On Mon, 2020-02-03 at 21:28 -0500, xiubli@xxxxxxxxxx wrote:
> From: Xiubo Li <xiubli@xxxxxxxxxx>
>
> In O_APPEND & O_DIRECT mode, the data from different writers will
> be possiblly overlapping each other with shared lock.
>
> For example, both Writer1 and Writer2 are in O_APPEND and O_DIRECT
> mode:
>
> Writer1 Writer2
>
> shared_lock() shared_lock()
> getattr(CAP_SIZE) getattr(CAP_SIZE)
> iocb->ki_pos = EOF iocb->ki_pos = EOF
> write(data1)
> write(data2)
> shared_unlock() shared_unlock()
>
> The data2 will overlap the data1 from the same file offset, the
> old EOF.
>
> Switch to exclusive lock instead when O_APPEND is specified.
>
> Signed-off-by: Xiubo Li <xiubli@xxxxxxxxxx>
> ---
>
> Changed in V2:
> - fix the commit comment
> - add more detail in the commit comment
> - s/direct_lock/shared_lock/g
>
> fs/ceph/file.c | 17 +++++++++++------
> 1 file changed, 11 insertions(+), 6 deletions(-)
>
> diff --git a/fs/ceph/file.c b/fs/ceph/file.c
> index ac7fe8b8081c..e3e67ef215dd 100644
> --- a/fs/ceph/file.c
> +++ b/fs/ceph/file.c
> @@ -1475,6 +1475,7 @@ static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
> struct ceph_cap_flush *prealloc_cf;
> ssize_t count, written = 0;
> int err, want, got;
> + bool shared_lock = false;
> loff_t pos;
> loff_t limit = max(i_size_read(inode), fsc->max_file_size);
>
> @@ -1485,8 +1486,11 @@ static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
> if (!prealloc_cf)
> return -ENOMEM;
>
> + if ((iocb->ki_flags & (IOCB_DIRECT | IOCB_APPEND)) == IOCB_DIRECT)
> + shared_lock = true;
> +
> retry_snap:
> - if (iocb->ki_flags & IOCB_DIRECT)
> + if (shared_lock)
> ceph_start_io_direct(inode);
> else
> ceph_start_io_write(inode);
> @@ -1576,14 +1580,15 @@ static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
>
> /* we might need to revert back to that point */
> data = *from;
> - if (iocb->ki_flags & IOCB_DIRECT) {
> + if (iocb->ki_flags & IOCB_DIRECT)
> written = ceph_direct_read_write(iocb, &data, snapc,
> &prealloc_cf);
> - ceph_end_io_direct(inode);
> - } else {
> + else
> written = ceph_sync_write(iocb, &data, pos, snapc);
> + if (shared_lock)
> + ceph_end_io_direct(inode);
> + else
> ceph_end_io_write(inode);
> - }
> if (written > 0)
> iov_iter_advance(from, written);
> ceph_put_snap_context(snapc);
> @@ -1634,7 +1639,7 @@ static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
>
> goto out_unlocked;
> out:
> - if (iocb->ki_flags & IOCB_DIRECT)
> + if (shared_lock)
> ceph_end_io_direct(inode);
> else
> ceph_end_io_write(inode);
Ok, I think this looks reasonable, but I actually preferred the
"direct_lock" name you had before. I'm going to do some testing today
and will probably merge this (with s/shared_lock/direct_lock/) if it
tests out ok.
--
Jeff Layton <jlayton@xxxxxxxxxx>
