1
/* net/sched/sch_atm.c - ATM VC selection "queueing discipline" */
3
/* Written 1998-2000 by Werner Almesberger, EPFL ICA */
5
#include <linux/module.h>
6
#include <linux/slab.h>
7
#include <linux/init.h>
8
#include <linux/interrupt.h>
9
#include <linux/string.h>
10
#include <linux/errno.h>
11
#include <linux/skbuff.h>
12
#include <linux/atmdev.h>
13
#include <linux/atmclip.h>
14
#include <linux/rtnetlink.h>
15
#include <linux/file.h> /* for fput */
16
#include <net/netlink.h>
17
#include <net/pkt_sched.h>
19
extern struct socket *sockfd_lookup(int fd, int *err); /* @@@ fix this */
22
* The ATM queuing discipline provides a framework for invoking classifiers
23
* (aka "filters"), which in turn select classes of this queuing discipline.
24
* Each class maps the flow(s) it is handling to a given VC. Multiple classes
25
* may share the same VC.
27
* When creating a class, VCs are specified by passing the number of the open
28
* socket descriptor by which the calling process references the VC. The kernel
29
* keeps the VC open at least until all classes using it are removed.
31
* In this file, most functions are named atm_tc_* to avoid confusion with all
32
* the atm_* in net/atm. This naming convention differs from what's used in the
36
* - sometimes messes up the IP stack
37
* - any manipulations besides the few operations described in the README, are
38
* untested and likely to crash the system
39
* - should lock the flow while there is data in the queue (?)
42
#define VCC2FLOW(vcc) ((struct atm_flow_data *) ((vcc)->user_back))
44
struct atm_flow_data {
45
struct Qdisc *q; /* FIFO, TBF, etc. */
46
struct tcf_proto *filter_list;
47
struct atm_vcc *vcc; /* VCC; NULL if VCC is closed */
48
void (*old_pop)(struct atm_vcc *vcc,
49
struct sk_buff *skb); /* chaining */
50
struct atm_qdisc_data *parent; /* parent qdisc */
51
struct socket *sock; /* for closing */
52
u32 classid; /* x:y type ID */
53
int ref; /* reference count */
54
struct gnet_stats_basic_packed bstats;
55
struct gnet_stats_queue qstats;
56
struct list_head list;
57
struct atm_flow_data *excess; /* flow for excess traffic;
58
NULL to set CLP instead */
60
unsigned char hdr[0]; /* header data; MUST BE LAST */
63
struct atm_qdisc_data {
64
struct atm_flow_data link; /* unclassified skbs go here */
65
struct list_head flows; /* NB: "link" is also on this
67
struct tasklet_struct task; /* dequeue tasklet */
70
/* ------------------------- Class/flow operations ------------------------- */
72
static inline struct atm_flow_data *lookup_flow(struct Qdisc *sch, u32 classid)
74
struct atm_qdisc_data *p = qdisc_priv(sch);
75
struct atm_flow_data *flow;
77
list_for_each_entry(flow, &p->flows, list) {
78
if (flow->classid == classid)
84
static int atm_tc_graft(struct Qdisc *sch, unsigned long arg,
85
struct Qdisc *new, struct Qdisc **old)
87
struct atm_qdisc_data *p = qdisc_priv(sch);
88
struct atm_flow_data *flow = (struct atm_flow_data *)arg;
90
pr_debug("atm_tc_graft(sch %p,[qdisc %p],flow %p,new %p,old %p)\n",
91
sch, p, flow, new, old);
92
if (list_empty(&flow->list))
103
static struct Qdisc *atm_tc_leaf(struct Qdisc *sch, unsigned long cl)
105
struct atm_flow_data *flow = (struct atm_flow_data *)cl;
107
pr_debug("atm_tc_leaf(sch %p,flow %p)\n", sch, flow);
108
return flow ? flow->q : NULL;
111
static unsigned long atm_tc_get(struct Qdisc *sch, u32 classid)
113
struct atm_qdisc_data *p __maybe_unused = qdisc_priv(sch);
114
struct atm_flow_data *flow;
116
pr_debug("atm_tc_get(sch %p,[qdisc %p],classid %x)\n", sch, p, classid);
117
flow = lookup_flow(sch, classid);
120
pr_debug("atm_tc_get: flow %p\n", flow);
121
return (unsigned long)flow;
124
static unsigned long atm_tc_bind_filter(struct Qdisc *sch,
125
unsigned long parent, u32 classid)
127
return atm_tc_get(sch, classid);
131
* atm_tc_put handles all destructions, including the ones that are explicitly
132
* requested (atm_tc_destroy, etc.). The assumption here is that we never drop
133
* anything that still seems to be in use.
135
static void atm_tc_put(struct Qdisc *sch, unsigned long cl)
137
struct atm_qdisc_data *p = qdisc_priv(sch);
138
struct atm_flow_data *flow = (struct atm_flow_data *)cl;
140
pr_debug("atm_tc_put(sch %p,[qdisc %p],flow %p)\n", sch, p, flow);
143
pr_debug("atm_tc_put: destroying\n");
144
list_del_init(&flow->list);
145
pr_debug("atm_tc_put: qdisc %p\n", flow->q);
146
qdisc_destroy(flow->q);
147
tcf_destroy_chain(&flow->filter_list);
149
pr_debug("atm_tc_put: f_count %ld\n",
150
file_count(flow->sock->file));
151
flow->vcc->pop = flow->old_pop;
152
sockfd_put(flow->sock);
155
atm_tc_put(sch, (unsigned long)flow->excess);
156
if (flow != &p->link)
159
* If flow == &p->link, the qdisc no longer works at this point and
160
* needs to be removed. (By the caller of atm_tc_put.)
164
static void sch_atm_pop(struct atm_vcc *vcc, struct sk_buff *skb)
166
struct atm_qdisc_data *p = VCC2FLOW(vcc)->parent;
168
pr_debug("sch_atm_pop(vcc %p,skb %p,[qdisc %p])\n", vcc, skb, p);
169
VCC2FLOW(vcc)->old_pop(vcc, skb);
170
tasklet_schedule(&p->task);
173
static const u8 llc_oui_ip[] = {
174
0xaa, /* DSAP: non-ISO */
175
0xaa, /* SSAP: non-ISO */
176
0x03, /* Ctrl: Unnumbered Information Command PDU */
177
0x00, /* OUI: EtherType */
180
}; /* Ethertype IP (0800) */
182
static const struct nla_policy atm_policy[TCA_ATM_MAX + 1] = {
183
[TCA_ATM_FD] = { .type = NLA_U32 },
184
[TCA_ATM_EXCESS] = { .type = NLA_U32 },
187
static int atm_tc_change(struct Qdisc *sch, u32 classid, u32 parent,
188
struct nlattr **tca, unsigned long *arg)
190
struct atm_qdisc_data *p = qdisc_priv(sch);
191
struct atm_flow_data *flow = (struct atm_flow_data *)*arg;
192
struct atm_flow_data *excess = NULL;
193
struct nlattr *opt = tca[TCA_OPTIONS];
194
struct nlattr *tb[TCA_ATM_MAX + 1];
196
int fd, error, hdr_len;
199
pr_debug("atm_tc_change(sch %p,[qdisc %p],classid %x,parent %x,"
200
"flow %p,opt %p)\n", sch, p, classid, parent, flow, opt);
202
* The concept of parents doesn't apply for this qdisc.
204
if (parent && parent != TC_H_ROOT && parent != sch->handle)
207
* ATM classes cannot be changed. In order to change properties of the
208
* ATM connection, that socket needs to be modified directly (via the
209
* native ATM API. In order to send a flow to a different VC, the old
210
* class needs to be removed and a new one added. (This may be changed
218
error = nla_parse_nested(tb, TCA_ATM_MAX, opt, atm_policy);
224
fd = nla_get_u32(tb[TCA_ATM_FD]);
225
pr_debug("atm_tc_change: fd %d\n", fd);
226
if (tb[TCA_ATM_HDR]) {
227
hdr_len = nla_len(tb[TCA_ATM_HDR]);
228
hdr = nla_data(tb[TCA_ATM_HDR]);
230
hdr_len = RFC1483LLC_LEN;
231
hdr = NULL; /* default LLC/SNAP for IP */
233
if (!tb[TCA_ATM_EXCESS])
236
excess = (struct atm_flow_data *)
237
atm_tc_get(sch, nla_get_u32(tb[TCA_ATM_EXCESS]));
241
pr_debug("atm_tc_change: type %d, payload %d, hdr_len %d\n",
242
opt->nla_type, nla_len(opt), hdr_len);
243
sock = sockfd_lookup(fd, &error);
245
return error; /* f_count++ */
246
pr_debug("atm_tc_change: f_count %ld\n", file_count(sock->file));
247
if (sock->ops->family != PF_ATMSVC && sock->ops->family != PF_ATMPVC) {
251
/* @@@ should check if the socket is really operational or we'll crash
254
if (TC_H_MAJ(classid ^ sch->handle)) {
255
pr_debug("atm_tc_change: classid mismatch\n");
263
for (i = 1; i < 0x8000; i++) {
264
classid = TC_H_MAKE(sch->handle, 0x8000 | i);
265
cl = atm_tc_get(sch, classid);
271
pr_debug("atm_tc_change: new id %x\n", classid);
272
flow = kzalloc(sizeof(struct atm_flow_data) + hdr_len, GFP_KERNEL);
273
pr_debug("atm_tc_change: flow %p\n", flow);
278
flow->filter_list = NULL;
279
flow->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, classid);
281
flow->q = &noop_qdisc;
282
pr_debug("atm_tc_change: qdisc %p\n", flow->q);
284
flow->vcc = ATM_SD(sock); /* speedup */
285
flow->vcc->user_back = flow;
286
pr_debug("atm_tc_change: vcc %p\n", flow->vcc);
287
flow->old_pop = flow->vcc->pop;
289
flow->vcc->pop = sch_atm_pop;
290
flow->classid = classid;
292
flow->excess = excess;
293
list_add(&flow->list, &p->link.list);
294
flow->hdr_len = hdr_len;
296
memcpy(flow->hdr, hdr, hdr_len);
298
memcpy(flow->hdr, llc_oui_ip, sizeof(llc_oui_ip));
299
*arg = (unsigned long)flow;
303
atm_tc_put(sch, (unsigned long)excess);
308
static int atm_tc_delete(struct Qdisc *sch, unsigned long arg)
310
struct atm_qdisc_data *p = qdisc_priv(sch);
311
struct atm_flow_data *flow = (struct atm_flow_data *)arg;
313
pr_debug("atm_tc_delete(sch %p,[qdisc %p],flow %p)\n", sch, p, flow);
314
if (list_empty(&flow->list))
316
if (flow->filter_list || flow == &p->link)
319
* Reference count must be 2: one for "keepalive" (set at class
320
* creation), and one for the reference held when calling delete.
323
pr_err("atm_tc_delete: flow->ref == %d\n", flow->ref);
327
return -EBUSY; /* catch references via excess, etc. */
328
atm_tc_put(sch, arg);
332
static void atm_tc_walk(struct Qdisc *sch, struct qdisc_walker *walker)
334
struct atm_qdisc_data *p = qdisc_priv(sch);
335
struct atm_flow_data *flow;
337
pr_debug("atm_tc_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
340
list_for_each_entry(flow, &p->flows, list) {
341
if (walker->count >= walker->skip &&
342
walker->fn(sch, (unsigned long)flow, walker) < 0) {
350
static struct tcf_proto **atm_tc_find_tcf(struct Qdisc *sch, unsigned long cl)
352
struct atm_qdisc_data *p = qdisc_priv(sch);
353
struct atm_flow_data *flow = (struct atm_flow_data *)cl;
355
pr_debug("atm_tc_find_tcf(sch %p,[qdisc %p],flow %p)\n", sch, p, flow);
356
return flow ? &flow->filter_list : &p->link.filter_list;
359
/* --------------------------- Qdisc operations ---------------------------- */
361
static int atm_tc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
363
struct atm_qdisc_data *p = qdisc_priv(sch);
364
struct atm_flow_data *flow;
365
struct tcf_result res;
367
int ret = NET_XMIT_POLICED;
369
pr_debug("atm_tc_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
370
result = TC_POLICE_OK; /* be nice to gcc */
372
if (TC_H_MAJ(skb->priority) != sch->handle ||
373
!(flow = (struct atm_flow_data *)atm_tc_get(sch, skb->priority))) {
374
list_for_each_entry(flow, &p->flows, list) {
375
if (flow->filter_list) {
376
result = tc_classify_compat(skb,
381
flow = (struct atm_flow_data *)res.class;
383
flow = lookup_flow(sch, res.classid);
395
ATM_SKB(skb)->atm_options = flow->vcc->atm_options;
396
/*@@@ looks good ... but it's not supposed to work :-) */
397
#ifdef CONFIG_NET_CLS_ACT
402
return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
406
case TC_POLICE_RECLASSIFY:
410
ATM_SKB(skb)->atm_options |= ATM_ATMOPT_CLP;
416
ret = qdisc_enqueue(skb, flow->q);
417
if (ret != NET_XMIT_SUCCESS) {
419
if (net_xmit_drop_count(ret)) {
422
flow->qstats.drops++;
426
qdisc_bstats_update(sch, skb);
427
bstats_update(&flow->bstats, skb);
429
* Okay, this may seem weird. We pretend we've dropped the packet if
430
* it goes via ATM. The reason for this is that the outer qdisc
431
* expects to be able to q->dequeue the packet later on if we return
432
* success at this place. Also, sch->q.qdisc needs to reflect whether
433
* there is a packet egligible for dequeuing or not. Note that the
434
* statistics of the outer qdisc are necessarily wrong because of all
435
* this. There's currently no correct solution for this.
437
if (flow == &p->link) {
439
return NET_XMIT_SUCCESS;
441
tasklet_schedule(&p->task);
442
return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
446
* Dequeue packets and send them over ATM. Note that we quite deliberately
447
* avoid checking net_device's flow control here, simply because sch_atm
448
* uses its own channels, which have nothing to do with any CLIP/LANE/or
449
* non-ATM interfaces.
452
static void sch_atm_dequeue(unsigned long data)
454
struct Qdisc *sch = (struct Qdisc *)data;
455
struct atm_qdisc_data *p = qdisc_priv(sch);
456
struct atm_flow_data *flow;
459
pr_debug("sch_atm_dequeue(sch %p,[qdisc %p])\n", sch, p);
460
list_for_each_entry(flow, &p->flows, list) {
461
if (flow == &p->link)
464
* If traffic is properly shaped, this won't generate nasty
465
* little bursts. Otherwise, it may ... (but that's okay)
467
while ((skb = flow->q->ops->peek(flow->q))) {
468
if (!atm_may_send(flow->vcc, skb->truesize))
471
skb = qdisc_dequeue_peeked(flow->q);
475
pr_debug("atm_tc_dequeue: sending on class %p\n", flow);
476
/* remove any LL header somebody else has attached */
477
skb_pull(skb, skb_network_offset(skb));
478
if (skb_headroom(skb) < flow->hdr_len) {
481
new = skb_realloc_headroom(skb, flow->hdr_len);
487
pr_debug("sch_atm_dequeue: ip %p, data %p\n",
488
skb_network_header(skb), skb->data);
489
ATM_SKB(skb)->vcc = flow->vcc;
490
memcpy(skb_push(skb, flow->hdr_len), flow->hdr,
492
atomic_add(skb->truesize,
493
&sk_atm(flow->vcc)->sk_wmem_alloc);
494
/* atm.atm_options are already set by atm_tc_enqueue */
495
flow->vcc->send(flow->vcc, skb);
500
static struct sk_buff *atm_tc_dequeue(struct Qdisc *sch)
502
struct atm_qdisc_data *p = qdisc_priv(sch);
505
pr_debug("atm_tc_dequeue(sch %p,[qdisc %p])\n", sch, p);
506
tasklet_schedule(&p->task);
507
skb = qdisc_dequeue_peeked(p->link.q);
513
static struct sk_buff *atm_tc_peek(struct Qdisc *sch)
515
struct atm_qdisc_data *p = qdisc_priv(sch);
517
pr_debug("atm_tc_peek(sch %p,[qdisc %p])\n", sch, p);
519
return p->link.q->ops->peek(p->link.q);
522
static unsigned int atm_tc_drop(struct Qdisc *sch)
524
struct atm_qdisc_data *p = qdisc_priv(sch);
525
struct atm_flow_data *flow;
528
pr_debug("atm_tc_drop(sch %p,[qdisc %p])\n", sch, p);
529
list_for_each_entry(flow, &p->flows, list) {
530
if (flow->q->ops->drop && (len = flow->q->ops->drop(flow->q)))
536
static int atm_tc_init(struct Qdisc *sch, struct nlattr *opt)
538
struct atm_qdisc_data *p = qdisc_priv(sch);
540
pr_debug("atm_tc_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
541
INIT_LIST_HEAD(&p->flows);
542
INIT_LIST_HEAD(&p->link.list);
543
list_add(&p->link.list, &p->flows);
544
p->link.q = qdisc_create_dflt(sch->dev_queue,
545
&pfifo_qdisc_ops, sch->handle);
547
p->link.q = &noop_qdisc;
548
pr_debug("atm_tc_init: link (%p) qdisc %p\n", &p->link, p->link.q);
549
p->link.filter_list = NULL;
552
p->link.classid = sch->handle;
554
tasklet_init(&p->task, sch_atm_dequeue, (unsigned long)sch);
558
static void atm_tc_reset(struct Qdisc *sch)
560
struct atm_qdisc_data *p = qdisc_priv(sch);
561
struct atm_flow_data *flow;
563
pr_debug("atm_tc_reset(sch %p,[qdisc %p])\n", sch, p);
564
list_for_each_entry(flow, &p->flows, list)
565
qdisc_reset(flow->q);
569
static void atm_tc_destroy(struct Qdisc *sch)
571
struct atm_qdisc_data *p = qdisc_priv(sch);
572
struct atm_flow_data *flow, *tmp;
574
pr_debug("atm_tc_destroy(sch %p,[qdisc %p])\n", sch, p);
575
list_for_each_entry(flow, &p->flows, list)
576
tcf_destroy_chain(&flow->filter_list);
578
list_for_each_entry_safe(flow, tmp, &p->flows, list) {
580
pr_err("atm_destroy: %p->ref = %d\n", flow, flow->ref);
581
atm_tc_put(sch, (unsigned long)flow);
583
tasklet_kill(&p->task);
586
static int atm_tc_dump_class(struct Qdisc *sch, unsigned long cl,
587
struct sk_buff *skb, struct tcmsg *tcm)
589
struct atm_qdisc_data *p = qdisc_priv(sch);
590
struct atm_flow_data *flow = (struct atm_flow_data *)cl;
593
pr_debug("atm_tc_dump_class(sch %p,[qdisc %p],flow %p,skb %p,tcm %p)\n",
594
sch, p, flow, skb, tcm);
595
if (list_empty(&flow->list))
597
tcm->tcm_handle = flow->classid;
598
tcm->tcm_info = flow->q->handle;
600
nest = nla_nest_start(skb, TCA_OPTIONS);
602
goto nla_put_failure;
604
NLA_PUT(skb, TCA_ATM_HDR, flow->hdr_len, flow->hdr);
606
struct sockaddr_atmpvc pvc;
609
pvc.sap_family = AF_ATMPVC;
610
pvc.sap_addr.itf = flow->vcc->dev ? flow->vcc->dev->number : -1;
611
pvc.sap_addr.vpi = flow->vcc->vpi;
612
pvc.sap_addr.vci = flow->vcc->vci;
613
NLA_PUT(skb, TCA_ATM_ADDR, sizeof(pvc), &pvc);
614
state = ATM_VF2VS(flow->vcc->flags);
615
NLA_PUT_U32(skb, TCA_ATM_STATE, state);
618
NLA_PUT_U32(skb, TCA_ATM_EXCESS, flow->classid);
620
NLA_PUT_U32(skb, TCA_ATM_EXCESS, 0);
622
nla_nest_end(skb, nest);
626
nla_nest_cancel(skb, nest);
630
atm_tc_dump_class_stats(struct Qdisc *sch, unsigned long arg,
633
struct atm_flow_data *flow = (struct atm_flow_data *)arg;
635
flow->qstats.qlen = flow->q->q.qlen;
637
if (gnet_stats_copy_basic(d, &flow->bstats) < 0 ||
638
gnet_stats_copy_queue(d, &flow->qstats) < 0)
644
static int atm_tc_dump(struct Qdisc *sch, struct sk_buff *skb)
649
static const struct Qdisc_class_ops atm_class_ops = {
650
.graft = atm_tc_graft,
654
.change = atm_tc_change,
655
.delete = atm_tc_delete,
657
.tcf_chain = atm_tc_find_tcf,
658
.bind_tcf = atm_tc_bind_filter,
659
.unbind_tcf = atm_tc_put,
660
.dump = atm_tc_dump_class,
661
.dump_stats = atm_tc_dump_class_stats,
664
static struct Qdisc_ops atm_qdisc_ops __read_mostly = {
665
.cl_ops = &atm_class_ops,
667
.priv_size = sizeof(struct atm_qdisc_data),
668
.enqueue = atm_tc_enqueue,
669
.dequeue = atm_tc_dequeue,
673
.reset = atm_tc_reset,
674
.destroy = atm_tc_destroy,
676
.owner = THIS_MODULE,
679
static int __init atm_init(void)
681
return register_qdisc(&atm_qdisc_ops);
684
static void __exit atm_exit(void)
686
unregister_qdisc(&atm_qdisc_ops);
689
module_init(atm_init)
690
module_exit(atm_exit)
691
MODULE_LICENSE("GPL");