On 07/06/16 12:37, Wang Xiaoguang wrote: > Below test scripts can reproduce this false ENOSPC: > #!/bin/bash > dd if=/dev/zero of=fs.img bs=$((1024*1024)) count=128 > dev=$(losetup --show -f fs.img) > mkfs.btrfs -f -M $dev > mkdir /tmp/mntpoint > mount /dev/loop0 /tmp/mntpoint > cd mntpoint > xfs_io -f -c "falloc 0 $((40*1024*1024))" testfile > > Above fallocate(2) operation will fail for ENOSPC reason, but indeed > fs still has free space to satisfy this request. The reason is > btrfs_fallocate() dose not decrease btrfs_space_info's bytes_may_use > just in time, and it calls btrfs_free_reserved_data_space_noquota() in > the end of btrfs_fallocate(), which is too late and have already added > false unnecessary pressure to enospc system. See call graph: > btrfs_fallocate() > |-> btrfs_alloc_data_chunk_ondemand() > It will add btrfs_space_info's bytes_may_use accordingly. > |-> btrfs_prealloc_file_range() > It will call btrfs_reserve_extent(), but note that alloc type is > RESERVE_ALLOC_NO_ACCOUNT, so btrfs_update_reserved_bytes() will > only increase btrfs_space_info's bytes_reserved accordingly, but > will not decrease btrfs_space_info's bytes_may_use, then obviously > we have overestimated real needed disk space, and it'll impact > other processes who do write(2) or fallocate(2) operations, also > can impact metadata reservation in mixed mode, and bytes_max_use > will only be decreased in the end of btrfs_fallocate(). To fix > this false ENOSPC, we need to decrease btrfs_space_info's > bytes_may_use in btrfs_prealloc_file_range() in time, as what we > do in cow_file_range(), > See call graph in : > cow_file_range() > |-> extent_clear_unlock_delalloc() > |-> clear_extent_bit() > |-> btrfs_clear_bit_hook() > |-> btrfs_free_reserved_data_space_noquota() > This function will decrease bytes_may_use accordingly. > > So this patch choose to call btrfs_free_reserved_data_space() in > __btrfs_prealloc_file_range() for both successful and failed path. > > Also this patch removes some old and useless comments. > > Signed-off-by: Wang Xiaoguang <wangxg.f...@cn.fujitsu.com>
Verified that the reproducer script indeed fails (with btrfs ~4.7) and the patch (on top of 1/2) fixes it. Also ran a bunch of other fallocating things without problem. Free space also still seems sane, as far as I could tell. So for both patches: Tested-by: Holger Hoffstätte <hol...@applied-asynchrony.com> cheers, Holger -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
