[llvm-objdump] Complete -chained_fixups support

[BertalanD]

Summary

This commit adds definitions for the dyld_chained_import* structs.
The imports array is now printed with llvm-otool -chained_fixups. This
completes this option's implementation.

A slight difference from cctools otool is that we don't yet dump the
raw bytes of the imports entries.

When Apple's effort to upstream their chained fixups code continues,
we'll replace this code with the then-upstreamed code. But we need
something in the meantime for testing ld64.lld's chained fixups code.

[thakis]

Very nice!

llvm/include/llvm/BinaryFormat/MachO.h

2131

It'd be nice to use bit_cast from llvm/ADT/bit.h here, which will ensure that Raw is the same size as dyld_chained_import. Here this is fairly easy I think:

auto Raw = bit_cast<uint32_t>(C);
sys::swapByteOrder(Raw);
C = bit_cast<dyld_chained_import>(Raw);

Below it's a bit more tricky, but I think this might work (all untested though):

auto Raw = bit_cast<std::array<uint32_t, 2>>(C);
sys::swapByteOrder(Raw[0]);
sys::swapByteOrder(Raw[1]);
C = bit_cast<(dyld_chained_import_addend>(Raw);

(and analogously for the 64-bit version)

llvm/lib/Object/MachOObjectFile.cpp

4935

nit: no else after return

This function looks weird! It almost does if (V == a) return a a bunch of times followed by return V, and the special cases look pretty pointless at first. (I understand they're needed to sign-extend the 3 special values to int, while the input is 8 or 16 bits.) It feels like there should be a nicer way to write this.

Maybe

if (Value == static_cast<T>(MachO::BIND_SPECIAL_DYLIB_MAIN_EXECUTABLE) ||
    Value == static_cast<T>(MachO::BIND_SPECIAL_DYLIB_FLAT_LOOKUP) ||
     third case)
  return SignExtend32<sizeof(T) * 8>(Value);
return Value;

Not sure if that's actually better, but it makes the difference a bit more explicit maybe.

Up to you :)

4969

Isn't sizeof(dyld_chained_import) clearer here? (likewise in the other two cases)

5005

This has GetEncodedOrdinal with upper-case G while the function above has it with lower-case g. Can this build? Maybe you uploaded a half-committed patch?

5028

nit: should this return a malformedError instead? it's data-dependent, so it's not truly unreachable. (but up to you)

llvm/tools/llvm-objdump/MachODump.cpp

1295

It makes sense to me to wait with this until the upstreaming has happened to see how to best implement this then.

[thakis]

(All comments optional and up to you, in case that wasn't clear.)

[BertalanD]

llvm/include/llvm/BinaryFormat/MachO.h

2131

I'm not convinced that swapping the bytes is enough for endianness conversion. https://www.naic.edu/~phil/notes/bitfieldStorage.html suggests that the bits within the bytes would also need to be reversed. I'm going to update this diff to use bit masks and shifts like D132036.

llvm/lib/Object/MachOObjectFile.cpp

4935

I didn't know about SignExtend32, thank you for mentioning it. Your version does indeed look nicer than what I wrote.

5028

I'll add an error message to where we're computing ImportSize, and then this branch will be truly unreachable.

[thakis]

llvm/include/llvm/BinaryFormat/MachO.h

2131

Huh, TIL, I suppose.

What I used to believe up to now: There are two separate things: in-memory byte order, and order in which bitfields get assigned to their underlying words.

The c99 standard (https://www.open-std.org/jtc1/sc22/wg14/www/docs/n1124.pdf), 6.7.2.1.10: says about the latter "The order of allocation of bit-fields within a unit (high-order to low-order or low-order to high-order) is implementation-defined."

So I thought that the order bitfields are assigned is implementation-defined, but independent of byte order.

But sysv abi (https://www.uclibc.org/docs/psABI-x86_64.pdf) paragraph 3.1 says: "bit-fields are allocated from right to left". So I thought, in practice, this settles things: Bit fields are assigned right-to-left (in a register, say), and then they assume system endianness when written to memory, conceptually.

(In theory, gcc has all kinds of options here: https://gcc.gnu.org/onlinedocs/gccint/Storage-Layout.html BITS_BIG_ENDIAN, PCC_BITFIELD_TYPE_MATTERS, TARGET_MS_BITFIELD_LAYOUT_P, …)

But https://godbolt.org/z/6sqo7nEd4 shows that this clearly isn't true. (Also, what's with the 65535 in the big-endian ppc64 output there? That looks completely wrong?)

And CGRecordLayoutBuilder.cpp has:

void CGRecordLowering::setBitFieldInfo(
...
  // Reverse the bit offsets for big endian machines. Because we represent
  // a bitfield as a single large integer load, we can imagine the bits
  // counting from the most-significant-bit instead of the
  // least-significant-bit.
  if (DataLayout.isBigEndian())
    Info.Offset = Info.StorageSize - (Info.Offset + Info.Size);

And gcc seems to match clang's behavior (see godbolt link), so that seems very intentional.

Does anyone reading this know what specifies this?

…aha, the ppc sysv abi (http://refspecs.linux-foundation.org/elf/elfspec_ppc.pdf) does say: " Bit-fields are allocated from right to left (least to most significant) on Little-Endian implementations and from left to right (most to least significant) on Big-Endian implementations."

I wish I hadn't learned that!

llvm/lib/Object/MachOObjectFile.cpp

4935

(FWIW, I didn't know about SignExtend32 either while I wrote the "looks weird" paragraph. I thought about how this could be written differently, then ran rg -i signextend llvm to see how other code handles this and then found SignExtend32 in there and figured it matches well enough.)

5004

The in-register view (i.e. not in-memory, we can ignore the byte-swapping of big-endian vs little-endian here) of

struct dyld_chained_import {
  uint32_t lib_ordinal : 8;
  uint32_t weak_import : 1;
  uint32_t name_offset : 23;
};

is:

• In little-endian (bits assigned right-to-left):
  • most significant 23 bits are name_offset
  • then 1 bit weak_import
  • then 8 bit lib_ordinal in the lowest 8 bits

• In big-endian (bits assigned left-to-right):
  • 8 bit lib_ordinal in the highest 8 bits
  • then 1 bit weak_import
  • least significant 23 bits are name_offset

So I think even with the manual masking, this currently gets things wrong.

Also, it's weird that we don't use the dyld_chained_import structs at all. Strange for greppability, using something like http://llvm-cs.pcc.me.uk/ for cross-references, etc.

What do you think about adding back the Swap<> functions and manually swapping the bitfields around too after swapping the bytes? That assumes sysv abi, but that'll work almost everywhere (can't think of a place where it wouldn't? Maybe windows big endian? Didn't try that, but I wouldn't be surprised if clang-cl didn't work great there either), and if someone needs this to work on some obscure system, they can worry about it then?

[thakis]

llvm/lib/Object/MachOObjectFile.cpp

5004

Alternatively, something like https://github.com/freebsd/freebsd-src/blob/master/sys/netinet/ip.h#L51-L59 might also be an option.

[thakis]

We chatted a bit offline. Rough summary

• https://github.com/freebsd/freebsd-src/blob/master/sys/netinet/ip.h#L51-L59 works if the on-disk format is the same for BE and LE, but since $(xcrun -show-sdk-path)/usr/include/mach-o/fixup-chains.h doesn't have any ifdefs, that's likely not the case here
• …but all chained fixup apple platforms are little endian, so it's really unknowable
• the current code does work on all hosts for when the on-disk data is little-endian
• it'd be nice to do the swapping anyways, since that puts all the endianness handling in a single place
• Bitfield swapping is different for little-endian-file-on-big-endian-host and the other way round, and the swapStruct() wrappers in BinaryFormat/MachO.h don't know which direction they're swapping in, so there's have to be a dedicated wrapper for each struct. Example for dyld_chained_import:

// Little-endian file running on big-endian host:
Raw = ((Raw & 0xff) << 24) | ((Raw & 0x100) << 15) || (Raw >> 9);

// Big-endian file running on little-endian host:
Raw = (Raw >> 24) | ((Raw & 0x800000) >> 15) || ((Raw & 0x7fffff << 9);

• Since there's a bunch of those structs, it'd be nice to make a general helper for this. This seems to roughly work:

#include <stdio.h>

template<typename T, unsigned W>
T bitfield_swap_le(T t) {
  static_assert(W == 0, "");
  return 0;
}

template<typename T, unsigned W, unsigned First, int... Sizes>
T bitfield_swap_le(T t) {
  // Peel off rightmost field, add it on left, recurse.
  return ((t & ((1u << First) - 1)) << (W - First)) |
         bitfield_swap_le<T, W - First, Sizes...>(t >> First);
}

template<typename T, unsigned W>
T bitfield_swap_be(T t) {
  static_assert(W == 0, "");
  return 0;
}

template<typename T, unsigned W, unsigned First, int... Sizes>
T bitfield_swap_be(T t) {
  // Peel off leftmost field, add it on right, recurse.
  return ((t >> (W - First)) |
         (bitfield_swap_be<T, W - First, Sizes...>(t & ((1u << (W - First)) - 1))) << First);
}

int main() {
  {
    unsigned i = bitfield_swap_le<unsigned, 32, 8, 1, 23>(0xff);
    unsigned j = bitfield_swap_le<unsigned, 32, 8, 1, 23>(0x100);
    unsigned k = bitfield_swap_le<unsigned, 32, 8, 1, 23>(0xffff'fe00);
    printf("%08x %08x %08x\n", i, j, k);
  }
  {
    unsigned i = bitfield_swap_be<unsigned, 32, 8, 1, 23>(0xff00'0000);
    unsigned j = bitfield_swap_be<unsigned, 32, 8, 1, 23>(0x0080'0000);
    unsigned k = bitfield_swap_be<unsigned, 32, 8, 1, 23>(0x007f'ffff);
    printf("%08x %08x %08x\n", i, j, k);
  }
}

% clang bitfield_swap.cc -std=c++17
% ./a.out                          
ff000000 00800000 007fffff
000000ff 00000100 fffffe00

• Then there could be a bitfield_swap<> member template that calls either of those based on file and host bitness, and that'd make it convenient to write a swapStruct() wrapper for each type with a bitfield

• However, all likely host systems are little-endian too, so it also seems fine to just go back to the original code, add a comment, mark the test REQUIRES: host-byteorder-little-endian and do the rest later (or have someone who actually needs it do it)

[BertalanD]

Added some extra checks. Addends are now printed as signed integers, as in Apple otool.

[thakis]

Please land :)

llvm/lib/Object/MachOObjectFile.cpp

4996

(As discussed elsewhere, I think this is the worst of the approaches discussed above, but it's also not terribly important, so fine.)

llvm/include/llvm/BinaryFormat/MachO.h

struct dyld_chained_starts_in_segment {
   uint16_t pointer_format;    ///< DYLD_CHAINED_PTR*
   uint64_t segment_offset;    ///< VM offset from the __TEXT segment
   uint32_t max_valid_pointer; ///< Values beyond this are not pointers on 32-bit
   uint16_t page_count;        ///< Length of the page_start array
   uint16_t page_start[1];     ///< Page offset of first fixup on each page, or
                               ///< DYLD_CHAINED_PTR_START_NONE if no fixups
 };
 
+// DYLD_CHAINED_IMPORT
+struct dyld_chained_import {
+  uint32_t lib_ordinal : 8;
+  uint32_t weak_import : 1;
+  uint32_t name_offset : 23;
+};
+
+// DYLD_CHAINED_IMPORT_ADDEND
+struct dyld_chained_import_addend {
+  uint32_t lib_ordinal : 8;
+  uint32_t weak_import : 1;
+  uint32_t name_offset : 23;
+  int32_t addend;
+};
+
+// DYLD_CHAINED_IMPORT_ADDEND64
+struct dyld_chained_import_addend64 {
+  uint64_t lib_ordinal : 16;
+  uint64_t weak_import : 1;
+  uint64_t reserved : 15;
+  uint64_t name_offset : 32;
+  uint64_t addend;
+};
+
 // Byte order swapping functions for MachO structs
 
 inline void swapStruct(fat_header &mh) {
   sys::swapByteOrder(mh.magic);
   sys::swapByteOrder(mh.nfat_arch);
 }
 
 inline void swapStruct(fat_arch &mh) {
inline void swapStruct(dyld_chained_starts_in_segment &C) {
   sys::swapByteOrder(C.pointer_format);
   sys::swapByteOrder(C.segment_offset);
   sys::swapByteOrder(C.max_valid_pointer);
   sys::swapByteOrder(C.page_count);
   // seg_info_offset entries must be byte swapped manually.
 }
 
 /* code signing attributes of a process */
 
 enum CodeSignAttrs {
   CS_VALID = 0x00000001,          /* dynamically valid */
   CS_ADHOC = 0x00000002,          /* ad hoc signed */
   CS_GET_TASK_ALLOW = 0x00000004, /* has get-task-allow entitlement */
   CS_INSTALLER = 0x00000008,      /* has installer entitlement */
 
   CS_FORCED_LV =
       0x00000010, /* Library Validation required by Hardened System Policy */

llvm/include/llvm/Object/MachO.h

 ///   WeakImport == true
 ///     The associated bind may be set to 0 if this symbol is missing from its
 ///     parent library. This is called a "weak import."
 ///   LibOrdinal == BIND_SPECIAL_DYLIB_WEAK_LOOKUP
 ///     This symbol may be coalesced with other libraries vending the same
 ///     symbol. E.g., C++'s "operator new". This is called a "weak bind."
 struct ChainedFixupTarget {
 public:
-  ChainedFixupTarget(int LibOrdinal, StringRef Symbol, uint64_t Addend,
-                     bool WeakImport)
-      : LibOrdinal(LibOrdinal), SymbolName(Symbol), Addend(Addend),
-        WeakImport(WeakImport) {}
+  ChainedFixupTarget(int LibOrdinal, uint32_t NameOffset, StringRef Symbol,
+                     uint64_t Addend, bool WeakImport)
+      : LibOrdinal(LibOrdinal), NameOffset(NameOffset), SymbolName(Symbol),
+        Addend(Addend), WeakImport(WeakImport) {}
 
   int libOrdinal() { return LibOrdinal; }
+  uint32_t nameOffset() { return NameOffset; }
   StringRef symbolName() { return SymbolName; }
   uint64_t addend() { return Addend; }
   bool weakImport() { return WeakImport; }
   bool weakBind() {
     return LibOrdinal == MachO::BIND_SPECIAL_DYLIB_WEAK_LOOKUP;
   }
 
 private:
   int LibOrdinal;
+  uint32_t NameOffset;
   StringRef SymbolName;
   uint64_t Addend;
   bool WeakImport;
 };
 
 /// MachOAbstractFixupEntry is an abstract class representing a fixup in a
 /// MH_DYLDLINK file. Fixups generally represent rebases and binds. Binds also
 /// subdivide into additional subtypes (weak, lazy, reexport).

llvm/lib/Object/MachOObjectFile.cpp

 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/None.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/StringRef.h"
 #include "llvm/ADT/StringSwitch.h"
 #include "llvm/ADT/Triple.h"
 #include "llvm/ADT/Twine.h"
+#include "llvm/ADT/bit.h"
 #include "llvm/BinaryFormat/MachO.h"
 #include "llvm/BinaryFormat/Swift.h"
 #include "llvm/Object/Error.h"
 #include "llvm/Object/MachO.h"
 #include "llvm/Object/ObjectFile.h"
 #include "llvm/Object/SymbolicFile.h"
 #include "llvm/Support/DataExtractor.h"
 #include "llvm/Support/Debug.h"
for (size_t I = 0, N = ImageStarts.seg_count; I < N; ++I) {
     }
 
     Segments.emplace_back(I, *OffOrErr, Seg, std::move(PageStarts));
   }
 
   return std::make_pair(ImageStarts.seg_count, Segments);
 }
 
+// The special library ordinals have a negative value, but they are encoded in
+// an unsigned bitfield, so we need to sign extend the value.
+template <typename T> static int getEncodedOrdinal(T Value) {
+  if (Value == static_cast<T>(MachO::BIND_SPECIAL_DYLIB_MAIN_EXECUTABLE) ||
+      Value == static_cast<T>(MachO::BIND_SPECIAL_DYLIB_FLAT_LOOKUP) ||
+      Value == static_cast<T>(MachO::BIND_SPECIAL_DYLIB_WEAK_LOOKUP))
+    return SignExtend32<sizeof(T) * CHAR_BIT>(Value);
+  return Value;
+}
+
+template <typename T, unsigned N>
+std::array<T, N> getArray(const MachOObjectFile &O, const void *Ptr) {
+  std::array<T, N> RawValue;
+  memcpy(RawValue.data(), Ptr, N * sizeof(T));
+  if (O.isLittleEndian() != sys::IsLittleEndianHost)
+    for (auto &Element : RawValue)
+      sys::swapByteOrder(Element);
+  return RawValue;
+}
+
 Expected<std::vector<ChainedFixupTarget>>
 MachOObjectFile::getDyldChainedFixupTargets() const {
+  auto CFOrErr = getChainedFixupsLoadCommand();
+  if (!CFOrErr)
+    return CFOrErr.takeError();
+
+  std::vector<ChainedFixupTarget> Targets;
+  if (!CFOrErr->has_value())
+    return Targets;
+
+  const MachO::linkedit_data_command &DyldChainedFixups = **CFOrErr;
+
   auto CFHeaderOrErr = getChainedFixupsHeader();
   if (!CFHeaderOrErr)
     return CFHeaderOrErr.takeError();
-  std::vector<ChainedFixupTarget> Targets;
   if (!(*CFHeaderOrErr))
     return Targets;
+  const MachO::dyld_chained_fixups_header &Header = **CFHeaderOrErr;
+
+  size_t ImportSize = 0;
+  if (Header.imports_format == MachO::DYLD_CHAINED_IMPORT)
+    ImportSize = sizeof(MachO::dyld_chained_import);
+  else if (Header.imports_format == MachO::DYLD_CHAINED_IMPORT_ADDEND)
+    ImportSize = sizeof(MachO::dyld_chained_import_addend);
+  else if (Header.imports_format == MachO::DYLD_CHAINED_IMPORT_ADDEND64)
+    ImportSize = sizeof(MachO::dyld_chained_import_addend64);
+  else
+    return malformedError("bad chained fixups: unknown imports format: " +
+                          Twine(Header.imports_format));
+
+  const char *Contents = getPtr(*this, DyldChainedFixups.dataoff);
+  const char *Imports = Contents + Header.imports_offset;
+  size_t ImportsEndOffset =
+      Header.imports_offset + ImportSize * Header.imports_count;
+  const char *ImportsEnd = Contents + ImportsEndOffset;
+  const char *Symbols = Contents + Header.symbols_offset;
+  const char *SymbolsEnd = Contents + DyldChainedFixups.datasize;
+
+  if (ImportsEnd > Symbols)
+    return malformedError("bad chained fixups: imports end " +
+                          Twine(ImportsEndOffset) + " extends past end " +
+                          Twine(DyldChainedFixups.datasize));
+
+  if (ImportsEnd > Symbols)
+    return malformedError("bad chained fixups: imports end " +
+                          Twine(ImportsEndOffset) + " overlaps with symbols");
+
+  for (const char *ImportPtr = Imports; ImportPtr < ImportsEnd;
+       ImportPtr += ImportSize) {
+    int LibOrdinal;
+    bool WeakImport;
+    uint32_t NameOffset;
+    uint64_t Addend;
+    if (Header.imports_format == MachO::DYLD_CHAINED_IMPORT) {
+      auto RawValue = getArray<uint32_t, 1>(*this, ImportPtr);
+
+      LibOrdinal = getEncodedOrdinal<uint8_t>(RawValue[0] & 0xFF);
+      WeakImport = (RawValue[0] >> 8) & 1;
+      NameOffset = RawValue[0] >> 9;
+      Addend = 0;
+    } else if (Header.imports_format == MachO::DYLD_CHAINED_IMPORT_ADDEND) {
+      auto RawValue = getArray<uint32_t, 2>(*this, ImportPtr);
+
+      LibOrdinal = getEncodedOrdinal<uint8_t>(RawValue[0] & 0xFF);
+      WeakImport = (RawValue[0] >> 8) & 1;
+      NameOffset = RawValue[0] >> 9;
+      Addend = bit_cast<int32_t>(RawValue[1]);
+    } else if (Header.imports_format == MachO::DYLD_CHAINED_IMPORT_ADDEND64) {
+      auto RawValue = getArray<uint64_t, 2>(*this, ImportPtr);
+
+      LibOrdinal = getEncodedOrdinal<uint16_t>(RawValue[0] & 0xFFFF);
+      NameOffset = (RawValue[0] >> 16) & 1;
+      WeakImport = RawValue[0] >> 17;
+      Addend = RawValue[1];
+    } else {
+      llvm_unreachable("Import format should have been checked");
+    }
+
+    const char *Str = Symbols + NameOffset;
+    if (Str >= SymbolsEnd)
+      return malformedError("bad chained fixups: symbol offset " +
+                            Twine(NameOffset) + " extends past end " +
+                            Twine(DyldChainedFixups.datasize));
+    Targets.emplace_back(LibOrdinal, NameOffset, Str, Addend, WeakImport);
+  }
+
   return Targets;
 }
 
 ArrayRef<uint8_t> MachOObjectFile::getDyldInfoExportsTrie() const {
   if (!DyldInfoLoadCmd)
     return None;
 
   auto DyldInfoOrErr =

llvm/test/tools/llvm-objdump/MachO/chained-fixups.test

 DETAILS-NEXT:   pointer_format = 6 (DYLD_CHAINED_PTR_64_OFFSET)
 DETAILS-NEXT:   segment_offset = 0x3f0
 DETAILS-NEXT:   max_valid_pointer = 0
 DETAILS-NEXT:   page_count = 4
 DETAILS-NEXT:     page_start[0] = 0
 DETAILS-NEXT:     page_start[1] = 32
 DETAILS-NEXT:     page_start[2] = 65535 (DYLD_CHAINED_PTR_START_NONE)
 DETAILS-NEXT:     page_start[3] = 32
+DETAILS-NEXT: dyld chained import[0]
+DETAILS-NEXT:   lib_ordinal = -2 (flat-namespace)
+DETAILS-NEXT:   weak_import = 0
+DETAILS-NEXT:   name_offset = 1 (_dynamicLookup)
+DETAILS-NEXT: dyld chained import[1]
+DETAILS-NEXT:   lib_ordinal = 1 (libdylib)
+DETAILS-NEXT:   weak_import = 1
+DETAILS-NEXT:   name_offset = 16 (_weakImport)
+DETAILS-NEXT: dyld chained import[2]
+DETAILS-NEXT:   lib_ordinal = 1 (libdylib)
+DETAILS-NEXT:   weak_import = 0
+DETAILS-NEXT:   name_offset = 28 (_dylib)
+DETAILS-NEXT: dyld chained import[3]
+DETAILS-NEXT:   lib_ordinal = -3 (weak)
+DETAILS-NEXT:   weak_import = 0
+DETAILS-NEXT:   name_offset = 35 (_weakLocal)
+DETAILS-NEXT: dyld chained import[4]
+DETAILS-NEXT:   lib_ordinal = -3 (weak)
+DETAILS-NEXT:   weak_import = 0
+DETAILS-NEXT:   name_offset = 46 (_weak)
 
 ## This test checks that the output is identical to that of cctools-1001.2 (XCode 14)
+## FIXME: Print encoded values of the dyld_chained_import* entries
+##
 ## The input was generated from the following files:
 ##
 ## --- dylib.s:
 ##   .data
 ##   .globl _weak, _dylib, _weakImport
 ##   .weak_definition _weak
 ##   _weak:
 ##   _dylib:

llvm/tools/llvm-objdump/MachODump.cpp

 bool objdump::ObjcMetaData;
 std::string objdump::DisSymName;
 bool objdump::SymbolicOperands;
 static std::vector<std::string> ArchFlags;
 
 static bool ArchAll = false;
 static std::string ThumbTripleName;
 
+static StringRef ordinalName(const object::MachOObjectFile *, int);
+
 void objdump::parseMachOOptions(const llvm::opt::InputArgList &InputArgs) {
   FirstPrivateHeader = InputArgs.hasArg(OBJDUMP_private_header);
   ExportsTrie = InputArgs.hasArg(OBJDUMP_exports_trie);
   Rebase = InputArgs.hasArg(OBJDUMP_rebase);
   Rpaths = InputArgs.hasArg(OBJDUMP_rpaths);
   Bind = InputArgs.hasArg(OBJDUMP_bind);
   LazyBind = InputArgs.hasArg(OBJDUMP_lazy_bind);
   WeakBind = InputArgs.hasArg(OBJDUMP_weak_bind);
for (auto [Index, PageStart] : enumerate(Segment.PageStarts)) {
     outs() << "    page_start[" << Index << "] = " << PageStart;
     // FIXME: Support DYLD_CHAINED_PTR_START_MULTI (32-bit only)
     if (PageStart == MachO::DYLD_CHAINED_PTR_START_NONE)
       outs() << " (DYLD_CHAINED_PTR_START_NONE)";
     outs() << '\n';
   }
 }
 
+static void PrintChainedFixupTarget(ChainedFixupTarget &Target, size_t Idx,
+                                    int Format, MachOObjectFile *O) {
+  if (Format == MachO::DYLD_CHAINED_IMPORT)
+    outs() << "dyld chained import";
+  else if (Format == MachO::DYLD_CHAINED_IMPORT_ADDEND)
+    outs() << "dyld chained import addend";
+  else if (Format == MachO::DYLD_CHAINED_IMPORT_ADDEND64)
+    outs() << "dyld chained import addend64";
+  // FIXME: otool prints the encoded value as well.
+  outs() << '[' << Idx << "]\n";
+
+  outs() << "  lib_ordinal = " << Target.libOrdinal() << " ("
+         << ordinalName(O, Target.libOrdinal()) << ")\n";
+  outs() << "  weak_import = " << Target.weakImport() << '\n';
+  outs() << "  name_offset = " << Target.nameOffset() << " ("
+         << Target.symbolName() << ")\n";
+  if (Format != MachO::DYLD_CHAINED_IMPORT)
+    outs() << "  addend      = " << Target.addend() << '\n';
+}
+
 static void PrintChainedFixups(MachOObjectFile *O) {
   // MachOObjectFile::getChainedFixupsHeader() reads LC_DYLD_CHAINED_FIXUPS.
   // FIXME: Support chained fixups in __TEXT,__chain_starts section too.
   auto ChainedFixupHeader =
       unwrapOrError(O->getChainedFixupsHeader(), O->getFileName());
   if (!ChainedFixupHeader)
     return;
 
for (size_t I = 0; I < SegCount; ++I) {
 
     outs() << "    seg_offset[" << I << "] = " << SegOffset << " ("
            << SegNames[I] << ")\n";
   }
 
   for (const MachOObjectFile::ChainedFixupsSegment &S : Segments)
     PrintChainedFixupsSegment(S, SegNames[S.SegIdx]);
 
-  // FIXME: Print more things.
+  auto FixupTargets =
+      unwrapOrError(O->getDyldChainedFixupTargets(), O->getFileName());
+
+  uint32_t ImportsFormat = ChainedFixupHeader->imports_format;
+  for (auto [Idx, Target] : enumerate(FixupTargets))
+    PrintChainedFixupTarget(Target, Idx, ImportsFormat, O);
 }
 
 static void PrintDyldInfo(MachOObjectFile *O) {
   outs() << "dyld information:" << '\n';
   printMachOChainedFixups(O);
 }
 
 static void PrintDylibs(MachOObjectFile *O, bool JustId) {
static StringRef ordinalName(const object::MachOObjectFile *Obj, int Ordinal) {
   StringRef DylibName;
   switch (Ordinal) {
   case MachO::BIND_SPECIAL_DYLIB_SELF:
     return "this-image";
   case MachO::BIND_SPECIAL_DYLIB_MAIN_EXECUTABLE:
     return "main-executable";
   case MachO::BIND_SPECIAL_DYLIB_FLAT_LOOKUP:
     return "flat-namespace";
+  case MachO::BIND_SPECIAL_DYLIB_WEAK_LOOKUP:
+    return "weak";
   default:
     if (Ordinal > 0) {
       std::error_code EC =
           Obj->getLibraryShortNameByIndex(Ordinal - 1, DylibName);
       if (EC)
         return "<<bad library ordinal>>";
       return DylibName;
     }