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/slab.h>
7 #include <linux/init.h>
8 #include <linux/interrupt.h>
9 #include <linux/string.h>
10 #include <linux/errno.h>
11 #include <linux/skbuff.h>
12 #include <linux/atmdev.h>
13 #include <linux/atmclip.h>
14 #include <linux/rtnetlink.h>
15 #include <linux/file.h> /* for fput */
16 #include <net/netlink.h>
17 #include <net/pkt_sched.h>
20 * The ATM queuing discipline provides a framework for invoking classifiers
21 * (aka "filters"), which in turn select classes of this queuing discipline.
22 * Each class maps the flow(s) it is handling to a given VC. Multiple classes
23 * may share the same VC.
25 * When creating a class, VCs are specified by passing the number of the open
26 * socket descriptor by which the calling process references the VC. The kernel
27 * keeps the VC open at least until all classes using it are removed.
29 * In this file, most functions are named atm_tc_* to avoid confusion with all
30 * the atm_* in net/atm. This naming convention differs from what's used in the
34 * - sometimes messes up the IP stack
35 * - any manipulations besides the few operations described in the README, are
36 * untested and likely to crash the system
37 * - should lock the flow while there is data in the queue (?)
40 #define VCC2FLOW(vcc) ((struct atm_flow_data *) ((vcc)->user_back))
42 struct atm_flow_data
{
43 struct Qdisc
*q
; /* FIFO, TBF, etc. */
44 struct tcf_proto __rcu
*filter_list
;
45 struct atm_vcc
*vcc
; /* VCC; NULL if VCC is closed */
46 void (*old_pop
)(struct atm_vcc
*vcc
,
47 struct sk_buff
*skb
); /* chaining */
48 struct atm_qdisc_data
*parent
; /* parent qdisc */
49 struct socket
*sock
; /* for closing */
50 u32 classid
; /* x:y type ID */
51 int ref
; /* reference count */
52 struct gnet_stats_basic_packed bstats
;
53 struct gnet_stats_queue qstats
;
54 struct list_head list
;
55 struct atm_flow_data
*excess
; /* flow for excess traffic;
56 NULL to set CLP instead */
58 unsigned char hdr
[0]; /* header data; MUST BE LAST */
61 struct atm_qdisc_data
{
62 struct atm_flow_data link
; /* unclassified skbs go here */
63 struct list_head flows
; /* NB: "link" is also on this
65 struct tasklet_struct task
; /* dequeue tasklet */
68 /* ------------------------- Class/flow operations ------------------------- */
70 static inline struct atm_flow_data
*lookup_flow(struct Qdisc
*sch
, u32 classid
)
72 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
73 struct atm_flow_data
*flow
;
75 list_for_each_entry(flow
, &p
->flows
, list
) {
76 if (flow
->classid
== classid
)
82 static int atm_tc_graft(struct Qdisc
*sch
, unsigned long arg
,
83 struct Qdisc
*new, struct Qdisc
**old
)
85 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
86 struct atm_flow_data
*flow
= (struct atm_flow_data
*)arg
;
88 pr_debug("atm_tc_graft(sch %p,[qdisc %p],flow %p,new %p,old %p)\n",
89 sch
, p
, flow
, new, old
);
90 if (list_empty(&flow
->list
))
101 static struct Qdisc
*atm_tc_leaf(struct Qdisc
*sch
, unsigned long cl
)
103 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
105 pr_debug("atm_tc_leaf(sch %p,flow %p)\n", sch
, flow
);
106 return flow
? flow
->q
: NULL
;
109 static unsigned long atm_tc_get(struct Qdisc
*sch
, u32 classid
)
111 struct atm_qdisc_data
*p __maybe_unused
= qdisc_priv(sch
);
112 struct atm_flow_data
*flow
;
114 pr_debug("atm_tc_get(sch %p,[qdisc %p],classid %x)\n", sch
, p
, classid
);
115 flow
= lookup_flow(sch
, classid
);
118 pr_debug("atm_tc_get: flow %p\n", flow
);
119 return (unsigned long)flow
;
122 static unsigned long atm_tc_bind_filter(struct Qdisc
*sch
,
123 unsigned long parent
, u32 classid
)
125 return atm_tc_get(sch
, classid
);
129 * atm_tc_put handles all destructions, including the ones that are explicitly
130 * requested (atm_tc_destroy, etc.). The assumption here is that we never drop
131 * anything that still seems to be in use.
133 static void atm_tc_put(struct Qdisc
*sch
, unsigned long cl
)
135 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
136 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
138 pr_debug("atm_tc_put(sch %p,[qdisc %p],flow %p)\n", sch
, p
, flow
);
141 pr_debug("atm_tc_put: destroying\n");
142 list_del_init(&flow
->list
);
143 pr_debug("atm_tc_put: qdisc %p\n", flow
->q
);
144 qdisc_destroy(flow
->q
);
145 tcf_destroy_chain(&flow
->filter_list
);
147 pr_debug("atm_tc_put: f_count %ld\n",
148 file_count(flow
->sock
->file
));
149 flow
->vcc
->pop
= flow
->old_pop
;
150 sockfd_put(flow
->sock
);
153 atm_tc_put(sch
, (unsigned long)flow
->excess
);
154 if (flow
!= &p
->link
)
157 * If flow == &p->link, the qdisc no longer works at this point and
158 * needs to be removed. (By the caller of atm_tc_put.)
162 static void sch_atm_pop(struct atm_vcc
*vcc
, struct sk_buff
*skb
)
164 struct atm_qdisc_data
*p
= VCC2FLOW(vcc
)->parent
;
166 pr_debug("sch_atm_pop(vcc %p,skb %p,[qdisc %p])\n", vcc
, skb
, p
);
167 VCC2FLOW(vcc
)->old_pop(vcc
, skb
);
168 tasklet_schedule(&p
->task
);
171 static const u8 llc_oui_ip
[] = {
172 0xaa, /* DSAP: non-ISO */
173 0xaa, /* SSAP: non-ISO */
174 0x03, /* Ctrl: Unnumbered Information Command PDU */
175 0x00, /* OUI: EtherType */
178 }; /* Ethertype IP (0800) */
180 static const struct nla_policy atm_policy
[TCA_ATM_MAX
+ 1] = {
181 [TCA_ATM_FD
] = { .type
= NLA_U32
},
182 [TCA_ATM_EXCESS
] = { .type
= NLA_U32
},
185 static int atm_tc_change(struct Qdisc
*sch
, u32 classid
, u32 parent
,
186 struct nlattr
**tca
, unsigned long *arg
)
188 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
189 struct atm_flow_data
*flow
= (struct atm_flow_data
*)*arg
;
190 struct atm_flow_data
*excess
= NULL
;
191 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
192 struct nlattr
*tb
[TCA_ATM_MAX
+ 1];
194 int fd
, error
, hdr_len
;
197 pr_debug("atm_tc_change(sch %p,[qdisc %p],classid %x,parent %x,"
198 "flow %p,opt %p)\n", sch
, p
, classid
, parent
, flow
, opt
);
200 * The concept of parents doesn't apply for this qdisc.
202 if (parent
&& parent
!= TC_H_ROOT
&& parent
!= sch
->handle
)
205 * ATM classes cannot be changed. In order to change properties of the
206 * ATM connection, that socket needs to be modified directly (via the
207 * native ATM API. In order to send a flow to a different VC, the old
208 * class needs to be removed and a new one added. (This may be changed
216 error
= nla_parse_nested(tb
, TCA_ATM_MAX
, opt
, atm_policy
);
222 fd
= nla_get_u32(tb
[TCA_ATM_FD
]);
223 pr_debug("atm_tc_change: fd %d\n", fd
);
224 if (tb
[TCA_ATM_HDR
]) {
225 hdr_len
= nla_len(tb
[TCA_ATM_HDR
]);
226 hdr
= nla_data(tb
[TCA_ATM_HDR
]);
228 hdr_len
= RFC1483LLC_LEN
;
229 hdr
= NULL
; /* default LLC/SNAP for IP */
231 if (!tb
[TCA_ATM_EXCESS
])
234 excess
= (struct atm_flow_data
*)
235 atm_tc_get(sch
, nla_get_u32(tb
[TCA_ATM_EXCESS
]));
239 pr_debug("atm_tc_change: type %d, payload %d, hdr_len %d\n",
240 opt
->nla_type
, nla_len(opt
), hdr_len
);
241 sock
= sockfd_lookup(fd
, &error
);
243 return error
; /* f_count++ */
244 pr_debug("atm_tc_change: f_count %ld\n", file_count(sock
->file
));
245 if (sock
->ops
->family
!= PF_ATMSVC
&& sock
->ops
->family
!= PF_ATMPVC
) {
249 /* @@@ should check if the socket is really operational or we'll crash
252 if (TC_H_MAJ(classid
^ sch
->handle
)) {
253 pr_debug("atm_tc_change: classid mismatch\n");
261 for (i
= 1; i
< 0x8000; i
++) {
262 classid
= TC_H_MAKE(sch
->handle
, 0x8000 | i
);
263 cl
= atm_tc_get(sch
, classid
);
269 pr_debug("atm_tc_change: new id %x\n", classid
);
270 flow
= kzalloc(sizeof(struct atm_flow_data
) + hdr_len
, GFP_KERNEL
);
271 pr_debug("atm_tc_change: flow %p\n", flow
);
276 RCU_INIT_POINTER(flow
->filter_list
, NULL
);
277 flow
->q
= qdisc_create_dflt(sch
->dev_queue
, &pfifo_qdisc_ops
, classid
);
279 flow
->q
= &noop_qdisc
;
280 pr_debug("atm_tc_change: qdisc %p\n", flow
->q
);
282 flow
->vcc
= ATM_SD(sock
); /* speedup */
283 flow
->vcc
->user_back
= flow
;
284 pr_debug("atm_tc_change: vcc %p\n", flow
->vcc
);
285 flow
->old_pop
= flow
->vcc
->pop
;
287 flow
->vcc
->pop
= sch_atm_pop
;
288 flow
->classid
= classid
;
290 flow
->excess
= excess
;
291 list_add(&flow
->list
, &p
->link
.list
);
292 flow
->hdr_len
= hdr_len
;
294 memcpy(flow
->hdr
, hdr
, hdr_len
);
296 memcpy(flow
->hdr
, llc_oui_ip
, sizeof(llc_oui_ip
));
297 *arg
= (unsigned long)flow
;
301 atm_tc_put(sch
, (unsigned long)excess
);
306 static int atm_tc_delete(struct Qdisc
*sch
, unsigned long arg
)
308 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
309 struct atm_flow_data
*flow
= (struct atm_flow_data
*)arg
;
311 pr_debug("atm_tc_delete(sch %p,[qdisc %p],flow %p)\n", sch
, p
, flow
);
312 if (list_empty(&flow
->list
))
314 if (rcu_access_pointer(flow
->filter_list
) || flow
== &p
->link
)
317 * Reference count must be 2: one for "keepalive" (set at class
318 * creation), and one for the reference held when calling delete.
321 pr_err("atm_tc_delete: flow->ref == %d\n", flow
->ref
);
325 return -EBUSY
; /* catch references via excess, etc. */
326 atm_tc_put(sch
, arg
);
330 static void atm_tc_walk(struct Qdisc
*sch
, struct qdisc_walker
*walker
)
332 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
333 struct atm_flow_data
*flow
;
335 pr_debug("atm_tc_walk(sch %p,[qdisc %p],walker %p)\n", sch
, p
, walker
);
338 list_for_each_entry(flow
, &p
->flows
, list
) {
339 if (walker
->count
>= walker
->skip
&&
340 walker
->fn(sch
, (unsigned long)flow
, walker
) < 0) {
348 static struct tcf_proto __rcu
**atm_tc_find_tcf(struct Qdisc
*sch
,
351 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
352 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
354 pr_debug("atm_tc_find_tcf(sch %p,[qdisc %p],flow %p)\n", sch
, p
, flow
);
355 return flow
? &flow
->filter_list
: &p
->link
.filter_list
;
358 /* --------------------------- Qdisc operations ---------------------------- */
360 static int atm_tc_enqueue(struct sk_buff
*skb
, struct Qdisc
*sch
,
361 struct sk_buff
**to_free
)
363 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
364 struct atm_flow_data
*flow
;
365 struct tcf_result res
;
367 int ret
= NET_XMIT_SUCCESS
| __NET_XMIT_BYPASS
;
369 pr_debug("atm_tc_enqueue(skb %p,sch %p,[qdisc %p])\n", skb
, sch
, p
);
370 result
= TC_ACT_OK
; /* be nice to gcc */
372 if (TC_H_MAJ(skb
->priority
) != sch
->handle
||
373 !(flow
= (struct atm_flow_data
*)atm_tc_get(sch
, skb
->priority
))) {
374 struct tcf_proto
*fl
;
376 list_for_each_entry(flow
, &p
->flows
, list
) {
377 fl
= rcu_dereference_bh(flow
->filter_list
);
379 result
= tc_classify(skb
, fl
, &res
, true);
382 flow
= (struct atm_flow_data
*)res
.class;
384 flow
= lookup_flow(sch
, res
.classid
);
396 ATM_SKB(skb
)->atm_options
= flow
->vcc
->atm_options
;
397 /*@@@ looks good ... but it's not supposed to work :-) */
398 #ifdef CONFIG_NET_CLS_ACT
402 __qdisc_drop(skb
, to_free
);
403 return NET_XMIT_SUCCESS
| __NET_XMIT_STOLEN
;
405 __qdisc_drop(skb
, to_free
);
407 case TC_ACT_RECLASSIFY
:
411 ATM_SKB(skb
)->atm_options
|= ATM_ATMOPT_CLP
;
417 ret
= qdisc_enqueue(skb
, flow
->q
, to_free
);
418 if (ret
!= NET_XMIT_SUCCESS
) {
420 if (net_xmit_drop_count(ret
)) {
421 qdisc_qstats_drop(sch
);
423 flow
->qstats
.drops
++;
428 * Okay, this may seem weird. We pretend we've dropped the packet if
429 * it goes via ATM. The reason for this is that the outer qdisc
430 * expects to be able to q->dequeue the packet later on if we return
431 * success at this place. Also, sch->q.qdisc needs to reflect whether
432 * there is a packet egligible for dequeuing or not. Note that the
433 * statistics of the outer qdisc are necessarily wrong because of all
434 * this. There's currently no correct solution for this.
436 if (flow
== &p
->link
) {
438 return NET_XMIT_SUCCESS
;
440 tasklet_schedule(&p
->task
);
441 return NET_XMIT_SUCCESS
| __NET_XMIT_BYPASS
;
445 * Dequeue packets and send them over ATM. Note that we quite deliberately
446 * avoid checking net_device's flow control here, simply because sch_atm
447 * uses its own channels, which have nothing to do with any CLIP/LANE/or
448 * non-ATM interfaces.
451 static void sch_atm_dequeue(unsigned long data
)
453 struct Qdisc
*sch
= (struct Qdisc
*)data
;
454 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
455 struct atm_flow_data
*flow
;
458 pr_debug("sch_atm_dequeue(sch %p,[qdisc %p])\n", sch
, p
);
459 list_for_each_entry(flow
, &p
->flows
, list
) {
460 if (flow
== &p
->link
)
463 * If traffic is properly shaped, this won't generate nasty
464 * little bursts. Otherwise, it may ... (but that's okay)
466 while ((skb
= flow
->q
->ops
->peek(flow
->q
))) {
467 if (!atm_may_send(flow
->vcc
, skb
->truesize
))
470 skb
= qdisc_dequeue_peeked(flow
->q
);
474 qdisc_bstats_update(sch
, skb
);
475 bstats_update(&flow
->bstats
, skb
);
476 pr_debug("atm_tc_dequeue: sending on class %p\n", flow
);
477 /* remove any LL header somebody else has attached */
478 skb_pull(skb
, skb_network_offset(skb
));
479 if (skb_headroom(skb
) < flow
->hdr_len
) {
482 new = skb_realloc_headroom(skb
, flow
->hdr_len
);
488 pr_debug("sch_atm_dequeue: ip %p, data %p\n",
489 skb_network_header(skb
), skb
->data
);
490 ATM_SKB(skb
)->vcc
= flow
->vcc
;
491 memcpy(skb_push(skb
, flow
->hdr_len
), flow
->hdr
,
493 atomic_add(skb
->truesize
,
494 &sk_atm(flow
->vcc
)->sk_wmem_alloc
);
495 /* atm.atm_options are already set by atm_tc_enqueue */
496 flow
->vcc
->send(flow
->vcc
, skb
);
501 static struct sk_buff
*atm_tc_dequeue(struct Qdisc
*sch
)
503 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
506 pr_debug("atm_tc_dequeue(sch %p,[qdisc %p])\n", sch
, p
);
507 tasklet_schedule(&p
->task
);
508 skb
= qdisc_dequeue_peeked(p
->link
.q
);
514 static struct sk_buff
*atm_tc_peek(struct Qdisc
*sch
)
516 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
518 pr_debug("atm_tc_peek(sch %p,[qdisc %p])\n", sch
, p
);
520 return p
->link
.q
->ops
->peek(p
->link
.q
);
523 static int atm_tc_init(struct Qdisc
*sch
, struct nlattr
*opt
)
525 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
527 pr_debug("atm_tc_init(sch %p,[qdisc %p],opt %p)\n", sch
, p
, opt
);
528 INIT_LIST_HEAD(&p
->flows
);
529 INIT_LIST_HEAD(&p
->link
.list
);
530 list_add(&p
->link
.list
, &p
->flows
);
531 p
->link
.q
= qdisc_create_dflt(sch
->dev_queue
,
532 &pfifo_qdisc_ops
, sch
->handle
);
534 p
->link
.q
= &noop_qdisc
;
535 pr_debug("atm_tc_init: link (%p) qdisc %p\n", &p
->link
, p
->link
.q
);
536 RCU_INIT_POINTER(p
->link
.filter_list
, NULL
);
539 p
->link
.classid
= sch
->handle
;
541 tasklet_init(&p
->task
, sch_atm_dequeue
, (unsigned long)sch
);
545 static void atm_tc_reset(struct Qdisc
*sch
)
547 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
548 struct atm_flow_data
*flow
;
550 pr_debug("atm_tc_reset(sch %p,[qdisc %p])\n", sch
, p
);
551 list_for_each_entry(flow
, &p
->flows
, list
)
552 qdisc_reset(flow
->q
);
556 static void atm_tc_destroy(struct Qdisc
*sch
)
558 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
559 struct atm_flow_data
*flow
, *tmp
;
561 pr_debug("atm_tc_destroy(sch %p,[qdisc %p])\n", sch
, p
);
562 list_for_each_entry(flow
, &p
->flows
, list
)
563 tcf_destroy_chain(&flow
->filter_list
);
565 list_for_each_entry_safe(flow
, tmp
, &p
->flows
, list
) {
567 pr_err("atm_destroy: %p->ref = %d\n", flow
, flow
->ref
);
568 atm_tc_put(sch
, (unsigned long)flow
);
570 tasklet_kill(&p
->task
);
573 static int atm_tc_dump_class(struct Qdisc
*sch
, unsigned long cl
,
574 struct sk_buff
*skb
, struct tcmsg
*tcm
)
576 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
577 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
580 pr_debug("atm_tc_dump_class(sch %p,[qdisc %p],flow %p,skb %p,tcm %p)\n",
581 sch
, p
, flow
, skb
, tcm
);
582 if (list_empty(&flow
->list
))
584 tcm
->tcm_handle
= flow
->classid
;
585 tcm
->tcm_info
= flow
->q
->handle
;
587 nest
= nla_nest_start(skb
, TCA_OPTIONS
);
589 goto nla_put_failure
;
591 if (nla_put(skb
, TCA_ATM_HDR
, flow
->hdr_len
, flow
->hdr
))
592 goto nla_put_failure
;
594 struct sockaddr_atmpvc pvc
;
597 memset(&pvc
, 0, sizeof(pvc
));
598 pvc
.sap_family
= AF_ATMPVC
;
599 pvc
.sap_addr
.itf
= flow
->vcc
->dev
? flow
->vcc
->dev
->number
: -1;
600 pvc
.sap_addr
.vpi
= flow
->vcc
->vpi
;
601 pvc
.sap_addr
.vci
= flow
->vcc
->vci
;
602 if (nla_put(skb
, TCA_ATM_ADDR
, sizeof(pvc
), &pvc
))
603 goto nla_put_failure
;
604 state
= ATM_VF2VS(flow
->vcc
->flags
);
605 if (nla_put_u32(skb
, TCA_ATM_STATE
, state
))
606 goto nla_put_failure
;
609 if (nla_put_u32(skb
, TCA_ATM_EXCESS
, flow
->classid
))
610 goto nla_put_failure
;
612 if (nla_put_u32(skb
, TCA_ATM_EXCESS
, 0))
613 goto nla_put_failure
;
615 return nla_nest_end(skb
, nest
);
618 nla_nest_cancel(skb
, nest
);
622 atm_tc_dump_class_stats(struct Qdisc
*sch
, unsigned long arg
,
625 struct atm_flow_data
*flow
= (struct atm_flow_data
*)arg
;
627 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch
),
628 d
, NULL
, &flow
->bstats
) < 0 ||
629 gnet_stats_copy_queue(d
, NULL
, &flow
->qstats
, flow
->q
->q
.qlen
) < 0)
635 static int atm_tc_dump(struct Qdisc
*sch
, struct sk_buff
*skb
)
640 static const struct Qdisc_class_ops atm_class_ops
= {
641 .graft
= atm_tc_graft
,
645 .change
= atm_tc_change
,
646 .delete = atm_tc_delete
,
648 .tcf_chain
= atm_tc_find_tcf
,
649 .bind_tcf
= atm_tc_bind_filter
,
650 .unbind_tcf
= atm_tc_put
,
651 .dump
= atm_tc_dump_class
,
652 .dump_stats
= atm_tc_dump_class_stats
,
655 static struct Qdisc_ops atm_qdisc_ops __read_mostly
= {
656 .cl_ops
= &atm_class_ops
,
658 .priv_size
= sizeof(struct atm_qdisc_data
),
659 .enqueue
= atm_tc_enqueue
,
660 .dequeue
= atm_tc_dequeue
,
663 .reset
= atm_tc_reset
,
664 .destroy
= atm_tc_destroy
,
666 .owner
= THIS_MODULE
,
669 static int __init
atm_init(void)
671 return register_qdisc(&atm_qdisc_ops
);
674 static void __exit
atm_exit(void)
676 unregister_qdisc(&atm_qdisc_ops
);
679 module_init(atm_init
)
680 module_exit(atm_exit
)
681 MODULE_LICENSE("GPL");