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 #define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
22 #define DPRINTK(format,args...)
26 #define D2PRINTK(format,args...) printk(KERN_DEBUG format,##args)
28 #define D2PRINTK(format,args...)
32 * The ATM queuing discipline provides a framework for invoking classifiers
33 * (aka "filters"), which in turn select classes of this queuing discipline.
34 * Each class maps the flow(s) it is handling to a given VC. Multiple classes
35 * may share the same VC.
37 * When creating a class, VCs are specified by passing the number of the open
38 * socket descriptor by which the calling process references the VC. The kernel
39 * keeps the VC open at least until all classes using it are removed.
41 * In this file, most functions are named atm_tc_* to avoid confusion with all
42 * the atm_* in net/atm. This naming convention differs from what's used in the
46 * - sometimes messes up the IP stack
47 * - any manipulations besides the few operations described in the README, are
48 * untested and likely to crash the system
49 * - should lock the flow while there is data in the queue (?)
52 #define PRIV(sch) qdisc_priv(sch)
53 #define VCC2FLOW(vcc) ((struct atm_flow_data *) ((vcc)->user_back))
55 struct atm_flow_data {
56 struct Qdisc *q; /* FIFO, TBF, etc. */
57 struct tcf_proto *filter_list;
58 struct atm_vcc *vcc; /* VCC; NULL if VCC is closed */
59 void (*old_pop)(struct atm_vcc *vcc,
60 struct sk_buff * skb); /* chaining */
61 struct atm_qdisc_data *parent; /* parent qdisc */
62 struct socket *sock; /* for closing */
63 u32 classid; /* x:y type ID */
64 int ref; /* reference count */
65 struct gnet_stats_basic bstats;
66 struct gnet_stats_queue qstats;
67 struct atm_flow_data *next;
68 struct atm_flow_data *excess; /* flow for excess traffic;
69 NULL to set CLP instead */
71 unsigned char hdr[0]; /* header data; MUST BE LAST */
74 struct atm_qdisc_data {
75 struct atm_flow_data link; /* unclassified skbs go here */
76 struct atm_flow_data *flows; /* NB: "link" is also on this
78 struct tasklet_struct task; /* requeue tasklet */
81 /* ------------------------- Class/flow operations ------------------------- */
83 static int find_flow(struct atm_qdisc_data *qdisc, struct atm_flow_data *flow)
85 struct atm_flow_data *walk;
87 DPRINTK("find_flow(qdisc %p,flow %p)\n", qdisc, flow);
88 for (walk = qdisc->flows; walk; walk = walk->next)
91 DPRINTK("find_flow: not found\n");
95 static inline struct atm_flow_data *lookup_flow(struct Qdisc *sch, u32 classid)
97 struct atm_qdisc_data *p = PRIV(sch);
98 struct atm_flow_data *flow;
100 for (flow = p->flows; flow; flow = flow->next)
101 if (flow->classid == classid)
106 static int atm_tc_graft(struct Qdisc *sch, unsigned long arg,
107 struct Qdisc *new, struct Qdisc **old)
109 struct atm_qdisc_data *p = PRIV(sch);
110 struct atm_flow_data *flow = (struct atm_flow_data *)arg;
112 DPRINTK("atm_tc_graft(sch %p,[qdisc %p],flow %p,new %p,old %p)\n",
113 sch, p, flow, new, old);
114 if (!find_flow(p, flow))
118 *old = xchg(&flow->q, new);
124 static struct Qdisc *atm_tc_leaf(struct Qdisc *sch, unsigned long cl)
126 struct atm_flow_data *flow = (struct atm_flow_data *)cl;
128 DPRINTK("atm_tc_leaf(sch %p,flow %p)\n", sch, flow);
129 return flow ? flow->q : NULL;
132 static unsigned long atm_tc_get(struct Qdisc *sch, u32 classid)
134 struct atm_qdisc_data *p __maybe_unused = PRIV(sch);
135 struct atm_flow_data *flow;
137 DPRINTK("atm_tc_get(sch %p,[qdisc %p],classid %x)\n", sch, p, classid);
138 flow = lookup_flow(sch, classid);
141 DPRINTK("atm_tc_get: flow %p\n", flow);
142 return (unsigned long)flow;
145 static unsigned long atm_tc_bind_filter(struct Qdisc *sch,
146 unsigned long parent, u32 classid)
148 return atm_tc_get(sch, classid);
152 * atm_tc_put handles all destructions, including the ones that are explicitly
153 * requested (atm_tc_destroy, etc.). The assumption here is that we never drop
154 * anything that still seems to be in use.
156 static void atm_tc_put(struct Qdisc *sch, unsigned long cl)
158 struct atm_qdisc_data *p = PRIV(sch);
159 struct atm_flow_data *flow = (struct atm_flow_data *)cl;
160 struct atm_flow_data **prev;
162 DPRINTK("atm_tc_put(sch %p,[qdisc %p],flow %p)\n", sch, p, flow);
165 DPRINTK("atm_tc_put: destroying\n");
166 for (prev = &p->flows; *prev; prev = &(*prev)->next)
170 printk(KERN_CRIT "atm_tc_put: class %p not found\n", flow);
174 DPRINTK("atm_tc_put: qdisc %p\n", flow->q);
175 qdisc_destroy(flow->q);
176 tcf_destroy_chain(flow->filter_list);
178 DPRINTK("atm_tc_put: f_count %d\n",
179 file_count(flow->sock->file));
180 flow->vcc->pop = flow->old_pop;
181 sockfd_put(flow->sock);
184 atm_tc_put(sch, (unsigned long)flow->excess);
185 if (flow != &p->link)
188 * If flow == &p->link, the qdisc no longer works at this point and
189 * needs to be removed. (By the caller of atm_tc_put.)
193 static void sch_atm_pop(struct atm_vcc *vcc, struct sk_buff *skb)
195 struct atm_qdisc_data *p = VCC2FLOW(vcc)->parent;
197 D2PRINTK("sch_atm_pop(vcc %p,skb %p,[qdisc %p])\n", vcc, skb, p);
198 VCC2FLOW(vcc)->old_pop(vcc, skb);
199 tasklet_schedule(&p->task);
202 static const u8 llc_oui_ip[] = {
203 0xaa, /* DSAP: non-ISO */
204 0xaa, /* SSAP: non-ISO */
205 0x03, /* Ctrl: Unnumbered Information Command PDU */
206 0x00, /* OUI: EtherType */
209 }; /* Ethertype IP (0800) */
211 static int atm_tc_change(struct Qdisc *sch, u32 classid, u32 parent,
212 struct rtattr **tca, unsigned long *arg)
214 struct atm_qdisc_data *p = PRIV(sch);
215 struct atm_flow_data *flow = (struct atm_flow_data *)*arg;
216 struct atm_flow_data *excess = NULL;
217 struct rtattr *opt = tca[TCA_OPTIONS - 1];
218 struct rtattr *tb[TCA_ATM_MAX];
220 int fd, error, hdr_len;
223 DPRINTK("atm_tc_change(sch %p,[qdisc %p],classid %x,parent %x,"
224 "flow %p,opt %p)\n", sch, p, classid, parent, flow, opt);
226 * The concept of parents doesn't apply for this qdisc.
228 if (parent && parent != TC_H_ROOT && parent != sch->handle)
231 * ATM classes cannot be changed. In order to change properties of the
232 * ATM connection, that socket needs to be modified directly (via the
233 * native ATM API. In order to send a flow to a different VC, the old
234 * class needs to be removed and a new one added. (This may be changed
239 if (opt == NULL || rtattr_parse_nested(tb, TCA_ATM_MAX, opt))
241 if (!tb[TCA_ATM_FD - 1] || RTA_PAYLOAD(tb[TCA_ATM_FD - 1]) < sizeof(fd))
243 fd = *(int *)RTA_DATA(tb[TCA_ATM_FD - 1]);
244 DPRINTK("atm_tc_change: fd %d\n", fd);
245 if (tb[TCA_ATM_HDR - 1]) {
246 hdr_len = RTA_PAYLOAD(tb[TCA_ATM_HDR - 1]);
247 hdr = RTA_DATA(tb[TCA_ATM_HDR - 1]);
249 hdr_len = RFC1483LLC_LEN;
250 hdr = NULL; /* default LLC/SNAP for IP */
252 if (!tb[TCA_ATM_EXCESS - 1])
255 if (RTA_PAYLOAD(tb[TCA_ATM_EXCESS - 1]) != sizeof(u32))
257 excess = (struct atm_flow_data *)
258 atm_tc_get(sch, *(u32 *)RTA_DATA(tb[TCA_ATM_EXCESS - 1]));
262 DPRINTK("atm_tc_change: type %d, payload %d, hdr_len %d\n",
263 opt->rta_type, RTA_PAYLOAD(opt), hdr_len);
264 if (!(sock = sockfd_lookup(fd, &error)))
265 return error; /* f_count++ */
266 DPRINTK("atm_tc_change: f_count %d\n", file_count(sock->file));
267 if (sock->ops->family != PF_ATMSVC && sock->ops->family != PF_ATMPVC) {
271 /* @@@ should check if the socket is really operational or we'll crash
274 if (TC_H_MAJ(classid ^ sch->handle)) {
275 DPRINTK("atm_tc_change: classid mismatch\n");
279 if (find_flow(p, flow)) {
287 for (i = 1; i < 0x8000; i++) {
288 classid = TC_H_MAKE(sch->handle, 0x8000 | i);
289 if (!(cl = atm_tc_get(sch, classid)))
294 DPRINTK("atm_tc_change: new id %x\n", classid);
295 flow = kzalloc(sizeof(struct atm_flow_data) + hdr_len, GFP_KERNEL);
296 DPRINTK("atm_tc_change: flow %p\n", flow);
301 flow->filter_list = NULL;
302 if (!(flow->q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, classid)))
303 flow->q = &noop_qdisc;
304 DPRINTK("atm_tc_change: qdisc %p\n", flow->q);
306 flow->vcc = ATM_SD(sock); /* speedup */
307 flow->vcc->user_back = flow;
308 DPRINTK("atm_tc_change: vcc %p\n", flow->vcc);
309 flow->old_pop = flow->vcc->pop;
311 flow->vcc->pop = sch_atm_pop;
312 flow->classid = classid;
314 flow->excess = excess;
315 flow->next = p->link.next;
317 flow->hdr_len = hdr_len;
319 memcpy(flow->hdr, hdr, hdr_len);
321 memcpy(flow->hdr, llc_oui_ip, sizeof(llc_oui_ip));
322 *arg = (unsigned long)flow;
326 atm_tc_put(sch, (unsigned long)excess);
331 static int atm_tc_delete(struct Qdisc *sch, unsigned long arg)
333 struct atm_qdisc_data *p = PRIV(sch);
334 struct atm_flow_data *flow = (struct atm_flow_data *)arg;
336 DPRINTK("atm_tc_delete(sch %p,[qdisc %p],flow %p)\n", sch, p, flow);
337 if (!find_flow(PRIV(sch), flow))
339 if (flow->filter_list || flow == &p->link)
342 * Reference count must be 2: one for "keepalive" (set at class
343 * creation), and one for the reference held when calling delete.
346 printk(KERN_ERR "atm_tc_delete: flow->ref == %d\n", flow->ref);
350 return -EBUSY; /* catch references via excess, etc. */
351 atm_tc_put(sch, arg);
355 static void atm_tc_walk(struct Qdisc *sch, struct qdisc_walker *walker)
357 struct atm_qdisc_data *p = PRIV(sch);
358 struct atm_flow_data *flow;
360 DPRINTK("atm_tc_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
363 for (flow = p->flows; flow; flow = flow->next) {
364 if (walker->count >= walker->skip)
365 if (walker->fn(sch, (unsigned long)flow, walker) < 0) {
373 static struct tcf_proto **atm_tc_find_tcf(struct Qdisc *sch, unsigned long cl)
375 struct atm_qdisc_data *p = PRIV(sch);
376 struct atm_flow_data *flow = (struct atm_flow_data *)cl;
378 DPRINTK("atm_tc_find_tcf(sch %p,[qdisc %p],flow %p)\n", sch, p, flow);
379 return flow ? &flow->filter_list : &p->link.filter_list;
382 /* --------------------------- Qdisc operations ---------------------------- */
384 static int atm_tc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
386 struct atm_qdisc_data *p = PRIV(sch);
387 struct atm_flow_data *flow = NULL; /* @@@ */
388 struct tcf_result res;
390 int ret = NET_XMIT_POLICED;
392 D2PRINTK("atm_tc_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
393 result = TC_POLICE_OK; /* be nice to gcc */
394 if (TC_H_MAJ(skb->priority) != sch->handle ||
395 !(flow = (struct atm_flow_data *)atm_tc_get(sch, skb->priority)))
396 for (flow = p->flows; flow; flow = flow->next)
397 if (flow->filter_list) {
398 result = tc_classify_compat(skb,
403 flow = (struct atm_flow_data *)res.class;
405 flow = lookup_flow(sch, res.classid);
412 ATM_SKB(skb)->atm_options = flow->vcc->atm_options;
413 /*@@@ looks good ... but it's not supposed to work :-) */
414 #ifdef CONFIG_NET_CLS_ACT
419 return NET_XMIT_SUCCESS;
423 case TC_POLICE_RECLASSIFY:
427 ATM_SKB(skb)->atm_options |= ATM_ATMOPT_CLP;
433 if ((ret = flow->q->enqueue(skb, flow->q)) != 0) {
437 flow->qstats.drops++;
440 sch->bstats.bytes += skb->len;
441 sch->bstats.packets++;
442 flow->bstats.bytes += skb->len;
443 flow->bstats.packets++;
445 * Okay, this may seem weird. We pretend we've dropped the packet if
446 * it goes via ATM. The reason for this is that the outer qdisc
447 * expects to be able to q->dequeue the packet later on if we return
448 * success at this place. Also, sch->q.qdisc needs to reflect whether
449 * there is a packet egligible for dequeuing or not. Note that the
450 * statistics of the outer qdisc are necessarily wrong because of all
451 * this. There's currently no correct solution for this.
453 if (flow == &p->link) {
457 tasklet_schedule(&p->task);
458 return NET_XMIT_BYPASS;
462 * Dequeue packets and send them over ATM. Note that we quite deliberately
463 * avoid checking net_device's flow control here, simply because sch_atm
464 * uses its own channels, which have nothing to do with any CLIP/LANE/or
465 * non-ATM interfaces.
468 static void sch_atm_dequeue(unsigned long data)
470 struct Qdisc *sch = (struct Qdisc *)data;
471 struct atm_qdisc_data *p = PRIV(sch);
472 struct atm_flow_data *flow;
475 D2PRINTK("sch_atm_dequeue(sch %p,[qdisc %p])\n", sch, p);
476 for (flow = p->link.next; flow; flow = flow->next)
478 * If traffic is properly shaped, this won't generate nasty
479 * little bursts. Otherwise, it may ... (but that's okay)
481 while ((skb = flow->q->dequeue(flow->q))) {
482 if (!atm_may_send(flow->vcc, skb->truesize)) {
483 (void)flow->q->ops->requeue(skb, flow->q);
486 D2PRINTK("atm_tc_dequeue: sending on class %p\n", flow);
487 /* remove any LL header somebody else has attached */
488 skb_pull(skb, skb_network_offset(skb));
489 if (skb_headroom(skb) < flow->hdr_len) {
492 new = skb_realloc_headroom(skb, flow->hdr_len);
498 D2PRINTK("sch_atm_dequeue: ip %p, data %p\n",
499 skb_network_header(skb), skb->data);
500 ATM_SKB(skb)->vcc = flow->vcc;
501 memcpy(skb_push(skb, flow->hdr_len), flow->hdr,
503 atomic_add(skb->truesize,
504 &sk_atm(flow->vcc)->sk_wmem_alloc);
505 /* atm.atm_options are already set by atm_tc_enqueue */
506 flow->vcc->send(flow->vcc, skb);
510 static struct sk_buff *atm_tc_dequeue(struct Qdisc *sch)
512 struct atm_qdisc_data *p = PRIV(sch);
515 D2PRINTK("atm_tc_dequeue(sch %p,[qdisc %p])\n", sch, p);
516 tasklet_schedule(&p->task);
517 skb = p->link.q->dequeue(p->link.q);
523 static int atm_tc_requeue(struct sk_buff *skb, struct Qdisc *sch)
525 struct atm_qdisc_data *p = PRIV(sch);
528 D2PRINTK("atm_tc_requeue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
529 ret = p->link.q->ops->requeue(skb, p->link.q);
532 sch->qstats.requeues++;
535 p->link.qstats.drops++;
540 static unsigned int atm_tc_drop(struct Qdisc *sch)
542 struct atm_qdisc_data *p = PRIV(sch);
543 struct atm_flow_data *flow;
546 DPRINTK("atm_tc_drop(sch %p,[qdisc %p])\n", sch, p);
547 for (flow = p->flows; flow; flow = flow->next)
548 if (flow->q->ops->drop && (len = flow->q->ops->drop(flow->q)))
553 static int atm_tc_init(struct Qdisc *sch, struct rtattr *opt)
555 struct atm_qdisc_data *p = PRIV(sch);
557 DPRINTK("atm_tc_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
559 if (!(p->link.q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
561 p->link.q = &noop_qdisc;
562 DPRINTK("atm_tc_init: link (%p) qdisc %p\n", &p->link, p->link.q);
563 p->link.filter_list = NULL;
566 p->link.classid = sch->handle;
569 tasklet_init(&p->task, sch_atm_dequeue, (unsigned long)sch);
573 static void atm_tc_reset(struct Qdisc *sch)
575 struct atm_qdisc_data *p = PRIV(sch);
576 struct atm_flow_data *flow;
578 DPRINTK("atm_tc_reset(sch %p,[qdisc %p])\n", sch, p);
579 for (flow = p->flows; flow; flow = flow->next)
580 qdisc_reset(flow->q);
584 static void atm_tc_destroy(struct Qdisc *sch)
586 struct atm_qdisc_data *p = PRIV(sch);
587 struct atm_flow_data *flow;
589 DPRINTK("atm_tc_destroy(sch %p,[qdisc %p])\n", sch, p);
591 while ((flow = p->flows)) {
592 tcf_destroy_chain(flow->filter_list);
593 flow->filter_list = NULL;
595 printk(KERN_ERR "atm_destroy: %p->ref = %d\n", flow,
597 atm_tc_put(sch, (unsigned long)flow);
598 if (p->flows == flow) {
599 printk(KERN_ERR "atm_destroy: putting flow %p didn't "
601 p->flows = flow->next; /* brute force */
605 tasklet_kill(&p->task);
608 static int atm_tc_dump_class(struct Qdisc *sch, unsigned long cl,
609 struct sk_buff *skb, struct tcmsg *tcm)
611 struct atm_qdisc_data *p = PRIV(sch);
612 struct atm_flow_data *flow = (struct atm_flow_data *)cl;
613 unsigned char *b = skb_tail_pointer(skb);
616 DPRINTK("atm_tc_dump_class(sch %p,[qdisc %p],flow %p,skb %p,tcm %p)\n",
617 sch, p, flow, skb, tcm);
618 if (!find_flow(p, flow))
620 tcm->tcm_handle = flow->classid;
621 tcm->tcm_info = flow->q->handle;
622 rta = (struct rtattr *)b;
623 RTA_PUT(skb, TCA_OPTIONS, 0, NULL);
624 RTA_PUT(skb, TCA_ATM_HDR, flow->hdr_len, flow->hdr);
626 struct sockaddr_atmpvc pvc;
629 pvc.sap_family = AF_ATMPVC;
630 pvc.sap_addr.itf = flow->vcc->dev ? flow->vcc->dev->number : -1;
631 pvc.sap_addr.vpi = flow->vcc->vpi;
632 pvc.sap_addr.vci = flow->vcc->vci;
633 RTA_PUT(skb, TCA_ATM_ADDR, sizeof(pvc), &pvc);
634 state = ATM_VF2VS(flow->vcc->flags);
635 RTA_PUT(skb, TCA_ATM_STATE, sizeof(state), &state);
638 RTA_PUT(skb, TCA_ATM_EXCESS, sizeof(u32), &flow->classid);
642 RTA_PUT(skb, TCA_ATM_EXCESS, sizeof(zero), &zero);
644 rta->rta_len = skb_tail_pointer(skb) - b;
652 atm_tc_dump_class_stats(struct Qdisc *sch, unsigned long arg,
655 struct atm_flow_data *flow = (struct atm_flow_data *)arg;
657 flow->qstats.qlen = flow->q->q.qlen;
659 if (gnet_stats_copy_basic(d, &flow->bstats) < 0 ||
660 gnet_stats_copy_queue(d, &flow->qstats) < 0)
666 static int atm_tc_dump(struct Qdisc *sch, struct sk_buff *skb)
671 static struct Qdisc_class_ops atm_class_ops = {
672 .graft = atm_tc_graft,
676 .change = atm_tc_change,
677 .delete = atm_tc_delete,
679 .tcf_chain = atm_tc_find_tcf,
680 .bind_tcf = atm_tc_bind_filter,
681 .unbind_tcf = atm_tc_put,
682 .dump = atm_tc_dump_class,
683 .dump_stats = atm_tc_dump_class_stats,
686 static struct Qdisc_ops atm_qdisc_ops = {
687 .cl_ops = &atm_class_ops,
689 .priv_size = sizeof(struct atm_qdisc_data),
690 .enqueue = atm_tc_enqueue,
691 .dequeue = atm_tc_dequeue,
692 .requeue = atm_tc_requeue,
695 .reset = atm_tc_reset,
696 .destroy = atm_tc_destroy,
698 .owner = THIS_MODULE,
701 static int __init atm_init(void)
703 return register_qdisc(&atm_qdisc_ops);
706 static void __exit atm_exit(void)
708 unregister_qdisc(&atm_qdisc_ops);
711 module_init(atm_init)
712 module_exit(atm_exit)
713 MODULE_LICENSE("GPL");