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/init.h>
7 #include <linux/string.h>
8 #include <linux/errno.h>
9 #include <linux/skbuff.h>
10 #include <linux/atmdev.h>
11 #include <linux/atmclip.h>
12 #include <linux/rtnetlink.h>
13 #include <linux/file.h> /* for fput */
14 #include <net/netlink.h>
15 #include <net/pkt_sched.h>
17 extern struct socket *sockfd_lookup(int fd, int *err); /* @@@ fix this */
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 bstats;
53 struct gnet_stats_queue qstats;
54 struct atm_flow_data *next;
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 atm_flow_data *flows; /* NB: "link" is also on this
65 struct tasklet_struct task; /* requeue tasklet */
68 /* ------------------------- Class/flow operations ------------------------- */
70 static int find_flow(struct atm_qdisc_data *qdisc, struct atm_flow_data *flow)
72 struct atm_flow_data *walk;
74 pr_debug("find_flow(qdisc %p,flow %p)\n", qdisc, flow);
75 for (walk = qdisc->flows; walk; walk = walk->next)
78 pr_debug("find_flow: not found\n");
82 static inline struct atm_flow_data *lookup_flow(struct Qdisc *sch, u32 classid)
84 struct atm_qdisc_data *p = qdisc_priv(sch);
85 struct atm_flow_data *flow;
87 for (flow = p->flows; flow; flow = flow->next)
88 if (flow->classid == classid)
93 static int atm_tc_graft(struct Qdisc *sch, unsigned long arg,
94 struct Qdisc *new, struct Qdisc **old)
96 struct atm_qdisc_data *p = qdisc_priv(sch);
97 struct atm_flow_data *flow = (struct atm_flow_data *)arg;
99 pr_debug("atm_tc_graft(sch %p,[qdisc %p],flow %p,new %p,old %p)\n",
100 sch, p, flow, new, old);
101 if (!find_flow(p, flow))
105 *old = xchg(&flow->q, new);
111 static struct Qdisc *atm_tc_leaf(struct Qdisc *sch, unsigned long cl)
113 struct atm_flow_data *flow = (struct atm_flow_data *)cl;
115 pr_debug("atm_tc_leaf(sch %p,flow %p)\n", sch, flow);
116 return flow ? flow->q : NULL;
119 static unsigned long atm_tc_get(struct Qdisc *sch, u32 classid)
121 struct atm_qdisc_data *p __maybe_unused = qdisc_priv(sch);
122 struct atm_flow_data *flow;
124 pr_debug("atm_tc_get(sch %p,[qdisc %p],classid %x)\n", sch, p, classid);
125 flow = lookup_flow(sch, classid);
128 pr_debug("atm_tc_get: flow %p\n", flow);
129 return (unsigned long)flow;
132 static unsigned long atm_tc_bind_filter(struct Qdisc *sch,
133 unsigned long parent, u32 classid)
135 return atm_tc_get(sch, classid);
139 * atm_tc_put handles all destructions, including the ones that are explicitly
140 * requested (atm_tc_destroy, etc.). The assumption here is that we never drop
141 * anything that still seems to be in use.
143 static void atm_tc_put(struct Qdisc *sch, unsigned long cl)
145 struct atm_qdisc_data *p = qdisc_priv(sch);
146 struct atm_flow_data *flow = (struct atm_flow_data *)cl;
147 struct atm_flow_data **prev;
149 pr_debug("atm_tc_put(sch %p,[qdisc %p],flow %p)\n", sch, p, flow);
152 pr_debug("atm_tc_put: destroying\n");
153 for (prev = &p->flows; *prev; prev = &(*prev)->next)
157 printk(KERN_CRIT "atm_tc_put: class %p not found\n", flow);
161 pr_debug("atm_tc_put: qdisc %p\n", flow->q);
162 qdisc_destroy(flow->q);
163 tcf_destroy_chain(flow->filter_list);
165 pr_debug("atm_tc_put: f_count %d\n",
166 file_count(flow->sock->file));
167 flow->vcc->pop = flow->old_pop;
168 sockfd_put(flow->sock);
171 atm_tc_put(sch, (unsigned long)flow->excess);
172 if (flow != &p->link)
175 * If flow == &p->link, the qdisc no longer works at this point and
176 * needs to be removed. (By the caller of atm_tc_put.)
180 static void sch_atm_pop(struct atm_vcc *vcc, struct sk_buff *skb)
182 struct atm_qdisc_data *p = VCC2FLOW(vcc)->parent;
184 pr_debug("sch_atm_pop(vcc %p,skb %p,[qdisc %p])\n", vcc, skb, p);
185 VCC2FLOW(vcc)->old_pop(vcc, skb);
186 tasklet_schedule(&p->task);
189 static const u8 llc_oui_ip[] = {
190 0xaa, /* DSAP: non-ISO */
191 0xaa, /* SSAP: non-ISO */
192 0x03, /* Ctrl: Unnumbered Information Command PDU */
193 0x00, /* OUI: EtherType */
196 }; /* Ethertype IP (0800) */
198 static int atm_tc_change(struct Qdisc *sch, u32 classid, u32 parent,
199 struct nlattr **tca, unsigned long *arg)
201 struct atm_qdisc_data *p = qdisc_priv(sch);
202 struct atm_flow_data *flow = (struct atm_flow_data *)*arg;
203 struct atm_flow_data *excess = NULL;
204 struct nlattr *opt = tca[TCA_OPTIONS];
205 struct nlattr *tb[TCA_ATM_MAX + 1];
207 int fd, error, hdr_len;
210 pr_debug("atm_tc_change(sch %p,[qdisc %p],classid %x,parent %x,"
211 "flow %p,opt %p)\n", sch, p, classid, parent, flow, opt);
213 * The concept of parents doesn't apply for this qdisc.
215 if (parent && parent != TC_H_ROOT && parent != sch->handle)
218 * ATM classes cannot be changed. In order to change properties of the
219 * ATM connection, that socket needs to be modified directly (via the
220 * native ATM API. In order to send a flow to a different VC, the old
221 * class needs to be removed and a new one added. (This may be changed
228 error = nla_parse_nested(tb, TCA_ATM_MAX, opt, NULL);
232 if (!tb[TCA_ATM_FD] || nla_len(tb[TCA_ATM_FD]) < sizeof(fd))
234 fd = *(int *)nla_data(tb[TCA_ATM_FD]);
235 pr_debug("atm_tc_change: fd %d\n", fd);
236 if (tb[TCA_ATM_HDR]) {
237 hdr_len = nla_len(tb[TCA_ATM_HDR]);
238 hdr = nla_data(tb[TCA_ATM_HDR]);
240 hdr_len = RFC1483LLC_LEN;
241 hdr = NULL; /* default LLC/SNAP for IP */
243 if (!tb[TCA_ATM_EXCESS])
246 if (nla_len(tb[TCA_ATM_EXCESS]) != sizeof(u32))
248 excess = (struct atm_flow_data *)
249 atm_tc_get(sch, *(u32 *)nla_data(tb[TCA_ATM_EXCESS]));
253 pr_debug("atm_tc_change: type %d, payload %d, hdr_len %d\n",
254 opt->nla_type, nla_len(opt), hdr_len);
255 sock = sockfd_lookup(fd, &error);
257 return error; /* f_count++ */
258 pr_debug("atm_tc_change: f_count %d\n", file_count(sock->file));
259 if (sock->ops->family != PF_ATMSVC && sock->ops->family != PF_ATMPVC) {
263 /* @@@ should check if the socket is really operational or we'll crash
266 if (TC_H_MAJ(classid ^ sch->handle)) {
267 pr_debug("atm_tc_change: classid mismatch\n");
271 if (find_flow(p, flow)) {
279 for (i = 1; i < 0x8000; i++) {
280 classid = TC_H_MAKE(sch->handle, 0x8000 | i);
281 cl = atm_tc_get(sch, classid);
287 pr_debug("atm_tc_change: new id %x\n", classid);
288 flow = kzalloc(sizeof(struct atm_flow_data) + hdr_len, GFP_KERNEL);
289 pr_debug("atm_tc_change: flow %p\n", flow);
294 flow->filter_list = NULL;
295 flow->q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, classid);
297 flow->q = &noop_qdisc;
298 pr_debug("atm_tc_change: qdisc %p\n", flow->q);
300 flow->vcc = ATM_SD(sock); /* speedup */
301 flow->vcc->user_back = flow;
302 pr_debug("atm_tc_change: vcc %p\n", flow->vcc);
303 flow->old_pop = flow->vcc->pop;
305 flow->vcc->pop = sch_atm_pop;
306 flow->classid = classid;
308 flow->excess = excess;
309 flow->next = p->link.next;
311 flow->hdr_len = hdr_len;
313 memcpy(flow->hdr, hdr, hdr_len);
315 memcpy(flow->hdr, llc_oui_ip, sizeof(llc_oui_ip));
316 *arg = (unsigned long)flow;
320 atm_tc_put(sch, (unsigned long)excess);
325 static int atm_tc_delete(struct Qdisc *sch, unsigned long arg)
327 struct atm_qdisc_data *p = qdisc_priv(sch);
328 struct atm_flow_data *flow = (struct atm_flow_data *)arg;
330 pr_debug("atm_tc_delete(sch %p,[qdisc %p],flow %p)\n", sch, p, flow);
331 if (!find_flow(qdisc_priv(sch), flow))
333 if (flow->filter_list || flow == &p->link)
336 * Reference count must be 2: one for "keepalive" (set at class
337 * creation), and one for the reference held when calling delete.
340 printk(KERN_ERR "atm_tc_delete: flow->ref == %d\n", flow->ref);
344 return -EBUSY; /* catch references via excess, etc. */
345 atm_tc_put(sch, arg);
349 static void atm_tc_walk(struct Qdisc *sch, struct qdisc_walker *walker)
351 struct atm_qdisc_data *p = qdisc_priv(sch);
352 struct atm_flow_data *flow;
354 pr_debug("atm_tc_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
357 for (flow = p->flows; flow; flow = flow->next) {
358 if (walker->count >= walker->skip)
359 if (walker->fn(sch, (unsigned long)flow, walker) < 0) {
367 static struct tcf_proto **atm_tc_find_tcf(struct Qdisc *sch, unsigned long cl)
369 struct atm_qdisc_data *p = qdisc_priv(sch);
370 struct atm_flow_data *flow = (struct atm_flow_data *)cl;
372 pr_debug("atm_tc_find_tcf(sch %p,[qdisc %p],flow %p)\n", sch, p, flow);
373 return flow ? &flow->filter_list : &p->link.filter_list;
376 /* --------------------------- Qdisc operations ---------------------------- */
378 static int atm_tc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
380 struct atm_qdisc_data *p = qdisc_priv(sch);
381 struct atm_flow_data *flow = NULL; /* @@@ */
382 struct tcf_result res;
384 int ret = NET_XMIT_POLICED;
386 pr_debug("atm_tc_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
387 result = TC_POLICE_OK; /* be nice to gcc */
388 if (TC_H_MAJ(skb->priority) != sch->handle ||
389 !(flow = (struct atm_flow_data *)atm_tc_get(sch, skb->priority)))
390 for (flow = p->flows; flow; flow = flow->next)
391 if (flow->filter_list) {
392 result = tc_classify_compat(skb,
397 flow = (struct atm_flow_data *)res.class;
399 flow = lookup_flow(sch, res.classid);
406 ATM_SKB(skb)->atm_options = flow->vcc->atm_options;
407 /*@@@ looks good ... but it's not supposed to work :-) */
408 #ifdef CONFIG_NET_CLS_ACT
413 return NET_XMIT_SUCCESS;
417 case TC_POLICE_RECLASSIFY:
421 ATM_SKB(skb)->atm_options |= ATM_ATMOPT_CLP;
427 ret = flow->q->enqueue(skb, flow->q);
432 flow->qstats.drops++;
435 sch->bstats.bytes += skb->len;
436 sch->bstats.packets++;
437 flow->bstats.bytes += skb->len;
438 flow->bstats.packets++;
440 * Okay, this may seem weird. We pretend we've dropped the packet if
441 * it goes via ATM. The reason for this is that the outer qdisc
442 * expects to be able to q->dequeue the packet later on if we return
443 * success at this place. Also, sch->q.qdisc needs to reflect whether
444 * there is a packet egligible for dequeuing or not. Note that the
445 * statistics of the outer qdisc are necessarily wrong because of all
446 * this. There's currently no correct solution for this.
448 if (flow == &p->link) {
452 tasklet_schedule(&p->task);
453 return NET_XMIT_BYPASS;
457 * Dequeue packets and send them over ATM. Note that we quite deliberately
458 * avoid checking net_device's flow control here, simply because sch_atm
459 * uses its own channels, which have nothing to do with any CLIP/LANE/or
460 * non-ATM interfaces.
463 static void sch_atm_dequeue(unsigned long data)
465 struct Qdisc *sch = (struct Qdisc *)data;
466 struct atm_qdisc_data *p = qdisc_priv(sch);
467 struct atm_flow_data *flow;
470 pr_debug("sch_atm_dequeue(sch %p,[qdisc %p])\n", sch, p);
471 for (flow = p->link.next; flow; flow = flow->next)
473 * If traffic is properly shaped, this won't generate nasty
474 * little bursts. Otherwise, it may ... (but that's okay)
476 while ((skb = flow->q->dequeue(flow->q))) {
477 if (!atm_may_send(flow->vcc, skb->truesize)) {
478 (void)flow->q->ops->requeue(skb, flow->q);
481 pr_debug("atm_tc_dequeue: sending on class %p\n", flow);
482 /* remove any LL header somebody else has attached */
483 skb_pull(skb, skb_network_offset(skb));
484 if (skb_headroom(skb) < flow->hdr_len) {
487 new = skb_realloc_headroom(skb, flow->hdr_len);
493 pr_debug("sch_atm_dequeue: ip %p, data %p\n",
494 skb_network_header(skb), skb->data);
495 ATM_SKB(skb)->vcc = flow->vcc;
496 memcpy(skb_push(skb, flow->hdr_len), flow->hdr,
498 atomic_add(skb->truesize,
499 &sk_atm(flow->vcc)->sk_wmem_alloc);
500 /* atm.atm_options are already set by atm_tc_enqueue */
501 flow->vcc->send(flow->vcc, skb);
505 static struct sk_buff *atm_tc_dequeue(struct Qdisc *sch)
507 struct atm_qdisc_data *p = qdisc_priv(sch);
510 pr_debug("atm_tc_dequeue(sch %p,[qdisc %p])\n", sch, p);
511 tasklet_schedule(&p->task);
512 skb = p->link.q->dequeue(p->link.q);
518 static int atm_tc_requeue(struct sk_buff *skb, struct Qdisc *sch)
520 struct atm_qdisc_data *p = qdisc_priv(sch);
523 pr_debug("atm_tc_requeue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
524 ret = p->link.q->ops->requeue(skb, p->link.q);
527 sch->qstats.requeues++;
530 p->link.qstats.drops++;
535 static unsigned int atm_tc_drop(struct Qdisc *sch)
537 struct atm_qdisc_data *p = qdisc_priv(sch);
538 struct atm_flow_data *flow;
541 pr_debug("atm_tc_drop(sch %p,[qdisc %p])\n", sch, p);
542 for (flow = p->flows; flow; flow = flow->next)
543 if (flow->q->ops->drop && (len = flow->q->ops->drop(flow->q)))
548 static int atm_tc_init(struct Qdisc *sch, struct nlattr *opt)
550 struct atm_qdisc_data *p = qdisc_priv(sch);
552 pr_debug("atm_tc_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
554 p->link.q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, sch->handle);
556 p->link.q = &noop_qdisc;
557 pr_debug("atm_tc_init: link (%p) qdisc %p\n", &p->link, p->link.q);
558 p->link.filter_list = NULL;
561 p->link.classid = sch->handle;
564 tasklet_init(&p->task, sch_atm_dequeue, (unsigned long)sch);
568 static void atm_tc_reset(struct Qdisc *sch)
570 struct atm_qdisc_data *p = qdisc_priv(sch);
571 struct atm_flow_data *flow;
573 pr_debug("atm_tc_reset(sch %p,[qdisc %p])\n", sch, p);
574 for (flow = p->flows; flow; flow = flow->next)
575 qdisc_reset(flow->q);
579 static void atm_tc_destroy(struct Qdisc *sch)
581 struct atm_qdisc_data *p = qdisc_priv(sch);
582 struct atm_flow_data *flow;
584 pr_debug("atm_tc_destroy(sch %p,[qdisc %p])\n", sch, p);
586 while ((flow = p->flows)) {
587 tcf_destroy_chain(flow->filter_list);
588 flow->filter_list = NULL;
590 printk(KERN_ERR "atm_destroy: %p->ref = %d\n", flow,
592 atm_tc_put(sch, (unsigned long)flow);
593 if (p->flows == flow) {
594 printk(KERN_ERR "atm_destroy: putting flow %p didn't "
596 p->flows = flow->next; /* brute force */
600 tasklet_kill(&p->task);
603 static int atm_tc_dump_class(struct Qdisc *sch, unsigned long cl,
604 struct sk_buff *skb, struct tcmsg *tcm)
606 struct atm_qdisc_data *p = qdisc_priv(sch);
607 struct atm_flow_data *flow = (struct atm_flow_data *)cl;
610 pr_debug("atm_tc_dump_class(sch %p,[qdisc %p],flow %p,skb %p,tcm %p)\n",
611 sch, p, flow, skb, tcm);
612 if (!find_flow(p, flow))
614 tcm->tcm_handle = flow->classid;
615 tcm->tcm_info = flow->q->handle;
617 nest = nla_nest_start(skb, TCA_OPTIONS);
619 goto nla_put_failure;
621 NLA_PUT(skb, TCA_ATM_HDR, flow->hdr_len, flow->hdr);
623 struct sockaddr_atmpvc pvc;
626 pvc.sap_family = AF_ATMPVC;
627 pvc.sap_addr.itf = flow->vcc->dev ? flow->vcc->dev->number : -1;
628 pvc.sap_addr.vpi = flow->vcc->vpi;
629 pvc.sap_addr.vci = flow->vcc->vci;
630 NLA_PUT(skb, TCA_ATM_ADDR, sizeof(pvc), &pvc);
631 state = ATM_VF2VS(flow->vcc->flags);
632 NLA_PUT(skb, TCA_ATM_STATE, sizeof(state), &state);
635 NLA_PUT(skb, TCA_ATM_EXCESS, sizeof(u32), &flow->classid);
639 NLA_PUT(skb, TCA_ATM_EXCESS, sizeof(zero), &zero);
642 nla_nest_end(skb, nest);
646 nla_nest_cancel(skb, nest);
650 atm_tc_dump_class_stats(struct Qdisc *sch, unsigned long arg,
653 struct atm_flow_data *flow = (struct atm_flow_data *)arg;
655 flow->qstats.qlen = flow->q->q.qlen;
657 if (gnet_stats_copy_basic(d, &flow->bstats) < 0 ||
658 gnet_stats_copy_queue(d, &flow->qstats) < 0)
664 static int atm_tc_dump(struct Qdisc *sch, struct sk_buff *skb)
669 static const struct Qdisc_class_ops atm_class_ops = {
670 .graft = atm_tc_graft,
674 .change = atm_tc_change,
675 .delete = atm_tc_delete,
677 .tcf_chain = atm_tc_find_tcf,
678 .bind_tcf = atm_tc_bind_filter,
679 .unbind_tcf = atm_tc_put,
680 .dump = atm_tc_dump_class,
681 .dump_stats = atm_tc_dump_class_stats,
684 static struct Qdisc_ops atm_qdisc_ops __read_mostly = {
685 .cl_ops = &atm_class_ops,
687 .priv_size = sizeof(struct atm_qdisc_data),
688 .enqueue = atm_tc_enqueue,
689 .dequeue = atm_tc_dequeue,
690 .requeue = atm_tc_requeue,
693 .reset = atm_tc_reset,
694 .destroy = atm_tc_destroy,
696 .owner = THIS_MODULE,
699 static int __init atm_init(void)
701 return register_qdisc(&atm_qdisc_ops);
704 static void __exit atm_exit(void)
706 unregister_qdisc(&atm_qdisc_ops);
709 module_init(atm_init)
710 module_exit(atm_exit)
711 MODULE_LICENSE("GPL");