mbox series

[v2,0/5] fuse: Further reducing contention of fc->lock

Message ID 154175935554.21228.11270624605198037428.stgit@localhost.localdomain (mailing list archive)
Headers show
Series fuse: Further reducing contention of fc->lock | expand

Message

Kirill Tkhai Nov. 9, 2018, 10:33 a.m. UTC
v2: Removed patch [1/6] from v1, and [4/6] from v1 is replaced
    with new patch (which is [2/5] in this v2).

There was introduced fc->bg_lock to reduce the contention,
and this patchset continues this efforts.

This patchset introduces per fuse_inode lock to protect
inode metadata, synchronization with background writes, etc.
All of the above is related to a single inode, and there
is no a reason, that inodes are concurrents of each other
to make some of actions exclusive.

So, here we introduce fuse_inode::lock spinlock and get rid
of fc->lock in many places.

---

Kirill Tkhai (5):
      fuse: Add fuse_inode argument to fuse_prepare_release()
      fuse: Convert fuse_conn::attr_version into atomic64_t
      fuse: Introduce fuse_inode::lock to protect write related fields and statistics
      fuse: Protect fuse_inode::nlookup with fuse_inode::lock
      fuse: Protect fuse_file::reserved_req via corresponding fuse_inode::lock


 fs/fuse/cuse.c    |    3 +
 fs/fuse/dev.c     |   10 +++--
 fs/fuse/dir.c     |   54 ++++++++++----------------
 fs/fuse/file.c    |  111 ++++++++++++++++++++++++++++-------------------------
 fs/fuse/fuse_i.h  |   23 ++++++++---
 fs/fuse/inode.c   |   17 +++++---
 fs/fuse/readdir.c |    4 +-
 7 files changed, 115 insertions(+), 107 deletions(-)

--
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>

Comments

Kirill Tkhai Dec. 10, 2018, 4:20 p.m. UTC | #1
ping

On 09.11.2018 13:33, Kirill Tkhai wrote:
> v2: Removed patch [1/6] from v1, and [4/6] from v1 is replaced
>     with new patch (which is [2/5] in this v2).
> 
> There was introduced fc->bg_lock to reduce the contention,
> and this patchset continues this efforts.
> 
> This patchset introduces per fuse_inode lock to protect
> inode metadata, synchronization with background writes, etc.
> All of the above is related to a single inode, and there
> is no a reason, that inodes are concurrents of each other
> to make some of actions exclusive.
> 
> So, here we introduce fuse_inode::lock spinlock and get rid
> of fc->lock in many places.
> 
> ---
> 
> Kirill Tkhai (5):
>       fuse: Add fuse_inode argument to fuse_prepare_release()
>       fuse: Convert fuse_conn::attr_version into atomic64_t
>       fuse: Introduce fuse_inode::lock to protect write related fields and statistics
>       fuse: Protect fuse_inode::nlookup with fuse_inode::lock
>       fuse: Protect fuse_file::reserved_req via corresponding fuse_inode::lock
> 
> 
>  fs/fuse/cuse.c    |    3 +
>  fs/fuse/dev.c     |   10 +++--
>  fs/fuse/dir.c     |   54 ++++++++++----------------
>  fs/fuse/file.c    |  111 ++++++++++++++++++++++++++++-------------------------
>  fs/fuse/fuse_i.h  |   23 ++++++++---
>  fs/fuse/inode.c   |   17 +++++---
>  fs/fuse/readdir.c |    4 +-
>  7 files changed, 115 insertions(+), 107 deletions(-)
> 
> --
> Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
>
Miklos Szeredi Jan. 10, 2019, 8:38 a.m. UTC | #2
On Mon, Dec 10, 2018 at 5:20 PM Kirill Tkhai <ktkhai@virtuozzo.com> wrote:
>
> ping
>
> On 09.11.2018 13:33, Kirill Tkhai wrote:
> > v2: Removed patch [1/6] from v1, and [4/6] from v1 is replaced
> >     with new patch (which is [2/5] in this v2).
> >
> > There was introduced fc->bg_lock to reduce the contention,
> > and this patchset continues this efforts.
> >
> > This patchset introduces per fuse_inode lock to protect
> > inode metadata, synchronization with background writes, etc.
> > All of the above is related to a single inode, and there
> > is no a reason, that inodes are concurrents of each other
> > to make some of actions exclusive.
> >
> > So, here we introduce fuse_inode::lock spinlock and get rid
> > of fc->lock in many places.

Pushed to fuse.git#for-next.

Thanks,
Miklos