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>
18 #include <net/pkt_cls.h>
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_class_common common
;
45 struct Qdisc
*q
; /* FIFO, TBF, etc. */
46 struct tcf_proto __rcu
*filter_list
;
47 struct tcf_block
*block
;
48 struct atm_vcc
*vcc
; /* VCC; NULL if VCC is closed */
49 void (*old_pop
)(struct atm_vcc
*vcc
,
50 struct sk_buff
*skb
); /* chaining */
51 struct atm_qdisc_data
*parent
; /* parent qdisc */
52 struct socket
*sock
; /* for closing */
53 int ref
; /* reference count */
54 struct gnet_stats_basic_packed bstats
;
55 struct gnet_stats_queue qstats
;
56 struct list_head list
;
57 struct atm_flow_data
*excess
; /* flow for excess traffic;
58 NULL to set CLP instead */
60 unsigned char hdr
[0]; /* header data; MUST BE LAST */
63 struct atm_qdisc_data
{
64 struct atm_flow_data link
; /* unclassified skbs go here */
65 struct list_head flows
; /* NB: "link" is also on this
67 struct tasklet_struct task
; /* dequeue tasklet */
70 /* ------------------------- Class/flow operations ------------------------- */
72 static inline struct atm_flow_data
*lookup_flow(struct Qdisc
*sch
, u32 classid
)
74 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
75 struct atm_flow_data
*flow
;
77 list_for_each_entry(flow
, &p
->flows
, list
) {
78 if (flow
->common
.classid
== classid
)
84 static int atm_tc_graft(struct Qdisc
*sch
, unsigned long arg
,
85 struct Qdisc
*new, struct Qdisc
**old
)
87 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
88 struct atm_flow_data
*flow
= (struct atm_flow_data
*)arg
;
90 pr_debug("atm_tc_graft(sch %p,[qdisc %p],flow %p,new %p,old %p)\n",
91 sch
, p
, flow
, new, old
);
92 if (list_empty(&flow
->list
))
103 static struct Qdisc
*atm_tc_leaf(struct Qdisc
*sch
, unsigned long cl
)
105 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
107 pr_debug("atm_tc_leaf(sch %p,flow %p)\n", sch
, flow
);
108 return flow
? flow
->q
: NULL
;
111 static unsigned long atm_tc_find(struct Qdisc
*sch
, u32 classid
)
113 struct atm_qdisc_data
*p __maybe_unused
= qdisc_priv(sch
);
114 struct atm_flow_data
*flow
;
116 pr_debug("%s(sch %p,[qdisc %p],classid %x)\n", __func__
, sch
, p
, classid
);
117 flow
= lookup_flow(sch
, classid
);
118 pr_debug("%s: flow %p\n", __func__
, flow
);
119 return (unsigned long)flow
;
122 static unsigned long atm_tc_bind_filter(struct Qdisc
*sch
,
123 unsigned long parent
, u32 classid
)
125 struct atm_qdisc_data
*p __maybe_unused
= qdisc_priv(sch
);
126 struct atm_flow_data
*flow
;
128 pr_debug("%s(sch %p,[qdisc %p],classid %x)\n", __func__
, sch
, p
, classid
);
129 flow
= lookup_flow(sch
, classid
);
132 pr_debug("%s: flow %p\n", __func__
, flow
);
133 return (unsigned long)flow
;
137 * atm_tc_put handles all destructions, including the ones that are explicitly
138 * requested (atm_tc_destroy, etc.). The assumption here is that we never drop
139 * anything that still seems to be in use.
141 static void atm_tc_put(struct Qdisc
*sch
, unsigned long cl
)
143 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
144 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
146 pr_debug("atm_tc_put(sch %p,[qdisc %p],flow %p)\n", sch
, p
, flow
);
149 pr_debug("atm_tc_put: destroying\n");
150 list_del_init(&flow
->list
);
151 pr_debug("atm_tc_put: qdisc %p\n", flow
->q
);
152 qdisc_destroy(flow
->q
);
153 tcf_block_put(flow
->block
);
155 pr_debug("atm_tc_put: f_count %ld\n",
156 file_count(flow
->sock
->file
));
157 flow
->vcc
->pop
= flow
->old_pop
;
158 sockfd_put(flow
->sock
);
161 atm_tc_put(sch
, (unsigned long)flow
->excess
);
162 if (flow
!= &p
->link
)
165 * If flow == &p->link, the qdisc no longer works at this point and
166 * needs to be removed. (By the caller of atm_tc_put.)
170 static void sch_atm_pop(struct atm_vcc
*vcc
, struct sk_buff
*skb
)
172 struct atm_qdisc_data
*p
= VCC2FLOW(vcc
)->parent
;
174 pr_debug("sch_atm_pop(vcc %p,skb %p,[qdisc %p])\n", vcc
, skb
, p
);
175 VCC2FLOW(vcc
)->old_pop(vcc
, skb
);
176 tasklet_schedule(&p
->task
);
179 static const u8 llc_oui_ip
[] = {
180 0xaa, /* DSAP: non-ISO */
181 0xaa, /* SSAP: non-ISO */
182 0x03, /* Ctrl: Unnumbered Information Command PDU */
183 0x00, /* OUI: EtherType */
186 }; /* Ethertype IP (0800) */
188 static const struct nla_policy atm_policy
[TCA_ATM_MAX
+ 1] = {
189 [TCA_ATM_FD
] = { .type
= NLA_U32
},
190 [TCA_ATM_EXCESS
] = { .type
= NLA_U32
},
193 static int atm_tc_change(struct Qdisc
*sch
, u32 classid
, u32 parent
,
194 struct nlattr
**tca
, unsigned long *arg
)
196 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
197 struct atm_flow_data
*flow
= (struct atm_flow_data
*)*arg
;
198 struct atm_flow_data
*excess
= NULL
;
199 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
200 struct nlattr
*tb
[TCA_ATM_MAX
+ 1];
202 int fd
, error
, hdr_len
;
205 pr_debug("atm_tc_change(sch %p,[qdisc %p],classid %x,parent %x,"
206 "flow %p,opt %p)\n", sch
, p
, classid
, parent
, flow
, opt
);
208 * The concept of parents doesn't apply for this qdisc.
210 if (parent
&& parent
!= TC_H_ROOT
&& parent
!= sch
->handle
)
213 * ATM classes cannot be changed. In order to change properties of the
214 * ATM connection, that socket needs to be modified directly (via the
215 * native ATM API. In order to send a flow to a different VC, the old
216 * class needs to be removed and a new one added. (This may be changed
224 error
= nla_parse_nested(tb
, TCA_ATM_MAX
, opt
, atm_policy
, NULL
);
230 fd
= nla_get_u32(tb
[TCA_ATM_FD
]);
231 pr_debug("atm_tc_change: fd %d\n", fd
);
232 if (tb
[TCA_ATM_HDR
]) {
233 hdr_len
= nla_len(tb
[TCA_ATM_HDR
]);
234 hdr
= nla_data(tb
[TCA_ATM_HDR
]);
236 hdr_len
= RFC1483LLC_LEN
;
237 hdr
= NULL
; /* default LLC/SNAP for IP */
239 if (!tb
[TCA_ATM_EXCESS
])
242 excess
= (struct atm_flow_data
*)
243 atm_tc_find(sch
, nla_get_u32(tb
[TCA_ATM_EXCESS
]));
247 pr_debug("atm_tc_change: type %d, payload %d, hdr_len %d\n",
248 opt
->nla_type
, nla_len(opt
), hdr_len
);
249 sock
= sockfd_lookup(fd
, &error
);
251 return error
; /* f_count++ */
252 pr_debug("atm_tc_change: f_count %ld\n", file_count(sock
->file
));
253 if (sock
->ops
->family
!= PF_ATMSVC
&& sock
->ops
->family
!= PF_ATMPVC
) {
257 /* @@@ should check if the socket is really operational or we'll crash
260 if (TC_H_MAJ(classid
^ sch
->handle
)) {
261 pr_debug("atm_tc_change: classid mismatch\n");
269 for (i
= 1; i
< 0x8000; i
++) {
270 classid
= TC_H_MAKE(sch
->handle
, 0x8000 | i
);
271 cl
= atm_tc_find(sch
, classid
);
276 pr_debug("atm_tc_change: new id %x\n", classid
);
277 flow
= kzalloc(sizeof(struct atm_flow_data
) + hdr_len
, GFP_KERNEL
);
278 pr_debug("atm_tc_change: flow %p\n", flow
);
284 error
= tcf_block_get(&flow
->block
, &flow
->filter_list
);
290 flow
->q
= qdisc_create_dflt(sch
->dev_queue
, &pfifo_qdisc_ops
, classid
);
292 flow
->q
= &noop_qdisc
;
293 pr_debug("atm_tc_change: qdisc %p\n", flow
->q
);
295 flow
->vcc
= ATM_SD(sock
); /* speedup */
296 flow
->vcc
->user_back
= flow
;
297 pr_debug("atm_tc_change: vcc %p\n", flow
->vcc
);
298 flow
->old_pop
= flow
->vcc
->pop
;
300 flow
->vcc
->pop
= sch_atm_pop
;
301 flow
->common
.classid
= classid
;
303 flow
->excess
= excess
;
304 list_add(&flow
->list
, &p
->link
.list
);
305 flow
->hdr_len
= hdr_len
;
307 memcpy(flow
->hdr
, hdr
, hdr_len
);
309 memcpy(flow
->hdr
, llc_oui_ip
, sizeof(llc_oui_ip
));
310 *arg
= (unsigned long)flow
;
317 static int atm_tc_delete(struct Qdisc
*sch
, unsigned long arg
)
319 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
320 struct atm_flow_data
*flow
= (struct atm_flow_data
*)arg
;
322 pr_debug("atm_tc_delete(sch %p,[qdisc %p],flow %p)\n", sch
, p
, flow
);
323 if (list_empty(&flow
->list
))
325 if (rcu_access_pointer(flow
->filter_list
) || flow
== &p
->link
)
328 * Reference count must be 2: one for "keepalive" (set at class
329 * creation), and one for the reference held when calling delete.
332 pr_err("atm_tc_delete: flow->ref == %d\n", flow
->ref
);
336 return -EBUSY
; /* catch references via excess, etc. */
337 atm_tc_put(sch
, arg
);
341 static void atm_tc_walk(struct Qdisc
*sch
, struct qdisc_walker
*walker
)
343 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
344 struct atm_flow_data
*flow
;
346 pr_debug("atm_tc_walk(sch %p,[qdisc %p],walker %p)\n", sch
, p
, walker
);
349 list_for_each_entry(flow
, &p
->flows
, list
) {
350 if (walker
->count
>= walker
->skip
&&
351 walker
->fn(sch
, (unsigned long)flow
, walker
) < 0) {
359 static struct tcf_block
*atm_tc_tcf_block(struct Qdisc
*sch
, unsigned long cl
)
361 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
362 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
364 pr_debug("atm_tc_find_tcf(sch %p,[qdisc %p],flow %p)\n", sch
, p
, flow
);
365 return flow
? flow
->block
: p
->link
.block
;
368 /* --------------------------- Qdisc operations ---------------------------- */
370 static int atm_tc_enqueue(struct sk_buff
*skb
, struct Qdisc
*sch
,
371 struct sk_buff
**to_free
)
373 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
374 struct atm_flow_data
*flow
;
375 struct tcf_result res
;
377 int ret
= NET_XMIT_SUCCESS
| __NET_XMIT_BYPASS
;
379 pr_debug("atm_tc_enqueue(skb %p,sch %p,[qdisc %p])\n", skb
, sch
, p
);
380 result
= TC_ACT_OK
; /* be nice to gcc */
382 if (TC_H_MAJ(skb
->priority
) != sch
->handle
||
383 !(flow
= (struct atm_flow_data
*)atm_tc_find(sch
, skb
->priority
))) {
384 struct tcf_proto
*fl
;
386 list_for_each_entry(flow
, &p
->flows
, list
) {
387 fl
= rcu_dereference_bh(flow
->filter_list
);
389 result
= tcf_classify(skb
, fl
, &res
, true);
392 flow
= (struct atm_flow_data
*)res
.class;
394 flow
= lookup_flow(sch
, res
.classid
);
406 ATM_SKB(skb
)->atm_options
= flow
->vcc
->atm_options
;
407 /*@@@ looks good ... but it's not supposed to work :-) */
408 #ifdef CONFIG_NET_CLS_ACT
413 __qdisc_drop(skb
, to_free
);
414 return NET_XMIT_SUCCESS
| __NET_XMIT_STOLEN
;
416 __qdisc_drop(skb
, to_free
);
418 case TC_ACT_RECLASSIFY
:
422 ATM_SKB(skb
)->atm_options
|= ATM_ATMOPT_CLP
;
428 ret
= qdisc_enqueue(skb
, flow
->q
, to_free
);
429 if (ret
!= NET_XMIT_SUCCESS
) {
431 if (net_xmit_drop_count(ret
)) {
432 qdisc_qstats_drop(sch
);
434 flow
->qstats
.drops
++;
439 * Okay, this may seem weird. We pretend we've dropped the packet if
440 * it goes via ATM. The reason for this is that the outer qdisc
441 * expects to be able to q->dequeue the packet later on if we return
442 * success at this place. Also, sch->q.qdisc needs to reflect whether
443 * there is a packet egligible for dequeuing or not. Note that the
444 * statistics of the outer qdisc are necessarily wrong because of all
445 * this. There's currently no correct solution for this.
447 if (flow
== &p
->link
) {
449 return NET_XMIT_SUCCESS
;
451 tasklet_schedule(&p
->task
);
452 return NET_XMIT_SUCCESS
| __NET_XMIT_BYPASS
;
456 * Dequeue packets and send them over ATM. Note that we quite deliberately
457 * avoid checking net_device's flow control here, simply because sch_atm
458 * uses its own channels, which have nothing to do with any CLIP/LANE/or
459 * non-ATM interfaces.
462 static void sch_atm_dequeue(unsigned long data
)
464 struct Qdisc
*sch
= (struct Qdisc
*)data
;
465 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
466 struct atm_flow_data
*flow
;
469 pr_debug("sch_atm_dequeue(sch %p,[qdisc %p])\n", sch
, p
);
470 list_for_each_entry(flow
, &p
->flows
, list
) {
471 if (flow
== &p
->link
)
474 * If traffic is properly shaped, this won't generate nasty
475 * little bursts. Otherwise, it may ... (but that's okay)
477 while ((skb
= flow
->q
->ops
->peek(flow
->q
))) {
478 if (!atm_may_send(flow
->vcc
, skb
->truesize
))
481 skb
= qdisc_dequeue_peeked(flow
->q
);
485 qdisc_bstats_update(sch
, skb
);
486 bstats_update(&flow
->bstats
, skb
);
487 pr_debug("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 pr_debug("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 refcount_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
);
512 static struct sk_buff
*atm_tc_dequeue(struct Qdisc
*sch
)
514 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
517 pr_debug("atm_tc_dequeue(sch %p,[qdisc %p])\n", sch
, p
);
518 tasklet_schedule(&p
->task
);
519 skb
= qdisc_dequeue_peeked(p
->link
.q
);
525 static struct sk_buff
*atm_tc_peek(struct Qdisc
*sch
)
527 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
529 pr_debug("atm_tc_peek(sch %p,[qdisc %p])\n", sch
, p
);
531 return p
->link
.q
->ops
->peek(p
->link
.q
);
534 static int atm_tc_init(struct Qdisc
*sch
, struct nlattr
*opt
)
536 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
539 pr_debug("atm_tc_init(sch %p,[qdisc %p],opt %p)\n", sch
, p
, opt
);
540 INIT_LIST_HEAD(&p
->flows
);
541 INIT_LIST_HEAD(&p
->link
.list
);
542 list_add(&p
->link
.list
, &p
->flows
);
543 p
->link
.q
= qdisc_create_dflt(sch
->dev_queue
,
544 &pfifo_qdisc_ops
, sch
->handle
);
546 p
->link
.q
= &noop_qdisc
;
547 pr_debug("atm_tc_init: link (%p) qdisc %p\n", &p
->link
, p
->link
.q
);
549 err
= tcf_block_get(&p
->link
.block
, &p
->link
.filter_list
);
555 p
->link
.common
.classid
= sch
->handle
;
557 tasklet_init(&p
->task
, sch_atm_dequeue
, (unsigned long)sch
);
561 static void atm_tc_reset(struct Qdisc
*sch
)
563 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
564 struct atm_flow_data
*flow
;
566 pr_debug("atm_tc_reset(sch %p,[qdisc %p])\n", sch
, p
);
567 list_for_each_entry(flow
, &p
->flows
, list
)
568 qdisc_reset(flow
->q
);
572 static void atm_tc_destroy(struct Qdisc
*sch
)
574 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
575 struct atm_flow_data
*flow
, *tmp
;
577 pr_debug("atm_tc_destroy(sch %p,[qdisc %p])\n", sch
, p
);
578 list_for_each_entry(flow
, &p
->flows
, list
) {
579 tcf_block_put(flow
->block
);
583 list_for_each_entry_safe(flow
, tmp
, &p
->flows
, list
) {
585 pr_err("atm_destroy: %p->ref = %d\n", flow
, flow
->ref
);
586 atm_tc_put(sch
, (unsigned long)flow
);
588 tasklet_kill(&p
->task
);
591 static int atm_tc_dump_class(struct Qdisc
*sch
, unsigned long cl
,
592 struct sk_buff
*skb
, struct tcmsg
*tcm
)
594 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
595 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
598 pr_debug("atm_tc_dump_class(sch %p,[qdisc %p],flow %p,skb %p,tcm %p)\n",
599 sch
, p
, flow
, skb
, tcm
);
600 if (list_empty(&flow
->list
))
602 tcm
->tcm_handle
= flow
->common
.classid
;
603 tcm
->tcm_info
= flow
->q
->handle
;
605 nest
= nla_nest_start(skb
, TCA_OPTIONS
);
607 goto nla_put_failure
;
609 if (nla_put(skb
, TCA_ATM_HDR
, flow
->hdr_len
, flow
->hdr
))
610 goto nla_put_failure
;
612 struct sockaddr_atmpvc pvc
;
615 memset(&pvc
, 0, sizeof(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 if (nla_put(skb
, TCA_ATM_ADDR
, sizeof(pvc
), &pvc
))
621 goto nla_put_failure
;
622 state
= ATM_VF2VS(flow
->vcc
->flags
);
623 if (nla_put_u32(skb
, TCA_ATM_STATE
, state
))
624 goto nla_put_failure
;
627 if (nla_put_u32(skb
, TCA_ATM_EXCESS
, flow
->common
.classid
))
628 goto nla_put_failure
;
630 if (nla_put_u32(skb
, TCA_ATM_EXCESS
, 0))
631 goto nla_put_failure
;
633 return nla_nest_end(skb
, nest
);
636 nla_nest_cancel(skb
, nest
);
640 atm_tc_dump_class_stats(struct Qdisc
*sch
, unsigned long arg
,
643 struct atm_flow_data
*flow
= (struct atm_flow_data
*)arg
;
645 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch
),
646 d
, NULL
, &flow
->bstats
) < 0 ||
647 gnet_stats_copy_queue(d
, NULL
, &flow
->qstats
, flow
->q
->q
.qlen
) < 0)
653 static int atm_tc_dump(struct Qdisc
*sch
, struct sk_buff
*skb
)
658 static const struct Qdisc_class_ops atm_class_ops
= {
659 .graft
= atm_tc_graft
,
662 .change
= atm_tc_change
,
663 .delete = atm_tc_delete
,
665 .tcf_block
= atm_tc_tcf_block
,
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
,
680 .reset
= atm_tc_reset
,
681 .destroy
= atm_tc_destroy
,
683 .owner
= THIS_MODULE
,
686 static int __init
atm_init(void)
688 return register_qdisc(&atm_qdisc_ops
);
691 static void __exit
atm_exit(void)
693 unregister_qdisc(&atm_qdisc_ops
);
696 module_init(atm_init
)
697 module_exit(atm_exit
)
698 MODULE_LICENSE("GPL");