1 /* net/sched/sch_atm.c - ATM VC selection "queueing discipline" */
3 /* Written 1998-2000 by Werner Almesberger, EPFL ICA */
6 #include <linux/module.h>
7 #include <linux/init.h>
8 #include <linux/string.h>
9 #include <linux/errno.h>
10 #include <linux/skbuff.h>
11 #include <linux/interrupt.h>
12 #include <linux/atmdev.h>
13 #include <linux/atmclip.h>
14 #include <linux/netdevice.h>
15 #include <linux/rtnetlink.h>
16 #include <linux/file.h> /* for fput */
17 #include <net/netlink.h>
18 #include <net/pkt_sched.h>
22 extern struct socket *sockfd_lookup(int fd, int *err); /* @@@ fix this */
25 #define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
27 #define DPRINTK(format,args...)
31 #define D2PRINTK(format,args...) printk(KERN_DEBUG format,##args)
33 #define D2PRINTK(format,args...)
38 * The ATM queuing discipline provides a framework for invoking classifiers
39 * (aka "filters"), which in turn select classes of this queuing discipline.
40 * Each class maps the flow(s) it is handling to a given VC. Multiple classes
41 * may share the same VC.
43 * When creating a class, VCs are specified by passing the number of the open
44 * socket descriptor by which the calling process references the VC. The kernel
45 * keeps the VC open at least until all classes using it are removed.
47 * In this file, most functions are named atm_tc_* to avoid confusion with all
48 * the atm_* in net/atm. This naming convention differs from what's used in the
52 * - sometimes messes up the IP stack
53 * - any manipulations besides the few operations described in the README, are
54 * untested and likely to crash the system
55 * - should lock the flow while there is data in the queue (?)
59 #define PRIV(sch) qdisc_priv(sch)
60 #define VCC2FLOW(vcc) ((struct atm_flow_data *) ((vcc)->user_back))
63 struct atm_flow_data {
64 struct Qdisc *q; /* FIFO, TBF, etc. */
65 struct tcf_proto *filter_list;
66 struct atm_vcc *vcc; /* VCC; NULL if VCC is closed */
67 void (*old_pop)(struct atm_vcc *vcc,struct sk_buff *skb); /* chaining */
68 struct atm_qdisc_data *parent; /* parent qdisc */
69 struct socket *sock; /* for closing */
70 u32 classid; /* x:y type ID */
71 int ref; /* reference count */
72 struct gnet_stats_basic bstats;
73 struct gnet_stats_queue qstats;
74 spinlock_t *stats_lock;
75 struct atm_flow_data *next;
76 struct atm_flow_data *excess; /* flow for excess traffic;
77 NULL to set CLP instead */
79 unsigned char hdr[0]; /* header data; MUST BE LAST */
82 struct atm_qdisc_data {
83 struct atm_flow_data link; /* unclassified skbs go here */
84 struct atm_flow_data *flows; /* NB: "link" is also on this
86 struct tasklet_struct task; /* requeue tasklet */
90 /* ------------------------- Class/flow operations ------------------------- */
93 static int find_flow(struct atm_qdisc_data *qdisc,struct atm_flow_data *flow)
95 struct atm_flow_data *walk;
97 DPRINTK("find_flow(qdisc %p,flow %p)\n",qdisc,flow);
98 for (walk = qdisc->flows; walk; walk = walk->next)
99 if (walk == flow) return 1;
100 DPRINTK("find_flow: not found\n");
105 static __inline__ struct atm_flow_data *lookup_flow(struct Qdisc *sch,
108 struct atm_qdisc_data *p = PRIV(sch);
109 struct atm_flow_data *flow;
111 for (flow = p->flows; flow; flow = flow->next)
112 if (flow->classid == classid) break;
117 static int atm_tc_graft(struct Qdisc *sch,unsigned long arg,
118 struct Qdisc *new,struct Qdisc **old)
120 struct atm_qdisc_data *p = PRIV(sch);
121 struct atm_flow_data *flow = (struct atm_flow_data *) arg;
123 DPRINTK("atm_tc_graft(sch %p,[qdisc %p],flow %p,new %p,old %p)\n",sch,
125 if (!find_flow(p,flow)) return -EINVAL;
126 if (!new) new = &noop_qdisc;
127 *old = xchg(&flow->q,new);
128 if (*old) qdisc_reset(*old);
133 static struct Qdisc *atm_tc_leaf(struct Qdisc *sch,unsigned long cl)
135 struct atm_flow_data *flow = (struct atm_flow_data *) cl;
137 DPRINTK("atm_tc_leaf(sch %p,flow %p)\n",sch,flow);
138 return flow ? flow->q : NULL;
142 static unsigned long atm_tc_get(struct Qdisc *sch,u32 classid)
144 struct atm_qdisc_data *p __attribute__((unused)) = PRIV(sch);
145 struct atm_flow_data *flow;
147 DPRINTK("atm_tc_get(sch %p,[qdisc %p],classid %x)\n",sch,p,classid);
148 flow = lookup_flow(sch,classid);
149 if (flow) flow->ref++;
150 DPRINTK("atm_tc_get: flow %p\n",flow);
151 return (unsigned long) flow;
155 static unsigned long atm_tc_bind_filter(struct Qdisc *sch,
156 unsigned long parent, u32 classid)
158 return atm_tc_get(sch,classid);
162 * atm_tc_put handles all destructions, including the ones that are explicitly
163 * requested (atm_tc_destroy, etc.). The assumption here is that we never drop
164 * anything that still seems to be in use.
167 static void atm_tc_put(struct Qdisc *sch, unsigned long cl)
169 struct atm_qdisc_data *p = PRIV(sch);
170 struct atm_flow_data *flow = (struct atm_flow_data *) cl;
171 struct atm_flow_data **prev;
173 DPRINTK("atm_tc_put(sch %p,[qdisc %p],flow %p)\n",sch,p,flow);
174 if (--flow->ref) return;
175 DPRINTK("atm_tc_put: destroying\n");
176 for (prev = &p->flows; *prev; prev = &(*prev)->next)
177 if (*prev == flow) break;
179 printk(KERN_CRIT "atm_tc_put: class %p not found\n",flow);
183 DPRINTK("atm_tc_put: qdisc %p\n",flow->q);
184 qdisc_destroy(flow->q);
185 tcf_destroy_chain(flow->filter_list);
187 DPRINTK("atm_tc_put: f_count %d\n",
188 file_count(flow->sock->file));
189 flow->vcc->pop = flow->old_pop;
190 sockfd_put(flow->sock);
192 if (flow->excess) atm_tc_put(sch,(unsigned long) flow->excess);
193 if (flow != &p->link) kfree(flow);
195 * If flow == &p->link, the qdisc no longer works at this point and
196 * needs to be removed. (By the caller of atm_tc_put.)
201 static void sch_atm_pop(struct atm_vcc *vcc,struct sk_buff *skb)
203 struct atm_qdisc_data *p = VCC2FLOW(vcc)->parent;
205 D2PRINTK("sch_atm_pop(vcc %p,skb %p,[qdisc %p])\n",vcc,skb,p);
206 VCC2FLOW(vcc)->old_pop(vcc,skb);
207 tasklet_schedule(&p->task);
210 static const u8 llc_oui_ip[] = {
211 0xaa, /* DSAP: non-ISO */
212 0xaa, /* SSAP: non-ISO */
213 0x03, /* Ctrl: Unnumbered Information Command PDU */
214 0x00, /* OUI: EtherType */
216 0x08, 0x00 }; /* Ethertype IP (0800) */
218 static int atm_tc_change(struct Qdisc *sch, u32 classid, u32 parent,
219 struct rtattr **tca, unsigned long *arg)
221 struct atm_qdisc_data *p = PRIV(sch);
222 struct atm_flow_data *flow = (struct atm_flow_data *) *arg;
223 struct atm_flow_data *excess = NULL;
224 struct rtattr *opt = tca[TCA_OPTIONS-1];
225 struct rtattr *tb[TCA_ATM_MAX];
227 int fd,error,hdr_len;
230 DPRINTK("atm_tc_change(sch %p,[qdisc %p],classid %x,parent %x,"
231 "flow %p,opt %p)\n",sch,p,classid,parent,flow,opt);
233 * The concept of parents doesn't apply for this qdisc.
235 if (parent && parent != TC_H_ROOT && parent != sch->handle)
238 * ATM classes cannot be changed. In order to change properties of the
239 * ATM connection, that socket needs to be modified directly (via the
240 * native ATM API. In order to send a flow to a different VC, the old
241 * class needs to be removed and a new one added. (This may be changed
244 if (flow) return -EBUSY;
245 if (opt == NULL || rtattr_parse_nested(tb, TCA_ATM_MAX, opt))
247 if (!tb[TCA_ATM_FD-1] || RTA_PAYLOAD(tb[TCA_ATM_FD-1]) < sizeof(fd))
249 fd = *(int *) RTA_DATA(tb[TCA_ATM_FD-1]);
250 DPRINTK("atm_tc_change: fd %d\n",fd);
251 if (tb[TCA_ATM_HDR-1]) {
252 hdr_len = RTA_PAYLOAD(tb[TCA_ATM_HDR-1]);
253 hdr = RTA_DATA(tb[TCA_ATM_HDR-1]);
256 hdr_len = RFC1483LLC_LEN;
257 hdr = NULL; /* default LLC/SNAP for IP */
259 if (!tb[TCA_ATM_EXCESS-1]) excess = NULL;
261 if (RTA_PAYLOAD(tb[TCA_ATM_EXCESS-1]) != sizeof(u32))
263 excess = (struct atm_flow_data *) atm_tc_get(sch,
264 *(u32 *) RTA_DATA(tb[TCA_ATM_EXCESS-1]));
265 if (!excess) return -ENOENT;
267 DPRINTK("atm_tc_change: type %d, payload %d, hdr_len %d\n",
268 opt->rta_type,RTA_PAYLOAD(opt),hdr_len);
269 if (!(sock = sockfd_lookup(fd,&error))) return error; /* f_count++ */
270 DPRINTK("atm_tc_change: f_count %d\n",file_count(sock->file));
271 if (sock->ops->family != PF_ATMSVC && sock->ops->family != PF_ATMPVC) {
275 /* @@@ should check if the socket is really operational or we'll crash
278 if (TC_H_MAJ(classid ^ sch->handle)) {
279 DPRINTK("atm_tc_change: classid mismatch\n");
283 if (find_flow(p,flow)) {
292 for (i = 1; i < 0x8000; i++) {
293 classid = TC_H_MAKE(sch->handle,0x8000 | i);
294 if (!(cl = atm_tc_get(sch,classid))) break;
298 DPRINTK("atm_tc_change: new id %x\n",classid);
299 flow = kmalloc(sizeof(struct atm_flow_data)+hdr_len,GFP_KERNEL);
300 DPRINTK("atm_tc_change: flow %p\n",flow);
305 memset(flow,0,sizeof(*flow));
306 flow->filter_list = NULL;
307 if (!(flow->q = qdisc_create_dflt(sch->dev,&pfifo_qdisc_ops,classid)))
308 flow->q = &noop_qdisc;
309 DPRINTK("atm_tc_change: qdisc %p\n",flow->q);
311 flow->vcc = ATM_SD(sock); /* speedup */
312 flow->vcc->user_back = flow;
313 DPRINTK("atm_tc_change: vcc %p\n",flow->vcc);
314 flow->old_pop = flow->vcc->pop;
316 flow->vcc->pop = sch_atm_pop;
317 flow->classid = classid;
319 flow->excess = excess;
320 flow->next = p->link.next;
322 flow->hdr_len = hdr_len;
324 memcpy(flow->hdr,hdr,hdr_len);
326 memcpy(flow->hdr,llc_oui_ip,sizeof(llc_oui_ip));
327 *arg = (unsigned long) flow;
330 if (excess) atm_tc_put(sch,(unsigned long) excess);
336 static int atm_tc_delete(struct Qdisc *sch,unsigned long arg)
338 struct atm_qdisc_data *p = PRIV(sch);
339 struct atm_flow_data *flow = (struct atm_flow_data *) arg;
341 DPRINTK("atm_tc_delete(sch %p,[qdisc %p],flow %p)\n",sch,p,flow);
342 if (!find_flow(PRIV(sch),flow)) return -EINVAL;
343 if (flow->filter_list || flow == &p->link) return -EBUSY;
345 * Reference count must be 2: one for "keepalive" (set at class
346 * creation), and one for the reference held when calling delete.
349 printk(KERN_ERR "atm_tc_delete: flow->ref == %d\n",flow->ref);
352 if (flow->ref > 2) return -EBUSY; /* catch references via excess, etc.*/
358 static void atm_tc_walk(struct Qdisc *sch,struct qdisc_walker *walker)
360 struct atm_qdisc_data *p = PRIV(sch);
361 struct atm_flow_data *flow;
363 DPRINTK("atm_tc_walk(sch %p,[qdisc %p],walker %p)\n",sch,p,walker);
364 if (walker->stop) return;
365 for (flow = p->flows; flow; flow = flow->next) {
366 if (walker->count >= walker->skip)
367 if (walker->fn(sch,(unsigned long) flow,walker) < 0) {
376 static struct tcf_proto **atm_tc_find_tcf(struct Qdisc *sch,unsigned long cl)
378 struct atm_qdisc_data *p = PRIV(sch);
379 struct atm_flow_data *flow = (struct atm_flow_data *) cl;
381 DPRINTK("atm_tc_find_tcf(sch %p,[qdisc %p],flow %p)\n",sch,p,flow);
382 return flow ? &flow->filter_list : &p->link.filter_list;
386 /* --------------------------- Qdisc operations ---------------------------- */
389 static int atm_tc_enqueue(struct sk_buff *skb,struct Qdisc *sch)
391 struct atm_qdisc_data *p = PRIV(sch);
392 struct atm_flow_data *flow = NULL ; /* @@@ */
393 struct tcf_result res;
395 int ret = NET_XMIT_POLICED;
397 D2PRINTK("atm_tc_enqueue(skb %p,sch %p,[qdisc %p])\n",skb,sch,p);
398 result = TC_POLICE_OK; /* be nice to gcc */
399 if (TC_H_MAJ(skb->priority) != sch->handle ||
400 !(flow = (struct atm_flow_data *) atm_tc_get(sch,skb->priority)))
401 for (flow = p->flows; flow; flow = flow->next)
402 if (flow->filter_list) {
403 result = tc_classify(skb,flow->filter_list,
405 if (result < 0) continue;
406 flow = (struct atm_flow_data *) res.class;
407 if (!flow) flow = lookup_flow(sch,res.classid);
410 if (!flow) flow = &p->link;
413 ATM_SKB(skb)->atm_options = flow->vcc->atm_options;
414 /*@@@ looks good ... but it's not supposed to work :-)*/
415 #ifdef CONFIG_NET_CLS_POLICE
420 case TC_POLICE_RECLASSIFY:
421 if (flow->excess) flow = flow->excess;
423 ATM_SKB(skb)->atm_options |=
436 #ifdef CONFIG_NET_CLS_POLICE
437 result == TC_POLICE_SHOT ||
439 (ret = flow->q->enqueue(skb,flow->q)) != 0) {
441 if (flow) flow->qstats.drops++;
444 sch->bstats.bytes += skb->len;
445 sch->bstats.packets++;
446 flow->bstats.bytes += skb->len;
447 flow->bstats.packets++;
449 * Okay, this may seem weird. We pretend we've dropped the packet if
450 * it goes via ATM. The reason for this is that the outer qdisc
451 * expects to be able to q->dequeue the packet later on if we return
452 * success at this place. Also, sch->q.qdisc needs to reflect whether
453 * there is a packet egligible for dequeuing or not. Note that the
454 * statistics of the outer qdisc are necessarily wrong because of all
455 * this. There's currently no correct solution for this.
457 if (flow == &p->link) {
461 tasklet_schedule(&p->task);
462 return NET_XMIT_BYPASS;
467 * Dequeue packets and send them over ATM. Note that we quite deliberately
468 * avoid checking net_device's flow control here, simply because sch_atm
469 * uses its own channels, which have nothing to do with any CLIP/LANE/or
470 * non-ATM interfaces.
474 static void sch_atm_dequeue(unsigned long data)
476 struct Qdisc *sch = (struct Qdisc *) data;
477 struct atm_qdisc_data *p = PRIV(sch);
478 struct atm_flow_data *flow;
481 D2PRINTK("sch_atm_dequeue(sch %p,[qdisc %p])\n",sch,p);
482 for (flow = p->link.next; flow; flow = flow->next)
484 * If traffic is properly shaped, this won't generate nasty
485 * little bursts. Otherwise, it may ... (but that's okay)
487 while ((skb = flow->q->dequeue(flow->q))) {
488 if (!atm_may_send(flow->vcc,skb->truesize)) {
489 (void) flow->q->ops->requeue(skb,flow->q);
492 D2PRINTK("atm_tc_dequeue: sending on class %p\n",flow);
493 /* remove any LL header somebody else has attached */
494 skb_pull(skb, skb_network_offset(skb));
495 if (skb_headroom(skb) < flow->hdr_len) {
498 new = skb_realloc_headroom(skb,flow->hdr_len);
503 D2PRINTK("sch_atm_dequeue: ip %p, data %p\n",
504 skb_network_header(skb), skb->data);
505 ATM_SKB(skb)->vcc = flow->vcc;
506 memcpy(skb_push(skb,flow->hdr_len),flow->hdr,
508 atomic_add(skb->truesize,
509 &sk_atm(flow->vcc)->sk_wmem_alloc);
510 /* atm.atm_options are already set by atm_tc_enqueue */
511 (void) flow->vcc->send(flow->vcc,skb);
516 static struct sk_buff *atm_tc_dequeue(struct Qdisc *sch)
518 struct atm_qdisc_data *p = PRIV(sch);
521 D2PRINTK("atm_tc_dequeue(sch %p,[qdisc %p])\n",sch,p);
522 tasklet_schedule(&p->task);
523 skb = p->link.q->dequeue(p->link.q);
524 if (skb) sch->q.qlen--;
529 static int atm_tc_requeue(struct sk_buff *skb,struct Qdisc *sch)
531 struct atm_qdisc_data *p = PRIV(sch);
534 D2PRINTK("atm_tc_requeue(skb %p,sch %p,[qdisc %p])\n",skb,sch,p);
535 ret = p->link.q->ops->requeue(skb,p->link.q);
538 sch->qstats.requeues++;
541 p->link.qstats.drops++;
547 static unsigned int atm_tc_drop(struct Qdisc *sch)
549 struct atm_qdisc_data *p = PRIV(sch);
550 struct atm_flow_data *flow;
553 DPRINTK("atm_tc_drop(sch %p,[qdisc %p])\n",sch,p);
554 for (flow = p->flows; flow; flow = flow->next)
555 if (flow->q->ops->drop && (len = flow->q->ops->drop(flow->q)))
561 static int atm_tc_init(struct Qdisc *sch,struct rtattr *opt)
563 struct atm_qdisc_data *p = PRIV(sch);
565 DPRINTK("atm_tc_init(sch %p,[qdisc %p],opt %p)\n",sch,p,opt);
567 if(!(p->link.q = qdisc_create_dflt(sch->dev,&pfifo_qdisc_ops,
569 p->link.q = &noop_qdisc;
570 DPRINTK("atm_tc_init: link (%p) qdisc %p\n",&p->link,p->link.q);
571 p->link.filter_list = NULL;
574 p->link.classid = sch->handle;
577 tasklet_init(&p->task,sch_atm_dequeue,(unsigned long) sch);
582 static void atm_tc_reset(struct Qdisc *sch)
584 struct atm_qdisc_data *p = PRIV(sch);
585 struct atm_flow_data *flow;
587 DPRINTK("atm_tc_reset(sch %p,[qdisc %p])\n",sch,p);
588 for (flow = p->flows; flow; flow = flow->next) qdisc_reset(flow->q);
593 static void atm_tc_destroy(struct Qdisc *sch)
595 struct atm_qdisc_data *p = PRIV(sch);
596 struct atm_flow_data *flow;
598 DPRINTK("atm_tc_destroy(sch %p,[qdisc %p])\n",sch,p);
600 while ((flow = p->flows)) {
601 tcf_destroy_chain(flow->filter_list);
602 flow->filter_list = NULL;
604 printk(KERN_ERR "atm_destroy: %p->ref = %d\n",flow,
606 atm_tc_put(sch,(unsigned long) flow);
607 if (p->flows == flow) {
608 printk(KERN_ERR "atm_destroy: putting flow %p didn't "
610 p->flows = flow->next; /* brute force */
614 tasklet_kill(&p->task);
618 static int atm_tc_dump_class(struct Qdisc *sch, unsigned long cl,
619 struct sk_buff *skb, struct tcmsg *tcm)
621 struct atm_qdisc_data *p = PRIV(sch);
622 struct atm_flow_data *flow = (struct atm_flow_data *) cl;
623 unsigned char *b = skb_tail_pointer(skb);
626 DPRINTK("atm_tc_dump_class(sch %p,[qdisc %p],flow %p,skb %p,tcm %p)\n",
628 if (!find_flow(p,flow)) return -EINVAL;
629 tcm->tcm_handle = flow->classid;
630 tcm->tcm_info = flow->q->handle;
631 rta = (struct rtattr *) b;
632 RTA_PUT(skb,TCA_OPTIONS,0,NULL);
633 RTA_PUT(skb,TCA_ATM_HDR,flow->hdr_len,flow->hdr);
635 struct sockaddr_atmpvc pvc;
638 pvc.sap_family = AF_ATMPVC;
639 pvc.sap_addr.itf = flow->vcc->dev ? flow->vcc->dev->number : -1;
640 pvc.sap_addr.vpi = flow->vcc->vpi;
641 pvc.sap_addr.vci = flow->vcc->vci;
642 RTA_PUT(skb,TCA_ATM_ADDR,sizeof(pvc),&pvc);
643 state = ATM_VF2VS(flow->vcc->flags);
644 RTA_PUT(skb,TCA_ATM_STATE,sizeof(state),&state);
647 RTA_PUT(skb,TCA_ATM_EXCESS,sizeof(u32),&flow->classid);
651 RTA_PUT(skb,TCA_ATM_EXCESS,sizeof(zero),&zero);
653 rta->rta_len = skb_tail_pointer(skb) - b;
661 atm_tc_dump_class_stats(struct Qdisc *sch, unsigned long arg,
664 struct atm_flow_data *flow = (struct atm_flow_data *) arg;
666 flow->qstats.qlen = flow->q->q.qlen;
668 if (gnet_stats_copy_basic(d, &flow->bstats) < 0 ||
669 gnet_stats_copy_queue(d, &flow->qstats) < 0)
675 static int atm_tc_dump(struct Qdisc *sch, struct sk_buff *skb)
680 static struct Qdisc_class_ops atm_class_ops = {
681 .graft = atm_tc_graft,
685 .change = atm_tc_change,
686 .delete = atm_tc_delete,
688 .tcf_chain = atm_tc_find_tcf,
689 .bind_tcf = atm_tc_bind_filter,
690 .unbind_tcf = atm_tc_put,
691 .dump = atm_tc_dump_class,
692 .dump_stats = atm_tc_dump_class_stats,
695 static struct Qdisc_ops atm_qdisc_ops = {
697 .cl_ops = &atm_class_ops,
699 .priv_size = sizeof(struct atm_qdisc_data),
700 .enqueue = atm_tc_enqueue,
701 .dequeue = atm_tc_dequeue,
702 .requeue = atm_tc_requeue,
705 .reset = atm_tc_reset,
706 .destroy = atm_tc_destroy,
709 .owner = THIS_MODULE,
713 static int __init atm_init(void)
715 return register_qdisc(&atm_qdisc_ops);
718 static void __exit atm_exit(void)
720 unregister_qdisc(&atm_qdisc_ops);
723 module_init(atm_init)
724 module_exit(atm_exit)
725 MODULE_LICENSE("GPL");