xref: /linux/include/net/pkt_sched.h (revision d8e473182ab9e85708067be81d20424045d939fa)
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 static inline void *qdisc_priv(struct Qdisc *q)
24 {
25 	return &q->privdata;
26 }
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 	u64		last_expires;
68 	struct hrtimer	timer;
69 	struct Qdisc	*qdisc;
70 };
71 
72 void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
73 				 clockid_t clockid);
74 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc);
75 
76 void qdisc_watchdog_schedule_range_ns(struct qdisc_watchdog *wd, u64 expires,
77 				      u64 delta_ns);
78 
79 static inline void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd,
80 					      u64 expires)
81 {
82 	return qdisc_watchdog_schedule_range_ns(wd, expires, 0ULL);
83 }
84 
85 static inline void qdisc_watchdog_schedule(struct qdisc_watchdog *wd,
86 					   psched_time_t expires)
87 {
88 	qdisc_watchdog_schedule_ns(wd, PSCHED_TICKS2NS(expires));
89 }
90 
91 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd);
92 
93 extern struct Qdisc_ops pfifo_qdisc_ops;
94 extern struct Qdisc_ops bfifo_qdisc_ops;
95 extern struct Qdisc_ops pfifo_head_drop_qdisc_ops;
96 
97 int fifo_set_limit(struct Qdisc *q, unsigned int limit);
98 struct Qdisc *fifo_create_dflt(struct Qdisc *sch, struct Qdisc_ops *ops,
99 			       unsigned int limit,
100 			       struct netlink_ext_ack *extack);
101 
102 int register_qdisc(struct Qdisc_ops *qops);
103 void unregister_qdisc(struct Qdisc_ops *qops);
104 void qdisc_get_default(char *id, size_t len);
105 int qdisc_set_default(const char *id);
106 
107 void qdisc_hash_add(struct Qdisc *q, bool invisible);
108 void qdisc_hash_del(struct Qdisc *q);
109 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle);
110 struct Qdisc *qdisc_lookup_rcu(struct net_device *dev, u32 handle);
111 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
112 					struct nlattr *tab,
113 					struct netlink_ext_ack *extack);
114 void qdisc_put_rtab(struct qdisc_rate_table *tab);
115 void qdisc_put_stab(struct qdisc_size_table *tab);
116 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc);
117 bool sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
118 		     struct net_device *dev, struct netdev_queue *txq,
119 		     spinlock_t *root_lock, bool validate);
120 
121 void __qdisc_run(struct Qdisc *q);
122 
123 static inline void qdisc_run(struct Qdisc *q)
124 {
125 	if (qdisc_run_begin(q)) {
126 		__qdisc_run(q);
127 		qdisc_run_end(q);
128 	}
129 }
130 
131 /* Calculate maximal size of packet seen by hard_start_xmit
132    routine of this device.
133  */
134 static inline unsigned int psched_mtu(const struct net_device *dev)
135 {
136 	return dev->mtu + dev->hard_header_len;
137 }
138 
139 static inline struct net *qdisc_net(struct Qdisc *q)
140 {
141 	return dev_net(q->dev_queue->dev);
142 }
143 
144 struct tc_query_caps_base {
145 	enum tc_setup_type type;
146 	void *caps;
147 };
148 
149 struct tc_cbs_qopt_offload {
150 	u8 enable;
151 	s32 queue;
152 	s32 hicredit;
153 	s32 locredit;
154 	s32 idleslope;
155 	s32 sendslope;
156 };
157 
158 struct tc_etf_qopt_offload {
159 	u8 enable;
160 	s32 queue;
161 };
162 
163 struct tc_mqprio_caps {
164 	bool validate_queue_counts:1;
165 };
166 
167 struct tc_mqprio_qopt_offload {
168 	/* struct tc_mqprio_qopt must always be the first element */
169 	struct tc_mqprio_qopt qopt;
170 	u16 mode;
171 	u16 shaper;
172 	u32 flags;
173 	u64 min_rate[TC_QOPT_MAX_QUEUE];
174 	u64 max_rate[TC_QOPT_MAX_QUEUE];
175 };
176 
177 struct tc_taprio_caps {
178 	bool supports_queue_max_sdu:1;
179 	bool gate_mask_per_txq:1;
180 	/* Device expects lower TXQ numbers to have higher priority over higher
181 	 * TXQs, regardless of their TC mapping. DO NOT USE FOR NEW DRIVERS,
182 	 * INSTEAD ENFORCE A PROPER TC:TXQ MAPPING COMING FROM USER SPACE.
183 	 */
184 	bool broken_mqprio:1;
185 };
186 
187 struct tc_taprio_sched_entry {
188 	u8 command; /* TC_TAPRIO_CMD_* */
189 
190 	/* The gate_mask in the offloading side refers to traffic classes */
191 	u32 gate_mask;
192 	u32 interval;
193 };
194 
195 struct tc_taprio_qopt_offload {
196 	struct tc_mqprio_qopt_offload mqprio;
197 	u8 enable;
198 	ktime_t base_time;
199 	u64 cycle_time;
200 	u64 cycle_time_extension;
201 	u32 max_sdu[TC_MAX_QUEUE];
202 
203 	size_t num_entries;
204 	struct tc_taprio_sched_entry entries[];
205 };
206 
207 #if IS_ENABLED(CONFIG_NET_SCH_TAPRIO)
208 
209 /* Reference counting */
210 struct tc_taprio_qopt_offload *taprio_offload_get(struct tc_taprio_qopt_offload
211 						  *offload);
212 void taprio_offload_free(struct tc_taprio_qopt_offload *offload);
213 
214 #else
215 
216 /* Reference counting */
217 static inline struct tc_taprio_qopt_offload *
218 taprio_offload_get(struct tc_taprio_qopt_offload *offload)
219 {
220 	return NULL;
221 }
222 
223 static inline void taprio_offload_free(struct tc_taprio_qopt_offload *offload)
224 {
225 }
226 
227 #endif
228 
229 /* Ensure skb_mstamp_ns, which might have been populated with the txtime, is
230  * not mistaken for a software timestamp, because this will otherwise prevent
231  * the dispatch of hardware timestamps to the socket.
232  */
233 static inline void skb_txtime_consumed(struct sk_buff *skb)
234 {
235 	skb->tstamp = ktime_set(0, 0);
236 }
237 
238 struct tc_skb_cb {
239 	struct qdisc_skb_cb qdisc_cb;
240 
241 	u16 mru;
242 	u8 post_ct:1;
243 	u8 post_ct_snat:1;
244 	u8 post_ct_dnat:1;
245 	u16 zone; /* Only valid if post_ct = true */
246 };
247 
248 static inline struct tc_skb_cb *tc_skb_cb(const struct sk_buff *skb)
249 {
250 	struct tc_skb_cb *cb = (struct tc_skb_cb *)skb->cb;
251 
252 	BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
253 	return cb;
254 }
255 
256 static inline bool tc_qdisc_stats_dump(struct Qdisc *sch,
257 				       unsigned long cl,
258 				       struct qdisc_walker *arg)
259 {
260 	if (arg->count >= arg->skip && arg->fn(sch, cl, arg) < 0) {
261 		arg->stop = 1;
262 		return false;
263 	}
264 
265 	arg->count++;
266 	return true;
267 }
268 
269 #endif
270