qcow2-refcount: avoid eating RAM

qcow2_inc_refcounts_imrt() (through realloc_refcount_array()) can eat
an unpredictable amount of memory on corrupted table entries, which are
referencing regions far beyond the end of file.

Prevent this, by skipping such regions from further processing.

Interesting that iotest 138 checks exactly the behavior which we fix
here. So, change the test appropriately.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190227131433.197063-3-vsementsov@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

block/qcow2-refcount.c

@@ -1520,12 +1520,31 @@ int qcow2_inc_refcounts_imrt(BlockDriverState *bs, BdrvCheckResult *res,
 {
     BDRVQcow2State *s = bs->opaque;
     uint64_t start, last, cluster_offset, k, refcount;
+    int64_t file_len;
     int ret;
 
     if (size <= 0) {
         return 0;
     }
 
+    file_len = bdrv_getlength(bs->file->bs);
+    if (file_len < 0) {
+        return file_len;
+    }
+
+    /*
+     * Last cluster of qcow2 image may be semi-allocated, so it may be OK to
+     * reference some space after file end but it should be less than one
+     * cluster.
+     */
+    if (offset + size - file_len >= s->cluster_size) {
+        fprintf(stderr, "ERROR: counting reference for region exceeding the "
+                "end of the file by one cluster or more: offset 0x%" PRIx64
+                " size 0x%" PRIx64 "\n", offset, size);
+        res->corruptions++;
+        return 0;
+    }
+
     start = start_of_cluster(s, offset);
     last = start_of_cluster(s, offset + size - 1);
     for(cluster_offset = start; cluster_offset <= last;

tests/qemu-iotests/138

@@ -54,15 +54,13 @@ $QEMU_IO -c 'write 0 512' "$TEST_IMG" | _filter_qemu_io
 # Put the data cluster at a multiple of 2 TB, resulting in the image apparently
 # having a multiple of 2^32 clusters
 # (To be more specific: It is at 32 PB)
-poke_file "$TEST_IMG" 2048 "\x80\x80\x00\x00\x00\x00\x00\x00"
+poke_file "$TEST_IMG" $((2048 + 8)) "\x00\x80\x00\x00\x00\x00\x00\x00"
 
 # An offset of 32 PB results in qemu-img check having to allocate an in-memory
-# refcount table of 128 TB (16 bit refcounts, 512 byte clusters).
+# refcount table of 128 TB (16 bit refcounts, 512 byte clusters), if qemu-img
-# This should be generally too much for any system and thus fail.
+# don't check that referenced data cluster is far beyond the end of file.
-# What this test is checking is that the qcow2 driver actually tries to allocate
+# But starting from 4.0, qemu-img does this check, and instead of "Cannot
-# such a large amount of memory (and is consequently aborting) instead of having
+# allocate memory", we have an error showing that l2 entry is invalid.
-# truncated the cluster count somewhere (which would result in much less memory
-# being allocated and then a segfault occurring).
 _check_test_img
 
 # success, all done

tests/qemu-iotests/138.out

@@ -5,5 +5,8 @@ QA output created by 138
 Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=512
 wrote 512/512 bytes at offset 0
 512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
-qemu-img: Check failed: Cannot allocate memory
+ERROR: counting reference for region exceeding the end of the file by one cluster or more: offset 0x80000000000000 size 0x200
+
+1 errors were found on the image.
+Data may be corrupted, or further writes to the image may corrupt it.
 *** done