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 = kmalloc(sizeof(struct atm_flow_data) + hdr_len, GFP_KERNEL);
296 DPRINTK("atm_tc_change: flow %p\n", flow);
301 memset(flow, 0, sizeof(*flow));
302 flow->filter_list = NULL;
303 if (!(flow->q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, classid)))
304 flow->q = &noop_qdisc;
305 DPRINTK("atm_tc_change: qdisc %p\n", flow->q);
307 flow->vcc = ATM_SD(sock); /* speedup */
308 flow->vcc->user_back = flow;
309 DPRINTK("atm_tc_change: vcc %p\n", flow->vcc);
310 flow->old_pop = flow->vcc->pop;
312 flow->vcc->pop = sch_atm_pop;
313 flow->classid = classid;
315 flow->excess = excess;
316 flow->next = p->link.next;
318 flow->hdr_len = hdr_len;
320 memcpy(flow->hdr, hdr, hdr_len);
322 memcpy(flow->hdr, llc_oui_ip, sizeof(llc_oui_ip));
323 *arg = (unsigned long)flow;
327 atm_tc_put(sch, (unsigned long)excess);
332 static int atm_tc_delete(struct Qdisc *sch, unsigned long arg)
334 struct atm_qdisc_data *p = PRIV(sch);
335 struct atm_flow_data *flow = (struct atm_flow_data *)arg;
337 DPRINTK("atm_tc_delete(sch %p,[qdisc %p],flow %p)\n", sch, p, flow);
338 if (!find_flow(PRIV(sch), flow))
340 if (flow->filter_list || flow == &p->link)
343 * Reference count must be 2: one for "keepalive" (set at class
344 * creation), and one for the reference held when calling delete.
347 printk(KERN_ERR "atm_tc_delete: flow->ref == %d\n", flow->ref);
351 return -EBUSY; /* catch references via excess, etc. */
352 atm_tc_put(sch, arg);
356 static void atm_tc_walk(struct Qdisc *sch, struct qdisc_walker *walker)
358 struct atm_qdisc_data *p = PRIV(sch);
359 struct atm_flow_data *flow;
361 DPRINTK("atm_tc_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
364 for (flow = p->flows; flow; flow = flow->next) {
365 if (walker->count >= walker->skip)
366 if (walker->fn(sch, (unsigned long)flow, walker) < 0) {
374 static struct tcf_proto **atm_tc_find_tcf(struct Qdisc *sch, unsigned long cl)
376 struct atm_qdisc_data *p = PRIV(sch);
377 struct atm_flow_data *flow = (struct atm_flow_data *)cl;
379 DPRINTK("atm_tc_find_tcf(sch %p,[qdisc %p],flow %p)\n", sch, p, flow);
380 return flow ? &flow->filter_list : &p->link.filter_list;
383 /* --------------------------- Qdisc operations ---------------------------- */
385 static int atm_tc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
387 struct atm_qdisc_data *p = PRIV(sch);
388 struct atm_flow_data *flow = NULL; /* @@@ */
389 struct tcf_result res;
391 int ret = NET_XMIT_POLICED;
393 D2PRINTK("atm_tc_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
394 result = TC_POLICE_OK; /* be nice to gcc */
395 if (TC_H_MAJ(skb->priority) != sch->handle ||
396 !(flow = (struct atm_flow_data *)atm_tc_get(sch, skb->priority)))
397 for (flow = p->flows; flow; flow = flow->next)
398 if (flow->filter_list) {
399 result = tc_classify_compat(skb,
404 flow = (struct atm_flow_data *)res.class;
406 flow = lookup_flow(sch, res.classid);
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_ACT
420 return NET_XMIT_SUCCESS;
424 case TC_POLICE_RECLASSIFY:
428 ATM_SKB(skb)->atm_options |= ATM_ATMOPT_CLP;
434 if ((ret = flow->q->enqueue(skb, flow->q)) != 0) {
438 flow->qstats.drops++;
441 sch->bstats.bytes += skb->len;
442 sch->bstats.packets++;
443 flow->bstats.bytes += skb->len;
444 flow->bstats.packets++;
446 * Okay, this may seem weird. We pretend we've dropped the packet if
447 * it goes via ATM. The reason for this is that the outer qdisc
448 * expects to be able to q->dequeue the packet later on if we return
449 * success at this place. Also, sch->q.qdisc needs to reflect whether
450 * there is a packet egligible for dequeuing or not. Note that the
451 * statistics of the outer qdisc are necessarily wrong because of all
452 * this. There's currently no correct solution for this.
454 if (flow == &p->link) {
458 tasklet_schedule(&p->task);
459 return NET_XMIT_BYPASS;
463 * Dequeue packets and send them over ATM. Note that we quite deliberately
464 * avoid checking net_device's flow control here, simply because sch_atm
465 * uses its own channels, which have nothing to do with any CLIP/LANE/or
466 * non-ATM interfaces.
469 static void sch_atm_dequeue(unsigned long data)
471 struct Qdisc *sch = (struct Qdisc *)data;
472 struct atm_qdisc_data *p = PRIV(sch);
473 struct atm_flow_data *flow;
476 D2PRINTK("sch_atm_dequeue(sch %p,[qdisc %p])\n", sch, p);
477 for (flow = p->link.next; flow; flow = flow->next)
479 * If traffic is properly shaped, this won't generate nasty
480 * little bursts. Otherwise, it may ... (but that's okay)
482 while ((skb = flow->q->dequeue(flow->q))) {
483 if (!atm_may_send(flow->vcc, skb->truesize)) {
484 (void)flow->q->ops->requeue(skb, flow->q);
487 D2PRINTK("atm_tc_dequeue: sending on class %p\n", flow);
488 /* remove any LL header somebody else has attached */
489 skb_pull(skb, skb_network_offset(skb));
490 if (skb_headroom(skb) < flow->hdr_len) {
493 new = skb_realloc_headroom(skb, flow->hdr_len);
499 D2PRINTK("sch_atm_dequeue: ip %p, data %p\n",
500 skb_network_header(skb), skb->data);
501 ATM_SKB(skb)->vcc = flow->vcc;
502 memcpy(skb_push(skb, flow->hdr_len), flow->hdr,
504 atomic_add(skb->truesize,
505 &sk_atm(flow->vcc)->sk_wmem_alloc);
506 /* atm.atm_options are already set by atm_tc_enqueue */
507 flow->vcc->send(flow->vcc, skb);
511 static struct sk_buff *atm_tc_dequeue(struct Qdisc *sch)
513 struct atm_qdisc_data *p = PRIV(sch);
516 D2PRINTK("atm_tc_dequeue(sch %p,[qdisc %p])\n", sch, p);
517 tasklet_schedule(&p->task);
518 skb = p->link.q->dequeue(p->link.q);
524 static int atm_tc_requeue(struct sk_buff *skb, struct Qdisc *sch)
526 struct atm_qdisc_data *p = PRIV(sch);
529 D2PRINTK("atm_tc_requeue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
530 ret = p->link.q->ops->requeue(skb, p->link.q);
533 sch->qstats.requeues++;
536 p->link.qstats.drops++;
541 static unsigned int atm_tc_drop(struct Qdisc *sch)
543 struct atm_qdisc_data *p = PRIV(sch);
544 struct atm_flow_data *flow;
547 DPRINTK("atm_tc_drop(sch %p,[qdisc %p])\n", sch, p);
548 for (flow = p->flows; flow; flow = flow->next)
549 if (flow->q->ops->drop && (len = flow->q->ops->drop(flow->q)))
554 static int atm_tc_init(struct Qdisc *sch, struct rtattr *opt)
556 struct atm_qdisc_data *p = PRIV(sch);
558 DPRINTK("atm_tc_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
560 if (!(p->link.q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
562 p->link.q = &noop_qdisc;
563 DPRINTK("atm_tc_init: link (%p) qdisc %p\n", &p->link, p->link.q);
564 p->link.filter_list = NULL;
567 p->link.classid = sch->handle;
570 tasklet_init(&p->task, sch_atm_dequeue, (unsigned long)sch);
574 static void atm_tc_reset(struct Qdisc *sch)
576 struct atm_qdisc_data *p = PRIV(sch);
577 struct atm_flow_data *flow;
579 DPRINTK("atm_tc_reset(sch %p,[qdisc %p])\n", sch, p);
580 for (flow = p->flows; flow; flow = flow->next)
581 qdisc_reset(flow->q);
585 static void atm_tc_destroy(struct Qdisc *sch)
587 struct atm_qdisc_data *p = PRIV(sch);
588 struct atm_flow_data *flow;
590 DPRINTK("atm_tc_destroy(sch %p,[qdisc %p])\n", sch, p);
592 while ((flow = p->flows)) {
593 tcf_destroy_chain(flow->filter_list);
594 flow->filter_list = NULL;
596 printk(KERN_ERR "atm_destroy: %p->ref = %d\n", flow,
598 atm_tc_put(sch, (unsigned long)flow);
599 if (p->flows == flow) {
600 printk(KERN_ERR "atm_destroy: putting flow %p didn't "
602 p->flows = flow->next; /* brute force */
606 tasklet_kill(&p->task);
609 static int atm_tc_dump_class(struct Qdisc *sch, unsigned long cl,
610 struct sk_buff *skb, struct tcmsg *tcm)
612 struct atm_qdisc_data *p = PRIV(sch);
613 struct atm_flow_data *flow = (struct atm_flow_data *)cl;
614 unsigned char *b = skb_tail_pointer(skb);
617 DPRINTK("atm_tc_dump_class(sch %p,[qdisc %p],flow %p,skb %p,tcm %p)\n",
618 sch, p, flow, skb, tcm);
619 if (!find_flow(p, flow))
621 tcm->tcm_handle = flow->classid;
622 tcm->tcm_info = flow->q->handle;
623 rta = (struct rtattr *)b;
624 RTA_PUT(skb, TCA_OPTIONS, 0, NULL);
625 RTA_PUT(skb, TCA_ATM_HDR, flow->hdr_len, flow->hdr);
627 struct sockaddr_atmpvc pvc;
630 pvc.sap_family = AF_ATMPVC;
631 pvc.sap_addr.itf = flow->vcc->dev ? flow->vcc->dev->number : -1;
632 pvc.sap_addr.vpi = flow->vcc->vpi;
633 pvc.sap_addr.vci = flow->vcc->vci;
634 RTA_PUT(skb, TCA_ATM_ADDR, sizeof(pvc), &pvc);
635 state = ATM_VF2VS(flow->vcc->flags);
636 RTA_PUT(skb, TCA_ATM_STATE, sizeof(state), &state);
639 RTA_PUT(skb, TCA_ATM_EXCESS, sizeof(u32), &flow->classid);
643 RTA_PUT(skb, TCA_ATM_EXCESS, sizeof(zero), &zero);
645 rta->rta_len = skb_tail_pointer(skb) - b;
653 atm_tc_dump_class_stats(struct Qdisc *sch, unsigned long arg,
656 struct atm_flow_data *flow = (struct atm_flow_data *)arg;
658 flow->qstats.qlen = flow->q->q.qlen;
660 if (gnet_stats_copy_basic(d, &flow->bstats) < 0 ||
661 gnet_stats_copy_queue(d, &flow->qstats) < 0)
667 static int atm_tc_dump(struct Qdisc *sch, struct sk_buff *skb)
672 static struct Qdisc_class_ops atm_class_ops = {
673 .graft = atm_tc_graft,
677 .change = atm_tc_change,
678 .delete = atm_tc_delete,
680 .tcf_chain = atm_tc_find_tcf,
681 .bind_tcf = atm_tc_bind_filter,
682 .unbind_tcf = atm_tc_put,
683 .dump = atm_tc_dump_class,
684 .dump_stats = atm_tc_dump_class_stats,
687 static struct Qdisc_ops atm_qdisc_ops = {
688 .cl_ops = &atm_class_ops,
690 .priv_size = sizeof(struct atm_qdisc_data),
691 .enqueue = atm_tc_enqueue,
692 .dequeue = atm_tc_dequeue,
693 .requeue = atm_tc_requeue,
696 .reset = atm_tc_reset,
697 .destroy = atm_tc_destroy,
699 .owner = THIS_MODULE,
702 static int __init atm_init(void)
704 return register_qdisc(&atm_qdisc_ops);
707 static void __exit atm_exit(void)
709 unregister_qdisc(&atm_qdisc_ops);
712 module_init(atm_init)
713 module_exit(atm_exit)
714 MODULE_LICENSE("GPL");