1 /* net/sched/sch_atm.c - ATM VC selection "queueing discipline" */
3 /* Written 1998-2000 by Werner Almesberger, EPFL ICA */
6 #include <linux/config.h>
7 #include <linux/module.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/pkt_sched.h>
21 extern struct socket *sockfd_lookup(int fd, int *err); /* @@@ fix this */
22 #define sockfd_put(sock) fput((sock)->file) /* @@@ copied because it's
23 __inline__ in socket.c */
27 #define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
29 #define DPRINTK(format,args...)
33 #define D2PRINTK(format,args...) printk(KERN_DEBUG format,##args)
35 #define D2PRINTK(format,args...)
40 * The ATM queuing discipline provides a framework for invoking classifiers
41 * (aka "filters"), which in turn select classes of this queuing discipline.
42 * Each class maps the flow(s) it is handling to a given VC. Multiple classes
43 * may share the same VC.
45 * When creating a class, VCs are specified by passing the number of the open
46 * socket descriptor by which the calling process references the VC. The kernel
47 * keeps the VC open at least until all classes using it are removed.
49 * In this file, most functions are named atm_tc_* to avoid confusion with all
50 * the atm_* in net/atm. This naming convention differs from what's used in the
54 * - sometimes messes up the IP stack
55 * - any manipulations besides the few operations described in the README, are
56 * untested and likely to crash the system
57 * - should lock the flow while there is data in the queue (?)
61 #define PRIV(sch) ((struct atm_qdisc_data *) (sch)->data)
62 #define VCC2FLOW(vcc) ((struct atm_flow_data *) ((vcc)->user_back))
65 struct atm_flow_data {
66 struct Qdisc *q; /* FIFO, TBF, etc. */
67 struct tcf_proto *filter_list;
68 struct atm_vcc *vcc; /* VCC; NULL if VCC is closed */
69 void (*old_pop)(struct atm_vcc *vcc,struct sk_buff *skb); /* chaining */
70 struct atm_qdisc_data *parent; /* parent qdisc */
71 struct socket *sock; /* for closing */
72 u32 classid; /* x:y type ID */
73 int ref; /* reference count */
74 struct tc_stats stats;
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_flow_data *flow;
110 for (flow = PRIV(sch)->flows; flow; flow = flow->next)
111 if (flow->classid == classid) break;
116 static int atm_tc_graft(struct Qdisc *sch,unsigned long arg,
117 struct Qdisc *new,struct Qdisc **old)
119 struct atm_qdisc_data *p = PRIV(sch);
120 struct atm_flow_data *flow = (struct atm_flow_data *) arg;
122 DPRINTK("atm_tc_graft(sch %p,[qdisc %p],flow %p,new %p,old %p)\n",sch,
124 if (!find_flow(p,flow)) return -EINVAL;
125 if (!new) new = &noop_qdisc;
126 *old = xchg(&flow->q,new);
127 if (*old) qdisc_reset(*old);
132 static struct Qdisc *atm_tc_leaf(struct Qdisc *sch,unsigned long cl)
134 struct atm_flow_data *flow = (struct atm_flow_data *) cl;
136 DPRINTK("atm_tc_leaf(sch %p,flow %p)\n",sch,flow);
137 return flow ? flow->q : NULL;
141 static unsigned long atm_tc_get(struct Qdisc *sch,u32 classid)
143 struct atm_qdisc_data *p __attribute__((unused)) = PRIV(sch);
144 struct atm_flow_data *flow;
146 DPRINTK("atm_tc_get(sch %p,[qdisc %p],classid %x)\n",sch,p,classid);
147 flow = lookup_flow(sch,classid);
148 if (flow) flow->ref++;
149 DPRINTK("atm_tc_get: flow %p\n",flow);
150 return (unsigned long) flow;
154 static unsigned long atm_tc_bind_filter(struct Qdisc *sch,
155 unsigned long parent, u32 classid)
157 return atm_tc_get(sch,classid);
161 static void destroy_filters(struct atm_flow_data *flow)
163 struct tcf_proto *filter;
165 while ((filter = flow->filter_list)) {
166 DPRINTK("destroy_filters: destroying filter %p\n",filter);
167 flow->filter_list = filter->next;
174 * atm_tc_put handles all destructions, including the ones that are explicitly
175 * requested (atm_tc_destroy, etc.). The assumption here is that we never drop
176 * anything that still seems to be in use.
179 static void atm_tc_put(struct Qdisc *sch, unsigned long cl)
181 struct atm_qdisc_data *p = PRIV(sch);
182 struct atm_flow_data *flow = (struct atm_flow_data *) cl;
183 struct atm_flow_data **prev;
185 DPRINTK("atm_tc_put(sch %p,[qdisc %p],flow %p)\n",sch,p,flow);
186 if (--flow->ref) return;
187 DPRINTK("atm_tc_put: destroying\n");
188 for (prev = &p->flows; *prev; prev = &(*prev)->next)
189 if (*prev == flow) break;
191 printk(KERN_CRIT "atm_tc_put: class %p not found\n",flow);
195 DPRINTK("atm_tc_put: qdisc %p\n",flow->q);
196 qdisc_destroy(flow->q);
197 destroy_filters(flow);
199 DPRINTK("atm_tc_put: f_count %d\n",
200 file_count(flow->sock->file));
201 flow->vcc->pop = flow->old_pop;
202 sockfd_put(flow->sock);
204 if (flow->excess) atm_tc_put(sch,(unsigned long) flow->excess);
205 if (flow != &p->link) kfree(flow);
207 * If flow == &p->link, the qdisc no longer works at this point and
208 * needs to be removed. (By the caller of atm_tc_put.)
213 static void sch_atm_pop(struct atm_vcc *vcc,struct sk_buff *skb)
215 struct atm_qdisc_data *p = VCC2FLOW(vcc)->parent;
217 D2PRINTK("sch_atm_pop(vcc %p,skb %p,[qdisc %p])\n",vcc,skb,p);
218 VCC2FLOW(vcc)->old_pop(vcc,skb);
219 tasklet_schedule(&p->task);
223 static int atm_tc_change(struct Qdisc *sch, u32 classid, u32 parent,
224 struct rtattr **tca, unsigned long *arg)
226 struct atm_qdisc_data *p = PRIV(sch);
227 struct atm_flow_data *flow = (struct atm_flow_data *) *arg;
228 struct atm_flow_data *excess = NULL;
229 struct rtattr *opt = tca[TCA_OPTIONS-1];
230 struct rtattr *tb[TCA_ATM_MAX];
232 int fd,error,hdr_len;
235 DPRINTK("atm_tc_change(sch %p,[qdisc %p],classid %x,parent %x,"
236 "flow %p,opt %p)\n",sch,p,classid,parent,flow,opt);
238 * The concept of parents doesn't apply for this qdisc.
240 if (parent && parent != TC_H_ROOT && parent != sch->handle)
243 * ATM classes cannot be changed. In order to change properties of the
244 * ATM connection, that socket needs to be modified directly (via the
245 * native ATM API. In order to send a flow to a different VC, the old
246 * class needs to be removed and a new one added. (This may be changed
249 if (flow) return -EBUSY;
250 if (opt == NULL || rtattr_parse(tb,TCA_ATM_MAX,RTA_DATA(opt),
251 RTA_PAYLOAD(opt))) return -EINVAL;
252 if (!tb[TCA_ATM_FD-1] || RTA_PAYLOAD(tb[TCA_ATM_FD-1]) < sizeof(fd))
254 fd = *(int *) RTA_DATA(tb[TCA_ATM_FD-1]);
255 DPRINTK("atm_tc_change: fd %d\n",fd);
256 if (tb[TCA_ATM_HDR-1]) {
257 hdr_len = RTA_PAYLOAD(tb[TCA_ATM_HDR-1]);
258 hdr = RTA_DATA(tb[TCA_ATM_HDR-1]);
261 hdr_len = RFC1483LLC_LEN;
262 hdr = NULL; /* default LLC/SNAP for IP */
264 if (!tb[TCA_ATM_EXCESS-1]) excess = NULL;
266 if (RTA_PAYLOAD(tb[TCA_ATM_EXCESS-1]) != sizeof(u32))
268 excess = (struct atm_flow_data *) atm_tc_get(sch,
269 *(u32 *) RTA_DATA(tb[TCA_ATM_EXCESS-1]));
270 if (!excess) return -ENOENT;
272 DPRINTK("atm_tc_change: type %d, payload %d, hdr_len %d\n",
273 opt->rta_type,RTA_PAYLOAD(opt),hdr_len);
274 if (!(sock = sockfd_lookup(fd,&error))) return error; /* f_count++ */
275 DPRINTK("atm_tc_change: f_count %d\n",file_count(sock->file));
276 if (sock->ops->family != PF_ATMSVC && sock->ops->family != PF_ATMPVC) {
280 /* @@@ should check if the socket is really operational or we'll crash
283 if (TC_H_MAJ(classid ^ sch->handle)) {
284 DPRINTK("atm_tc_change: classid mismatch\n");
288 if (find_flow(p,flow)) {
297 for (i = 1; i < 0x8000; i++) {
298 classid = TC_H_MAKE(sch->handle,0x8000 | i);
299 if (!(cl = atm_tc_get(sch,classid))) break;
303 DPRINTK("atm_tc_change: new id %x\n",classid);
304 flow = kmalloc(sizeof(struct atm_flow_data)+hdr_len,GFP_KERNEL);
305 DPRINTK("atm_tc_change: flow %p\n",flow);
310 memset(flow,0,sizeof(*flow));
311 flow->filter_list = NULL;
312 if (!(flow->q = qdisc_create_dflt(sch->dev,&pfifo_qdisc_ops)))
313 flow->q = &noop_qdisc;
314 DPRINTK("atm_tc_change: qdisc %p\n",flow->q);
316 flow->vcc = ATM_SD(sock); /* speedup */
317 flow->vcc->user_back = flow;
318 DPRINTK("atm_tc_change: vcc %p\n",flow->vcc);
319 flow->old_pop = flow->vcc->pop;
321 flow->vcc->pop = sch_atm_pop;
322 flow->classid = classid;
324 flow->excess = excess;
325 flow->next = p->link.next;
327 flow->hdr_len = hdr_len;
328 if (hdr) memcpy(flow->hdr,hdr,hdr_len);
330 memcpy(flow->hdr,llc_oui,sizeof(llc_oui));
331 ((u16 *) flow->hdr)[3] = htons(ETH_P_IP);
333 *arg = (unsigned long) flow;
336 if (excess) atm_tc_put(sch,(unsigned long) excess);
342 static int atm_tc_delete(struct Qdisc *sch,unsigned long arg)
344 struct atm_qdisc_data *p = PRIV(sch);
345 struct atm_flow_data *flow = (struct atm_flow_data *) arg;
347 DPRINTK("atm_tc_delete(sch %p,[qdisc %p],flow %p)\n",sch,p,flow);
348 if (!find_flow(PRIV(sch),flow)) return -EINVAL;
349 if (flow->filter_list || flow == &p->link) return -EBUSY;
351 * Reference count must be 2: one for "keepalive" (set at class
352 * creation), and one for the reference held when calling delete.
355 printk(KERN_ERR "atm_tc_delete: flow->ref == %d\n",flow->ref);
358 if (flow->ref > 2) return -EBUSY; /* catch references via excess, etc.*/
364 static void atm_tc_walk(struct Qdisc *sch,struct qdisc_walker *walker)
366 struct atm_qdisc_data *p = PRIV(sch);
367 struct atm_flow_data *flow;
369 DPRINTK("atm_tc_walk(sch %p,[qdisc %p],walker %p)\n",sch,p,walker);
370 if (walker->stop) return;
371 for (flow = p->flows; flow; flow = flow->next) {
372 if (walker->count >= walker->skip)
373 if (walker->fn(sch,(unsigned long) flow,walker) < 0) {
382 static struct tcf_proto **atm_tc_find_tcf(struct Qdisc *sch,unsigned long cl)
384 struct atm_qdisc_data *p = PRIV(sch);
385 struct atm_flow_data *flow = (struct atm_flow_data *) cl;
387 DPRINTK("atm_tc_find_tcf(sch %p,[qdisc %p],flow %p)\n",sch,p,flow);
388 return flow ? &flow->filter_list : &p->link.filter_list;
392 /* --------------------------- Qdisc operations ---------------------------- */
395 static int atm_tc_enqueue(struct sk_buff *skb,struct Qdisc *sch)
397 struct atm_qdisc_data *p = PRIV(sch);
398 struct atm_flow_data *flow = NULL ; /* @@@ */
399 struct tcf_result res;
401 int ret = NET_XMIT_POLICED;
403 D2PRINTK("atm_tc_enqueue(skb %p,sch %p,[qdisc %p])\n",skb,sch,p);
404 result = TC_POLICE_OK; /* be nice to gcc */
405 if (TC_H_MAJ(skb->priority) != sch->handle ||
406 !(flow = (struct atm_flow_data *) atm_tc_get(sch,skb->priority)))
407 for (flow = p->flows; flow; flow = flow->next)
408 if (flow->filter_list) {
409 result = tc_classify(skb,flow->filter_list,
411 if (result < 0) continue;
412 flow = (struct atm_flow_data *) res.class;
413 if (!flow) flow = lookup_flow(sch,res.classid);
416 if (!flow) flow = &p->link;
419 ATM_SKB(skb)->atm_options = flow->vcc->atm_options;
420 /*@@@ looks good ... but it's not supposed to work :-)*/
421 #ifdef CONFIG_NET_CLS_POLICE
426 case TC_POLICE_RECLASSIFY:
427 if (flow->excess) flow = flow->excess;
429 ATM_SKB(skb)->atm_options |=
442 #ifdef CONFIG_NET_CLS_POLICE
443 result == TC_POLICE_SHOT ||
445 (ret = flow->q->enqueue(skb,flow->q)) != 0) {
447 if (flow) flow->stats.drops++;
450 sch->stats.bytes += skb->len;
451 sch->stats.packets++;
452 flow->stats.bytes += skb->len;
453 flow->stats.packets++;
455 * Okay, this may seem weird. We pretend we've dropped the packet if
456 * it goes via ATM. The reason for this is that the outer qdisc
457 * expects to be able to q->dequeue the packet later on if we return
458 * success at this place. Also, sch->q.qdisc needs to reflect whether
459 * there is a packet egligible for dequeuing or not. Note that the
460 * statistics of the outer qdisc are necessarily wrong because of all
461 * this. There's currently no correct solution for this.
463 if (flow == &p->link) {
467 tasklet_schedule(&p->task);
468 return NET_XMIT_BYPASS;
473 * Dequeue packets and send them over ATM. Note that we quite deliberately
474 * avoid checking net_device's flow control here, simply because sch_atm
475 * uses its own channels, which have nothing to do with any CLIP/LANE/or
476 * non-ATM interfaces.
480 static void sch_atm_dequeue(unsigned long data)
482 struct Qdisc *sch = (struct Qdisc *) data;
483 struct atm_qdisc_data *p = PRIV(sch);
484 struct atm_flow_data *flow;
487 D2PRINTK("sch_atm_dequeue(sch %p,[qdisc %p])\n",sch,p);
488 for (flow = p->link.next; flow; flow = flow->next)
490 * If traffic is properly shaped, this won't generate nasty
491 * little bursts. Otherwise, it may ... (but that's okay)
493 while ((skb = flow->q->dequeue(flow->q))) {
494 if (!atm_may_send(flow->vcc,skb->truesize)) {
495 (void) flow->q->ops->requeue(skb,flow->q);
498 D2PRINTK("atm_tc_deqeueue: sending on class %p\n",flow);
499 /* remove any LL header somebody else has attached */
500 skb_pull(skb,(char *) skb->nh.iph-(char *) skb->data);
501 if (skb_headroom(skb) < flow->hdr_len) {
504 new = skb_realloc_headroom(skb,flow->hdr_len);
509 D2PRINTK("sch_atm_dequeue: ip %p, data %p\n",
510 skb->nh.iph,skb->data);
511 ATM_SKB(skb)->vcc = flow->vcc;
512 memcpy(skb_push(skb,flow->hdr_len),flow->hdr,
514 atomic_add(skb->truesize,&flow->vcc->sk->wmem_alloc);
515 /* atm.atm_options are already set by atm_tc_enqueue */
516 (void) flow->vcc->send(flow->vcc,skb);
521 static struct sk_buff *atm_tc_dequeue(struct Qdisc *sch)
523 struct atm_qdisc_data *p = PRIV(sch);
526 D2PRINTK("atm_tc_dequeue(sch %p,[qdisc %p])\n",sch,p);
527 tasklet_schedule(&p->task);
528 skb = p->link.q->dequeue(p->link.q);
529 if (skb) sch->q.qlen--;
534 static int atm_tc_requeue(struct sk_buff *skb,struct Qdisc *sch)
536 struct atm_qdisc_data *p = PRIV(sch);
539 D2PRINTK("atm_tc_requeue(skb %p,sch %p,[qdisc %p])\n",skb,sch,p);
540 ret = p->link.q->ops->requeue(skb,p->link.q);
541 if (!ret) sch->q.qlen++;
544 p->link.stats.drops++;
550 static unsigned int atm_tc_drop(struct Qdisc *sch)
552 struct atm_qdisc_data *p = PRIV(sch);
553 struct atm_flow_data *flow;
556 DPRINTK("atm_tc_drop(sch %p,[qdisc %p])\n",sch,p);
557 for (flow = p->flows; flow; flow = flow->next)
558 if (flow->q->ops->drop && (len = flow->q->ops->drop(flow->q)))
564 static int atm_tc_init(struct Qdisc *sch,struct rtattr *opt)
566 struct atm_qdisc_data *p = PRIV(sch);
568 DPRINTK("atm_tc_init(sch %p,[qdisc %p],opt %p)\n",sch,p,opt);
570 if(!(p->link.q = qdisc_create_dflt(sch->dev,&pfifo_qdisc_ops)))
571 p->link.q = &noop_qdisc;
572 DPRINTK("atm_tc_init: link (%p) qdisc %p\n",&p->link,p->link.q);
573 p->link.filter_list = NULL;
576 p->link.classid = sch->handle;
579 tasklet_init(&p->task,sch_atm_dequeue,(unsigned long) sch);
585 static void atm_tc_reset(struct Qdisc *sch)
587 struct atm_qdisc_data *p = PRIV(sch);
588 struct atm_flow_data *flow;
590 DPRINTK("atm_tc_reset(sch %p,[qdisc %p])\n",sch,p);
591 for (flow = p->flows; flow; flow = flow->next) qdisc_reset(flow->q);
596 static void atm_tc_destroy(struct Qdisc *sch)
598 struct atm_qdisc_data *p = PRIV(sch);
599 struct atm_flow_data *flow;
601 DPRINTK("atm_tc_destroy(sch %p,[qdisc %p])\n",sch,p);
603 while ((flow = p->flows)) {
604 destroy_filters(flow);
606 printk(KERN_ERR "atm_destroy: %p->ref = %d\n",flow,
608 atm_tc_put(sch,(unsigned long) flow);
609 if (p->flows == flow) {
610 printk(KERN_ERR "atm_destroy: putting flow %p didn't "
612 p->flows = flow->next; /* brute force */
616 tasklet_kill(&p->task);
621 static int atm_tc_dump_class(struct Qdisc *sch, unsigned long cl,
622 struct sk_buff *skb, struct tcmsg *tcm)
624 struct atm_qdisc_data *p = PRIV(sch);
625 struct atm_flow_data *flow = (struct atm_flow_data *) cl;
626 unsigned char *b = skb->tail;
629 DPRINTK("atm_tc_dump_class(sch %p,[qdisc %p],flow %p,skb %p,tcm %p)\n",
631 if (!find_flow(p,flow)) return -EINVAL;
632 tcm->tcm_handle = flow->classid;
633 rta = (struct rtattr *) b;
634 RTA_PUT(skb,TCA_OPTIONS,0,NULL);
635 RTA_PUT(skb,TCA_ATM_HDR,flow->hdr_len,flow->hdr);
637 struct sockaddr_atmpvc pvc;
640 pvc.sap_family = AF_ATMPVC;
641 pvc.sap_addr.itf = flow->vcc->dev ? flow->vcc->dev->number : -1;
642 pvc.sap_addr.vpi = flow->vcc->vpi;
643 pvc.sap_addr.vci = flow->vcc->vci;
644 RTA_PUT(skb,TCA_ATM_ADDR,sizeof(pvc),&pvc);
645 state = ATM_VF2VS(flow->vcc->flags);
646 RTA_PUT(skb,TCA_ATM_STATE,sizeof(state),&state);
649 RTA_PUT(skb,TCA_ATM_EXCESS,sizeof(u32),&flow->classid);
653 RTA_PUT(skb,TCA_ATM_EXCESS,sizeof(zero),&zero);
655 rta->rta_len = skb->tail-b;
659 skb_trim(skb,b-skb->data);
663 static int atm_tc_dump(struct Qdisc *sch, struct sk_buff *skb)
668 static struct Qdisc_class_ops atm_class_ops =
670 atm_tc_graft, /* graft */
671 atm_tc_leaf, /* leaf */
672 atm_tc_get, /* get */
673 atm_tc_put, /* put */
674 atm_tc_change, /* change */
675 atm_tc_delete, /* delete */
676 atm_tc_walk, /* walk */
678 atm_tc_find_tcf, /* tcf_chain */
679 atm_tc_bind_filter, /* bind_tcf */
680 atm_tc_put, /* unbind_tcf */
682 atm_tc_dump_class, /* dump */
685 struct Qdisc_ops atm_qdisc_ops =
688 &atm_class_ops, /* cl_ops */
690 sizeof(struct atm_qdisc_data),
692 atm_tc_enqueue, /* enqueue */
693 atm_tc_dequeue, /* dequeue */
694 atm_tc_requeue, /* requeue */
695 atm_tc_drop, /* drop */
697 atm_tc_init, /* init */
698 atm_tc_reset, /* reset */
699 atm_tc_destroy, /* destroy */
702 atm_tc_dump /* dump */
707 int init_module(void)
709 return register_qdisc(&atm_qdisc_ops);
713 void cleanup_module(void)
715 unregister_qdisc(&atm_qdisc_ops);