Import 2.3.18pre1
[davej-history.git] / net / sched / sch_atm.c
blob79901319da1dc0ae99d6709d9d88e39cf8f1559b
1 /* net/sched/sch_atm.c - ATM VC selection "queueing discipline" */
3 /* Written 1998,1999 by Werner Almesberger, EPFL ICA */
6 #include <linux/config.h>
7 #include <linux/module.h>
8 #include <linux/skbuff.h>
9 #include <linux/atmdev.h>
10 #include <linux/atmclip.h>
11 #include <linux/netdevice.h>
12 #include <linux/rtnetlink.h>
13 #include <linux/file.h> /* for fput */
14 #include <net/pkt_sched.h>
15 #include <net/sock.h>
18 extern struct socket *sockfd_lookup(int fd, int *err); /* @@@ fix this */
19 #define sockfd_put(sock) fput((sock)->file) /* @@@ copied because it's
20 __inline__ in socket.c */
23 #if 1 /* control */
24 #define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
25 #else
26 #define DPRINTK(format,args...)
27 #endif
29 #if 0 /* data */
30 #define D2PRINTK(format,args...) printk(KERN_DEBUG format,##args)
31 #else
32 #define D2PRINTK(format,args...)
33 #endif
37 * The ATM queuing discipline provides a framework for invoking classifiers
38 * (aka "filters"), which in turn select classes of this queuing discipline.
39 * Each class maps the flow(s) it is handling to a given VC. Multiple classes
40 * may share the same VC.
42 * When creating a class, VCs are specified by passing the number of the open
43 * socket descriptor by which the calling process references the VC. The kernel
44 * keeps the VC open at least until all classes using it are removed.
46 * In this file, most functions are named atm_tc_* to avoid confusion with all
47 * the atm_* in net/atm. This naming convention differs from what's used in the
48 * rest of net/sched.
50 * Known bugs:
51 * - sometimes messes up the IP stack
52 * - any manipulations besides the few operations described in the README, are
53 * untested and likely to crash the system
54 * - should lock the flow while there is data in the queue (?)
58 #define PRIV(sch) ((struct atm_qdisc_data *) (sch)->data)
61 struct atm_flow_data {
62 struct Qdisc *q; /* FIFO, TBF, etc. */
63 struct tcf_proto *filter_list;
64 struct atm_vcc *vcc; /* VCC; NULL if VCC is closed */
65 struct socket *sock; /* for closing */
66 u32 classid; /* x:y type ID */
67 int ref; /* reference count */
68 struct tc_stats stats;
69 struct atm_flow_data *next;
70 struct atm_flow_data *excess; /* flow for excess traffic;
71 NULL to set CLP instead */
72 int hdr_len;
73 unsigned char hdr[0]; /* header data; MUST BE LAST */
76 struct atm_qdisc_data {
77 struct atm_flow_data link; /* unclassified skbs go here */
78 struct atm_flow_data *flows; /* NB: "link" is also on this
79 list */
83 /* ------------------------- Class/flow operations ------------------------- */
86 static int find_flow(struct atm_qdisc_data *qdisc,struct atm_flow_data *flow)
88 struct atm_flow_data *walk;
90 DPRINTK("find_flow(qdisc %p,flow %p)\n",qdisc,flow);
91 for (walk = qdisc->flows; walk; walk = walk->next)
92 if (walk == flow) return 1;
93 DPRINTK("find_flow: not found\n");
94 return 0;
98 static __inline__ struct atm_flow_data *lookup_flow(struct Qdisc *sch,
99 u32 classid)
101 struct atm_flow_data *flow;
103 for (flow = PRIV(sch)->flows; flow; flow = flow->next)
104 if (flow->classid == classid) break;
105 return flow;
109 static int atm_tc_graft(struct Qdisc *sch,unsigned long arg,
110 struct Qdisc *new,struct Qdisc **old)
112 struct atm_qdisc_data *p = PRIV(sch);
113 struct atm_flow_data *flow = (struct atm_flow_data *) arg;
115 DPRINTK("atm_tc_graft(sch %p,[qdisc %p],flow %p,new %p,old %p)\n",sch,
116 p,flow,new,old);
117 if (!find_flow(p,flow)) return -EINVAL;
118 if (!new) new = &noop_qdisc;
119 *old = xchg(&flow->q,new);
120 if (*old) qdisc_reset(*old);
121 return 0;
125 static struct Qdisc *atm_tc_leaf(struct Qdisc *sch,unsigned long cl)
127 struct atm_flow_data *flow = (struct atm_flow_data *) cl;
129 DPRINTK("atm_tc_leaf(sch %p,flow %p)\n",sch,flow);
130 return flow ? flow->q : NULL;
134 static unsigned long atm_tc_get(struct Qdisc *sch,u32 classid)
136 struct atm_qdisc_data *p = PRIV(sch);
137 struct atm_flow_data *flow;
139 DPRINTK("atm_tc_get(sch %p,[qdisc %p],classid %x)\n",sch,p,classid);
140 flow = lookup_flow(sch,classid);
141 if (flow) flow->ref++;
142 DPRINTK("atm_tc_get: flow %p\n",flow);
143 return (unsigned long) flow;
147 static unsigned long atm_tc_bind_filter(struct Qdisc *sch,
148 unsigned long parent, u32 classid)
150 return atm_tc_get(sch,classid);
155 * atm_tc_put handles all destructions, including the ones that are explicitly
156 * requested (atm_tc_destroy, etc.). The assumption here is that we never drop
157 * anything that still seems to be in use.
160 static void atm_tc_put(struct Qdisc *sch, unsigned long cl)
162 struct atm_qdisc_data *p = PRIV(sch);
163 struct atm_flow_data *flow = (struct atm_flow_data *) cl;
164 struct atm_flow_data **prev;
165 struct tcf_proto *filter;
167 DPRINTK("atm_tc_put(sch %p,[qdisc %p],flow %p)\n",sch,p,flow);
168 if (--flow->ref) return;
169 DPRINTK("atm_tc_put: destroying\n");
170 for (prev = &p->flows; *prev; prev = &(*prev)->next)
171 if (*prev == flow) break;
172 if (!*prev) {
173 printk(KERN_CRIT "atm_tc_put: class %p not found\n",flow);
174 return;
176 *prev = flow->next;
177 DPRINTK("atm_tc_put: qdisc %p\n",flow->q);
178 qdisc_destroy(flow->q);
179 while ((filter = flow->filter_list)) {
180 DPRINTK("atm_tc_put: destroying filter %p\n",filter);
181 flow->filter_list = filter->next;
182 DPRINTK("atm_tc_put: filter %p\n",filter);
183 filter->ops->destroy(filter);
185 if (flow->sock) {
186 DPRINTK("atm_tc_put: f_count %d\n",file_count(flow->sock->file));
187 sockfd_put(flow->sock);
189 if (flow->excess) atm_tc_put(sch,(unsigned long) flow->excess);
190 if (flow != &p->link) kfree(flow);
192 * If flow == &p->link, the qdisc no longer works at this point and
193 * needs to be removed. (By the caller of atm_tc_put.)
198 static int atm_tc_change(struct Qdisc *sch, u32 classid, u32 parent,
199 struct rtattr **tca, unsigned long *arg)
201 struct atm_qdisc_data *p = PRIV(sch);
202 struct atm_flow_data *flow = (struct atm_flow_data *) *arg;
203 struct atm_flow_data *excess = NULL;
204 struct rtattr *opt = tca[TCA_OPTIONS-1];
205 struct rtattr *tb[TCA_ATM_MAX];
206 struct socket *sock;
207 int fd,error,hdr_len;
208 void *hdr;
210 DPRINTK("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)
216 return -EINVAL;
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
222 * later.)
224 if (flow) return -EBUSY;
225 if (opt == NULL || rtattr_parse(tb,TCA_ATM_MAX,RTA_DATA(opt),
226 RTA_PAYLOAD(opt))) return -EINVAL;
227 if (!tb[TCA_ATM_FD-1] || RTA_PAYLOAD(tb[TCA_ATM_FD-1]) < sizeof(fd))
228 return -EINVAL;
229 fd = *(int *) RTA_DATA(tb[TCA_ATM_FD-1]);
230 DPRINTK("atm_tc_change: fd %d\n",fd);
231 if (tb[TCA_ATM_HDR-1]) {
232 hdr_len = RTA_PAYLOAD(tb[TCA_ATM_HDR-1]);
233 hdr = RTA_DATA(tb[TCA_ATM_HDR-1]);
235 else {
236 hdr_len = RFC1483LLC_LEN;
237 hdr = NULL; /* default LLC/SNAP for IP */
239 if (!tb[TCA_ATM_EXCESS-1]) excess = NULL;
240 else {
241 if (RTA_PAYLOAD(tb[TCA_ATM_EXCESS-1]) != sizeof(u32))
242 return -EINVAL;
243 excess = (struct atm_flow_data *) atm_tc_get(sch,
244 *(u32 *) RTA_DATA(tb[TCA_ATM_EXCESS-1]));
245 if (!excess) return -ENOENT;
247 DPRINTK("atm_tc_change: type %d, payload %d, hdr_len %d\n",
248 opt->rta_type,RTA_PAYLOAD(opt),hdr_len);
249 if (!(sock = sockfd_lookup(fd,&error))) return error; /* f_count++ */
250 DPRINTK("atm_tc_change: f_count %d\n",file_count(sock->file));
251 if (sock->ops->family != PF_ATMSVC && sock->ops->family != PF_ATMPVC) {
252 error = -EPROTOTYPE;
253 goto err_out;
255 /* @@@ should check if the socket is really operational or we'll crash
256 on vcc->dev->ops->send */
257 if (classid) {
258 if (TC_H_MAJ(classid ^ sch->handle)) {
259 DPRINTK("atm_tc_change: classid mismatch\n");
260 error = -EINVAL;
261 goto err_out;
263 if (find_flow(p,flow)) {
264 error = -EEXIST;
265 goto err_out;
268 else {
269 int i;
270 unsigned long cl;
272 for (i = 1; i < 0x8000; i++) {
273 classid = TC_H_MAKE(sch->handle,0x8000 | i);
274 if (!(cl = atm_tc_get(sch,classid))) break;
275 atm_tc_put(sch,cl);
278 DPRINTK("atm_tc_change: new id %x\n",classid);
279 flow = kmalloc(sizeof(struct atm_flow_data)+hdr_len,GFP_KERNEL);
280 DPRINTK("atm_tc_change: flow %p\n",flow);
281 if (!flow) {
282 error = -ENOBUFS;
283 goto err_out;
285 memset(flow,0,sizeof(*flow));
286 flow->filter_list = NULL;
287 if (!(flow->q = qdisc_create_dflt(sch->dev,&pfifo_qdisc_ops)))
288 flow->q = &noop_qdisc;
289 DPRINTK("atm_tc_change: qdisc %p\n",flow->q);
290 flow->sock = sock;
291 flow->vcc = ATM_SD(sock); /* speedup */
292 DPRINTK("atm_tc_change: vcc %p\n",flow->vcc);
293 flow->classid = classid;
294 flow->ref = 1;
295 flow->excess = excess;
296 flow->next = p->link.next;
297 p->link.next = flow;
298 flow->hdr_len = hdr_len;
299 if (hdr) memcpy(flow->hdr,hdr,hdr_len);
300 else {
301 memcpy(flow->hdr,llc_oui,sizeof(llc_oui));
302 ((u16 *) flow->hdr)[3] = htons(ETH_P_IP);
304 *arg = (unsigned long) flow;
305 return 0;
306 err_out:
307 if (excess) atm_tc_put(sch,(unsigned long) excess);
308 sockfd_put(sock);
309 return error;
313 static int atm_tc_delete(struct Qdisc *sch,unsigned long arg)
315 struct atm_qdisc_data *p = PRIV(sch);
316 struct atm_flow_data *flow = (struct atm_flow_data *) arg;
318 DPRINTK("atm_tc_delete(sch %p,[qdisc %p],flow %p)\n",sch,p,flow);
319 if (!find_flow(PRIV(sch),flow)) return -EINVAL;
320 if (flow->filter_list || flow == &p->link) return -EBUSY;
322 * Reference count must be 2: one for "keepalive" (set at class
323 * creation), and one for the reference held when calling delete.
325 if (flow->ref < 2) {
326 printk(KERN_ERR "atm_tc_delete: flow->ref == %d\n",flow->ref);
327 return -EINVAL;
329 if (flow->ref > 2) return -EBUSY; /* catch references via excess, etc.*/
330 atm_tc_put(sch,arg);
331 return 0;
335 static void atm_tc_walk(struct Qdisc *sch,struct qdisc_walker *walker)
337 struct atm_qdisc_data *p = PRIV(sch);
338 struct atm_flow_data *flow;
340 DPRINTK("atm_tc_walk(sch %p,[qdisc %p],walker %p)\n",sch,p,walker);
341 if (walker->stop) return;
342 for (flow = p->flows; flow; flow = flow->next) {
343 if (walker->count >= walker->skip)
344 if (walker->fn(sch,(unsigned long) flow,walker) < 0) {
345 walker->stop = 1;
346 break;
348 walker->count++;
353 static struct tcf_proto **atm_tc_find_tcf(struct Qdisc *sch,unsigned long cl)
355 struct atm_qdisc_data *p = PRIV(sch);
356 struct atm_flow_data *flow = (struct atm_flow_data *) cl;
358 DPRINTK("atm_tc_find_tcf(sch %p,[qdisc %p],flow %p)\n",sch,p,flow);
359 return flow ? &flow->filter_list : &p->link.filter_list;
363 /* --------------------------- Qdisc operations ---------------------------- */
366 static int atm_tc_enqueue(struct sk_buff *skb,struct Qdisc *sch)
368 struct atm_qdisc_data *p = PRIV(sch);
369 struct atm_flow_data *flow = NULL ; /* @@@ */
370 struct tcf_result res;
371 int result;
372 int ret = NET_XMIT_POLICED;
374 D2PRINTK("atm_tc_enqueue(skb %p,sch %p,[qdisc %p])\n",skb,sch,p);
375 result = TC_POLICE_OK; /* be nice to gcc */
376 if (TC_H_MAJ(skb->priority) != sch->handle ||
377 !(flow = (struct atm_flow_data *) atm_tc_get(sch,skb->priority)))
378 for (flow = p->flows; flow; flow = flow->next)
379 if (flow->filter_list) {
380 result = tc_classify(skb,flow->filter_list,
381 &res);
382 if (result < 0) continue;
383 flow = (struct atm_flow_data *) res.class;
384 if (!flow) flow = lookup_flow(sch,res.classid);
385 break;
387 if (!flow) flow = &p->link;
388 else {
389 if (flow->vcc)
390 ATM_SKB(skb)->atm_options = flow->vcc->atm_options;
391 /*@@@ looks good ... but it's not supposed to work :-)*/
392 #ifdef CONFIG_NET_CLS_POLICE
393 switch (result) {
394 case TC_POLICE_SHOT:
395 kfree_skb(skb);
396 break;
397 case TC_POLICE_RECLASSIFY:
398 if (flow->excess) flow = flow->excess;
399 else {
400 ATM_SKB(skb)->atm_options |=
401 ATM_ATMOPT_CLP;
402 break;
404 /* fall through */
405 case TC_POLICE_OK:
406 /* fall through */
407 default:
408 break;
410 #endif
412 if (
413 #ifdef CONFIG_NET_CLS_POLICE
414 result == TC_POLICE_SHOT ||
415 #endif
416 (ret = flow->q->enqueue(skb,flow->q)) != 0) {
417 sch->stats.drops++;
418 if (flow) flow->stats.drops++;
419 return ret;
421 sch->stats.bytes += skb->len;
422 sch->stats.packets++;
423 flow->stats.bytes += skb->len;
424 flow->stats.packets++;
425 sch->q.qlen++;
426 return 0;
430 static struct sk_buff *atm_tc_dequeue(struct Qdisc *sch)
432 struct atm_qdisc_data *p = PRIV(sch);
433 struct atm_flow_data *flow;
434 struct sk_buff *skb;
436 D2PRINTK("atm_tc_dequeue(sch %p,[qdisc %p])\n",sch,p);
437 for (flow = p->link.next; flow; flow = flow->next)
439 * If traffic is properly shaped, this won't generate nasty
440 * little bursts. Otherwise, it may ... @@@
442 while ((skb = flow->q->dequeue(flow->q))) {
443 sch->q.qlen--;
444 D2PRINTK("atm_tc_deqeueue: sending on class %p\n",flow);
445 /* remove any LL header somebody else has attached */
446 skb_pull(skb,(char *) skb->nh.iph-(char *) skb->data);
447 if (skb_headroom(skb) < flow->hdr_len) {
448 struct sk_buff *new;
450 new = skb_realloc_headroom(skb,flow->hdr_len);
451 dev_kfree_skb(skb);
452 if (!new) continue;
453 skb = new;
455 D2PRINTK("atm_tc_dequeue: ip %p, data %p\n",
456 skb->nh.iph,skb->data);
457 ATM_SKB(skb)->vcc = flow->vcc;
458 memcpy(skb_push(skb,flow->hdr_len),flow->hdr,
459 flow->hdr_len);
460 atomic_add(skb->truesize,&flow->vcc->tx_inuse);
461 ATM_SKB(skb)->iovcnt = 0;
462 /* atm.atm_options are already set by atm_tc_enqueue */
463 (void) flow->vcc->dev->ops->send(flow->vcc,skb);
465 skb = p->link.q->dequeue(p->link.q);
466 if (skb) sch->q.qlen--;
467 return skb;
471 static int atm_tc_drop(struct Qdisc *sch)
473 struct atm_qdisc_data *p = PRIV(sch);
474 struct atm_flow_data *flow;
476 DPRINTK("atm_tc_drop(sch %p,[qdisc %p])\n",sch,p);
477 for (flow = p->flows; flow; flow = flow->next)
478 if (flow->q->ops->drop && flow->q->ops->drop(flow->q))
479 return 1;
480 return 0;
484 static int atm_tc_init(struct Qdisc *sch,struct rtattr *opt)
486 struct atm_qdisc_data *p = PRIV(sch);
488 DPRINTK("atm_tc_init(sch %p,[qdisc %p],opt %p)\n",sch,p,opt);
489 memset(p,0,sizeof(*p));
490 p->flows = &p->link;
491 if(!(p->link.q = qdisc_create_dflt(sch->dev,&pfifo_qdisc_ops)))
492 p->link.q = &noop_qdisc;
493 DPRINTK("atm_tc_init: link (%p) qdisc %p\n",&p->link,p->link.q);
494 p->link.filter_list = NULL;
495 p->link.vcc = NULL;
496 p->link.sock = NULL;
497 p->link.classid = sch->handle;
498 p->link.ref = 1;
499 p->link.next = NULL;
500 MOD_INC_USE_COUNT;
501 return 0;
505 static void atm_tc_reset(struct Qdisc *sch)
507 struct atm_qdisc_data *p = PRIV(sch);
508 struct atm_flow_data *flow;
510 DPRINTK("atm_tc_reset(sch %p,[qdisc %p])\n",sch,p);
511 for (flow = p->flows; flow; flow = flow->next) qdisc_reset(flow->q);
512 sch->q.qlen = 0;
516 static void atm_tc_destroy(struct Qdisc *sch)
518 struct atm_qdisc_data *p = PRIV(sch);
519 struct atm_flow_data *flow;
521 DPRINTK("atm_tc_destroy(sch %p,[qdisc %p])\n",sch,p);
522 /* races ? */
523 while ((flow = p->flows)) {
524 if (flow->ref > 1)
525 printk(KERN_ERR "atm_destroy: %p->ref = %d\n",flow,
526 flow->ref);
527 atm_tc_put(sch,(unsigned long) flow);
528 if (p->flows == flow) {
529 printk(KERN_ERR "atm_destroy: putting flow %p didn't "
530 "kill it\n",flow);
531 p->flows = flow->next; /* brute force */
532 break;
535 MOD_DEC_USE_COUNT;
539 #ifdef CONFIG_RTNETLINK
541 static int atm_tc_dump_class(struct Qdisc *sch, unsigned long cl,
542 struct sk_buff *skb, struct tcmsg *tcm)
544 struct atm_qdisc_data *p = PRIV(sch);
545 struct atm_flow_data *flow = (struct atm_flow_data *) cl;
546 unsigned char *b = skb->tail;
547 struct rtattr *rta;
549 DPRINTK("atm_tc_dump_class(sch %p,[qdisc %p],flow %p,skb %p,tcm %p)\n",
550 sch,p,flow,skb,tcm);
551 if (!find_flow(p,flow)) return -EINVAL;
552 tcm->tcm_handle = flow->classid;
553 rta = (struct rtattr *) b;
554 RTA_PUT(skb,TCA_OPTIONS,0,NULL);
555 RTA_PUT(skb,TCA_ATM_HDR,flow->hdr_len,flow->hdr);
556 if (flow->vcc) {
557 struct sockaddr_atmpvc pvc;
558 int state;
560 pvc.sap_family = AF_ATMPVC;
561 pvc.sap_addr.itf = flow->vcc->dev ? flow->vcc->dev->number : -1;
562 pvc.sap_addr.vpi = flow->vcc->vpi;
563 pvc.sap_addr.vci = flow->vcc->vci;
564 RTA_PUT(skb,TCA_ATM_ADDR,sizeof(pvc),&pvc);
565 state = ATM_VF2VS(flow->vcc->flags);
566 RTA_PUT(skb,TCA_ATM_STATE,sizeof(state),&state);
568 if (flow->excess)
569 RTA_PUT(skb,TCA_ATM_EXCESS,sizeof(u32),&flow->classid);
570 else {
571 static u32 zero = 0;
573 RTA_PUT(skb,TCA_ATM_EXCESS,sizeof(zero),&zero);
575 rta->rta_len = skb->tail-b;
576 return skb->len;
578 rtattr_failure:
579 skb_trim(skb,b-skb->data);
580 return -1;
583 static int atm_tc_dump(struct Qdisc *sch, struct sk_buff *skb)
585 return 0;
588 #endif
591 static struct Qdisc_class_ops atm_class_ops =
593 atm_tc_graft, /* graft */
594 atm_tc_leaf, /* leaf */
595 atm_tc_get, /* get */
596 atm_tc_put, /* put */
597 atm_tc_change, /* change */
598 atm_tc_delete, /* delete */
599 atm_tc_walk, /* walk */
601 atm_tc_find_tcf, /* tcf_chain */
602 atm_tc_bind_filter, /* bind_tcf */
603 atm_tc_put, /* unbind_tcf */
605 #ifdef CONFIG_RTNETLINK
606 atm_tc_dump_class, /* dump */
607 #endif
610 struct Qdisc_ops atm_qdisc_ops =
612 NULL, /* next */
613 &atm_class_ops, /* cl_ops */
614 "atm",
615 sizeof(struct atm_qdisc_data),
617 atm_tc_enqueue, /* enqueue */
618 atm_tc_dequeue, /* dequeue */
619 atm_tc_enqueue, /* requeue; we're cheating a little */
620 atm_tc_drop, /* drop */
622 atm_tc_init, /* init */
623 atm_tc_reset, /* reset */
624 atm_tc_destroy, /* destroy */
625 NULL, /* change */
627 #ifdef CONFIG_RTNETLINK
628 atm_tc_dump /* dump */
629 #endif
633 #ifdef MODULE
634 int init_module(void)
636 return register_qdisc(&atm_qdisc_ops);
640 void cleanup_module(void)
642 unregister_qdisc(&atm_qdisc_ops);
644 #endif