHECK2-NEXT: LOAD 0x000000 0x0000000000000000 0x0000000000000000 0x000202 0x000202 R E 0x1000
+# CHECK2-NEXT: LOAD 0x000202 0x0000000000000202 0x0000000000000000 0x00000e 0x00000e RW  0x1
+
+--- !ELF
+FileHeader:
+  Class:   ELFCLASS64
+  Data:    ELFDATA2LSB
+  Type:    ET_DYN
+  Machine: EM_X86_64
+Sections:
+  - Name:         .text
+    Type:         SHT_PROGBITS
+    Flags:        [ SHF_ALLOC, SHF_EXECINSTR ]
+    Address:      0x200
+    AddressAlign: 0x200
+    Content:      c3
+  - Name:         .note
+    Type:         SHT_NOTE
+    Flags:        [ SHF_ALLOC ]
+    Address:      0x201
+    Content:      01
+  - Name:         .debug_info
+    Type:         SHT_PROGBITS
+    AddressAlign: 0x20
+    Content:      02
+ProgramHeaders:
+  - Type:     PT_PHDR
+    Flags:    [ PF_R ]
+    Offset:   0x40
+    VAddr:    0x40
+    # 3 * sizeof(Elf64_Phdr) = 0xa8
+    FileSize: 0xa8
+    MemSize:  0xa8
+    Align:    8
+  - Type:     PT_LOAD
+    Flags:    [ PF_R, PF_X ]
+    Offset:   0
+    Align:    4096
+    Sections:
+      - Section: .text
+      - Section: .note
+  - Type:     PT_LOAD
+    Flags:    [ PF_R, PF_W ]
+    Offset:   0x202
+    VAddr:    0x202
+    FileSize: 14
+    MemSize:  14
+...
+
+## If .symtab or .strtab has the SHF_ALLOC flag, it will be changed to SHT_NOBITS.
+# RUN: yaml2obj --docnum=3 %s -o %t3
+# RUN: llvm-objcopy --only-keep-debug %t3 %t3.dbg
+# RUN: llvm-readelf -S -l %t3.dbg | FileCheck --check-prefix=CHECK3 %s
+
+# CHECK3:      [Nr] Name          Type     Address          Off    Size   ES Flg Lk Inf Al
+# CHECK3:      [ 1] .dynsym       NOBITS   0000000000000000 000040 000018 18   A  2   1 1024
+# CHECK3-NEXT: [ 2] .dynstr       NOBITS   0000000000000000 000040 000001 00   A  0   0  0
+# CHECK3-NEXT: [ 3] .symtab       NOBITS   0000000000000000 000040 000018 18   A  4   1  0
+# CHECK3-NEXT: [ 4] .strtab       NOBITS   0000000000000000 000040 000001 00   A  0   0  0
+# CHECK3-NEXT: [ 5] .shstrtab     STRTAB   0000000000000000 000040 00002b 00      0   0  1
+
+--- !ELF
+FileHeader:
+  Class:   ELFCLASS64
+  Data:    ELFDATA2LSB
+  Type:    ET_DYN
+  Machine: EM_X86_64
+Sections:
+  - Name:  .dynsym
+    Type:  SHT_DYNSYM
+    Flags: [ SHF_ALLOC ]
+    Link:  .dynstr
+    AddressAlign: 0x400
+  - Name:  .dynstr
+    Type:  SHT_STRTAB
+    Flags: [ SHF_ALLOC ]
+  - Name:  .symtab
+    Type:  SHT_STRTAB
+    Flags: [ SHF_ALLOC ]
+    Link:  .strtab
+  - Name:  .strtab
+    Type:  SHT_STRTAB
+    Flags: [ SHF_ALLOC ]
+DynamicSymbols: []
+Symbols: []