OSDN Git Service

e2fsck: only release clusters when shortening a directory during a rehash
authorDarrick J. Wong <darrick.wong@oracle.com>
Wed, 11 Dec 2013 01:23:12 +0000 (17:23 -0800)
committerTheodore Ts'o <tytso@mit.edu>
Mon, 16 Dec 2013 04:52:23 +0000 (23:52 -0500)
When the rehash process is running on a bigalloc filesystem, it
compresses all the directory entries and hash structures into the
beginning of the directory file and then uses block_iterate3() to free
the blocks off the end of the file.  It seems to call
ext2fs_block_alloc_stats2() for every block in a cluster, which is
unfortunate because this function allocates and frees entire clusters
(and updates the summary counts accordingly).  In this case e2fsck
writes out incorrect summary counts.

Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
e2fsck/rehash.c

index ac0ff31..3aafbb1 100644 (file)
@@ -648,10 +648,18 @@ static int write_dir_block(ext2_filsys fs,
        if (blockcnt >= wd->outdir->num) {
                e2fsck_read_bitmaps(wd->ctx);
                blk = *block_nr;
-               ext2fs_unmark_block_bitmap2(wd->ctx->block_found_map, blk);
-               ext2fs_block_alloc_stats2(fs, blk, -1);
+               /*
+                * In theory, we only release blocks from the end of the
+                * directory file, so it's fine to clobber a whole cluster at
+                * once.
+                */
+               if (blk % EXT2FS_CLUSTER_RATIO(fs) == 0) {
+                       ext2fs_unmark_block_bitmap2(wd->ctx->block_found_map,
+                                                   blk);
+                       ext2fs_block_alloc_stats2(fs, blk, -1);
+                       wd->cleared++;
+               }
                *block_nr = 0;
-               wd->cleared++;
                return BLOCK_CHANGED;
        }
        if (blockcnt < 0)