]> git.karo-electronics.de Git - linux-beck.git/commitdiff
ext4: teach ext4_statfs() to deal with clusters if bigalloc is enabled
authorTheodore Ts'o <tytso@mit.edu>
Fri, 9 Sep 2011 23:00:51 +0000 (19:00 -0400)
committerTheodore Ts'o <tytso@mit.edu>
Fri, 9 Sep 2011 23:00:51 +0000 (19:00 -0400)
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
fs/ext4/super.c

index d7e0e045b11b030d073768a91702bc787ed47fec..6810957e0ac75c713a76811b2314bbf52d6cf19a 100644 (file)
@@ -4580,16 +4580,34 @@ restore_opts:
        return err;
 }
 
+/*
+ * Note: calculating the overhead so we can be compatible with
+ * historical BSD practice is quite difficult in the face of
+ * clusters/bigalloc.  This is because multiple metadata blocks from
+ * different block group can end up in the same allocation cluster.
+ * Calculating the exact overhead in the face of clustered allocation
+ * requires either O(all block bitmaps) in memory or O(number of block
+ * groups**2) in time.  We will still calculate the superblock for
+ * older file systems --- and if we come across with a bigalloc file
+ * system with zero in s_overhead_clusters the estimate will be close to
+ * correct especially for very large cluster sizes --- but for newer
+ * file systems, it's better to calculate this figure once at mkfs
+ * time, and store it in the superblock.  If the superblock value is
+ * present (even for non-bigalloc file systems), we will use it.
+ */
 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
 {
        struct super_block *sb = dentry->d_sb;
        struct ext4_sb_info *sbi = EXT4_SB(sb);
        struct ext4_super_block *es = sbi->s_es;
+       struct ext4_group_desc *gdp;
        u64 fsid;
        s64 bfree;
 
        if (test_opt(sb, MINIX_DF)) {
                sbi->s_overhead_last = 0;
+       } else if (es->s_overhead_clusters) {
+               sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
        } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
                ext4_group_t i, ngroups = ext4_get_groups_count(sb);
                ext4_fsblk_t overhead = 0;
@@ -4604,24 +4622,16 @@ static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
                 * All of the blocks before first_data_block are
                 * overhead
                 */
-               overhead = le32_to_cpu(es->s_first_data_block);
+               overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
 
                /*
-                * Add the overhead attributed to the superblock and
-                * block group descriptors.  If the sparse superblocks
-                * feature is turned on, then not all groups have this.
+                * Add the overhead found in each block group
                 */
                for (i = 0; i < ngroups; i++) {
-                       overhead += ext4_bg_has_super(sb, i) +
-                               ext4_bg_num_gdb(sb, i);
+                       gdp = ext4_get_group_desc(sb, i, NULL);
+                       overhead += ext4_num_overhead_clusters(sb, i, gdp);
                        cond_resched();
                }
-
-               /*
-                * Every block group has an inode bitmap, a block
-                * bitmap, and an inode table.
-                */
-               overhead += ngroups * (2 + sbi->s_itb_per_group);
                sbi->s_overhead_last = overhead;
                smp_wmb();
                sbi->s_blocks_last = ext4_blocks_count(es);
@@ -4629,7 +4639,8 @@ static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
 
        buf->f_type = EXT4_SUPER_MAGIC;
        buf->f_bsize = sb->s_blocksize;
-       buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
+       buf->f_blocks = (ext4_blocks_count(es) -
+                        EXT4_C2B(sbi, sbi->s_overhead_last));
        bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
                percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
        /* prevent underflow in case that few free space is available */