MDEV-20813: Do not rotate keys for unallocated pages

fil_crypt_rotate_page(): Skip the key rotation for pages that carry 0
in FIL_PAGE_TYPE. This avoids not only unnecessary writes, but also
failures of the recently added debug assertion in
buf_flush_init_for_writing() that the FIL_PAGE_TYPE should be nonzero.

Note: the debug assertion can fail if the file was originally created
before MySQL 5.5. In old InnoDB versions, FIL_PAGE_TYPE was only
initialized for B-tree pages, to FIL_PAGE_INDEX. For any other pages,
the field could be garbage, including FIL_PAGE_INDEX. In MariaDB 10.2
and later, buf_flush_init_for_writing() would initialize the
FIL_PAGE_TYPE on such old pages, but only after passing the debug
assertion that insists that pages have a nonzero FIL_PAGE_TYPE.
Thus, the debug assertion at the start of buf_flush_init_for_writing()
can fail when upgrading from very old debug files. This assertion is
only present in debug builds, not release builds.

storage/innobase/fil/fil0crypt.cc

@@ -1773,8 +1773,6 @@ fil_crypt_rotate_page(
 		return;
 	}
 
-	ut_d(const bool was_free = fseg_page_is_free(space, offset));
-
 	mtr_t mtr;
 	mtr.start();
 	if (buf_block_t* block = fil_crypt_get_page_throttle(state,
@@ -1789,9 +1787,9 @@ fil_crypt_rotate_page(
 		if (space->is_stopping()) {
 			/* The tablespace is closing (in DROP TABLE or
 			TRUNCATE TABLE or similar): avoid further access */
-		} else if (!*reinterpret_cast<uint32_t*>(FIL_PAGE_OFFSET
-							 + frame)) {
-			/* It looks like this page was never
+		} else if (!kv && !*reinterpret_cast<uint16_t*>
+			   (&frame[FIL_PAGE_TYPE])) {
+			/* It looks like this page is not
 			allocated. Because key rotation is accessing
 			pages in a pattern that is unlike the normal
 			B-tree and undo log access pattern, we cannot
@@ -1801,9 +1799,20 @@ fil_crypt_rotate_page(
 			tablespace latch before acquiring block->lock,
 			then the fseg_page_is_free() information
 			could be stale already. */
-			ut_ad(was_free);
-			ut_ad(kv == 0);
-			ut_ad(page_get_space_id(frame) == 0);
+
+			/* If the data file was originally created
+			before MariaDB 10.0 or MySQL 5.6, some
+			allocated data pages could carry 0 in
+			FIL_PAGE_TYPE. The FIL_PAGE_TYPE on those
+			pages will be updated in
+			buf_flush_init_for_writing() when the page
+			is modified the next time.
+
+			Also, when the doublewrite buffer pages are
+			allocated on bootstrap in a non-debug build,
+			some dummy pages will be allocated, with 0 in
+			the FIL_PAGE_TYPE. Those pages should be
+			skipped from key rotation forever. */
 		} else if (fil_crypt_needs_rotation(
 				crypt_data,
 				kv,