write in 30M

hc

2024-02-20 ea08eeccae9297f7aabd2ef7f0c2517ac4549acc

 kernel/net/sched/sch_etf.c
..
..
@@ -22,15 +22,17 @@
22
22
 
23
23
 #define DEADLINE_MODE_IS_ON(x) ((x)->flags & TC_ETF_DEADLINE_MODE_ON)
24
24
 #define OFFLOAD_IS_ON(x) ((x)->flags & TC_ETF_OFFLOAD_ON)
25
+#define SKIP_SOCK_CHECK_IS_SET(x) ((x)->flags & TC_ETF_SKIP_SOCK_CHECK)
25
26
 
26
27
 struct etf_sched_data {
27
28
 	bool offload;
28
29
 	bool deadline_mode;
30
+	bool skip_sock_check;
29
31
 	int clockid;
30
32
 	int queue;
31
33
 	s32 delta; /* in ns */
32
34
 	ktime_t last; /* The txtime of the last skb sent to the netdevice. */
33
-	struct rb_root head;
35
+	struct rb_root_cached head;
34
36
 	struct qdisc_watchdog watchdog;
35
37
 	ktime_t (*get_time)(void);
36
38
 };
..
..
@@ -77,6 +79,9 @@
77
79
 	struct sock *sk = nskb->sk;
78
80
 	ktime_t now;
79
81
 
82
+	if (q->skip_sock_check)
83
+		goto skip;
84
+
80
85
 	if (!sk || !sk_fullsock(sk))
81
86
 		return false;
82
87
 
..
..
@@ -92,6 +97,7 @@
92
97
 	if (sk->sk_txtime_deadline_mode != q->deadline_mode)
93
98
 		return false;
94
99
 
100
+skip:
95
101
 	now = q->get_time();
96
102
 	if (ktime_before(txtime, now) || ktime_before(txtime, q->last))
97
103
 		return false;
..
..
@@ -104,7 +110,7 @@
104
110
 	struct etf_sched_data *q = qdisc_priv(sch);
105
111
 	struct rb_node *p;
106
112
 
107
-	p = rb_first(&q->head);
113
+	p = rb_first_cached(&q->head);
108
114
 	if (!p)
109
115
 		return NULL;
110
116
 
..
..
@@ -117,8 +123,10 @@
117
123
 	struct sk_buff *skb = etf_peek_timesortedlist(sch);
118
124
 	ktime_t next;
119
125
 
120
-	if (!skb)
126
+	if (!skb) {
127
+		qdisc_watchdog_cancel(&q->watchdog);
121
128
 		return;
129
+	}
122
130
 
123
131
 	next = ktime_sub_ns(skb->tstamp, q->delta);
124
132
 	qdisc_watchdog_schedule_ns(&q->watchdog, ktime_to_ns(next));
..
..
@@ -155,8 +163,9 @@
155
163
 				      struct sk_buff **to_free)
156
164
 {
157
165
 	struct etf_sched_data *q = qdisc_priv(sch);
158
-	struct rb_node **p = &q->head.rb_node, *parent = NULL;
166
+	struct rb_node **p = &q->head.rb_root.rb_node, *parent = NULL;
159
167
 	ktime_t txtime = nskb->tstamp;
168
+	bool leftmost = true;
160
169
 
161
170
 	if (!is_packet_valid(sch, nskb)) {
162
171
 		report_sock_error(nskb, EINVAL,
..
..
@@ -169,13 +178,15 @@
169
178
 
170
179
 		parent = *p;
171
180
 		skb = rb_to_skb(parent);
172
-		if (ktime_after(txtime, skb->tstamp))
181
+		if (ktime_compare(txtime, skb->tstamp) >= 0) {
173
182
 			p = &parent->rb_right;
174
-		else
183
+			leftmost = false;
184
+		} else {
175
185
 			p = &parent->rb_left;
186
+		}
176
187
 	}
177
188
 	rb_link_node(&nskb->rbnode, parent, p);
178
-	rb_insert_color(&nskb->rbnode, &q->head);
189
+	rb_insert_color_cached(&nskb->rbnode, &q->head, leftmost);
179
190
 
180
191
 	qdisc_qstats_backlog_inc(sch, nskb);
181
192
 	sch->q.qlen++;
..
..
@@ -186,12 +197,42 @@
186
197
 	return NET_XMIT_SUCCESS;
187
198
 }
188
199
 
189
-static void timesortedlist_erase(struct Qdisc *sch, struct sk_buff *skb,
190
-				 bool drop)
200
+static void timesortedlist_drop(struct Qdisc *sch, struct sk_buff *skb,
201
+				ktime_t now)
202
+{
203
+	struct etf_sched_data *q = qdisc_priv(sch);
204
+	struct sk_buff *to_free = NULL;
205
+	struct sk_buff *tmp = NULL;
206
+
207
+	skb_rbtree_walk_from_safe(skb, tmp) {
208
+		if (ktime_after(skb->tstamp, now))
209
+			break;
210
+
211
+		rb_erase_cached(&skb->rbnode, &q->head);
212
+
213
+		/* The rbnode field in the skb re-uses these fields, now that
214
+		 * we are done with the rbnode, reset them.
215
+		 */
216
+		skb->next = NULL;
217
+		skb->prev = NULL;
218
+		skb->dev = qdisc_dev(sch);
219
+
220
+		report_sock_error(skb, ECANCELED, SO_EE_CODE_TXTIME_MISSED);
221
+
222
+		qdisc_qstats_backlog_dec(sch, skb);
223
+		qdisc_drop(skb, sch, &to_free);
224
+		qdisc_qstats_overlimit(sch);
225
+		sch->q.qlen--;
226
+	}
227
+
228
+	kfree_skb_list(to_free);
229
+}
230
+
231
+static void timesortedlist_remove(struct Qdisc *sch, struct sk_buff *skb)
191
232
 {
192
233
 	struct etf_sched_data *q = qdisc_priv(sch);
193
234
 
194
-	rb_erase(&skb->rbnode, &q->head);
235
+	rb_erase_cached(&skb->rbnode, &q->head);
195
236
 
196
237
 	/* The rbnode field in the skb re-uses these fields, now that
197
238
 	 * we are done with the rbnode, reset them.
..
..
@@ -202,19 +243,9 @@
202
243
 
203
244
 	qdisc_qstats_backlog_dec(sch, skb);
204
245
 
205
-	if (drop) {
206
-		struct sk_buff *to_free = NULL;
246
+	qdisc_bstats_update(sch, skb);
207
247
 
208
-		report_sock_error(skb, ECANCELED, SO_EE_CODE_TXTIME_MISSED);
209
-
210
-		qdisc_drop(skb, sch, &to_free);
211
-		kfree_skb_list(to_free);
212
-		qdisc_qstats_overlimit(sch);
213
-	} else {
214
-		qdisc_bstats_update(sch, skb);
215
-
216
-		q->last = skb->tstamp;
217
-	}
248
+	q->last = skb->tstamp;
218
249
 
219
250
 	sch->q.qlen--;
220
251
 }
..
..
@@ -233,7 +264,7 @@
233
264
 
234
265
 	/* Drop if packet has expired while in queue. */
235
266
 	if (ktime_before(skb->tstamp, now)) {
236
-		timesortedlist_erase(sch, skb, true);
267
+		timesortedlist_drop(sch, skb, now);
237
268
 		skb = NULL;
238
269
 		goto out;
239
270
 	}
..
..
@@ -242,7 +273,7 @@
242
273
 	 * txtime from deadline to (now + delta).
243
274
 	 */
244
275
 	if (q->deadline_mode) {
245
-		timesortedlist_erase(sch, skb, false);
276
+		timesortedlist_remove(sch, skb);
246
277
 		skb->tstamp = now;
247
278
 		goto out;
248
279
 	}
..
..
@@ -251,7 +282,7 @@
251
282
 
252
283
 	/* Dequeue only if now is within the [txtime - delta, txtime] range. */
253
284
 	if (ktime_after(now, next))
254
-		timesortedlist_erase(sch, skb, false);
285
+		timesortedlist_remove(sch, skb);
255
286
 	else
256
287
 		skb = NULL;
257
288
 
..
..
@@ -327,7 +358,8 @@
327
358
 		return -EINVAL;
328
359
 	}
329
360
 
330
-	err = nla_parse_nested(tb, TCA_ETF_MAX, opt, etf_policy, extack);
361
+	err = nla_parse_nested_deprecated(tb, TCA_ETF_MAX, opt, etf_policy,
362
+					  extack);
331
363
 	if (err < 0)
332
364
 		return err;
333
365
 
..
..
@@ -360,6 +392,7 @@
360
392
 	q->clockid = qopt->clockid;
361
393
 	q->offload = OFFLOAD_IS_ON(qopt);
362
394
 	q->deadline_mode = DEADLINE_MODE_IS_ON(qopt);
395
+	q->skip_sock_check = SKIP_SOCK_CHECK_IS_SET(qopt);
363
396
 
364
397
 	switch (q->clockid) {
365
398
 	case CLOCK_REALTIME:
..
..
@@ -387,14 +420,14 @@
387
420
 static void timesortedlist_clear(struct Qdisc *sch)
388
421
 {
389
422
 	struct etf_sched_data *q = qdisc_priv(sch);
390
-	struct rb_node *p = rb_first(&q->head);
423
+	struct rb_node *p = rb_first_cached(&q->head);
391
424
 
392
425
 	while (p) {
393
426
 		struct sk_buff *skb = rb_to_skb(p);
394
427
 
395
428
 		p = rb_next(p);
396
429
 
397
-		rb_erase(&skb->rbnode, &q->head);
430
+		rb_erase_cached(&skb->rbnode, &q->head);
398
431
 		rtnl_kfree_skbs(skb, skb);
399
432
 		sch->q.qlen--;
400
433
 	}
..
..
@@ -411,9 +444,6 @@
411
444
 	/* No matter which mode we are on, it's safe to clear both lists. */
412
445
 	timesortedlist_clear(sch);
413
446
 	__qdisc_reset_queue(&sch->q);
414
-
415
-	sch->qstats.backlog = 0;
416
-	sch->q.qlen = 0;
417
447
 
418
448
 	q->last = 0;
419
449
 }
..
..
@@ -436,7 +466,7 @@
436
466
 	struct tc_etf_qopt opt = { };
437
467
 	struct nlattr *nest;
438
468
 
439
-	nest = nla_nest_start(skb, TCA_OPTIONS);
469
+	nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
440
470
 	if (!nest)
441
471
 		goto nla_put_failure;
442
472
 
..
..
@@ -448,6 +478,9 @@
448
478
 	if (q->deadline_mode)
449
479
 		opt.flags |= TC_ETF_DEADLINE_MODE_ON;
450
480
 
481
+	if (q->skip_sock_check)
482
+		opt.flags |= TC_ETF_SKIP_SOCK_CHECK;
483
+
451
484
 	if (nla_put(skb, TCA_ETF_PARMS, sizeof(opt), &opt))
452
485
 		goto nla_put_failure;
453
486