fix hardlinked CACHEDIR.TAG processing, fixes #4911

src/borg/archive.py

@@ -1842,24 +1842,33 @@ class ArchiveRecreater:
         matcher = self.matcher
         tag_files = []
         tagged_dirs = []
-        # build hardlink masters, but only for paths ending in CACHE_TAG_NAME, so we can read hard-linked TAGs
+
+        # to support reading hard-linked CACHEDIR.TAGs (aka CACHE_TAG_NAME), similar to hardlink_masters:
         cachedir_masters = {}
 
+        if self.exclude_caches:
+            # sadly, due to how CACHEDIR.TAG works (filename AND file [header] contents) and
+            # how borg deals with hardlinks (slave hardlinks referring back to master hardlinks),
+            # we need to pass over the archive collecting hardlink master paths.
+            # as seen in issue #4911, the master paths can have an arbitrary filenames,
+            # not just CACHEDIR.TAG.
+            for item in archive.iter_items(filter=lambda item: os.path.basename(item.path) == CACHE_TAG_NAME):
+                if stat.S_ISREG(item.mode) and 'chunks' not in item and 'source' in item:
+                    # this is a hardlink slave, referring back to its hardlink master (via item.source)
+                    cachedir_masters[item.source] = None  # we know the key (path), but not the value (item) yet
+
         for item in archive.iter_items(
-                filter=lambda item: item.path.endswith(CACHE_TAG_NAME) or matcher.match(item.path)):
-            if item.path.endswith(CACHE_TAG_NAME):
+                filter=lambda item: os.path.basename(item.path) == CACHE_TAG_NAME or matcher.match(item.path)):
+            if self.exclude_caches and item.path in cachedir_masters:
                 cachedir_masters[item.path] = item
             dir, tag_file = os.path.split(item.path)
             if tag_file in self.exclude_if_present:
                 exclude(dir, item)
-            if stat.S_ISREG(item.mode):
-                if self.exclude_caches and tag_file == CACHE_TAG_NAME:
-                    if 'chunks' in item:
-                        file = open_item(archive, item)
-                    else:
-                        file = open_item(archive, cachedir_masters[item.source])
-                    if file.read(len(CACHE_TAG_CONTENTS)).startswith(CACHE_TAG_CONTENTS):
-                        exclude(dir, item)
+            elif self.exclude_caches and tag_file == CACHE_TAG_NAME and stat.S_ISREG(item.mode):
+                content_item = item if 'chunks' in item else cachedir_masters[item.source]
+                file = open_item(archive, content_item)
+                if file.read(len(CACHE_TAG_CONTENTS)) == CACHE_TAG_CONTENTS:
+                    exclude(dir, item)
         matcher.add(tag_files, IECommand.Include)
         matcher.add(tagged_dirs, IECommand.ExcludeNoRecurse)