On Fri, Sep 18, 2020 at 7:09 AM Ritesh Harjani <riteshh@xxxxxxxxxxxxx> wrote:
>
> In case if the file already has underlying blocks/extents allocated
> then we don't need to start a journal txn and can directly return
> the underlying mapping. Currently ext4_iomap_begin() is used by
> both DAX & DIO path. We can check if the write request is an
> overwrite & then directly return the mapping information.
>
> This could give a significant perf boost for multi-threaded writes
> specially random overwrites.
> On PPC64 VM with simulated pmem(DAX) device, ~10x perf improvement
> could be seen in random writes (overwrite). Also bcoz this optimizes
> away the spinlock contention during jbd2 slab cache allocation
> (jbd2_journal_handle). On x86 VM, ~2x perf improvement was observed.
>
> Reported-by: Dan Williams <dan.j.williams@xxxxxxxxx>
> Suggested-by: Jan Kara <jack@xxxxxxx>
> Signed-off-by: Ritesh Harjani <riteshh@xxxxxxxxxxxxx>
I have applied your patch on top of recent Linus Git and boot-tested on x86-64.
Here I have LTP installed.
If you have a LTP filesystem test-/use-case you know for testing,
please let me know.
Yes, I have seen the FIO config in the cover-letter.
Maybe you have a different FIO config - 16G AFAIK is too big here.
Feel free to add...
Tested-by: Sedat Dilek <sedat.dilek@xxxxxxxxx> # Compile and boot on
x86-64 Debian/unstable
Thanks.
- Sedat -
> ---
> fs/ext4/inode.c | 18 +++++++++++++++---
> 1 file changed, 15 insertions(+), 3 deletions(-)
>
> diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
> index 10dd470876b3..6eae17758ece 100644
> --- a/fs/ext4/inode.c
> +++ b/fs/ext4/inode.c
> @@ -3437,14 +3437,26 @@ static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
> map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
> EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
>
> - if (flags & IOMAP_WRITE)
> + if (flags & IOMAP_WRITE) {
> + /*
> + * We check here if the blocks are already allocated, then we
> + * don't need to start a journal txn and we can directly return
> + * the mapping information. This could boost performance
> + * especially in multi-threaded overwrite requests.
> + */
> + if (offset + length <= i_size_read(inode)) {
> + ret = ext4_map_blocks(NULL, inode, &map, 0);
> + if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
> + goto out;
> + }
> ret = ext4_iomap_alloc(inode, &map, flags);
> - else
> + } else {
> ret = ext4_map_blocks(NULL, inode, &map, 0);
> + }
>
> if (ret < 0)
> return ret;
> -
> +out:
> ext4_set_iomap(inode, iomap, &map, offset, length);
>
> return 0;
> --
> 2.26.2
>
