Thomas Gleixner | d2912cb | 2019-06-04 10:11:33 +0200 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0-only |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 2 | /* |
| 3 | * net/sched/sch_choke.c CHOKE scheduler |
| 4 | * |
| 5 | * Copyright (c) 2011 Stephen Hemminger <shemminger@vyatta.com> |
| 6 | * Copyright (c) 2011 Eric Dumazet <eric.dumazet@gmail.com> |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 7 | */ |
| 8 | |
| 9 | #include <linux/module.h> |
| 10 | #include <linux/types.h> |
| 11 | #include <linux/kernel.h> |
| 12 | #include <linux/skbuff.h> |
David S. Miller | cdfb74d | 2011-02-02 23:06:31 -0800 | [diff] [blame] | 13 | #include <linux/vmalloc.h> |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 14 | #include <net/pkt_sched.h> |
Jiri Pirko | cf1facd | 2017-02-09 14:38:56 +0100 | [diff] [blame] | 15 | #include <net/pkt_cls.h> |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 16 | #include <net/inet_ecn.h> |
| 17 | #include <net/red.h> |
Jiri Pirko | 1bd758e | 2015-05-12 14:56:07 +0200 | [diff] [blame] | 18 | #include <net/flow_dissector.h> |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 19 | |
| 20 | /* |
| 21 | CHOKe stateless AQM for fair bandwidth allocation |
| 22 | ================================================= |
| 23 | |
| 24 | CHOKe (CHOose and Keep for responsive flows, CHOose and Kill for |
| 25 | unresponsive flows) is a variant of RED that penalizes misbehaving flows but |
| 26 | maintains no flow state. The difference from RED is an additional step |
| 27 | during the enqueuing process. If average queue size is over the |
| 28 | low threshold (qmin), a packet is chosen at random from the queue. |
| 29 | If both the new and chosen packet are from the same flow, both |
| 30 | are dropped. Unlike RED, CHOKe is not really a "classful" qdisc because it |
| 31 | needs to access packets in queue randomly. It has a minimal class |
| 32 | interface to allow overriding the builtin flow classifier with |
| 33 | filters. |
| 34 | |
| 35 | Source: |
| 36 | R. Pan, B. Prabhakar, and K. Psounis, "CHOKe, A Stateless |
| 37 | Active Queue Management Scheme for Approximating Fair Bandwidth Allocation", |
| 38 | IEEE INFOCOM, 2000. |
| 39 | |
| 40 | A. Tang, J. Wang, S. Low, "Understanding CHOKe: Throughput and Spatial |
| 41 | Characteristics", IEEE/ACM Transactions on Networking, 2004 |
| 42 | |
| 43 | */ |
| 44 | |
| 45 | /* Upper bound on size of sk_buff table (packets) */ |
| 46 | #define CHOKE_MAX_QUEUE (128*1024 - 1) |
| 47 | |
| 48 | struct choke_sched_data { |
| 49 | /* Parameters */ |
| 50 | u32 limit; |
| 51 | unsigned char flags; |
| 52 | |
| 53 | struct red_parms parms; |
| 54 | |
| 55 | /* Variables */ |
Eric Dumazet | eeca668 | 2012-01-05 02:25:16 +0000 | [diff] [blame] | 56 | struct red_vars vars; |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 57 | struct { |
| 58 | u32 prob_drop; /* Early probability drops */ |
| 59 | u32 prob_mark; /* Early probability marks */ |
| 60 | u32 forced_drop; /* Forced drops, qavg > max_thresh */ |
| 61 | u32 forced_mark; /* Forced marks, qavg > max_thresh */ |
| 62 | u32 pdrop; /* Drops due to queue limits */ |
| 63 | u32 other; /* Drops due to drop() calls */ |
| 64 | u32 matched; /* Drops to flow match */ |
| 65 | } stats; |
| 66 | |
| 67 | unsigned int head; |
| 68 | unsigned int tail; |
| 69 | |
| 70 | unsigned int tab_mask; /* size - 1 */ |
| 71 | |
| 72 | struct sk_buff **tab; |
| 73 | }; |
| 74 | |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 75 | /* number of elements in queue including holes */ |
| 76 | static unsigned int choke_len(const struct choke_sched_data *q) |
| 77 | { |
| 78 | return (q->tail - q->head) & q->tab_mask; |
| 79 | } |
| 80 | |
| 81 | /* Is ECN parameter configured */ |
| 82 | static int use_ecn(const struct choke_sched_data *q) |
| 83 | { |
| 84 | return q->flags & TC_RED_ECN; |
| 85 | } |
| 86 | |
| 87 | /* Should packets over max just be dropped (versus marked) */ |
| 88 | static int use_harddrop(const struct choke_sched_data *q) |
| 89 | { |
| 90 | return q->flags & TC_RED_HARDDROP; |
| 91 | } |
| 92 | |
| 93 | /* Move head pointer forward to skip over holes */ |
| 94 | static void choke_zap_head_holes(struct choke_sched_data *q) |
| 95 | { |
| 96 | do { |
| 97 | q->head = (q->head + 1) & q->tab_mask; |
| 98 | if (q->head == q->tail) |
| 99 | break; |
| 100 | } while (q->tab[q->head] == NULL); |
| 101 | } |
| 102 | |
| 103 | /* Move tail pointer backwards to reuse holes */ |
| 104 | static void choke_zap_tail_holes(struct choke_sched_data *q) |
| 105 | { |
| 106 | do { |
| 107 | q->tail = (q->tail - 1) & q->tab_mask; |
| 108 | if (q->head == q->tail) |
| 109 | break; |
| 110 | } while (q->tab[q->tail] == NULL); |
| 111 | } |
| 112 | |
| 113 | /* Drop packet from queue array by creating a "hole" */ |
Eric Dumazet | 520ac30 | 2016-06-21 23:16:49 -0700 | [diff] [blame] | 114 | static void choke_drop_by_idx(struct Qdisc *sch, unsigned int idx, |
| 115 | struct sk_buff **to_free) |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 116 | { |
| 117 | struct choke_sched_data *q = qdisc_priv(sch); |
| 118 | struct sk_buff *skb = q->tab[idx]; |
| 119 | |
| 120 | q->tab[idx] = NULL; |
| 121 | |
| 122 | if (idx == q->head) |
| 123 | choke_zap_head_holes(q); |
| 124 | if (idx == q->tail) |
| 125 | choke_zap_tail_holes(q); |
| 126 | |
John Fastabend | 25331d6 | 2014-09-28 11:53:29 -0700 | [diff] [blame] | 127 | qdisc_qstats_backlog_dec(sch, skb); |
WANG Cong | 2ccccf5 | 2016-02-25 14:55:01 -0800 | [diff] [blame] | 128 | qdisc_tree_reduce_backlog(sch, 1, qdisc_pkt_len(skb)); |
Eric Dumazet | 520ac30 | 2016-06-21 23:16:49 -0700 | [diff] [blame] | 129 | qdisc_drop(skb, sch, to_free); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 130 | --sch->q.qlen; |
| 131 | } |
| 132 | |
Eric Dumazet | 26f70e1 | 2011-02-24 17:45:41 +0000 | [diff] [blame] | 133 | struct choke_skb_cb { |
Eric Dumazet | 2bcc34b | 2011-11-29 04:22:15 +0000 | [diff] [blame] | 134 | u8 keys_valid; |
Tom Herbert | 2e99403 | 2015-05-01 11:30:18 -0700 | [diff] [blame] | 135 | struct flow_keys_digest keys; |
Eric Dumazet | 26f70e1 | 2011-02-24 17:45:41 +0000 | [diff] [blame] | 136 | }; |
| 137 | |
| 138 | static inline struct choke_skb_cb *choke_skb_cb(const struct sk_buff *skb) |
| 139 | { |
David S. Miller | 16bda13 | 2012-02-06 15:14:37 -0500 | [diff] [blame] | 140 | qdisc_cb_private_validate(skb, sizeof(struct choke_skb_cb)); |
Eric Dumazet | 26f70e1 | 2011-02-24 17:45:41 +0000 | [diff] [blame] | 141 | return (struct choke_skb_cb *)qdisc_skb_cb(skb)->data; |
| 142 | } |
| 143 | |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 144 | /* |
Eric Dumazet | 2bcc34b | 2011-11-29 04:22:15 +0000 | [diff] [blame] | 145 | * Compare flow of two packets |
| 146 | * Returns true only if source and destination address and port match. |
| 147 | * false for special cases |
| 148 | */ |
| 149 | static bool choke_match_flow(struct sk_buff *skb1, |
| 150 | struct sk_buff *skb2) |
| 151 | { |
Eric Dumazet | 2571178 | 2014-09-18 08:02:05 -0700 | [diff] [blame] | 152 | struct flow_keys temp; |
| 153 | |
Eric Dumazet | 2bcc34b | 2011-11-29 04:22:15 +0000 | [diff] [blame] | 154 | if (skb1->protocol != skb2->protocol) |
| 155 | return false; |
| 156 | |
| 157 | if (!choke_skb_cb(skb1)->keys_valid) { |
| 158 | choke_skb_cb(skb1)->keys_valid = 1; |
Tom Herbert | cd79a23 | 2015-09-01 09:24:27 -0700 | [diff] [blame] | 159 | skb_flow_dissect_flow_keys(skb1, &temp, 0); |
Tom Herbert | 2e99403 | 2015-05-01 11:30:18 -0700 | [diff] [blame] | 160 | make_flow_keys_digest(&choke_skb_cb(skb1)->keys, &temp); |
Eric Dumazet | 2bcc34b | 2011-11-29 04:22:15 +0000 | [diff] [blame] | 161 | } |
| 162 | |
| 163 | if (!choke_skb_cb(skb2)->keys_valid) { |
| 164 | choke_skb_cb(skb2)->keys_valid = 1; |
Tom Herbert | cd79a23 | 2015-09-01 09:24:27 -0700 | [diff] [blame] | 165 | skb_flow_dissect_flow_keys(skb2, &temp, 0); |
Tom Herbert | 2e99403 | 2015-05-01 11:30:18 -0700 | [diff] [blame] | 166 | make_flow_keys_digest(&choke_skb_cb(skb2)->keys, &temp); |
Eric Dumazet | 2bcc34b | 2011-11-29 04:22:15 +0000 | [diff] [blame] | 167 | } |
| 168 | |
| 169 | return !memcmp(&choke_skb_cb(skb1)->keys, |
| 170 | &choke_skb_cb(skb2)->keys, |
Tom Herbert | 2e99403 | 2015-05-01 11:30:18 -0700 | [diff] [blame] | 171 | sizeof(choke_skb_cb(skb1)->keys)); |
Eric Dumazet | 2bcc34b | 2011-11-29 04:22:15 +0000 | [diff] [blame] | 172 | } |
| 173 | |
| 174 | /* |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 175 | * Select a packet at random from queue |
| 176 | * HACK: since queue can have holes from previous deletion; retry several |
| 177 | * times to find a random skb but then just give up and return the head |
| 178 | * Will return NULL if queue is empty (q->head == q->tail) |
| 179 | */ |
| 180 | static struct sk_buff *choke_peek_random(const struct choke_sched_data *q, |
| 181 | unsigned int *pidx) |
| 182 | { |
| 183 | struct sk_buff *skb; |
| 184 | int retrys = 3; |
| 185 | |
| 186 | do { |
Daniel Borkmann | f337db6 | 2014-01-22 02:29:39 +0100 | [diff] [blame] | 187 | *pidx = (q->head + prandom_u32_max(choke_len(q))) & q->tab_mask; |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 188 | skb = q->tab[*pidx]; |
| 189 | if (skb) |
| 190 | return skb; |
| 191 | } while (--retrys > 0); |
| 192 | |
| 193 | return q->tab[*pidx = q->head]; |
| 194 | } |
| 195 | |
| 196 | /* |
| 197 | * Compare new packet with random packet in queue |
| 198 | * returns true if matched and sets *pidx |
| 199 | */ |
| 200 | static bool choke_match_random(const struct choke_sched_data *q, |
| 201 | struct sk_buff *nskb, |
| 202 | unsigned int *pidx) |
| 203 | { |
| 204 | struct sk_buff *oskb; |
| 205 | |
| 206 | if (q->head == q->tail) |
| 207 | return false; |
| 208 | |
| 209 | oskb = choke_peek_random(q, pidx); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 210 | return choke_match_flow(oskb, nskb); |
| 211 | } |
| 212 | |
Eric Dumazet | 520ac30 | 2016-06-21 23:16:49 -0700 | [diff] [blame] | 213 | static int choke_enqueue(struct sk_buff *skb, struct Qdisc *sch, |
| 214 | struct sk_buff **to_free) |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 215 | { |
| 216 | struct choke_sched_data *q = qdisc_priv(sch); |
Eric Dumazet | eeca668 | 2012-01-05 02:25:16 +0000 | [diff] [blame] | 217 | const struct red_parms *p = &q->parms; |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 218 | |
Eric Dumazet | 2bcc34b | 2011-11-29 04:22:15 +0000 | [diff] [blame] | 219 | choke_skb_cb(skb)->keys_valid = 0; |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 220 | /* Compute average queue usage (see RED) */ |
Eric Dumazet | eeca668 | 2012-01-05 02:25:16 +0000 | [diff] [blame] | 221 | q->vars.qavg = red_calc_qavg(p, &q->vars, sch->q.qlen); |
| 222 | if (red_is_idling(&q->vars)) |
| 223 | red_end_of_idle_period(&q->vars); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 224 | |
| 225 | /* Is queue small? */ |
Eric Dumazet | eeca668 | 2012-01-05 02:25:16 +0000 | [diff] [blame] | 226 | if (q->vars.qavg <= p->qth_min) |
| 227 | q->vars.qcount = -1; |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 228 | else { |
| 229 | unsigned int idx; |
| 230 | |
| 231 | /* Draw a packet at random from queue and compare flow */ |
| 232 | if (choke_match_random(q, skb, &idx)) { |
| 233 | q->stats.matched++; |
Eric Dumazet | 520ac30 | 2016-06-21 23:16:49 -0700 | [diff] [blame] | 234 | choke_drop_by_idx(sch, idx, to_free); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 235 | goto congestion_drop; |
| 236 | } |
| 237 | |
| 238 | /* Queue is large, always mark/drop */ |
Eric Dumazet | eeca668 | 2012-01-05 02:25:16 +0000 | [diff] [blame] | 239 | if (q->vars.qavg > p->qth_max) { |
| 240 | q->vars.qcount = -1; |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 241 | |
John Fastabend | 25331d6 | 2014-09-28 11:53:29 -0700 | [diff] [blame] | 242 | qdisc_qstats_overlimit(sch); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 243 | if (use_harddrop(q) || !use_ecn(q) || |
| 244 | !INET_ECN_set_ce(skb)) { |
| 245 | q->stats.forced_drop++; |
| 246 | goto congestion_drop; |
| 247 | } |
| 248 | |
| 249 | q->stats.forced_mark++; |
Eric Dumazet | eeca668 | 2012-01-05 02:25:16 +0000 | [diff] [blame] | 250 | } else if (++q->vars.qcount) { |
| 251 | if (red_mark_probability(p, &q->vars, q->vars.qavg)) { |
| 252 | q->vars.qcount = 0; |
| 253 | q->vars.qR = red_random(p); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 254 | |
John Fastabend | 25331d6 | 2014-09-28 11:53:29 -0700 | [diff] [blame] | 255 | qdisc_qstats_overlimit(sch); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 256 | if (!use_ecn(q) || !INET_ECN_set_ce(skb)) { |
| 257 | q->stats.prob_drop++; |
| 258 | goto congestion_drop; |
| 259 | } |
| 260 | |
| 261 | q->stats.prob_mark++; |
| 262 | } |
| 263 | } else |
Eric Dumazet | eeca668 | 2012-01-05 02:25:16 +0000 | [diff] [blame] | 264 | q->vars.qR = red_random(p); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 265 | } |
| 266 | |
| 267 | /* Admit new packet */ |
| 268 | if (sch->q.qlen < q->limit) { |
| 269 | q->tab[q->tail] = skb; |
| 270 | q->tail = (q->tail + 1) & q->tab_mask; |
| 271 | ++sch->q.qlen; |
John Fastabend | 25331d6 | 2014-09-28 11:53:29 -0700 | [diff] [blame] | 272 | qdisc_qstats_backlog_inc(sch, skb); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 273 | return NET_XMIT_SUCCESS; |
| 274 | } |
| 275 | |
| 276 | q->stats.pdrop++; |
Eric Dumazet | 520ac30 | 2016-06-21 23:16:49 -0700 | [diff] [blame] | 277 | return qdisc_drop(skb, sch, to_free); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 278 | |
Eric Dumazet | 1704575 | 2012-05-04 04:37:21 +0000 | [diff] [blame] | 279 | congestion_drop: |
Eric Dumazet | 520ac30 | 2016-06-21 23:16:49 -0700 | [diff] [blame] | 280 | qdisc_drop(skb, sch, to_free); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 281 | return NET_XMIT_CN; |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 282 | } |
| 283 | |
| 284 | static struct sk_buff *choke_dequeue(struct Qdisc *sch) |
| 285 | { |
| 286 | struct choke_sched_data *q = qdisc_priv(sch); |
| 287 | struct sk_buff *skb; |
| 288 | |
| 289 | if (q->head == q->tail) { |
Eric Dumazet | eeca668 | 2012-01-05 02:25:16 +0000 | [diff] [blame] | 290 | if (!red_is_idling(&q->vars)) |
| 291 | red_start_of_idle_period(&q->vars); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 292 | return NULL; |
| 293 | } |
| 294 | |
| 295 | skb = q->tab[q->head]; |
| 296 | q->tab[q->head] = NULL; |
| 297 | choke_zap_head_holes(q); |
| 298 | --sch->q.qlen; |
John Fastabend | 25331d6 | 2014-09-28 11:53:29 -0700 | [diff] [blame] | 299 | qdisc_qstats_backlog_dec(sch, skb); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 300 | qdisc_bstats_update(sch, skb); |
| 301 | |
| 302 | return skb; |
| 303 | } |
| 304 | |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 305 | static void choke_reset(struct Qdisc *sch) |
| 306 | { |
| 307 | struct choke_sched_data *q = qdisc_priv(sch); |
| 308 | |
WANG Cong | 77e62da | 2015-07-21 16:52:43 -0700 | [diff] [blame] | 309 | while (q->head != q->tail) { |
| 310 | struct sk_buff *skb = q->tab[q->head]; |
| 311 | |
| 312 | q->head = (q->head + 1) & q->tab_mask; |
| 313 | if (!skb) |
| 314 | continue; |
Eric Dumazet | f9aed31 | 2016-06-13 20:21:51 -0700 | [diff] [blame] | 315 | rtnl_qdisc_drop(skb, sch); |
WANG Cong | 77e62da | 2015-07-21 16:52:43 -0700 | [diff] [blame] | 316 | } |
| 317 | |
Eric Dumazet | f9aed31 | 2016-06-13 20:21:51 -0700 | [diff] [blame] | 318 | sch->q.qlen = 0; |
| 319 | sch->qstats.backlog = 0; |
Eric Dumazet | 8738c85 | 2020-04-25 15:19:51 -0700 | [diff] [blame] | 320 | if (q->tab) |
| 321 | memset(q->tab, 0, (q->tab_mask + 1) * sizeof(struct sk_buff *)); |
WANG Cong | 77e62da | 2015-07-21 16:52:43 -0700 | [diff] [blame] | 322 | q->head = q->tail = 0; |
Eric Dumazet | eeca668 | 2012-01-05 02:25:16 +0000 | [diff] [blame] | 323 | red_restart(&q->vars); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 324 | } |
| 325 | |
| 326 | static const struct nla_policy choke_policy[TCA_CHOKE_MAX + 1] = { |
| 327 | [TCA_CHOKE_PARMS] = { .len = sizeof(struct tc_red_qopt) }, |
| 328 | [TCA_CHOKE_STAB] = { .len = RED_STAB_SIZE }, |
Eric Dumazet | a73ed26 | 2011-12-09 02:46:45 +0000 | [diff] [blame] | 329 | [TCA_CHOKE_MAX_P] = { .type = NLA_U32 }, |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 330 | }; |
| 331 | |
| 332 | |
| 333 | static void choke_free(void *addr) |
| 334 | { |
WANG Cong | 4cb2897 | 2014-06-02 15:55:22 -0700 | [diff] [blame] | 335 | kvfree(addr); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 336 | } |
| 337 | |
Alexander Aring | 2030721 | 2017-12-20 12:35:14 -0500 | [diff] [blame] | 338 | static int choke_change(struct Qdisc *sch, struct nlattr *opt, |
| 339 | struct netlink_ext_ack *extack) |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 340 | { |
| 341 | struct choke_sched_data *q = qdisc_priv(sch); |
| 342 | struct nlattr *tb[TCA_CHOKE_MAX + 1]; |
| 343 | const struct tc_red_qopt *ctl; |
| 344 | int err; |
| 345 | struct sk_buff **old = NULL; |
| 346 | unsigned int mask; |
Eric Dumazet | a73ed26 | 2011-12-09 02:46:45 +0000 | [diff] [blame] | 347 | u32 max_P; |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 348 | |
| 349 | if (opt == NULL) |
| 350 | return -EINVAL; |
| 351 | |
Johannes Berg | 8cb0817 | 2019-04-26 14:07:28 +0200 | [diff] [blame] | 352 | err = nla_parse_nested_deprecated(tb, TCA_CHOKE_MAX, opt, |
| 353 | choke_policy, NULL); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 354 | if (err < 0) |
| 355 | return err; |
| 356 | |
| 357 | if (tb[TCA_CHOKE_PARMS] == NULL || |
| 358 | tb[TCA_CHOKE_STAB] == NULL) |
| 359 | return -EINVAL; |
| 360 | |
Eric Dumazet | a73ed26 | 2011-12-09 02:46:45 +0000 | [diff] [blame] | 361 | max_P = tb[TCA_CHOKE_MAX_P] ? nla_get_u32(tb[TCA_CHOKE_MAX_P]) : 0; |
| 362 | |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 363 | ctl = nla_data(tb[TCA_CHOKE_PARMS]); |
| 364 | |
Nogah Frankel | 8afa10c | 2017-12-04 13:31:11 +0200 | [diff] [blame] | 365 | if (!red_check_params(ctl->qth_min, ctl->qth_max, ctl->Wlog)) |
| 366 | return -EINVAL; |
| 367 | |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 368 | if (ctl->limit > CHOKE_MAX_QUEUE) |
| 369 | return -EINVAL; |
| 370 | |
| 371 | mask = roundup_pow_of_two(ctl->limit + 1) - 1; |
| 372 | if (mask != q->tab_mask) { |
| 373 | struct sk_buff **ntab; |
| 374 | |
Joe Perches | 793da4b | 2020-01-28 11:12:03 -0800 | [diff] [blame] | 375 | ntab = kvcalloc(mask + 1, sizeof(struct sk_buff *), GFP_KERNEL); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 376 | if (!ntab) |
| 377 | return -ENOMEM; |
| 378 | |
| 379 | sch_tree_lock(sch); |
| 380 | old = q->tab; |
| 381 | if (old) { |
| 382 | unsigned int oqlen = sch->q.qlen, tail = 0; |
WANG Cong | 2ccccf5 | 2016-02-25 14:55:01 -0800 | [diff] [blame] | 383 | unsigned dropped = 0; |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 384 | |
| 385 | while (q->head != q->tail) { |
| 386 | struct sk_buff *skb = q->tab[q->head]; |
| 387 | |
| 388 | q->head = (q->head + 1) & q->tab_mask; |
| 389 | if (!skb) |
| 390 | continue; |
| 391 | if (tail < mask) { |
| 392 | ntab[tail++] = skb; |
| 393 | continue; |
| 394 | } |
WANG Cong | 2ccccf5 | 2016-02-25 14:55:01 -0800 | [diff] [blame] | 395 | dropped += qdisc_pkt_len(skb); |
John Fastabend | 25331d6 | 2014-09-28 11:53:29 -0700 | [diff] [blame] | 396 | qdisc_qstats_backlog_dec(sch, skb); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 397 | --sch->q.qlen; |
Eric Dumazet | f9aed31 | 2016-06-13 20:21:51 -0700 | [diff] [blame] | 398 | rtnl_qdisc_drop(skb, sch); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 399 | } |
WANG Cong | 2ccccf5 | 2016-02-25 14:55:01 -0800 | [diff] [blame] | 400 | qdisc_tree_reduce_backlog(sch, oqlen - sch->q.qlen, dropped); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 401 | q->head = 0; |
| 402 | q->tail = tail; |
| 403 | } |
| 404 | |
| 405 | q->tab_mask = mask; |
| 406 | q->tab = ntab; |
| 407 | } else |
| 408 | sch_tree_lock(sch); |
| 409 | |
| 410 | q->flags = ctl->flags; |
| 411 | q->limit = ctl->limit; |
| 412 | |
| 413 | red_set_parms(&q->parms, ctl->qth_min, ctl->qth_max, ctl->Wlog, |
| 414 | ctl->Plog, ctl->Scell_log, |
Eric Dumazet | a73ed26 | 2011-12-09 02:46:45 +0000 | [diff] [blame] | 415 | nla_data(tb[TCA_CHOKE_STAB]), |
| 416 | max_P); |
Eric Dumazet | eeca668 | 2012-01-05 02:25:16 +0000 | [diff] [blame] | 417 | red_set_vars(&q->vars); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 418 | |
| 419 | if (q->head == q->tail) |
Eric Dumazet | eeca668 | 2012-01-05 02:25:16 +0000 | [diff] [blame] | 420 | red_end_of_idle_period(&q->vars); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 421 | |
| 422 | sch_tree_unlock(sch); |
| 423 | choke_free(old); |
| 424 | return 0; |
| 425 | } |
| 426 | |
Alexander Aring | e63d7dfd | 2017-12-20 12:35:13 -0500 | [diff] [blame] | 427 | static int choke_init(struct Qdisc *sch, struct nlattr *opt, |
| 428 | struct netlink_ext_ack *extack) |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 429 | { |
Alexander Aring | 2030721 | 2017-12-20 12:35:14 -0500 | [diff] [blame] | 430 | return choke_change(sch, opt, extack); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 431 | } |
| 432 | |
| 433 | static int choke_dump(struct Qdisc *sch, struct sk_buff *skb) |
| 434 | { |
| 435 | struct choke_sched_data *q = qdisc_priv(sch); |
| 436 | struct nlattr *opts = NULL; |
| 437 | struct tc_red_qopt opt = { |
| 438 | .limit = q->limit, |
| 439 | .flags = q->flags, |
| 440 | .qth_min = q->parms.qth_min >> q->parms.Wlog, |
| 441 | .qth_max = q->parms.qth_max >> q->parms.Wlog, |
| 442 | .Wlog = q->parms.Wlog, |
| 443 | .Plog = q->parms.Plog, |
| 444 | .Scell_log = q->parms.Scell_log, |
| 445 | }; |
| 446 | |
Michal Kubecek | ae0be8d | 2019-04-26 11:13:06 +0200 | [diff] [blame] | 447 | opts = nla_nest_start_noflag(skb, TCA_OPTIONS); |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 448 | if (opts == NULL) |
| 449 | goto nla_put_failure; |
| 450 | |
David S. Miller | 1b34ec4 | 2012-03-29 05:11:39 -0400 | [diff] [blame] | 451 | if (nla_put(skb, TCA_CHOKE_PARMS, sizeof(opt), &opt) || |
| 452 | nla_put_u32(skb, TCA_CHOKE_MAX_P, q->parms.max_P)) |
| 453 | goto nla_put_failure; |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 454 | return nla_nest_end(skb, opts); |
| 455 | |
| 456 | nla_put_failure: |
| 457 | nla_nest_cancel(skb, opts); |
| 458 | return -EMSGSIZE; |
| 459 | } |
| 460 | |
| 461 | static int choke_dump_stats(struct Qdisc *sch, struct gnet_dump *d) |
| 462 | { |
| 463 | struct choke_sched_data *q = qdisc_priv(sch); |
| 464 | struct tc_choke_xstats st = { |
| 465 | .early = q->stats.prob_drop + q->stats.forced_drop, |
| 466 | .marked = q->stats.prob_mark + q->stats.forced_mark, |
| 467 | .pdrop = q->stats.pdrop, |
| 468 | .other = q->stats.other, |
| 469 | .matched = q->stats.matched, |
| 470 | }; |
| 471 | |
| 472 | return gnet_stats_copy_app(d, &st, sizeof(st)); |
| 473 | } |
| 474 | |
| 475 | static void choke_destroy(struct Qdisc *sch) |
| 476 | { |
| 477 | struct choke_sched_data *q = qdisc_priv(sch); |
| 478 | |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 479 | choke_free(q->tab); |
| 480 | } |
| 481 | |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 482 | static struct sk_buff *choke_peek_head(struct Qdisc *sch) |
| 483 | { |
| 484 | struct choke_sched_data *q = qdisc_priv(sch); |
| 485 | |
| 486 | return (q->head != q->tail) ? q->tab[q->head] : NULL; |
| 487 | } |
| 488 | |
| 489 | static struct Qdisc_ops choke_qdisc_ops __read_mostly = { |
| 490 | .id = "choke", |
| 491 | .priv_size = sizeof(struct choke_sched_data), |
| 492 | |
| 493 | .enqueue = choke_enqueue, |
| 494 | .dequeue = choke_dequeue, |
| 495 | .peek = choke_peek_head, |
stephen hemminger | 45e1443 | 2011-02-02 15:21:10 +0000 | [diff] [blame] | 496 | .init = choke_init, |
| 497 | .destroy = choke_destroy, |
| 498 | .reset = choke_reset, |
| 499 | .change = choke_change, |
| 500 | .dump = choke_dump, |
| 501 | .dump_stats = choke_dump_stats, |
| 502 | .owner = THIS_MODULE, |
| 503 | }; |
| 504 | |
| 505 | static int __init choke_module_init(void) |
| 506 | { |
| 507 | return register_qdisc(&choke_qdisc_ops); |
| 508 | } |
| 509 | |
| 510 | static void __exit choke_module_exit(void) |
| 511 | { |
| 512 | unregister_qdisc(&choke_qdisc_ops); |
| 513 | } |
| 514 | |
| 515 | module_init(choke_module_init) |
| 516 | module_exit(choke_module_exit) |
| 517 | |
| 518 | MODULE_LICENSE("GPL"); |