1 // SPDX-License-Identifier: GPL-2.0-only
2 /* net/sched/sch_atm.c - ATM VC selection "queueing discipline" */
4 /* Written 1998-2000 by Werner Almesberger, EPFL ICA */
6 #include <linux/module.h>
7 #include <linux/slab.h>
8 #include <linux/init.h>
9 #include <linux/interrupt.h>
10 #include <linux/string.h>
11 #include <linux/errno.h>
12 #include <linux/skbuff.h>
13 #include <linux/atmdev.h>
14 #include <linux/atmclip.h>
15 #include <linux/rtnetlink.h>
16 #include <linux/file.h> /* for fput */
17 #include <net/netlink.h>
18 #include <net/pkt_sched.h>
19 #include <net/pkt_cls.h>
22 * The ATM queuing discipline provides a framework for invoking classifiers
23 * (aka "filters"), which in turn select classes of this queuing discipline.
24 * Each class maps the flow(s) it is handling to a given VC. Multiple classes
25 * may share the same VC.
27 * When creating a class, VCs are specified by passing the number of the open
28 * socket descriptor by which the calling process references the VC. The kernel
29 * keeps the VC open at least until all classes using it are removed.
31 * In this file, most functions are named atm_tc_* to avoid confusion with all
32 * the atm_* in net/atm. This naming convention differs from what's used in the
36 * - sometimes messes up the IP stack
37 * - any manipulations besides the few operations described in the README, are
38 * untested and likely to crash the system
39 * - should lock the flow while there is data in the queue (?)
42 #define VCC2FLOW(vcc) ((struct atm_flow_data *) ((vcc)->user_back))
44 struct atm_flow_data
{
45 struct Qdisc_class_common common
;
46 struct Qdisc
*q
; /* FIFO, TBF, etc. */
47 struct tcf_proto __rcu
*filter_list
;
48 struct tcf_block
*block
;
49 struct atm_vcc
*vcc
; /* VCC; NULL if VCC is closed */
50 void (*old_pop
)(struct atm_vcc
*vcc
,
51 struct sk_buff
*skb
); /* chaining */
52 struct atm_qdisc_data
*parent
; /* parent qdisc */
53 struct socket
*sock
; /* for closing */
54 int ref
; /* reference count */
55 struct gnet_stats_basic_packed bstats
;
56 struct gnet_stats_queue qstats
;
57 struct list_head list
;
58 struct atm_flow_data
*excess
; /* flow for excess traffic;
59 NULL to set CLP instead */
61 unsigned char hdr
[]; /* header data; MUST BE LAST */
64 struct atm_qdisc_data
{
65 struct atm_flow_data link
; /* unclassified skbs go here */
66 struct list_head flows
; /* NB: "link" is also on this
68 struct tasklet_struct task
; /* dequeue tasklet */
71 /* ------------------------- Class/flow operations ------------------------- */
73 static inline struct atm_flow_data
*lookup_flow(struct Qdisc
*sch
, u32 classid
)
75 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
76 struct atm_flow_data
*flow
;
78 list_for_each_entry(flow
, &p
->flows
, list
) {
79 if (flow
->common
.classid
== classid
)
85 static int atm_tc_graft(struct Qdisc
*sch
, unsigned long arg
,
86 struct Qdisc
*new, struct Qdisc
**old
,
87 struct netlink_ext_ack
*extack
)
89 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
90 struct atm_flow_data
*flow
= (struct atm_flow_data
*)arg
;
92 pr_debug("atm_tc_graft(sch %p,[qdisc %p],flow %p,new %p,old %p)\n",
93 sch
, p
, flow
, new, old
);
94 if (list_empty(&flow
->list
))
105 static struct Qdisc
*atm_tc_leaf(struct Qdisc
*sch
, unsigned long cl
)
107 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
109 pr_debug("atm_tc_leaf(sch %p,flow %p)\n", sch
, flow
);
110 return flow
? flow
->q
: NULL
;
113 static unsigned long atm_tc_find(struct Qdisc
*sch
, u32 classid
)
115 struct atm_qdisc_data
*p __maybe_unused
= qdisc_priv(sch
);
116 struct atm_flow_data
*flow
;
118 pr_debug("%s(sch %p,[qdisc %p],classid %x)\n", __func__
, sch
, p
, classid
);
119 flow
= lookup_flow(sch
, classid
);
120 pr_debug("%s: flow %p\n", __func__
, flow
);
121 return (unsigned long)flow
;
124 static unsigned long atm_tc_bind_filter(struct Qdisc
*sch
,
125 unsigned long parent
, u32 classid
)
127 struct atm_qdisc_data
*p __maybe_unused
= qdisc_priv(sch
);
128 struct atm_flow_data
*flow
;
130 pr_debug("%s(sch %p,[qdisc %p],classid %x)\n", __func__
, sch
, p
, classid
);
131 flow
= lookup_flow(sch
, classid
);
134 pr_debug("%s: flow %p\n", __func__
, flow
);
135 return (unsigned long)flow
;
139 * atm_tc_put handles all destructions, including the ones that are explicitly
140 * requested (atm_tc_destroy, etc.). The assumption here is that we never drop
141 * anything that still seems to be in use.
143 static void atm_tc_put(struct Qdisc
*sch
, unsigned long cl
)
145 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
146 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
148 pr_debug("atm_tc_put(sch %p,[qdisc %p],flow %p)\n", sch
, p
, flow
);
151 pr_debug("atm_tc_put: destroying\n");
152 list_del_init(&flow
->list
);
153 pr_debug("atm_tc_put: qdisc %p\n", flow
->q
);
155 tcf_block_put(flow
->block
);
157 pr_debug("atm_tc_put: f_count %ld\n",
158 file_count(flow
->sock
->file
));
159 flow
->vcc
->pop
= flow
->old_pop
;
160 sockfd_put(flow
->sock
);
163 atm_tc_put(sch
, (unsigned long)flow
->excess
);
164 if (flow
!= &p
->link
)
167 * If flow == &p->link, the qdisc no longer works at this point and
168 * needs to be removed. (By the caller of atm_tc_put.)
172 static void sch_atm_pop(struct atm_vcc
*vcc
, struct sk_buff
*skb
)
174 struct atm_qdisc_data
*p
= VCC2FLOW(vcc
)->parent
;
176 pr_debug("sch_atm_pop(vcc %p,skb %p,[qdisc %p])\n", vcc
, skb
, p
);
177 VCC2FLOW(vcc
)->old_pop(vcc
, skb
);
178 tasklet_schedule(&p
->task
);
181 static const u8 llc_oui_ip
[] = {
182 0xaa, /* DSAP: non-ISO */
183 0xaa, /* SSAP: non-ISO */
184 0x03, /* Ctrl: Unnumbered Information Command PDU */
185 0x00, /* OUI: EtherType */
188 }; /* Ethertype IP (0800) */
190 static const struct nla_policy atm_policy
[TCA_ATM_MAX
+ 1] = {
191 [TCA_ATM_FD
] = { .type
= NLA_U32
},
192 [TCA_ATM_EXCESS
] = { .type
= NLA_U32
},
195 static int atm_tc_change(struct Qdisc
*sch
, u32 classid
, u32 parent
,
196 struct nlattr
**tca
, unsigned long *arg
,
197 struct netlink_ext_ack
*extack
)
199 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
200 struct atm_flow_data
*flow
= (struct atm_flow_data
*)*arg
;
201 struct atm_flow_data
*excess
= NULL
;
202 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
203 struct nlattr
*tb
[TCA_ATM_MAX
+ 1];
205 int fd
, error
, hdr_len
;
208 pr_debug("atm_tc_change(sch %p,[qdisc %p],classid %x,parent %x,"
209 "flow %p,opt %p)\n", sch
, p
, classid
, parent
, flow
, opt
);
211 * The concept of parents doesn't apply for this qdisc.
213 if (parent
&& parent
!= TC_H_ROOT
&& parent
!= sch
->handle
)
216 * ATM classes cannot be changed. In order to change properties of the
217 * ATM connection, that socket needs to be modified directly (via the
218 * native ATM API. In order to send a flow to a different VC, the old
219 * class needs to be removed and a new one added. (This may be changed
227 error
= nla_parse_nested_deprecated(tb
, TCA_ATM_MAX
, opt
, atm_policy
,
234 fd
= nla_get_u32(tb
[TCA_ATM_FD
]);
235 pr_debug("atm_tc_change: fd %d\n", fd
);
236 if (tb
[TCA_ATM_HDR
]) {
237 hdr_len
= nla_len(tb
[TCA_ATM_HDR
]);
238 hdr
= nla_data(tb
[TCA_ATM_HDR
]);
240 hdr_len
= RFC1483LLC_LEN
;
241 hdr
= NULL
; /* default LLC/SNAP for IP */
243 if (!tb
[TCA_ATM_EXCESS
])
246 excess
= (struct atm_flow_data
*)
247 atm_tc_find(sch
, nla_get_u32(tb
[TCA_ATM_EXCESS
]));
251 pr_debug("atm_tc_change: type %d, payload %d, hdr_len %d\n",
252 opt
->nla_type
, nla_len(opt
), hdr_len
);
253 sock
= sockfd_lookup(fd
, &error
);
255 return error
; /* f_count++ */
256 pr_debug("atm_tc_change: f_count %ld\n", file_count(sock
->file
));
257 if (sock
->ops
->family
!= PF_ATMSVC
&& sock
->ops
->family
!= PF_ATMPVC
) {
261 /* @@@ should check if the socket is really operational or we'll crash
264 if (TC_H_MAJ(classid
^ sch
->handle
)) {
265 pr_debug("atm_tc_change: classid mismatch\n");
273 for (i
= 1; i
< 0x8000; i
++) {
274 classid
= TC_H_MAKE(sch
->handle
, 0x8000 | i
);
275 cl
= atm_tc_find(sch
, classid
);
280 pr_debug("atm_tc_change: new id %x\n", classid
);
281 flow
= kzalloc(sizeof(struct atm_flow_data
) + hdr_len
, GFP_KERNEL
);
282 pr_debug("atm_tc_change: flow %p\n", flow
);
288 error
= tcf_block_get(&flow
->block
, &flow
->filter_list
, sch
,
295 flow
->q
= qdisc_create_dflt(sch
->dev_queue
, &pfifo_qdisc_ops
, classid
,
298 flow
->q
= &noop_qdisc
;
299 pr_debug("atm_tc_change: qdisc %p\n", flow
->q
);
301 flow
->vcc
= ATM_SD(sock
); /* speedup */
302 flow
->vcc
->user_back
= flow
;
303 pr_debug("atm_tc_change: vcc %p\n", flow
->vcc
);
304 flow
->old_pop
= flow
->vcc
->pop
;
306 flow
->vcc
->pop
= sch_atm_pop
;
307 flow
->common
.classid
= classid
;
309 flow
->excess
= excess
;
310 list_add(&flow
->list
, &p
->link
.list
);
311 flow
->hdr_len
= hdr_len
;
313 memcpy(flow
->hdr
, hdr
, hdr_len
);
315 memcpy(flow
->hdr
, llc_oui_ip
, sizeof(llc_oui_ip
));
316 *arg
= (unsigned long)flow
;
323 static int atm_tc_delete(struct Qdisc
*sch
, unsigned long arg
)
325 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
326 struct atm_flow_data
*flow
= (struct atm_flow_data
*)arg
;
328 pr_debug("atm_tc_delete(sch %p,[qdisc %p],flow %p)\n", sch
, p
, flow
);
329 if (list_empty(&flow
->list
))
331 if (rcu_access_pointer(flow
->filter_list
) || flow
== &p
->link
)
334 * Reference count must be 2: one for "keepalive" (set at class
335 * creation), and one for the reference held when calling delete.
338 pr_err("atm_tc_delete: flow->ref == %d\n", flow
->ref
);
342 return -EBUSY
; /* catch references via excess, etc. */
343 atm_tc_put(sch
, arg
);
347 static void atm_tc_walk(struct Qdisc
*sch
, struct qdisc_walker
*walker
)
349 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
350 struct atm_flow_data
*flow
;
352 pr_debug("atm_tc_walk(sch %p,[qdisc %p],walker %p)\n", sch
, p
, walker
);
355 list_for_each_entry(flow
, &p
->flows
, list
) {
356 if (walker
->count
>= walker
->skip
&&
357 walker
->fn(sch
, (unsigned long)flow
, walker
) < 0) {
365 static struct tcf_block
*atm_tc_tcf_block(struct Qdisc
*sch
, unsigned long cl
,
366 struct netlink_ext_ack
*extack
)
368 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
369 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
371 pr_debug("atm_tc_find_tcf(sch %p,[qdisc %p],flow %p)\n", sch
, p
, flow
);
372 return flow
? flow
->block
: p
->link
.block
;
375 /* --------------------------- Qdisc operations ---------------------------- */
377 static int atm_tc_enqueue(struct sk_buff
*skb
, struct Qdisc
*sch
,
378 struct sk_buff
**to_free
)
380 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
381 struct atm_flow_data
*flow
;
382 struct tcf_result res
;
384 int ret
= NET_XMIT_SUCCESS
| __NET_XMIT_BYPASS
;
386 pr_debug("atm_tc_enqueue(skb %p,sch %p,[qdisc %p])\n", skb
, sch
, p
);
387 result
= TC_ACT_OK
; /* be nice to gcc */
389 if (TC_H_MAJ(skb
->priority
) != sch
->handle
||
390 !(flow
= (struct atm_flow_data
*)atm_tc_find(sch
, skb
->priority
))) {
391 struct tcf_proto
*fl
;
393 list_for_each_entry(flow
, &p
->flows
, list
) {
394 fl
= rcu_dereference_bh(flow
->filter_list
);
396 result
= tcf_classify(skb
, fl
, &res
, true);
399 flow
= (struct atm_flow_data
*)res
.class;
401 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 __qdisc_drop(skb
, to_free
);
421 return NET_XMIT_SUCCESS
| __NET_XMIT_STOLEN
;
423 __qdisc_drop(skb
, to_free
);
425 case TC_ACT_RECLASSIFY
:
429 ATM_SKB(skb
)->atm_options
|= ATM_ATMOPT_CLP
;
435 ret
= qdisc_enqueue(skb
, flow
->q
, to_free
);
436 if (ret
!= NET_XMIT_SUCCESS
) {
438 if (net_xmit_drop_count(ret
)) {
439 qdisc_qstats_drop(sch
);
441 flow
->qstats
.drops
++;
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
) {
456 return NET_XMIT_SUCCESS
;
458 tasklet_schedule(&p
->task
);
459 return NET_XMIT_SUCCESS
| __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
= qdisc_priv(sch
);
473 struct atm_flow_data
*flow
;
476 pr_debug("sch_atm_dequeue(sch %p,[qdisc %p])\n", sch
, p
);
477 list_for_each_entry(flow
, &p
->flows
, list
) {
478 if (flow
== &p
->link
)
481 * If traffic is properly shaped, this won't generate nasty
482 * little bursts. Otherwise, it may ... (but that's okay)
484 while ((skb
= flow
->q
->ops
->peek(flow
->q
))) {
485 if (!atm_may_send(flow
->vcc
, skb
->truesize
))
488 skb
= qdisc_dequeue_peeked(flow
->q
);
492 qdisc_bstats_update(sch
, skb
);
493 bstats_update(&flow
->bstats
, skb
);
494 pr_debug("atm_tc_dequeue: sending on class %p\n", flow
);
495 /* remove any LL header somebody else has attached */
496 skb_pull(skb
, skb_network_offset(skb
));
497 if (skb_headroom(skb
) < flow
->hdr_len
) {
500 new = skb_realloc_headroom(skb
, flow
->hdr_len
);
506 pr_debug("sch_atm_dequeue: ip %p, data %p\n",
507 skb_network_header(skb
), skb
->data
);
508 ATM_SKB(skb
)->vcc
= flow
->vcc
;
509 memcpy(skb_push(skb
, flow
->hdr_len
), flow
->hdr
,
511 refcount_add(skb
->truesize
,
512 &sk_atm(flow
->vcc
)->sk_wmem_alloc
);
513 /* atm.atm_options are already set by atm_tc_enqueue */
514 flow
->vcc
->send(flow
->vcc
, skb
);
519 static struct sk_buff
*atm_tc_dequeue(struct Qdisc
*sch
)
521 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
524 pr_debug("atm_tc_dequeue(sch %p,[qdisc %p])\n", sch
, p
);
525 tasklet_schedule(&p
->task
);
526 skb
= qdisc_dequeue_peeked(p
->link
.q
);
532 static struct sk_buff
*atm_tc_peek(struct Qdisc
*sch
)
534 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
536 pr_debug("atm_tc_peek(sch %p,[qdisc %p])\n", sch
, p
);
538 return p
->link
.q
->ops
->peek(p
->link
.q
);
541 static int atm_tc_init(struct Qdisc
*sch
, struct nlattr
*opt
,
542 struct netlink_ext_ack
*extack
)
544 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
547 pr_debug("atm_tc_init(sch %p,[qdisc %p],opt %p)\n", sch
, p
, opt
);
548 INIT_LIST_HEAD(&p
->flows
);
549 INIT_LIST_HEAD(&p
->link
.list
);
550 list_add(&p
->link
.list
, &p
->flows
);
551 p
->link
.q
= qdisc_create_dflt(sch
->dev_queue
,
552 &pfifo_qdisc_ops
, sch
->handle
, extack
);
554 p
->link
.q
= &noop_qdisc
;
555 pr_debug("atm_tc_init: link (%p) qdisc %p\n", &p
->link
, p
->link
.q
);
557 err
= tcf_block_get(&p
->link
.block
, &p
->link
.filter_list
, sch
,
564 p
->link
.common
.classid
= sch
->handle
;
566 tasklet_init(&p
->task
, sch_atm_dequeue
, (unsigned long)sch
);
570 static void atm_tc_reset(struct Qdisc
*sch
)
572 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
573 struct atm_flow_data
*flow
;
575 pr_debug("atm_tc_reset(sch %p,[qdisc %p])\n", sch
, p
);
576 list_for_each_entry(flow
, &p
->flows
, list
)
577 qdisc_reset(flow
->q
);
581 static void atm_tc_destroy(struct Qdisc
*sch
)
583 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
584 struct atm_flow_data
*flow
, *tmp
;
586 pr_debug("atm_tc_destroy(sch %p,[qdisc %p])\n", sch
, p
);
587 list_for_each_entry(flow
, &p
->flows
, list
) {
588 tcf_block_put(flow
->block
);
592 list_for_each_entry_safe(flow
, tmp
, &p
->flows
, list
) {
594 pr_err("atm_destroy: %p->ref = %d\n", flow
, flow
->ref
);
595 atm_tc_put(sch
, (unsigned long)flow
);
597 tasklet_kill(&p
->task
);
600 static int atm_tc_dump_class(struct Qdisc
*sch
, unsigned long cl
,
601 struct sk_buff
*skb
, struct tcmsg
*tcm
)
603 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
604 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
607 pr_debug("atm_tc_dump_class(sch %p,[qdisc %p],flow %p,skb %p,tcm %p)\n",
608 sch
, p
, flow
, skb
, tcm
);
609 if (list_empty(&flow
->list
))
611 tcm
->tcm_handle
= flow
->common
.classid
;
612 tcm
->tcm_info
= flow
->q
->handle
;
614 nest
= nla_nest_start_noflag(skb
, TCA_OPTIONS
);
616 goto nla_put_failure
;
618 if (nla_put(skb
, TCA_ATM_HDR
, flow
->hdr_len
, flow
->hdr
))
619 goto nla_put_failure
;
621 struct sockaddr_atmpvc pvc
;
624 memset(&pvc
, 0, sizeof(pvc
));
625 pvc
.sap_family
= AF_ATMPVC
;
626 pvc
.sap_addr
.itf
= flow
->vcc
->dev
? flow
->vcc
->dev
->number
: -1;
627 pvc
.sap_addr
.vpi
= flow
->vcc
->vpi
;
628 pvc
.sap_addr
.vci
= flow
->vcc
->vci
;
629 if (nla_put(skb
, TCA_ATM_ADDR
, sizeof(pvc
), &pvc
))
630 goto nla_put_failure
;
631 state
= ATM_VF2VS(flow
->vcc
->flags
);
632 if (nla_put_u32(skb
, TCA_ATM_STATE
, state
))
633 goto nla_put_failure
;
636 if (nla_put_u32(skb
, TCA_ATM_EXCESS
, flow
->common
.classid
))
637 goto nla_put_failure
;
639 if (nla_put_u32(skb
, TCA_ATM_EXCESS
, 0))
640 goto nla_put_failure
;
642 return nla_nest_end(skb
, nest
);
645 nla_nest_cancel(skb
, nest
);
649 atm_tc_dump_class_stats(struct Qdisc
*sch
, unsigned long arg
,
652 struct atm_flow_data
*flow
= (struct atm_flow_data
*)arg
;
654 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch
),
655 d
, NULL
, &flow
->bstats
) < 0 ||
656 gnet_stats_copy_queue(d
, NULL
, &flow
->qstats
, flow
->q
->q
.qlen
) < 0)
662 static int atm_tc_dump(struct Qdisc
*sch
, struct sk_buff
*skb
)
667 static const struct Qdisc_class_ops atm_class_ops
= {
668 .graft
= atm_tc_graft
,
671 .change
= atm_tc_change
,
672 .delete = atm_tc_delete
,
674 .tcf_block
= atm_tc_tcf_block
,
675 .bind_tcf
= atm_tc_bind_filter
,
676 .unbind_tcf
= atm_tc_put
,
677 .dump
= atm_tc_dump_class
,
678 .dump_stats
= atm_tc_dump_class_stats
,
681 static struct Qdisc_ops atm_qdisc_ops __read_mostly
= {
682 .cl_ops
= &atm_class_ops
,
684 .priv_size
= sizeof(struct atm_qdisc_data
),
685 .enqueue
= atm_tc_enqueue
,
686 .dequeue
= atm_tc_dequeue
,
689 .reset
= atm_tc_reset
,
690 .destroy
= atm_tc_destroy
,
692 .owner
= THIS_MODULE
,
695 static int __init
atm_init(void)
697 return register_qdisc(&atm_qdisc_ops
);
700 static void __exit
atm_exit(void)
702 unregister_qdisc(&atm_qdisc_ops
);
705 module_init(atm_init
)
706 module_exit(atm_exit
)
707 MODULE_LICENSE("GPL");