include/net/pkt_sched.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef __NET_PKT_SCHED_H
   3 #define __NET_PKT_SCHED_H
   4
   5 #include <linux/jiffies.h>
   6 #include <linux/ktime.h>
   7 #include <linux/if_vlan.h>
   8 #include <linux/netdevice.h>
   9 #include <net/sch_generic.h>
  10 #include <net/net_namespace.h>
  11 #include <uapi/linux/pkt_sched.h>
  12
  13 #define DEFAULT_TX_QUEUE_LEN    1000
  14 #define STAB_SIZE_LOG_MAX       30
  15
  16 struct qdisc_walker {
  17         int     stop;
  18         int     skip;
  19         int     count;
  20         int     (*fn)(struct Qdisc *, unsigned long cl, struct qdisc_walker *);
  21 };
  22
  23 #define qdisc_priv(q)                                                   \
  24         _Generic(q,                                                     \
  25                  const struct Qdisc * : (const void *)&q->privdata,     \
  26                  struct Qdisc * : (void *)&q->privdata)
  27
  28 static inline struct Qdisc *qdisc_from_priv(void *priv)
  29 {
  30         return container_of(priv, struct Qdisc, privdata);
  31 }
  32
  33 /*
  34    Timer resolution MUST BE < 10% of min_schedulable_packet_size/bandwidth
  35
  36    Normal IP packet size ~ 512byte, hence:
  37
  38    0.5Kbyte/1Mbyte/sec = 0.5msec, so that we need 50usec timer for
  39    10Mbit ethernet.
  40
  41    10msec resolution -> <50Kbit/sec.
  42
  43    The result: [34]86 is not good choice for QoS router :-(
  44
  45    The things are not so bad, because we may use artificial
  46    clock evaluated by integration of network data flow
  47    in the most critical places.
  48  */
  49
  50 typedef u64     psched_time_t;
  51 typedef long    psched_tdiff_t;
  52
  53 /* Avoid doing 64 bit divide */
  54 #define PSCHED_SHIFT                    6
  55 #define PSCHED_TICKS2NS(x)              ((s64)(x) << PSCHED_SHIFT)
  56 #define PSCHED_NS2TICKS(x)              ((x) >> PSCHED_SHIFT)
  57
  58 #define PSCHED_TICKS_PER_SEC            PSCHED_NS2TICKS(NSEC_PER_SEC)
  59 #define PSCHED_PASTPERFECT              0
  60
  61 static inline psched_time_t psched_get_time(void)
  62 {
  63         return PSCHED_NS2TICKS(ktime_get_ns());
  64 }
  65
  66 struct qdisc_watchdog {
  67         struct hrtimer  timer;
  68         struct Qdisc    *qdisc;
  69 };
  70
  71 void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
  72                                  clockid_t clockid);
  73 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc);
  74
  75 void qdisc_watchdog_schedule_range_ns(struct qdisc_watchdog *wd, u64 expires,
  76                                       u64 delta_ns);
  77
  78 static inline void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd,
  79                                               u64 expires)
  80 {
  81         return qdisc_watchdog_schedule_range_ns(wd, expires, 0ULL);
  82 }
  83
  84 static inline void qdisc_watchdog_schedule(struct qdisc_watchdog *wd,
  85                                            psched_time_t expires)
  86 {
  87         qdisc_watchdog_schedule_ns(wd, PSCHED_TICKS2NS(expires));
  88 }
  89
  90 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd);
  91
  92 extern struct Qdisc_ops pfifo_qdisc_ops;
  93 extern struct Qdisc_ops bfifo_qdisc_ops;
  94 extern struct Qdisc_ops pfifo_head_drop_qdisc_ops;
  95
  96 int fifo_set_limit(struct Qdisc *q, unsigned int limit);
  97 struct Qdisc *fifo_create_dflt(struct Qdisc *sch, struct Qdisc_ops *ops,
  98                                unsigned int limit,
  99                                struct netlink_ext_ack *extack);
 100
 101 int register_qdisc(struct Qdisc_ops *qops);
 102 void unregister_qdisc(struct Qdisc_ops *qops);
 103 #define NET_SCH_ALIAS_PREFIX "net-sch-"
 104 #define MODULE_ALIAS_NET_SCH(id)        MODULE_ALIAS(NET_SCH_ALIAS_PREFIX id)
 105 void qdisc_get_default(char *id, size_t len);
 106 int qdisc_set_default(const char *id);
 107
 108 void qdisc_hash_add(struct Qdisc *q, bool invisible);
 109 void qdisc_hash_del(struct Qdisc *q);
 110 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle);
 111 struct Qdisc *qdisc_lookup_rcu(struct net_device *dev, u32 handle);
 112 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
 113                                         struct nlattr *tab,
 114                                         struct netlink_ext_ack *extack);
 115 void qdisc_put_rtab(struct qdisc_rate_table *tab);
 116 void qdisc_put_stab(struct qdisc_size_table *tab);
 117 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc);
 118 bool sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
 119                      struct net_device *dev, struct netdev_queue *txq,
 120                      spinlock_t *root_lock, bool validate);
 121
 122 void __qdisc_run(struct Qdisc *q);
 123
 124 static inline void qdisc_run(struct Qdisc *q)
 125 {
 126         if (qdisc_run_begin(q)) {
 127                 __qdisc_run(q);
 128                 qdisc_run_end(q);
 129         }
 130 }
 131
 132 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
 133
 134 /* Calculate maximal size of packet seen by hard_start_xmit
 135    routine of this device.
 136  */
 137 static inline unsigned int psched_mtu(const struct net_device *dev)
 138 {
 139         return READ_ONCE(dev->mtu) + dev->hard_header_len;
 140 }
 141
 142 static inline struct net *qdisc_net(struct Qdisc *q)
 143 {
 144         return dev_net(q->dev_queue->dev);
 145 }
 146
 147 struct tc_query_caps_base {
 148         enum tc_setup_type type;
 149         void *caps;
 150 };
 151
 152 struct tc_cbs_qopt_offload {
 153         u8 enable;
 154         s32 queue;
 155         s32 hicredit;
 156         s32 locredit;
 157         s32 idleslope;
 158         s32 sendslope;
 159 };
 160
 161 struct tc_etf_qopt_offload {
 162         u8 enable;
 163         s32 queue;
 164 };
 165
 166 struct tc_mqprio_caps {
 167         bool validate_queue_counts:1;
 168 };
 169
 170 struct tc_mqprio_qopt_offload {
 171         /* struct tc_mqprio_qopt must always be the first element */
 172         struct tc_mqprio_qopt qopt;
 173         struct netlink_ext_ack *extack;
 174         u16 mode;
 175         u16 shaper;
 176         u32 flags;
 177         u64 min_rate[TC_QOPT_MAX_QUEUE];
 178         u64 max_rate[TC_QOPT_MAX_QUEUE];
 179         unsigned long preemptible_tcs;
 180 };
 181
 182 struct tc_taprio_caps {
 183         bool supports_queue_max_sdu:1;
 184         bool gate_mask_per_txq:1;
 185         /* Device expects lower TXQ numbers to have higher priority over higher
 186          * TXQs, regardless of their TC mapping. DO NOT USE FOR NEW DRIVERS,
 187          * INSTEAD ENFORCE A PROPER TC:TXQ MAPPING COMING FROM USER SPACE.
 188          */
 189         bool broken_mqprio:1;
 190 };
 191
 192 enum tc_taprio_qopt_cmd {
 193         TAPRIO_CMD_REPLACE,
 194         TAPRIO_CMD_DESTROY,
 195         TAPRIO_CMD_STATS,
 196         TAPRIO_CMD_QUEUE_STATS,
 197 };
 198
 199 /**
 200  * struct tc_taprio_qopt_stats - IEEE 802.1Qbv statistics
 201  * @window_drops: Frames that were dropped because they were too large to be
 202  *      transmitted in any of the allotted time windows (open gates) for their
 203  *      traffic class.
 204  * @tx_overruns: Frames still being transmitted by the MAC after the
 205  *      transmission gate associated with their traffic class has closed.
 206  *      Equivalent to `12.29.1.1.2 TransmissionOverrun` from 802.1Q-2018.
 207  */
 208 struct tc_taprio_qopt_stats {
 209         u64 window_drops;
 210         u64 tx_overruns;
 211 };
 212
 213 struct tc_taprio_qopt_queue_stats {
 214         int queue;
 215         struct tc_taprio_qopt_stats stats;
 216 };
 217
 218 struct tc_taprio_sched_entry {
 219         u8 command; /* TC_TAPRIO_CMD_* */
 220
 221         /* The gate_mask in the offloading side refers to traffic classes */
 222         u32 gate_mask;
 223         u32 interval;
 224 };
 225
 226 struct tc_taprio_qopt_offload {
 227         enum tc_taprio_qopt_cmd cmd;
 228
 229         union {
 230                 /* TAPRIO_CMD_STATS */
 231                 struct tc_taprio_qopt_stats stats;
 232                 /* TAPRIO_CMD_QUEUE_STATS */
 233                 struct tc_taprio_qopt_queue_stats queue_stats;
 234                 /* TAPRIO_CMD_REPLACE */
 235                 struct {
 236                         struct tc_mqprio_qopt_offload mqprio;
 237                         struct netlink_ext_ack *extack;
 238                         ktime_t base_time;
 239                         u64 cycle_time;
 240                         u64 cycle_time_extension;
 241                         u32 max_sdu[TC_MAX_QUEUE];
 242
 243                         size_t num_entries;
 244                         struct tc_taprio_sched_entry entries[];
 245                 };
 246         };
 247 };
 248
 249 #if IS_ENABLED(CONFIG_NET_SCH_TAPRIO)
 250
 251 /* Reference counting */
 252 struct tc_taprio_qopt_offload *taprio_offload_get(struct tc_taprio_qopt_offload
 253                                                   *offload);
 254 void taprio_offload_free(struct tc_taprio_qopt_offload *offload);
 255
 256 #else
 257
 258 /* Reference counting */
 259 static inline struct tc_taprio_qopt_offload *
 260 taprio_offload_get(struct tc_taprio_qopt_offload *offload)
 261 {
 262         return NULL;
 263 }
 264
 265 static inline void taprio_offload_free(struct tc_taprio_qopt_offload *offload)
 266 {
 267 }
 268
 269 #endif
 270
 271 /* Ensure skb_mstamp_ns, which might have been populated with the txtime, is
 272  * not mistaken for a software timestamp, because this will otherwise prevent
 273  * the dispatch of hardware timestamps to the socket.
 274  */
 275 static inline void skb_txtime_consumed(struct sk_buff *skb)
 276 {
 277         skb->tstamp = ktime_set(0, 0);
 278 }
 279
 280 static inline bool tc_qdisc_stats_dump(struct Qdisc *sch,
 281                                        unsigned long cl,
 282                                        struct qdisc_walker *arg)
 283 {
 284         if (arg->count >= arg->skip && arg->fn(sch, cl, arg) < 0) {
 285                 arg->stop = 1;
 286                 return false;
 287         }
 288
 289         arg->count++;
 290         return true;
 291 }
 292
 293 #endif