[lld-macho]Fix bug in finding "chained" re-exported libs.

[oontvoo]

Details:
When we have the following scenario:

• lib_a re-exports lib_b
• lib_b re-exports @rpath/lib_c
  • lib_b contains LC_RPATH

Previously, lld-macho cannot find lib_c because it was attempting to resolve the '@rpath' from lib_b (which had no LC_RPATH defined). The change here is to also consider all the LC_RPATH rom lib_b when trying to find lib_c.

Inspired by real-life example when linking with libXCTestSwiftSupport.dylib (which re-exports XCTest, which re-exports XCTestCore)

[llvmbot]

@llvm/pr-subscribers-lld

Author: Vy Nguyen (oontvoo)

Changes

Details:
When we have the following scenario:

• lib_a re-exports lib_b
• lib_b re-exports @rpath/lib_c
  • lib_b contains LC_RPATH

Previously, lld-macho cannot find lib_c because it was attempting to resolve the '@rpath' from lib_b (which had no LC_RPATH defined). The change here is to also consider all the LC_RPATH rom lib_b when trying to find lib_c.

Inspired by real-life example when linking with libXCTestSwiftSupport.dylib (which re-exports XCTest, which re-exports XCTestCore)

---

Full diff: https://github.com/llvm/llvm-project/pull/135241.diff

2 Files Affected:

• (modified) lld/MachO/InputFiles.cpp (+21-3)
• (added) lld/test/MachO/reexport-with-rpath.s (+37)

diff --git a/lld/MachO/InputFiles.cpp b/lld/MachO/InputFiles.cpp
index 9adfbc9d3f6f5..9afa5feaa1693 100644
--- a/lld/MachO/InputFiles.cpp
+++ b/lld/MachO/InputFiles.cpp
@@ -1633,6 +1633,17 @@ static DylibFile *findDylib(StringRef path, DylibFile *umbrella,
       if (std::optional<StringRef> dylibPath = resolveDylibPath(newPath.str()))
         return loadDylib(*dylibPath, umbrella);
     }
+    // If not found in umbrella, try the rpaths specified via -rpath too.
+    for (StringRef rpath : config->runtimePaths) {
+      newPath.clear();
+      if (rpath.consume_front("@loader_path/")) {
+        fs::real_path(umbrella->getName(), newPath);
+        path::remove_filename(newPath);
+      }
+      path::append(newPath, rpath, path.drop_front(strlen("@rpath/")));
+      if (std::optional<StringRef> dylibPath = resolveDylibPath(newPath.str()))
+        return loadDylib(*dylibPath, umbrella);
+    }
   }
 
   // FIXME: Should this be further up?
@@ -1678,9 +1689,16 @@ static bool isImplicitlyLinked(StringRef path) {
 void DylibFile::loadReexport(StringRef path, DylibFile *umbrella,
                          const InterfaceFile *currentTopLevelTapi) {
   DylibFile *reexport = findDylib(path, umbrella, currentTopLevelTapi);
-  if (!reexport)
-    error(toString(this) + ": unable to locate re-export with install name " +
-          path);
+  if (!reexport) {
+    // If not found in umbrella, retry since some rpaths might have been
+    // defined in "this" dyblib (which contains the LC_REEXPORT_DYLIB cmd) and
+    // not in the umbrella.
+    DylibFile *reexport2 = findDylib(path, this, currentTopLevelTapi);
+    if (!reexport2) {
+      error(toString(this) + ": unable to locate re-export with install name " +
+            path);
+    }
+  }
 }
 
 DylibFile::DylibFile(MemoryBufferRef mb, DylibFile *umbrella,
diff --git a/lld/test/MachO/reexport-with-rpath.s b/lld/test/MachO/reexport-with-rpath.s
new file mode 100644
index 0000000000000..f1b00ddc9f7e5
--- /dev/null
+++ b/lld/test/MachO/reexport-with-rpath.s
@@ -0,0 +1,37 @@
+# REQUIRES: x86
+# RUN: rm -rf %t; split-file %s %t
+# RUN: mkdir -p %t/cc/two/three
+# RUN: mkdir -p %t/bb/two/three
+# RUN: mkdir -p %t/aa/two/three
+
+# RUN: llvm-mc -filetype=obj -triple=x86_64-apple-darwin %t/c.s -o %t/c.o
+# RUN: llvm-mc -filetype=obj -triple=x86_64-apple-darwin %t/b.s -o %t/b.o
+# RUN: llvm-mc -filetype=obj -triple=x86_64-apple-darwin %t/a.s -o %t/a.o
+
+# RUN: %lld -dylib -install_name @rpath/two/three/libCee.dylib %t/c.o -o %t/cc/two/three/libCee.dylib 
+# RUN: %lld -dylib -install_name @rpath/two/three/libBee.dylib -L%t/cc/two/three -sub_library libCee -lCee %t/b.o -o %t/bb/two/three/libBee.dylib -rpath %t/cc
+# RUN: %lld -dylib -install_name @rpath/two/three/libAee.dylib -L%t/bb/two/three -sub_library libBee -lBee %t/a.o -o %t/aa/two/three/libAee.dylib
+##################
+
+
+#--- c.s
+.text
+.global _c_func
+_c_func:
+  mov $0, %rax
+  ret
+
+#--- b.s
+.text
+.global _b_func
+
+_b_func:
+  mov $0, %rax
+  ret
+
+#--- a.s
+.text
+.global _a_func
+_a_func:
+  mov $0, %rax
+  ret

[llvmbot]

@llvm/pr-subscribers-lld-macho

Author: Vy Nguyen (oontvoo)

Changes

Details:
When we have the following scenario:

• lib_a re-exports lib_b
• lib_b re-exports @rpath/lib_c
  • lib_b contains LC_RPATH

Previously, lld-macho cannot find lib_c because it was attempting to resolve the '@rpath' from lib_b (which had no LC_RPATH defined). The change here is to also consider all the LC_RPATH rom lib_b when trying to find lib_c.

Inspired by real-life example when linking with libXCTestSwiftSupport.dylib (which re-exports XCTest, which re-exports XCTestCore)

---

Full diff: https://github.com/llvm/llvm-project/pull/135241.diff

2 Files Affected:

• (modified) lld/MachO/InputFiles.cpp (+21-3)
• (added) lld/test/MachO/reexport-with-rpath.s (+37)

diff --git a/lld/MachO/InputFiles.cpp b/lld/MachO/InputFiles.cpp
index 9adfbc9d3f6f5..9afa5feaa1693 100644
--- a/lld/MachO/InputFiles.cpp
+++ b/lld/MachO/InputFiles.cpp
@@ -1633,6 +1633,17 @@ static DylibFile *findDylib(StringRef path, DylibFile *umbrella,
       if (std::optional<StringRef> dylibPath = resolveDylibPath(newPath.str()))
         return loadDylib(*dylibPath, umbrella);
     }
+    // If not found in umbrella, try the rpaths specified via -rpath too.
+    for (StringRef rpath : config->runtimePaths) {
+      newPath.clear();
+      if (rpath.consume_front("@loader_path/")) {
+        fs::real_path(umbrella->getName(), newPath);
+        path::remove_filename(newPath);
+      }
+      path::append(newPath, rpath, path.drop_front(strlen("@rpath/")));
+      if (std::optional<StringRef> dylibPath = resolveDylibPath(newPath.str()))
+        return loadDylib(*dylibPath, umbrella);
+    }
   }
 
   // FIXME: Should this be further up?
@@ -1678,9 +1689,16 @@ static bool isImplicitlyLinked(StringRef path) {
 void DylibFile::loadReexport(StringRef path, DylibFile *umbrella,
                          const InterfaceFile *currentTopLevelTapi) {
   DylibFile *reexport = findDylib(path, umbrella, currentTopLevelTapi);
-  if (!reexport)
-    error(toString(this) + ": unable to locate re-export with install name " +
-          path);
+  if (!reexport) {
+    // If not found in umbrella, retry since some rpaths might have been
+    // defined in "this" dyblib (which contains the LC_REEXPORT_DYLIB cmd) and
+    // not in the umbrella.
+    DylibFile *reexport2 = findDylib(path, this, currentTopLevelTapi);
+    if (!reexport2) {
+      error(toString(this) + ": unable to locate re-export with install name " +
+            path);
+    }
+  }
 }
 
 DylibFile::DylibFile(MemoryBufferRef mb, DylibFile *umbrella,
diff --git a/lld/test/MachO/reexport-with-rpath.s b/lld/test/MachO/reexport-with-rpath.s
new file mode 100644
index 0000000000000..f1b00ddc9f7e5
--- /dev/null
+++ b/lld/test/MachO/reexport-with-rpath.s
@@ -0,0 +1,37 @@
+# REQUIRES: x86
+# RUN: rm -rf %t; split-file %s %t
+# RUN: mkdir -p %t/cc/two/three
+# RUN: mkdir -p %t/bb/two/three
+# RUN: mkdir -p %t/aa/two/three
+
+# RUN: llvm-mc -filetype=obj -triple=x86_64-apple-darwin %t/c.s -o %t/c.o
+# RUN: llvm-mc -filetype=obj -triple=x86_64-apple-darwin %t/b.s -o %t/b.o
+# RUN: llvm-mc -filetype=obj -triple=x86_64-apple-darwin %t/a.s -o %t/a.o
+
+# RUN: %lld -dylib -install_name @rpath/two/three/libCee.dylib %t/c.o -o %t/cc/two/three/libCee.dylib 
+# RUN: %lld -dylib -install_name @rpath/two/three/libBee.dylib -L%t/cc/two/three -sub_library libCee -lCee %t/b.o -o %t/bb/two/three/libBee.dylib -rpath %t/cc
+# RUN: %lld -dylib -install_name @rpath/two/three/libAee.dylib -L%t/bb/two/three -sub_library libBee -lBee %t/a.o -o %t/aa/two/three/libAee.dylib
+##################
+
+
+#--- c.s
+.text
+.global _c_func
+_c_func:
+  mov $0, %rax
+  ret
+
+#--- b.s
+.text
+.global _b_func
+
+_b_func:
+  mov $0, %rax
+  ret
+
+#--- a.s
+.text
+.global _a_func
+_a_func:
+  mov $0, %rax
+  ret

[nico]

#134925 recently was kind of similar. But that was about tbd files and this here is about mach-o files.

I wonder if our behavior here is consistent across .tbd and .dylib inputs.

[oontvoo]

Whoops - the last commit didn't get pushed successfully - I've added it in PR/137672