include/hw/virtio/virtio-net.h

   1 /*
   2  * Virtio Network Device
   3  *
   4  * Copyright IBM, Corp. 2007
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  */
  13
  14 #ifndef QEMU_VIRTIO_NET_H
  15 #define QEMU_VIRTIO_NET_H
  16
  17 #include "qemu/units.h"
  18 #include "standard-headers/linux/virtio_net.h"
  19 #include "hw/virtio/virtio.h"
  20 #include "net/announce.h"
  21 #include "qemu/option_int.h"
  22 #include "qom/object.h"
  23
  24 #include "ebpf/ebpf_rss.h"
  25
  26 #define TYPE_VIRTIO_NET "virtio-net-device"
  27 OBJECT_DECLARE_SIMPLE_TYPE(VirtIONet, VIRTIO_NET)
  28
  29 #define TX_TIMER_INTERVAL 150000 /* 150 us */
  30
  31 /* Limit the number of packets that can be sent via a single flush
  32  * of the TX queue.  This gives us a guaranteed exit condition and
  33  * ensures fairness in the io path.  256 conveniently matches the
  34  * length of the TX queue and shows a good balance of performance
  35  * and latency. */
  36 #define TX_BURST 256
  37
  38 typedef struct virtio_net_conf
  39 {
  40     uint32_t txtimer;
  41     int32_t txburst;
  42     char *tx;
  43     uint16_t rx_queue_size;
  44     uint16_t tx_queue_size;
  45     uint16_t mtu;
  46     int32_t speed;
  47     char *duplex_str;
  48     uint8_t duplex;
  49     char *primary_id_str;
  50 } virtio_net_conf;
  51
  52 /* Coalesced packets type & status */
  53 typedef enum {
  54     RSC_COALESCE,           /* Data been coalesced */
  55     RSC_FINAL,              /* Will terminate current connection */
  56     RSC_NO_MATCH,           /* No matched in the buffer pool */
  57     RSC_BYPASS,             /* Packet to be bypass, not tcp, tcp ctrl, etc */
  58     RSC_CANDIDATE                /* Data want to be coalesced */
  59 } CoalesceStatus;
  60
  61 typedef struct VirtioNetRscStat {
  62     uint32_t received;
  63     uint32_t coalesced;
  64     uint32_t over_size;
  65     uint32_t cache;
  66     uint32_t empty_cache;
  67     uint32_t no_match_cache;
  68     uint32_t win_update;
  69     uint32_t no_match;
  70     uint32_t tcp_syn;
  71     uint32_t tcp_ctrl_drain;
  72     uint32_t dup_ack;
  73     uint32_t dup_ack1;
  74     uint32_t dup_ack2;
  75     uint32_t pure_ack;
  76     uint32_t ack_out_of_win;
  77     uint32_t data_out_of_win;
  78     uint32_t data_out_of_order;
  79     uint32_t data_after_pure_ack;
  80     uint32_t bypass_not_tcp;
  81     uint32_t tcp_option;
  82     uint32_t tcp_all_opt;
  83     uint32_t ip_frag;
  84     uint32_t ip_ecn;
  85     uint32_t ip_hacked;
  86     uint32_t ip_option;
  87     uint32_t purge_failed;
  88     uint32_t drain_failed;
  89     uint32_t final_failed;
  90     int64_t  timer;
  91 } VirtioNetRscStat;
  92
  93 /* Rsc unit general info used to checking if can coalescing */
  94 typedef struct VirtioNetRscUnit {
  95     void *ip;   /* ip header */
  96     uint16_t *ip_plen;      /* data len pointer in ip header field */
  97     struct tcp_header *tcp; /* tcp header */
  98     uint16_t tcp_hdrlen;    /* tcp header len */
  99     uint16_t payload;       /* pure payload without virtio/eth/ip/tcp */
 100 } VirtioNetRscUnit;
 101
 102 /* Coalesced segment */
 103 typedef struct VirtioNetRscSeg {
 104     QTAILQ_ENTRY(VirtioNetRscSeg) next;
 105     void *buf;
 106     size_t size;
 107     uint16_t packets;
 108     uint16_t dup_ack;
 109     bool is_coalesced;      /* need recal ipv4 header checksum, mark here */
 110     VirtioNetRscUnit unit;
 111     NetClientState *nc;
 112 } VirtioNetRscSeg;
 113
 114
 115 /* Chain is divided by protocol(ipv4/v6) and NetClientInfo */
 116 typedef struct VirtioNetRscChain {
 117     QTAILQ_ENTRY(VirtioNetRscChain) next;
 118     VirtIONet *n;                            /* VirtIONet */
 119     uint16_t proto;
 120     uint8_t  gso_type;
 121     uint16_t max_payload;
 122     QEMUTimer *drain_timer;
 123     QTAILQ_HEAD(, VirtioNetRscSeg) buffers;
 124     VirtioNetRscStat stat;
 125 } VirtioNetRscChain;
 126
 127 /* Maximum packet size we can receive from tap device: header + 64k */
 128 #define VIRTIO_NET_MAX_BUFSIZE (sizeof(struct virtio_net_hdr) + (64 * KiB))
 129
 130 #define VIRTIO_NET_RSS_MAX_KEY_SIZE     40
 131 #define VIRTIO_NET_RSS_MAX_TABLE_LEN    128
 132
 133 typedef struct VirtioNetRssData {
 134     bool    enabled;
 135     bool    enabled_software_rss;
 136     bool    redirect;
 137     bool    populate_hash;
 138     uint32_t hash_types;
 139     uint8_t key[VIRTIO_NET_RSS_MAX_KEY_SIZE];
 140     uint16_t indirections_len;
 141     uint16_t *indirections_table;
 142     uint16_t default_queue;
 143 } VirtioNetRssData;
 144
 145 typedef struct VirtIONetQueue {
 146     VirtQueue *rx_vq;
 147     VirtQueue *tx_vq;
 148     QEMUTimer *tx_timer;
 149     QEMUBH *tx_bh;
 150     uint32_t tx_waiting;
 151     struct {
 152         VirtQueueElement *elem;
 153     } async_tx;
 154     struct VirtIONet *n;
 155 } VirtIONetQueue;
 156
 157 struct VirtIONet {
 158     VirtIODevice parent_obj;
 159     uint8_t mac[ETH_ALEN];
 160     uint16_t status;
 161     VirtIONetQueue *vqs;
 162     VirtQueue *ctrl_vq;
 163     NICState *nic;
 164     /* RSC Chains - temporary storage of coalesced data,
 165        all these data are lost in case of migration */
 166     QTAILQ_HEAD(, VirtioNetRscChain) rsc_chains;
 167     uint32_t tx_timeout;
 168     int32_t tx_burst;
 169     uint32_t has_vnet_hdr;
 170     size_t host_hdr_len;
 171     size_t guest_hdr_len;
 172     uint64_t host_features;
 173     uint32_t rsc_timeout;
 174     uint8_t rsc4_enabled;
 175     uint8_t rsc6_enabled;
 176     uint8_t has_ufo;
 177     uint32_t mergeable_rx_bufs;
 178     uint8_t promisc;
 179     uint8_t allmulti;
 180     uint8_t alluni;
 181     uint8_t nomulti;
 182     uint8_t nouni;
 183     uint8_t nobcast;
 184     uint8_t vhost_started;
 185     struct {
 186         uint32_t in_use;
 187         uint32_t first_multi;
 188         uint8_t multi_overflow;
 189         uint8_t uni_overflow;
 190         uint8_t *macs;
 191     } mac_table;
 192     uint32_t *vlans;
 193     virtio_net_conf net_conf;
 194     NICConf nic_conf;
 195     DeviceState *qdev;
 196     int multiqueue;
 197     uint16_t max_queues;
 198     uint16_t curr_queues;
 199     size_t config_size;
 200     char *netclient_name;
 201     char *netclient_type;
 202     uint64_t curr_guest_offloads;
 203     /* used on saved state restore phase to preserve the curr_guest_offloads */
 204     uint64_t saved_guest_offloads;
 205     AnnounceTimer announce_timer;
 206     bool needs_vnet_hdr_swap;
 207     bool mtu_bypass_backend;
 208     /* primary failover device is hidden*/
 209     bool failover_primary_hidden;
 210     bool failover;
 211     DeviceListener primary_listener;
 212     QDict *primary_opts;
 213     bool primary_opts_from_json;
 214     Notifier migration_state;
 215     VirtioNetRssData rss_data;
 216     struct NetRxPkt *rx_pkt;
 217     struct EBPFRSSContext ebpf_rss;
 218 };
 219
 220 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
 221                                    const char *type);
 222
 223 #endif