diff options
| -rw-r--r-- | net/vhost-vdpa.c | 162 |
1 files changed, 161 insertions, 1 deletions
diff --git a/net/vhost-vdpa.c b/net/vhost-vdpa.c index eee4b2a09e..7610589a60 100644 --- a/net/vhost-vdpa.c +++ b/net/vhost-vdpa.c @@ -922,6 +922,148 @@ static NetClientInfo net_vhost_vdpa_cvq_info = { .check_peer_type = vhost_vdpa_check_peer_type, }; +/* + * Forward the excessive VIRTIO_NET_CTRL_MAC_TABLE_SET CVQ command to + * vdpa device. + * + * Considering that QEMU cannot send the entire filter table to the + * vdpa device, it should send the VIRTIO_NET_CTRL_RX_PROMISC CVQ + * command to enable promiscuous mode to receive all packets, + * according to VirtIO standard, "Since there are no guarantees, + * it can use a hash filter or silently switch to allmulti or + * promiscuous mode if it is given too many addresses.". + * + * Since QEMU ignores MAC addresses beyond `MAC_TABLE_ENTRIES` and + * marks `n->mac_table.x_overflow` accordingly, it should have + * the same effect on the device model to receive + * (`MAC_TABLE_ENTRIES` + 1) or more non-multicast MAC addresses. + * The same applies to multicast MAC addresses. + * + * Therefore, QEMU can provide the device model with a fake + * VIRTIO_NET_CTRL_MAC_TABLE_SET command with (`MAC_TABLE_ENTRIES` + 1) + * non-multicast MAC addresses and (`MAC_TABLE_ENTRIES` + 1) multicast + * MAC addresses. This ensures that the device model marks + * `n->mac_table.uni_overflow` and `n->mac_table.multi_overflow`, + * allowing all packets to be received, which aligns with the + * state of the vdpa device. + */ +static int vhost_vdpa_net_excessive_mac_filter_cvq_add(VhostVDPAState *s, + VirtQueueElement *elem, + struct iovec *out) +{ + struct virtio_net_ctrl_mac mac_data, *mac_ptr; + struct virtio_net_ctrl_hdr *hdr_ptr; + uint32_t cursor; + ssize_t r; + + /* parse the non-multicast MAC address entries from CVQ command */ + cursor = sizeof(*hdr_ptr); + r = iov_to_buf(elem->out_sg, elem->out_num, cursor, + &mac_data, sizeof(mac_data)); + if (unlikely(r != sizeof(mac_data))) { + /* + * If the CVQ command is invalid, we should simulate the vdpa device + * to reject the VIRTIO_NET_CTRL_MAC_TABLE_SET CVQ command + */ + *s->status = VIRTIO_NET_ERR; + return sizeof(*s->status); + } + cursor += sizeof(mac_data) + le32_to_cpu(mac_data.entries) * ETH_ALEN; + + /* parse the multicast MAC address entries from CVQ command */ + r = iov_to_buf(elem->out_sg, elem->out_num, cursor, + &mac_data, sizeof(mac_data)); + if (r != sizeof(mac_data)) { + /* + * If the CVQ command is invalid, we should simulate the vdpa device + * to reject the VIRTIO_NET_CTRL_MAC_TABLE_SET CVQ command + */ + *s->status = VIRTIO_NET_ERR; + return sizeof(*s->status); + } + cursor += sizeof(mac_data) + le32_to_cpu(mac_data.entries) * ETH_ALEN; + + /* validate the CVQ command */ + if (iov_size(elem->out_sg, elem->out_num) != cursor) { + /* + * If the CVQ command is invalid, we should simulate the vdpa device + * to reject the VIRTIO_NET_CTRL_MAC_TABLE_SET CVQ command + */ + *s->status = VIRTIO_NET_ERR; + return sizeof(*s->status); + } + + /* + * According to VirtIO standard, "Since there are no guarantees, + * it can use a hash filter or silently switch to allmulti or + * promiscuous mode if it is given too many addresses.". + * + * Therefore, considering that QEMU is unable to send the entire + * filter table to the vdpa device, it should send the + * VIRTIO_NET_CTRL_RX_PROMISC CVQ command to enable promiscuous mode + */ + r = vhost_vdpa_net_load_rx_mode(s, VIRTIO_NET_CTRL_RX_PROMISC, 1); + if (unlikely(r < 0)) { + return r; + } + if (*s->status != VIRTIO_NET_OK) { + return sizeof(*s->status); + } + + /* + * QEMU should also send a fake VIRTIO_NET_CTRL_MAC_TABLE_SET CVQ + * command to the device model, including (`MAC_TABLE_ENTRIES` + 1) + * non-multicast MAC addresses and (`MAC_TABLE_ENTRIES` + 1) + * multicast MAC addresses. + * + * By doing so, the device model can mark `n->mac_table.uni_overflow` + * and `n->mac_table.multi_overflow`, enabling all packets to be + * received, which aligns with the state of the vdpa device. + */ + cursor = 0; + uint32_t fake_uni_entries = MAC_TABLE_ENTRIES + 1, + fake_mul_entries = MAC_TABLE_ENTRIES + 1, + fake_cvq_size = sizeof(struct virtio_net_ctrl_hdr) + + sizeof(mac_data) + fake_uni_entries * ETH_ALEN + + sizeof(mac_data) + fake_mul_entries * ETH_ALEN; + + assert(fake_cvq_size < vhost_vdpa_net_cvq_cmd_page_len()); + out->iov_len = fake_cvq_size; + + /* pack the header for fake CVQ command */ + hdr_ptr = out->iov_base + cursor; + hdr_ptr->class = VIRTIO_NET_CTRL_MAC; + hdr_ptr->cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET; + cursor += sizeof(*hdr_ptr); + + /* + * Pack the non-multicast MAC addresses part for fake CVQ command. + * + * According to virtio_net_handle_mac(), QEMU doesn't verify the MAC + * addresses provieded in CVQ command. Therefore, only the entries + * field need to be prepared in the CVQ command. + */ + mac_ptr = out->iov_base + cursor; + mac_ptr->entries = cpu_to_le32(fake_uni_entries); + cursor += sizeof(*mac_ptr) + fake_uni_entries * ETH_ALEN; + + /* + * Pack the multicast MAC addresses part for fake CVQ command. + * + * According to virtio_net_handle_mac(), QEMU doesn't verify the MAC + * addresses provieded in CVQ command. Therefore, only the entries + * field need to be prepared in the CVQ command. + */ + mac_ptr = out->iov_base + cursor; + mac_ptr->entries = cpu_to_le32(fake_mul_entries); + + /* + * Simulating QEMU poll a vdpa device used buffer + * for VIRTIO_NET_CTRL_MAC_TABLE_SET CVQ command + */ + return sizeof(*s->status); +} + /** * Validate and copy control virtqueue commands. * @@ -949,7 +1091,7 @@ static int vhost_vdpa_net_handle_ctrl_avail(VhostShadowVirtqueue *svq, out.iov_len = iov_to_buf(elem->out_sg, elem->out_num, 0, s->cvq_cmd_out_buffer, - vhost_vdpa_net_cvq_cmd_len()); + vhost_vdpa_net_cvq_cmd_page_len()); ctrl = s->cvq_cmd_out_buffer; if (ctrl->class == VIRTIO_NET_CTRL_ANNOUNCE) { @@ -959,6 +1101,24 @@ static int vhost_vdpa_net_handle_ctrl_avail(VhostShadowVirtqueue *svq, */ dev_written = sizeof(status); *s->status = VIRTIO_NET_OK; + } else if (unlikely(ctrl->class == VIRTIO_NET_CTRL_MAC && + ctrl->cmd == VIRTIO_NET_CTRL_MAC_TABLE_SET && + iov_size(elem->out_sg, elem->out_num) > out.iov_len)) { + /* + * Due to the size limitation of the out buffer sent to the vdpa device, + * which is determined by vhost_vdpa_net_cvq_cmd_page_len(), excessive + * MAC addresses set by the driver for the filter table can cause + * truncation of the CVQ command in QEMU. As a result, the vdpa device + * rejects the flawed CVQ command. + * + * Therefore, QEMU must handle this situation instead of sending + * the CVQ command direclty. + */ + dev_written = vhost_vdpa_net_excessive_mac_filter_cvq_add(s, elem, + &out); + if (unlikely(dev_written < 0)) { + goto out; + } } else { dev_written = vhost_vdpa_net_cvq_add(s, out.iov_len, sizeof(status)); if (unlikely(dev_written < 0)) { |