extract: skip a hard link item that refers to itself, refs #10393 (1.4-maint)

src/borg/archive.py

@@ -917,6 +917,12 @@ def extract_item(self, item, restore_attrs=True, dry_run=False, stdout=False, sp
         # overwrite files outside the extraction directory (e.g. /etc/passwd).
         self.safe_dirs.discard(path)  # path is about to be (re)created; never trust a stale entry for it
         self._check_safe_parent(item.path)
+        if 'source' in item and hardlinkable(item.mode):
+            source = os.path.join(dest, *item.source.split(os.sep)[stripped_components:])
+            if hardlink_masters.get(item.source, (None, source))[1] == path:
+                # duplicate item: path was extracted already and is the link target of its hard link group,
+                # removing it would lose the file (and hard linking it to itself can not work).
+                return
         # Attempt to remove existing files, ignore errors on failure.
         # A trailing "." component means the extraction directory itself (the archive's
         # "." item, as created by "borg create ARCHIVE ."). posix requires rmdir() to

src/borg/archiver.py

@@ -1149,6 +1149,10 @@ def item_to_tarinfo(item, original_path):
                         linkname = source
                     else:
                         chunks, linkname = hardlink_masters.get(item.source, (None, source))
+                    if linkname == item.path:
+                        # duplicate item: path was added already and is the link target of its hard link group,
+                        # a hard link to itself would be useless (and some tar implementations refuse it).
+                        return None, stream
                     if linkname:
                         # Master was already added to the archive, add a hardlink reference to it.
                         tarinfo.type = tarfile.LNKTYPE

src/borg/fuse.py

@@ -421,6 +421,10 @@ def make_versioned_name(name, version, add_dir=False):
         if 'source' in item and hardlinkable(item.mode):
             source = os.sep.join(item.source.split(os.sep)[stripped_components:])
             chunks, link_target = hardlink_masters.get(item.source, (None, source))
+            if link_target == path:
+                # duplicate item: path was added already and is the link target of its hard link group,
+                # adding it again would only increase the link count.
+                return
             if link_target:
                 # Hard link was extracted previously, just link
                 link_target = os.fsencode(link_target)

src/borg/testsuite/archiver.py

@@ -1208,6 +1208,50 @@ def test_extract_hardlinks_twice(self):
             assert os.stat('input/a/hardlink').st_nlink == 2
             assert os.stat('input/b/hardlink').st_nlink == 2
 
+    @requires_hardlinks
+    def test_extract_duplicate_hardlink_item(self):
+        # an archive can have the same hard link item twice (here: the same path given twice via stdin).
+        # extracting the duplicate must not remove the file that is the link target of its hard link group (#10393).
+        self.create_regular_file('file1', contents=b'123456')
+        os.link(os.path.join(self.input_path, 'file1'), os.path.join(self.input_path, 'file2'))
+        self.cmd('init', '--encryption=none', self.repository_location)
+        self.cmd('create', '--paths-from-stdin', self.repository_location + '::test',
+                 input=b'input/file1\ninput/file1\ninput/file2\n')
+        output = self.cmd('list', '--format={path}{NL}', self.repository_location + '::test')
+        assert output.splitlines() == ['input/file1'] * 2 + ['input/file2']
+        with changedir('output'):
+            self.cmd('extract', self.repository_location + '::test')
+            assert os.stat('input/file1').st_nlink == 2
+            assert os.stat('input/file2').st_nlink == 2
+            with open('input/file1', 'rb') as f:
+                assert f.read() == b'123456'
+            self.cmd('extract', self.repository_location + '::test', '--strip-components', '1')
+            assert os.stat('file1').st_nlink == 2
+            assert os.stat('file2').st_nlink == 2
+            with open('file1', 'rb') as f:
+                assert f.read() == b'123456'
+
+    @requires_hardlinks
+    @unittest.skipUnless(llfuse, 'llfuse not installed')
+    def test_fuse_mount_duplicate_hardlink_item(self):
+        # an archive can have the same hard link item twice (here: the same path given twice via stdin).
+        # the duplicate must not increase the link count of its hard link group (#10393).
+        self.create_regular_file('file1', contents=b'123456')
+        os.link(os.path.join(self.input_path, 'file1'), os.path.join(self.input_path, 'file2'))
+        self.cmd('init', '--encryption=none', self.repository_location)
+        self.cmd('create', '--paths-from-stdin', self.repository_location + '::test',
+                 input=b'input/file1\ninput/file1\ninput/file2\n')
+        mountpoint = os.path.join(self.tmpdir, 'mountpoint')
+        if sys.platform == 'darwin':
+            ignore_perms = ['-o', 'ignore_permissions,defer_permissions']
+        else:
+            ignore_perms = ['-o', 'ignore_permissions']
+        with self.fuse_mount(self.repository_location + '::test', mountpoint, *ignore_perms), \
+             changedir(mountpoint):
+            assert os.stat('input/file1').st_nlink == 2
+            assert os.stat('input/file2').st_nlink == 2
+            assert open('input/file1', 'rb').read() == b'123456'
+
     def test_extract_include_exclude(self):
         self.cmd('init', '--encryption=repokey', self.repository_location)
         self.create_regular_file('file1', size=1024 * 80)
@@ -4410,6 +4454,20 @@ def test_extract_hardlinks_tar(self):
             assert os.stat('input/dir1/aaaa').st_nlink == 2
             assert os.stat('input/dir1/source2').st_nlink == 2
 
+    @requires_hardlinks
+    def test_export_tar_duplicate_hardlink_item(self):
+        # an archive can have the same hard link item twice (here: the same path given twice via stdin).
+        # export-tar must not add the duplicate as a hard link to itself (#10393).
+        self.create_regular_file('file1', contents=b'123456')
+        os.link(os.path.join(self.input_path, 'file1'), os.path.join(self.input_path, 'file2'))
+        self.cmd('init', '--encryption=none', self.repository_location)
+        self.cmd('create', '--paths-from-stdin', self.repository_location + '::test',
+                 input=b'input/file1\ninput/file1\ninput/file2\n')
+        self.cmd('export-tar', self.repository_location + '::test', 'output.tar')
+        with tarfile.open('output.tar') as tar:
+            members = [(m.name, m.type, m.linkname) for m in tar]
+        assert members == [('input/file1', tarfile.REGTYPE, ''), ('input/file2', tarfile.LNKTYPE, 'input/file1')]
+
     def test_import_tar(self):
         self.create_test_files()
         os.unlink('input/flagfile')