]>
Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
4b549a2e ED |
2 | /* |
3 | * Fair Queue CoDel discipline | |
4 | * | |
80ba92fa | 5 | * Copyright (C) 2012,2015 Eric Dumazet <edumazet@google.com> |
4b549a2e ED |
6 | */ |
7 | ||
8 | #include <linux/module.h> | |
9 | #include <linux/types.h> | |
10 | #include <linux/kernel.h> | |
11 | #include <linux/jiffies.h> | |
12 | #include <linux/string.h> | |
13 | #include <linux/in.h> | |
14 | #include <linux/errno.h> | |
15 | #include <linux/init.h> | |
16 | #include <linux/skbuff.h> | |
17 | #include <linux/jhash.h> | |
18 | #include <linux/slab.h> | |
19 | #include <linux/vmalloc.h> | |
20 | #include <net/netlink.h> | |
21 | #include <net/pkt_sched.h> | |
cf1facda | 22 | #include <net/pkt_cls.h> |
4b549a2e | 23 | #include <net/codel.h> |
d068ca2a MK |
24 | #include <net/codel_impl.h> |
25 | #include <net/codel_qdisc.h> | |
4b549a2e ED |
26 | |
27 | /* Fair Queue CoDel. | |
28 | * | |
29 | * Principles : | |
30 | * Packets are classified (internal classifier or external) on flows. | |
31 | * This is a Stochastic model (as we use a hash, several flows | |
32 | * might be hashed on same slot) | |
33 | * Each flow has a CoDel managed queue. | |
34 | * Flows are linked onto two (Round Robin) lists, | |
35 | * so that new flows have priority on old ones. | |
36 | * | |
37 | * For a given flow, packets are not reordered (CoDel uses a FIFO) | |
38 | * head drops only. | |
39 | * ECN capability is on by default. | |
40 | * Low memory footprint (64 bytes per flow) | |
41 | */ | |
42 | ||
43 | struct fq_codel_flow { | |
44 | struct sk_buff *head; | |
45 | struct sk_buff *tail; | |
46 | struct list_head flowchain; | |
47 | int deficit; | |
48 | u32 dropped; /* number of drops (or ECN marks) on this flow */ | |
49 | struct codel_vars cvars; | |
50 | }; /* please try to keep this structure <= 64 bytes */ | |
51 | ||
52 | struct fq_codel_sched_data { | |
25d8c0d5 | 53 | struct tcf_proto __rcu *filter_list; /* optional external classifier */ |
6529eaba | 54 | struct tcf_block *block; |
4b549a2e ED |
55 | struct fq_codel_flow *flows; /* Flows table [flows_cnt] */ |
56 | u32 *backlogs; /* backlog table [flows_cnt] */ | |
57 | u32 flows_cnt; /* number of flows */ | |
4b549a2e | 58 | u32 quantum; /* psched_mtu(qdisc_dev(sch)); */ |
9d18562a | 59 | u32 drop_batch_size; |
95b58430 | 60 | u32 memory_limit; |
4b549a2e ED |
61 | struct codel_params cparams; |
62 | struct codel_stats cstats; | |
95b58430 ED |
63 | u32 memory_usage; |
64 | u32 drop_overmemory; | |
4b549a2e ED |
65 | u32 drop_overlimit; |
66 | u32 new_flow_count; | |
67 | ||
68 | struct list_head new_flows; /* list of new flows */ | |
69 | struct list_head old_flows; /* list of old flows */ | |
70 | }; | |
71 | ||
72 | static unsigned int fq_codel_hash(const struct fq_codel_sched_data *q, | |
342db221 | 73 | struct sk_buff *skb) |
4b549a2e | 74 | { |
264b87fa | 75 | return reciprocal_scale(skb_get_hash(skb), q->flows_cnt); |
4b549a2e ED |
76 | } |
77 | ||
78 | static unsigned int fq_codel_classify(struct sk_buff *skb, struct Qdisc *sch, | |
79 | int *qerr) | |
80 | { | |
81 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
25d8c0d5 | 82 | struct tcf_proto *filter; |
4b549a2e ED |
83 | struct tcf_result res; |
84 | int result; | |
85 | ||
86 | if (TC_H_MAJ(skb->priority) == sch->handle && | |
87 | TC_H_MIN(skb->priority) > 0 && | |
88 | TC_H_MIN(skb->priority) <= q->flows_cnt) | |
89 | return TC_H_MIN(skb->priority); | |
90 | ||
69204cf7 | 91 | filter = rcu_dereference_bh(q->filter_list); |
25d8c0d5 | 92 | if (!filter) |
4b549a2e ED |
93 | return fq_codel_hash(q, skb) + 1; |
94 | ||
95 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; | |
87d83093 | 96 | result = tcf_classify(skb, filter, &res, false); |
4b549a2e ED |
97 | if (result >= 0) { |
98 | #ifdef CONFIG_NET_CLS_ACT | |
99 | switch (result) { | |
100 | case TC_ACT_STOLEN: | |
101 | case TC_ACT_QUEUED: | |
e25ea21f | 102 | case TC_ACT_TRAP: |
4b549a2e | 103 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; |
f3ae608e | 104 | /* fall through */ |
4b549a2e ED |
105 | case TC_ACT_SHOT: |
106 | return 0; | |
107 | } | |
108 | #endif | |
109 | if (TC_H_MIN(res.classid) <= q->flows_cnt) | |
110 | return TC_H_MIN(res.classid); | |
111 | } | |
112 | return 0; | |
113 | } | |
114 | ||
115 | /* helper functions : might be changed when/if skb use a standard list_head */ | |
116 | ||
117 | /* remove one skb from head of slot queue */ | |
118 | static inline struct sk_buff *dequeue_head(struct fq_codel_flow *flow) | |
119 | { | |
120 | struct sk_buff *skb = flow->head; | |
121 | ||
122 | flow->head = skb->next; | |
a8305bff | 123 | skb_mark_not_on_list(skb); |
4b549a2e ED |
124 | return skb; |
125 | } | |
126 | ||
127 | /* add skb to flow queue (tail add) */ | |
128 | static inline void flow_queue_add(struct fq_codel_flow *flow, | |
129 | struct sk_buff *skb) | |
130 | { | |
131 | if (flow->head == NULL) | |
132 | flow->head = skb; | |
133 | else | |
134 | flow->tail->next = skb; | |
135 | flow->tail = skb; | |
136 | skb->next = NULL; | |
137 | } | |
138 | ||
520ac30f ED |
139 | static unsigned int fq_codel_drop(struct Qdisc *sch, unsigned int max_packets, |
140 | struct sk_buff **to_free) | |
4b549a2e ED |
141 | { |
142 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
143 | struct sk_buff *skb; | |
144 | unsigned int maxbacklog = 0, idx = 0, i, len; | |
145 | struct fq_codel_flow *flow; | |
9d18562a | 146 | unsigned int threshold; |
95b58430 | 147 | unsigned int mem = 0; |
4b549a2e | 148 | |
9d18562a | 149 | /* Queue is full! Find the fat flow and drop packet(s) from it. |
4b549a2e ED |
150 | * This might sound expensive, but with 1024 flows, we scan |
151 | * 4KB of memory, and we dont need to handle a complex tree | |
152 | * in fast path (packet queue/enqueue) with many cache misses. | |
9d18562a ED |
153 | * In stress mode, we'll try to drop 64 packets from the flow, |
154 | * amortizing this linear lookup to one cache line per drop. | |
4b549a2e ED |
155 | */ |
156 | for (i = 0; i < q->flows_cnt; i++) { | |
157 | if (q->backlogs[i] > maxbacklog) { | |
158 | maxbacklog = q->backlogs[i]; | |
159 | idx = i; | |
160 | } | |
161 | } | |
9d18562a ED |
162 | |
163 | /* Our goal is to drop half of this fat flow backlog */ | |
164 | threshold = maxbacklog >> 1; | |
165 | ||
4b549a2e | 166 | flow = &q->flows[idx]; |
9d18562a ED |
167 | len = 0; |
168 | i = 0; | |
169 | do { | |
170 | skb = dequeue_head(flow); | |
171 | len += qdisc_pkt_len(skb); | |
008830bc | 172 | mem += get_codel_cb(skb)->mem_usage; |
520ac30f | 173 | __qdisc_drop(skb, to_free); |
9d18562a ED |
174 | } while (++i < max_packets && len < threshold); |
175 | ||
176 | flow->dropped += i; | |
4b549a2e | 177 | q->backlogs[idx] -= len; |
95b58430 | 178 | q->memory_usage -= mem; |
9d18562a ED |
179 | sch->qstats.drops += i; |
180 | sch->qstats.backlog -= len; | |
181 | sch->q.qlen -= i; | |
4b549a2e ED |
182 | return idx; |
183 | } | |
184 | ||
520ac30f ED |
185 | static int fq_codel_enqueue(struct sk_buff *skb, struct Qdisc *sch, |
186 | struct sk_buff **to_free) | |
4b549a2e ED |
187 | { |
188 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
9d18562a | 189 | unsigned int idx, prev_backlog, prev_qlen; |
4b549a2e ED |
190 | struct fq_codel_flow *flow; |
191 | int uninitialized_var(ret); | |
80e509db | 192 | unsigned int pkt_len; |
95b58430 | 193 | bool memory_limited; |
4b549a2e ED |
194 | |
195 | idx = fq_codel_classify(skb, sch, &ret); | |
196 | if (idx == 0) { | |
197 | if (ret & __NET_XMIT_BYPASS) | |
25331d6c | 198 | qdisc_qstats_drop(sch); |
520ac30f | 199 | __qdisc_drop(skb, to_free); |
4b549a2e ED |
200 | return ret; |
201 | } | |
202 | idx--; | |
203 | ||
204 | codel_set_enqueue_time(skb); | |
205 | flow = &q->flows[idx]; | |
206 | flow_queue_add(flow, skb); | |
207 | q->backlogs[idx] += qdisc_pkt_len(skb); | |
25331d6c | 208 | qdisc_qstats_backlog_inc(sch, skb); |
4b549a2e ED |
209 | |
210 | if (list_empty(&flow->flowchain)) { | |
211 | list_add_tail(&flow->flowchain, &q->new_flows); | |
4b549a2e ED |
212 | q->new_flow_count++; |
213 | flow->deficit = q->quantum; | |
214 | flow->dropped = 0; | |
215 | } | |
008830bc ED |
216 | get_codel_cb(skb)->mem_usage = skb->truesize; |
217 | q->memory_usage += get_codel_cb(skb)->mem_usage; | |
95b58430 ED |
218 | memory_limited = q->memory_usage > q->memory_limit; |
219 | if (++sch->q.qlen <= sch->limit && !memory_limited) | |
4b549a2e ED |
220 | return NET_XMIT_SUCCESS; |
221 | ||
2ccccf5f | 222 | prev_backlog = sch->qstats.backlog; |
9d18562a ED |
223 | prev_qlen = sch->q.qlen; |
224 | ||
80e509db ED |
225 | /* save this packet length as it might be dropped by fq_codel_drop() */ |
226 | pkt_len = qdisc_pkt_len(skb); | |
9d18562a ED |
227 | /* fq_codel_drop() is quite expensive, as it performs a linear search |
228 | * in q->backlogs[] to find a fat flow. | |
229 | * So instead of dropping a single packet, drop half of its backlog | |
230 | * with a 64 packets limit to not add a too big cpu spike here. | |
4b549a2e | 231 | */ |
520ac30f | 232 | ret = fq_codel_drop(sch, q->drop_batch_size, to_free); |
9d18562a | 233 | |
80e509db ED |
234 | prev_qlen -= sch->q.qlen; |
235 | prev_backlog -= sch->qstats.backlog; | |
236 | q->drop_overlimit += prev_qlen; | |
95b58430 | 237 | if (memory_limited) |
80e509db | 238 | q->drop_overmemory += prev_qlen; |
9d18562a | 239 | |
80e509db ED |
240 | /* As we dropped packet(s), better let upper stack know this. |
241 | * If we dropped a packet for this flow, return NET_XMIT_CN, | |
242 | * but in this case, our parents wont increase their backlogs. | |
243 | */ | |
244 | if (ret == idx) { | |
245 | qdisc_tree_reduce_backlog(sch, prev_qlen - 1, | |
246 | prev_backlog - pkt_len); | |
247 | return NET_XMIT_CN; | |
248 | } | |
249 | qdisc_tree_reduce_backlog(sch, prev_qlen, prev_backlog); | |
250 | return NET_XMIT_SUCCESS; | |
4b549a2e ED |
251 | } |
252 | ||
253 | /* This is the specific function called from codel_dequeue() | |
254 | * to dequeue a packet from queue. Note: backlog is handled in | |
255 | * codel, we dont need to reduce it here. | |
256 | */ | |
79bdc4c8 | 257 | static struct sk_buff *dequeue_func(struct codel_vars *vars, void *ctx) |
4b549a2e | 258 | { |
79bdc4c8 | 259 | struct Qdisc *sch = ctx; |
865ec552 | 260 | struct fq_codel_sched_data *q = qdisc_priv(sch); |
4b549a2e ED |
261 | struct fq_codel_flow *flow; |
262 | struct sk_buff *skb = NULL; | |
263 | ||
264 | flow = container_of(vars, struct fq_codel_flow, cvars); | |
265 | if (flow->head) { | |
266 | skb = dequeue_head(flow); | |
865ec552 | 267 | q->backlogs[flow - q->flows] -= qdisc_pkt_len(skb); |
008830bc | 268 | q->memory_usage -= get_codel_cb(skb)->mem_usage; |
4b549a2e | 269 | sch->q.qlen--; |
79bdc4c8 | 270 | sch->qstats.backlog -= qdisc_pkt_len(skb); |
4b549a2e ED |
271 | } |
272 | return skb; | |
273 | } | |
274 | ||
79bdc4c8 MK |
275 | static void drop_func(struct sk_buff *skb, void *ctx) |
276 | { | |
277 | struct Qdisc *sch = ctx; | |
278 | ||
520ac30f ED |
279 | kfree_skb(skb); |
280 | qdisc_qstats_drop(sch); | |
79bdc4c8 MK |
281 | } |
282 | ||
4b549a2e ED |
283 | static struct sk_buff *fq_codel_dequeue(struct Qdisc *sch) |
284 | { | |
285 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
286 | struct sk_buff *skb; | |
287 | struct fq_codel_flow *flow; | |
288 | struct list_head *head; | |
289 | u32 prev_drop_count, prev_ecn_mark; | |
290 | ||
291 | begin: | |
292 | head = &q->new_flows; | |
293 | if (list_empty(head)) { | |
294 | head = &q->old_flows; | |
295 | if (list_empty(head)) | |
296 | return NULL; | |
297 | } | |
298 | flow = list_first_entry(head, struct fq_codel_flow, flowchain); | |
299 | ||
300 | if (flow->deficit <= 0) { | |
301 | flow->deficit += q->quantum; | |
302 | list_move_tail(&flow->flowchain, &q->old_flows); | |
303 | goto begin; | |
304 | } | |
305 | ||
306 | prev_drop_count = q->cstats.drop_count; | |
307 | prev_ecn_mark = q->cstats.ecn_mark; | |
308 | ||
79bdc4c8 MK |
309 | skb = codel_dequeue(sch, &sch->qstats.backlog, &q->cparams, |
310 | &flow->cvars, &q->cstats, qdisc_pkt_len, | |
311 | codel_get_enqueue_time, drop_func, dequeue_func); | |
4b549a2e ED |
312 | |
313 | flow->dropped += q->cstats.drop_count - prev_drop_count; | |
314 | flow->dropped += q->cstats.ecn_mark - prev_ecn_mark; | |
315 | ||
316 | if (!skb) { | |
317 | /* force a pass through old_flows to prevent starvation */ | |
318 | if ((head == &q->new_flows) && !list_empty(&q->old_flows)) | |
319 | list_move_tail(&flow->flowchain, &q->old_flows); | |
320 | else | |
321 | list_del_init(&flow->flowchain); | |
322 | goto begin; | |
323 | } | |
324 | qdisc_bstats_update(sch, skb); | |
325 | flow->deficit -= qdisc_pkt_len(skb); | |
2ccccf5f | 326 | /* We cant call qdisc_tree_reduce_backlog() if our qlen is 0, |
4b549a2e ED |
327 | * or HTB crashes. Defer it for next round. |
328 | */ | |
329 | if (q->cstats.drop_count && sch->q.qlen) { | |
2ccccf5f WC |
330 | qdisc_tree_reduce_backlog(sch, q->cstats.drop_count, |
331 | q->cstats.drop_len); | |
4b549a2e | 332 | q->cstats.drop_count = 0; |
2ccccf5f | 333 | q->cstats.drop_len = 0; |
4b549a2e ED |
334 | } |
335 | return skb; | |
336 | } | |
337 | ||
ece5d4c7 ED |
338 | static void fq_codel_flow_purge(struct fq_codel_flow *flow) |
339 | { | |
340 | rtnl_kfree_skbs(flow->head, flow->tail); | |
341 | flow->head = NULL; | |
342 | } | |
343 | ||
4b549a2e ED |
344 | static void fq_codel_reset(struct Qdisc *sch) |
345 | { | |
3d0e0af4 ED |
346 | struct fq_codel_sched_data *q = qdisc_priv(sch); |
347 | int i; | |
4b549a2e | 348 | |
3d0e0af4 ED |
349 | INIT_LIST_HEAD(&q->new_flows); |
350 | INIT_LIST_HEAD(&q->old_flows); | |
351 | for (i = 0; i < q->flows_cnt; i++) { | |
352 | struct fq_codel_flow *flow = q->flows + i; | |
353 | ||
ece5d4c7 | 354 | fq_codel_flow_purge(flow); |
3d0e0af4 ED |
355 | INIT_LIST_HEAD(&flow->flowchain); |
356 | codel_vars_init(&flow->cvars); | |
357 | } | |
358 | memset(q->backlogs, 0, q->flows_cnt * sizeof(u32)); | |
359 | sch->q.qlen = 0; | |
ece5d4c7 | 360 | sch->qstats.backlog = 0; |
77f57761 | 361 | q->memory_usage = 0; |
4b549a2e ED |
362 | } |
363 | ||
364 | static const struct nla_policy fq_codel_policy[TCA_FQ_CODEL_MAX + 1] = { | |
365 | [TCA_FQ_CODEL_TARGET] = { .type = NLA_U32 }, | |
366 | [TCA_FQ_CODEL_LIMIT] = { .type = NLA_U32 }, | |
367 | [TCA_FQ_CODEL_INTERVAL] = { .type = NLA_U32 }, | |
368 | [TCA_FQ_CODEL_ECN] = { .type = NLA_U32 }, | |
369 | [TCA_FQ_CODEL_FLOWS] = { .type = NLA_U32 }, | |
370 | [TCA_FQ_CODEL_QUANTUM] = { .type = NLA_U32 }, | |
80ba92fa | 371 | [TCA_FQ_CODEL_CE_THRESHOLD] = { .type = NLA_U32 }, |
9d18562a | 372 | [TCA_FQ_CODEL_DROP_BATCH_SIZE] = { .type = NLA_U32 }, |
95b58430 | 373 | [TCA_FQ_CODEL_MEMORY_LIMIT] = { .type = NLA_U32 }, |
4b549a2e ED |
374 | }; |
375 | ||
2030721c AA |
376 | static int fq_codel_change(struct Qdisc *sch, struct nlattr *opt, |
377 | struct netlink_ext_ack *extack) | |
4b549a2e ED |
378 | { |
379 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
380 | struct nlattr *tb[TCA_FQ_CODEL_MAX + 1]; | |
381 | int err; | |
382 | ||
383 | if (!opt) | |
384 | return -EINVAL; | |
385 | ||
8cb08174 JB |
386 | err = nla_parse_nested_deprecated(tb, TCA_FQ_CODEL_MAX, opt, |
387 | fq_codel_policy, NULL); | |
4b549a2e ED |
388 | if (err < 0) |
389 | return err; | |
390 | if (tb[TCA_FQ_CODEL_FLOWS]) { | |
391 | if (q->flows) | |
392 | return -EINVAL; | |
393 | q->flows_cnt = nla_get_u32(tb[TCA_FQ_CODEL_FLOWS]); | |
394 | if (!q->flows_cnt || | |
395 | q->flows_cnt > 65536) | |
396 | return -EINVAL; | |
397 | } | |
398 | sch_tree_lock(sch); | |
399 | ||
400 | if (tb[TCA_FQ_CODEL_TARGET]) { | |
401 | u64 target = nla_get_u32(tb[TCA_FQ_CODEL_TARGET]); | |
402 | ||
403 | q->cparams.target = (target * NSEC_PER_USEC) >> CODEL_SHIFT; | |
404 | } | |
405 | ||
80ba92fa ED |
406 | if (tb[TCA_FQ_CODEL_CE_THRESHOLD]) { |
407 | u64 val = nla_get_u32(tb[TCA_FQ_CODEL_CE_THRESHOLD]); | |
408 | ||
409 | q->cparams.ce_threshold = (val * NSEC_PER_USEC) >> CODEL_SHIFT; | |
410 | } | |
411 | ||
4b549a2e ED |
412 | if (tb[TCA_FQ_CODEL_INTERVAL]) { |
413 | u64 interval = nla_get_u32(tb[TCA_FQ_CODEL_INTERVAL]); | |
414 | ||
415 | q->cparams.interval = (interval * NSEC_PER_USEC) >> CODEL_SHIFT; | |
416 | } | |
417 | ||
418 | if (tb[TCA_FQ_CODEL_LIMIT]) | |
419 | sch->limit = nla_get_u32(tb[TCA_FQ_CODEL_LIMIT]); | |
420 | ||
421 | if (tb[TCA_FQ_CODEL_ECN]) | |
422 | q->cparams.ecn = !!nla_get_u32(tb[TCA_FQ_CODEL_ECN]); | |
423 | ||
424 | if (tb[TCA_FQ_CODEL_QUANTUM]) | |
425 | q->quantum = max(256U, nla_get_u32(tb[TCA_FQ_CODEL_QUANTUM])); | |
426 | ||
9d18562a ED |
427 | if (tb[TCA_FQ_CODEL_DROP_BATCH_SIZE]) |
428 | q->drop_batch_size = min(1U, nla_get_u32(tb[TCA_FQ_CODEL_DROP_BATCH_SIZE])); | |
429 | ||
95b58430 ED |
430 | if (tb[TCA_FQ_CODEL_MEMORY_LIMIT]) |
431 | q->memory_limit = min(1U << 31, nla_get_u32(tb[TCA_FQ_CODEL_MEMORY_LIMIT])); | |
432 | ||
433 | while (sch->q.qlen > sch->limit || | |
434 | q->memory_usage > q->memory_limit) { | |
4b549a2e ED |
435 | struct sk_buff *skb = fq_codel_dequeue(sch); |
436 | ||
2ccccf5f | 437 | q->cstats.drop_len += qdisc_pkt_len(skb); |
ece5d4c7 | 438 | rtnl_kfree_skbs(skb, skb); |
4b549a2e ED |
439 | q->cstats.drop_count++; |
440 | } | |
2ccccf5f | 441 | qdisc_tree_reduce_backlog(sch, q->cstats.drop_count, q->cstats.drop_len); |
4b549a2e | 442 | q->cstats.drop_count = 0; |
2ccccf5f | 443 | q->cstats.drop_len = 0; |
4b549a2e ED |
444 | |
445 | sch_tree_unlock(sch); | |
446 | return 0; | |
447 | } | |
448 | ||
4b549a2e ED |
449 | static void fq_codel_destroy(struct Qdisc *sch) |
450 | { | |
451 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
452 | ||
6529eaba | 453 | tcf_block_put(q->block); |
752ade68 MH |
454 | kvfree(q->backlogs); |
455 | kvfree(q->flows); | |
4b549a2e ED |
456 | } |
457 | ||
e63d7dfd AA |
458 | static int fq_codel_init(struct Qdisc *sch, struct nlattr *opt, |
459 | struct netlink_ext_ack *extack) | |
4b549a2e ED |
460 | { |
461 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
462 | int i; | |
6529eaba | 463 | int err; |
4b549a2e ED |
464 | |
465 | sch->limit = 10*1024; | |
466 | q->flows_cnt = 1024; | |
95b58430 | 467 | q->memory_limit = 32 << 20; /* 32 MBytes */ |
9d18562a | 468 | q->drop_batch_size = 64; |
4b549a2e | 469 | q->quantum = psched_mtu(qdisc_dev(sch)); |
4b549a2e ED |
470 | INIT_LIST_HEAD(&q->new_flows); |
471 | INIT_LIST_HEAD(&q->old_flows); | |
79bdc4c8 | 472 | codel_params_init(&q->cparams); |
4b549a2e ED |
473 | codel_stats_init(&q->cstats); |
474 | q->cparams.ecn = true; | |
79bdc4c8 | 475 | q->cparams.mtu = psched_mtu(qdisc_dev(sch)); |
4b549a2e ED |
476 | |
477 | if (opt) { | |
83fe6b87 | 478 | err = fq_codel_change(sch, opt, extack); |
4b549a2e | 479 | if (err) |
83fe6b87 | 480 | goto init_failure; |
4b549a2e ED |
481 | } |
482 | ||
8d1a77f9 | 483 | err = tcf_block_get(&q->block, &q->filter_list, sch, extack); |
6529eaba | 484 | if (err) |
83fe6b87 | 485 | goto init_failure; |
6529eaba | 486 | |
4b549a2e | 487 | if (!q->flows) { |
778e1cdd KC |
488 | q->flows = kvcalloc(q->flows_cnt, |
489 | sizeof(struct fq_codel_flow), | |
490 | GFP_KERNEL); | |
83fe6b87 JK |
491 | if (!q->flows) { |
492 | err = -ENOMEM; | |
493 | goto init_failure; | |
494 | } | |
778e1cdd | 495 | q->backlogs = kvcalloc(q->flows_cnt, sizeof(u32), GFP_KERNEL); |
83fe6b87 JK |
496 | if (!q->backlogs) { |
497 | err = -ENOMEM; | |
498 | goto alloc_failure; | |
499 | } | |
4b549a2e ED |
500 | for (i = 0; i < q->flows_cnt; i++) { |
501 | struct fq_codel_flow *flow = q->flows + i; | |
502 | ||
503 | INIT_LIST_HEAD(&flow->flowchain); | |
b379135c | 504 | codel_vars_init(&flow->cvars); |
4b549a2e ED |
505 | } |
506 | } | |
507 | if (sch->limit >= 1) | |
508 | sch->flags |= TCQ_F_CAN_BYPASS; | |
509 | else | |
510 | sch->flags &= ~TCQ_F_CAN_BYPASS; | |
511 | return 0; | |
83fe6b87 JK |
512 | |
513 | alloc_failure: | |
514 | kvfree(q->flows); | |
515 | q->flows = NULL; | |
516 | init_failure: | |
517 | q->flows_cnt = 0; | |
518 | return err; | |
4b549a2e ED |
519 | } |
520 | ||
521 | static int fq_codel_dump(struct Qdisc *sch, struct sk_buff *skb) | |
522 | { | |
523 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
524 | struct nlattr *opts; | |
525 | ||
ae0be8de | 526 | opts = nla_nest_start_noflag(skb, TCA_OPTIONS); |
4b549a2e ED |
527 | if (opts == NULL) |
528 | goto nla_put_failure; | |
529 | ||
530 | if (nla_put_u32(skb, TCA_FQ_CODEL_TARGET, | |
531 | codel_time_to_us(q->cparams.target)) || | |
532 | nla_put_u32(skb, TCA_FQ_CODEL_LIMIT, | |
533 | sch->limit) || | |
534 | nla_put_u32(skb, TCA_FQ_CODEL_INTERVAL, | |
535 | codel_time_to_us(q->cparams.interval)) || | |
536 | nla_put_u32(skb, TCA_FQ_CODEL_ECN, | |
537 | q->cparams.ecn) || | |
538 | nla_put_u32(skb, TCA_FQ_CODEL_QUANTUM, | |
539 | q->quantum) || | |
9d18562a ED |
540 | nla_put_u32(skb, TCA_FQ_CODEL_DROP_BATCH_SIZE, |
541 | q->drop_batch_size) || | |
95b58430 ED |
542 | nla_put_u32(skb, TCA_FQ_CODEL_MEMORY_LIMIT, |
543 | q->memory_limit) || | |
4b549a2e ED |
544 | nla_put_u32(skb, TCA_FQ_CODEL_FLOWS, |
545 | q->flows_cnt)) | |
546 | goto nla_put_failure; | |
547 | ||
80ba92fa ED |
548 | if (q->cparams.ce_threshold != CODEL_DISABLED_THRESHOLD && |
549 | nla_put_u32(skb, TCA_FQ_CODEL_CE_THRESHOLD, | |
550 | codel_time_to_us(q->cparams.ce_threshold))) | |
551 | goto nla_put_failure; | |
552 | ||
d59b7d80 | 553 | return nla_nest_end(skb, opts); |
4b549a2e ED |
554 | |
555 | nla_put_failure: | |
556 | return -1; | |
557 | } | |
558 | ||
559 | static int fq_codel_dump_stats(struct Qdisc *sch, struct gnet_dump *d) | |
560 | { | |
561 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
562 | struct tc_fq_codel_xstats st = { | |
563 | .type = TCA_FQ_CODEL_XSTATS_QDISC, | |
4b549a2e ED |
564 | }; |
565 | struct list_head *pos; | |
566 | ||
669d67bf SL |
567 | st.qdisc_stats.maxpacket = q->cstats.maxpacket; |
568 | st.qdisc_stats.drop_overlimit = q->drop_overlimit; | |
569 | st.qdisc_stats.ecn_mark = q->cstats.ecn_mark; | |
570 | st.qdisc_stats.new_flow_count = q->new_flow_count; | |
80ba92fa | 571 | st.qdisc_stats.ce_mark = q->cstats.ce_mark; |
95b58430 ED |
572 | st.qdisc_stats.memory_usage = q->memory_usage; |
573 | st.qdisc_stats.drop_overmemory = q->drop_overmemory; | |
669d67bf | 574 | |
edb09eb1 | 575 | sch_tree_lock(sch); |
4b549a2e ED |
576 | list_for_each(pos, &q->new_flows) |
577 | st.qdisc_stats.new_flows_len++; | |
578 | ||
579 | list_for_each(pos, &q->old_flows) | |
580 | st.qdisc_stats.old_flows_len++; | |
edb09eb1 | 581 | sch_tree_unlock(sch); |
4b549a2e ED |
582 | |
583 | return gnet_stats_copy_app(d, &st, sizeof(st)); | |
584 | } | |
585 | ||
586 | static struct Qdisc *fq_codel_leaf(struct Qdisc *sch, unsigned long arg) | |
587 | { | |
588 | return NULL; | |
589 | } | |
590 | ||
143976ce | 591 | static unsigned long fq_codel_find(struct Qdisc *sch, u32 classid) |
4b549a2e ED |
592 | { |
593 | return 0; | |
594 | } | |
595 | ||
596 | static unsigned long fq_codel_bind(struct Qdisc *sch, unsigned long parent, | |
597 | u32 classid) | |
598 | { | |
599 | /* we cannot bypass queue discipline anymore */ | |
600 | sch->flags &= ~TCQ_F_CAN_BYPASS; | |
601 | return 0; | |
602 | } | |
603 | ||
143976ce | 604 | static void fq_codel_unbind(struct Qdisc *q, unsigned long cl) |
4b549a2e ED |
605 | { |
606 | } | |
607 | ||
cbaacc4e AA |
608 | static struct tcf_block *fq_codel_tcf_block(struct Qdisc *sch, unsigned long cl, |
609 | struct netlink_ext_ack *extack) | |
4b549a2e ED |
610 | { |
611 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
612 | ||
613 | if (cl) | |
614 | return NULL; | |
6529eaba | 615 | return q->block; |
4b549a2e ED |
616 | } |
617 | ||
618 | static int fq_codel_dump_class(struct Qdisc *sch, unsigned long cl, | |
619 | struct sk_buff *skb, struct tcmsg *tcm) | |
620 | { | |
621 | tcm->tcm_handle |= TC_H_MIN(cl); | |
622 | return 0; | |
623 | } | |
624 | ||
625 | static int fq_codel_dump_class_stats(struct Qdisc *sch, unsigned long cl, | |
626 | struct gnet_dump *d) | |
627 | { | |
628 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
629 | u32 idx = cl - 1; | |
630 | struct gnet_stats_queue qs = { 0 }; | |
631 | struct tc_fq_codel_xstats xstats; | |
632 | ||
633 | if (idx < q->flows_cnt) { | |
634 | const struct fq_codel_flow *flow = &q->flows[idx]; | |
edb09eb1 | 635 | const struct sk_buff *skb; |
4b549a2e ED |
636 | |
637 | memset(&xstats, 0, sizeof(xstats)); | |
638 | xstats.type = TCA_FQ_CODEL_XSTATS_CLASS; | |
639 | xstats.class_stats.deficit = flow->deficit; | |
640 | xstats.class_stats.ldelay = | |
641 | codel_time_to_us(flow->cvars.ldelay); | |
642 | xstats.class_stats.count = flow->cvars.count; | |
643 | xstats.class_stats.lastcount = flow->cvars.lastcount; | |
644 | xstats.class_stats.dropping = flow->cvars.dropping; | |
645 | if (flow->cvars.dropping) { | |
646 | codel_tdiff_t delta = flow->cvars.drop_next - | |
647 | codel_get_time(); | |
648 | ||
649 | xstats.class_stats.drop_next = (delta >= 0) ? | |
650 | codel_time_to_us(delta) : | |
651 | -codel_time_to_us(-delta); | |
652 | } | |
edb09eb1 ED |
653 | if (flow->head) { |
654 | sch_tree_lock(sch); | |
655 | skb = flow->head; | |
656 | while (skb) { | |
657 | qs.qlen++; | |
658 | skb = skb->next; | |
659 | } | |
660 | sch_tree_unlock(sch); | |
4b549a2e ED |
661 | } |
662 | qs.backlog = q->backlogs[idx]; | |
663 | qs.drops = flow->dropped; | |
664 | } | |
aafddbf0 | 665 | if (gnet_stats_copy_queue(d, NULL, &qs, qs.qlen) < 0) |
4b549a2e ED |
666 | return -1; |
667 | if (idx < q->flows_cnt) | |
668 | return gnet_stats_copy_app(d, &xstats, sizeof(xstats)); | |
669 | return 0; | |
670 | } | |
671 | ||
672 | static void fq_codel_walk(struct Qdisc *sch, struct qdisc_walker *arg) | |
673 | { | |
674 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
675 | unsigned int i; | |
676 | ||
677 | if (arg->stop) | |
678 | return; | |
679 | ||
680 | for (i = 0; i < q->flows_cnt; i++) { | |
681 | if (list_empty(&q->flows[i].flowchain) || | |
682 | arg->count < arg->skip) { | |
683 | arg->count++; | |
684 | continue; | |
685 | } | |
686 | if (arg->fn(sch, i + 1, arg) < 0) { | |
687 | arg->stop = 1; | |
688 | break; | |
689 | } | |
690 | arg->count++; | |
691 | } | |
692 | } | |
693 | ||
694 | static const struct Qdisc_class_ops fq_codel_class_ops = { | |
695 | .leaf = fq_codel_leaf, | |
143976ce | 696 | .find = fq_codel_find, |
6529eaba | 697 | .tcf_block = fq_codel_tcf_block, |
4b549a2e | 698 | .bind_tcf = fq_codel_bind, |
143976ce | 699 | .unbind_tcf = fq_codel_unbind, |
4b549a2e ED |
700 | .dump = fq_codel_dump_class, |
701 | .dump_stats = fq_codel_dump_class_stats, | |
702 | .walk = fq_codel_walk, | |
703 | }; | |
704 | ||
705 | static struct Qdisc_ops fq_codel_qdisc_ops __read_mostly = { | |
706 | .cl_ops = &fq_codel_class_ops, | |
707 | .id = "fq_codel", | |
708 | .priv_size = sizeof(struct fq_codel_sched_data), | |
709 | .enqueue = fq_codel_enqueue, | |
710 | .dequeue = fq_codel_dequeue, | |
711 | .peek = qdisc_peek_dequeued, | |
4b549a2e ED |
712 | .init = fq_codel_init, |
713 | .reset = fq_codel_reset, | |
714 | .destroy = fq_codel_destroy, | |
715 | .change = fq_codel_change, | |
716 | .dump = fq_codel_dump, | |
717 | .dump_stats = fq_codel_dump_stats, | |
718 | .owner = THIS_MODULE, | |
719 | }; | |
720 | ||
721 | static int __init fq_codel_module_init(void) | |
722 | { | |
723 | return register_qdisc(&fq_codel_qdisc_ops); | |
724 | } | |
725 | ||
726 | static void __exit fq_codel_module_exit(void) | |
727 | { | |
728 | unregister_qdisc(&fq_codel_qdisc_ops); | |
729 | } | |
730 | ||
731 | module_init(fq_codel_module_init) | |
732 | module_exit(fq_codel_module_exit) | |
733 | MODULE_AUTHOR("Eric Dumazet"); | |
734 | MODULE_LICENSE("GPL"); |