blob: fb0294d0b55e6e71ed87c9b7e1206e755ad5561b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * net/sched/sch_netem.c Network emulator
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
Stephen Hemminger798b6b12006-10-22 20:16:57 -07007 * 2 of the License.
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 *
9 * Many of the algorithms and ideas for this came from
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +090010 * NIST Net which is not copyrighted.
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 *
12 * Authors: Stephen Hemminger <shemminger@osdl.org>
13 * Catalin(ux aka Dino) BOIE <catab at umbrella dot ro>
14 */
15
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/types.h>
18#include <linux/kernel.h>
19#include <linux/errno.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/skbuff.h>
21#include <linux/rtnetlink.h>
22
Arnaldo Carvalho de Melodc5fc572007-03-25 23:06:12 -070023#include <net/netlink.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <net/pkt_sched.h>
25
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -080026#define VERSION "1.2"
Stephen Hemmingereb229c42005-11-03 13:49:01 -080027
Linus Torvalds1da177e2005-04-16 15:20:36 -070028/* Network Emulation Queuing algorithm.
29 ====================================
30
31 Sources: [1] Mark Carson, Darrin Santay, "NIST Net - A Linux-based
32 Network Emulation Tool
33 [2] Luigi Rizzo, DummyNet for FreeBSD
34
35 ----------------------------------------------------------------
36
37 This started out as a simple way to delay outgoing packets to
38 test TCP but has grown to include most of the functionality
39 of a full blown network emulator like NISTnet. It can delay
40 packets and add random jitter (and correlation). The random
41 distribution can be loaded from a table as well to provide
42 normal, Pareto, or experimental curves. Packet loss,
43 duplication, and reordering can also be emulated.
44
45 This qdisc does not do classification that can be handled in
46 layering other disciplines. It does not need to do bandwidth
47 control either since that can be handled by using token
48 bucket or other rate control.
49
50 The simulator is limited by the Linux timer resolution
51 and will create packet bursts on the HZ boundary (1ms).
52*/
53
54struct netem_sched_data {
55 struct Qdisc *qdisc;
Patrick McHardy59cb5c62007-03-16 01:20:31 -070056 struct qdisc_watchdog watchdog;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Stephen Hemmingerb4076212007-03-22 12:16:21 -070058 psched_tdiff_t latency;
59 psched_tdiff_t jitter;
60
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 u32 loss;
62 u32 limit;
63 u32 counter;
64 u32 gap;
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 u32 duplicate;
Stephen Hemminger0dca51d2005-05-26 12:55:48 -070066 u32 reorder;
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -080067 u32 corrupt;
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
69 struct crndstate {
Stephen Hemmingerb4076212007-03-22 12:16:21 -070070 u32 last;
71 u32 rho;
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -080072 } delay_cor, loss_cor, dup_cor, reorder_cor, corrupt_cor;
Linus Torvalds1da177e2005-04-16 15:20:36 -070073
74 struct disttable {
75 u32 size;
76 s16 table[0];
77 } *delay_dist;
78};
79
80/* Time stamp put into socket buffer control block */
81struct netem_skb_cb {
82 psched_time_t time_to_send;
83};
84
Jussi Kivilinna5f861732008-07-20 00:08:04 -070085static inline struct netem_skb_cb *netem_skb_cb(struct sk_buff *skb)
86{
Jussi Kivilinna175f9c12008-07-20 00:08:47 -070087 BUILD_BUG_ON(sizeof(skb->cb) <
88 sizeof(struct qdisc_skb_cb) + sizeof(struct netem_skb_cb));
89 return (struct netem_skb_cb *)qdisc_skb_cb(skb)->data;
Jussi Kivilinna5f861732008-07-20 00:08:04 -070090}
91
Linus Torvalds1da177e2005-04-16 15:20:36 -070092/* init_crandom - initialize correlated random number generator
93 * Use entropy source for initial seed.
94 */
95static void init_crandom(struct crndstate *state, unsigned long rho)
96{
97 state->rho = rho;
98 state->last = net_random();
99}
100
101/* get_crandom - correlated random number generator
102 * Next number depends on last value.
103 * rho is scaled to avoid floating point.
104 */
Stephen Hemmingerb4076212007-03-22 12:16:21 -0700105static u32 get_crandom(struct crndstate *state)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106{
107 u64 value, rho;
108 unsigned long answer;
109
Stephen Hemmingerbb2f8cc2007-03-23 00:12:09 -0700110 if (state->rho == 0) /* no correlation */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 return net_random();
112
113 value = net_random();
114 rho = (u64)state->rho + 1;
115 answer = (value * ((1ull<<32) - rho) + state->last * rho) >> 32;
116 state->last = answer;
117 return answer;
118}
119
120/* tabledist - return a pseudo-randomly distributed value with mean mu and
121 * std deviation sigma. Uses table lookup to approximate the desired
122 * distribution, and a uniformly-distributed pseudo-random source.
123 */
Stephen Hemmingerb4076212007-03-22 12:16:21 -0700124static psched_tdiff_t tabledist(psched_tdiff_t mu, psched_tdiff_t sigma,
125 struct crndstate *state,
126 const struct disttable *dist)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127{
Stephen Hemmingerb4076212007-03-22 12:16:21 -0700128 psched_tdiff_t x;
129 long t;
130 u32 rnd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
132 if (sigma == 0)
133 return mu;
134
135 rnd = get_crandom(state);
136
137 /* default uniform distribution */
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900138 if (dist == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 return (rnd % (2*sigma)) - sigma + mu;
140
141 t = dist->table[rnd % dist->size];
142 x = (sigma % NETEM_DIST_SCALE) * t;
143 if (x >= 0)
144 x += NETEM_DIST_SCALE/2;
145 else
146 x -= NETEM_DIST_SCALE/2;
147
148 return x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * t + mu;
149}
150
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700151/*
152 * Insert one skb into qdisc.
153 * Note: parent depends on return value to account for queue length.
154 * NET_XMIT_DROP: queue length didn't change.
155 * NET_XMIT_SUCCESS: one skb was queued.
156 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch)
158{
159 struct netem_sched_data *q = qdisc_priv(sch);
Guillaume Chazarain89e1df72006-07-21 14:45:25 -0700160 /* We don't fill cb now as skb_unshare() may invalidate it */
161 struct netem_skb_cb *cb;
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700162 struct sk_buff *skb2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 int ret;
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700164 int count = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165
Stephen Hemminger771018e2005-05-03 16:24:32 -0700166 pr_debug("netem_enqueue skb=%p\n", skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700168 /* Random duplication */
169 if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor))
170 ++count;
171
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 /* Random packet drop 0 => none, ~0 => all */
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700173 if (q->loss && q->loss >= get_crandom(&q->loss_cor))
174 --count;
175
176 if (count == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 sch->qstats.drops++;
178 kfree_skb(skb);
Jarek Poplawskic27f3392008-08-04 22:39:11 -0700179 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 }
181
David S. Miller4e8a5202006-10-22 21:00:33 -0700182 skb_orphan(skb);
183
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700184 /*
185 * If we need to duplicate packet, then re-insert at top of the
186 * qdisc tree, since parent queuer expects that only one
187 * skb will be queued.
188 */
189 if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) {
David S. Miller7698b4f2008-07-16 01:42:40 -0700190 struct Qdisc *rootq = qdisc_root(sch);
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700191 u32 dupsave = q->duplicate; /* prevent duplicating a dup... */
192 q->duplicate = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
Jussi Kivilinna5f861732008-07-20 00:08:04 -0700194 qdisc_enqueue_root(skb2, rootq);
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700195 q->duplicate = dupsave;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 }
197
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800198 /*
199 * Randomized packet corruption.
200 * Make copy if needed since we are modifying
201 * If packet is going to be hardware checksummed, then
202 * do it now in software before we mangle it.
203 */
204 if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) {
205 if (!(skb = skb_unshare(skb, GFP_ATOMIC))
Patrick McHardy84fa7932006-08-29 16:44:56 -0700206 || (skb->ip_summed == CHECKSUM_PARTIAL
207 && skb_checksum_help(skb))) {
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800208 sch->qstats.drops++;
209 return NET_XMIT_DROP;
210 }
211
212 skb->data[net_random() % skb_headlen(skb)] ^= 1<<(net_random() % 8);
213 }
214
Jussi Kivilinna5f861732008-07-20 00:08:04 -0700215 cb = netem_skb_cb(skb);
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700216 if (q->gap == 0 /* not doing reordering */
217 || q->counter < q->gap /* inside last reordering gap */
218 || q->reorder < get_crandom(&q->reorder_cor)) {
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700219 psched_time_t now;
Stephen Hemminger07aaa112005-11-03 13:43:07 -0800220 psched_tdiff_t delay;
221
222 delay = tabledist(q->latency, q->jitter,
223 &q->delay_cor, q->delay_dist);
224
Patrick McHardy3bebcda2007-03-23 11:29:25 -0700225 now = psched_get_time();
Patrick McHardy7c59e252007-03-23 11:27:45 -0700226 cb->time_to_send = now + delay;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 ++q->counter;
Jussi Kivilinna5f861732008-07-20 00:08:04 -0700228 ret = qdisc_enqueue(skb, q->qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 } else {
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900230 /*
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700231 * Do re-ordering by putting one out of N packets at the front
232 * of the queue.
233 */
Patrick McHardy3bebcda2007-03-23 11:29:25 -0700234 cb->time_to_send = psched_get_time();
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700235 q->counter = 0;
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700236 ret = q->qdisc->ops->requeue(skb, q->qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 }
238
239 if (likely(ret == NET_XMIT_SUCCESS)) {
240 sch->q.qlen++;
Jussi Kivilinna0abf77e2008-07-20 00:08:27 -0700241 sch->bstats.bytes += qdisc_pkt_len(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 sch->bstats.packets++;
Jarek Poplawski378a2f02008-08-04 22:31:03 -0700243 } else if (net_xmit_drop_count(ret)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 sch->qstats.drops++;
Jarek Poplawski378a2f02008-08-04 22:31:03 -0700245 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
Stephen Hemmingerd5d75cd2005-05-03 16:24:57 -0700247 pr_debug("netem: enqueue ret %d\n", ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 return ret;
249}
250
251/* Requeue packets but don't change time stamp */
252static int netem_requeue(struct sk_buff *skb, struct Qdisc *sch)
253{
254 struct netem_sched_data *q = qdisc_priv(sch);
255 int ret;
256
257 if ((ret = q->qdisc->ops->requeue(skb, q->qdisc)) == 0) {
258 sch->q.qlen++;
259 sch->qstats.requeues++;
260 }
261
262 return ret;
263}
264
265static unsigned int netem_drop(struct Qdisc* sch)
266{
267 struct netem_sched_data *q = qdisc_priv(sch);
Patrick McHardy6d037a22006-03-20 19:00:49 -0800268 unsigned int len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
Patrick McHardy6d037a22006-03-20 19:00:49 -0800270 if (q->qdisc->ops->drop && (len = q->qdisc->ops->drop(q->qdisc)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 sch->q.qlen--;
272 sch->qstats.drops++;
273 }
274 return len;
275}
276
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277static struct sk_buff *netem_dequeue(struct Qdisc *sch)
278{
279 struct netem_sched_data *q = qdisc_priv(sch);
280 struct sk_buff *skb;
281
Stephen Hemminger11274e52007-03-22 12:17:42 -0700282 smp_mb();
283 if (sch->flags & TCQ_F_THROTTLED)
284 return NULL;
285
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 skb = q->qdisc->dequeue(q->qdisc);
Stephen Hemminger771018e2005-05-03 16:24:32 -0700287 if (skb) {
Jussi Kivilinna5f861732008-07-20 00:08:04 -0700288 const struct netem_skb_cb *cb = netem_skb_cb(skb);
Patrick McHardy3bebcda2007-03-23 11:29:25 -0700289 psched_time_t now = psched_get_time();
Stephen Hemminger771018e2005-05-03 16:24:32 -0700290
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700291 /* if more time remaining? */
Patrick McHardy104e0872007-03-23 11:28:07 -0700292 if (cb->time_to_send <= now) {
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700293 pr_debug("netem_dequeue: return skb=%p\n", skb);
294 sch->q.qlen--;
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700295 return skb;
296 }
Stephen Hemminger11274e52007-03-22 12:17:42 -0700297
298 if (unlikely(q->qdisc->ops->requeue(skb, q->qdisc) != NET_XMIT_SUCCESS)) {
299 qdisc_tree_decrease_qlen(q->qdisc, 1);
300 sch->qstats.drops++;
301 printk(KERN_ERR "netem: %s could not requeue\n",
302 q->qdisc->ops->id);
303 }
304
305 qdisc_watchdog_schedule(&q->watchdog, cb->time_to_send);
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700306 }
307
308 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309}
310
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311static void netem_reset(struct Qdisc *sch)
312{
313 struct netem_sched_data *q = qdisc_priv(sch);
314
315 qdisc_reset(q->qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 sch->q.qlen = 0;
Patrick McHardy59cb5c62007-03-16 01:20:31 -0700317 qdisc_watchdog_cancel(&q->watchdog);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318}
319
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320/*
321 * Distribution data is a variable size payload containing
322 * signed 16 bit values.
323 */
Patrick McHardy1e904742008-01-22 22:11:17 -0800324static int get_dist_table(struct Qdisc *sch, const struct nlattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325{
326 struct netem_sched_data *q = qdisc_priv(sch);
Patrick McHardy1e904742008-01-22 22:11:17 -0800327 unsigned long n = nla_len(attr)/sizeof(__s16);
328 const __s16 *data = nla_data(attr);
David S. Miller7698b4f2008-07-16 01:42:40 -0700329 spinlock_t *root_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 struct disttable *d;
331 int i;
332
333 if (n > 65536)
334 return -EINVAL;
335
336 d = kmalloc(sizeof(*d) + n*sizeof(d->table[0]), GFP_KERNEL);
337 if (!d)
338 return -ENOMEM;
339
340 d->size = n;
341 for (i = 0; i < n; i++)
342 d->table[i] = data[i];
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900343
David S. Miller7698b4f2008-07-16 01:42:40 -0700344 root_lock = qdisc_root_lock(sch);
345
346 spin_lock_bh(root_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 d = xchg(&q->delay_dist, d);
David S. Miller7698b4f2008-07-16 01:42:40 -0700348 spin_unlock_bh(root_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349
350 kfree(d);
351 return 0;
352}
353
Patrick McHardy1e904742008-01-22 22:11:17 -0800354static int get_correlation(struct Qdisc *sch, const struct nlattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355{
356 struct netem_sched_data *q = qdisc_priv(sch);
Patrick McHardy1e904742008-01-22 22:11:17 -0800357 const struct tc_netem_corr *c = nla_data(attr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 init_crandom(&q->delay_cor, c->delay_corr);
360 init_crandom(&q->loss_cor, c->loss_corr);
361 init_crandom(&q->dup_cor, c->dup_corr);
362 return 0;
363}
364
Patrick McHardy1e904742008-01-22 22:11:17 -0800365static int get_reorder(struct Qdisc *sch, const struct nlattr *attr)
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700366{
367 struct netem_sched_data *q = qdisc_priv(sch);
Patrick McHardy1e904742008-01-22 22:11:17 -0800368 const struct tc_netem_reorder *r = nla_data(attr);
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700369
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700370 q->reorder = r->probability;
371 init_crandom(&q->reorder_cor, r->correlation);
372 return 0;
373}
374
Patrick McHardy1e904742008-01-22 22:11:17 -0800375static int get_corrupt(struct Qdisc *sch, const struct nlattr *attr)
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800376{
377 struct netem_sched_data *q = qdisc_priv(sch);
Patrick McHardy1e904742008-01-22 22:11:17 -0800378 const struct tc_netem_corrupt *r = nla_data(attr);
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800379
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800380 q->corrupt = r->probability;
381 init_crandom(&q->corrupt_cor, r->correlation);
382 return 0;
383}
384
Patrick McHardy27a34212008-01-23 20:35:39 -0800385static const struct nla_policy netem_policy[TCA_NETEM_MAX + 1] = {
386 [TCA_NETEM_CORR] = { .len = sizeof(struct tc_netem_corr) },
387 [TCA_NETEM_REORDER] = { .len = sizeof(struct tc_netem_reorder) },
388 [TCA_NETEM_CORRUPT] = { .len = sizeof(struct tc_netem_corrupt) },
389};
390
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800391/* Parse netlink message to set options */
Patrick McHardy1e904742008-01-22 22:11:17 -0800392static int netem_change(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393{
394 struct netem_sched_data *q = qdisc_priv(sch);
Patrick McHardyb03f4672008-01-23 20:32:21 -0800395 struct nlattr *tb[TCA_NETEM_MAX + 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 struct tc_netem_qopt *qopt;
397 int ret;
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900398
Patrick McHardyb03f4672008-01-23 20:32:21 -0800399 if (opt == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 return -EINVAL;
401
Patrick McHardy27a34212008-01-23 20:35:39 -0800402 ret = nla_parse_nested_compat(tb, TCA_NETEM_MAX, opt, netem_policy,
403 qopt, sizeof(*qopt));
Patrick McHardyb03f4672008-01-23 20:32:21 -0800404 if (ret < 0)
405 return ret;
406
Patrick McHardyfb0305c2008-07-05 23:40:21 -0700407 ret = fifo_set_limit(q->qdisc, qopt->limit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 if (ret) {
409 pr_debug("netem: can't set fifo limit\n");
410 return ret;
411 }
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900412
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 q->latency = qopt->latency;
414 q->jitter = qopt->jitter;
415 q->limit = qopt->limit;
416 q->gap = qopt->gap;
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700417 q->counter = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 q->loss = qopt->loss;
419 q->duplicate = qopt->duplicate;
420
Stephen Hemmingerbb2f8cc2007-03-23 00:12:09 -0700421 /* for compatibility with earlier versions.
422 * if gap is set, need to assume 100% probability
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700423 */
Stephen Hemmingera362e0a2007-03-22 12:15:45 -0700424 if (q->gap)
425 q->reorder = ~0;
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700426
Patrick McHardyb03f4672008-01-23 20:32:21 -0800427 if (tb[TCA_NETEM_CORR]) {
428 ret = get_correlation(sch, tb[TCA_NETEM_CORR]);
429 if (ret)
430 return ret;
431 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
Patrick McHardyb03f4672008-01-23 20:32:21 -0800433 if (tb[TCA_NETEM_DELAY_DIST]) {
434 ret = get_dist_table(sch, tb[TCA_NETEM_DELAY_DIST]);
435 if (ret)
436 return ret;
437 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
Patrick McHardyb03f4672008-01-23 20:32:21 -0800439 if (tb[TCA_NETEM_REORDER]) {
440 ret = get_reorder(sch, tb[TCA_NETEM_REORDER]);
441 if (ret)
442 return ret;
443 }
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800444
Patrick McHardyb03f4672008-01-23 20:32:21 -0800445 if (tb[TCA_NETEM_CORRUPT]) {
446 ret = get_corrupt(sch, tb[TCA_NETEM_CORRUPT]);
447 if (ret)
448 return ret;
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800449 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
451 return 0;
452}
453
Stephen Hemminger300ce172005-10-30 13:47:34 -0800454/*
455 * Special case version of FIFO queue for use by netem.
456 * It queues in order based on timestamps in skb's
457 */
458struct fifo_sched_data {
459 u32 limit;
Stephen Hemminger075aa572007-03-22 12:17:05 -0700460 psched_time_t oldest;
Stephen Hemminger300ce172005-10-30 13:47:34 -0800461};
462
463static int tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch)
464{
465 struct fifo_sched_data *q = qdisc_priv(sch);
466 struct sk_buff_head *list = &sch->q;
Jussi Kivilinna5f861732008-07-20 00:08:04 -0700467 psched_time_t tnext = netem_skb_cb(nskb)->time_to_send;
Stephen Hemminger300ce172005-10-30 13:47:34 -0800468 struct sk_buff *skb;
469
470 if (likely(skb_queue_len(list) < q->limit)) {
Stephen Hemminger075aa572007-03-22 12:17:05 -0700471 /* Optimize for add at tail */
Patrick McHardy104e0872007-03-23 11:28:07 -0700472 if (likely(skb_queue_empty(list) || tnext >= q->oldest)) {
Stephen Hemminger075aa572007-03-22 12:17:05 -0700473 q->oldest = tnext;
474 return qdisc_enqueue_tail(nskb, sch);
475 }
476
Stephen Hemminger300ce172005-10-30 13:47:34 -0800477 skb_queue_reverse_walk(list, skb) {
Jussi Kivilinna5f861732008-07-20 00:08:04 -0700478 const struct netem_skb_cb *cb = netem_skb_cb(skb);
Stephen Hemminger300ce172005-10-30 13:47:34 -0800479
Patrick McHardy104e0872007-03-23 11:28:07 -0700480 if (tnext >= cb->time_to_send)
Stephen Hemminger300ce172005-10-30 13:47:34 -0800481 break;
482 }
483
484 __skb_queue_after(list, skb, nskb);
485
Jussi Kivilinna0abf77e2008-07-20 00:08:27 -0700486 sch->qstats.backlog += qdisc_pkt_len(nskb);
487 sch->bstats.bytes += qdisc_pkt_len(nskb);
Stephen Hemminger300ce172005-10-30 13:47:34 -0800488 sch->bstats.packets++;
489
490 return NET_XMIT_SUCCESS;
491 }
492
Stephen Hemminger075aa572007-03-22 12:17:05 -0700493 return qdisc_reshape_fail(nskb, sch);
Stephen Hemminger300ce172005-10-30 13:47:34 -0800494}
495
Patrick McHardy1e904742008-01-22 22:11:17 -0800496static int tfifo_init(struct Qdisc *sch, struct nlattr *opt)
Stephen Hemminger300ce172005-10-30 13:47:34 -0800497{
498 struct fifo_sched_data *q = qdisc_priv(sch);
499
500 if (opt) {
Patrick McHardy1e904742008-01-22 22:11:17 -0800501 struct tc_fifo_qopt *ctl = nla_data(opt);
502 if (nla_len(opt) < sizeof(*ctl))
Stephen Hemminger300ce172005-10-30 13:47:34 -0800503 return -EINVAL;
504
505 q->limit = ctl->limit;
506 } else
David S. Miller5ce2d482008-07-08 17:06:30 -0700507 q->limit = max_t(u32, qdisc_dev(sch)->tx_queue_len, 1);
Stephen Hemminger300ce172005-10-30 13:47:34 -0800508
Patrick McHardya0849802007-03-23 11:28:30 -0700509 q->oldest = PSCHED_PASTPERFECT;
Stephen Hemminger300ce172005-10-30 13:47:34 -0800510 return 0;
511}
512
513static int tfifo_dump(struct Qdisc *sch, struct sk_buff *skb)
514{
515 struct fifo_sched_data *q = qdisc_priv(sch);
516 struct tc_fifo_qopt opt = { .limit = q->limit };
517
Patrick McHardy1e904742008-01-22 22:11:17 -0800518 NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
Stephen Hemminger300ce172005-10-30 13:47:34 -0800519 return skb->len;
520
Patrick McHardy1e904742008-01-22 22:11:17 -0800521nla_put_failure:
Stephen Hemminger300ce172005-10-30 13:47:34 -0800522 return -1;
523}
524
Eric Dumazet20fea082007-11-14 01:44:41 -0800525static struct Qdisc_ops tfifo_qdisc_ops __read_mostly = {
Stephen Hemminger300ce172005-10-30 13:47:34 -0800526 .id = "tfifo",
527 .priv_size = sizeof(struct fifo_sched_data),
528 .enqueue = tfifo_enqueue,
529 .dequeue = qdisc_dequeue_head,
530 .requeue = qdisc_requeue,
531 .drop = qdisc_queue_drop,
532 .init = tfifo_init,
533 .reset = qdisc_reset_queue,
534 .change = tfifo_init,
535 .dump = tfifo_dump,
536};
537
Patrick McHardy1e904742008-01-22 22:11:17 -0800538static int netem_init(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539{
540 struct netem_sched_data *q = qdisc_priv(sch);
541 int ret;
542
543 if (!opt)
544 return -EINVAL;
545
Patrick McHardy59cb5c62007-03-16 01:20:31 -0700546 qdisc_watchdog_init(&q->watchdog, sch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
David S. Miller5ce2d482008-07-08 17:06:30 -0700548 q->qdisc = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
David S. Millerbb949fb2008-07-08 16:55:56 -0700549 &tfifo_qdisc_ops,
Patrick McHardy9f9afec2006-11-29 17:35:18 -0800550 TC_H_MAKE(sch->handle, 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 if (!q->qdisc) {
552 pr_debug("netem: qdisc create failed\n");
553 return -ENOMEM;
554 }
555
556 ret = netem_change(sch, opt);
557 if (ret) {
558 pr_debug("netem: change failed\n");
559 qdisc_destroy(q->qdisc);
560 }
561 return ret;
562}
563
564static void netem_destroy(struct Qdisc *sch)
565{
566 struct netem_sched_data *q = qdisc_priv(sch);
567
Patrick McHardy59cb5c62007-03-16 01:20:31 -0700568 qdisc_watchdog_cancel(&q->watchdog);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 qdisc_destroy(q->qdisc);
570 kfree(q->delay_dist);
571}
572
573static int netem_dump(struct Qdisc *sch, struct sk_buff *skb)
574{
575 const struct netem_sched_data *q = qdisc_priv(sch);
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700576 unsigned char *b = skb_tail_pointer(skb);
Patrick McHardy1e904742008-01-22 22:11:17 -0800577 struct nlattr *nla = (struct nlattr *) b;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 struct tc_netem_qopt qopt;
579 struct tc_netem_corr cor;
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700580 struct tc_netem_reorder reorder;
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800581 struct tc_netem_corrupt corrupt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582
583 qopt.latency = q->latency;
584 qopt.jitter = q->jitter;
585 qopt.limit = q->limit;
586 qopt.loss = q->loss;
587 qopt.gap = q->gap;
588 qopt.duplicate = q->duplicate;
Patrick McHardy1e904742008-01-22 22:11:17 -0800589 NLA_PUT(skb, TCA_OPTIONS, sizeof(qopt), &qopt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590
591 cor.delay_corr = q->delay_cor.rho;
592 cor.loss_corr = q->loss_cor.rho;
593 cor.dup_corr = q->dup_cor.rho;
Patrick McHardy1e904742008-01-22 22:11:17 -0800594 NLA_PUT(skb, TCA_NETEM_CORR, sizeof(cor), &cor);
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700595
596 reorder.probability = q->reorder;
597 reorder.correlation = q->reorder_cor.rho;
Patrick McHardy1e904742008-01-22 22:11:17 -0800598 NLA_PUT(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder);
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700599
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800600 corrupt.probability = q->corrupt;
601 corrupt.correlation = q->corrupt_cor.rho;
Patrick McHardy1e904742008-01-22 22:11:17 -0800602 NLA_PUT(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt);
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800603
Patrick McHardy1e904742008-01-22 22:11:17 -0800604 nla->nla_len = skb_tail_pointer(skb) - b;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605
606 return skb->len;
607
Patrick McHardy1e904742008-01-22 22:11:17 -0800608nla_put_failure:
Arnaldo Carvalho de Melodc5fc572007-03-25 23:06:12 -0700609 nlmsg_trim(skb, b);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 return -1;
611}
612
613static int netem_dump_class(struct Qdisc *sch, unsigned long cl,
614 struct sk_buff *skb, struct tcmsg *tcm)
615{
616 struct netem_sched_data *q = qdisc_priv(sch);
617
618 if (cl != 1) /* only one class */
619 return -ENOENT;
620
621 tcm->tcm_handle |= TC_H_MIN(1);
622 tcm->tcm_info = q->qdisc->handle;
623
624 return 0;
625}
626
627static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
628 struct Qdisc **old)
629{
630 struct netem_sched_data *q = qdisc_priv(sch);
631
632 if (new == NULL)
633 new = &noop_qdisc;
634
635 sch_tree_lock(sch);
636 *old = xchg(&q->qdisc, new);
Patrick McHardy5e50da02006-11-29 17:36:20 -0800637 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 qdisc_reset(*old);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 sch_tree_unlock(sch);
640
641 return 0;
642}
643
644static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg)
645{
646 struct netem_sched_data *q = qdisc_priv(sch);
647 return q->qdisc;
648}
649
650static unsigned long netem_get(struct Qdisc *sch, u32 classid)
651{
652 return 1;
653}
654
655static void netem_put(struct Qdisc *sch, unsigned long arg)
656{
657}
658
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900659static int netem_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
Patrick McHardy1e904742008-01-22 22:11:17 -0800660 struct nlattr **tca, unsigned long *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661{
662 return -ENOSYS;
663}
664
665static int netem_delete(struct Qdisc *sch, unsigned long arg)
666{
667 return -ENOSYS;
668}
669
670static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker)
671{
672 if (!walker->stop) {
673 if (walker->count >= walker->skip)
674 if (walker->fn(sch, 1, walker) < 0) {
675 walker->stop = 1;
676 return;
677 }
678 walker->count++;
679 }
680}
681
682static struct tcf_proto **netem_find_tcf(struct Qdisc *sch, unsigned long cl)
683{
684 return NULL;
685}
686
Eric Dumazet20fea082007-11-14 01:44:41 -0800687static const struct Qdisc_class_ops netem_class_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 .graft = netem_graft,
689 .leaf = netem_leaf,
690 .get = netem_get,
691 .put = netem_put,
692 .change = netem_change_class,
693 .delete = netem_delete,
694 .walk = netem_walk,
695 .tcf_chain = netem_find_tcf,
696 .dump = netem_dump_class,
697};
698
Eric Dumazet20fea082007-11-14 01:44:41 -0800699static struct Qdisc_ops netem_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 .id = "netem",
701 .cl_ops = &netem_class_ops,
702 .priv_size = sizeof(struct netem_sched_data),
703 .enqueue = netem_enqueue,
704 .dequeue = netem_dequeue,
705 .requeue = netem_requeue,
706 .drop = netem_drop,
707 .init = netem_init,
708 .reset = netem_reset,
709 .destroy = netem_destroy,
710 .change = netem_change,
711 .dump = netem_dump,
712 .owner = THIS_MODULE,
713};
714
715
716static int __init netem_module_init(void)
717{
Stephen Hemmingereb229c42005-11-03 13:49:01 -0800718 pr_info("netem: version " VERSION "\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 return register_qdisc(&netem_qdisc_ops);
720}
721static void __exit netem_module_exit(void)
722{
723 unregister_qdisc(&netem_qdisc_ops);
724}
725module_init(netem_module_init)
726module_exit(netem_module_exit)
727MODULE_LICENSE("GPL");