Add support for clobbered blocks

In ext4 filesystems, some blocks might be changed even being mounted
R/O, such as the superblock (block 0). We need to exclude such blocks
from integrity verification. Plus such blocks should always be
written to the target by copying instead of patching.

Bug: http://b/20939131
Change-Id: I991169ec307dfb231b2fe8908a0668595ecb2060

tools/releasetools/blockimgdiff.py

@@ -82,6 +82,7 @@ class EmptyImage(Image):
   """A zero-length image."""
   blocksize = 4096
   care_map = RangeSet()
+  clobbered_blocks = RangeSet()
   total_blocks = 0
   file_map = {}
   def ReadRangeSet(self, ranges):
@@ -114,6 +115,7 @@ class DataImage(Image):
 
     self.total_blocks = len(self.data) / self.blocksize
     self.care_map = RangeSet(data=(0, self.total_blocks))
+    self.clobbered_blocks = RangeSet()
 
     zero_blocks = []
     nonzero_blocks = []
@@ -135,6 +137,8 @@ class DataImage(Image):
     return [self.data[s*self.blocksize:e*self.blocksize] for (s, e) in ranges]
 
   def TotalSha1(self):
+    # DataImage always carries empty clobbered_blocks.
+    assert self.clobbered_blocks.size() == 0
     return sha1(self.data).hexdigest()
 
 
@@ -184,6 +188,10 @@ class Transfer(object):
 #      (Typically a domain is a file, and the key in file_map is the
 #      pathname.)
 #
+#    clobbered_blocks: a RangeSet containing which blocks contain data
+#      but may be altered by the FS. They need to be excluded when
+#      verifying the partition integrity.
+#
 #    ReadRangeSet(): a function that takes a RangeSet and returns the
 #      data contained in the image blocks of that RangeSet.  The data
 #      is returned as a list or tuple of strings; concatenating the
@@ -193,7 +201,7 @@ class Transfer(object):
 #
 #    TotalSha1(): a function that returns (as a hex string) the SHA-1
 #      hash of all the data in the image (ie, all the blocks in the
-#      care_map)
+#      care_map minus clobbered_blocks).
 #
 # When creating a BlockImageDiff, the src image may be None, in which
 # case the list of transfers produced will never read from the
@@ -445,7 +453,6 @@ class BlockImageDiff(object):
       if free_string:
         out.append("".join(free_string))
 
-
       # sanity check: abort if we're going to need more than 512 MB if
       # stash space
       assert max_stashed_blocks * self.tgt.blocksize < (512 << 20)
@@ -845,6 +852,12 @@ class BlockImageDiff(object):
                  "zero", self.transfers)
         continue
 
+      elif tgt_fn == "__COPY":
+        # "__COPY" domain includes all the blocks not contained in any
+        # file and that need to be copied unconditionally to the target.
+        Transfer(tgt_fn, None, tgt_ranges, empty, "new", self.transfers)
+        continue
+
       elif tgt_fn in self.src.file_map:
         # Look for an exact pathname match in the source.
         Transfer(tgt_fn, tgt_fn, tgt_ranges, self.src.file_map[tgt_fn],

tools/releasetools/common.py

@@ -1141,6 +1141,9 @@ class BlockDifference(object):
     self.partition = partition
     self.check_first_block = check_first_block
 
+    # Due to http://b/20939131, check_first_block is disabled temporarily.
+    assert not self.check_first_block
+
     if version is None:
       version = 1
       if OPTIONS.info_dict:
@@ -1174,18 +1177,18 @@ class BlockDifference(object):
     if not self.src:
       script.Print("Image %s will be patched unconditionally." % (partition,))
     else:
+      ranges = self.src.care_map.subtract(self.src.clobbered_blocks)
+      ranges_str = ranges.to_string_raw()
       if self.version >= 3:
         script.AppendExtra(('if (range_sha1("%s", "%s") == "%s" || '
                             'block_image_verify("%s", '
                             'package_extract_file("%s.transfer.list"), '
                             '"%s.new.dat", "%s.patch.dat")) then') % (
-                            self.device, self.src.care_map.to_string_raw(),
+                            self.device, ranges_str, self.src.TotalSha1(),
-                            self.src.TotalSha1(),
                             self.device, partition, partition, partition))
       else:
         script.AppendExtra('if range_sha1("%s", "%s") == "%s" then' % (
-            self.device, self.src.care_map.to_string_raw(),
+                           self.device, ranges_str, self.src.TotalSha1()))
-            self.src.TotalSha1()))
       script.Print('Verified %s image...' % (partition,))
       script.AppendExtra('else')
 
@@ -1233,6 +1236,9 @@ class BlockDifference(object):
 
     return ctx.hexdigest()
 
+  # TODO(tbao): Due to http://b/20939131, block 0 may be changed without
+  # remounting R/W. Will change the checking to a finer-grained way to
+  # mask off those bits.
   def _CheckFirstBlock(self, script):
     r = rangelib.RangeSet((0, 1))
     srchash = self._HashBlocks(self.src, r)

tools/releasetools/ota_from_target_files.py

@@ -475,7 +475,13 @@ def GetImage(which, tmpdir, info_dict):
       path = add_img_to_target_files.BuildVendor(
           tmpdir, info_dict, block_list=mappath)
 
-  return sparse_img.SparseImage(path, mappath)
+  # Bug: http://b/20939131
+  # In ext4 filesystems, block 0 might be changed even being mounted
+  # R/O. We add it to clobbered_blocks so that it will be written to the
+  # target unconditionally. Note that they are still part of care_map.
+  clobbered_blocks = "0"
+
+  return sparse_img.SparseImage(path, mappath, clobbered_blocks)
 
 
 def WriteFullOTAPackage(input_zip, output_zip):
@@ -773,7 +779,6 @@ def WriteBlockIncrementalOTAPackage(target_zip, source_zip, output_zip):
         OPTIONS.info_dict.get("blockimgdiff_versions", "1").split(","))
 
   system_diff = common.BlockDifference("system", system_tgt, system_src,
-                                       check_first_block=True,
                                        version=blockimgdiff_version)
 
   if HasVendorPartition(target_zip):
@@ -784,7 +789,6 @@ def WriteBlockIncrementalOTAPackage(target_zip, source_zip, output_zip):
     vendor_tgt = GetImage("vendor", OPTIONS.target_tmp,
                           OPTIONS.target_info_dict)
     vendor_diff = common.BlockDifference("vendor", vendor_tgt, vendor_src,
-                                         check_first_block=True,
                                          version=blockimgdiff_version)
   else:
     vendor_diff = None

tools/releasetools/sparse_img.py

@@ -21,10 +21,17 @@ import rangelib
 
 
 class SparseImage(object):
-  """Wraps a sparse image file (and optional file map) into an image
+  """Wraps a sparse image file into an image object.
-  object suitable for passing to BlockImageDiff."""
 
-  def __init__(self, simg_fn, file_map_fn=None):
+  Wraps a sparse image file (and optional file map and clobbered_blocks) into
+  an image object suitable for passing to BlockImageDiff. file_map contains
+  the mapping between files and their blocks. clobbered_blocks contains the set
+  of blocks that should be always written to the target regardless of the old
+  contents (i.e. copying instead of patching). clobbered_blocks should be in
+  the form of a string like "0" or "0 1-5 8".
+  """
+
+  def __init__(self, simg_fn, file_map_fn=None, clobbered_blocks=None):
     self.simg_f = f = open(simg_fn, "rb")
 
     header_bin = f.read(28)
@@ -57,6 +64,7 @@ class SparseImage(object):
     pos = 0   # in blocks
     care_data = []
     self.offset_map = offset_map = []
+    self.clobbered_blocks = rangelib.RangeSet(data=clobbered_blocks)
 
     for i in range(total_chunks):
       header_bin = f.read(12)
@@ -103,7 +111,7 @@ class SparseImage(object):
     self.offset_index = [i[0] for i in offset_map]
 
     if file_map_fn:
-      self.LoadFileBlockMap(file_map_fn)
+      self.LoadFileBlockMap(file_map_fn, self.clobbered_blocks)
     else:
       self.file_map = {"__DATA": self.care_map}
 
@@ -111,9 +119,10 @@ class SparseImage(object):
     return [d for d in self._GetRangeData(ranges)]
 
   def TotalSha1(self):
-    """Return the SHA-1 hash of all data in the 'care' regions of this image."""
+    """Return the SHA-1 hash of all data in the 'care' regions but not in
+    clobbered_blocks of this image."""
     h = sha1()
-    for d in self._GetRangeData(self.care_map):
+    for d in self._GetRangeData(self.care_map.subtract(self.clobbered_blocks)):
       h.update(d)
     return h.hexdigest()
 
@@ -156,7 +165,7 @@ class SparseImage(object):
           yield fill_data * (this_read * (self.blocksize >> 2))
         to_read -= this_read
 
-  def LoadFileBlockMap(self, fn):
+  def LoadFileBlockMap(self, fn, clobbered_blocks):
     remaining = self.care_map
     self.file_map = out = {}
 
@@ -166,14 +175,20 @@ class SparseImage(object):
         ranges = rangelib.RangeSet.parse(ranges)
         out[fn] = ranges
         assert ranges.size() == ranges.intersect(remaining).size()
+
+        # Currently we assume that blocks in clobbered_blocks are not part of
+        # any file.
+        assert not clobbered_blocks.overlaps(ranges)
         remaining = remaining.subtract(ranges)
 
+    remaining = remaining.subtract(clobbered_blocks)
+
     # For all the remaining blocks in the care_map (ie, those that
-    # aren't part of the data for any file), divide them into blocks
+    # aren't part of the data for any file nor part of the clobbered_blocks),
-    # that are all zero and blocks that aren't.  (Zero blocks are
+    # divide them into blocks that are all zero and blocks that aren't.
-    # handled specially because (1) there are usually a lot of them
+    # (Zero blocks are handled specially because (1) there are usually
-    # and (2) bsdiff handles files with long sequences of repeated
+    # a lot of them and (2) bsdiff handles files with long sequences of
-    # bytes especially poorly.)
+    # repeated bytes especially poorly.)
 
     zero_blocks = []
     nonzero_blocks = []
@@ -203,6 +218,7 @@ class SparseImage(object):
 
     out["__ZERO"] = rangelib.RangeSet(data=zero_blocks)
     out["__NONZERO"] = rangelib.RangeSet(data=nonzero_blocks)
+    out["__COPY"] = clobbered_blocks
 
   def ResetFileMap(self):
     """Throw away the file map and treat the entire image as