btrfs: free exchange changeset on failures
authorJohannes Thumshirn <johannes.thumshirn@wdc.com>
Fri, 3 Dec 2021 10:55:33 +0000 (02:55 -0800)
committerDavid Sterba <dsterba@suse.com>
Wed, 8 Dec 2021 14:42:32 +0000 (15:42 +0100)
Fstests runs on my VMs have show several kmemleak reports like the following.

  unreferenced object 0xffff88811ae59080 (size 64):
    comm "xfs_io", pid 12124, jiffies 4294987392 (age 6.368s)
    hex dump (first 32 bytes):
      00 c0 1c 00 00 00 00 00 ff cf 1c 00 00 00 00 00  ................
      90 97 e5 1a 81 88 ff ff 90 97 e5 1a 81 88 ff ff  ................
    backtrace:
      [<00000000ac0176d2>] ulist_add_merge+0x60/0x150 [btrfs]
      [<0000000076e9f312>] set_state_bits+0x86/0xc0 [btrfs]
      [<0000000014fe73d6>] set_extent_bit+0x270/0x690 [btrfs]
      [<000000004f675208>] set_record_extent_bits+0x19/0x20 [btrfs]
      [<00000000b96137b1>] qgroup_reserve_data+0x274/0x310 [btrfs]
      [<0000000057e9dcbb>] btrfs_check_data_free_space+0x5c/0xa0 [btrfs]
      [<0000000019c4511d>] btrfs_delalloc_reserve_space+0x1b/0xa0 [btrfs]
      [<000000006d37e007>] btrfs_dio_iomap_begin+0x415/0x970 [btrfs]
      [<00000000fb8a74b8>] iomap_iter+0x161/0x1e0
      [<0000000071dff6ff>] __iomap_dio_rw+0x1df/0x700
      [<000000002567ba53>] iomap_dio_rw+0x5/0x20
      [<0000000072e555f8>] btrfs_file_write_iter+0x290/0x530 [btrfs]
      [<000000005eb3d845>] new_sync_write+0x106/0x180
      [<000000003fb505bf>] vfs_write+0x24d/0x2f0
      [<000000009bb57d37>] __x64_sys_pwrite64+0x69/0xa0
      [<000000003eba3fdf>] do_syscall_64+0x43/0x90

In case brtfs_qgroup_reserve_data() or btrfs_delalloc_reserve_metadata()
fail the allocated extent_changeset will not be freed.

So in btrfs_check_data_free_space() and btrfs_delalloc_reserve_space()
free the allocated extent_changeset to get rid of the allocated memory.

The issue currently only happens in the direct IO write path, but only
after 65b3c08606e5 ("btrfs: fix ENOSPC failure when attempting direct IO
write into NOCOW range"), and also at defrag_one_locked_target(). Every
other place is always calling extent_changeset_free() even if its call
to btrfs_delalloc_reserve_space() or btrfs_check_data_free_space() has
failed.

CC: stable@vger.kernel.org # 5.15+
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
fs/btrfs/delalloc-space.c

index 2059d1504149a6326cfc2fc2bc34d9b9f1d8eb67..40c4d6ba3fb9a79d639f5b8d8de2cb018c8904be 100644 (file)
@@ -143,10 +143,13 @@ int btrfs_check_data_free_space(struct btrfs_inode *inode,
 
        /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
        ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
-       if (ret < 0)
+       if (ret < 0) {
                btrfs_free_reserved_data_space_noquota(fs_info, len);
-       else
+               extent_changeset_free(*reserved);
+               *reserved = NULL;
+       } else {
                ret = 0;
+       }
        return ret;
 }
 
@@ -452,8 +455,11 @@ int btrfs_delalloc_reserve_space(struct btrfs_inode *inode,
        if (ret < 0)
                return ret;
        ret = btrfs_delalloc_reserve_metadata(inode, len);
-       if (ret < 0)
+       if (ret < 0) {
                btrfs_free_reserved_data_space(inode, *reserved, start, len);
+               extent_changeset_free(*reserved);
+               *reserved = NULL;
+       }
        return ret;
 }