xref: /linux/include/net/pkt_sched.h (revision 0e50474fa514822e9d990874e554bf8043a201d7)
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 /*
29    Timer resolution MUST BE < 10% of min_schedulable_packet_size/bandwidth
30 
31    Normal IP packet size ~ 512byte, hence:
32 
33    0.5Kbyte/1Mbyte/sec = 0.5msec, so that we need 50usec timer for
34    10Mbit ethernet.
35 
36    10msec resolution -> <50Kbit/sec.
37 
38    The result: [34]86 is not good choice for QoS router :-(
39 
40    The things are not so bad, because we may use artificial
41    clock evaluated by integration of network data flow
42    in the most critical places.
43  */
44 
45 typedef u64	psched_time_t;
46 
47 /* Avoid doing 64 bit divide */
48 #define PSCHED_SHIFT			6
49 #define PSCHED_TICKS2NS(x)		((s64)(x) << PSCHED_SHIFT)
50 #define PSCHED_NS2TICKS(x)		((x) >> PSCHED_SHIFT)
51 
52 #define PSCHED_TICKS_PER_SEC		PSCHED_NS2TICKS(NSEC_PER_SEC)
53 #define PSCHED_PASTPERFECT		0
54 
55 static inline psched_time_t psched_get_time(void)
56 {
57 	return PSCHED_NS2TICKS(ktime_get_ns());
58 }
59 
60 struct qdisc_watchdog {
61 	struct hrtimer	timer;
62 	struct Qdisc	*qdisc;
63 };
64 
65 void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
66 				 clockid_t clockid);
67 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc);
68 
69 void qdisc_watchdog_schedule_range_ns(struct qdisc_watchdog *wd, u64 expires,
70 				      u64 delta_ns);
71 
72 static inline void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd,
73 					      u64 expires)
74 {
75 	return qdisc_watchdog_schedule_range_ns(wd, expires, 0ULL);
76 }
77 
78 static inline void qdisc_watchdog_schedule(struct qdisc_watchdog *wd,
79 					   psched_time_t expires)
80 {
81 	qdisc_watchdog_schedule_ns(wd, PSCHED_TICKS2NS(expires));
82 }
83 
84 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd);
85 
86 extern struct Qdisc_ops pfifo_qdisc_ops;
87 extern struct Qdisc_ops bfifo_qdisc_ops;
88 extern struct Qdisc_ops pfifo_head_drop_qdisc_ops;
89 
90 int fifo_set_limit(struct Qdisc *q, unsigned int limit);
91 struct Qdisc *fifo_create_dflt(struct Qdisc *sch, struct Qdisc_ops *ops,
92 			       unsigned int limit,
93 			       struct netlink_ext_ack *extack);
94 
95 int register_qdisc(struct Qdisc_ops *qops);
96 void unregister_qdisc(struct Qdisc_ops *qops);
97 #define NET_SCH_ALIAS_PREFIX "net-sch-"
98 #define MODULE_ALIAS_NET_SCH(id)	MODULE_ALIAS(NET_SCH_ALIAS_PREFIX id)
99 void qdisc_get_default(char *id, size_t len);
100 int qdisc_set_default(const char *id);
101 
102 void qdisc_hash_add(struct Qdisc *q, bool invisible);
103 void qdisc_hash_del(struct Qdisc *q);
104 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle);
105 struct Qdisc *qdisc_lookup_rcu(struct net_device *dev, u32 handle);
106 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
107 					struct nlattr *tab,
108 					struct netlink_ext_ack *extack);
109 void qdisc_put_rtab(struct qdisc_rate_table *tab);
110 void qdisc_put_stab(struct qdisc_size_table *tab);
111 bool sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
112 		     struct net_device *dev, struct netdev_queue *txq,
113 		     spinlock_t *root_lock, bool validate);
114 
115 void __qdisc_run(struct Qdisc *q);
116 
117 static inline void qdisc_run(struct Qdisc *q)
118 {
119 	if (qdisc_run_begin(q)) {
120 		__qdisc_run(q);
121 		qdisc_run_end(q);
122 	}
123 }
124 
125 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
126 
127 /* Calculate maximal size of packet seen by hard_start_xmit
128    routine of this device.
129  */
130 static inline unsigned int psched_mtu(const struct net_device *dev)
131 {
132 	return READ_ONCE(dev->mtu) + dev->hard_header_len;
133 }
134 
135 static inline struct net *qdisc_net(struct Qdisc *q)
136 {
137 	return dev_net(q->dev_queue->dev);
138 }
139 
140 struct tc_query_caps_base {
141 	enum tc_setup_type type;
142 	void *caps;
143 };
144 
145 struct tc_cbs_qopt_offload {
146 	u8 enable;
147 	s32 queue;
148 	s32 hicredit;
149 	s32 locredit;
150 	s32 idleslope;
151 	s32 sendslope;
152 };
153 
154 struct tc_etf_qopt_offload {
155 	u8 enable;
156 	s32 queue;
157 };
158 
159 struct tc_mqprio_caps {
160 	bool validate_queue_counts:1;
161 };
162 
163 struct tc_mqprio_qopt_offload {
164 	/* struct tc_mqprio_qopt must always be the first element */
165 	struct tc_mqprio_qopt qopt;
166 	struct netlink_ext_ack *extack;
167 	u16 mode;
168 	u16 shaper;
169 	u32 flags;
170 	u64 min_rate[TC_QOPT_MAX_QUEUE];
171 	u64 max_rate[TC_QOPT_MAX_QUEUE];
172 	unsigned long preemptible_tcs;
173 };
174 
175 struct tc_taprio_caps {
176 	bool supports_queue_max_sdu:1;
177 	bool gate_mask_per_txq:1;
178 	/* Device expects lower TXQ numbers to have higher priority over higher
179 	 * TXQs, regardless of their TC mapping. DO NOT USE FOR NEW DRIVERS,
180 	 * INSTEAD ENFORCE A PROPER TC:TXQ MAPPING COMING FROM USER SPACE.
181 	 */
182 	bool broken_mqprio:1;
183 };
184 
185 enum tc_taprio_qopt_cmd {
186 	TAPRIO_CMD_REPLACE,
187 	TAPRIO_CMD_DESTROY,
188 	TAPRIO_CMD_STATS,
189 	TAPRIO_CMD_QUEUE_STATS,
190 };
191 
192 /**
193  * struct tc_taprio_qopt_stats - IEEE 802.1Qbv statistics
194  * @window_drops: Frames that were dropped because they were too large to be
195  *	transmitted in any of the allotted time windows (open gates) for their
196  *	traffic class.
197  * @tx_overruns: Frames still being transmitted by the MAC after the
198  *	transmission gate associated with their traffic class has closed.
199  *	Equivalent to `12.29.1.1.2 TransmissionOverrun` from 802.1Q-2018.
200  */
201 struct tc_taprio_qopt_stats {
202 	u64 window_drops;
203 	u64 tx_overruns;
204 };
205 
206 struct tc_taprio_qopt_queue_stats {
207 	int queue;
208 	struct tc_taprio_qopt_stats stats;
209 };
210 
211 struct tc_taprio_sched_entry {
212 	u8 command; /* TC_TAPRIO_CMD_* */
213 
214 	/* The gate_mask in the offloading side refers to traffic classes */
215 	u32 gate_mask;
216 	u32 interval;
217 };
218 
219 struct tc_taprio_qopt_offload {
220 	enum tc_taprio_qopt_cmd cmd;
221 
222 	union {
223 		/* TAPRIO_CMD_STATS */
224 		struct tc_taprio_qopt_stats stats;
225 		/* TAPRIO_CMD_QUEUE_STATS */
226 		struct tc_taprio_qopt_queue_stats queue_stats;
227 		/* TAPRIO_CMD_REPLACE */
228 		struct {
229 			struct tc_mqprio_qopt_offload mqprio;
230 			struct netlink_ext_ack *extack;
231 			ktime_t base_time;
232 			u64 cycle_time;
233 			u64 cycle_time_extension;
234 			u32 max_sdu[TC_MAX_QUEUE];
235 
236 			size_t num_entries;
237 			struct tc_taprio_sched_entry entries[];
238 		};
239 	};
240 };
241 
242 #if IS_ENABLED(CONFIG_NET_SCH_TAPRIO)
243 
244 /* Reference counting */
245 struct tc_taprio_qopt_offload *taprio_offload_get(struct tc_taprio_qopt_offload
246 						  *offload);
247 void taprio_offload_free(struct tc_taprio_qopt_offload *offload);
248 
249 #else
250 
251 /* Reference counting */
252 static inline struct tc_taprio_qopt_offload *
253 taprio_offload_get(struct tc_taprio_qopt_offload *offload)
254 {
255 	return NULL;
256 }
257 
258 static inline void taprio_offload_free(struct tc_taprio_qopt_offload *offload)
259 {
260 }
261 
262 #endif
263 
264 /* Ensure skb_mstamp_ns, which might have been populated with the txtime, is
265  * not mistaken for a software timestamp, because this will otherwise prevent
266  * the dispatch of hardware timestamps to the socket.
267  */
268 static inline void skb_txtime_consumed(struct sk_buff *skb)
269 {
270 	skb->tstamp = ktime_set(0, 0);
271 }
272 
273 static inline bool tc_qdisc_stats_dump(struct Qdisc *sch,
274 				       unsigned long cl,
275 				       struct qdisc_walker *arg)
276 {
277 	if (arg->count >= arg->skip && arg->fn(sch, cl, arg) < 0) {
278 		arg->stop = 1;
279 		return false;
280 	}
281 
282 	arg->count++;
283 	return true;
284 }
285 
286 static inline void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
287 {
288 	if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
289 		pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
290 			txt, qdisc->ops->id, qdisc->handle >> 16);
291 		qdisc->flags |= TCQ_F_WARN_NONWC;
292 	}
293 }
294 
295 static inline unsigned int qdisc_peek_len(struct Qdisc *sch)
296 {
297 	struct sk_buff *skb;
298 	unsigned int len;
299 
300 	skb = sch->ops->peek(sch);
301 	if (unlikely(skb == NULL)) {
302 		qdisc_warn_nonwc("qdisc_peek_len", sch);
303 		return 0;
304 	}
305 	len = qdisc_pkt_len(skb);
306 
307 	return len;
308 }
309 
310 #endif
311