[PATCH] Fix stat.st_blocks for files compressed with CompactOS method

[Christian Franke]

Cygwin 2.8.0 returns stat.st_blocks = 0 if a file is compressed with 
CompactOS method (at least on Win10 1607):

Testcase:

$ ls -ls file
280 -rw-r--r-- 1 ... ... 285363 Apr 22 13:52 file

$ compact /c file
...
$ ls -ls file
56 -rw-r--r-- 1 ... ... 285363 Apr 22 13:52 file

$ compact /u file
...
$ compact /c /exe file
...
$ ls -ls file
0 -rw-r--r-- 1 ... ... 285363 Apr 22 13:52 file

This is because StandardInformation.AllocationSize is always 0 for 
theses files. CompressedFileSize returns the correct value.

This is likely related to the interesting method how these files are 
encoded in the MFT:
The default $DATA stream is a sparse stream with original size but no 
allocated blocks.
An alternate $DATA stream WofCompressedData contains the compressed data.
An additional $REPARSE_POINT possibly marks this file a special and lets 
accesses fail on older Windows releases (and on Linux, most current 
forensic tools, ...).

With the attached patch, stat.st_blocks work as expected:

$ ls -ls file
48 -rw-r--r-- 1 ... ... 285363 Apr 22 13:52 file

The only drawback is an unnecessary FileCompressionInformation query for 
sparse files with no blocks.

Christian

-------------- next part --------------
Always retrieve FileCompressionInformation for non-empty
files if FileStandardInformation returns 0 allocated blocks.
This fixes stat.st_blocks for files compressed with CompactOS method.

Signed-off-by: Christian Franke <franke@computer.org>
---
 winsup/cygwin/fhandler_disk_file.cc | 17 ++++++++++++-----
 1 file changed, 12 insertions(+), 5 deletions(-)

diff --git a/winsup/cygwin/fhandler_disk_file.cc b/winsup/cygwin/fhandler_disk_file.cc
index fcbe15c..bf5f988 100644
--- a/winsup/cygwin/fhandler_disk_file.cc
+++ b/winsup/cygwin/fhandler_disk_file.cc
@@ -463,18 +463,25 @@ fhandler_base::fstat_helper (struct stat *buf)
 
   buf->st_blksize = PREFERRED_IO_BLKSIZE;
 
-  if (pfai->StandardInformation.AllocationSize.QuadPart >= 0LL)
+  if (buf->st_size == 0
+      && pfai->StandardInformation.AllocationSize.QuadPart == 0LL)
+    /* File is empty and no blocks are preallocated. */
+    buf->st_blocks = 0;
+  else if (pfai->StandardInformation.AllocationSize.QuadPart > 0LL)
     /* A successful NtQueryInformationFile returns the allocation size
-       correctly for compressed and sparse files as well. */
+       correctly for compressed and sparse files as well.
+       Allocation size 0 is ignored here because (at least) Windows 10
+       1607 always returns 0 for CompactOS compressed files. */
     buf->st_blocks = (pfai->StandardInformation.AllocationSize.QuadPart
 		      + S_BLKSIZE - 1) / S_BLKSIZE;
-  else if (::has_attribute (attributes, FILE_ATTRIBUTE_COMPRESSED
-					| FILE_ATTRIBUTE_SPARSE_FILE)
+  else if ((pfai->StandardInformation.AllocationSize.QuadPart == 0LL
+	    || ::has_attribute (attributes, FILE_ATTRIBUTE_COMPRESSED
+					  | FILE_ATTRIBUTE_SPARSE_FILE))
 	   && h && !is_fs_special ()
 	   && !NtQueryInformationFile (h, &st, (PVOID) &fci, sizeof fci,
 				       FileCompressionInformation))
     /* Otherwise we request the actual amount of bytes allocated for
-       compressed and sparsed files. */
+       compressed, sparsed and CompactOS files. */
     buf->st_blocks = (fci.CompressedFileSize.QuadPart + S_BLKSIZE - 1)
 		     / S_BLKSIZE;
   else