[SystemZ][z/OS] Update autoconversion functions to improve support for UTF-8 (#98652)

This fixes the following error when reading source and header files on
z/OS: error: source file is not valid UTF-8

Author: abhina-sree

clang/include/clang/Basic/FileEntry.h

@@ -70,6 +70,11 @@ class FileEntryRef {
   const FileEntry &getFileEntry() const {
     return *getBaseMapEntry().second->V.get<FileEntry *>();
   }
+
+  // This function is used if the buffer size needs to be increased
+  // due to potential z/OS EBCDIC -> UTF-8 conversion
+  inline void updateFileEntryBufferSize(unsigned BufferSize);
+
   DirectoryEntryRef getDir() const { return ME->second->Dir; }
 
   inline off_t getSize() const;
@@ -323,6 +328,8 @@ class FileEntry {
 
   StringRef tryGetRealPathName() const { return RealPathName; }
   off_t getSize() const { return Size; }
+  // Size may increase due to potential z/OS EBCDIC -> UTF-8 conversion.
+  void setSize(off_t NewSize) { Size = NewSize; }
   unsigned getUID() const { return UID; }
   const llvm::sys::fs::UniqueID &getUniqueID() const { return UniqueID; }
   time_t getModificationTime() const { return ModTime; }
@@ -353,6 +360,10 @@ bool FileEntryRef::isNamedPipe() const { return getFileEntry().isNamedPipe(); }
 
 void FileEntryRef::closeFile() const { getFileEntry().closeFile(); }
 
+void FileEntryRef::updateFileEntryBufferSize(unsigned BufferSize) {
+  getBaseMapEntry().second->V.get<FileEntry *>()->setSize(BufferSize);
+}
+
 } // end namespace clang
 
 #endif // LLVM_CLANG_BASIC_FILEENTRY_H

clang/lib/Basic/SourceManager.cpp

@@ -24,6 +24,7 @@
 #include "llvm/ADT/StringRef.h"
 #include "llvm/ADT/StringSwitch.h"
 #include "llvm/Support/Allocator.h"
+#include "llvm/Support/AutoConvert.h"
 #include "llvm/Support/Capacity.h"
 #include "llvm/Support/Compiler.h"
 #include "llvm/Support/Endian.h"
@@ -156,8 +157,11 @@ ContentCache::getBufferOrNone(DiagnosticsEngine &Diag, FileManager &FM,
   // Unless this is a named pipe (in which case we can handle a mismatch),
   // check that the file's size is the same as in the file entry (which may
   // have come from a stat cache).
+  // The buffer will always be larger than the file size on z/OS in the presence
+  // of characters outside the base character set.
+  assert(Buffer->getBufferSize() >= (size_t)ContentsEntry->getSize());
   if (!ContentsEntry->isNamedPipe() &&
-      Buffer->getBufferSize() != (size_t)ContentsEntry->getSize()) {
+      Buffer->getBufferSize() < (size_t)ContentsEntry->getSize()) {
     Diag.Report(Loc, diag::err_file_modified) << ContentsEntry->getName();
 
     return std::nullopt;
@@ -583,6 +587,18 @@ SourceManager::getOrCreateFileID(FileEntryRef SourceFile,
 					  FileCharacter);
 }
 
+/// Helper function to determine if an input file requires conversion
+bool needConversion(StringRef Filename) {
+#ifdef __MVS__
+  llvm::ErrorOr<bool> NeedConversion =
+      llvm::needzOSConversion(Filename.str().c_str());
+  assert(NeedConversion && "Filename was not found");
+  return *NeedConversion;
+#else
+  return false;
+#endif
+}
+
 /// createFileID - Create a new FileID for the specified ContentCache and
 /// include position.  This works regardless of whether the ContentCache
 /// corresponds to a file or some other input source.
@@ -602,6 +618,20 @@ FileID SourceManager::createFileIDImpl(ContentCache &File, StringRef Filename,
     return FileID::get(LoadedID);
   }
   unsigned FileSize = File.getSize();
+  bool NeedConversion = needConversion(Filename);
+  if (NeedConversion) {
+    // Buffer size may increase due to potential z/OS EBCDIC to UTF-8
+    // conversion.
+    if (std::optional<llvm::MemoryBufferRef> Buffer =
+            File.getBufferOrNone(Diag, getFileManager())) {
+      unsigned BufSize = Buffer->getBufferSize();
+      if (BufSize > FileSize) {
+        if (File.ContentsEntry.has_value())
+          File.ContentsEntry->updateFileEntryBufferSize(BufSize);
+        FileSize = BufSize;
+      }
+    }
+  }
   if (!(NextLocalOffset + FileSize + 1 > NextLocalOffset &&
         NextLocalOffset + FileSize + 1 <= CurrentLoadedOffset)) {
     Diag.Report(IncludePos, diag::err_sloc_space_too_large);

llvm/include/llvm/Support/AutoConvert.h

@@ -17,6 +17,7 @@
 #ifdef __MVS__
 #include <_Ccsid.h>
 #ifdef __cplusplus
+#include "llvm/Support/ErrorOr.h"
 #include <system_error>
 #endif /* __cplusplus */
 
@@ -54,8 +55,14 @@ std::error_code restorezOSStdHandleAutoConversion(int FD);
 /** \brief Set the tag information for a file descriptor. */
 std::error_code setzOSFileTag(int FD, int CCSID, bool Text);
 
-} /* namespace llvm */
-#endif /* __cplusplus */
+// Get the the tag ccsid for a file name or a file descriptor.
+ErrorOr<__ccsid_t> getzOSFileTag(const char *FileName, const int FD = -1);
+
+// Query the file tag to determine if it needs conversion to UTF-8 codepage.
+ErrorOr<bool> needzOSConversion(const char *FileName, const int FD = -1);
+
+} // namespace llvm
+#endif // __cplusplus
 
 #endif /* __MVS__ */
 

llvm/lib/Support/AutoConvert.cpp

@@ -20,6 +20,8 @@
 #include <sys/stat.h>
 #include <unistd.h>
 
+using namespace llvm;
+
 static int savedStdHandleAutoConversionMode[3] = {-1, -1, -1};
 
 int disablezOSAutoConversion(int FD) {
@@ -116,4 +118,40 @@ std::error_code llvm::setzOSFileTag(int FD, int CCSID, bool Text) {
   return std::error_code();
 }
 
-#endif // __MVS__
+ErrorOr<__ccsid_t> llvm::getzOSFileTag(const char *FileName, const int FD) {
+  // If we have a file descriptor, use it to find out file tagging. Otherwise we
+  // need to use stat() with the file path.
+  if (FD != -1) {
+    struct f_cnvrt Query = {
+        QUERYCVT, // cvtcmd
+        0,        // pccsid
+        0,        // fccsid
+    };
+    if (fcntl(FD, F_CONTROL_CVT, &Query) == -1)
+      return std::error_code(errno, std::generic_category());
+    return Query.fccsid;
+  }
+  struct stat Attr;
+  if (stat(FileName, &Attr) == -1)
+    return std::error_code(errno, std::generic_category());
+  return Attr.st_tag.ft_ccsid;
+}
+
+ErrorOr<bool> llvm::needzOSConversion(const char *FileName, const int FD) {
+  ErrorOr<__ccsid_t> Ccsid = getzOSFileTag(FileName, FD);
+  if (std::error_code EC = Ccsid.getError())
+    return EC;
+  // We don't need conversion for UTF-8 tagged files or binary files.
+  // TODO: Remove the assumption of ISO8859-1 = UTF-8 here when we fully resolve
+  // problems related to UTF-8 tagged source files.
+  switch (*Ccsid) {
+  case CCSID_UTF_8:
+  case CCSID_ISO8859_1:
+  case FT_BINARY:
+    return false;
+  default:
+    return true;
+  }
+}
+
+#endif //__MVS__

llvm/lib/Support/MemoryBuffer.cpp

@@ -361,6 +361,11 @@ static bool shouldUseMmap(sys::fs::file_t FD,
                           bool RequiresNullTerminator,
                           int PageSize,
                           bool IsVolatile) {
+#if defined(__MVS__)
+  // zOS Enhanced ASCII auto convert does not support mmap.
+  return false;
+#endif
+
   // mmap may leave the buffer without null terminator if the file size changed
   // by the time the last page is mapped in, so avoid it if the file size is
   // likely to change.
@@ -503,9 +508,16 @@ getOpenFileImpl(sys::fs::file_t FD, const Twine &Filename, uint64_t FileSize,
   }
 
 #ifdef __MVS__
-  // Set codepage auto-conversion for z/OS.
-  if (auto EC = llvm::enablezOSAutoConversion(FD))
+  ErrorOr<bool> NeedConversion = needzOSConversion(Filename.str().c_str(), FD);
+  if (std::error_code EC = NeedConversion.getError())
     return EC;
+  // File size may increase due to EBCDIC -> UTF-8 conversion, therefore we
+  // cannot trust the file size and we create the memory buffer by copying
+  // off the stream.
+  // Note: This only works with the assumption of reading a full file (i.e,
+  // Offset == 0 and MapSize == FileSize). Reading a file slice does not work.
+  if (Offset == 0 && MapSize == FileSize && *NeedConversion)
+    return getMemoryBufferForStream(FD, Filename);
 #endif
 
   auto Buf =