]>
Commit | Line | Data |
---|---|---|
4b549a2e ED |
1 | /* |
2 | * Fair Queue CoDel discipline | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
80ba92fa | 9 | * Copyright (C) 2012,2015 Eric Dumazet <edumazet@google.com> |
4b549a2e ED |
10 | */ |
11 | ||
12 | #include <linux/module.h> | |
13 | #include <linux/types.h> | |
14 | #include <linux/kernel.h> | |
15 | #include <linux/jiffies.h> | |
16 | #include <linux/string.h> | |
17 | #include <linux/in.h> | |
18 | #include <linux/errno.h> | |
19 | #include <linux/init.h> | |
20 | #include <linux/skbuff.h> | |
21 | #include <linux/jhash.h> | |
22 | #include <linux/slab.h> | |
23 | #include <linux/vmalloc.h> | |
24 | #include <net/netlink.h> | |
25 | #include <net/pkt_sched.h> | |
4b549a2e | 26 | #include <net/codel.h> |
d068ca2a MK |
27 | #include <net/codel_impl.h> |
28 | #include <net/codel_qdisc.h> | |
4b549a2e ED |
29 | |
30 | /* Fair Queue CoDel. | |
31 | * | |
32 | * Principles : | |
33 | * Packets are classified (internal classifier or external) on flows. | |
34 | * This is a Stochastic model (as we use a hash, several flows | |
35 | * might be hashed on same slot) | |
36 | * Each flow has a CoDel managed queue. | |
37 | * Flows are linked onto two (Round Robin) lists, | |
38 | * so that new flows have priority on old ones. | |
39 | * | |
40 | * For a given flow, packets are not reordered (CoDel uses a FIFO) | |
41 | * head drops only. | |
42 | * ECN capability is on by default. | |
43 | * Low memory footprint (64 bytes per flow) | |
44 | */ | |
45 | ||
46 | struct fq_codel_flow { | |
47 | struct sk_buff *head; | |
48 | struct sk_buff *tail; | |
49 | struct list_head flowchain; | |
50 | int deficit; | |
51 | u32 dropped; /* number of drops (or ECN marks) on this flow */ | |
52 | struct codel_vars cvars; | |
53 | }; /* please try to keep this structure <= 64 bytes */ | |
54 | ||
55 | struct fq_codel_sched_data { | |
25d8c0d5 | 56 | struct tcf_proto __rcu *filter_list; /* optional external classifier */ |
4b549a2e ED |
57 | struct fq_codel_flow *flows; /* Flows table [flows_cnt] */ |
58 | u32 *backlogs; /* backlog table [flows_cnt] */ | |
59 | u32 flows_cnt; /* number of flows */ | |
4b549a2e | 60 | u32 quantum; /* psched_mtu(qdisc_dev(sch)); */ |
9d18562a | 61 | u32 drop_batch_size; |
95b58430 | 62 | u32 memory_limit; |
4b549a2e ED |
63 | struct codel_params cparams; |
64 | struct codel_stats cstats; | |
95b58430 ED |
65 | u32 memory_usage; |
66 | u32 drop_overmemory; | |
4b549a2e ED |
67 | u32 drop_overlimit; |
68 | u32 new_flow_count; | |
69 | ||
70 | struct list_head new_flows; /* list of new flows */ | |
71 | struct list_head old_flows; /* list of old flows */ | |
72 | }; | |
73 | ||
74 | static unsigned int fq_codel_hash(const struct fq_codel_sched_data *q, | |
342db221 | 75 | struct sk_buff *skb) |
4b549a2e | 76 | { |
264b87fa | 77 | return reciprocal_scale(skb_get_hash(skb), q->flows_cnt); |
4b549a2e ED |
78 | } |
79 | ||
80 | static unsigned int fq_codel_classify(struct sk_buff *skb, struct Qdisc *sch, | |
81 | int *qerr) | |
82 | { | |
83 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
25d8c0d5 | 84 | struct tcf_proto *filter; |
4b549a2e ED |
85 | struct tcf_result res; |
86 | int result; | |
87 | ||
88 | if (TC_H_MAJ(skb->priority) == sch->handle && | |
89 | TC_H_MIN(skb->priority) > 0 && | |
90 | TC_H_MIN(skb->priority) <= q->flows_cnt) | |
91 | return TC_H_MIN(skb->priority); | |
92 | ||
69204cf7 | 93 | filter = rcu_dereference_bh(q->filter_list); |
25d8c0d5 | 94 | if (!filter) |
4b549a2e ED |
95 | return fq_codel_hash(q, skb) + 1; |
96 | ||
97 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; | |
3b3ae880 | 98 | result = tc_classify(skb, filter, &res, false); |
4b549a2e ED |
99 | if (result >= 0) { |
100 | #ifdef CONFIG_NET_CLS_ACT | |
101 | switch (result) { | |
102 | case TC_ACT_STOLEN: | |
103 | case TC_ACT_QUEUED: | |
104 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; | |
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; | |
123 | skb->next = NULL; | |
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; | |
2ccccf5f | 290 | unsigned int prev_backlog; |
4b549a2e ED |
291 | |
292 | begin: | |
293 | head = &q->new_flows; | |
294 | if (list_empty(head)) { | |
295 | head = &q->old_flows; | |
296 | if (list_empty(head)) | |
297 | return NULL; | |
298 | } | |
299 | flow = list_first_entry(head, struct fq_codel_flow, flowchain); | |
300 | ||
301 | if (flow->deficit <= 0) { | |
302 | flow->deficit += q->quantum; | |
303 | list_move_tail(&flow->flowchain, &q->old_flows); | |
304 | goto begin; | |
305 | } | |
306 | ||
307 | prev_drop_count = q->cstats.drop_count; | |
308 | prev_ecn_mark = q->cstats.ecn_mark; | |
2ccccf5f | 309 | prev_backlog = sch->qstats.backlog; |
4b549a2e | 310 | |
79bdc4c8 MK |
311 | skb = codel_dequeue(sch, &sch->qstats.backlog, &q->cparams, |
312 | &flow->cvars, &q->cstats, qdisc_pkt_len, | |
313 | codel_get_enqueue_time, drop_func, dequeue_func); | |
4b549a2e ED |
314 | |
315 | flow->dropped += q->cstats.drop_count - prev_drop_count; | |
316 | flow->dropped += q->cstats.ecn_mark - prev_ecn_mark; | |
317 | ||
318 | if (!skb) { | |
319 | /* force a pass through old_flows to prevent starvation */ | |
320 | if ((head == &q->new_flows) && !list_empty(&q->old_flows)) | |
321 | list_move_tail(&flow->flowchain, &q->old_flows); | |
322 | else | |
323 | list_del_init(&flow->flowchain); | |
324 | goto begin; | |
325 | } | |
326 | qdisc_bstats_update(sch, skb); | |
327 | flow->deficit -= qdisc_pkt_len(skb); | |
2ccccf5f | 328 | /* We cant call qdisc_tree_reduce_backlog() if our qlen is 0, |
4b549a2e ED |
329 | * or HTB crashes. Defer it for next round. |
330 | */ | |
331 | if (q->cstats.drop_count && sch->q.qlen) { | |
2ccccf5f WC |
332 | qdisc_tree_reduce_backlog(sch, q->cstats.drop_count, |
333 | q->cstats.drop_len); | |
4b549a2e | 334 | q->cstats.drop_count = 0; |
2ccccf5f | 335 | q->cstats.drop_len = 0; |
4b549a2e ED |
336 | } |
337 | return skb; | |
338 | } | |
339 | ||
ece5d4c7 ED |
340 | static void fq_codel_flow_purge(struct fq_codel_flow *flow) |
341 | { | |
342 | rtnl_kfree_skbs(flow->head, flow->tail); | |
343 | flow->head = NULL; | |
344 | } | |
345 | ||
4b549a2e ED |
346 | static void fq_codel_reset(struct Qdisc *sch) |
347 | { | |
3d0e0af4 ED |
348 | struct fq_codel_sched_data *q = qdisc_priv(sch); |
349 | int i; | |
4b549a2e | 350 | |
3d0e0af4 ED |
351 | INIT_LIST_HEAD(&q->new_flows); |
352 | INIT_LIST_HEAD(&q->old_flows); | |
353 | for (i = 0; i < q->flows_cnt; i++) { | |
354 | struct fq_codel_flow *flow = q->flows + i; | |
355 | ||
ece5d4c7 | 356 | fq_codel_flow_purge(flow); |
3d0e0af4 ED |
357 | INIT_LIST_HEAD(&flow->flowchain); |
358 | codel_vars_init(&flow->cvars); | |
359 | } | |
360 | memset(q->backlogs, 0, q->flows_cnt * sizeof(u32)); | |
361 | sch->q.qlen = 0; | |
ece5d4c7 | 362 | sch->qstats.backlog = 0; |
77f57761 | 363 | q->memory_usage = 0; |
4b549a2e ED |
364 | } |
365 | ||
366 | static const struct nla_policy fq_codel_policy[TCA_FQ_CODEL_MAX + 1] = { | |
367 | [TCA_FQ_CODEL_TARGET] = { .type = NLA_U32 }, | |
368 | [TCA_FQ_CODEL_LIMIT] = { .type = NLA_U32 }, | |
369 | [TCA_FQ_CODEL_INTERVAL] = { .type = NLA_U32 }, | |
370 | [TCA_FQ_CODEL_ECN] = { .type = NLA_U32 }, | |
371 | [TCA_FQ_CODEL_FLOWS] = { .type = NLA_U32 }, | |
372 | [TCA_FQ_CODEL_QUANTUM] = { .type = NLA_U32 }, | |
80ba92fa | 373 | [TCA_FQ_CODEL_CE_THRESHOLD] = { .type = NLA_U32 }, |
9d18562a | 374 | [TCA_FQ_CODEL_DROP_BATCH_SIZE] = { .type = NLA_U32 }, |
95b58430 | 375 | [TCA_FQ_CODEL_MEMORY_LIMIT] = { .type = NLA_U32 }, |
4b549a2e ED |
376 | }; |
377 | ||
378 | static int fq_codel_change(struct Qdisc *sch, struct nlattr *opt) | |
379 | { | |
380 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
381 | struct nlattr *tb[TCA_FQ_CODEL_MAX + 1]; | |
382 | int err; | |
383 | ||
384 | if (!opt) | |
385 | return -EINVAL; | |
386 | ||
387 | err = nla_parse_nested(tb, TCA_FQ_CODEL_MAX, opt, fq_codel_policy); | |
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 | ||
449 | static void *fq_codel_zalloc(size_t sz) | |
450 | { | |
451 | void *ptr = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN); | |
452 | ||
453 | if (!ptr) | |
454 | ptr = vzalloc(sz); | |
455 | return ptr; | |
456 | } | |
457 | ||
458 | static void fq_codel_free(void *addr) | |
459 | { | |
4cb28970 | 460 | kvfree(addr); |
4b549a2e ED |
461 | } |
462 | ||
463 | static void fq_codel_destroy(struct Qdisc *sch) | |
464 | { | |
465 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
466 | ||
467 | tcf_destroy_chain(&q->filter_list); | |
468 | fq_codel_free(q->backlogs); | |
469 | fq_codel_free(q->flows); | |
470 | } | |
471 | ||
472 | static int fq_codel_init(struct Qdisc *sch, struct nlattr *opt) | |
473 | { | |
474 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
475 | int i; | |
476 | ||
477 | sch->limit = 10*1024; | |
478 | q->flows_cnt = 1024; | |
95b58430 | 479 | q->memory_limit = 32 << 20; /* 32 MBytes */ |
9d18562a | 480 | q->drop_batch_size = 64; |
4b549a2e | 481 | q->quantum = psched_mtu(qdisc_dev(sch)); |
4b549a2e ED |
482 | INIT_LIST_HEAD(&q->new_flows); |
483 | INIT_LIST_HEAD(&q->old_flows); | |
79bdc4c8 | 484 | codel_params_init(&q->cparams); |
4b549a2e ED |
485 | codel_stats_init(&q->cstats); |
486 | q->cparams.ecn = true; | |
79bdc4c8 | 487 | q->cparams.mtu = psched_mtu(qdisc_dev(sch)); |
4b549a2e ED |
488 | |
489 | if (opt) { | |
490 | int err = fq_codel_change(sch, opt); | |
491 | if (err) | |
492 | return err; | |
493 | } | |
494 | ||
495 | if (!q->flows) { | |
496 | q->flows = fq_codel_zalloc(q->flows_cnt * | |
497 | sizeof(struct fq_codel_flow)); | |
498 | if (!q->flows) | |
499 | return -ENOMEM; | |
500 | q->backlogs = fq_codel_zalloc(q->flows_cnt * sizeof(u32)); | |
501 | if (!q->backlogs) { | |
502 | fq_codel_free(q->flows); | |
503 | return -ENOMEM; | |
504 | } | |
505 | for (i = 0; i < q->flows_cnt; i++) { | |
506 | struct fq_codel_flow *flow = q->flows + i; | |
507 | ||
508 | INIT_LIST_HEAD(&flow->flowchain); | |
b379135c | 509 | codel_vars_init(&flow->cvars); |
4b549a2e ED |
510 | } |
511 | } | |
512 | if (sch->limit >= 1) | |
513 | sch->flags |= TCQ_F_CAN_BYPASS; | |
514 | else | |
515 | sch->flags &= ~TCQ_F_CAN_BYPASS; | |
516 | return 0; | |
517 | } | |
518 | ||
519 | static int fq_codel_dump(struct Qdisc *sch, struct sk_buff *skb) | |
520 | { | |
521 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
522 | struct nlattr *opts; | |
523 | ||
524 | opts = nla_nest_start(skb, TCA_OPTIONS); | |
525 | if (opts == NULL) | |
526 | goto nla_put_failure; | |
527 | ||
528 | if (nla_put_u32(skb, TCA_FQ_CODEL_TARGET, | |
529 | codel_time_to_us(q->cparams.target)) || | |
530 | nla_put_u32(skb, TCA_FQ_CODEL_LIMIT, | |
531 | sch->limit) || | |
532 | nla_put_u32(skb, TCA_FQ_CODEL_INTERVAL, | |
533 | codel_time_to_us(q->cparams.interval)) || | |
534 | nla_put_u32(skb, TCA_FQ_CODEL_ECN, | |
535 | q->cparams.ecn) || | |
536 | nla_put_u32(skb, TCA_FQ_CODEL_QUANTUM, | |
537 | q->quantum) || | |
9d18562a ED |
538 | nla_put_u32(skb, TCA_FQ_CODEL_DROP_BATCH_SIZE, |
539 | q->drop_batch_size) || | |
95b58430 ED |
540 | nla_put_u32(skb, TCA_FQ_CODEL_MEMORY_LIMIT, |
541 | q->memory_limit) || | |
4b549a2e ED |
542 | nla_put_u32(skb, TCA_FQ_CODEL_FLOWS, |
543 | q->flows_cnt)) | |
544 | goto nla_put_failure; | |
545 | ||
80ba92fa ED |
546 | if (q->cparams.ce_threshold != CODEL_DISABLED_THRESHOLD && |
547 | nla_put_u32(skb, TCA_FQ_CODEL_CE_THRESHOLD, | |
548 | codel_time_to_us(q->cparams.ce_threshold))) | |
549 | goto nla_put_failure; | |
550 | ||
d59b7d80 | 551 | return nla_nest_end(skb, opts); |
4b549a2e ED |
552 | |
553 | nla_put_failure: | |
554 | return -1; | |
555 | } | |
556 | ||
557 | static int fq_codel_dump_stats(struct Qdisc *sch, struct gnet_dump *d) | |
558 | { | |
559 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
560 | struct tc_fq_codel_xstats st = { | |
561 | .type = TCA_FQ_CODEL_XSTATS_QDISC, | |
4b549a2e ED |
562 | }; |
563 | struct list_head *pos; | |
564 | ||
669d67bf SL |
565 | st.qdisc_stats.maxpacket = q->cstats.maxpacket; |
566 | st.qdisc_stats.drop_overlimit = q->drop_overlimit; | |
567 | st.qdisc_stats.ecn_mark = q->cstats.ecn_mark; | |
568 | st.qdisc_stats.new_flow_count = q->new_flow_count; | |
80ba92fa | 569 | st.qdisc_stats.ce_mark = q->cstats.ce_mark; |
95b58430 ED |
570 | st.qdisc_stats.memory_usage = q->memory_usage; |
571 | st.qdisc_stats.drop_overmemory = q->drop_overmemory; | |
669d67bf | 572 | |
edb09eb1 | 573 | sch_tree_lock(sch); |
4b549a2e ED |
574 | list_for_each(pos, &q->new_flows) |
575 | st.qdisc_stats.new_flows_len++; | |
576 | ||
577 | list_for_each(pos, &q->old_flows) | |
578 | st.qdisc_stats.old_flows_len++; | |
edb09eb1 | 579 | sch_tree_unlock(sch); |
4b549a2e ED |
580 | |
581 | return gnet_stats_copy_app(d, &st, sizeof(st)); | |
582 | } | |
583 | ||
584 | static struct Qdisc *fq_codel_leaf(struct Qdisc *sch, unsigned long arg) | |
585 | { | |
586 | return NULL; | |
587 | } | |
588 | ||
589 | static unsigned long fq_codel_get(struct Qdisc *sch, u32 classid) | |
590 | { | |
591 | return 0; | |
592 | } | |
593 | ||
594 | static unsigned long fq_codel_bind(struct Qdisc *sch, unsigned long parent, | |
595 | u32 classid) | |
596 | { | |
597 | /* we cannot bypass queue discipline anymore */ | |
598 | sch->flags &= ~TCQ_F_CAN_BYPASS; | |
599 | return 0; | |
600 | } | |
601 | ||
602 | static void fq_codel_put(struct Qdisc *q, unsigned long cl) | |
603 | { | |
604 | } | |
605 | ||
25d8c0d5 JF |
606 | static struct tcf_proto __rcu **fq_codel_find_tcf(struct Qdisc *sch, |
607 | unsigned long cl) | |
4b549a2e ED |
608 | { |
609 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
610 | ||
611 | if (cl) | |
612 | return NULL; | |
613 | return &q->filter_list; | |
614 | } | |
615 | ||
616 | static int fq_codel_dump_class(struct Qdisc *sch, unsigned long cl, | |
617 | struct sk_buff *skb, struct tcmsg *tcm) | |
618 | { | |
619 | tcm->tcm_handle |= TC_H_MIN(cl); | |
620 | return 0; | |
621 | } | |
622 | ||
623 | static int fq_codel_dump_class_stats(struct Qdisc *sch, unsigned long cl, | |
624 | struct gnet_dump *d) | |
625 | { | |
626 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
627 | u32 idx = cl - 1; | |
628 | struct gnet_stats_queue qs = { 0 }; | |
629 | struct tc_fq_codel_xstats xstats; | |
630 | ||
631 | if (idx < q->flows_cnt) { | |
632 | const struct fq_codel_flow *flow = &q->flows[idx]; | |
edb09eb1 | 633 | const struct sk_buff *skb; |
4b549a2e ED |
634 | |
635 | memset(&xstats, 0, sizeof(xstats)); | |
636 | xstats.type = TCA_FQ_CODEL_XSTATS_CLASS; | |
637 | xstats.class_stats.deficit = flow->deficit; | |
638 | xstats.class_stats.ldelay = | |
639 | codel_time_to_us(flow->cvars.ldelay); | |
640 | xstats.class_stats.count = flow->cvars.count; | |
641 | xstats.class_stats.lastcount = flow->cvars.lastcount; | |
642 | xstats.class_stats.dropping = flow->cvars.dropping; | |
643 | if (flow->cvars.dropping) { | |
644 | codel_tdiff_t delta = flow->cvars.drop_next - | |
645 | codel_get_time(); | |
646 | ||
647 | xstats.class_stats.drop_next = (delta >= 0) ? | |
648 | codel_time_to_us(delta) : | |
649 | -codel_time_to_us(-delta); | |
650 | } | |
edb09eb1 ED |
651 | if (flow->head) { |
652 | sch_tree_lock(sch); | |
653 | skb = flow->head; | |
654 | while (skb) { | |
655 | qs.qlen++; | |
656 | skb = skb->next; | |
657 | } | |
658 | sch_tree_unlock(sch); | |
4b549a2e ED |
659 | } |
660 | qs.backlog = q->backlogs[idx]; | |
661 | qs.drops = flow->dropped; | |
662 | } | |
aafddbf0 | 663 | if (gnet_stats_copy_queue(d, NULL, &qs, qs.qlen) < 0) |
4b549a2e ED |
664 | return -1; |
665 | if (idx < q->flows_cnt) | |
666 | return gnet_stats_copy_app(d, &xstats, sizeof(xstats)); | |
667 | return 0; | |
668 | } | |
669 | ||
670 | static void fq_codel_walk(struct Qdisc *sch, struct qdisc_walker *arg) | |
671 | { | |
672 | struct fq_codel_sched_data *q = qdisc_priv(sch); | |
673 | unsigned int i; | |
674 | ||
675 | if (arg->stop) | |
676 | return; | |
677 | ||
678 | for (i = 0; i < q->flows_cnt; i++) { | |
679 | if (list_empty(&q->flows[i].flowchain) || | |
680 | arg->count < arg->skip) { | |
681 | arg->count++; | |
682 | continue; | |
683 | } | |
684 | if (arg->fn(sch, i + 1, arg) < 0) { | |
685 | arg->stop = 1; | |
686 | break; | |
687 | } | |
688 | arg->count++; | |
689 | } | |
690 | } | |
691 | ||
692 | static const struct Qdisc_class_ops fq_codel_class_ops = { | |
693 | .leaf = fq_codel_leaf, | |
694 | .get = fq_codel_get, | |
695 | .put = fq_codel_put, | |
696 | .tcf_chain = fq_codel_find_tcf, | |
697 | .bind_tcf = fq_codel_bind, | |
698 | .unbind_tcf = fq_codel_put, | |
699 | .dump = fq_codel_dump_class, | |
700 | .dump_stats = fq_codel_dump_class_stats, | |
701 | .walk = fq_codel_walk, | |
702 | }; | |
703 | ||
704 | static struct Qdisc_ops fq_codel_qdisc_ops __read_mostly = { | |
705 | .cl_ops = &fq_codel_class_ops, | |
706 | .id = "fq_codel", | |
707 | .priv_size = sizeof(struct fq_codel_sched_data), | |
708 | .enqueue = fq_codel_enqueue, | |
709 | .dequeue = fq_codel_dequeue, | |
710 | .peek = qdisc_peek_dequeued, | |
4b549a2e ED |
711 | .init = fq_codel_init, |
712 | .reset = fq_codel_reset, | |
713 | .destroy = fq_codel_destroy, | |
714 | .change = fq_codel_change, | |
715 | .dump = fq_codel_dump, | |
716 | .dump_stats = fq_codel_dump_stats, | |
717 | .owner = THIS_MODULE, | |
718 | }; | |
719 | ||
720 | static int __init fq_codel_module_init(void) | |
721 | { | |
722 | return register_qdisc(&fq_codel_qdisc_ops); | |
723 | } | |
724 | ||
725 | static void __exit fq_codel_module_exit(void) | |
726 | { | |
727 | unregister_qdisc(&fq_codel_qdisc_ops); | |
728 | } | |
729 | ||
730 | module_init(fq_codel_module_init) | |
731 | module_exit(fq_codel_module_exit) | |
732 | MODULE_AUTHOR("Eric Dumazet"); | |
733 | MODULE_LICENSE("GPL"); |