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>
20 * The ATM queuing discipline provides a framework for invoking classifiers
21 * (aka "filters"), which in turn select classes of this queuing discipline.
22 * Each class maps the flow(s) it is handling to a given VC. Multiple classes
23 * may share the same VC.
25 * When creating a class, VCs are specified by passing the number of the open
26 * socket descriptor by which the calling process references the VC. The kernel
27 * keeps the VC open at least until all classes using it are removed.
29 * In this file, most functions are named atm_tc_* to avoid confusion with all
30 * the atm_* in net/atm. This naming convention differs from what's used in the
34 * - sometimes messes up the IP stack
35 * - any manipulations besides the few operations described in the README, are
36 * untested and likely to crash the system
37 * - should lock the flow while there is data in the queue (?)
40 #define VCC2FLOW(vcc) ((struct atm_flow_data *) ((vcc)->user_back))
42 struct atm_flow_data {
43 struct Qdisc *q; /* FIFO, TBF, etc. */
44 struct tcf_proto *filter_list;
45 struct atm_vcc *vcc; /* VCC; NULL if VCC is closed */
46 void (*old_pop)(struct atm_vcc *vcc,
47 struct sk_buff *skb); /* chaining */
48 struct atm_qdisc_data *parent; /* parent qdisc */
49 struct socket *sock; /* for closing */
50 u32 classid; /* x:y type ID */
51 int ref; /* reference count */
52 struct gnet_stats_basic_packed bstats;
53 struct gnet_stats_queue qstats;
54 struct list_head list;
55 struct atm_flow_data *excess; /* flow for excess traffic;
56 NULL to set CLP instead */
58 unsigned char hdr[0]; /* header data; MUST BE LAST */
61 struct atm_qdisc_data {
62 struct atm_flow_data link; /* unclassified skbs go here */
63 struct list_head flows; /* NB: "link" is also on this
65 struct tasklet_struct task; /* dequeue tasklet */
68 /* ------------------------- Class/flow operations ------------------------- */
70 static inline struct atm_flow_data *lookup_flow(struct Qdisc *sch, u32 classid)
72 struct atm_qdisc_data *p = qdisc_priv(sch);
73 struct atm_flow_data *flow;
75 list_for_each_entry(flow, &p->flows, list) {
76 if (flow->classid == classid)
82 static int atm_tc_graft(struct Qdisc *sch, unsigned long arg,
83 struct Qdisc *new, struct Qdisc **old)
85 struct atm_qdisc_data *p = qdisc_priv(sch);
86 struct atm_flow_data *flow = (struct atm_flow_data *)arg;
88 pr_debug("atm_tc_graft(sch %p,[qdisc %p],flow %p,new %p,old %p)\n",
89 sch, p, flow, new, old);
90 if (list_empty(&flow->list))
101 static struct Qdisc *atm_tc_leaf(struct Qdisc *sch, unsigned long cl)
103 struct atm_flow_data *flow = (struct atm_flow_data *)cl;
105 pr_debug("atm_tc_leaf(sch %p,flow %p)\n", sch, flow);
106 return flow ? flow->q : NULL;
109 static unsigned long atm_tc_get(struct Qdisc *sch, u32 classid)
111 struct atm_qdisc_data *p __maybe_unused = qdisc_priv(sch);
112 struct atm_flow_data *flow;
114 pr_debug("atm_tc_get(sch %p,[qdisc %p],classid %x)\n", sch, p, classid);
115 flow = lookup_flow(sch, classid);
118 pr_debug("atm_tc_get: flow %p\n", flow);
119 return (unsigned long)flow;
122 static unsigned long atm_tc_bind_filter(struct Qdisc *sch,
123 unsigned long parent, u32 classid)
125 return atm_tc_get(sch, classid);
129 * atm_tc_put handles all destructions, including the ones that are explicitly
130 * requested (atm_tc_destroy, etc.). The assumption here is that we never drop
131 * anything that still seems to be in use.
133 static void atm_tc_put(struct Qdisc *sch, unsigned long cl)
135 struct atm_qdisc_data *p = qdisc_priv(sch);
136 struct atm_flow_data *flow = (struct atm_flow_data *)cl;
138 pr_debug("atm_tc_put(sch %p,[qdisc %p],flow %p)\n", sch, p, flow);
141 pr_debug("atm_tc_put: destroying\n");
142 list_del_init(&flow->list);
143 pr_debug("atm_tc_put: qdisc %p\n", flow->q);
144 qdisc_destroy(flow->q);
145 tcf_destroy_chain(&flow->filter_list);
147 pr_debug("atm_tc_put: f_count %ld\n",
148 file_count(flow->sock->file));
149 flow->vcc->pop = flow->old_pop;
150 sockfd_put(flow->sock);
153 atm_tc_put(sch, (unsigned long)flow->excess);
154 if (flow != &p->link)
157 * If flow == &p->link, the qdisc no longer works at this point and
158 * needs to be removed. (By the caller of atm_tc_put.)
162 static void sch_atm_pop(struct atm_vcc *vcc, struct sk_buff *skb)
164 struct atm_qdisc_data *p = VCC2FLOW(vcc)->parent;
166 pr_debug("sch_atm_pop(vcc %p,skb %p,[qdisc %p])\n", vcc, skb, p);
167 VCC2FLOW(vcc)->old_pop(vcc, skb);
168 tasklet_schedule(&p->task);
171 static const u8 llc_oui_ip[] = {
172 0xaa, /* DSAP: non-ISO */
173 0xaa, /* SSAP: non-ISO */
174 0x03, /* Ctrl: Unnumbered Information Command PDU */
175 0x00, /* OUI: EtherType */
178 }; /* Ethertype IP (0800) */
180 static const struct nla_policy atm_policy[TCA_ATM_MAX + 1] = {
181 [TCA_ATM_FD] = { .type = NLA_U32 },
182 [TCA_ATM_EXCESS] = { .type = NLA_U32 },
185 static int atm_tc_change(struct Qdisc *sch, u32 classid, u32 parent,
186 struct nlattr **tca, unsigned long *arg)
188 struct atm_qdisc_data *p = qdisc_priv(sch);
189 struct atm_flow_data *flow = (struct atm_flow_data *)*arg;
190 struct atm_flow_data *excess = NULL;
191 struct nlattr *opt = tca[TCA_OPTIONS];
192 struct nlattr *tb[TCA_ATM_MAX + 1];
194 int fd, error, hdr_len;
197 pr_debug("atm_tc_change(sch %p,[qdisc %p],classid %x,parent %x,"
198 "flow %p,opt %p)\n", sch, p, classid, parent, flow, opt);
200 * The concept of parents doesn't apply for this qdisc.
202 if (parent && parent != TC_H_ROOT && parent != sch->handle)
205 * ATM classes cannot be changed. In order to change properties of the
206 * ATM connection, that socket needs to be modified directly (via the
207 * native ATM API. In order to send a flow to a different VC, the old
208 * class needs to be removed and a new one added. (This may be changed
216 error = nla_parse_nested(tb, TCA_ATM_MAX, opt, atm_policy);
222 fd = nla_get_u32(tb[TCA_ATM_FD]);
223 pr_debug("atm_tc_change: fd %d\n", fd);
224 if (tb[TCA_ATM_HDR]) {
225 hdr_len = nla_len(tb[TCA_ATM_HDR]);
226 hdr = nla_data(tb[TCA_ATM_HDR]);
228 hdr_len = RFC1483LLC_LEN;
229 hdr = NULL; /* default LLC/SNAP for IP */
231 if (!tb[TCA_ATM_EXCESS])
234 excess = (struct atm_flow_data *)
235 atm_tc_get(sch, nla_get_u32(tb[TCA_ATM_EXCESS]));
239 pr_debug("atm_tc_change: type %d, payload %d, hdr_len %d\n",
240 opt->nla_type, nla_len(opt), hdr_len);
241 sock = sockfd_lookup(fd, &error);
243 return error; /* f_count++ */
244 pr_debug("atm_tc_change: f_count %ld\n", file_count(sock->file));
245 if (sock->ops->family != PF_ATMSVC && sock->ops->family != PF_ATMPVC) {
249 /* @@@ should check if the socket is really operational or we'll crash
252 if (TC_H_MAJ(classid ^ sch->handle)) {
253 pr_debug("atm_tc_change: classid mismatch\n");
261 for (i = 1; i < 0x8000; i++) {
262 classid = TC_H_MAKE(sch->handle, 0x8000 | i);
263 cl = atm_tc_get(sch, classid);
269 pr_debug("atm_tc_change: new id %x\n", classid);
270 flow = kzalloc(sizeof(struct atm_flow_data) + hdr_len, GFP_KERNEL);
271 pr_debug("atm_tc_change: flow %p\n", flow);
276 flow->filter_list = NULL;
277 flow->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, classid);
279 flow->q = &noop_qdisc;
280 pr_debug("atm_tc_change: qdisc %p\n", flow->q);
282 flow->vcc = ATM_SD(sock); /* speedup */
283 flow->vcc->user_back = flow;
284 pr_debug("atm_tc_change: vcc %p\n", flow->vcc);
285 flow->old_pop = flow->vcc->pop;
287 flow->vcc->pop = sch_atm_pop;
288 flow->classid = classid;
290 flow->excess = excess;
291 list_add(&flow->list, &p->link.list);
292 flow->hdr_len = hdr_len;
294 memcpy(flow->hdr, hdr, hdr_len);
296 memcpy(flow->hdr, llc_oui_ip, sizeof(llc_oui_ip));
297 *arg = (unsigned long)flow;
301 atm_tc_put(sch, (unsigned long)excess);
306 static int atm_tc_delete(struct Qdisc *sch, unsigned long arg)
308 struct atm_qdisc_data *p = qdisc_priv(sch);
309 struct atm_flow_data *flow = (struct atm_flow_data *)arg;
311 pr_debug("atm_tc_delete(sch %p,[qdisc %p],flow %p)\n", sch, p, flow);
312 if (list_empty(&flow->list))
314 if (flow->filter_list || flow == &p->link)
317 * Reference count must be 2: one for "keepalive" (set at class
318 * creation), and one for the reference held when calling delete.
321 pr_err("atm_tc_delete: flow->ref == %d\n", flow->ref);
325 return -EBUSY; /* catch references via excess, etc. */
326 atm_tc_put(sch, arg);
330 static void atm_tc_walk(struct Qdisc *sch, struct qdisc_walker *walker)
332 struct atm_qdisc_data *p = qdisc_priv(sch);
333 struct atm_flow_data *flow;
335 pr_debug("atm_tc_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
338 list_for_each_entry(flow, &p->flows, list) {
339 if (walker->count >= walker->skip &&
340 walker->fn(sch, (unsigned long)flow, walker) < 0) {
348 static struct tcf_proto **atm_tc_find_tcf(struct Qdisc *sch, unsigned long cl)
350 struct atm_qdisc_data *p = qdisc_priv(sch);
351 struct atm_flow_data *flow = (struct atm_flow_data *)cl;
353 pr_debug("atm_tc_find_tcf(sch %p,[qdisc %p],flow %p)\n", sch, p, flow);
354 return flow ? &flow->filter_list : &p->link.filter_list;
357 /* --------------------------- Qdisc operations ---------------------------- */
359 static int atm_tc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
361 struct atm_qdisc_data *p = qdisc_priv(sch);
362 struct atm_flow_data *flow;
363 struct tcf_result res;
365 int ret = NET_XMIT_POLICED;
367 pr_debug("atm_tc_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
368 result = TC_POLICE_OK; /* be nice to gcc */
370 if (TC_H_MAJ(skb->priority) != sch->handle ||
371 !(flow = (struct atm_flow_data *)atm_tc_get(sch, skb->priority))) {
372 list_for_each_entry(flow, &p->flows, list) {
373 if (flow->filter_list) {
374 result = tc_classify_compat(skb,
379 flow = (struct atm_flow_data *)res.class;
381 flow = lookup_flow(sch, res.classid);
393 ATM_SKB(skb)->atm_options = flow->vcc->atm_options;
394 /*@@@ looks good ... but it's not supposed to work :-) */
395 #ifdef CONFIG_NET_CLS_ACT
400 return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
404 case TC_POLICE_RECLASSIFY:
408 ATM_SKB(skb)->atm_options |= ATM_ATMOPT_CLP;
414 ret = qdisc_enqueue(skb, flow->q);
415 if (ret != NET_XMIT_SUCCESS) {
417 if (net_xmit_drop_count(ret)) {
420 flow->qstats.drops++;
425 * Okay, this may seem weird. We pretend we've dropped the packet if
426 * it goes via ATM. The reason for this is that the outer qdisc
427 * expects to be able to q->dequeue the packet later on if we return
428 * success at this place. Also, sch->q.qdisc needs to reflect whether
429 * there is a packet egligible for dequeuing or not. Note that the
430 * statistics of the outer qdisc are necessarily wrong because of all
431 * this. There's currently no correct solution for this.
433 if (flow == &p->link) {
435 return NET_XMIT_SUCCESS;
437 tasklet_schedule(&p->task);
438 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
442 * Dequeue packets and send them over ATM. Note that we quite deliberately
443 * avoid checking net_device's flow control here, simply because sch_atm
444 * uses its own channels, which have nothing to do with any CLIP/LANE/or
445 * non-ATM interfaces.
448 static void sch_atm_dequeue(unsigned long data)
450 struct Qdisc *sch = (struct Qdisc *)data;
451 struct atm_qdisc_data *p = qdisc_priv(sch);
452 struct atm_flow_data *flow;
455 pr_debug("sch_atm_dequeue(sch %p,[qdisc %p])\n", sch, p);
456 list_for_each_entry(flow, &p->flows, list) {
457 if (flow == &p->link)
460 * If traffic is properly shaped, this won't generate nasty
461 * little bursts. Otherwise, it may ... (but that's okay)
463 while ((skb = flow->q->ops->peek(flow->q))) {
464 if (!atm_may_send(flow->vcc, skb->truesize))
467 skb = qdisc_dequeue_peeked(flow->q);
471 qdisc_bstats_update(sch, skb);
472 bstats_update(&flow->bstats, skb);
473 pr_debug("atm_tc_dequeue: sending on class %p\n", flow);
474 /* remove any LL header somebody else has attached */
475 skb_pull(skb, skb_network_offset(skb));
476 if (skb_headroom(skb) < flow->hdr_len) {
479 new = skb_realloc_headroom(skb, flow->hdr_len);
485 pr_debug("sch_atm_dequeue: ip %p, data %p\n",
486 skb_network_header(skb), skb->data);
487 ATM_SKB(skb)->vcc = flow->vcc;
488 memcpy(skb_push(skb, flow->hdr_len), flow->hdr,
490 atomic_add(skb->truesize,
491 &sk_atm(flow->vcc)->sk_wmem_alloc);
492 /* atm.atm_options are already set by atm_tc_enqueue */
493 flow->vcc->send(flow->vcc, skb);
498 static struct sk_buff *atm_tc_dequeue(struct Qdisc *sch)
500 struct atm_qdisc_data *p = qdisc_priv(sch);
503 pr_debug("atm_tc_dequeue(sch %p,[qdisc %p])\n", sch, p);
504 tasklet_schedule(&p->task);
505 skb = qdisc_dequeue_peeked(p->link.q);
511 static struct sk_buff *atm_tc_peek(struct Qdisc *sch)
513 struct atm_qdisc_data *p = qdisc_priv(sch);
515 pr_debug("atm_tc_peek(sch %p,[qdisc %p])\n", sch, p);
517 return p->link.q->ops->peek(p->link.q);
520 static unsigned int atm_tc_drop(struct Qdisc *sch)
522 struct atm_qdisc_data *p = qdisc_priv(sch);
523 struct atm_flow_data *flow;
526 pr_debug("atm_tc_drop(sch %p,[qdisc %p])\n", sch, p);
527 list_for_each_entry(flow, &p->flows, list) {
528 if (flow->q->ops->drop && (len = flow->q->ops->drop(flow->q)))
534 static int atm_tc_init(struct Qdisc *sch, struct nlattr *opt)
536 struct atm_qdisc_data *p = qdisc_priv(sch);
538 pr_debug("atm_tc_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
539 INIT_LIST_HEAD(&p->flows);
540 INIT_LIST_HEAD(&p->link.list);
541 list_add(&p->link.list, &p->flows);
542 p->link.q = qdisc_create_dflt(sch->dev_queue,
543 &pfifo_qdisc_ops, sch->handle);
545 p->link.q = &noop_qdisc;
546 pr_debug("atm_tc_init: link (%p) qdisc %p\n", &p->link, p->link.q);
547 p->link.filter_list = NULL;
550 p->link.classid = sch->handle;
552 tasklet_init(&p->task, sch_atm_dequeue, (unsigned long)sch);
556 static void atm_tc_reset(struct Qdisc *sch)
558 struct atm_qdisc_data *p = qdisc_priv(sch);
559 struct atm_flow_data *flow;
561 pr_debug("atm_tc_reset(sch %p,[qdisc %p])\n", sch, p);
562 list_for_each_entry(flow, &p->flows, list)
563 qdisc_reset(flow->q);
567 static void atm_tc_destroy(struct Qdisc *sch)
569 struct atm_qdisc_data *p = qdisc_priv(sch);
570 struct atm_flow_data *flow, *tmp;
572 pr_debug("atm_tc_destroy(sch %p,[qdisc %p])\n", sch, p);
573 list_for_each_entry(flow, &p->flows, list)
574 tcf_destroy_chain(&flow->filter_list);
576 list_for_each_entry_safe(flow, tmp, &p->flows, list) {
578 pr_err("atm_destroy: %p->ref = %d\n", flow, flow->ref);
579 atm_tc_put(sch, (unsigned long)flow);
581 tasklet_kill(&p->task);
584 static int atm_tc_dump_class(struct Qdisc *sch, unsigned long cl,
585 struct sk_buff *skb, struct tcmsg *tcm)
587 struct atm_qdisc_data *p = qdisc_priv(sch);
588 struct atm_flow_data *flow = (struct atm_flow_data *)cl;
591 pr_debug("atm_tc_dump_class(sch %p,[qdisc %p],flow %p,skb %p,tcm %p)\n",
592 sch, p, flow, skb, tcm);
593 if (list_empty(&flow->list))
595 tcm->tcm_handle = flow->classid;
596 tcm->tcm_info = flow->q->handle;
598 nest = nla_nest_start(skb, TCA_OPTIONS);
600 goto nla_put_failure;
602 if (nla_put(skb, TCA_ATM_HDR, flow->hdr_len, flow->hdr))
603 goto nla_put_failure;
605 struct sockaddr_atmpvc pvc;
608 pvc.sap_family = AF_ATMPVC;
609 pvc.sap_addr.itf = flow->vcc->dev ? flow->vcc->dev->number : -1;
610 pvc.sap_addr.vpi = flow->vcc->vpi;
611 pvc.sap_addr.vci = flow->vcc->vci;
612 if (nla_put(skb, TCA_ATM_ADDR, sizeof(pvc), &pvc))
613 goto nla_put_failure;
614 state = ATM_VF2VS(flow->vcc->flags);
615 if (nla_put_u32(skb, TCA_ATM_STATE, state))
616 goto nla_put_failure;
619 if (nla_put_u32(skb, TCA_ATM_EXCESS, flow->classid))
620 goto nla_put_failure;
622 if (nla_put_u32(skb, TCA_ATM_EXCESS, 0))
623 goto nla_put_failure;
625 nla_nest_end(skb, nest);
629 nla_nest_cancel(skb, nest);
633 atm_tc_dump_class_stats(struct Qdisc *sch, unsigned long arg,
636 struct atm_flow_data *flow = (struct atm_flow_data *)arg;
638 flow->qstats.qlen = flow->q->q.qlen;
640 if (gnet_stats_copy_basic(d, &flow->bstats) < 0 ||
641 gnet_stats_copy_queue(d, &flow->qstats) < 0)
647 static int atm_tc_dump(struct Qdisc *sch, struct sk_buff *skb)
652 static const struct Qdisc_class_ops atm_class_ops = {
653 .graft = atm_tc_graft,
657 .change = atm_tc_change,
658 .delete = atm_tc_delete,
660 .tcf_chain = atm_tc_find_tcf,
661 .bind_tcf = atm_tc_bind_filter,
662 .unbind_tcf = atm_tc_put,
663 .dump = atm_tc_dump_class,
664 .dump_stats = atm_tc_dump_class_stats,
667 static struct Qdisc_ops atm_qdisc_ops __read_mostly = {
668 .cl_ops = &atm_class_ops,
670 .priv_size = sizeof(struct atm_qdisc_data),
671 .enqueue = atm_tc_enqueue,
672 .dequeue = atm_tc_dequeue,
676 .reset = atm_tc_reset,
677 .destroy = atm_tc_destroy,
679 .owner = THIS_MODULE,
682 static int __init atm_init(void)
684 return register_qdisc(&atm_qdisc_ops);
687 static void __exit atm_exit(void)
689 unregister_qdisc(&atm_qdisc_ops);
692 module_init(atm_init)
693 module_exit(atm_exit)
694 MODULE_LICENSE("GPL");