packed-backend: check whether the refname contains NULL binaries

We have already implemented the header consistency check for the raw
"packed-refs" file. Before we implement the consistency check for each
ref entry, let's analysis [1] which reports that "git fsck" cannot
detect some binary zeros.

"packed-backend.c::next_record" will use "check_refname_format" to check
the consistency of the refname. If it is not OK, the program will die.
So, we already have the code path and we must miss out something.

We use the following code to get the refname:

    strbuf_add(&iter->refname_buf, p, eol - p);
    iter->base.refname = iter->refname_buf.buf

In the above code, `p` is the start pointer of the refname and `eol` is
the next newline pointer. We calculate the length of the refname by
subtracting the two pointers. Then we add the memory range between `p`
and `eol` to get the refname.

However, if there are some NULL binaries in the memory range between `p`
and `eol`, we will see the refname as a valid ref name as long as the
memory range between `p` and the first occurred NULL binary is valid.

In order to catch above corruption, create a new function
"refname_contains_null" by checking whether the "refname.len" equals to
the length of the raw string pointer "refname.buf". If not equal, there
must be some NULL binaries in the refname.

Use this function in "next_record" function to die the program if
"refname_contains_null" returns true.

[1] https://lore.kernel.org/git/[email protected]/

Reported-by: R. Diez <[email protected]>
Mentored-by: Patrick Steinhardt <[email protected]>
Mentored-by: Karthik Nayak <[email protected]>
Signed-off-by: shejialuo <[email protected]>
Signed-off-by: Junio C Hamano <[email protected]>

Author: shejialuo

refs/packed-backend.c

@@ -494,6 +494,23 @@ static void verify_buffer_safe(struct snapshot *snapshot)
 				 last_line, eof - last_line);
 }
 
+/*
+ * When parsing the "packed-refs" file, we will parse it line by line.
+ * Because we know the start pointer of the refname and the next
+ * newline pointer, we could calculate the length of the refname by
+ * subtracting the two pointers. However, there is a corner case where
+ * the refname contains corrupted embedded NULL binaries. And
+ * `check_refname_format()` will not catch this when the truncated
+ * refname is still a valid refname. To prevent this, we need to check
+ * whether the refname contains the NULL binaries.
+ */
+static int refname_contains_null(struct strbuf refname)
+{
+	if (refname.len != strlen(refname.buf))
+		return 1;
+	return 0;
+}
+
 #define SMALL_FILE_SIZE (32*1024)
 
 /*
@@ -895,6 +912,9 @@ static int next_record(struct packed_ref_iterator *iter)
 	strbuf_add(&iter->refname_buf, p, eol - p);
 	iter->base.refname = iter->refname_buf.buf;
 
+	if (refname_contains_null(iter->refname_buf))
+		die("packed refname contains embedded NULL: %s", iter->base.refname);
+
 	if (check_refname_format(iter->base.refname, REFNAME_ALLOW_ONELEVEL)) {
 		if (!refname_is_safe(iter->base.refname))
 			die("packed refname is dangerous: %s",