[llvm-objcopy] Add --compress-sections

--compress-sections is similar to --compress-debug-sections but applies
to arbitrary sections.

* `--compress-sections <section>=none`: decompress sections
* `--compress-sections <section>=[zlib|zstd]`: compress sections with zlib/zstd

Like `--remove-section`, the pattern is by default a glob, but a regex
when --regex is specified.

For `--remove-section` like options, `!` prevents matches and is not
dependent on ordering (see `ELF/wildcard-syntax.test`). Since
`--compress-sections a=zlib --compress-sections a=none` naturally allows
overriding, having an order-independent `!` would be confusing.
Therefore, `!` is disallowed.

Sections within a segment are effectively immutable. Report an error for
an attempt to (de)compress them. `SHF_ALLOC` sections in a relocatable
file can be compressed, but linkers usually reject them.

Note: Before this patch, a compressed relocation section is recognized
as a `RelocationSectionBase` as well and `removeSections` `!ToRemove(*ToRelSec)`
may incorrectly interpret a `CompressedSections` as `RelocationSectionBase`,
leading to ubsan failure for the new test. Fix this by setting
`OriginalFlags` in CompressedSection::CompressedSection.

Link: https://discourse.llvm.org/t/rfc-compress-arbitrary-sections-with-ld-lld-compress-sections/71674

Pull Request: llvm#85036

llvm/docs/CommandGuide/llvm-objcopy.rst

@@ -309,6 +309,14 @@ them.
  Compress DWARF debug sections in the output, using the specified format.
  Supported formats are ``zlib`` and ``zstd``. Use ``zlib`` if ``<format>`` is omitted.
 
+.. option:: --compress-sections <section>=<format>
+
+ Compress or decompress sections matched by ``<section>`` using the specified
+ format. Supported formats are ``zlib`` and ``zstd``. Specify ``none`` for
+ decompression. When a section is matched by multiple options, the last one
+ wins. A wildcard ``<section>`` starting with '!' is disallowed.
+ Sections within a segment cannot be (de)compressed.
+
 .. option:: --decompress-debug-sections
 
  Decompress any compressed DWARF debug sections in the output.

llvm/docs/ReleaseNotes.rst

@@ -186,6 +186,10 @@ Changes to the LLVM tools
   for ELF input to skip the specified symbols when executing other options
   that can change a symbol's name, binding or visibility.
 
+* llvm-objcopy now supports ``--compress-sections`` to compress or decompress
+  arbitrary sections not within a segment.
+  (`#85036 <https://github.com/llvm/llvm-project/pull/85036>`_.)
+
 * llvm-profgen now supports COFF+DWARF binaries. This enables Sample-based PGO
   on Windows using Intel VTune's SEP. For details on usage, see the `end-user
   documentation for SPGO

llvm/include/llvm/ObjCopy/CommonConfig.h

@@ -262,6 +262,9 @@ struct CommonConfig {
   bool DecompressDebugSections = false;
 
   DebugCompressionType CompressionType = DebugCompressionType::None;
+
+  SmallVector<std::pair<NameMatcher, llvm::DebugCompressionType>, 0>
+      compressSections;
 };
 
 } // namespace objcopy

llvm/lib/ObjCopy/ELF/ELFObjcopy.cpp

@@ -215,23 +215,41 @@ static Error dumpSectionToFile(StringRef SecName, StringRef Filename,
 }
 
 Error Object::compressOrDecompressSections(const CommonConfig &Config) {
-  // Build a list of the debug sections we are going to replace.
-  // We can't call `AddSection` while iterating over sections,
+  // Build a list of sections we are going to replace.
+  // We can't call `addSection` while iterating over sections,
   // because it would mutate the sections array.
   SmallVector<std::pair<SectionBase *, std::function<SectionBase *()>>, 0>
       ToReplace;
   for (SectionBase &Sec : sections()) {
-    if ((Sec.Flags & SHF_ALLOC) || !StringRef(Sec.Name).starts_with(".debug"))
+    std::optional<DebugCompressionType> CType;
+    for (auto &[Matcher, T] : Config.compressSections)
+      if (Matcher.matches(Sec.Name))
+        CType = T;
+    // Handle --compress-debug-sections and --decompress-debug-sections, which
+    // apply to non-ALLOC debug sections.
+    if (!(Sec.Flags & SHF_ALLOC) && StringRef(Sec.Name).starts_with(".debug")) {
+      if (Config.CompressionType != DebugCompressionType::None)
+        CType = Config.CompressionType;
+      else if (Config.DecompressDebugSections)
+        CType = DebugCompressionType::None;
+    }
+    if (!CType)
       continue;
+
+    if (Sec.ParentSegment)
+      return createStringError(
+          errc::invalid_argument,
+          "section '" + Sec.Name +
+              "' within a segment cannot be (de)compressed");
+
     if (auto *CS = dyn_cast<CompressedSection>(&Sec)) {
-      if (Config.DecompressDebugSections) {
+      if (*CType == DebugCompressionType::None)
         ToReplace.emplace_back(
             &Sec, [=] { return &addSection<DecompressedSection>(*CS); });
-      }
-    } else if (Config.CompressionType != DebugCompressionType::None) {
-      ToReplace.emplace_back(&Sec, [&, S = &Sec] {
+    } else if (*CType != DebugCompressionType::None) {
+      ToReplace.emplace_back(&Sec, [=, S = &Sec] {
         return &addSection<CompressedSection>(
-            CompressedSection(*S, Config.CompressionType, Is64Bits));
+            CompressedSection(*S, *CType, Is64Bits));
       });
     }
   }

llvm/lib/ObjCopy/ELF/ELFObject.cpp

@@ -548,6 +548,7 @@ CompressedSection::CompressedSection(const SectionBase &Sec,
                         CompressedData);
 
   Flags |= ELF::SHF_COMPRESSED;
+  OriginalFlags |= ELF::SHF_COMPRESSED;
   size_t ChdrSize = Is64Bits ? sizeof(object::Elf_Chdr_Impl<object::ELF64LE>)
                              : sizeof(object::Elf_Chdr_Impl<object::ELF32LE>);
   Size = ChdrSize + CompressedData.size();
@@ -2161,6 +2162,10 @@ Error Object::removeSections(
       std::begin(Sections), std::end(Sections), [=](const SecPtr &Sec) {
         if (ToRemove(*Sec))
           return false;
+        // TODO: A compressed relocation section may be recognized as
+        // RelocationSectionBase. We don't want such a section to be removed.
+        if (isa<CompressedSection>(Sec))
+          return true;
         if (auto RelSec = dyn_cast<RelocationSectionBase>(Sec.get())) {
           if (auto ToRelSec = RelSec->getSection())
             return !ToRemove(*ToRelSec);

llvm/test/tools/llvm-objcopy/ELF/compress-sections-within-segment.s

@@ -0,0 +1,38 @@
+## Disallow (de)compression for sections within a segment as they are
+## effectively immutable.
+# RUN: rm -rf %t && mkdir %t && cd %t
+# RUN: yaml2obj %s -o a
+# RUN: not llvm-objcopy a /dev/null --compress-sections .text=zlib 2>&1 | FileCheck %s --implicit-check-not=error:
+
+# CHECK: error: 'a': section '.text' within a segment cannot be (de)compressed
+
+# RUN: not llvm-objcopy a /dev/null --compress-sections foo=none 2>&1 | FileCheck %s --check-prefix=CHECK2 --implicit-check-not=error:
+
+# CHECK2: error: 'a': section 'foo' within a segment cannot be (de)compressed
+
+## There is an error even if 'foo' is already compressed with zlib.
+# RUN: not llvm-objcopy a /dev/null --compress-sections foo=zlib 2>&1 | FileCheck %s --check-prefix=CHECK3 --implicit-check-not=error:
+
+# CHECK3: error: 'a': section 'foo' within a segment cannot be (de)compressed
+
+--- !ELF
+FileHeader:
+  Class:      ELFCLASS64
+  Data:       ELFDATA2LSB
+  Type:       ET_EXEC
+  Machine:    EM_X86_64
+ProgramHeaders:
+  - Type:     PT_LOAD
+    FirstSec: .text
+    LastSec:  foo
+    Align:    0x1000
+    Offset:   0x1000
+Sections:
+  - Name:     .text
+    Type:     SHT_PROGBITS
+    Offset:   0x1000
+    Content:  C3
+  - Name:     foo
+    Type:     SHT_PROGBITS
+    Flags:    [ SHF_COMPRESSED ]
+    Content:  010000000000000040000000000000000100000000000000789cd36280002d3269002f800151

llvm/test/tools/llvm-objcopy/ELF/compress-sections.s

@@ -0,0 +1,128 @@
+# REQUIRES: x86-registered-target, zlib, zstd
+
+# RUN: rm -rf %t && mkdir %t && cd %t
+# RUN: llvm-mc -filetype=obj -triple=x86_64 %s -o a.o
+## '*0=none' wins because it is the last. '*0' sections are decompressed (if originally compressed) or kept unchanged (if uncompressed).
+## No section is named 'nomatch'. The third option is a no-op.
+# RUN: llvm-objcopy a.o out --compress-sections='*0=zlib' --compress-sections '*0=none' --compress-sections 'nomatch=none' 2>&1 | count 0
+# RUN: llvm-readelf -S out | FileCheck %s --check-prefix=CHECK1
+
+# CHECK1:      Name           Type          Address     Off      Size     ES Flg Lk Inf Al
+# CHECK1:      .text          PROGBITS [[#%x,TEXT:]]    [[#%x,]] [[#%x,]] 00 AX   0   0  4
+# CHECK1:      foo0           PROGBITS [[#%x,FOO0:]]    [[#%x,]] [[#%x,]] 00 A    0   0  8
+# CHECK1-NEXT: .relafoo0      RELA     [[#%x,]]         [[#%x,]] [[#%x,]] 18  I  11   3  8
+# CHECK1-NEXT: foo1           PROGBITS [[#%x,FOO1:]]    [[#%x,]] [[#%x,]] 00 A    0   0  8
+# CHECK1-NEXT: .relafoo1      RELA     [[#%x,]]         [[#%x,]] [[#%x,]] 18  I  11   5  8
+# CHECK1:      nonalloc0      PROGBITS 0000000000000000 [[#%x,]] [[#%x,]] 00      0   0  8
+# CHECK1-NEXT: .relanonalloc0 RELA     [[#%x,]]         [[#%x,]] [[#%x,]] 18  I  11   7  8
+# CHECK1-NEXT: nonalloc1      PROGBITS 0000000000000000 [[#%x,]] [[#%x,]] 00      0   0  8
+# CHECK1-NEXT: .debug_str     PROGBITS 0000000000000000 [[#%x,]] [[#%x,]] 01 MS   0   0  1
+
+## Mixing zlib and zstd.
+# RUN: llvm-objcopy a.o out2 --compress-sections '*c0=zlib' --compress-sections .debug_str=zstd
+# RUN: llvm-readelf -Sr -x nonalloc0 -x .debug_str out2 2>&1 | FileCheck %s --check-prefix=CHECK2
+# RUN: llvm-readelf -z -x nonalloc0 -x .debug_str out2 | FileCheck %s --check-prefix=CHECK2DE
+
+# CHECK2:      Name           Type          Address     Off      Size     ES Flg Lk Inf Al
+# CHECK2:      .text          PROGBITS [[#%x,TEXT:]]    [[#%x,]] [[#%x,]] 00 AX   0   0  4
+# CHECK2:      foo0           PROGBITS [[#%x,FOO0:]]    [[#%x,]] [[#%x,]] 00 A    0   0  8
+# CHECK2-NEXT: .relafoo0      RELA     [[#%x,]]         [[#%x,]] [[#%x,]] 18  I  11   3  8
+# CHECK2-NEXT: foo1           PROGBITS [[#%x,FOO1:]]    [[#%x,]] [[#%x,]] 00 A    0   0  8
+# CHECK2-NEXT: .relafoo1      RELA     [[#%x,]]         [[#%x,]] [[#%x,]] 18  I  11   5  8
+# CHECK2:      nonalloc0      PROGBITS 0000000000000000 [[#%x,]] [[#%x,]] 00   C  0   0  8
+# CHECK2-NEXT: .relanonalloc0 RELA     [[#%x,]]         [[#%x,]] [[#%x,]] 18  IC 11   7  8
+# CHECK2-NEXT: nonalloc1      PROGBITS 0000000000000000 [[#%x,]] [[#%x,]] 00      0   0  8
+# CHECK2-NEXT: .debug_str     PROGBITS 0000000000000000 [[#%x,]] [[#%x,]] 01 MSC  0   0  8
+
+## llvm-readelf -r doesn't support SHF_COMPRESSED SHT_RELA.
+# CHECK2: warning: {{.*}}: unable to read relocations from SHT_RELA section with index 8: section [index 8] has an invalid sh_size ([[#]]) which is not a multiple of its sh_entsize (24)
+
+# CHECK2:      Hex dump of section 'nonalloc0':
+## zlib with ch_size=0x10
+# CHECK2-NEXT: 01000000 00000000 10000000 00000000
+# CHECK2-NEXT: 08000000 00000000 {{.*}}
+# CHECK2:      Hex dump of section '.debug_str':
+## zstd with ch_size=0x38
+# CHECK2-NEXT: 02000000 00000000 38000000 00000000
+# CHECK2-NEXT: 01000000 00000000 {{.*}}
+
+# CHECK2DE:       Hex dump of section 'nonalloc0':
+# CHECK2DE-NEXT:  0x00000000 00000000 00000000 00000000 00000000 ................
+# CHECK2DE-EMPTY:
+# CHECK2DE-NEXT:  Hex dump of section '.debug_str':
+# CHECK2DE-NEXT:  0x00000000 41414141 41414141 41414141 41414141 AAAAAAAAAAAAAAAA
+
+## --decompress-debug-sections takes precedence, even if it is before --compress-sections.
+# RUN: llvm-objcopy a.o out3 --decompress-debug-sections --compress-sections .debug_str=zstd
+# RUN: llvm-readelf -S out3 | FileCheck %s --check-prefix=CHECK3
+
+# CHECK3:      .debug_str PROGBITS 0000000000000000 [[#%x,]] [[#%x,]] 01 MS   0   0  1
+
+# RUN: llvm-objcopy a.o out4 --compress-sections '*0=zlib'
+# RUN: llvm-readelf -S out4 | FileCheck %s --check-prefix=CHECK4
+
+# CHECK4:      Name           Type          Address     Off      Size     ES Flg Lk Inf Al
+# CHECK4:      .text          PROGBITS [[#%x,TEXT:]]    [[#%x,]] [[#%x,]] 00  AX  0   0  4
+# CHECK4:      foo0           PROGBITS [[#%x,FOO0:]]    [[#%x,]] [[#%x,]] 00  AC  0   0  8
+# CHECK4-NEXT: .relafoo0      RELA     [[#%x,]]         [[#%x,]] [[#%x,]] 18  IC 11   3  8
+# CHECK4-NEXT: foo1           PROGBITS [[#%x,FOO1:]]    [[#%x,]] [[#%x,]] 00   A  0   0  8
+# CHECK4-NEXT: .relafoo1      RELA     [[#%x,]]         [[#%x,]] [[#%x,]] 18   I 11   5  8
+# CHECK4:      nonalloc0      PROGBITS 0000000000000000 [[#%x,]] [[#%x,]] 00   C  0   0  8
+# CHECK4-NEXT: .relanonalloc0 RELA     [[#%x,]]         [[#%x,]] [[#%x,]] 18  IC 11   7  8
+# CHECK4-NEXT: nonalloc1      PROGBITS 0000000000000000 [[#%x,]] [[#%x,]] 00      0   0  8
+# CHECK4-NEXT: .debug_str     PROGBITS 0000000000000000 [[#%x,]] [[#%x,]] 01  MS  0   0  1
+
+## If a section is already compressed, compression request for another format is ignored.
+# RUN: llvm-objcopy a.o out5 --compress-sections 'nonalloc0=zlib'
+# RUN: llvm-readelf -x nonalloc0 out5 | FileCheck %s --check-prefix=CHECK5
+# RUN: llvm-objcopy out5 out5a --compress-sections 'nonalloc0=zstd'
+# RUN: cmp out5 out5a
+
+# CHECK5:      Hex dump of section 'nonalloc0':
+## zlib with ch_size=0x10
+# CHECK5-NEXT: 01000000 00000000 10000000 00000000
+# CHECK5-NEXT: 08000000 00000000 {{.*}}
+
+# RUN: not llvm-objcopy --compress-sections=foo a.o out 2>&1 | \
+# RUN:   FileCheck %s --check-prefix=ERR1 --implicit-check-not=error:
+# ERR1:      error: --compress-sections: parse error, not 'section-glob=[none|zlib|zstd]'
+
+# RUN: llvm-objcopy --compress-sections 'a[=zlib' a.o out 2>&1 | \
+# RUN:   FileCheck %s --check-prefix=ERR2 --implicit-check-not=error:
+# ERR2:      warning: invalid glob pattern, unmatched '['
+
+# RUN: not llvm-objcopy a.o out --compress-sections='.debug*=zlib-gabi' --compress-sections='.debug*=' 2>&1 | \
+# RUN:   FileCheck -check-prefix=ERR3 %s
+# ERR3:      error: invalid or unsupported --compress-sections format: .debug*=zlib-gabi
+
+# RUN: not llvm-objcopy a.o out --compress-sections='!.debug*=zlib' 2>&1 | \
+# RUN:   FileCheck -check-prefix=ERR4 %s
+# ERR4:      error: --compress-sections: negative pattern is unsupported
+
+.globl _start
+_start:
+  ret
+
+.section foo0,"a"
+.balign 8
+.quad .text-.
+.quad .text-.
+.section foo1,"a"
+.balign 8
+.quad .text-.
+.quad .text-.
+.section nonalloc0,""
+.balign 8
+.quad .text+1
+.quad .text+2
+sym0:
+.section nonalloc1,""
+.balign 8
+.quad 42
+sym1:
+
+.section .debug_str,"MS",@progbits,1
+.Linfo_string0:
+  .asciz "AAAAAAAAAAAAAAAAAAAAAAAAAAA"
+.Linfo_string1:
+  .asciz "BBBBBBBBBBBBBBBBBBBBBBBBBBB"

llvm/test/tools/llvm-objcopy/ELF/decompress-sections.test

@@ -4,13 +4,42 @@
 # RUN: yaml2obj %s -o %t
 # RUN: llvm-objcopy --decompress-debug-sections %t %t.de
 # RUN: llvm-readelf -S %t.de | FileCheck %s
+# RUN: llvm-objcopy --compress-sections '*nonalloc=none' --compress-sections .debugx=none %t %t.1.de
+# RUN: cmp %t.de %t.1.de
 
 # CHECK:        Name              Type            Address          Off      Size     ES Flg Lk Inf Al
 # CHECK:        .debug_alloc      PROGBITS        0000000000000000 [[#%x,]] [[#%x,]] 00  AC  0   0  0
 # CHECK-NEXT:   .debug_nonalloc   PROGBITS        0000000000000000 [[#%x,]] [[#%x,]] 00      0   0  1
 # CHECK-NEXT:   .debugx           PROGBITS        0000000000000000 [[#%x,]] [[#%x,]] 00      0   0  1
 # CHECK-NEXT:   nodebug           PROGBITS        0000000000000000 [[#%x,]] [[#%x,]] 00   C  0   0  0
 
+# RUN: llvm-objcopy --compress-sections '.debug*=none' %t %t2.de
+# RUN: llvm-readelf -S -x .debug_alloc -x .debug_nonalloc -x .debugx %t2.de | FileCheck %s --check-prefix=CHECK2
+
+# CHECK2:        Name              Type            Address          Off      Size     ES Flg Lk Inf Al
+# CHECK2:        .debug_alloc      PROGBITS        0000000000000000 [[#%x,]] [[#%x,]] 00  A   0   0  1
+# CHECK2-NEXT:   .debug_nonalloc   PROGBITS        0000000000000000 [[#%x,]] [[#%x,]] 00      0   0  1
+# CHECK2-NEXT:   .debugx           PROGBITS        0000000000000000 [[#%x,]] [[#%x,]] 00      0   0  1
+# CHECK2-NEXT:   nodebug           PROGBITS        0000000000000000 [[#%x,]] [[#%x,]] 00   C  0   0  0
+
+# CHECK2:       Hex dump of section '.debug_alloc':
+# CHECK2-NEXT:  0x00000000 2a000000 00000000 2a000000 00000000 *.......*.......
+# CHECK2-NEXT:  0x00000010 2a000000 00000000 2a000000 00000000 *.......*.......
+# CHECK2-NEXT:  0x00000020 2a000000 00000000 2a000000 00000000 *.......*.......
+# CHECK2-NEXT:  0x00000030 2a000000 00000000 2a000000 00000000 *.......*.......
+# CHECK2-EMPTY:
+# CHECK2:       Hex dump of section '.debug_nonalloc':
+# CHECK2-NEXT:  0x00000000 2a000000 00000000 2a000000 00000000 *.......*.......
+# CHECK2-NEXT:  0x00000010 2a000000 00000000 2a000000 00000000 *.......*.......
+# CHECK2-NEXT:  0x00000020 2a000000 00000000 2a000000 00000000 *.......*.......
+# CHECK2-NEXT:  0x00000030 2a000000 00000000 2a000000 00000000 *.......*.......
+# CHECK2-EMPTY:
+# CHECK2-NEXT:  Hex dump of section '.debugx':
+# CHECK2-NEXT:  0x00000000 2a000000 00000000 2a000000 00000000 *.......*.......
+# CHECK2-NEXT:  0x00000010 2a000000 00000000 2a000000 00000000 *.......*.......
+# CHECK2-NEXT:  0x00000020 2a000000 00000000 2a000000 00000000 *.......*.......
+# CHECK2-NEXT:  0x00000030 2a000000 00000000 2a000000 00000000 *.......*.......
+
 --- !ELF
 FileHeader:
   Class:   ELFCLASS64

llvm/tools/llvm-objcopy/ObjcopyOptions.cpp

@@ -736,6 +736,42 @@ objcopy::parseObjcopyOptions(ArrayRef<const char *> RawArgsArr,
       return createStringError(errc::invalid_argument, Reason);
   }
 
+  for (const auto *A : InputArgs.filtered(OBJCOPY_compress_sections)) {
+    SmallVector<StringRef, 0> Fields;
+    StringRef(A->getValue()).split(Fields, '=');
+    if (Fields.size() != 2 || Fields[1].empty()) {
+      return createStringError(
+          errc::invalid_argument,
+          A->getSpelling() +
+              ": parse error, not 'section-glob=[none|zlib|zstd]'");
+    }
+
+    auto Type = StringSwitch<DebugCompressionType>(Fields[1])
+                    .Case("zlib", DebugCompressionType::Zlib)
+                    .Case("zstd", DebugCompressionType::Zstd)
+                    .Default(DebugCompressionType::None);
+    if (Type == DebugCompressionType::None && Fields[1] != "none") {
+      return createStringError(
+          errc::invalid_argument,
+          "invalid or unsupported --compress-sections format: %s",
+          A->getValue());
+    }
+
+    auto &P = Config.compressSections.emplace_back();
+    P.second = Type;
+    auto Matcher =
+        NameOrPattern::create(Fields[0], SectionMatchStyle, ErrorCallback);
+    // =none allows overriding a previous =zlib or =zstd. Reject negative
+    // patterns, which would be confusing.
+    if (Matcher && !Matcher->isPositiveMatch()) {
+      return createStringError(
+          errc::invalid_argument,
+          "--compress-sections: negative pattern is unsupported");
+    }
+    if (Error E = P.first.addMatcher(std::move(Matcher)))
+      return std::move(E);
+  }
+
   Config.AddGnuDebugLink = InputArgs.getLastArgValue(OBJCOPY_add_gnu_debuglink);
   // The gnu_debuglink's target is expected to not change or else its CRC would
   // become invalidated and get rejected. We can avoid recalculating the

llvm/tools/llvm-objcopy/ObjcopyOpts.td

@@ -35,6 +35,12 @@ def : Flag<["--"], "compress-debug-sections">, Alias<compress_debug_sections>,
       AliasArgs<["zlib"]>;
 def decompress_debug_sections : Flag<["--"], "decompress-debug-sections">,
                                 HelpText<"Decompress DWARF debug sections">;
+defm compress_sections
+    : Eq<"compress-sections",
+         "Compress or decompress sections using specified format. Supported "
+         "formats: zlib, zstd. Specify 'none' for decompression">,
+      MetaVarName<"<section-glob>=<format>">;
+
 defm split_dwo
     : Eq<"split-dwo", "Equivalent to --extract-dwo and <dwo-file> as the output file and no other options, "
                       "and then --strip-dwo on the input file">,