diff options
Diffstat (limited to 'src/plugins/af_packet')
| -rw-r--r-- | src/plugins/af_packet/CMakeLists.txt | 38 | ||||
| -rw-r--r-- | src/plugins/af_packet/FEATURE.yaml | 16 | ||||
| -rw-r--r-- | src/plugins/af_packet/af_packet.api | 200 | ||||
| -rw-r--r-- | src/plugins/af_packet/af_packet.c | 1054 | ||||
| -rw-r--r-- | src/plugins/af_packet/af_packet.h | 182 | ||||
| -rw-r--r-- | src/plugins/af_packet/af_packet_api.c | 253 | ||||
| -rw-r--r-- | src/plugins/af_packet/cli.c | 341 | ||||
| -rw-r--r-- | src/plugins/af_packet/device.c | 793 | ||||
| -rw-r--r-- | src/plugins/af_packet/dir.dox | 29 | ||||
| -rw-r--r-- | src/plugins/af_packet/node.c | 832 | ||||
| -rw-r--r-- | src/plugins/af_packet/plugin.c | 12 |
11 files changed, 3750 insertions, 0 deletions
diff --git a/src/plugins/af_packet/CMakeLists.txt b/src/plugins/af_packet/CMakeLists.txt new file mode 100644 index 00000000000..818a03107d7 --- /dev/null +++ b/src/plugins/af_packet/CMakeLists.txt @@ -0,0 +1,38 @@ +# SPDX-License-Identifier: Apache-2.0 +# Copyright (c) 2022 Cisco and/or its affiliates. +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at: +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +add_vpp_plugin(af_packet + SOURCES + plugin.c + af_packet.c + device.c + node.c + cli.c + af_packet_api.c + + MULTIARCH_SOURCES + node.c + device.c + + INSTALL_HEADERS + af_packet.h + + API_FILES + af_packet.api + + # API_TEST_SOURCES + #af_packet_test_api.c + + SUPPORTED_OS_LIST Linux +) diff --git a/src/plugins/af_packet/FEATURE.yaml b/src/plugins/af_packet/FEATURE.yaml new file mode 100644 index 00000000000..4a11ea2beb5 --- /dev/null +++ b/src/plugins/af_packet/FEATURE.yaml @@ -0,0 +1,16 @@ +--- +name: host-interface Device AF_PACKET +maintainer: Damjan Marion <damarion@cisco.com> +features: + - L4 checksum offload + - GSO offload +description: "Create a host interface that will attach to a linux AF_PACKET + interface, one side of a veth pair. The veth pair must + already exist. Once created, a new host interface will + exist in VPP with the name 'host-<ifname>', where '<ifname>' + is the name of the specified veth pair. Use the 'show interface' + command to display host interface details." +missing: + - API dump details beyond sw_if_index and name +state: production +properties: [API, CLI, STATS, MULTITHREAD] diff --git a/src/plugins/af_packet/af_packet.api b/src/plugins/af_packet/af_packet.api new file mode 100644 index 00000000000..abc7d1a7206 --- /dev/null +++ b/src/plugins/af_packet/af_packet.api @@ -0,0 +1,200 @@ +/* + * Copyright (c) 2015-2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +option version = "2.0.0"; + +import "vnet/interface_types.api"; +import "vnet/ethernet/ethernet_types.api"; + +/** \brief Create host-interface + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request + @param hw_addr - interface MAC + @param use_random_hw_addr - use random generated MAC + @param host_if_name - interface name +*/ +define af_packet_create +{ + option deprecated; + + u32 client_index; + u32 context; + + vl_api_mac_address_t hw_addr; + bool use_random_hw_addr; + string host_if_name[64]; +}; + +/** \brief Create host-interface response + @param context - sender context, to match reply w/ request + @param retval - return value for request +*/ +define af_packet_create_reply +{ + option deprecated; + + u32 context; + i32 retval; + vl_api_interface_index_t sw_if_index; +}; + +/** \brief Create host-interface + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request + @param hw_addr - interface MAC + @param use_random_hw_addr - use random generated MAC + @param host_if_name - interface name + @param rx_frame_size - frame size for RX + @param tx_frame_size - frame size for TX + @param rx_frames_per_block - frames per block for RX + @param tx_frames_per_block - frames per block for TX + @param flags - flags for the af_packet interface creation + @param num_rx_queues - number of rx queues +*/ +define af_packet_create_v2 +{ + option deprecated; + + u32 client_index; + u32 context; + + vl_api_mac_address_t hw_addr; + bool use_random_hw_addr; + string host_if_name[64]; + u32 rx_frame_size; + u32 tx_frame_size; + u32 rx_frames_per_block; + u32 tx_frames_per_block; + u32 flags; + u16 num_rx_queues [default=1]; +}; + +/** \brief Create host-interface response + @param context - sender context, to match reply w/ request + @param retval - return value for request +*/ +define af_packet_create_v2_reply +{ + option deprecated; + + u32 context; + i32 retval; + vl_api_interface_index_t sw_if_index; +}; + +enum af_packet_mode { + AF_PACKET_API_MODE_ETHERNET = 1, /* mode ethernet */ + AF_PACKET_API_MODE_IP = 2, /* mode ip */ +}; + +enum af_packet_flags { + AF_PACKET_API_FLAG_QDISC_BYPASS = 1, /* enable the qdisc bypass */ + AF_PACKET_API_FLAG_CKSUM_GSO = 2, /* enable checksum/gso */ + AF_PACKET_API_FLAG_VERSION_2 = 8 [backwards_compatible], /* af packet v2, default is v3 */ +}; + +/** \brief Create host-interface + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request + @param mode - 1 - Ethernet, 2 - IP + @param hw_addr - interface MAC + @param use_random_hw_addr - use random generated MAC + @param host_if_name - interface name + @param rx_frame_size - frame size for RX + @param tx_frame_size - frame size for TX + @param rx_frames_per_block - frames per block for RX + @param tx_frames_per_block - frames per block for TX + @param flags - flags for the af_packet interface creation + @param num_rx_queues - number of rx queues + @param num_tx_queues - number of tx queues +*/ +define af_packet_create_v3 +{ + u32 client_index; + u32 context; + + vl_api_af_packet_mode_t mode; + vl_api_mac_address_t hw_addr; + bool use_random_hw_addr; + string host_if_name[64]; + u32 rx_frame_size; + u32 tx_frame_size; + u32 rx_frames_per_block; + u32 tx_frames_per_block; + vl_api_af_packet_flags_t flags; + u16 num_rx_queues [default=1]; + u16 num_tx_queues [default=1]; +}; + +/** \brief Create host-interface response + @param context - sender context, to match reply w/ request + @param retval - return value for request +*/ +define af_packet_create_v3_reply +{ + u32 context; + i32 retval; + vl_api_interface_index_t sw_if_index; +}; + +/** \brief Delete host-interface + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request + @param host_if_name - interface name +*/ +autoreply define af_packet_delete +{ + u32 client_index; + u32 context; + + string host_if_name[64]; +}; + +/** \brief Set l4 offload checksum calculation + @param client_index - opaque cookie to identify the sender + @param context - sender context, to match reply w/ request +*/ +autoreply define af_packet_set_l4_cksum_offload +{ + u32 client_index; + u32 context; + + vl_api_interface_index_t sw_if_index; + bool set; +}; + +/** \brief Dump af_packet interfaces request */ +define af_packet_dump +{ + u32 client_index; + u32 context; +}; + +/** \brief Reply for af_packet dump request + @param sw_if_index - software index of af_packet interface + @param host_if_name - interface name +*/ +define af_packet_details +{ + u32 context; + vl_api_interface_index_t sw_if_index; + string host_if_name[64]; +}; + +/* + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/plugins/af_packet/af_packet.c b/src/plugins/af_packet/af_packet.c new file mode 100644 index 00000000000..69245429918 --- /dev/null +++ b/src/plugins/af_packet/af_packet.c @@ -0,0 +1,1054 @@ +/* + *------------------------------------------------------------------ + * af_packet.c - linux kernel packet interface + * + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + *------------------------------------------------------------------ + */ + +#include <linux/if_ether.h> +#include <linux/if_packet.h> +#include <linux/ethtool.h> +#include <linux/sockios.h> +#include <sys/ioctl.h> +#include <net/if.h> +#include <dirent.h> +#include <sys/stat.h> +#include <sys/types.h> +#include <fcntl.h> + +#include <vppinfra/linux/sysfs.h> +#include <vlib/vlib.h> +#include <vlib/unix/unix.h> +#include <vnet/ip/ip.h> +#include <vnet/devices/netlink.h> +#include <vnet/ethernet/ethernet.h> +#include <vnet/interface/rx_queue_funcs.h> +#include <vnet/interface/tx_queue_funcs.h> + +#include <af_packet/af_packet.h> + +af_packet_main_t af_packet_main; + +VNET_HW_INTERFACE_CLASS (af_packet_ip_device_hw_interface_class, static) = { + .name = "af-packet-ip-device", + .flags = VNET_HW_INTERFACE_CLASS_FLAG_P2P, +}; + +#define AF_PACKET_DEFAULT_TX_FRAMES_PER_BLOCK 1024 +#define AF_PACKET_DEFAULT_TX_FRAME_SIZE (2048 * 33) // GSO packet of 64KB +#define AF_PACKET_TX_BLOCK_NR 1 + +#define AF_PACKET_DEFAULT_RX_FRAMES_PER_BLOCK_V2 1024 +#define AF_PACKET_DEFAULT_RX_FRAME_SIZE_V2 (2048 * 33) // GSO packet of 64KB +#define AF_PACKET_RX_BLOCK_NR_V2 1 + +#define AF_PACKET_DEFAULT_RX_FRAMES_PER_BLOCK 32 +#define AF_PACKET_DEFAULT_RX_FRAME_SIZE 2048 +#define AF_PACKET_RX_BLOCK_NR 160 + +/*defined in net/if.h but clashes with dpdk headers */ +unsigned int if_nametoindex (const char *ifname); + +#define AF_PACKET_OFFLOAD_FLAG_RXCKSUM (1 << 0) +#define AF_PACKET_OFFLOAD_FLAG_TXCKSUM (1 << 1) +#define AF_PACKET_OFFLOAD_FLAG_SG (1 << 2) +#define AF_PACKET_OFFLOAD_FLAG_TSO (1 << 3) +#define AF_PACKET_OFFLOAD_FLAG_UFO (1 << 4) +#define AF_PACKET_OFFLOAD_FLAG_GSO (1 << 5) +#define AF_PACKET_OFFLOAD_FLAG_GRO (1 << 6) + +#define AF_PACKET_OFFLOAD_FLAG_MASK \ + (AF_PACKET_OFFLOAD_FLAG_RXCKSUM | AF_PACKET_OFFLOAD_FLAG_TXCKSUM | \ + AF_PACKET_OFFLOAD_FLAG_SG | AF_PACKET_OFFLOAD_FLAG_TSO | \ + AF_PACKET_OFFLOAD_FLAG_UFO | AF_PACKET_OFFLOAD_FLAG_GSO | \ + AF_PACKET_OFFLOAD_FLAG_GRO) + +#define AF_PACKET_IOCTL(fd, a, ...) \ + if (ioctl (fd, a, __VA_ARGS__) < 0) \ + { \ + err = clib_error_return_unix (0, "ioctl(" #a ")"); \ + vlib_log_err (af_packet_main.log_class, "%U", format_clib_error, err); \ + goto done; \ + } + +static u32 +af_packet_get_if_capabilities (u8 *host_if_name) +{ + struct ifreq ifr; + struct ethtool_value e; // { __u32 cmd; __u32 data; }; + clib_error_t *err = 0; + int ctl_fd = -1; + u32 oflags = 0; + + if ((ctl_fd = socket (AF_INET, SOCK_STREAM, 0)) == -1) + { + clib_warning ("Cannot open control socket"); + goto done; + } + + clib_memset (&ifr, 0, sizeof (ifr)); + clib_memcpy (ifr.ifr_name, host_if_name, + strlen ((const char *) host_if_name)); + ifr.ifr_data = (void *) &e; + + e.cmd = ETHTOOL_GRXCSUM; + AF_PACKET_IOCTL (ctl_fd, SIOCETHTOOL, &ifr); + if (e.data) + oflags |= AF_PACKET_OFFLOAD_FLAG_RXCKSUM; + + e.cmd = ETHTOOL_GTXCSUM; + AF_PACKET_IOCTL (ctl_fd, SIOCETHTOOL, &ifr); + if (e.data) + oflags |= AF_PACKET_OFFLOAD_FLAG_TXCKSUM; + + e.cmd = ETHTOOL_GTSO; + AF_PACKET_IOCTL (ctl_fd, SIOCETHTOOL, &ifr); + if (e.data) + oflags |= AF_PACKET_OFFLOAD_FLAG_TSO; + + e.cmd = ETHTOOL_GGSO; + AF_PACKET_IOCTL (ctl_fd, SIOCETHTOOL, &ifr); + if (e.data) + oflags |= AF_PACKET_OFFLOAD_FLAG_GSO; + + e.cmd = ETHTOOL_GGRO; + AF_PACKET_IOCTL (ctl_fd, SIOCETHTOOL, &ifr); + if (e.data) + oflags |= AF_PACKET_OFFLOAD_FLAG_GRO; + +done: + if (ctl_fd != -1) + close (ctl_fd); + + return oflags; +} + +static clib_error_t * +af_packet_eth_set_max_frame_size (vnet_main_t *vnm, vnet_hw_interface_t *hi, + u32 frame_size) +{ + clib_error_t *error, *rv; + af_packet_main_t *apm = &af_packet_main; + af_packet_if_t *apif = pool_elt_at_index (apm->interfaces, hi->dev_instance); + + error = vnet_netlink_set_link_mtu (apif->host_if_index, + frame_size + hi->frame_overhead); + + if (error) + { + vlib_log_err (apm->log_class, "netlink failed to change MTU: %U", + format_clib_error, error); + rv = vnet_error (VNET_ERR_SYSCALL_ERROR_1, "netlink error: %U", + format_clib_error, error); + clib_error_free (error); + return rv; + } + else + apif->host_mtu = frame_size + hi->frame_overhead; + return 0; +} + +static int +af_packet_read_mtu (af_packet_if_t *apif) +{ + af_packet_main_t *apm = &af_packet_main; + clib_error_t *error; + error = vnet_netlink_get_link_mtu (apif->host_if_index, &apif->host_mtu); + if (error) + { + vlib_log_err (apm->log_class, "netlink failed to get MTU: %U", + format_clib_error, error); + clib_error_free (error); + return VNET_API_ERROR_SYSCALL_ERROR_1; + } + return 0; +} + +static clib_error_t * +af_packet_fd_read_ready (clib_file_t * uf) +{ + vnet_main_t *vnm = vnet_get_main (); + + /* Schedule the rx node */ + vnet_hw_if_rx_queue_set_int_pending (vnm, uf->private_data); + return 0; +} + +static clib_error_t * +af_packet_fd_error (clib_file_t *uf) +{ + af_packet_main_t *apm = &af_packet_main; + clib_error_t *err = 0; + u64 u64; + + int ret = read (uf->file_descriptor, (char *) &u64, sizeof (u64)); + + if (ret < 0) + { + err = clib_error_return_unix (0, ""); + vlib_log_notice (apm->log_class, "fd %u %U", uf->file_descriptor, + format_clib_error, err); + clib_error_free (err); + } + + return 0; +} + +static int +is_bridge (const u8 * host_if_name) +{ + u8 *s; + DIR *dir = NULL; + + s = format (0, "/sys/class/net/%s/bridge%c", host_if_name, 0); + dir = opendir ((char *) s); + vec_free (s); + + if (dir) + { + closedir (dir); + return 0; + } + + return -1; +} + +static void +af_packet_set_rx_queues (vlib_main_t *vm, af_packet_if_t *apif) +{ + vnet_main_t *vnm = vnet_get_main (); + af_packet_queue_t *rx_queue; + + vnet_hw_if_set_input_node (vnm, apif->hw_if_index, + af_packet_input_node.index); + + vec_foreach (rx_queue, apif->rx_queues) + { + rx_queue->queue_index = vnet_hw_if_register_rx_queue ( + vnm, apif->hw_if_index, rx_queue->queue_id, VNET_HW_IF_RXQ_THREAD_ANY); + + { + clib_file_t template = { 0 }; + template.read_function = af_packet_fd_read_ready; + template.error_function = af_packet_fd_error; + template.file_descriptor = rx_queue->fd; + template.private_data = rx_queue->queue_index; + template.description = + format (0, "%U queue %u", format_af_packet_device_name, + apif->dev_instance, rx_queue->queue_id); + rx_queue->clib_file_index = clib_file_add (&file_main, &template); + } + vnet_hw_if_set_rx_queue_file_index (vnm, rx_queue->queue_index, + rx_queue->clib_file_index); + vnet_hw_if_set_rx_queue_mode (vnm, rx_queue->queue_index, + VNET_HW_IF_RX_MODE_INTERRUPT); + rx_queue->mode = VNET_HW_IF_RX_MODE_INTERRUPT; + } + vnet_hw_if_update_runtime_data (vnm, apif->hw_if_index); +} + +static void +af_packet_set_tx_queues (vlib_main_t *vm, af_packet_if_t *apif) +{ + vnet_main_t *vnm = vnet_get_main (); + af_packet_main_t *apm = &af_packet_main; + af_packet_queue_t *tx_queue; + + vec_foreach (tx_queue, apif->tx_queues) + { + tx_queue->queue_index = vnet_hw_if_register_tx_queue ( + vnm, apif->hw_if_index, tx_queue->queue_id); + } + + if (apif->num_txqs == 0) + { + vlib_log_err (apm->log_class, "Interface %U has 0 txq", + format_vnet_hw_if_index_name, vnm, apif->hw_if_index); + return; + } + + for (u32 j = 0; j < vlib_get_n_threads (); j++) + { + u32 qi = apif->tx_queues[j % apif->num_txqs].queue_index; + vnet_hw_if_tx_queue_assign_thread (vnm, qi, j); + } + + vnet_hw_if_update_runtime_data (vnm, apif->hw_if_index); +} + +static int +create_packet_sock (int host_if_index, tpacket_req_u_t *rx_req, + tpacket_req_u_t *tx_req, int *fd, af_packet_ring_t *ring, + u32 fanout_id, af_packet_if_flags_t *flags, int ver) +{ + af_packet_main_t *apm = &af_packet_main; + struct sockaddr_ll sll; + socklen_t req_sz = sizeof (tpacket_req3_t); + int ret; + u32 ring_sz = 0; + + if ((*fd = socket (AF_PACKET, SOCK_RAW, htons (ETH_P_ALL))) < 0) + { + vlib_log_err (apm->log_class, + "Failed to create AF_PACKET socket: %s (errno %d)", + strerror (errno), errno); + ret = VNET_API_ERROR_SYSCALL_ERROR_1; + goto error; + } + + /* bind before rx ring is cfged so we don't receive packets from other interfaces */ + clib_memset (&sll, 0, sizeof (sll)); + sll.sll_family = PF_PACKET; + sll.sll_protocol = htons (ETH_P_ALL); + sll.sll_ifindex = host_if_index; + if (bind (*fd, (struct sockaddr *) &sll, sizeof (sll)) < 0) + { + vlib_log_err (apm->log_class, + "Failed to bind rx packet socket: %s (errno %d)", + strerror (errno), errno); + ret = VNET_API_ERROR_SYSCALL_ERROR_1; + goto error; + } + + if (setsockopt (*fd, SOL_PACKET, PACKET_VERSION, &ver, sizeof (ver)) < 0) + { + vlib_log_err (apm->log_class, + "Failed to set rx packet interface version: %s (errno %d)", + strerror (errno), errno); + ret = VNET_API_ERROR_SYSCALL_ERROR_1; + goto error; + } + + int opt = 1; + if (setsockopt (*fd, SOL_PACKET, PACKET_LOSS, &opt, sizeof (opt)) < 0) + { + vlib_log_err ( + apm->log_class, + "Failed to set packet tx ring error handling option: %s (errno %d)", + strerror (errno), errno); + ret = VNET_API_ERROR_SYSCALL_ERROR_1; + goto error; + } + + if (*flags & AF_PACKET_IF_FLAGS_CKSUM_GSO) + { + + int opt2 = 1; + if (setsockopt (*fd, SOL_PACKET, PACKET_VNET_HDR, &opt2, sizeof (opt2)) < + 0) + { + // remove the flag + *flags &= ~AF_PACKET_IF_FLAGS_CKSUM_GSO; + vlib_log_debug (apm->log_class, + "Failed to set packet vnet hdr error handling " + "option: %s (errno %d)", + strerror (errno), errno); + } + } + +#if defined(PACKET_QDISC_BYPASS) + if (*flags & AF_PACKET_IF_FLAGS_QDISC_BYPASS) + /* Introduced with Linux 3.14 so the ifdef should eventually be removed */ + if (setsockopt (*fd, SOL_PACKET, PACKET_QDISC_BYPASS, &opt, sizeof (opt)) < + 0) + { + // remove the flag + *flags &= ~AF_PACKET_IF_FLAGS_QDISC_BYPASS; + vlib_log_debug (apm->log_class, + "Failed to set qdisc bypass error " + "handling option: %s (errno %d)", + strerror (errno), errno); + } +#endif + + if (rx_req) + { + if (*flags & AF_PACKET_IF_FLAGS_FANOUT) + { + int fanout = ((fanout_id & 0xffff) | ((PACKET_FANOUT_HASH) << 16)); + if (setsockopt (*fd, SOL_PACKET, PACKET_FANOUT, &fanout, + sizeof (fanout)) < 0) + { + // remove the flag + *flags &= ~AF_PACKET_IF_FLAGS_FANOUT; + vlib_log_err (apm->log_class, + "Failed to set fanout options: %s (errno %d)", + strerror (errno), errno); + ret = VNET_API_ERROR_SYSCALL_ERROR_1; + goto error; + } + } + if (ver == TPACKET_V2) + { + req_sz = sizeof (tpacket_req_t); + ring_sz += rx_req->req.tp_block_size * rx_req->req.tp_block_nr; + } + else + ring_sz += rx_req->req3.tp_block_size * rx_req->req3.tp_block_nr; + if (setsockopt (*fd, SOL_PACKET, PACKET_RX_RING, rx_req, req_sz) < 0) + { + vlib_log_err (apm->log_class, + "Failed to set packet rx ring options: %s (errno %d)", + strerror (errno), errno); + ret = VNET_API_ERROR_SYSCALL_ERROR_1; + goto error; + } + } + + if (tx_req) + { + if (ver == TPACKET_V2) + { + req_sz = sizeof (tpacket_req_t); + ring_sz += tx_req->req.tp_block_size * tx_req->req.tp_block_nr; + } + else + ring_sz += tx_req->req3.tp_block_size * tx_req->req3.tp_block_nr; + if (setsockopt (*fd, SOL_PACKET, PACKET_TX_RING, tx_req, req_sz) < 0) + { + vlib_log_err (apm->log_class, + "Failed to set packet tx ring options: %s (errno %d)", + strerror (errno), errno); + ret = VNET_API_ERROR_SYSCALL_ERROR_1; + goto error; + } + } + ring->ring_start_addr = mmap (NULL, ring_sz, PROT_READ | PROT_WRITE, + MAP_SHARED | MAP_LOCKED, *fd, 0); + if (ring->ring_start_addr == MAP_FAILED) + { + vlib_log_err (apm->log_class, "mmap failure: %s (errno %d)", + strerror (errno), errno); + ret = VNET_API_ERROR_SYSCALL_ERROR_1; + goto error; + } + + ring->ring_size = ring_sz; + + return 0; +error: + if (*fd >= 0) + { + close (*fd); + *fd = -1; + } + return ret; +} + +int +af_packet_queue_init (vlib_main_t *vm, af_packet_if_t *apif, + af_packet_create_if_arg_t *arg, + af_packet_queue_t *rx_queue, af_packet_queue_t *tx_queue, + u8 queue_id) +{ + af_packet_main_t *apm = &af_packet_main; + tpacket_req_u_t *rx_req = 0; + tpacket_req_u_t *tx_req = 0; + int ret, fd = -1; + af_packet_ring_t ring = { 0 }; + u8 *ring_addr = 0; + u32 rx_frames_per_block, tx_frames_per_block; + u32 rx_frame_size, tx_frame_size; + u32 i = 0; + + if (rx_queue) + { + rx_frames_per_block = arg->rx_frames_per_block ? + arg->rx_frames_per_block : + ((apif->version == TPACKET_V3) ? + AF_PACKET_DEFAULT_RX_FRAMES_PER_BLOCK : + AF_PACKET_DEFAULT_RX_FRAMES_PER_BLOCK_V2); + + rx_frame_size = + arg->rx_frame_size ? + arg->rx_frame_size : + ((apif->version == TPACKET_V3) ? AF_PACKET_DEFAULT_RX_FRAME_SIZE : + AF_PACKET_DEFAULT_RX_FRAME_SIZE_V2); + vec_validate (rx_queue->rx_req, 0); + rx_queue->rx_req->req.tp_block_size = + rx_frame_size * rx_frames_per_block; + rx_queue->rx_req->req.tp_frame_size = rx_frame_size; + rx_queue->rx_req->req.tp_block_nr = (apif->version == TPACKET_V3) ? + AF_PACKET_RX_BLOCK_NR : + AF_PACKET_RX_BLOCK_NR_V2; + rx_queue->rx_req->req.tp_frame_nr = + rx_queue->rx_req->req.tp_block_nr * rx_frames_per_block; + if (apif->version == TPACKET_V3) + { + rx_queue->rx_req->req3.tp_retire_blk_tov = 1; // 1 ms block timout + rx_queue->rx_req->req3.tp_feature_req_word = 0; + rx_queue->rx_req->req3.tp_sizeof_priv = 0; + } + rx_req = rx_queue->rx_req; + } + if (tx_queue) + { + tx_frames_per_block = arg->tx_frames_per_block ? + arg->tx_frames_per_block : + AF_PACKET_DEFAULT_TX_FRAMES_PER_BLOCK; + tx_frame_size = arg->tx_frame_size ? arg->tx_frame_size : + AF_PACKET_DEFAULT_TX_FRAME_SIZE; + + vec_validate (tx_queue->tx_req, 0); + tx_queue->tx_req->req.tp_block_size = + tx_frame_size * tx_frames_per_block; + tx_queue->tx_req->req.tp_frame_size = tx_frame_size; + tx_queue->tx_req->req.tp_block_nr = AF_PACKET_TX_BLOCK_NR; + tx_queue->tx_req->req.tp_frame_nr = + AF_PACKET_TX_BLOCK_NR * tx_frames_per_block; + if (apif->version == TPACKET_V3) + { + tx_queue->tx_req->req3.tp_retire_blk_tov = 0; + tx_queue->tx_req->req3.tp_sizeof_priv = 0; + tx_queue->tx_req->req3.tp_feature_req_word = 0; + } + tx_req = tx_queue->tx_req; + } + + if (rx_queue || tx_queue) + { + ret = + create_packet_sock (apif->host_if_index, rx_req, tx_req, &fd, &ring, + apif->dev_instance, &arg->flags, apif->version); + + if (ret != 0) + goto error; + + vec_add1 (apif->fds, fd); + vec_add1 (apif->rings, ring); + ring_addr = ring.ring_start_addr; + } + + if (rx_queue) + { + rx_queue->fd = fd; + vec_validate (rx_queue->rx_ring, rx_queue->rx_req->req.tp_block_nr - 1); + vec_foreach_index (i, rx_queue->rx_ring) + { + rx_queue->rx_ring[i] = + ring_addr + i * rx_queue->rx_req->req.tp_block_size; + } + + rx_queue->next_rx_block = 0; + rx_queue->queue_id = queue_id; + rx_queue->is_rx_pending = 0; + ring_addr = ring_addr + rx_queue->rx_req->req.tp_block_size * + rx_queue->rx_req->req.tp_block_nr; + } + + if (tx_queue) + { + tx_queue->fd = fd; + vec_validate (tx_queue->tx_ring, tx_queue->tx_req->req.tp_block_nr - 1); + vec_foreach_index (i, tx_queue->tx_ring) + { + tx_queue->tx_ring[i] = + ring_addr + i * tx_queue->tx_req->req.tp_block_size; + } + + tx_queue->next_tx_frame = 0; + tx_queue->queue_id = queue_id; + tx_queue->is_tx_pending = 0; + clib_spinlock_init (&tx_queue->lockp); + } + + return 0; +error: + vlib_log_err (apm->log_class, "Failed to set queue %u error", queue_id); + if (rx_queue) + vec_free (rx_queue->rx_req); + if (tx_queue) + vec_free (tx_queue->tx_req); + return ret; +} + +int +af_packet_device_init (vlib_main_t *vm, af_packet_if_t *apif, + af_packet_create_if_arg_t *args) +{ + af_packet_main_t *apm = &af_packet_main; + af_packet_queue_t *rx_queue = 0; + af_packet_queue_t *tx_queue = 0; + u16 nq = clib_min (args->num_rxqs, args->num_txqs); + u16 i = 0; + int ret = 0; + + // enable fanout feature for multi-rxqs + if (args->num_rxqs > 1) + args->flags |= AF_PACKET_IF_FLAGS_FANOUT; + + vec_validate (apif->rx_queues, args->num_rxqs - 1); + vec_validate (apif->tx_queues, args->num_txqs - 1); + + for (; i < nq; i++) + { + rx_queue = vec_elt_at_index (apif->rx_queues, i); + tx_queue = vec_elt_at_index (apif->tx_queues, i); + ret = af_packet_queue_init (vm, apif, args, rx_queue, tx_queue, i); + if (ret != 0) + goto error; + } + + if (args->num_rxqs > args->num_txqs) + { + for (; i < args->num_rxqs; i++) + { + rx_queue = vec_elt_at_index (apif->rx_queues, i); + ret = af_packet_queue_init (vm, apif, args, rx_queue, 0, i); + if (ret != 0) + goto error; + } + } + else if (args->num_txqs > args->num_rxqs) + { + for (; i < args->num_txqs; i++) + { + tx_queue = vec_elt_at_index (apif->tx_queues, i); + ret = af_packet_queue_init (vm, apif, args, 0, tx_queue, i); + if (ret != 0) + goto error; + } + } + + apif->num_rxqs = args->num_rxqs; + apif->num_txqs = args->num_txqs; + + return 0; +error: + vlib_log_err (apm->log_class, "Failed to init device error"); + return ret; +} + +int +af_packet_create_if (af_packet_create_if_arg_t *arg) +{ + af_packet_main_t *apm = &af_packet_main; + vlib_main_t *vm = vlib_get_main (); + int fd2 = -1; + struct ifreq ifr; + af_packet_if_t *apif = 0; + u8 hw_addr[6]; + vnet_sw_interface_t *sw; + vnet_main_t *vnm = vnet_get_main (); + vnet_hw_if_caps_t caps = VNET_HW_IF_CAP_INT_MODE; + uword *p; + uword if_index; + u8 *host_if_name_dup = 0; + int host_if_index = -1; + int ret = 0; + u32 oflags = 0, i = 0; + + p = mhash_get (&apm->if_index_by_host_if_name, arg->host_if_name); + if (p) + { + apif = vec_elt_at_index (apm->interfaces, p[0]); + arg->sw_if_index = apif->sw_if_index; + return VNET_API_ERROR_IF_ALREADY_EXISTS; + } + + host_if_name_dup = vec_dup (arg->host_if_name); + + /* + * make sure host side of interface is 'UP' before binding AF_PACKET + * socket on it. + */ + if ((fd2 = socket (AF_UNIX, SOCK_DGRAM, 0)) < 0) + { + vlib_log_debug (apm->log_class, + "Failed to create AF_UNIX socket: %s (errno %d)", + strerror (errno), errno); + ret = VNET_API_ERROR_SYSCALL_ERROR_1; + goto error; + } + + clib_memcpy (ifr.ifr_name, (const char *) arg->host_if_name, + vec_len (arg->host_if_name)); + if (ioctl (fd2, SIOCGIFINDEX, &ifr) < 0) + { + vlib_log_debug ( + apm->log_class, + "Failed to retrieve the interface (%s) index: %s (errno %d)", + arg->host_if_name, strerror (errno), errno); + ret = VNET_API_ERROR_INVALID_INTERFACE; + goto error; + } + + host_if_index = ifr.ifr_ifindex; + if (ioctl (fd2, SIOCGIFFLAGS, &ifr) < 0) + { + vlib_log_debug (apm->log_class, + "Failed to get the active flag: %s (errno %d)", + strerror (errno), errno); + ret = VNET_API_ERROR_SYSCALL_ERROR_1; + goto error; + } + + if (!(ifr.ifr_flags & IFF_UP)) + { + ifr.ifr_flags |= IFF_UP; + if (ioctl (fd2, SIOCSIFFLAGS, &ifr) < 0) + { + vlib_log_debug (apm->log_class, + "Failed to set the active flag: %s (errno %d)", + strerror (errno), errno); + ret = VNET_API_ERROR_SYSCALL_ERROR_1; + goto error; + } + } + + if (fd2 > -1) + { + close (fd2); + fd2 = -1; + } + + // check the host interface capabilities + oflags = af_packet_get_if_capabilities (arg->host_if_name); + + ret = is_bridge (arg->host_if_name); + if (ret == 0) /* is a bridge, ignore state */ + host_if_index = -1; + + /* So far everything looks good, let's create interface */ + pool_get (apm->interfaces, apif); + if_index = apif - apm->interfaces; + + apif->dev_instance = if_index; + apif->host_if_index = host_if_index; + apif->host_if_name = host_if_name_dup; + apif->per_interface_next_index = ~0; + apif->mode = arg->mode; + apif->host_interface_oflags = oflags; + + if (arg->is_v2) + apif->version = TPACKET_V2; + else + apif->version = TPACKET_V3; + + ret = af_packet_device_init (vm, apif, arg); + if (ret != 0) + goto error; + + ret = af_packet_read_mtu (apif); + if (ret != 0) + goto error; + + + if (apif->mode != AF_PACKET_IF_MODE_IP) + { + vnet_eth_interface_registration_t eir = {}; + /*use configured or generate random MAC address */ + if (arg->hw_addr) + clib_memcpy (hw_addr, arg->hw_addr, 6); + else + { + f64 now = vlib_time_now (vm); + u32 rnd; + rnd = (u32) (now * 1e6); + rnd = random_u32 (&rnd); + + clib_memcpy (hw_addr + 2, &rnd, sizeof (rnd)); + hw_addr[0] = 2; + hw_addr[1] = 0xfe; + } + + eir.dev_class_index = af_packet_device_class.index; + eir.dev_instance = apif->dev_instance; + eir.address = hw_addr; + eir.cb.set_max_frame_size = af_packet_eth_set_max_frame_size; + apif->hw_if_index = vnet_eth_register_interface (vnm, &eir); + } + else + { + apif->hw_if_index = vnet_register_interface ( + vnm, af_packet_device_class.index, apif->dev_instance, + af_packet_ip_device_hw_interface_class.index, apif->dev_instance); + } + + sw = vnet_get_hw_sw_interface (vnm, apif->hw_if_index); + apif->sw_if_index = sw->sw_if_index; + + af_packet_set_rx_queues (vm, apif); + af_packet_set_tx_queues (vm, apif); + + if (arg->flags & AF_PACKET_IF_FLAGS_FANOUT) + apif->is_fanout_enabled = 1; + + apif->is_qdisc_bypass_enabled = + (arg->flags & AF_PACKET_IF_FLAGS_QDISC_BYPASS); + + if (arg->flags & AF_PACKET_IF_FLAGS_CKSUM_GSO) + { + if (apif->host_interface_oflags & AF_PACKET_OFFLOAD_FLAG_TXCKSUM) + { + apif->is_cksum_gso_enabled = 1; + caps |= VNET_HW_IF_CAP_TX_IP4_CKSUM | VNET_HW_IF_CAP_TX_TCP_CKSUM | + VNET_HW_IF_CAP_TX_UDP_CKSUM; + } + + if (apif->host_interface_oflags & AF_PACKET_OFFLOAD_FLAG_GSO) + { + apif->is_cksum_gso_enabled = 1; + caps |= VNET_HW_IF_CAP_TCP_GSO | VNET_HW_IF_CAP_TX_IP4_CKSUM | + VNET_HW_IF_CAP_TX_TCP_CKSUM | VNET_HW_IF_CAP_TX_UDP_CKSUM; + } + } + vnet_hw_if_set_caps (vnm, apif->hw_if_index, caps); + vnet_hw_interface_set_flags (vnm, apif->hw_if_index, + VNET_HW_INTERFACE_FLAG_LINK_UP); + + mhash_set_mem (&apm->if_index_by_host_if_name, host_if_name_dup, &if_index, + 0); + arg->sw_if_index = apif->sw_if_index; + + return 0; + +error: + if (fd2 > -1) + { + close (fd2); + fd2 = -1; + } + + vec_free (host_if_name_dup); + + if (apif) + { + vec_foreach_index (i, apif->fds) + if (apif->fds[i] != -1) + close (apif->fds[i]); + vec_free (apif->fds); + memset (apif, 0, sizeof (*apif)); + pool_put (apm->interfaces, apif); + } + return ret; +} + +static int +af_packet_rx_queue_free (af_packet_if_t *apif, af_packet_queue_t *rx_queue) +{ + clib_file_del_by_index (&file_main, rx_queue->clib_file_index); + rx_queue->fd = -1; + rx_queue->rx_ring = NULL; + vec_free (rx_queue->rx_req); + rx_queue->rx_req = NULL; + return 0; +} + +static int +af_packet_tx_queue_free (af_packet_if_t *apif, af_packet_queue_t *tx_queue) +{ + tx_queue->fd = -1; + clib_spinlock_free (&tx_queue->lockp); + tx_queue->tx_ring = NULL; + vec_free (tx_queue->tx_req); + tx_queue->tx_req = NULL; + return 0; +} + +static int +af_packet_ring_free (af_packet_if_t *apif, af_packet_ring_t *ring) +{ + af_packet_main_t *apm = &af_packet_main; + + if (ring) + { + // FIXME: unmap the memory + if (munmap (ring->ring_start_addr, ring->ring_size)) + vlib_log_warn (apm->log_class, + "Host interface %s could not free ring %p of size %u", + apif->host_if_name, ring->ring_start_addr, + ring->ring_size); + else + ring->ring_start_addr = 0; + } + + return 0; +} + +int +af_packet_delete_if (u8 *host_if_name) +{ + vnet_main_t *vnm = vnet_get_main (); + af_packet_main_t *apm = &af_packet_main; + af_packet_if_t *apif; + af_packet_queue_t *rx_queue; + af_packet_queue_t *tx_queue; + af_packet_ring_t *ring; + uword *p; + u32 i = 0; + + p = mhash_get (&apm->if_index_by_host_if_name, host_if_name); + if (p == NULL) + { + vlib_log_warn (apm->log_class, "Host interface %s does not exist", + host_if_name); + return VNET_API_ERROR_SYSCALL_ERROR_1; + } + apif = pool_elt_at_index (apm->interfaces, p[0]); + + /* bring down the interface */ + vnet_hw_interface_set_flags (vnm, apif->hw_if_index, 0); + if (apif->mode != AF_PACKET_IF_MODE_IP) + ethernet_delete_interface (vnm, apif->hw_if_index); + else + vnet_delete_hw_interface (vnm, apif->hw_if_index); + + /* clean up */ + vec_foreach_index (i, apif->fds) + if (apif->fds[i] != -1) + close (apif->fds[i]); + vec_foreach (rx_queue, apif->rx_queues) + af_packet_rx_queue_free (apif, rx_queue); + vec_foreach (tx_queue, apif->tx_queues) + af_packet_tx_queue_free (apif, tx_queue); + vec_foreach (ring, apif->rings) + af_packet_ring_free (apif, ring); + + vec_free (apif->fds); + apif->fds = NULL; + vec_free (apif->rx_queues); + apif->rx_queues = NULL; + vec_free (apif->tx_queues); + apif->tx_queues = NULL; + vec_free (apif->rings); + apif->rings = NULL; + + vec_free (apif->host_if_name); + apif->host_if_name = NULL; + apif->host_if_index = -1; + + mhash_unset (&apm->if_index_by_host_if_name, host_if_name, p); + + memset (apif, 0, sizeof (*apif)); + pool_put (apm->interfaces, apif); + + return 0; +} + +int +af_packet_enable_disable_qdisc_bypass (u32 sw_if_index, u8 enable_disable) +{ + af_packet_main_t *apm = &af_packet_main; + af_packet_if_t *apif; + vnet_main_t *vnm = vnet_get_main (); + vnet_hw_interface_t *hw; + u32 i; + + hw = vnet_get_sup_hw_interface_api_visible_or_null (vnm, sw_if_index); + + if (hw->dev_class_index != af_packet_device_class.index) + return VNET_API_ERROR_INVALID_INTERFACE; + + apif = pool_elt_at_index (apm->interfaces, hw->dev_instance); + +#if defined(PACKET_QDISC_BYPASS) + vec_foreach_index (i, apif->fds) + { + if (enable_disable) + { + int opt = 1; + + /* Introduced with Linux 3.14 so the ifdef should eventually be + * removed */ + if (setsockopt (apif->fds[i], SOL_PACKET, PACKET_QDISC_BYPASS, &opt, + sizeof (opt)) < 0) + { + vlib_log_err (apm->log_class, + "Failed to enable qdisc bypass error " + "handling option: %s (errno %d)", + strerror (errno), errno); + } + apif->is_qdisc_bypass_enabled = 1; + } + else + { + int opt = 0; + if (setsockopt (apif->fds[i], SOL_PACKET, PACKET_QDISC_BYPASS, &opt, + sizeof (opt)) < 0) + { + vlib_log_err (apm->log_class, + "Failed to disable qdisc bypass error " + "handling option: %s (errno %d)", + strerror (errno), errno); + } + apif->is_qdisc_bypass_enabled = 0; + } + } + +#endif + return 0; +} + +int +af_packet_set_l4_cksum_offload (u32 sw_if_index, u8 set) +{ + // deprecated ... + return 0; +} + +int +af_packet_dump_ifs (af_packet_if_detail_t ** out_af_packet_ifs) +{ + af_packet_main_t *apm = &af_packet_main; + af_packet_if_t *apif; + af_packet_if_detail_t *r_af_packet_ifs = NULL; + af_packet_if_detail_t *af_packet_if = NULL; + + pool_foreach (apif, apm->interfaces) + { + vec_add2 (r_af_packet_ifs, af_packet_if, 1); + af_packet_if->sw_if_index = apif->sw_if_index; + if (apif->host_if_name) + { + clib_memcpy (af_packet_if->host_if_name, apif->host_if_name, + MIN (ARRAY_LEN (af_packet_if->host_if_name) - 1, + strlen ((const char *) apif->host_if_name))); + } + } + + *out_af_packet_ifs = r_af_packet_ifs; + + return 0; +} + +static clib_error_t * +af_packet_init (vlib_main_t * vm) +{ + af_packet_main_t *apm = &af_packet_main; + vlib_thread_main_t *tm = vlib_get_thread_main (); + + clib_memset (apm, 0, sizeof (af_packet_main_t)); + + mhash_init_vec_string (&apm->if_index_by_host_if_name, sizeof (uword)); + + vec_validate_aligned (apm->rx_buffers, tm->n_vlib_mains - 1, + CLIB_CACHE_LINE_BYTES); + + apm->log_class = vlib_log_register_class ("af_packet", 0); + vlib_log_debug (apm->log_class, "initialized"); + + return 0; +} + +VLIB_INIT_FUNCTION (af_packet_init); + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/plugins/af_packet/af_packet.h b/src/plugins/af_packet/af_packet.h new file mode 100644 index 00000000000..e66a1119ba1 --- /dev/null +++ b/src/plugins/af_packet/af_packet.h @@ -0,0 +1,182 @@ +/* + *------------------------------------------------------------------ + * af_packet.h - linux kernel packet interface header file + * + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + *------------------------------------------------------------------ + */ + +#include <linux/if_packet.h> + +#include <vppinfra/lock.h> +#include <vlib/log.h> + +typedef struct tpacket_block_desc block_desc_t; +typedef struct tpacket_req tpacket_req_t; +typedef struct tpacket_req3 tpacket_req3_t; +typedef struct tpacket2_hdr tpacket2_hdr_t; +typedef struct tpacket3_hdr tpacket3_hdr_t; + +typedef union _tpacket_req_u +{ + tpacket_req_t req; + tpacket_req3_t req3; +} tpacket_req_u_t; + +typedef enum +{ + AF_PACKET_IF_MODE_ETHERNET = 1, + AF_PACKET_IF_MODE_IP = 2 +} af_packet_if_mode_t; + +typedef enum +{ + AF_PACKET_IF_FLAGS_QDISC_BYPASS = 1, + AF_PACKET_IF_FLAGS_CKSUM_GSO = 2, + AF_PACKET_IF_FLAGS_FANOUT = 4, + AF_PACKET_IF_FLAGS_VERSION_2 = 8, +} af_packet_if_flags_t; + +typedef struct +{ + u32 sw_if_index; + u8 host_if_name[64]; +} af_packet_if_detail_t; + +typedef struct +{ + u8 *ring_start_addr; + u32 ring_size; +} af_packet_ring_t; + +typedef struct +{ + CLIB_CACHE_LINE_ALIGN_MARK (cacheline0); + clib_spinlock_t lockp; + int fd; + union + { + tpacket_req_u_t *rx_req; + tpacket_req_u_t *tx_req; + }; + + union + { + u8 **rx_ring; + u8 **tx_ring; + }; + + union + { + u32 next_rx_block; + u32 next_rx_frame; + u32 next_tx_frame; + }; + + u16 queue_id; + u32 queue_index; + + u32 clib_file_index; + + u32 rx_frame_offset; + u16 num_rx_pkts; + u8 is_rx_pending; + u8 is_tx_pending; + vnet_hw_if_rx_mode mode; +} af_packet_queue_t; + +typedef struct +{ + CLIB_CACHE_LINE_ALIGN_MARK (cacheline0); + u32 hw_if_index; + u32 sw_if_index; + u32 per_interface_next_index; + af_packet_if_mode_t mode; + u8 is_admin_up; + u8 is_cksum_gso_enabled; + u8 version; + af_packet_queue_t *rx_queues; + af_packet_queue_t *tx_queues; + + u8 num_rxqs; + u8 num_txqs; + + u8 *host_if_name; + int host_if_index; + + u32 host_mtu; + u32 dev_instance; + + af_packet_ring_t *rings; + u8 is_qdisc_bypass_enabled; + u8 is_fanout_enabled; + int *fds; + u32 host_interface_oflags; +} af_packet_if_t; + +typedef struct +{ + CLIB_CACHE_LINE_ALIGN_MARK (cacheline0); + af_packet_if_t *interfaces; + + u32 polling_count; + /* rx buffer cache */ + u32 **rx_buffers; + + /* hash of host interface names */ + mhash_t if_index_by_host_if_name; + + /** log class */ + vlib_log_class_t log_class; +} af_packet_main_t; + +typedef struct +{ + u8 *host_if_name; + u8 *hw_addr; + u32 rx_frame_size; + u32 tx_frame_size; + u32 rx_frames_per_block; + u32 tx_frames_per_block; + u8 num_rxqs; + u8 num_txqs; + u8 is_v2; + af_packet_if_mode_t mode; + af_packet_if_flags_t flags; + + /* return */ + u32 sw_if_index; +} af_packet_create_if_arg_t; + +extern af_packet_main_t af_packet_main; +extern vnet_device_class_t af_packet_device_class; +extern vlib_node_registration_t af_packet_input_node; + +int af_packet_create_if (af_packet_create_if_arg_t *arg); +int af_packet_delete_if (u8 *host_if_name); +int af_packet_set_l4_cksum_offload (u32 sw_if_index, u8 set); +int af_packet_enable_disable_qdisc_bypass (u32 sw_if_index, u8 enable_disable); +int af_packet_dump_ifs (af_packet_if_detail_t ** out_af_packet_ifs); + +format_function_t format_af_packet_device_name; + +#define MIN(x,y) (((x)<(y))?(x):(y)) + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/plugins/af_packet/af_packet_api.c b/src/plugins/af_packet/af_packet_api.c new file mode 100644 index 00000000000..ede057659e8 --- /dev/null +++ b/src/plugins/af_packet/af_packet_api.c @@ -0,0 +1,253 @@ +/* + *------------------------------------------------------------------ + * af_packet_api.c - af-packet api + * + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + *------------------------------------------------------------------ + */ + +#include <vnet/vnet.h> +#include <vlibmemory/api.h> + +#include <vnet/interface.h> +#include <vnet/api_errno.h> +#include <af_packet/af_packet.h> + +#include <vnet/format_fns.h> +#include <af_packet/af_packet.api_enum.h> +#include <af_packet/af_packet.api_types.h> + +#define REPLY_MSG_ID_BASE msg_id_base +#include <vlibapi/api_helper_macros.h> + +static u16 msg_id_base; + +static void +vl_api_af_packet_create_t_handler (vl_api_af_packet_create_t * mp) +{ + af_packet_create_if_arg_t _arg, *arg = &_arg; + vl_api_af_packet_create_reply_t *rmp; + int rv = 0; + + clib_memset (arg, 0, sizeof (*arg)); + + arg->host_if_name = format (0, "%s", mp->host_if_name); + vec_add1 (arg->host_if_name, 0); + + arg->hw_addr = mp->use_random_hw_addr ? 0 : mp->hw_addr; + arg->mode = AF_PACKET_IF_MODE_ETHERNET; + // Default flags + arg->flags = AF_PACKET_IF_FLAGS_QDISC_BYPASS | AF_PACKET_IF_FLAGS_CKSUM_GSO; + rv = af_packet_create_if (arg); + + vec_free (arg->host_if_name); + + REPLY_MACRO2 (VL_API_AF_PACKET_CREATE_REPLY, ({ + rmp->sw_if_index = clib_host_to_net_u32 (arg->sw_if_index); + })); +} + +static void +vl_api_af_packet_create_v2_t_handler (vl_api_af_packet_create_v2_t *mp) +{ + af_packet_create_if_arg_t _arg, *arg = &_arg; + vl_api_af_packet_create_v2_reply_t *rmp; + int rv = 0; + + clib_memset (arg, 0, sizeof (*arg)); + + arg->host_if_name = format (0, "%s", mp->host_if_name); + vec_add1 (arg->host_if_name, 0); + + // Default number of rx/tx queue(s) + arg->num_rxqs = 1; + arg->num_txqs = 1; + arg->rx_frame_size = clib_net_to_host_u32 (mp->rx_frame_size); + arg->tx_frame_size = clib_net_to_host_u32 (mp->tx_frame_size); + arg->rx_frames_per_block = clib_net_to_host_u32 (mp->rx_frames_per_block); + arg->tx_frames_per_block = clib_net_to_host_u32 (mp->tx_frames_per_block); + arg->hw_addr = mp->use_random_hw_addr ? 0 : mp->hw_addr; + arg->mode = AF_PACKET_IF_MODE_ETHERNET; + // Default flags + arg->flags = AF_PACKET_IF_FLAGS_QDISC_BYPASS | AF_PACKET_IF_FLAGS_CKSUM_GSO; + + if (mp->num_rx_queues > 1) + arg->num_rxqs = clib_net_to_host_u16 (mp->num_rx_queues); + + rv = af_packet_create_if (arg); + + vec_free (arg->host_if_name); + REPLY_MACRO2 (VL_API_AF_PACKET_CREATE_V2_REPLY, ({ + rmp->sw_if_index = clib_host_to_net_u32 (arg->sw_if_index); + })); +} + +static void +vl_api_af_packet_create_v3_t_handler (vl_api_af_packet_create_v3_t *mp) +{ + af_packet_create_if_arg_t _arg, *arg = &_arg; + vl_api_af_packet_create_v3_reply_t *rmp; + int rv = 0; + + clib_memset (arg, 0, sizeof (*arg)); + + arg->host_if_name = format (0, "%s", mp->host_if_name); + vec_add1 (arg->host_if_name, 0); + + // Default number of rx/tx queue(s) + arg->num_rxqs = 1; + arg->num_txqs = 1; + arg->rx_frame_size = clib_net_to_host_u32 (mp->rx_frame_size); + arg->tx_frame_size = clib_net_to_host_u32 (mp->tx_frame_size); + arg->rx_frames_per_block = clib_net_to_host_u32 (mp->rx_frames_per_block); + arg->tx_frames_per_block = clib_net_to_host_u32 (mp->tx_frames_per_block); + arg->hw_addr = mp->use_random_hw_addr ? 0 : mp->hw_addr; + + switch (clib_net_to_host_u32 (mp->mode)) + { + case AF_PACKET_API_MODE_ETHERNET: + arg->mode = AF_PACKET_IF_MODE_ETHERNET; + break; + case AF_PACKET_API_MODE_IP: + arg->mode = AF_PACKET_IF_MODE_IP; + break; + default: + arg->sw_if_index = ~0; + rv = VNET_ERR_INVALID_VALUE; + goto error; + } + + STATIC_ASSERT (((int) AF_PACKET_API_FLAG_QDISC_BYPASS == + (int) AF_PACKET_IF_FLAGS_QDISC_BYPASS), + "af-packet qdisc-bypass api flag mismatch"); + STATIC_ASSERT ( + ((int) AF_PACKET_API_FLAG_CKSUM_GSO == (int) AF_PACKET_IF_FLAGS_CKSUM_GSO), + "af-packet checksum/gso offload api flag mismatch"); + + STATIC_ASSERT ( + ((int) AF_PACKET_API_FLAG_VERSION_2 == (int) AF_PACKET_IF_FLAGS_VERSION_2), + "af-packet version 2 api flag mismatch"); + + // Default flags + arg->flags = clib_net_to_host_u32 (mp->flags); + + if (clib_net_to_host_u16 (mp->num_rx_queues) > 1) + arg->num_rxqs = clib_net_to_host_u16 (mp->num_rx_queues); + + if (clib_net_to_host_u16 (mp->num_tx_queues) > 1) + arg->num_txqs = clib_net_to_host_u16 (mp->num_tx_queues); + + arg->is_v2 = (arg->flags & AF_PACKET_API_FLAG_VERSION_2) ? 1 : 0; + rv = af_packet_create_if (arg); + +error: + vec_free (arg->host_if_name); + REPLY_MACRO2 (VL_API_AF_PACKET_CREATE_V3_REPLY, ({ + rmp->sw_if_index = clib_host_to_net_u32 (arg->sw_if_index); + })); +} + +static void +vl_api_af_packet_delete_t_handler (vl_api_af_packet_delete_t * mp) +{ + vl_api_af_packet_delete_reply_t *rmp; + int rv = 0; + u8 *host_if_name = NULL; + + host_if_name = format (0, "%s", mp->host_if_name); + vec_add1 (host_if_name, 0); + + rv = af_packet_delete_if (host_if_name); + + vec_free (host_if_name); + + REPLY_MACRO (VL_API_AF_PACKET_DELETE_REPLY); +} + +static void + vl_api_af_packet_set_l4_cksum_offload_t_handler + (vl_api_af_packet_set_l4_cksum_offload_t * mp) +{ + vl_api_af_packet_delete_reply_t *rmp; + int rv = 0; + + rv = af_packet_set_l4_cksum_offload (ntohl (mp->sw_if_index), mp->set); + REPLY_MACRO (VL_API_AF_PACKET_SET_L4_CKSUM_OFFLOAD_REPLY); +} + +static void +af_packet_send_details (vpe_api_main_t * am, + vl_api_registration_t * reg, + af_packet_if_detail_t * af_packet_if, u32 context) +{ + vl_api_af_packet_details_t *mp; + mp = vl_msg_api_alloc (sizeof (*mp)); + clib_memset (mp, 0, sizeof (*mp)); + mp->_vl_msg_id = htons (REPLY_MSG_ID_BASE + VL_API_AF_PACKET_DETAILS); + mp->sw_if_index = htonl (af_packet_if->sw_if_index); + clib_memcpy (mp->host_if_name, af_packet_if->host_if_name, + MIN (ARRAY_LEN (mp->host_if_name) - 1, + strlen ((const char *) af_packet_if->host_if_name))); + + mp->context = context; + vl_api_send_msg (reg, (u8 *) mp); +} + + +static void +vl_api_af_packet_dump_t_handler (vl_api_af_packet_dump_t * mp) +{ + int rv; + vpe_api_main_t *am = &vpe_api_main; + vl_api_registration_t *reg; + af_packet_if_detail_t *out_af_packet_ifs = NULL; + af_packet_if_detail_t *af_packet_if = NULL; + + reg = vl_api_client_index_to_registration (mp->client_index); + if (!reg) + return; + + rv = af_packet_dump_ifs (&out_af_packet_ifs); + if (rv) + return; + + vec_foreach (af_packet_if, out_af_packet_ifs) + { + af_packet_send_details (am, reg, af_packet_if, mp->context); + } + + vec_free (out_af_packet_ifs); +} + +#include <af_packet/af_packet.api.c> +static clib_error_t * +af_packet_api_hookup (vlib_main_t * vm) +{ + /* + * Set up the (msg_name, crc, message-id) table + */ + REPLY_MSG_ID_BASE = setup_message_id_table (); + + return 0; +} + +VLIB_API_INIT_FUNCTION (af_packet_api_hookup); + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/plugins/af_packet/cli.c b/src/plugins/af_packet/cli.c new file mode 100644 index 00000000000..2af3fb17ee5 --- /dev/null +++ b/src/plugins/af_packet/cli.c @@ -0,0 +1,341 @@ +/* + *------------------------------------------------------------------ + * af_packet.c - linux kernel packet interface + * + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + *------------------------------------------------------------------ + */ + +#include <fcntl.h> /* for open */ +#include <sys/ioctl.h> +#include <sys/socket.h> +#include <sys/stat.h> +#include <sys/types.h> +#include <sys/uio.h> /* for iovec */ +#include <netinet/in.h> + +#include <vlib/vlib.h> +#include <vlib/unix/unix.h> +#include <vnet/ip/ip.h> +#include <vnet/ethernet/ethernet.h> + +#include <af_packet/af_packet.h> + +/** + * @file + * @brief CLI for Host Interface Device Driver. + * + * This file contains the source code for CLI for the host interface. + */ + +static clib_error_t * +af_packet_create_command_fn (vlib_main_t * vm, unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + unformat_input_t _line_input, *line_input = &_line_input; + af_packet_create_if_arg_t _arg, *arg = &_arg; + clib_error_t *error = NULL; + u8 hwaddr[6]; + int r; + + clib_memset (arg, 0, sizeof (*arg)); + + // Default mode + arg->mode = AF_PACKET_IF_MODE_ETHERNET; + + // Default number of rx/tx queue(s) + arg->num_rxqs = 1; + arg->num_txqs = 1; + + // Default flags + arg->flags = AF_PACKET_IF_FLAGS_QDISC_BYPASS | AF_PACKET_IF_FLAGS_CKSUM_GSO; + + /* Get a line of input. */ + if (!unformat_user (input, unformat_line_input, line_input)) + return 0; + + while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (line_input, "name %s", &arg->host_if_name)) + ; + else if (unformat (line_input, "rx-size %u", &arg->rx_frame_size)) + ; + else if (unformat (line_input, "tx-size %u", &arg->tx_frame_size)) + ; + else if (unformat (line_input, "rx-per-block %u", + &arg->rx_frames_per_block)) + ; + else if (unformat (line_input, "tx-per-block %u", + &arg->tx_frames_per_block)) + ; + else if (unformat (line_input, "num-rx-queues %u", &arg->num_rxqs)) + ; + else if (unformat (line_input, "num-tx-queues %u", &arg->num_txqs)) + ; + else if (unformat (line_input, "qdisc-bypass-disable")) + arg->flags &= ~AF_PACKET_IF_FLAGS_QDISC_BYPASS; + else if (unformat (line_input, "cksum-gso-disable")) + arg->flags &= ~AF_PACKET_IF_FLAGS_CKSUM_GSO; + else if (unformat (line_input, "mode ip")) + arg->mode = AF_PACKET_IF_MODE_IP; + else if (unformat (line_input, "v2")) + arg->is_v2 = 1; + else if (unformat (line_input, "hw-addr %U", unformat_ethernet_address, + hwaddr)) + arg->hw_addr = hwaddr; + else + { + error = clib_error_return (0, "unknown input `%U'", + format_unformat_error, line_input); + goto done; + } + } + + if (arg->host_if_name == NULL) + { + error = clib_error_return (0, "missing host interface name"); + goto done; + } + + r = af_packet_create_if (arg); + + if (r == VNET_API_ERROR_SYSCALL_ERROR_1) + { + error = clib_error_return (0, "%s (errno %d)", strerror (errno), errno); + goto done; + } + + if (r == VNET_API_ERROR_INVALID_INTERFACE) + { + error = clib_error_return (0, "Invalid interface name"); + goto done; + } + + if (r == VNET_API_ERROR_SUBIF_ALREADY_EXISTS) + { + error = clib_error_return (0, "Interface already exists"); + goto done; + } + + vlib_cli_output (vm, "%U\n", format_vnet_sw_if_index_name, vnet_get_main (), + arg->sw_if_index); + +done: + vec_free (arg->host_if_name); + unformat_free (line_input); + + return error; +} + +/*? + * Create a host interface that will attach to a linux AF_PACKET + * interface, one side of a veth pair. The veth pair must already + * exist. Once created, a new host interface will exist in VPP + * with the name '<em>host-<ifname></em>', where '<em><ifname></em>' + * is the name of the specified veth pair. Use the + * '<em>show interface</em>' command to display host interface details. + * + * This command has the following optional parameters: + * + * - <b>hw-addr <mac-addr></b> - Optional ethernet address, can be in either + * X:X:X:X:X:X unix or X.X.X cisco format. + * + * @cliexpar + * Example of how to create a host interface tied to one side of an + * existing linux veth pair named vpp1: + * @cliexstart{create host-interface name vpp1} + * host-vpp1 + * @cliexend + * Once the host interface is created, enable the interface using: + * @cliexcmd{set interface state host-vpp1 up} +?*/ +VLIB_CLI_COMMAND (af_packet_create_command, static) = { + .path = "create host-interface", + .short_help = "create host-interface [v2] name <ifname> [num-rx-queues <n>] " + "[num-tx-queues <n>] [hw-addr <mac-addr>] [mode ip] " + "[qdisc-bypass-disable] [cksum-gso-disable]", + .function = af_packet_create_command_fn, +}; + +static clib_error_t * +af_packet_delete_command_fn (vlib_main_t * vm, unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + unformat_input_t _line_input, *line_input = &_line_input; + u8 *host_if_name = NULL; + clib_error_t *error = NULL; + + /* Get a line of input. */ + if (!unformat_user (input, unformat_line_input, line_input)) + return 0; + + while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (line_input, "name %s", &host_if_name)) + ; + else + { + error = clib_error_return (0, "unknown input `%U'", + format_unformat_error, line_input); + goto done; + } + } + + if (host_if_name == NULL) + { + error = clib_error_return (0, "missing host interface name"); + goto done; + } + + af_packet_delete_if (host_if_name); + +done: + vec_free (host_if_name); + unformat_free (line_input); + + return error; +} + +/*? + * Delete a host interface. Use the linux interface name to identify + * the host interface to be deleted. In VPP, host interfaces are + * named as '<em>host-<ifname></em>', where '<em><ifname></em>' + * is the name of the linux interface. + * + * @cliexpar + * Example of how to delete a host interface named host-vpp1: + * @cliexcmd{delete host-interface name vpp1} +?*/ +VLIB_CLI_COMMAND (af_packet_delete_command, static) = { + .path = "delete host-interface", + .short_help = "delete host-interface name <ifname>", + .function = af_packet_delete_command_fn, +}; + +static clib_error_t * +af_packet_set_l4_cksum_offload_command_fn (vlib_main_t * vm, + unformat_input_t * input, + vlib_cli_command_t * cmd) +{ + unformat_input_t _line_input, *line_input = &_line_input; + u8 set = 0; + clib_error_t *error = NULL; + vnet_main_t *vnm = vnet_get_main (); + u32 sw_if_index; + + if (!unformat_user (input, unformat_line_input, line_input)) + return 0; + + while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (line_input, "%U", unformat_vnet_sw_interface, vnm, + &sw_if_index)) + ; + else if (unformat (line_input, "on")) + set = 1; + else if (unformat (line_input, "off")) + set = 0; + else + { + error = clib_error_return (0, "unknown input '%U'", + format_unformat_error, line_input); + goto done; + } + } + + if (af_packet_set_l4_cksum_offload (sw_if_index, set) < 0) + error = clib_error_return (0, "not an af_packet interface"); + +done: + unformat_free (line_input); + return error; +} + +/*? + * Set TCP/UDP offload checksum calculation. Use interface + * name to identify the interface to set TCP/UDP offload checksum + * calculation. + * + * @cliexpar + * Example of how to set TCP/UDP offload checksum calculation on host-vpp0: + * @cliexcmd{set host-interface l4-cksum-offload host-vpp0 off} + * @cliexcmd{set host-interface l4-cksum-offload host-vpp0 on} +?*/ +VLIB_CLI_COMMAND (af_packet_set_l4_cksum_offload_command, static) = { + .path = "set host-interface l4-cksum-offload", + .short_help = "set host-interface l4-cksum-offload <host-if-name> <on|off>", + .function = af_packet_set_l4_cksum_offload_command_fn, +}; + +static clib_error_t * +af_packet_enable_disable_qdisc_bypass_command_fn (vlib_main_t *vm, + unformat_input_t *input, + vlib_cli_command_t *cmd) +{ + unformat_input_t _line_input, *line_input = &_line_input; + u8 enable_disable = 0; + clib_error_t *error = NULL; + vnet_main_t *vnm = vnet_get_main (); + u32 sw_if_index; + + if (!unformat_user (input, unformat_line_input, line_input)) + return 0; + + while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (line_input, "%U", unformat_vnet_sw_interface, vnm, + &sw_if_index)) + ; + else if (unformat (line_input, "enable")) + enable_disable = 1; + else if (unformat (line_input, "disable")) + enable_disable = 0; + else + { + error = clib_error_return (0, "unknown input '%U'", + format_unformat_error, line_input); + goto done; + } + } + + if (af_packet_enable_disable_qdisc_bypass (sw_if_index, enable_disable) < 0) + error = clib_error_return (0, "not an af_packet interface"); + +done: + unformat_free (line_input); + return error; +} + +VLIB_CLI_COMMAND (af_packet_enable_disable_qdisc_bypass_command, static) = { + .path = "set host-interface qdisc-bypass", + .short_help = + "set host-interface qdisc-bypass <host-if-name> <enable|disable>", + .function = af_packet_enable_disable_qdisc_bypass_command_fn, +}; + +clib_error_t * +af_packet_cli_init (vlib_main_t * vm) +{ + return 0; +} + +VLIB_INIT_FUNCTION (af_packet_cli_init); + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/plugins/af_packet/device.c b/src/plugins/af_packet/device.c new file mode 100644 index 00000000000..d76dad3dde0 --- /dev/null +++ b/src/plugins/af_packet/device.c @@ -0,0 +1,793 @@ +/* + *------------------------------------------------------------------ + * af_packet.c - linux kernel packet interface + * + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + *------------------------------------------------------------------ + */ + +#include <linux/if_packet.h> +#include <sys/socket.h> +#include <sys/ioctl.h> +#include <net/if.h> +#include <net/if_arp.h> + +#include <vlib/vlib.h> +#include <vlib/unix/unix.h> +#include <vnet/ip/ip.h> +#include <vnet/ethernet/ethernet.h> +#include <vnet/ip/ip4_packet.h> +#include <vnet/ip/ip6_packet.h> +#include <vnet/ip/ip_psh_cksum.h> +#include <vnet/tcp/tcp_packet.h> +#include <vnet/udp/udp_packet.h> + +#include <af_packet/af_packet.h> +#include <vnet/devices/virtio/virtio_std.h> +#include <vnet/devices/netlink.h> + +#define foreach_af_packet_tx_func_error \ +_(FRAME_NOT_READY, "tx frame not ready") \ +_(TXRING_EAGAIN, "tx sendto temporary failure") \ +_(TXRING_FATAL, "tx sendto fatal failure") \ +_(TXRING_OVERRUN, "tx ring overrun") + +typedef enum +{ +#define _(f,s) AF_PACKET_TX_ERROR_##f, + foreach_af_packet_tx_func_error +#undef _ + AF_PACKET_TX_N_ERROR, +} af_packet_tx_func_error_t; + +static char *af_packet_tx_func_error_strings[] = { +#define _(n,s) s, + foreach_af_packet_tx_func_error +#undef _ +}; + +typedef struct +{ + u32 buffer_index; + u32 hw_if_index; + u16 queue_id; + u8 is_v2; + union + { + tpacket2_hdr_t tph2; + tpacket3_hdr_t tph3; + }; + vnet_virtio_net_hdr_t vnet_hdr; + vlib_buffer_t buffer; +} af_packet_tx_trace_t; + +#ifndef CLIB_MARCH_VARIANT +u8 * +format_af_packet_device_name (u8 * s, va_list * args) +{ + u32 i = va_arg (*args, u32); + af_packet_main_t *apm = &af_packet_main; + af_packet_if_t *apif = pool_elt_at_index (apm->interfaces, i); + + s = format (s, "host-%s", apif->host_if_name); + return s; +} +#endif /* CLIB_MARCH_VARIANT */ + +static u8 * +format_af_packet_device (u8 * s, va_list * args) +{ + u32 dev_instance = va_arg (*args, u32); + u32 indent = format_get_indent (s); + int __clib_unused verbose = va_arg (*args, int); + + af_packet_main_t *apm = &af_packet_main; + af_packet_if_t *apif = pool_elt_at_index (apm->interfaces, dev_instance); + af_packet_queue_t *rx_queue = 0; + af_packet_queue_t *tx_queue = 0; + + s = format (s, "Linux PACKET socket interface %s", + (apif->version == TPACKET_V2) ? "v2" : "v3"); + s = format (s, "\n%UFEATURES:", format_white_space, indent); + if (apif->is_qdisc_bypass_enabled) + s = format (s, "\n%Uqdisc-bpass-enabled", format_white_space, indent + 2); + if (apif->is_cksum_gso_enabled) + s = format (s, "\n%Ucksum-gso-enabled", format_white_space, indent + 2); + if (apif->is_fanout_enabled) + s = format (s, "\n%Ufanout-enabled", format_white_space, indent + 2); + + vec_foreach (rx_queue, apif->rx_queues) + { + u32 rx_block_size = rx_queue->rx_req->req.tp_block_size; + u32 rx_frame_size = rx_queue->rx_req->req.tp_frame_size; + u32 rx_frame_nr = rx_queue->rx_req->req.tp_frame_nr; + u32 rx_block_nr = rx_queue->rx_req->req.tp_block_nr; + + s = format (s, "\n%URX Queue %u:", format_white_space, indent, + rx_queue->queue_id); + s = format (s, "\n%Ublock size:%d nr:%d frame size:%d nr:%d", + format_white_space, indent + 2, rx_block_size, rx_block_nr, + rx_frame_size, rx_frame_nr); + if (apif->version == TPACKET_V2) + s = format (s, " next frame:%d", rx_queue->next_rx_frame); + else + s = format (s, " next block:%d", rx_queue->next_rx_block); + if (rx_queue->is_rx_pending) + { + s = format ( + s, "\n%UPending Request: num-rx-pkts:%d next-frame-offset:%d", + format_white_space, indent + 2, rx_queue->num_rx_pkts, + rx_queue->rx_frame_offset); + } + } + + vec_foreach (tx_queue, apif->tx_queues) + { + clib_spinlock_lock (&tx_queue->lockp); + u32 tx_block_sz = tx_queue->tx_req->req.tp_block_size; + u32 tx_frame_sz = tx_queue->tx_req->req.tp_frame_size; + u32 tx_frame_nr = tx_queue->tx_req->req.tp_frame_nr; + u32 tx_block_nr = tx_queue->tx_req->req.tp_block_nr; + int block = 0; + int n_send_req = 0, n_avail = 0, n_sending = 0, n_tot = 0, n_wrong = 0; + u8 *tx_block_start = tx_queue->tx_ring[block]; + u32 tx_frame = tx_queue->next_tx_frame; + tpacket3_hdr_t *tph3; + tpacket2_hdr_t *tph2; + + s = format (s, "\n%UTX Queue %u:", format_white_space, indent, + tx_queue->queue_id); + s = format (s, "\n%Ublock size:%d nr:%d frame size:%d nr:%d", + format_white_space, indent + 2, tx_block_sz, tx_block_nr, + tx_frame_sz, tx_frame_nr); + s = format (s, " next frame:%d", tx_queue->next_tx_frame); + if (apif->version & TPACKET_V3) + do + { + tph3 = + (tpacket3_hdr_t *) (tx_block_start + tx_frame * tx_frame_sz); + tx_frame = (tx_frame + 1) % tx_frame_nr; + if (tph3->tp_status == 0) + n_avail++; + else if (tph3->tp_status & TP_STATUS_SEND_REQUEST) + n_send_req++; + else if (tph3->tp_status & TP_STATUS_SENDING) + n_sending++; + else + n_wrong++; + n_tot++; + } + while (tx_frame != tx_queue->next_tx_frame); + else + do + { + tph2 = + (tpacket2_hdr_t *) (tx_block_start + tx_frame * tx_frame_sz); + tx_frame = (tx_frame + 1) % tx_frame_nr; + if (tph2->tp_status == 0) + n_avail++; + else if (tph2->tp_status & TP_STATUS_SEND_REQUEST) + n_send_req++; + else if (tph2->tp_status & TP_STATUS_SENDING) + n_sending++; + else + n_wrong++; + n_tot++; + } + while (tx_frame != tx_queue->next_tx_frame); + s = + format (s, "\n%Uavailable:%d request:%d sending:%d wrong:%d total:%d", + format_white_space, indent + 2, n_avail, n_send_req, n_sending, + n_wrong, n_tot); + clib_spinlock_unlock (&tx_queue->lockp); + } + return s; +} + +static u8 * +format_af_packet_tx_trace (u8 *s, va_list *va) +{ + CLIB_UNUSED (vlib_main_t * vm) = va_arg (*va, vlib_main_t *); + CLIB_UNUSED (vlib_node_t * node) = va_arg (*va, vlib_node_t *); + af_packet_tx_trace_t *t = va_arg (*va, af_packet_tx_trace_t *); + u32 indent = format_get_indent (s); + + s = format (s, "af_packet: hw_if_index %u tx-queue %u", t->hw_if_index, + t->queue_id); + + if (t->is_v2) + { + s = format ( + s, + "\n%Utpacket2_hdr:\n%Ustatus 0x%x len %u snaplen %u mac %u net %u" + "\n%Usec 0x%x nsec 0x%x vlan %U" +#ifdef TP_STATUS_VLAN_TPID_VALID + " vlan_tpid %u" +#endif + , + format_white_space, indent + 2, format_white_space, indent + 4, + t->tph2.tp_status, t->tph2.tp_len, t->tph2.tp_snaplen, t->tph2.tp_mac, + t->tph2.tp_net, format_white_space, indent + 4, t->tph2.tp_sec, + t->tph2.tp_nsec, format_ethernet_vlan_tci, t->tph2.tp_vlan_tci +#ifdef TP_STATUS_VLAN_TPID_VALID + , + t->tph2.tp_vlan_tpid +#endif + ); + } + else + { + s = format ( + s, + "\n%Utpacket3_hdr:\n%Ustatus 0x%x len %u snaplen %u mac %u net %u" + "\n%Usec 0x%x nsec 0x%x vlan %U" +#ifdef TP_STATUS_VLAN_TPID_VALID + " vlan_tpid %u" +#endif + , + format_white_space, indent + 2, format_white_space, indent + 4, + t->tph3.tp_status, t->tph3.tp_len, t->tph3.tp_snaplen, t->tph3.tp_mac, + t->tph3.tp_net, format_white_space, indent + 4, t->tph3.tp_sec, + t->tph3.tp_nsec, format_ethernet_vlan_tci, t->tph3.hv1.tp_vlan_tci +#ifdef TP_STATUS_VLAN_TPID_VALID + , + t->tph3.hv1.tp_vlan_tpid +#endif + ); + } + s = format (s, + "\n%Uvnet-hdr:\n%Uflags 0x%02x gso_type 0x%02x hdr_len %u" + "\n%Ugso_size %u csum_start %u csum_offset %u", + format_white_space, indent + 2, format_white_space, indent + 4, + t->vnet_hdr.flags, t->vnet_hdr.gso_type, t->vnet_hdr.hdr_len, + format_white_space, indent + 4, t->vnet_hdr.gso_size, + t->vnet_hdr.csum_start, t->vnet_hdr.csum_offset); + + s = format (s, "\n%Ubuffer 0x%x:\n%U%U", format_white_space, indent + 2, + t->buffer_index, format_white_space, indent + 4, + format_vnet_buffer_no_chain, &t->buffer); + s = format (s, "\n%U%U", format_white_space, indent + 2, + format_ethernet_header_with_length, t->buffer.pre_data, + sizeof (t->buffer.pre_data)); + return s; +} + +static void +af_packet_tx_trace (vlib_main_t *vm, vlib_node_runtime_t *node, + vlib_buffer_t *b0, u32 bi, void *tph, + vnet_virtio_net_hdr_t *vnet_hdr, u32 hw_if_index, + u16 queue_id, u8 is_v2) +{ + af_packet_tx_trace_t *t; + t = vlib_add_trace (vm, node, b0, sizeof (t[0])); + t->hw_if_index = hw_if_index; + t->queue_id = queue_id; + t->buffer_index = bi; + t->is_v2 = is_v2; + + if (is_v2) + clib_memcpy_fast (&t->tph2, (tpacket2_hdr_t *) tph, + sizeof (tpacket2_hdr_t)); + else + clib_memcpy_fast (&t->tph3, (tpacket3_hdr_t *) tph, + sizeof (tpacket3_hdr_t)); + clib_memcpy_fast (&t->vnet_hdr, vnet_hdr, sizeof (*vnet_hdr)); + clib_memcpy_fast (&t->buffer, b0, sizeof (*b0) - sizeof (b0->pre_data)); + clib_memcpy_fast (t->buffer.pre_data, vlib_buffer_get_current (b0), + sizeof (t->buffer.pre_data)); +} + +static_always_inline void +fill_gso_offload (vlib_buffer_t *b0, vnet_virtio_net_hdr_t *vnet_hdr) +{ + vnet_buffer_oflags_t oflags = vnet_buffer (b0)->oflags; + i16 l4_hdr_offset = vnet_buffer (b0)->l4_hdr_offset - b0->current_data; + if (b0->flags & VNET_BUFFER_F_IS_IP4) + { + ip4_header_t *ip4; + vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; + vnet_hdr->gso_size = vnet_buffer2 (b0)->gso_size; + vnet_hdr->hdr_len = l4_hdr_offset + vnet_buffer2 (b0)->gso_l4_hdr_sz; + vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; + vnet_hdr->csum_start = l4_hdr_offset; // 0x22; + vnet_hdr->csum_offset = STRUCT_OFFSET_OF (tcp_header_t, checksum); + ip4 = (ip4_header_t *) (b0->data + vnet_buffer (b0)->l3_hdr_offset); + if (oflags & VNET_BUFFER_OFFLOAD_F_IP_CKSUM) + ip4->checksum = ip4_header_checksum (ip4); + tcp_header_t *tcp = + (tcp_header_t *) (b0->data + vnet_buffer (b0)->l4_hdr_offset); + tcp->checksum = ip4_pseudo_header_cksum (ip4); + } + else if (b0->flags & VNET_BUFFER_F_IS_IP6) + { + ip6_header_t *ip6; + vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; + vnet_hdr->gso_size = vnet_buffer2 (b0)->gso_size; + vnet_hdr->hdr_len = l4_hdr_offset + vnet_buffer2 (b0)->gso_l4_hdr_sz; + vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; + vnet_hdr->csum_start = l4_hdr_offset; // 0x36; + vnet_hdr->csum_offset = STRUCT_OFFSET_OF (tcp_header_t, checksum); + ip6 = (ip6_header_t *) (b0->data + vnet_buffer (b0)->l3_hdr_offset); + tcp_header_t *tcp = + (tcp_header_t *) (b0->data + vnet_buffer (b0)->l4_hdr_offset); + tcp->checksum = ip6_pseudo_header_cksum (ip6); + } +} + +static_always_inline void +fill_cksum_offload (vlib_buffer_t *b0, vnet_virtio_net_hdr_t *vnet_hdr) +{ + vnet_buffer_oflags_t oflags = vnet_buffer (b0)->oflags; + i16 l4_hdr_offset = vnet_buffer (b0)->l4_hdr_offset - b0->current_data; + if (b0->flags & VNET_BUFFER_F_IS_IP4) + { + ip4_header_t *ip4; + ip4 = (ip4_header_t *) (b0->data + vnet_buffer (b0)->l3_hdr_offset); + if (oflags & VNET_BUFFER_OFFLOAD_F_IP_CKSUM) + ip4->checksum = ip4_header_checksum (ip4); + vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; + vnet_hdr->csum_start = l4_hdr_offset; + if (oflags & VNET_BUFFER_OFFLOAD_F_TCP_CKSUM) + { + tcp_header_t *tcp = + (tcp_header_t *) (b0->data + vnet_buffer (b0)->l4_hdr_offset); + tcp->checksum = ip4_pseudo_header_cksum (ip4); + vnet_hdr->csum_offset = STRUCT_OFFSET_OF (tcp_header_t, checksum); + vnet_hdr->hdr_len = l4_hdr_offset + tcp_header_bytes (tcp); + } + else if (oflags & VNET_BUFFER_OFFLOAD_F_UDP_CKSUM) + { + udp_header_t *udp = + (udp_header_t *) (b0->data + vnet_buffer (b0)->l4_hdr_offset); + udp->checksum = ip4_pseudo_header_cksum (ip4); + vnet_hdr->csum_offset = STRUCT_OFFSET_OF (udp_header_t, checksum); + vnet_hdr->hdr_len = l4_hdr_offset + sizeof (udp_header_t); + } + } + else if (b0->flags & VNET_BUFFER_F_IS_IP6) + { + ip6_header_t *ip6; + vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; + vnet_hdr->csum_start = l4_hdr_offset; + ip6 = (ip6_header_t *) (b0->data + vnet_buffer (b0)->l3_hdr_offset); + if (oflags & VNET_BUFFER_OFFLOAD_F_TCP_CKSUM) + { + tcp_header_t *tcp = + (tcp_header_t *) (b0->data + vnet_buffer (b0)->l4_hdr_offset); + tcp->checksum = ip6_pseudo_header_cksum (ip6); + vnet_hdr->csum_offset = STRUCT_OFFSET_OF (tcp_header_t, checksum); + vnet_hdr->hdr_len = l4_hdr_offset + tcp_header_bytes (tcp); + } + else if (oflags & VNET_BUFFER_OFFLOAD_F_UDP_CKSUM) + { + udp_header_t *udp = + (udp_header_t *) (b0->data + vnet_buffer (b0)->l4_hdr_offset); + udp->checksum = ip6_pseudo_header_cksum (ip6); + vnet_hdr->csum_offset = STRUCT_OFFSET_OF (udp_header_t, checksum); + vnet_hdr->hdr_len = l4_hdr_offset + sizeof (udp_header_t); + } + } +} + +VNET_DEVICE_CLASS_TX_FN (af_packet_device_class) (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * frame) +{ + af_packet_main_t *apm = &af_packet_main; + vnet_hw_if_tx_frame_t *tf = vlib_frame_scalar_args (frame); + u32 *buffers = vlib_frame_vector_args (frame); + u32 n_left = frame->n_vectors; + u32 n_sent = 0; + vnet_interface_output_runtime_t *rd = (void *) node->runtime_data; + af_packet_if_t *apif = + pool_elt_at_index (apm->interfaces, rd->dev_instance); + u16 queue_id = tf->queue_id; + af_packet_queue_t *tx_queue = vec_elt_at_index (apif->tx_queues, queue_id); + u32 block = 0, frame_size = 0, frame_num = 0, tx_frame = 0; + u8 *block_start = 0; + tpacket3_hdr_t *tph3 = 0; + tpacket2_hdr_t *tph2 = 0; + u32 frame_not_ready = 0; + u8 is_cksum_gso_enabled = (apif->is_cksum_gso_enabled == 1) ? 1 : 0; + u32 tpacket_align = 0; + u8 is_v2 = (apif->version == TPACKET_V2) ? 1 : 0; + + if (tf->shared_queue) + clib_spinlock_lock (&tx_queue->lockp); + + frame_size = tx_queue->tx_req->req.tp_frame_size; + frame_num = tx_queue->tx_req->req.tp_frame_nr; + block_start = tx_queue->tx_ring[block]; + tx_frame = tx_queue->next_tx_frame; + if (is_v2) + { + tpacket_align = TPACKET_ALIGN (sizeof (tpacket2_hdr_t)); + while (n_left) + { + u32 len; + vnet_virtio_net_hdr_t *vnet_hdr = 0; + u32 offset = 0; + vlib_buffer_t *b0 = 0, *b0_first = 0; + u32 bi, bi_first; + + bi = bi_first = buffers[0]; + n_left--; + buffers++; + + tph2 = (tpacket2_hdr_t *) (block_start + tx_frame * frame_size); + if (PREDICT_FALSE (tph2->tp_status & + (TP_STATUS_SEND_REQUEST | TP_STATUS_SENDING))) + { + frame_not_ready++; + goto nextv2; + } + + b0_first = b0 = vlib_get_buffer (vm, bi); + + if (PREDICT_TRUE (is_cksum_gso_enabled)) + { + vnet_hdr = + (vnet_virtio_net_hdr_t *) ((u8 *) tph2 + tpacket_align); + + clib_memset_u8 (vnet_hdr, 0, sizeof (vnet_virtio_net_hdr_t)); + offset = sizeof (vnet_virtio_net_hdr_t); + + if (b0->flags & VNET_BUFFER_F_GSO) + fill_gso_offload (b0, vnet_hdr); + else if (b0->flags & VNET_BUFFER_F_OFFLOAD) + fill_cksum_offload (b0, vnet_hdr); + } + + len = b0->current_length; + clib_memcpy_fast ((u8 *) tph2 + tpacket_align + offset, + vlib_buffer_get_current (b0), len); + offset += len; + + while (b0->flags & VLIB_BUFFER_NEXT_PRESENT) + { + b0 = vlib_get_buffer (vm, b0->next_buffer); + len = b0->current_length; + clib_memcpy_fast ((u8 *) tph2 + tpacket_align + offset, + vlib_buffer_get_current (b0), len); + offset += len; + } + + tph2->tp_len = tph2->tp_snaplen = offset; + tph2->tp_status = TP_STATUS_SEND_REQUEST; + n_sent++; + + if (PREDICT_FALSE (b0_first->flags & VLIB_BUFFER_IS_TRACED)) + { + if (PREDICT_TRUE (is_cksum_gso_enabled)) + af_packet_tx_trace (vm, node, b0_first, bi_first, tph2, + vnet_hdr, apif->hw_if_index, queue_id, 1); + else + { + vnet_virtio_net_hdr_t vnet_hdr2 = {}; + af_packet_tx_trace (vm, node, b0_first, bi_first, tph2, + &vnet_hdr2, apif->hw_if_index, queue_id, + 1); + } + } + tx_frame = (tx_frame + 1) % frame_num; + + nextv2: + /* check if we've exhausted the ring */ + if (PREDICT_FALSE (frame_not_ready + n_sent == frame_num)) + break; + } + } + else + { + tpacket_align = TPACKET_ALIGN (sizeof (tpacket3_hdr_t)); + + while (n_left) + { + u32 len; + vnet_virtio_net_hdr_t *vnet_hdr = 0; + u32 offset = 0; + vlib_buffer_t *b0 = 0, *b0_first = 0; + u32 bi, bi_first; + + bi = bi_first = buffers[0]; + n_left--; + buffers++; + + tph3 = (tpacket3_hdr_t *) (block_start + tx_frame * frame_size); + if (PREDICT_FALSE (tph3->tp_status & + (TP_STATUS_SEND_REQUEST | TP_STATUS_SENDING))) + { + frame_not_ready++; + goto nextv3; + } + + b0_first = b0 = vlib_get_buffer (vm, bi); + + if (PREDICT_TRUE (is_cksum_gso_enabled)) + { + vnet_hdr = + (vnet_virtio_net_hdr_t *) ((u8 *) tph3 + tpacket_align); + + clib_memset_u8 (vnet_hdr, 0, sizeof (vnet_virtio_net_hdr_t)); + offset = sizeof (vnet_virtio_net_hdr_t); + + if (b0->flags & VNET_BUFFER_F_GSO) + fill_gso_offload (b0, vnet_hdr); + else if (b0->flags & VNET_BUFFER_F_OFFLOAD) + fill_cksum_offload (b0, vnet_hdr); + } + + len = b0->current_length; + clib_memcpy_fast ((u8 *) tph3 + tpacket_align + offset, + vlib_buffer_get_current (b0), len); + offset += len; + + while (b0->flags & VLIB_BUFFER_NEXT_PRESENT) + { + b0 = vlib_get_buffer (vm, b0->next_buffer); + len = b0->current_length; + clib_memcpy_fast ((u8 *) tph3 + tpacket_align + offset, + vlib_buffer_get_current (b0), len); + offset += len; + } + + tph3->tp_len = tph3->tp_snaplen = offset; + tph3->tp_status = TP_STATUS_SEND_REQUEST; + n_sent++; + + if (PREDICT_FALSE (b0_first->flags & VLIB_BUFFER_IS_TRACED)) + { + if (PREDICT_TRUE (is_cksum_gso_enabled)) + af_packet_tx_trace (vm, node, b0_first, bi_first, tph3, + vnet_hdr, apif->hw_if_index, queue_id, 0); + else + { + vnet_virtio_net_hdr_t vnet_hdr2 = {}; + af_packet_tx_trace (vm, node, b0_first, bi_first, tph3, + &vnet_hdr2, apif->hw_if_index, queue_id, + 0); + } + } + tx_frame = (tx_frame + 1) % frame_num; + + nextv3: + /* check if we've exhausted the ring */ + if (PREDICT_FALSE (frame_not_ready + n_sent == frame_num)) + break; + } + } + CLIB_MEMORY_BARRIER (); + + if (PREDICT_TRUE (n_sent || tx_queue->is_tx_pending)) + { + tx_queue->next_tx_frame = tx_frame; + tx_queue->is_tx_pending = 0; + + if (PREDICT_FALSE ( + sendto (tx_queue->fd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1)) + { + /* Uh-oh, drop & move on, but count whether it was fatal or not. + * Note that we have no reliable way to properly determine the + * disposition of the packets we just enqueued for delivery. + */ + uword counter; + + if (unix_error_is_fatal (errno)) + { + counter = AF_PACKET_TX_ERROR_TXRING_FATAL; + } + else + { + counter = AF_PACKET_TX_ERROR_TXRING_EAGAIN; + /* non-fatal error: kick again next time + * note that you could still end up in a deadlock: if you do not + * try to send new packets (ie reschedule this tx node), eg. + * because your peer is waiting for the unsent packets to reply + * to you but your waiting for its reply etc., you are not going + * to kick again, and everybody is waiting for the other to talk + * 1st... */ + tx_queue->is_tx_pending = 1; + } + + vlib_error_count (vm, node->node_index, counter, 1); + } + } + + if (tf->shared_queue) + clib_spinlock_unlock (&tx_queue->lockp); + + if (PREDICT_FALSE (frame_not_ready)) + vlib_error_count (vm, node->node_index, + AF_PACKET_TX_ERROR_FRAME_NOT_READY, frame_not_ready); + + if (PREDICT_FALSE (frame_not_ready + n_sent == frame_num)) + vlib_error_count (vm, node->node_index, AF_PACKET_TX_ERROR_TXRING_OVERRUN, + n_left); + + vlib_buffer_free (vm, vlib_frame_vector_args (frame), frame->n_vectors); + return frame->n_vectors; +} + +static void +af_packet_set_interface_next_node (vnet_main_t * vnm, u32 hw_if_index, + u32 node_index) +{ + af_packet_main_t *apm = &af_packet_main; + vnet_hw_interface_t *hw = vnet_get_hw_interface (vnm, hw_if_index); + af_packet_if_t *apif = + pool_elt_at_index (apm->interfaces, hw->dev_instance); + + /* Shut off redirection */ + if (node_index == ~0) + { + apif->per_interface_next_index = node_index; + return; + } + + apif->per_interface_next_index = + vlib_node_add_next (vlib_get_main (), af_packet_input_node.index, + node_index); +} + +static void +af_packet_clear_hw_interface_counters (u32 instance) +{ + /* Nothing for now */ +} + +static clib_error_t * +af_packet_interface_admin_up_down (vnet_main_t * vnm, u32 hw_if_index, + u32 flags) +{ + af_packet_main_t *apm = &af_packet_main; + vnet_hw_interface_t *hw = vnet_get_hw_interface (vnm, hw_if_index); + af_packet_if_t *apif = + pool_elt_at_index (apm->interfaces, hw->dev_instance); + u32 hw_flags; + + if (apif->host_if_index < 0) + return 0; /* no error */ + + apif->is_admin_up = (flags & VNET_SW_INTERFACE_FLAG_ADMIN_UP) != 0; + + if (apif->is_admin_up) + { + hw_flags = VNET_HW_INTERFACE_FLAG_LINK_UP; + vnet_netlink_set_link_state (apif->host_if_index, 1); + } + else + { + hw_flags = 0; + vnet_netlink_set_link_state (apif->host_if_index, 0); + } + + vnet_hw_interface_set_flags (vnm, hw_if_index, hw_flags); + + return 0; /* no error */ +} + +static clib_error_t *af_packet_set_mac_address_function + (struct vnet_hw_interface_t *hi, const u8 * old_address, const u8 * address) +{ + af_packet_main_t *apm = &af_packet_main; + af_packet_if_t *apif = + pool_elt_at_index (apm->interfaces, hi->dev_instance); + int rv, fd; + struct ifreq ifr; + + if (apif->mode == AF_PACKET_IF_MODE_IP) + { + vlib_log_warn (apm->log_class, "af_packet_%s interface is in IP mode", + apif->host_if_name); + return clib_error_return (0, + " MAC update failed, interface is in IP mode"); + } + + fd = socket (AF_UNIX, SOCK_DGRAM, 0); + if (0 > fd) + { + vlib_log_warn (apm->log_class, "af_packet_%s could not open socket", + apif->host_if_name); + return 0; + } + + /* if interface is a bridge ignore */ + if (apif->host_if_index < 0) + goto error; /* no error */ + + /* use host_if_index in case host name has changed */ + ifr.ifr_ifindex = apif->host_if_index; + if ((rv = ioctl (fd, SIOCGIFNAME, &ifr)) < 0) + { + vlib_log_warn + (apm->log_class, + "af_packet_%s ioctl could not retrieve eth name, error: %d", + apif->host_if_name, rv); + goto error; + } + + clib_memcpy (ifr.ifr_hwaddr.sa_data, address, 6); + ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER; + + if ((rv = ioctl (fd, SIOCSIFHWADDR, &ifr)) < 0) + { + vlib_log_warn (apm->log_class, + "af_packet_%s ioctl could not set mac, error: %d", + apif->host_if_name, rv); + goto error; + } + +error: + + if (0 <= fd) + close (fd); + + return 0; /* no error */ +} + +static clib_error_t * +af_packet_interface_rx_mode_change (vnet_main_t *vnm, u32 hw_if_index, u32 qid, + vnet_hw_if_rx_mode mode) +{ + af_packet_main_t *apm = &af_packet_main; + vnet_hw_interface_t *hw = vnet_get_hw_interface (vnm, hw_if_index); + af_packet_if_t *apif; + + apif = vec_elt_at_index (apm->interfaces, hw->dev_instance); + + if (mode == VNET_HW_IF_RX_MODE_ADAPTIVE) + { + vlib_log_err (apm->log_class, + "af_packet_%s adaptive mode is not supported", + apif->host_if_name); + return clib_error_return ( + 0, "af_packet_%s adaptive mode is not supported", apif->host_if_name); + } + + af_packet_queue_t *rx_queue = vec_elt_at_index (apif->rx_queues, qid); + + if (rx_queue->mode != mode) + { + rx_queue->mode = mode; + + if (mode == VNET_HW_IF_RX_MODE_POLLING) + apm->polling_count++; + else if (mode == VNET_HW_IF_RX_MODE_INTERRUPT && apm->polling_count > 0) + apm->polling_count--; + } + + return 0; +} + +VNET_DEVICE_CLASS (af_packet_device_class) = { + .name = "af-packet", + .format_device_name = format_af_packet_device_name, + .format_device = format_af_packet_device, + .format_tx_trace = format_af_packet_tx_trace, + .tx_function_n_errors = AF_PACKET_TX_N_ERROR, + .tx_function_error_strings = af_packet_tx_func_error_strings, + .rx_redirect_to_node = af_packet_set_interface_next_node, + .clear_counters = af_packet_clear_hw_interface_counters, + .admin_up_down_function = af_packet_interface_admin_up_down, + .mac_addr_change_function = af_packet_set_mac_address_function, + .rx_mode_change_function = af_packet_interface_rx_mode_change, +}; + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/plugins/af_packet/dir.dox b/src/plugins/af_packet/dir.dox new file mode 100644 index 00000000000..78991c6d97f --- /dev/null +++ b/src/plugins/af_packet/dir.dox @@ -0,0 +1,29 @@ +/* + * Copyright (c) 2016 Cisco and/or its affiliates. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +/* Doxygen directory documentation */ + +/** +@dir +@brief Host Interface Implementation. + +This directory contains the source code for Host Interface driver. The +Host Interface driver leverages the DPDK AF_PACKET driver. + + +*/ +/*? %%clicmd:group_label Host Interface %% ?*/ +/*? %%syscfg:group_label Host Interface %% ?*/ diff --git a/src/plugins/af_packet/node.c b/src/plugins/af_packet/node.c new file mode 100644 index 00000000000..279f11c0183 --- /dev/null +++ b/src/plugins/af_packet/node.c @@ -0,0 +1,832 @@ +/* + *------------------------------------------------------------------ + * af_packet.c - linux kernel packet interface + * + * Copyright (c) 2016 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + *------------------------------------------------------------------ + */ + +#include <linux/if_packet.h> + +#include <vlib/vlib.h> +#include <vlib/unix/unix.h> +#include <vnet/ip/ip.h> +#include <vnet/ethernet/ethernet.h> +#include <vnet/interface/rx_queue_funcs.h> +#include <vnet/feature/feature.h> +#include <vnet/ethernet/packet.h> + +#include <af_packet/af_packet.h> +#include <vnet/devices/virtio/virtio_std.h> + +#define foreach_af_packet_input_error \ + _ (PARTIAL_PKT, "partial packet") \ + _ (TIMEDOUT_BLK, "timed out block") \ + _ (TOTAL_RECV_BLK, "total received block") +typedef enum +{ +#define _(f,s) AF_PACKET_INPUT_ERROR_##f, + foreach_af_packet_input_error +#undef _ + AF_PACKET_INPUT_N_ERROR, +} af_packet_input_error_t; + +static char *af_packet_input_error_strings[] = { +#define _(n,s) s, + foreach_af_packet_input_error +#undef _ +}; + +typedef struct +{ + u32 next_index; + u32 hw_if_index; + u16 queue_id; + int block; + u32 pkt_num; + void *block_start; + block_desc_t bd; + union + { + tpacket3_hdr_t tph3; + tpacket2_hdr_t tph2; + }; + vnet_virtio_net_hdr_t vnet_hdr; + u8 is_v3; +} af_packet_input_trace_t; + +static u8 * +format_af_packet_input_trace (u8 * s, va_list * args) +{ + CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *); + CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *); + af_packet_input_trace_t *t = va_arg (*args, af_packet_input_trace_t *); + u32 indent = format_get_indent (s); + + s = format (s, "af_packet: hw_if_index %d rx-queue %u next-index %d", + t->hw_if_index, t->queue_id, t->next_index); + + if (t->is_v3) + { + s = format ( + s, "\n%Ublock %u:\n%Uaddress %p version %u seq_num %lu pkt_num %u", + format_white_space, indent + 2, t->block, format_white_space, + indent + 4, t->block_start, t->bd.version, t->bd.hdr.bh1.seq_num, + t->pkt_num); + s = format ( + s, + "\n%Utpacket3_hdr:\n%Ustatus 0x%x len %u snaplen %u mac %u net %u" + "\n%Usec 0x%x nsec 0x%x vlan %U" +#ifdef TP_STATUS_VLAN_TPID_VALID + " vlan_tpid %u" +#endif + , + format_white_space, indent + 2, format_white_space, indent + 4, + t->tph3.tp_status, t->tph3.tp_len, t->tph3.tp_snaplen, t->tph3.tp_mac, + t->tph3.tp_net, format_white_space, indent + 4, t->tph3.tp_sec, + t->tph3.tp_nsec, format_ethernet_vlan_tci, t->tph3.hv1.tp_vlan_tci +#ifdef TP_STATUS_VLAN_TPID_VALID + , + t->tph3.hv1.tp_vlan_tpid +#endif + ); + } + else + { + s = format ( + s, + "\n%Utpacket2_hdr:\n%Ustatus 0x%x len %u snaplen %u mac %u net %u" + "\n%Usec 0x%x nsec 0x%x vlan %U" +#ifdef TP_STATUS_VLAN_TPID_VALID + " vlan_tpid %u" +#endif + , + format_white_space, indent + 2, format_white_space, indent + 4, + t->tph2.tp_status, t->tph2.tp_len, t->tph2.tp_snaplen, t->tph2.tp_mac, + t->tph2.tp_net, format_white_space, indent + 4, t->tph2.tp_sec, + t->tph2.tp_nsec, format_ethernet_vlan_tci, t->tph2.tp_vlan_tci +#ifdef TP_STATUS_VLAN_TPID_VALID + , + t->tph2.tp_vlan_tpid +#endif + ); + } + + s = format (s, + "\n%Uvnet-hdr:\n%Uflags 0x%02x gso_type 0x%02x hdr_len %u" + "\n%Ugso_size %u csum_start %u csum_offset %u", + format_white_space, indent + 2, format_white_space, indent + 4, + t->vnet_hdr.flags, t->vnet_hdr.gso_type, t->vnet_hdr.hdr_len, + format_white_space, indent + 4, t->vnet_hdr.gso_size, + t->vnet_hdr.csum_start, t->vnet_hdr.csum_offset); + return s; +} + +always_inline void +buffer_add_to_chain (vlib_buffer_t *b, vlib_buffer_t *first_b, + vlib_buffer_t *prev_b, u32 bi) +{ + /* update first buffer */ + first_b->total_length_not_including_first_buffer += b->current_length; + + /* update previous buffer */ + prev_b->next_buffer = bi; + prev_b->flags |= VLIB_BUFFER_NEXT_PRESENT; + + /* update current buffer */ + b->next_buffer = ~0; +} + +static_always_inline void +fill_gso_offload (vlib_buffer_t *b, u32 gso_size, u8 l4_hdr_sz) +{ + b->flags |= VNET_BUFFER_F_GSO; + vnet_buffer2 (b)->gso_size = gso_size; + vnet_buffer2 (b)->gso_l4_hdr_sz = l4_hdr_sz; +} + +static_always_inline void +fill_cksum_offload (vlib_buffer_t *b, u8 *l4_hdr_sz, u8 is_ip) +{ + vnet_buffer_oflags_t oflags = 0; + u16 l2hdr_sz = 0; + u16 ethertype = 0; + u8 l4_proto = 0; + + if (is_ip) + { + switch (b->data[0] & 0xf0) + { + case 0x40: + ethertype = ETHERNET_TYPE_IP4; + break; + case 0x60: + ethertype = ETHERNET_TYPE_IP6; + break; + } + } + else + { + ethernet_header_t *eth = (ethernet_header_t *) b->data; + ethertype = clib_net_to_host_u16 (eth->type); + l2hdr_sz = sizeof (ethernet_header_t); + if (ethernet_frame_is_tagged (ethertype)) + { + ethernet_vlan_header_t *vlan = (ethernet_vlan_header_t *) (eth + 1); + + ethertype = clib_net_to_host_u16 (vlan->type); + l2hdr_sz += sizeof (*vlan); + if (ethertype == ETHERNET_TYPE_VLAN) + { + vlan++; + ethertype = clib_net_to_host_u16 (vlan->type); + l2hdr_sz += sizeof (*vlan); + } + } + } + + vnet_buffer (b)->l2_hdr_offset = 0; + vnet_buffer (b)->l3_hdr_offset = l2hdr_sz; + + if (ethertype == ETHERNET_TYPE_IP4) + { + ip4_header_t *ip4 = (ip4_header_t *) (b->data + l2hdr_sz); + vnet_buffer (b)->l4_hdr_offset = l2hdr_sz + ip4_header_bytes (ip4); + b->flags |= (VNET_BUFFER_F_IS_IP4 | VNET_BUFFER_F_L2_HDR_OFFSET_VALID | + VNET_BUFFER_F_L3_HDR_OFFSET_VALID | + VNET_BUFFER_F_L4_HDR_OFFSET_VALID); + + l4_proto = ip4->protocol; + } + else if (ethertype == ETHERNET_TYPE_IP6) + { + ip6_header_t *ip6 = (ip6_header_t *) (b->data + l2hdr_sz); + b->flags |= (VNET_BUFFER_F_IS_IP6 | VNET_BUFFER_F_L2_HDR_OFFSET_VALID | + VNET_BUFFER_F_L3_HDR_OFFSET_VALID | + VNET_BUFFER_F_L4_HDR_OFFSET_VALID); + u16 ip6_hdr_len = sizeof (ip6_header_t); + + if (ip6_ext_hdr (ip6->protocol)) + { + ip6_ext_header_t *p = (void *) (ip6 + 1); + ip6_hdr_len += ip6_ext_header_len (p); + while (ip6_ext_hdr (p->next_hdr)) + { + ip6_hdr_len += ip6_ext_header_len (p); + p = ip6_ext_next_header (p); + } + l4_proto = p->next_hdr; + } + else + l4_proto = ip6->protocol; + vnet_buffer (b)->l4_hdr_offset = l2hdr_sz + ip6_hdr_len; + } + + if (l4_proto == IP_PROTOCOL_TCP) + { + oflags |= VNET_BUFFER_OFFLOAD_F_TCP_CKSUM; + tcp_header_t *tcp = + (tcp_header_t *) (b->data + vnet_buffer (b)->l4_hdr_offset); + *l4_hdr_sz = tcp_header_bytes (tcp); + } + else if (l4_proto == IP_PROTOCOL_UDP) + { + oflags |= VNET_BUFFER_OFFLOAD_F_UDP_CKSUM; + *l4_hdr_sz = sizeof (udp_header_t); + } + + if (oflags) + vnet_buffer_offload_flags_set (b, oflags); +} + +always_inline uword +af_packet_v3_device_input_fn (vlib_main_t *vm, vlib_node_runtime_t *node, + vlib_frame_t *frame, af_packet_if_t *apif, + u16 queue_id, u8 is_cksum_gso_enabled) +{ + af_packet_main_t *apm = &af_packet_main; + af_packet_queue_t *rx_queue = vec_elt_at_index (apif->rx_queues, queue_id); + tpacket3_hdr_t *tph; + u32 next_index; + u32 n_free_bufs; + u32 n_rx_packets = 0; + u32 n_rx_bytes = 0; + u32 timedout_blk = 0; + u32 total = 0; + u32 *to_next = 0; + u32 block = rx_queue->next_rx_block; + u32 block_nr = rx_queue->rx_req->req3.tp_block_nr; + u8 *block_start = 0; + uword n_trace = vlib_get_trace_count (vm, node); + u32 thread_index = vm->thread_index; + u32 n_buffer_bytes = vlib_buffer_get_default_data_size (vm); + u32 min_bufs = rx_queue->rx_req->req3.tp_frame_size / n_buffer_bytes; + u32 num_pkts = 0; + u32 rx_frame_offset = 0; + block_desc_t *bd = 0; + u32 sw_if_index = apif->sw_if_index; + u8 is_ip = (apif->mode == AF_PACKET_IF_MODE_IP); + + if (is_ip) + next_index = VNET_DEVICE_INPUT_NEXT_IP4_INPUT; + else + next_index = VNET_DEVICE_INPUT_NEXT_ETHERNET_INPUT; + + if ((((block_desc_t *) (block_start = rx_queue->rx_ring[block])) + ->hdr.bh1.block_status & + TP_STATUS_USER) != 0) + { + u32 n_required = 0; + bd = (block_desc_t *) block_start; + + if (PREDICT_FALSE (rx_queue->is_rx_pending)) + { + num_pkts = rx_queue->num_rx_pkts; + rx_frame_offset = rx_queue->rx_frame_offset; + rx_queue->is_rx_pending = 0; + } + else + { + num_pkts = bd->hdr.bh1.num_pkts; + rx_frame_offset = bd->hdr.bh1.offset_to_first_pkt; + total++; + + if (TP_STATUS_BLK_TMO & bd->hdr.bh1.block_status) + timedout_blk++; + } + + n_required = clib_max (num_pkts, VLIB_FRAME_SIZE); + n_free_bufs = vec_len (apm->rx_buffers[thread_index]); + if (PREDICT_FALSE (n_free_bufs < n_required)) + { + vec_validate (apm->rx_buffers[thread_index], + n_required + n_free_bufs - 1); + n_free_bufs += vlib_buffer_alloc ( + vm, &apm->rx_buffers[thread_index][n_free_bufs], n_required); + vec_set_len (apm->rx_buffers[thread_index], n_free_bufs); + } + + while (num_pkts && (n_free_bufs >= min_bufs)) + { + u32 next0 = next_index; + u32 n_left_to_next; + + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (num_pkts && n_left_to_next && (n_free_bufs >= min_bufs)) + { + tph = (tpacket3_hdr_t *) (block_start + rx_frame_offset); + + if (num_pkts > 1) + CLIB_PREFETCH (block_start + rx_frame_offset + + tph->tp_next_offset, + 2 * CLIB_CACHE_LINE_BYTES, LOAD); + + vlib_buffer_t *b0 = 0, *first_b0 = 0, *prev_b0 = 0; + vnet_virtio_net_hdr_t *vnet_hdr = 0; + u32 data_len = tph->tp_snaplen; + u32 offset = 0; + u32 bi0 = ~0, first_bi0 = ~0; + u8 l4_hdr_sz = 0; + + if (is_cksum_gso_enabled) + vnet_hdr = + (vnet_virtio_net_hdr_t *) ((u8 *) tph + tph->tp_mac - + sizeof (vnet_virtio_net_hdr_t)); + + // save current state and return + if (PREDICT_FALSE (((data_len / n_buffer_bytes) + 1) > + vec_len (apm->rx_buffers[thread_index]))) + { + rx_queue->rx_frame_offset = rx_frame_offset; + rx_queue->num_rx_pkts = num_pkts; + rx_queue->is_rx_pending = 1; + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + goto done; + } + + while (data_len) + { + /* grab free buffer */ + u32 last_empty_buffer = + vec_len (apm->rx_buffers[thread_index]) - 1; + bi0 = apm->rx_buffers[thread_index][last_empty_buffer]; + vec_set_len (apm->rx_buffers[thread_index], + last_empty_buffer); + n_free_bufs--; + + /* copy data */ + u32 bytes_to_copy = + data_len > n_buffer_bytes ? n_buffer_bytes : data_len; + u32 vlan_len = 0; + u32 bytes_copied = 0; + + b0 = vlib_get_buffer (vm, bi0); + b0->current_data = 0; + + /* Kernel removes VLAN headers, so reconstruct VLAN */ + if (PREDICT_FALSE (tph->tp_status & TP_STATUS_VLAN_VALID)) + { + if (PREDICT_TRUE (offset == 0)) + { + clib_memcpy_fast (vlib_buffer_get_current (b0), + (u8 *) tph + tph->tp_mac, + sizeof (ethernet_header_t)); + ethernet_header_t *eth = + vlib_buffer_get_current (b0); + ethernet_vlan_header_t *vlan = + (ethernet_vlan_header_t *) (eth + 1); + vlan->priority_cfi_and_id = + clib_host_to_net_u16 (tph->hv1.tp_vlan_tci); + vlan->type = eth->type; + eth->type = + clib_host_to_net_u16 (ETHERNET_TYPE_VLAN); + vlan_len = sizeof (ethernet_vlan_header_t); + bytes_copied = sizeof (ethernet_header_t); + } + } + clib_memcpy_fast (((u8 *) vlib_buffer_get_current (b0)) + + bytes_copied + vlan_len, + (u8 *) tph + tph->tp_mac + offset + + bytes_copied, + (bytes_to_copy - bytes_copied)); + + /* fill buffer header */ + b0->current_length = bytes_to_copy + vlan_len; + + if (offset == 0) + { + b0->total_length_not_including_first_buffer = 0; + b0->flags = VLIB_BUFFER_TOTAL_LENGTH_VALID; + vnet_buffer (b0)->sw_if_index[VLIB_RX] = sw_if_index; + vnet_buffer (b0)->sw_if_index[VLIB_TX] = (u32) ~0; + first_b0 = b0; + first_bi0 = bi0; + if (is_cksum_gso_enabled) + { + if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) + fill_cksum_offload (first_b0, &l4_hdr_sz, is_ip); + if (vnet_hdr->gso_type & (VIRTIO_NET_HDR_GSO_TCPV4 | + VIRTIO_NET_HDR_GSO_TCPV6)) + fill_gso_offload (first_b0, vnet_hdr->gso_size, + l4_hdr_sz); + } + } + else + buffer_add_to_chain (b0, first_b0, prev_b0, bi0); + + prev_b0 = b0; + offset += bytes_to_copy; + data_len -= bytes_to_copy; + } + n_rx_packets++; + n_rx_bytes += tph->tp_snaplen; + to_next[0] = first_bi0; + to_next += 1; + n_left_to_next--; + + /* drop partial packets */ + if (PREDICT_FALSE (tph->tp_len != tph->tp_snaplen)) + { + next0 = VNET_DEVICE_INPUT_NEXT_DROP; + first_b0->error = + node->errors[AF_PACKET_INPUT_ERROR_PARTIAL_PKT]; + } + else + { + if (PREDICT_FALSE (apif->mode == AF_PACKET_IF_MODE_IP)) + { + switch (first_b0->data[0] & 0xf0) + { + case 0x40: + next0 = VNET_DEVICE_INPUT_NEXT_IP4_INPUT; + break; + case 0x60: + next0 = VNET_DEVICE_INPUT_NEXT_IP6_INPUT; + break; + default: + next0 = VNET_DEVICE_INPUT_NEXT_DROP; + break; + } + if (PREDICT_FALSE (apif->per_interface_next_index != ~0)) + next0 = apif->per_interface_next_index; + } + else + { + next0 = VNET_DEVICE_INPUT_NEXT_ETHERNET_INPUT; + if (PREDICT_FALSE (apif->per_interface_next_index != ~0)) + next0 = apif->per_interface_next_index; + /* redirect if feature path enabled */ + vnet_feature_start_device_input (sw_if_index, &next0, + first_b0); + } + } + + /* trace */ + if (PREDICT_FALSE (n_trace > 0 && + vlib_trace_buffer (vm, node, next0, first_b0, + /* follow_chain */ 0))) + { + af_packet_input_trace_t *tr; + vlib_set_trace_count (vm, node, --n_trace); + tr = vlib_add_trace (vm, node, first_b0, sizeof (*tr)); + tr->is_v3 = 1; + tr->next_index = next0; + tr->hw_if_index = apif->hw_if_index; + tr->queue_id = queue_id; + tr->block = block; + tr->block_start = bd; + tr->pkt_num = bd->hdr.bh1.num_pkts - num_pkts; + clib_memcpy_fast (&tr->bd, bd, sizeof (block_desc_t)); + clib_memcpy_fast (&tr->tph3, tph, sizeof (tpacket3_hdr_t)); + if (is_cksum_gso_enabled) + clib_memcpy_fast (&tr->vnet_hdr, vnet_hdr, + sizeof (vnet_virtio_net_hdr_t)); + else + clib_memset_u8 (&tr->vnet_hdr, 0, + sizeof (vnet_virtio_net_hdr_t)); + } + + /* enque and take next packet */ + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, first_bi0, + next0); + + /* next packet */ + num_pkts--; + rx_frame_offset += tph->tp_next_offset; + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + if (PREDICT_TRUE (num_pkts == 0)) + { + bd->hdr.bh1.block_status = TP_STATUS_KERNEL; + block = (block + 1) % block_nr; + } + else + { + rx_queue->rx_frame_offset = rx_frame_offset; + rx_queue->num_rx_pkts = num_pkts; + rx_queue->is_rx_pending = 1; + } + } + + rx_queue->next_rx_block = block; + +done: + + if (apm->polling_count == 0) + { + if ((((block_desc_t *) (block_start = rx_queue->rx_ring[block])) + ->hdr.bh1.block_status & + TP_STATUS_USER) != 0) + vlib_node_set_state (vm, node->node_index, VLIB_NODE_STATE_POLLING); + else + vlib_node_set_state (vm, node->node_index, VLIB_NODE_STATE_INTERRUPT); + } + + vlib_error_count (vm, node->node_index, AF_PACKET_INPUT_ERROR_TOTAL_RECV_BLK, + total); + vlib_error_count (vm, node->node_index, AF_PACKET_INPUT_ERROR_TIMEDOUT_BLK, + timedout_blk); + + vlib_increment_combined_counter + (vnet_get_main ()->interface_main.combined_sw_if_counters + + VNET_INTERFACE_COUNTER_RX, + vlib_get_thread_index (), apif->hw_if_index, n_rx_packets, n_rx_bytes); + + vnet_device_increment_rx_packets (thread_index, n_rx_packets); + return n_rx_packets; +} + +always_inline uword +af_packet_v2_device_input_fn (vlib_main_t *vm, vlib_node_runtime_t *node, + vlib_frame_t *frame, af_packet_if_t *apif, + u16 queue_id, u8 is_cksum_gso_enabled) +{ + af_packet_main_t *apm = &af_packet_main; + af_packet_queue_t *rx_queue = vec_elt_at_index (apif->rx_queues, queue_id); + tpacket2_hdr_t *tph; + u32 next_index; + u32 block = 0; + u32 rx_frame; + u32 n_free_bufs; + u32 n_rx_packets = 0; + u32 n_rx_bytes = 0; + u32 *to_next = 0; + u32 frame_size = rx_queue->rx_req->req.tp_frame_size; + u32 frame_num = rx_queue->rx_req->req.tp_frame_nr; + u8 *block_start = rx_queue->rx_ring[block]; + uword n_trace = vlib_get_trace_count (vm, node); + u32 thread_index = vm->thread_index; + u32 n_buffer_bytes = vlib_buffer_get_default_data_size (vm); + u32 min_bufs = rx_queue->rx_req->req.tp_frame_size / n_buffer_bytes; + u32 sw_if_index = apif->sw_if_index; + u8 is_ip = (apif->mode == AF_PACKET_IF_MODE_IP); + + if (is_ip) + next_index = VNET_DEVICE_INPUT_NEXT_IP4_INPUT; + else + next_index = VNET_DEVICE_INPUT_NEXT_ETHERNET_INPUT; + + n_free_bufs = vec_len (apm->rx_buffers[thread_index]); + if (PREDICT_FALSE (n_free_bufs < VLIB_FRAME_SIZE)) + { + vec_validate (apm->rx_buffers[thread_index], + VLIB_FRAME_SIZE + n_free_bufs - 1); + n_free_bufs += vlib_buffer_alloc ( + vm, &apm->rx_buffers[thread_index][n_free_bufs], VLIB_FRAME_SIZE); + vec_set_len (apm->rx_buffers[thread_index], n_free_bufs); + } + + rx_frame = rx_queue->next_rx_frame; + tph = (tpacket2_hdr_t *) (block_start + rx_frame * frame_size); + while ((tph->tp_status & TP_STATUS_USER) && (n_free_bufs > min_bufs)) + { + vlib_buffer_t *b0 = 0, *first_b0 = 0, *prev_b0 = 0; + u32 next0 = next_index; + + u32 n_left_to_next; + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + while ((tph->tp_status & TP_STATUS_USER) && (n_free_bufs > min_bufs) && + n_left_to_next) + { + vnet_virtio_net_hdr_t *vnet_hdr = 0; + u32 data_len = tph->tp_snaplen; + u32 offset = 0; + u32 bi0 = 0, first_bi0 = 0; + u8 l4_hdr_sz = 0; + + if (is_cksum_gso_enabled) + vnet_hdr = + (vnet_virtio_net_hdr_t *) ((u8 *) tph + tph->tp_mac - + sizeof (vnet_virtio_net_hdr_t)); + while (data_len) + { + /* grab free buffer */ + u32 last_empty_buffer = + vec_len (apm->rx_buffers[thread_index]) - 1; + bi0 = apm->rx_buffers[thread_index][last_empty_buffer]; + b0 = vlib_get_buffer (vm, bi0); + vec_set_len (apm->rx_buffers[thread_index], last_empty_buffer); + n_free_bufs--; + + /* copy data */ + u32 bytes_to_copy = + data_len > n_buffer_bytes ? n_buffer_bytes : data_len; + u32 vlan_len = 0; + u32 bytes_copied = 0; + b0->current_data = 0; + /* Kernel removes VLAN headers, so reconstruct VLAN */ + if (PREDICT_FALSE (tph->tp_status & TP_STATUS_VLAN_VALID)) + { + if (PREDICT_TRUE (offset == 0)) + { + clib_memcpy_fast (vlib_buffer_get_current (b0), + (u8 *) tph + tph->tp_mac, + sizeof (ethernet_header_t)); + ethernet_header_t *eth = vlib_buffer_get_current (b0); + ethernet_vlan_header_t *vlan = + (ethernet_vlan_header_t *) (eth + 1); + vlan->priority_cfi_and_id = + clib_host_to_net_u16 (tph->tp_vlan_tci); + vlan->type = eth->type; + eth->type = clib_host_to_net_u16 (ETHERNET_TYPE_VLAN); + vlan_len = sizeof (ethernet_vlan_header_t); + bytes_copied = sizeof (ethernet_header_t); + } + } + clib_memcpy_fast (((u8 *) vlib_buffer_get_current (b0)) + + bytes_copied + vlan_len, + (u8 *) tph + tph->tp_mac + offset + + bytes_copied, + (bytes_to_copy - bytes_copied)); + + /* fill buffer header */ + b0->current_length = bytes_to_copy + vlan_len; + + if (offset == 0) + { + b0->total_length_not_including_first_buffer = 0; + b0->flags = VLIB_BUFFER_TOTAL_LENGTH_VALID; + vnet_buffer (b0)->sw_if_index[VLIB_RX] = sw_if_index; + vnet_buffer (b0)->sw_if_index[VLIB_TX] = (u32) ~0; + first_bi0 = bi0; + first_b0 = vlib_get_buffer (vm, first_bi0); + + if (is_cksum_gso_enabled) + { + if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) + fill_cksum_offload (first_b0, &l4_hdr_sz, is_ip); + if (vnet_hdr->gso_type & (VIRTIO_NET_HDR_GSO_TCPV4 | + VIRTIO_NET_HDR_GSO_TCPV6)) + fill_gso_offload (first_b0, vnet_hdr->gso_size, + l4_hdr_sz); + } + } + else + buffer_add_to_chain (b0, first_b0, prev_b0, bi0); + + prev_b0 = b0; + offset += bytes_to_copy; + data_len -= bytes_to_copy; + } + n_rx_packets++; + n_rx_bytes += tph->tp_snaplen; + to_next[0] = first_bi0; + to_next += 1; + n_left_to_next--; + + /* drop partial packets */ + if (PREDICT_FALSE (tph->tp_len != tph->tp_snaplen)) + { + next0 = VNET_DEVICE_INPUT_NEXT_DROP; + first_b0->error = + node->errors[AF_PACKET_INPUT_ERROR_PARTIAL_PKT]; + } + else + { + if (PREDICT_FALSE (is_ip)) + { + switch (first_b0->data[0] & 0xf0) + { + case 0x40: + next0 = VNET_DEVICE_INPUT_NEXT_IP4_INPUT; + break; + case 0x60: + next0 = VNET_DEVICE_INPUT_NEXT_IP6_INPUT; + break; + default: + next0 = VNET_DEVICE_INPUT_NEXT_DROP; + break; + } + if (PREDICT_FALSE (apif->per_interface_next_index != ~0)) + next0 = apif->per_interface_next_index; + } + else + { + next0 = VNET_DEVICE_INPUT_NEXT_ETHERNET_INPUT; + if (PREDICT_FALSE (apif->per_interface_next_index != ~0)) + next0 = apif->per_interface_next_index; + /* redirect if feature path enabled */ + vnet_feature_start_device_input (sw_if_index, &next0, + first_b0); + } + } + + /* trace */ + if (PREDICT_FALSE (n_trace > 0 && + vlib_trace_buffer (vm, node, next0, first_b0, + /* follow_chain */ 0))) + { + af_packet_input_trace_t *tr; + vlib_set_trace_count (vm, node, --n_trace); + tr = vlib_add_trace (vm, node, first_b0, sizeof (*tr)); + tr->is_v3 = 0; + tr->next_index = next0; + tr->hw_if_index = apif->hw_if_index; + tr->queue_id = queue_id; + clib_memcpy_fast (&tr->tph2, tph, sizeof (struct tpacket2_hdr)); + if (is_cksum_gso_enabled) + clib_memcpy_fast (&tr->vnet_hdr, vnet_hdr, + sizeof (vnet_virtio_net_hdr_t)); + else + clib_memset_u8 (&tr->vnet_hdr, 0, + sizeof (vnet_virtio_net_hdr_t)); + } + + /* enque and take next packet */ + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, + n_left_to_next, first_bi0, next0); + + /* next packet */ + tph->tp_status = TP_STATUS_KERNEL; + rx_frame = (rx_frame + 1) % frame_num; + tph = (struct tpacket2_hdr *) (block_start + rx_frame * frame_size); + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + rx_queue->next_rx_frame = rx_frame; + + vlib_increment_combined_counter ( + vnet_get_main ()->interface_main.combined_sw_if_counters + + VNET_INTERFACE_COUNTER_RX, + vlib_get_thread_index (), apif->hw_if_index, n_rx_packets, n_rx_bytes); + + vnet_device_increment_rx_packets (thread_index, n_rx_packets); + return n_rx_packets; +} + +always_inline uword +af_packet_device_input_fn (vlib_main_t *vm, vlib_node_runtime_t *node, + vlib_frame_t *frame, af_packet_if_t *apif, + u16 queue_id, u8 is_cksum_gso_enabled) + +{ + if (apif->version == TPACKET_V3) + return af_packet_v3_device_input_fn (vm, node, frame, apif, queue_id, + is_cksum_gso_enabled); + else + return af_packet_v2_device_input_fn (vm, node, frame, apif, queue_id, + is_cksum_gso_enabled); +} + +VLIB_NODE_FN (af_packet_input_node) (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * frame) +{ + u32 n_rx_packets = 0; + af_packet_main_t *apm = &af_packet_main; + vnet_hw_if_rxq_poll_vector_t *pv; + pv = vnet_hw_if_get_rxq_poll_vector (vm, node); + for (int i = 0; i < vec_len (pv); i++) + { + af_packet_if_t *apif; + apif = vec_elt_at_index (apm->interfaces, pv[i].dev_instance); + if (apif->is_admin_up) + { + if (apif->is_cksum_gso_enabled) + n_rx_packets += af_packet_device_input_fn (vm, node, frame, apif, + pv[i].queue_id, 1); + else + n_rx_packets += af_packet_device_input_fn (vm, node, frame, apif, + pv[i].queue_id, 0); + } + } + return n_rx_packets; +} + +VLIB_REGISTER_NODE (af_packet_input_node) = { + .name = "af-packet-input", + .flags = VLIB_NODE_FLAG_TRACE_SUPPORTED, + .sibling_of = "device-input", + .format_trace = format_af_packet_input_trace, + .type = VLIB_NODE_TYPE_INPUT, + .state = VLIB_NODE_STATE_INTERRUPT, + .n_errors = AF_PACKET_INPUT_N_ERROR, + .error_strings = af_packet_input_error_strings, +}; + + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/plugins/af_packet/plugin.c b/src/plugins/af_packet/plugin.c new file mode 100644 index 00000000000..ddad52404c7 --- /dev/null +++ b/src/plugins/af_packet/plugin.c @@ -0,0 +1,12 @@ +/* SPDX-License-Identifier: Apache-2.0 + * Copyright (c) 2022 Cisco Systems, Inc. + */ + +#include <vlib/vlib.h> +#include <vnet/plugin/plugin.h> +#include <vpp/app/version.h> + +VLIB_PLUGIN_REGISTER () = { + .version = VPP_BUILD_VER, + .description = "af-packet", +}; |