RE: [[PATCH v1] 21/37] [CIFS] SMBD: Implement API for upper layer to receive data

[Tom Talpey]

> -----Original Message-----
> From: linux-cifs-owner@xxxxxxxxxxxxxxx [mailto:linux-cifs-
> owner@xxxxxxxxxxxxxxx] On Behalf Of Long Li
> Sent: Wednesday, August 2, 2017 4:11 PM
> To: Steve French <sfrench@xxxxxxxxx>; linux-cifs@xxxxxxxxxxxxxxx; samba-
> technical@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx
> Cc: Long Li <longli@xxxxxxxxxxxxx>
> Subject: [[PATCH v1] 21/37] [CIFS] SMBD: Implement API for upper layer to
> receive data
> 
>  /*
> + * Read data from receive reassembly queue
> + * All the incoming data packets are placed in reassembly queue
> + * buf: the buffer to read data into
> + * size: the length of data to read
> + * return value: actual data read
> + */
> +int cifs_rdma_read(struct cifs_rdma_info *info, char *buf, unsigned int size)
> +{
>...
> +       spin_lock_irqsave(&info->reassembly_queue_lock, flags);
> +       log_cifs_read("size=%d info->reassembly_data_length=%d\n", size,
> +               atomic_read(&info->reassembly_data_length));
> +       if (atomic_read(&info->reassembly_data_length) >= size) {

If the reassembly queue is protected by a lock, why is an atomic_read() of
its length needed?

> +                       // this is for reading rfc1002 length
> +                       if (response->first_segment && size==4) {
> +                               unsigned int rfc1002_len =
> +                                       data_length + remaining_data_length;
> +                               *((__be32*)buf) = cpu_to_be32(rfc1002_len);
> +                               data_read = 4;
> +                               response->first_segment = false;
> +                               log_cifs_read("returning rfc1002 length %d\n",
> +                                       rfc1002_len);
> +                               goto read_rfc1002_done;
> +                       }

I am totally confused. What does RFC1002 framing have to do with
receiving an SMB Direct packet???

> +
> +                       to_copy = min_t(int, data_length - offset, to_read);
> +                       memcpy(
> +                               buf + data_read,
> +                               (char*)data_transfer + data_offset + offset,
> +                               to_copy);

Is it really necessary to perform all these data copies, especially under the
reassembly_queue spinlock? This seems quite inefficient. Can the receive
buffers not be loaned out and chained logically?

Tom.