blob: ae49be00022f153d830fe6327ab6adaeb79fed65 [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{
87 BUILD_BUG_ON(sizeof(skb->cb) < sizeof(struct netem_skb_cb));
88 return (struct netem_skb_cb *)skb->cb;
89}
90
Linus Torvalds1da177e2005-04-16 15:20:36 -070091/* init_crandom - initialize correlated random number generator
92 * Use entropy source for initial seed.
93 */
94static void init_crandom(struct crndstate *state, unsigned long rho)
95{
96 state->rho = rho;
97 state->last = net_random();
98}
99
100/* get_crandom - correlated random number generator
101 * Next number depends on last value.
102 * rho is scaled to avoid floating point.
103 */
Stephen Hemmingerb4076212007-03-22 12:16:21 -0700104static u32 get_crandom(struct crndstate *state)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105{
106 u64 value, rho;
107 unsigned long answer;
108
Stephen Hemmingerbb2f8cc2007-03-23 00:12:09 -0700109 if (state->rho == 0) /* no correlation */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 return net_random();
111
112 value = net_random();
113 rho = (u64)state->rho + 1;
114 answer = (value * ((1ull<<32) - rho) + state->last * rho) >> 32;
115 state->last = answer;
116 return answer;
117}
118
119/* tabledist - return a pseudo-randomly distributed value with mean mu and
120 * std deviation sigma. Uses table lookup to approximate the desired
121 * distribution, and a uniformly-distributed pseudo-random source.
122 */
Stephen Hemmingerb4076212007-03-22 12:16:21 -0700123static psched_tdiff_t tabledist(psched_tdiff_t mu, psched_tdiff_t sigma,
124 struct crndstate *state,
125 const struct disttable *dist)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126{
Stephen Hemmingerb4076212007-03-22 12:16:21 -0700127 psched_tdiff_t x;
128 long t;
129 u32 rnd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130
131 if (sigma == 0)
132 return mu;
133
134 rnd = get_crandom(state);
135
136 /* default uniform distribution */
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900137 if (dist == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 return (rnd % (2*sigma)) - sigma + mu;
139
140 t = dist->table[rnd % dist->size];
141 x = (sigma % NETEM_DIST_SCALE) * t;
142 if (x >= 0)
143 x += NETEM_DIST_SCALE/2;
144 else
145 x -= NETEM_DIST_SCALE/2;
146
147 return x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * t + mu;
148}
149
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700150/*
151 * Insert one skb into qdisc.
152 * Note: parent depends on return value to account for queue length.
153 * NET_XMIT_DROP: queue length didn't change.
154 * NET_XMIT_SUCCESS: one skb was queued.
155 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156static int netem_enqueue(struct sk_buff *skb, struct Qdisc *sch)
157{
158 struct netem_sched_data *q = qdisc_priv(sch);
Guillaume Chazarain89e1df72006-07-21 14:45:25 -0700159 /* We don't fill cb now as skb_unshare() may invalidate it */
160 struct netem_skb_cb *cb;
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700161 struct sk_buff *skb2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 int ret;
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700163 int count = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
Stephen Hemminger771018e2005-05-03 16:24:32 -0700165 pr_debug("netem_enqueue skb=%p\n", skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700167 /* Random duplication */
168 if (q->duplicate && q->duplicate >= get_crandom(&q->dup_cor))
169 ++count;
170
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 /* Random packet drop 0 => none, ~0 => all */
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700172 if (q->loss && q->loss >= get_crandom(&q->loss_cor))
173 --count;
174
175 if (count == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 sch->qstats.drops++;
177 kfree_skb(skb);
Stephen Hemminger89bbb0a32006-04-28 12:11:36 -0700178 return NET_XMIT_BYPASS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 }
180
David S. Miller4e8a5202006-10-22 21:00:33 -0700181 skb_orphan(skb);
182
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700183 /*
184 * If we need to duplicate packet, then re-insert at top of the
185 * qdisc tree, since parent queuer expects that only one
186 * skb will be queued.
187 */
188 if (count > 1 && (skb2 = skb_clone(skb, GFP_ATOMIC)) != NULL) {
David S. Miller7698b4f2008-07-16 01:42:40 -0700189 struct Qdisc *rootq = qdisc_root(sch);
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700190 u32 dupsave = q->duplicate; /* prevent duplicating a dup... */
191 q->duplicate = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
Jussi Kivilinna5f861732008-07-20 00:08:04 -0700193 qdisc_enqueue_root(skb2, rootq);
Stephen Hemminger0afb51e2005-05-26 12:53:49 -0700194 q->duplicate = dupsave;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 }
196
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800197 /*
198 * Randomized packet corruption.
199 * Make copy if needed since we are modifying
200 * If packet is going to be hardware checksummed, then
201 * do it now in software before we mangle it.
202 */
203 if (q->corrupt && q->corrupt >= get_crandom(&q->corrupt_cor)) {
204 if (!(skb = skb_unshare(skb, GFP_ATOMIC))
Patrick McHardy84fa7932006-08-29 16:44:56 -0700205 || (skb->ip_summed == CHECKSUM_PARTIAL
206 && skb_checksum_help(skb))) {
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800207 sch->qstats.drops++;
208 return NET_XMIT_DROP;
209 }
210
211 skb->data[net_random() % skb_headlen(skb)] ^= 1<<(net_random() % 8);
212 }
213
Jussi Kivilinna5f861732008-07-20 00:08:04 -0700214 cb = netem_skb_cb(skb);
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700215 if (q->gap == 0 /* not doing reordering */
216 || q->counter < q->gap /* inside last reordering gap */
217 || q->reorder < get_crandom(&q->reorder_cor)) {
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700218 psched_time_t now;
Stephen Hemminger07aaa112005-11-03 13:43:07 -0800219 psched_tdiff_t delay;
220
221 delay = tabledist(q->latency, q->jitter,
222 &q->delay_cor, q->delay_dist);
223
Patrick McHardy3bebcda2007-03-23 11:29:25 -0700224 now = psched_get_time();
Patrick McHardy7c59e252007-03-23 11:27:45 -0700225 cb->time_to_send = now + delay;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 ++q->counter;
Jussi Kivilinna5f861732008-07-20 00:08:04 -0700227 ret = qdisc_enqueue(skb, q->qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 } else {
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900229 /*
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700230 * Do re-ordering by putting one out of N packets at the front
231 * of the queue.
232 */
Patrick McHardy3bebcda2007-03-23 11:29:25 -0700233 cb->time_to_send = psched_get_time();
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700234 q->counter = 0;
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700235 ret = q->qdisc->ops->requeue(skb, q->qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 }
237
238 if (likely(ret == NET_XMIT_SUCCESS)) {
239 sch->q.qlen++;
Jussi Kivilinna0abf77e2008-07-20 00:08:27 -0700240 sch->bstats.bytes += qdisc_pkt_len(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 sch->bstats.packets++;
242 } else
243 sch->qstats.drops++;
244
Stephen Hemmingerd5d75cd2005-05-03 16:24:57 -0700245 pr_debug("netem: enqueue ret %d\n", ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 return ret;
247}
248
249/* Requeue packets but don't change time stamp */
250static int netem_requeue(struct sk_buff *skb, struct Qdisc *sch)
251{
252 struct netem_sched_data *q = qdisc_priv(sch);
253 int ret;
254
255 if ((ret = q->qdisc->ops->requeue(skb, q->qdisc)) == 0) {
256 sch->q.qlen++;
257 sch->qstats.requeues++;
258 }
259
260 return ret;
261}
262
263static unsigned int netem_drop(struct Qdisc* sch)
264{
265 struct netem_sched_data *q = qdisc_priv(sch);
Patrick McHardy6d037a22006-03-20 19:00:49 -0800266 unsigned int len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
Patrick McHardy6d037a22006-03-20 19:00:49 -0800268 if (q->qdisc->ops->drop && (len = q->qdisc->ops->drop(q->qdisc)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 sch->q.qlen--;
270 sch->qstats.drops++;
271 }
272 return len;
273}
274
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275static struct sk_buff *netem_dequeue(struct Qdisc *sch)
276{
277 struct netem_sched_data *q = qdisc_priv(sch);
278 struct sk_buff *skb;
279
Stephen Hemminger11274e52007-03-22 12:17:42 -0700280 smp_mb();
281 if (sch->flags & TCQ_F_THROTTLED)
282 return NULL;
283
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 skb = q->qdisc->dequeue(q->qdisc);
Stephen Hemminger771018e2005-05-03 16:24:32 -0700285 if (skb) {
Jussi Kivilinna5f861732008-07-20 00:08:04 -0700286 const struct netem_skb_cb *cb = netem_skb_cb(skb);
Patrick McHardy3bebcda2007-03-23 11:29:25 -0700287 psched_time_t now = psched_get_time();
Stephen Hemminger771018e2005-05-03 16:24:32 -0700288
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700289 /* if more time remaining? */
Patrick McHardy104e0872007-03-23 11:28:07 -0700290 if (cb->time_to_send <= now) {
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700291 pr_debug("netem_dequeue: return skb=%p\n", skb);
292 sch->q.qlen--;
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700293 return skb;
294 }
Stephen Hemminger11274e52007-03-22 12:17:42 -0700295
296 if (unlikely(q->qdisc->ops->requeue(skb, q->qdisc) != NET_XMIT_SUCCESS)) {
297 qdisc_tree_decrease_qlen(q->qdisc, 1);
298 sch->qstats.drops++;
299 printk(KERN_ERR "netem: %s could not requeue\n",
300 q->qdisc->ops->id);
301 }
302
303 qdisc_watchdog_schedule(&q->watchdog, cb->time_to_send);
Stephen Hemminger0f9f32a2005-05-26 12:55:01 -0700304 }
305
306 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307}
308
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309static void netem_reset(struct Qdisc *sch)
310{
311 struct netem_sched_data *q = qdisc_priv(sch);
312
313 qdisc_reset(q->qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 sch->q.qlen = 0;
Patrick McHardy59cb5c62007-03-16 01:20:31 -0700315 qdisc_watchdog_cancel(&q->watchdog);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316}
317
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318/*
319 * Distribution data is a variable size payload containing
320 * signed 16 bit values.
321 */
Patrick McHardy1e904742008-01-22 22:11:17 -0800322static int get_dist_table(struct Qdisc *sch, const struct nlattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323{
324 struct netem_sched_data *q = qdisc_priv(sch);
Patrick McHardy1e904742008-01-22 22:11:17 -0800325 unsigned long n = nla_len(attr)/sizeof(__s16);
326 const __s16 *data = nla_data(attr);
David S. Miller7698b4f2008-07-16 01:42:40 -0700327 spinlock_t *root_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 struct disttable *d;
329 int i;
330
331 if (n > 65536)
332 return -EINVAL;
333
334 d = kmalloc(sizeof(*d) + n*sizeof(d->table[0]), GFP_KERNEL);
335 if (!d)
336 return -ENOMEM;
337
338 d->size = n;
339 for (i = 0; i < n; i++)
340 d->table[i] = data[i];
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900341
David S. Miller7698b4f2008-07-16 01:42:40 -0700342 root_lock = qdisc_root_lock(sch);
343
344 spin_lock_bh(root_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 d = xchg(&q->delay_dist, d);
David S. Miller7698b4f2008-07-16 01:42:40 -0700346 spin_unlock_bh(root_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
348 kfree(d);
349 return 0;
350}
351
Patrick McHardy1e904742008-01-22 22:11:17 -0800352static int get_correlation(struct Qdisc *sch, const struct nlattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353{
354 struct netem_sched_data *q = qdisc_priv(sch);
Patrick McHardy1e904742008-01-22 22:11:17 -0800355 const struct tc_netem_corr *c = nla_data(attr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 init_crandom(&q->delay_cor, c->delay_corr);
358 init_crandom(&q->loss_cor, c->loss_corr);
359 init_crandom(&q->dup_cor, c->dup_corr);
360 return 0;
361}
362
Patrick McHardy1e904742008-01-22 22:11:17 -0800363static int get_reorder(struct Qdisc *sch, const struct nlattr *attr)
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700364{
365 struct netem_sched_data *q = qdisc_priv(sch);
Patrick McHardy1e904742008-01-22 22:11:17 -0800366 const struct tc_netem_reorder *r = nla_data(attr);
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700367
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700368 q->reorder = r->probability;
369 init_crandom(&q->reorder_cor, r->correlation);
370 return 0;
371}
372
Patrick McHardy1e904742008-01-22 22:11:17 -0800373static int get_corrupt(struct Qdisc *sch, const struct nlattr *attr)
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800374{
375 struct netem_sched_data *q = qdisc_priv(sch);
Patrick McHardy1e904742008-01-22 22:11:17 -0800376 const struct tc_netem_corrupt *r = nla_data(attr);
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800377
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800378 q->corrupt = r->probability;
379 init_crandom(&q->corrupt_cor, r->correlation);
380 return 0;
381}
382
Patrick McHardy27a34212008-01-23 20:35:39 -0800383static const struct nla_policy netem_policy[TCA_NETEM_MAX + 1] = {
384 [TCA_NETEM_CORR] = { .len = sizeof(struct tc_netem_corr) },
385 [TCA_NETEM_REORDER] = { .len = sizeof(struct tc_netem_reorder) },
386 [TCA_NETEM_CORRUPT] = { .len = sizeof(struct tc_netem_corrupt) },
387};
388
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800389/* Parse netlink message to set options */
Patrick McHardy1e904742008-01-22 22:11:17 -0800390static int netem_change(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391{
392 struct netem_sched_data *q = qdisc_priv(sch);
Patrick McHardyb03f4672008-01-23 20:32:21 -0800393 struct nlattr *tb[TCA_NETEM_MAX + 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 struct tc_netem_qopt *qopt;
395 int ret;
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900396
Patrick McHardyb03f4672008-01-23 20:32:21 -0800397 if (opt == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 return -EINVAL;
399
Patrick McHardy27a34212008-01-23 20:35:39 -0800400 ret = nla_parse_nested_compat(tb, TCA_NETEM_MAX, opt, netem_policy,
401 qopt, sizeof(*qopt));
Patrick McHardyb03f4672008-01-23 20:32:21 -0800402 if (ret < 0)
403 return ret;
404
Patrick McHardyfb0305c2008-07-05 23:40:21 -0700405 ret = fifo_set_limit(q->qdisc, qopt->limit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 if (ret) {
407 pr_debug("netem: can't set fifo limit\n");
408 return ret;
409 }
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900410
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 q->latency = qopt->latency;
412 q->jitter = qopt->jitter;
413 q->limit = qopt->limit;
414 q->gap = qopt->gap;
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700415 q->counter = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 q->loss = qopt->loss;
417 q->duplicate = qopt->duplicate;
418
Stephen Hemmingerbb2f8cc2007-03-23 00:12:09 -0700419 /* for compatibility with earlier versions.
420 * if gap is set, need to assume 100% probability
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700421 */
Stephen Hemmingera362e0a2007-03-22 12:15:45 -0700422 if (q->gap)
423 q->reorder = ~0;
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700424
Patrick McHardyb03f4672008-01-23 20:32:21 -0800425 if (tb[TCA_NETEM_CORR]) {
426 ret = get_correlation(sch, tb[TCA_NETEM_CORR]);
427 if (ret)
428 return ret;
429 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
Patrick McHardyb03f4672008-01-23 20:32:21 -0800431 if (tb[TCA_NETEM_DELAY_DIST]) {
432 ret = get_dist_table(sch, tb[TCA_NETEM_DELAY_DIST]);
433 if (ret)
434 return ret;
435 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436
Patrick McHardyb03f4672008-01-23 20:32:21 -0800437 if (tb[TCA_NETEM_REORDER]) {
438 ret = get_reorder(sch, tb[TCA_NETEM_REORDER]);
439 if (ret)
440 return ret;
441 }
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800442
Patrick McHardyb03f4672008-01-23 20:32:21 -0800443 if (tb[TCA_NETEM_CORRUPT]) {
444 ret = get_corrupt(sch, tb[TCA_NETEM_CORRUPT]);
445 if (ret)
446 return ret;
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800447 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
449 return 0;
450}
451
Stephen Hemminger300ce172005-10-30 13:47:34 -0800452/*
453 * Special case version of FIFO queue for use by netem.
454 * It queues in order based on timestamps in skb's
455 */
456struct fifo_sched_data {
457 u32 limit;
Stephen Hemminger075aa572007-03-22 12:17:05 -0700458 psched_time_t oldest;
Stephen Hemminger300ce172005-10-30 13:47:34 -0800459};
460
461static int tfifo_enqueue(struct sk_buff *nskb, struct Qdisc *sch)
462{
463 struct fifo_sched_data *q = qdisc_priv(sch);
464 struct sk_buff_head *list = &sch->q;
Jussi Kivilinna5f861732008-07-20 00:08:04 -0700465 psched_time_t tnext = netem_skb_cb(nskb)->time_to_send;
Stephen Hemminger300ce172005-10-30 13:47:34 -0800466 struct sk_buff *skb;
467
468 if (likely(skb_queue_len(list) < q->limit)) {
Stephen Hemminger075aa572007-03-22 12:17:05 -0700469 /* Optimize for add at tail */
Patrick McHardy104e0872007-03-23 11:28:07 -0700470 if (likely(skb_queue_empty(list) || tnext >= q->oldest)) {
Stephen Hemminger075aa572007-03-22 12:17:05 -0700471 q->oldest = tnext;
472 return qdisc_enqueue_tail(nskb, sch);
473 }
474
Stephen Hemminger300ce172005-10-30 13:47:34 -0800475 skb_queue_reverse_walk(list, skb) {
Jussi Kivilinna5f861732008-07-20 00:08:04 -0700476 const struct netem_skb_cb *cb = netem_skb_cb(skb);
Stephen Hemminger300ce172005-10-30 13:47:34 -0800477
Patrick McHardy104e0872007-03-23 11:28:07 -0700478 if (tnext >= cb->time_to_send)
Stephen Hemminger300ce172005-10-30 13:47:34 -0800479 break;
480 }
481
482 __skb_queue_after(list, skb, nskb);
483
Jussi Kivilinna0abf77e2008-07-20 00:08:27 -0700484 sch->qstats.backlog += qdisc_pkt_len(nskb);
485 sch->bstats.bytes += qdisc_pkt_len(nskb);
Stephen Hemminger300ce172005-10-30 13:47:34 -0800486 sch->bstats.packets++;
487
488 return NET_XMIT_SUCCESS;
489 }
490
Stephen Hemminger075aa572007-03-22 12:17:05 -0700491 return qdisc_reshape_fail(nskb, sch);
Stephen Hemminger300ce172005-10-30 13:47:34 -0800492}
493
Patrick McHardy1e904742008-01-22 22:11:17 -0800494static int tfifo_init(struct Qdisc *sch, struct nlattr *opt)
Stephen Hemminger300ce172005-10-30 13:47:34 -0800495{
496 struct fifo_sched_data *q = qdisc_priv(sch);
497
498 if (opt) {
Patrick McHardy1e904742008-01-22 22:11:17 -0800499 struct tc_fifo_qopt *ctl = nla_data(opt);
500 if (nla_len(opt) < sizeof(*ctl))
Stephen Hemminger300ce172005-10-30 13:47:34 -0800501 return -EINVAL;
502
503 q->limit = ctl->limit;
504 } else
David S. Miller5ce2d482008-07-08 17:06:30 -0700505 q->limit = max_t(u32, qdisc_dev(sch)->tx_queue_len, 1);
Stephen Hemminger300ce172005-10-30 13:47:34 -0800506
Patrick McHardya0849802007-03-23 11:28:30 -0700507 q->oldest = PSCHED_PASTPERFECT;
Stephen Hemminger300ce172005-10-30 13:47:34 -0800508 return 0;
509}
510
511static int tfifo_dump(struct Qdisc *sch, struct sk_buff *skb)
512{
513 struct fifo_sched_data *q = qdisc_priv(sch);
514 struct tc_fifo_qopt opt = { .limit = q->limit };
515
Patrick McHardy1e904742008-01-22 22:11:17 -0800516 NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
Stephen Hemminger300ce172005-10-30 13:47:34 -0800517 return skb->len;
518
Patrick McHardy1e904742008-01-22 22:11:17 -0800519nla_put_failure:
Stephen Hemminger300ce172005-10-30 13:47:34 -0800520 return -1;
521}
522
Eric Dumazet20fea082007-11-14 01:44:41 -0800523static struct Qdisc_ops tfifo_qdisc_ops __read_mostly = {
Stephen Hemminger300ce172005-10-30 13:47:34 -0800524 .id = "tfifo",
525 .priv_size = sizeof(struct fifo_sched_data),
526 .enqueue = tfifo_enqueue,
527 .dequeue = qdisc_dequeue_head,
528 .requeue = qdisc_requeue,
529 .drop = qdisc_queue_drop,
530 .init = tfifo_init,
531 .reset = qdisc_reset_queue,
532 .change = tfifo_init,
533 .dump = tfifo_dump,
534};
535
Patrick McHardy1e904742008-01-22 22:11:17 -0800536static int netem_init(struct Qdisc *sch, struct nlattr *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537{
538 struct netem_sched_data *q = qdisc_priv(sch);
539 int ret;
540
541 if (!opt)
542 return -EINVAL;
543
Patrick McHardy59cb5c62007-03-16 01:20:31 -0700544 qdisc_watchdog_init(&q->watchdog, sch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545
David S. Miller5ce2d482008-07-08 17:06:30 -0700546 q->qdisc = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
David S. Millerbb949fb2008-07-08 16:55:56 -0700547 &tfifo_qdisc_ops,
Patrick McHardy9f9afec2006-11-29 17:35:18 -0800548 TC_H_MAKE(sch->handle, 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 if (!q->qdisc) {
550 pr_debug("netem: qdisc create failed\n");
551 return -ENOMEM;
552 }
553
554 ret = netem_change(sch, opt);
555 if (ret) {
556 pr_debug("netem: change failed\n");
557 qdisc_destroy(q->qdisc);
558 }
559 return ret;
560}
561
562static void netem_destroy(struct Qdisc *sch)
563{
564 struct netem_sched_data *q = qdisc_priv(sch);
565
Patrick McHardy59cb5c62007-03-16 01:20:31 -0700566 qdisc_watchdog_cancel(&q->watchdog);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 qdisc_destroy(q->qdisc);
568 kfree(q->delay_dist);
569}
570
571static int netem_dump(struct Qdisc *sch, struct sk_buff *skb)
572{
573 const struct netem_sched_data *q = qdisc_priv(sch);
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700574 unsigned char *b = skb_tail_pointer(skb);
Patrick McHardy1e904742008-01-22 22:11:17 -0800575 struct nlattr *nla = (struct nlattr *) b;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 struct tc_netem_qopt qopt;
577 struct tc_netem_corr cor;
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700578 struct tc_netem_reorder reorder;
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800579 struct tc_netem_corrupt corrupt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
581 qopt.latency = q->latency;
582 qopt.jitter = q->jitter;
583 qopt.limit = q->limit;
584 qopt.loss = q->loss;
585 qopt.gap = q->gap;
586 qopt.duplicate = q->duplicate;
Patrick McHardy1e904742008-01-22 22:11:17 -0800587 NLA_PUT(skb, TCA_OPTIONS, sizeof(qopt), &qopt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
589 cor.delay_corr = q->delay_cor.rho;
590 cor.loss_corr = q->loss_cor.rho;
591 cor.dup_corr = q->dup_cor.rho;
Patrick McHardy1e904742008-01-22 22:11:17 -0800592 NLA_PUT(skb, TCA_NETEM_CORR, sizeof(cor), &cor);
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700593
594 reorder.probability = q->reorder;
595 reorder.correlation = q->reorder_cor.rho;
Patrick McHardy1e904742008-01-22 22:11:17 -0800596 NLA_PUT(skb, TCA_NETEM_REORDER, sizeof(reorder), &reorder);
Stephen Hemminger0dca51d2005-05-26 12:55:48 -0700597
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800598 corrupt.probability = q->corrupt;
599 corrupt.correlation = q->corrupt_cor.rho;
Patrick McHardy1e904742008-01-22 22:11:17 -0800600 NLA_PUT(skb, TCA_NETEM_CORRUPT, sizeof(corrupt), &corrupt);
Stephen Hemmingerc865e5d2005-12-21 19:03:44 -0800601
Patrick McHardy1e904742008-01-22 22:11:17 -0800602 nla->nla_len = skb_tail_pointer(skb) - b;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603
604 return skb->len;
605
Patrick McHardy1e904742008-01-22 22:11:17 -0800606nla_put_failure:
Arnaldo Carvalho de Melodc5fc572007-03-25 23:06:12 -0700607 nlmsg_trim(skb, b);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 return -1;
609}
610
611static int netem_dump_class(struct Qdisc *sch, unsigned long cl,
612 struct sk_buff *skb, struct tcmsg *tcm)
613{
614 struct netem_sched_data *q = qdisc_priv(sch);
615
616 if (cl != 1) /* only one class */
617 return -ENOENT;
618
619 tcm->tcm_handle |= TC_H_MIN(1);
620 tcm->tcm_info = q->qdisc->handle;
621
622 return 0;
623}
624
625static int netem_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
626 struct Qdisc **old)
627{
628 struct netem_sched_data *q = qdisc_priv(sch);
629
630 if (new == NULL)
631 new = &noop_qdisc;
632
633 sch_tree_lock(sch);
634 *old = xchg(&q->qdisc, new);
Patrick McHardy5e50da02006-11-29 17:36:20 -0800635 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 qdisc_reset(*old);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 sch_tree_unlock(sch);
638
639 return 0;
640}
641
642static struct Qdisc *netem_leaf(struct Qdisc *sch, unsigned long arg)
643{
644 struct netem_sched_data *q = qdisc_priv(sch);
645 return q->qdisc;
646}
647
648static unsigned long netem_get(struct Qdisc *sch, u32 classid)
649{
650 return 1;
651}
652
653static void netem_put(struct Qdisc *sch, unsigned long arg)
654{
655}
656
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900657static int netem_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
Patrick McHardy1e904742008-01-22 22:11:17 -0800658 struct nlattr **tca, unsigned long *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659{
660 return -ENOSYS;
661}
662
663static int netem_delete(struct Qdisc *sch, unsigned long arg)
664{
665 return -ENOSYS;
666}
667
668static void netem_walk(struct Qdisc *sch, struct qdisc_walker *walker)
669{
670 if (!walker->stop) {
671 if (walker->count >= walker->skip)
672 if (walker->fn(sch, 1, walker) < 0) {
673 walker->stop = 1;
674 return;
675 }
676 walker->count++;
677 }
678}
679
680static struct tcf_proto **netem_find_tcf(struct Qdisc *sch, unsigned long cl)
681{
682 return NULL;
683}
684
Eric Dumazet20fea082007-11-14 01:44:41 -0800685static const struct Qdisc_class_ops netem_class_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 .graft = netem_graft,
687 .leaf = netem_leaf,
688 .get = netem_get,
689 .put = netem_put,
690 .change = netem_change_class,
691 .delete = netem_delete,
692 .walk = netem_walk,
693 .tcf_chain = netem_find_tcf,
694 .dump = netem_dump_class,
695};
696
Eric Dumazet20fea082007-11-14 01:44:41 -0800697static struct Qdisc_ops netem_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 .id = "netem",
699 .cl_ops = &netem_class_ops,
700 .priv_size = sizeof(struct netem_sched_data),
701 .enqueue = netem_enqueue,
702 .dequeue = netem_dequeue,
703 .requeue = netem_requeue,
704 .drop = netem_drop,
705 .init = netem_init,
706 .reset = netem_reset,
707 .destroy = netem_destroy,
708 .change = netem_change,
709 .dump = netem_dump,
710 .owner = THIS_MODULE,
711};
712
713
714static int __init netem_module_init(void)
715{
Stephen Hemmingereb229c42005-11-03 13:49:01 -0800716 pr_info("netem: version " VERSION "\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 return register_qdisc(&netem_qdisc_ops);
718}
719static void __exit netem_module_exit(void)
720{
721 unregister_qdisc(&netem_qdisc_ops);
722}
723module_init(netem_module_init)
724module_exit(netem_module_exit)
725MODULE_LICENSE("GPL");