ext4: revert commit which was causing fs corruption after journal replays
authorTheodore Ts'o <tytso@mit.edu>
Fri, 11 Jul 2014 17:55:40 +0000 (13:55 -0400)
committerTheodore Ts'o <tytso@mit.edu>
Fri, 11 Jul 2014 17:55:40 +0000 (13:55 -0400)
Commit 007649375f6af2 ("ext4: initialize multi-block allocator before
checking block descriptors") causes the block group descriptor's count
of the number of free blocks to become inconsistent with the number of
free blocks in the allocation bitmap.  This is a harmless form of fs
corruption, but it causes the kernel to potentially remount the file
system read-only, or to panic, depending on the file systems's error
behavior.

Thanks to Eric Whitney for his tireless work to reproduce and to find
the guilty commit.

Fixes: 007649375f6af2 ("ext4: initialize multi-block allocator before checking block descriptors"
Cc: stable@vger.kernel.org # 3.15
Reported-by: David Jander <david@protonic.nl>
Reported-by: Matteo Croce <technoboy85@gmail.com>
Tested-by: Eric Whitney <enwlinux@gmail.com>
Suggested-by: Eric Whitney <enwlinux@gmail.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
fs/ext4/super.c

index 6297c07c937f1d962196fb92a29497a16eaa617c..6df7bc611dbdc12c743f786ba093694521292fbf 100644 (file)
@@ -3879,38 +3879,19 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
                        goto failed_mount2;
                }
        }
-
-       /*
-        * set up enough so that it can read an inode,
-        * and create new inode for buddy allocator
-        */
-       sbi->s_gdb_count = db_count;
-       if (!test_opt(sb, NOLOAD) &&
-           EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
-               sb->s_op = &ext4_sops;
-       else
-               sb->s_op = &ext4_nojournal_sops;
-
-       ext4_ext_init(sb);
-       err = ext4_mb_init(sb);
-       if (err) {
-               ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
-                        err);
-               goto failed_mount2;
-       }
-
        if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
                ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
-               goto failed_mount2a;
+               goto failed_mount2;
        }
        if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
                if (!ext4_fill_flex_info(sb)) {
                        ext4_msg(sb, KERN_ERR,
                               "unable to initialize "
                               "flex_bg meta info!");
-                       goto failed_mount2a;
+                       goto failed_mount2;
                }
 
+       sbi->s_gdb_count = db_count;
        get_random_bytes(&sbi->s_next_generation, sizeof(u32));
        spin_lock_init(&sbi->s_next_gen_lock);
 
@@ -3945,6 +3926,14 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
        sbi->s_stripe = ext4_get_stripe_size(sbi);
        sbi->s_extent_max_zeroout_kb = 32;
 
+       /*
+        * set up enough so that it can read an inode
+        */
+       if (!test_opt(sb, NOLOAD) &&
+           EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
+               sb->s_op = &ext4_sops;
+       else
+               sb->s_op = &ext4_nojournal_sops;
        sb->s_export_op = &ext4_export_ops;
        sb->s_xattr = ext4_xattr_handlers;
 #ifdef CONFIG_QUOTA
@@ -4134,13 +4123,21 @@ no_journal:
        if (err) {
                ext4_msg(sb, KERN_ERR, "failed to reserve %llu clusters for "
                         "reserved pool", ext4_calculate_resv_clusters(sb));
-               goto failed_mount5;
+               goto failed_mount4a;
        }
 
        err = ext4_setup_system_zone(sb);
        if (err) {
                ext4_msg(sb, KERN_ERR, "failed to initialize system "
                         "zone (%d)", err);
+               goto failed_mount4a;
+       }
+
+       ext4_ext_init(sb);
+       err = ext4_mb_init(sb);
+       if (err) {
+               ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
+                        err);
                goto failed_mount5;
        }
 
@@ -4217,8 +4214,11 @@ failed_mount8:
 failed_mount7:
        ext4_unregister_li_request(sb);
 failed_mount6:
-       ext4_release_system_zone(sb);
+       ext4_mb_release(sb);
 failed_mount5:
+       ext4_ext_release(sb);
+       ext4_release_system_zone(sb);
+failed_mount4a:
        dput(sb->s_root);
        sb->s_root = NULL;
 failed_mount4:
@@ -4242,14 +4242,11 @@ failed_mount3:
        percpu_counter_destroy(&sbi->s_extent_cache_cnt);
        if (sbi->s_mmp_tsk)
                kthread_stop(sbi->s_mmp_tsk);
-failed_mount2a:
-       ext4_mb_release(sb);
 failed_mount2:
        for (i = 0; i < db_count; i++)
                brelse(sbi->s_group_desc[i]);
        ext4_kvfree(sbi->s_group_desc);
 failed_mount:
-       ext4_ext_release(sb);
        if (sbi->s_chksum_driver)
                crypto_free_shash(sbi->s_chksum_driver);
        if (sbi->s_proc) {