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/string.h>
9 #include <linux/errno.h>
10 #include <linux/skbuff.h>
11 #include <linux/atmdev.h>
12 #include <linux/atmclip.h>
13 #include <linux/rtnetlink.h>
14 #include <linux/file.h> /* for fput */
15 #include <net/netlink.h>
16 #include <net/pkt_sched.h>
18 extern struct socket
*sockfd_lookup(int fd
, int *err
); /* @@@ fix this */
21 * The ATM queuing discipline provides a framework for invoking classifiers
22 * (aka "filters"), which in turn select classes of this queuing discipline.
23 * Each class maps the flow(s) it is handling to a given VC. Multiple classes
24 * may share the same VC.
26 * When creating a class, VCs are specified by passing the number of the open
27 * socket descriptor by which the calling process references the VC. The kernel
28 * keeps the VC open at least until all classes using it are removed.
30 * In this file, most functions are named atm_tc_* to avoid confusion with all
31 * the atm_* in net/atm. This naming convention differs from what's used in the
35 * - sometimes messes up the IP stack
36 * - any manipulations besides the few operations described in the README, are
37 * untested and likely to crash the system
38 * - should lock the flow while there is data in the queue (?)
41 #define VCC2FLOW(vcc) ((struct atm_flow_data *) ((vcc)->user_back))
43 struct atm_flow_data
{
44 struct Qdisc
*q
; /* FIFO, TBF, etc. */
45 struct tcf_proto
*filter_list
;
46 struct atm_vcc
*vcc
; /* VCC; NULL if VCC is closed */
47 void (*old_pop
)(struct atm_vcc
*vcc
,
48 struct sk_buff
*skb
); /* chaining */
49 struct atm_qdisc_data
*parent
; /* parent qdisc */
50 struct socket
*sock
; /* for closing */
51 u32 classid
; /* x:y type ID */
52 int ref
; /* reference count */
53 struct gnet_stats_basic_packed bstats
;
54 struct gnet_stats_queue qstats
;
55 struct list_head list
;
56 struct atm_flow_data
*excess
; /* flow for excess traffic;
57 NULL to set CLP instead */
59 unsigned char hdr
[0]; /* header data; MUST BE LAST */
62 struct atm_qdisc_data
{
63 struct atm_flow_data link
; /* unclassified skbs go here */
64 struct list_head flows
; /* NB: "link" is also on this
66 struct tasklet_struct task
; /* dequeue tasklet */
69 /* ------------------------- Class/flow operations ------------------------- */
71 static inline struct atm_flow_data
*lookup_flow(struct Qdisc
*sch
, u32 classid
)
73 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
74 struct atm_flow_data
*flow
;
76 list_for_each_entry(flow
, &p
->flows
, list
) {
77 if (flow
->classid
== classid
)
83 static int atm_tc_graft(struct Qdisc
*sch
, unsigned long arg
,
84 struct Qdisc
*new, struct Qdisc
**old
)
86 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
87 struct atm_flow_data
*flow
= (struct atm_flow_data
*)arg
;
89 pr_debug("atm_tc_graft(sch %p,[qdisc %p],flow %p,new %p,old %p)\n",
90 sch
, p
, flow
, new, old
);
91 if (list_empty(&flow
->list
))
102 static struct Qdisc
*atm_tc_leaf(struct Qdisc
*sch
, unsigned long cl
)
104 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
106 pr_debug("atm_tc_leaf(sch %p,flow %p)\n", sch
, flow
);
107 return flow
? flow
->q
: NULL
;
110 static unsigned long atm_tc_get(struct Qdisc
*sch
, u32 classid
)
112 struct atm_qdisc_data
*p __maybe_unused
= qdisc_priv(sch
);
113 struct atm_flow_data
*flow
;
115 pr_debug("atm_tc_get(sch %p,[qdisc %p],classid %x)\n", sch
, p
, classid
);
116 flow
= lookup_flow(sch
, classid
);
119 pr_debug("atm_tc_get: flow %p\n", flow
);
120 return (unsigned long)flow
;
123 static unsigned long atm_tc_bind_filter(struct Qdisc
*sch
,
124 unsigned long parent
, u32 classid
)
126 return atm_tc_get(sch
, classid
);
130 * atm_tc_put handles all destructions, including the ones that are explicitly
131 * requested (atm_tc_destroy, etc.). The assumption here is that we never drop
132 * anything that still seems to be in use.
134 static void atm_tc_put(struct Qdisc
*sch
, unsigned long cl
)
136 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
137 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
139 pr_debug("atm_tc_put(sch %p,[qdisc %p],flow %p)\n", sch
, p
, flow
);
142 pr_debug("atm_tc_put: destroying\n");
143 list_del_init(&flow
->list
);
144 pr_debug("atm_tc_put: qdisc %p\n", flow
->q
);
145 qdisc_destroy(flow
->q
);
146 tcf_destroy_chain(&flow
->filter_list
);
148 pr_debug("atm_tc_put: f_count %ld\n",
149 file_count(flow
->sock
->file
));
150 flow
->vcc
->pop
= flow
->old_pop
;
151 sockfd_put(flow
->sock
);
154 atm_tc_put(sch
, (unsigned long)flow
->excess
);
155 if (flow
!= &p
->link
)
158 * If flow == &p->link, the qdisc no longer works at this point and
159 * needs to be removed. (By the caller of atm_tc_put.)
163 static void sch_atm_pop(struct atm_vcc
*vcc
, struct sk_buff
*skb
)
165 struct atm_qdisc_data
*p
= VCC2FLOW(vcc
)->parent
;
167 pr_debug("sch_atm_pop(vcc %p,skb %p,[qdisc %p])\n", vcc
, skb
, p
);
168 VCC2FLOW(vcc
)->old_pop(vcc
, skb
);
169 tasklet_schedule(&p
->task
);
172 static const u8 llc_oui_ip
[] = {
173 0xaa, /* DSAP: non-ISO */
174 0xaa, /* SSAP: non-ISO */
175 0x03, /* Ctrl: Unnumbered Information Command PDU */
176 0x00, /* OUI: EtherType */
179 }; /* Ethertype IP (0800) */
181 static const struct nla_policy atm_policy
[TCA_ATM_MAX
+ 1] = {
182 [TCA_ATM_FD
] = { .type
= NLA_U32
},
183 [TCA_ATM_EXCESS
] = { .type
= NLA_U32
},
186 static int atm_tc_change(struct Qdisc
*sch
, u32 classid
, u32 parent
,
187 struct nlattr
**tca
, unsigned long *arg
)
189 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
190 struct atm_flow_data
*flow
= (struct atm_flow_data
*)*arg
;
191 struct atm_flow_data
*excess
= NULL
;
192 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
193 struct nlattr
*tb
[TCA_ATM_MAX
+ 1];
195 int fd
, error
, hdr_len
;
198 pr_debug("atm_tc_change(sch %p,[qdisc %p],classid %x,parent %x,"
199 "flow %p,opt %p)\n", sch
, p
, classid
, parent
, flow
, opt
);
201 * The concept of parents doesn't apply for this qdisc.
203 if (parent
&& parent
!= TC_H_ROOT
&& parent
!= sch
->handle
)
206 * ATM classes cannot be changed. In order to change properties of the
207 * ATM connection, that socket needs to be modified directly (via the
208 * native ATM API. In order to send a flow to a different VC, the old
209 * class needs to be removed and a new one added. (This may be changed
217 error
= nla_parse_nested(tb
, TCA_ATM_MAX
, opt
, atm_policy
);
223 fd
= nla_get_u32(tb
[TCA_ATM_FD
]);
224 pr_debug("atm_tc_change: fd %d\n", fd
);
225 if (tb
[TCA_ATM_HDR
]) {
226 hdr_len
= nla_len(tb
[TCA_ATM_HDR
]);
227 hdr
= nla_data(tb
[TCA_ATM_HDR
]);
229 hdr_len
= RFC1483LLC_LEN
;
230 hdr
= NULL
; /* default LLC/SNAP for IP */
232 if (!tb
[TCA_ATM_EXCESS
])
235 excess
= (struct atm_flow_data
*)
236 atm_tc_get(sch
, nla_get_u32(tb
[TCA_ATM_EXCESS
]));
240 pr_debug("atm_tc_change: type %d, payload %d, hdr_len %d\n",
241 opt
->nla_type
, nla_len(opt
), hdr_len
);
242 sock
= sockfd_lookup(fd
, &error
);
244 return error
; /* f_count++ */
245 pr_debug("atm_tc_change: f_count %ld\n", file_count(sock
->file
));
246 if (sock
->ops
->family
!= PF_ATMSVC
&& sock
->ops
->family
!= PF_ATMPVC
) {
250 /* @@@ should check if the socket is really operational or we'll crash
253 if (TC_H_MAJ(classid
^ sch
->handle
)) {
254 pr_debug("atm_tc_change: classid mismatch\n");
258 if (!list_empty(&flow
->list
)) {
266 for (i
= 1; i
< 0x8000; i
++) {
267 classid
= TC_H_MAKE(sch
->handle
, 0x8000 | i
);
268 cl
= atm_tc_get(sch
, classid
);
274 pr_debug("atm_tc_change: new id %x\n", classid
);
275 flow
= kzalloc(sizeof(struct atm_flow_data
) + hdr_len
, GFP_KERNEL
);
276 pr_debug("atm_tc_change: flow %p\n", flow
);
281 flow
->filter_list
= NULL
;
282 flow
->q
= qdisc_create_dflt(qdisc_dev(sch
), sch
->dev_queue
,
283 &pfifo_qdisc_ops
, classid
);
285 flow
->q
= &noop_qdisc
;
286 pr_debug("atm_tc_change: qdisc %p\n", flow
->q
);
288 flow
->vcc
= ATM_SD(sock
); /* speedup */
289 flow
->vcc
->user_back
= flow
;
290 pr_debug("atm_tc_change: vcc %p\n", flow
->vcc
);
291 flow
->old_pop
= flow
->vcc
->pop
;
293 flow
->vcc
->pop
= sch_atm_pop
;
294 flow
->classid
= classid
;
296 flow
->excess
= excess
;
297 list_add(&flow
->list
, &p
->link
.list
);
298 flow
->hdr_len
= hdr_len
;
300 memcpy(flow
->hdr
, hdr
, hdr_len
);
302 memcpy(flow
->hdr
, llc_oui_ip
, sizeof(llc_oui_ip
));
303 *arg
= (unsigned long)flow
;
307 atm_tc_put(sch
, (unsigned long)excess
);
312 static int atm_tc_delete(struct Qdisc
*sch
, unsigned long arg
)
314 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
315 struct atm_flow_data
*flow
= (struct atm_flow_data
*)arg
;
317 pr_debug("atm_tc_delete(sch %p,[qdisc %p],flow %p)\n", sch
, p
, flow
);
318 if (list_empty(&flow
->list
))
320 if (flow
->filter_list
|| flow
== &p
->link
)
323 * Reference count must be 2: one for "keepalive" (set at class
324 * creation), and one for the reference held when calling delete.
327 printk(KERN_ERR
"atm_tc_delete: flow->ref == %d\n", flow
->ref
);
331 return -EBUSY
; /* catch references via excess, etc. */
332 atm_tc_put(sch
, arg
);
336 static void atm_tc_walk(struct Qdisc
*sch
, struct qdisc_walker
*walker
)
338 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
339 struct atm_flow_data
*flow
;
341 pr_debug("atm_tc_walk(sch %p,[qdisc %p],walker %p)\n", sch
, p
, walker
);
344 list_for_each_entry(flow
, &p
->flows
, list
) {
345 if (walker
->count
>= walker
->skip
&&
346 walker
->fn(sch
, (unsigned long)flow
, walker
) < 0) {
354 static struct tcf_proto
**atm_tc_find_tcf(struct Qdisc
*sch
, unsigned long cl
)
356 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
357 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
359 pr_debug("atm_tc_find_tcf(sch %p,[qdisc %p],flow %p)\n", sch
, p
, flow
);
360 return flow
? &flow
->filter_list
: &p
->link
.filter_list
;
363 /* --------------------------- Qdisc operations ---------------------------- */
365 static int atm_tc_enqueue(struct sk_buff
*skb
, struct Qdisc
*sch
)
367 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
368 struct atm_flow_data
*flow
;
369 struct tcf_result res
;
371 int ret
= NET_XMIT_POLICED
;
373 pr_debug("atm_tc_enqueue(skb %p,sch %p,[qdisc %p])\n", skb
, sch
, p
);
374 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 list_for_each_entry(flow
, &p
->flows
, list
) {
379 if (flow
->filter_list
) {
380 result
= tc_classify_compat(skb
,
385 flow
= (struct atm_flow_data
*)res
.class;
387 flow
= lookup_flow(sch
, res
.classid
);
399 ATM_SKB(skb
)->atm_options
= flow
->vcc
->atm_options
;
400 /*@@@ looks good ... but it's not supposed to work :-) */
401 #ifdef CONFIG_NET_CLS_ACT
406 return NET_XMIT_SUCCESS
| __NET_XMIT_STOLEN
;
410 case TC_POLICE_RECLASSIFY
:
414 ATM_SKB(skb
)->atm_options
|= ATM_ATMOPT_CLP
;
420 ret
= qdisc_enqueue(skb
, flow
->q
);
421 if (ret
!= NET_XMIT_SUCCESS
) {
423 if (net_xmit_drop_count(ret
)) {
426 flow
->qstats
.drops
++;
430 sch
->bstats
.bytes
+= qdisc_pkt_len(skb
);
431 sch
->bstats
.packets
++;
432 flow
->bstats
.bytes
+= qdisc_pkt_len(skb
);
433 flow
->bstats
.packets
++;
435 * Okay, this may seem weird. We pretend we've dropped the packet if
436 * it goes via ATM. The reason for this is that the outer qdisc
437 * expects to be able to q->dequeue the packet later on if we return
438 * success at this place. Also, sch->q.qdisc needs to reflect whether
439 * there is a packet egligible for dequeuing or not. Note that the
440 * statistics of the outer qdisc are necessarily wrong because of all
441 * this. There's currently no correct solution for this.
443 if (flow
== &p
->link
) {
445 return NET_XMIT_SUCCESS
;
447 tasklet_schedule(&p
->task
);
448 return NET_XMIT_SUCCESS
| __NET_XMIT_BYPASS
;
452 * Dequeue packets and send them over ATM. Note that we quite deliberately
453 * avoid checking net_device's flow control here, simply because sch_atm
454 * uses its own channels, which have nothing to do with any CLIP/LANE/or
455 * non-ATM interfaces.
458 static void sch_atm_dequeue(unsigned long data
)
460 struct Qdisc
*sch
= (struct Qdisc
*)data
;
461 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
462 struct atm_flow_data
*flow
;
465 pr_debug("sch_atm_dequeue(sch %p,[qdisc %p])\n", sch
, p
);
466 list_for_each_entry(flow
, &p
->flows
, list
) {
467 if (flow
== &p
->link
)
470 * If traffic is properly shaped, this won't generate nasty
471 * little bursts. Otherwise, it may ... (but that's okay)
473 while ((skb
= flow
->q
->ops
->peek(flow
->q
))) {
474 if (!atm_may_send(flow
->vcc
, skb
->truesize
))
477 skb
= qdisc_dequeue_peeked(flow
->q
);
481 pr_debug("atm_tc_dequeue: sending on class %p\n", flow
);
482 /* remove any LL header somebody else has attached */
483 skb_pull(skb
, skb_network_offset(skb
));
484 if (skb_headroom(skb
) < flow
->hdr_len
) {
487 new = skb_realloc_headroom(skb
, flow
->hdr_len
);
493 pr_debug("sch_atm_dequeue: ip %p, data %p\n",
494 skb_network_header(skb
), skb
->data
);
495 ATM_SKB(skb
)->vcc
= flow
->vcc
;
496 memcpy(skb_push(skb
, flow
->hdr_len
), flow
->hdr
,
498 atomic_add(skb
->truesize
,
499 &sk_atm(flow
->vcc
)->sk_wmem_alloc
);
500 /* atm.atm_options are already set by atm_tc_enqueue */
501 flow
->vcc
->send(flow
->vcc
, skb
);
506 static struct sk_buff
*atm_tc_dequeue(struct Qdisc
*sch
)
508 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
511 pr_debug("atm_tc_dequeue(sch %p,[qdisc %p])\n", sch
, p
);
512 tasklet_schedule(&p
->task
);
513 skb
= qdisc_dequeue_peeked(p
->link
.q
);
519 static struct sk_buff
*atm_tc_peek(struct Qdisc
*sch
)
521 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
523 pr_debug("atm_tc_peek(sch %p,[qdisc %p])\n", sch
, p
);
525 return p
->link
.q
->ops
->peek(p
->link
.q
);
528 static unsigned int atm_tc_drop(struct Qdisc
*sch
)
530 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
531 struct atm_flow_data
*flow
;
534 pr_debug("atm_tc_drop(sch %p,[qdisc %p])\n", sch
, p
);
535 list_for_each_entry(flow
, &p
->flows
, list
) {
536 if (flow
->q
->ops
->drop
&& (len
= flow
->q
->ops
->drop(flow
->q
)))
542 static int atm_tc_init(struct Qdisc
*sch
, struct nlattr
*opt
)
544 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
546 pr_debug("atm_tc_init(sch %p,[qdisc %p],opt %p)\n", sch
, p
, opt
);
547 INIT_LIST_HEAD(&p
->flows
);
548 INIT_LIST_HEAD(&p
->link
.list
);
549 list_add(&p
->link
.list
, &p
->flows
);
550 p
->link
.q
= qdisc_create_dflt(qdisc_dev(sch
), sch
->dev_queue
,
551 &pfifo_qdisc_ops
, sch
->handle
);
553 p
->link
.q
= &noop_qdisc
;
554 pr_debug("atm_tc_init: link (%p) qdisc %p\n", &p
->link
, p
->link
.q
);
555 p
->link
.filter_list
= NULL
;
558 p
->link
.classid
= sch
->handle
;
560 tasklet_init(&p
->task
, sch_atm_dequeue
, (unsigned long)sch
);
564 static void atm_tc_reset(struct Qdisc
*sch
)
566 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
567 struct atm_flow_data
*flow
;
569 pr_debug("atm_tc_reset(sch %p,[qdisc %p])\n", sch
, p
);
570 list_for_each_entry(flow
, &p
->flows
, list
)
571 qdisc_reset(flow
->q
);
575 static void atm_tc_destroy(struct Qdisc
*sch
)
577 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
578 struct atm_flow_data
*flow
, *tmp
;
580 pr_debug("atm_tc_destroy(sch %p,[qdisc %p])\n", sch
, p
);
581 list_for_each_entry(flow
, &p
->flows
, list
)
582 tcf_destroy_chain(&flow
->filter_list
);
584 list_for_each_entry_safe(flow
, tmp
, &p
->flows
, list
) {
586 printk(KERN_ERR
"atm_destroy: %p->ref = %d\n", flow
,
588 atm_tc_put(sch
, (unsigned long)flow
);
590 tasklet_kill(&p
->task
);
593 static int atm_tc_dump_class(struct Qdisc
*sch
, unsigned long cl
,
594 struct sk_buff
*skb
, struct tcmsg
*tcm
)
596 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
597 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
600 pr_debug("atm_tc_dump_class(sch %p,[qdisc %p],flow %p,skb %p,tcm %p)\n",
601 sch
, p
, flow
, skb
, tcm
);
602 if (list_empty(&flow
->list
))
604 tcm
->tcm_handle
= flow
->classid
;
605 tcm
->tcm_info
= flow
->q
->handle
;
607 nest
= nla_nest_start(skb
, TCA_OPTIONS
);
609 goto nla_put_failure
;
611 NLA_PUT(skb
, TCA_ATM_HDR
, flow
->hdr_len
, flow
->hdr
);
613 struct sockaddr_atmpvc pvc
;
616 pvc
.sap_family
= AF_ATMPVC
;
617 pvc
.sap_addr
.itf
= flow
->vcc
->dev
? flow
->vcc
->dev
->number
: -1;
618 pvc
.sap_addr
.vpi
= flow
->vcc
->vpi
;
619 pvc
.sap_addr
.vci
= flow
->vcc
->vci
;
620 NLA_PUT(skb
, TCA_ATM_ADDR
, sizeof(pvc
), &pvc
);
621 state
= ATM_VF2VS(flow
->vcc
->flags
);
622 NLA_PUT_U32(skb
, TCA_ATM_STATE
, state
);
625 NLA_PUT_U32(skb
, TCA_ATM_EXCESS
, flow
->classid
);
627 NLA_PUT_U32(skb
, TCA_ATM_EXCESS
, 0);
630 nla_nest_end(skb
, nest
);
634 nla_nest_cancel(skb
, nest
);
638 atm_tc_dump_class_stats(struct Qdisc
*sch
, unsigned long arg
,
641 struct atm_flow_data
*flow
= (struct atm_flow_data
*)arg
;
643 flow
->qstats
.qlen
= flow
->q
->q
.qlen
;
645 if (gnet_stats_copy_basic(d
, &flow
->bstats
) < 0 ||
646 gnet_stats_copy_queue(d
, &flow
->qstats
) < 0)
652 static int atm_tc_dump(struct Qdisc
*sch
, struct sk_buff
*skb
)
657 static const struct Qdisc_class_ops atm_class_ops
= {
658 .graft
= atm_tc_graft
,
662 .change
= atm_tc_change
,
663 .delete = atm_tc_delete
,
665 .tcf_chain
= atm_tc_find_tcf
,
666 .bind_tcf
= atm_tc_bind_filter
,
667 .unbind_tcf
= atm_tc_put
,
668 .dump
= atm_tc_dump_class
,
669 .dump_stats
= atm_tc_dump_class_stats
,
672 static struct Qdisc_ops atm_qdisc_ops __read_mostly
= {
673 .cl_ops
= &atm_class_ops
,
675 .priv_size
= sizeof(struct atm_qdisc_data
),
676 .enqueue
= atm_tc_enqueue
,
677 .dequeue
= atm_tc_dequeue
,
681 .reset
= atm_tc_reset
,
682 .destroy
= atm_tc_destroy
,
684 .owner
= THIS_MODULE
,
687 static int __init
atm_init(void)
689 return register_qdisc(&atm_qdisc_ops
);
692 static void __exit
atm_exit(void)
694 unregister_qdisc(&atm_qdisc_ops
);
697 module_init(atm_init
)
698 module_exit(atm_exit
)
699 MODULE_LICENSE("GPL");