Re: [for-6.1 v2 2/2] virtiofsd: Add support for FUSE_SYNCFS request

[Vivek Goyal <vgoyal@redhat.com>]

On Mon, Apr 26, 2021 at 05:21:35PM +0200, Greg Kurz wrote:
> Honor the expected behavior of syncfs() to synchronously flush all
> data and metadata on linux systems.
> 
> Flushing is done with syncfs(). This is suboptimal as it will also
> flush writes performed by any other process on the same file system,
> and thus add an unbounded time penalty to syncfs(). This may be
> optimized in the future, but enforce correctness first.
> 
> Signed-off-by: Greg Kurz <groug@kaod.org>
> ---
>  tools/virtiofsd/fuse_lowlevel.c       | 19 ++++++++++++++++++
>  tools/virtiofsd/fuse_lowlevel.h       | 13 ++++++++++++
>  tools/virtiofsd/passthrough_ll.c      | 29 +++++++++++++++++++++++++++
>  tools/virtiofsd/passthrough_seccomp.c |  1 +
>  4 files changed, 62 insertions(+)
> 
> diff --git a/tools/virtiofsd/fuse_lowlevel.c b/tools/virtiofsd/fuse_lowlevel.c
> index 58e32fc96369..918ab11f54c2 100644
> --- a/tools/virtiofsd/fuse_lowlevel.c
> +++ b/tools/virtiofsd/fuse_lowlevel.c
> @@ -1870,6 +1870,24 @@ static void do_lseek(fuse_req_t req, fuse_ino_t nodeid,
>      }
>  }
>  
> +static void do_syncfs(fuse_req_t req, fuse_ino_t nodeid,
> +                      struct fuse_mbuf_iter *iter)
> +{
> +    struct fuse_syncfs_in *arg;
> +
> +    arg = fuse_mbuf_iter_advance(iter, sizeof(*arg));
> +    if (!arg) {
> +        fuse_reply_err(req, EINVAL);
> +        return;
> +    }
> +
> +    if (req->se->op.syncfs) {
> +        req->se->op.syncfs(req, arg->flags);
> +    } else {
> +        fuse_reply_err(req, ENOSYS);
> +    }
> +}
> +
>  static void do_init(fuse_req_t req, fuse_ino_t nodeid,
>                      struct fuse_mbuf_iter *iter)
>  {
> @@ -2267,6 +2285,7 @@ static struct {
>      [FUSE_RENAME2] = { do_rename2, "RENAME2" },
>      [FUSE_COPY_FILE_RANGE] = { do_copy_file_range, "COPY_FILE_RANGE" },
>      [FUSE_LSEEK] = { do_lseek, "LSEEK" },
> +    [FUSE_SYNCFS] = { do_syncfs, "SYNCFS" },
>  };
>  
>  #define FUSE_MAXOP (sizeof(fuse_ll_ops) / sizeof(fuse_ll_ops[0]))
> diff --git a/tools/virtiofsd/fuse_lowlevel.h b/tools/virtiofsd/fuse_lowlevel.h
> index 3bf786b03485..220bb3db4898 100644
> --- a/tools/virtiofsd/fuse_lowlevel.h
> +++ b/tools/virtiofsd/fuse_lowlevel.h
> @@ -1225,6 +1225,19 @@ struct fuse_lowlevel_ops {
>       */
>      void (*lseek)(fuse_req_t req, fuse_ino_t ino, off_t off, int whence,
>                    struct fuse_file_info *fi);
> +
> +    /**
> +     * Synchronize file system content
> +     *
> +     * If this request is answered with an error code of ENOSYS,
> +     * this is treated as success and future calls to syncfs() will
> +     * succeed automatically without being sent to the filesystem
> +     * process.
> +     *
> +     * @param req request handle
> +     * @param flags not used yet
> +     */
> +    void (*syncfs)(fuse_req_t req, uint64_t flags);
>  };
>  
>  /**
> diff --git a/tools/virtiofsd/passthrough_ll.c b/tools/virtiofsd/passthrough_ll.c
> index 1553d2ef454f..6790a2f6fe10 100644
> --- a/tools/virtiofsd/passthrough_ll.c
> +++ b/tools/virtiofsd/passthrough_ll.c
> @@ -3124,6 +3124,34 @@ static void lo_lseek(fuse_req_t req, fuse_ino_t ino, off_t off, int whence,
>      }
>  }
>  
> +static void lo_syncfs(fuse_req_t req, uint64_t flags)
> +{
> +    struct lo_data *lo = lo_data(req);
> +    int fd, ret;
> +
> +    /* No flags supported yet */
> +    if (flags) {
> +        fuse_reply_err(req, EINVAL);
> +        return;
> +    }
> +
> +    fd = lo_inode_open(lo, &lo->root, O_RDONLY);
> +    if (fd < 0) {
> +        fuse_reply_err(req, errno);
> +        return;
> +    }
> +
> +    /*
> +     * FIXME: this is suboptimal because it will also flush unrelated
> +     *        writes not coming from the client. This can dramatically
> +     *        increase the time spent in syncfs() if some process is
> +     *        writing lots of data on the same filesystem as virtiofsd.
> +     */
> +    ret = syncfs(fd);

Hi Greg,

As we discussed in the community call that this works only if there are
no other filesystems mounted as submounts under exported directory.

We proably need to find a way to call syncfs() on all the filesystems
which are submounts of exported directory. Might not be easy at all.

Just mentioning it here so that we have a note about the limitation of
current patch.

Vivek

> +    fuse_reply_err(req, ret < 0 ? errno : 0);
> +    close(fd);
> +}
> +
>  static void lo_destroy(void *userdata)
>  {
>      struct lo_data *lo = (struct lo_data *)userdata;
> @@ -3184,6 +3212,7 @@ static struct fuse_lowlevel_ops lo_oper = {
>      .copy_file_range = lo_copy_file_range,
>  #endif
>      .lseek = lo_lseek,
> +    .syncfs = lo_syncfs,
>      .destroy = lo_destroy,
>  };
>  
> diff --git a/tools/virtiofsd/passthrough_seccomp.c b/tools/virtiofsd/passthrough_seccomp.c
> index 62441cfcdb95..343188447901 100644
> --- a/tools/virtiofsd/passthrough_seccomp.c
> +++ b/tools/virtiofsd/passthrough_seccomp.c
> @@ -107,6 +107,7 @@ static const int syscall_allowlist[] = {
>      SCMP_SYS(set_robust_list),
>      SCMP_SYS(setxattr),
>      SCMP_SYS(symlinkat),
> +    SCMP_SYS(syncfs),
>      SCMP_SYS(time), /* Rarely needed, except on static builds */
>      SCMP_SYS(tgkill),
>      SCMP_SYS(unlinkat),
> -- 
> 2.26.3
>