[Virtio-fs] [PATCH 07/24] DAX: virtio-fs: Add vhost-user slave commands for mapping

Chirantan Ekbote chirantan at chromium.org
Mon Feb 15 10:35:53 UTC 2021


On Wed, Feb 10, 2021 at 4:04 AM Dr. David Alan Gilbert (git)
<dgilbert at redhat.com> wrote:
>
> From: "Dr. David Alan Gilbert" <dgilbert at redhat.com>
> +
> +typedef struct {
> +    /* Offsets within the file being mapped */
> +    uint64_t fd_offset[VHOST_USER_FS_SLAVE_ENTRIES];
> +    /* Offsets within the cache */
> +    uint64_t c_offset[VHOST_USER_FS_SLAVE_ENTRIES];
> +    /* Lengths of sections */
> +    uint64_t len[VHOST_USER_FS_SLAVE_ENTRIES];
> +    /* Flags, from VHOST_USER_FS_FLAG_* */
> +    uint64_t flags[VHOST_USER_FS_SLAVE_ENTRIES];
> +} VhostUserFSSlaveMsg;
> +

Is it too late to change this?  This struct allocates space for up to
8 entries but most of the time the server will only try to set up one
mapping at a time so only 32 out of the 256 bytes in the message are
actually being used.  We're just wasting time memcpy'ing bytes that
will never be used.  Is there some reason this can't be dynamically
sized?  Something like:

typedef struct {
    /* Number of mapping requests */
    uint16_t num_requests;
    /* `num_requests` mapping requests */
   MappingRequest requests[];
} VhostUserFSSlaveMsg;

typedef struct {
    /* Offset within the file being mapped */
    uint64_t fd_offset;
    /* Offset within the cache */
    uint64_t c_offset;
    /* Length of section */
    uint64_t len;
    /* Flags, from VHOST_USER_FS_FLAG_* */
    uint64_t flags;
} MappingRequest;

The current pre-allocated structure both wastes space when there are
fewer than 8 requests and requires extra messages to be sent when
there are more than 8 requests.  I realize that in the grand scheme of
things copying 224 extra bytes is basically not noticeable but it just
irks me that we could fix this really easily before it gets propagated
to too many other places.

Chirantan

> --
> 2.29.2
>
> _______________________________________________
> Virtio-fs mailing list
> Virtio-fs at redhat.com
> https://www.redhat.com/mailman/listinfo/virtio-fs
>




More information about the Virtio-fs mailing list