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 atm_flow_data
*next
;
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 atm_flow_data
*flows
; /* NB: "link" is also on this
66 struct tasklet_struct task
; /* dequeue tasklet */
69 /* ------------------------- Class/flow operations ------------------------- */
71 static int find_flow(struct atm_qdisc_data
*qdisc
, struct atm_flow_data
*flow
)
73 struct atm_flow_data
*walk
;
75 pr_debug("find_flow(qdisc %p,flow %p)\n", qdisc
, flow
);
76 for (walk
= qdisc
->flows
; walk
; walk
= walk
->next
)
79 pr_debug("find_flow: not found\n");
83 static inline struct atm_flow_data
*lookup_flow(struct Qdisc
*sch
, u32 classid
)
85 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
86 struct atm_flow_data
*flow
;
88 for (flow
= p
->flows
; flow
; flow
= flow
->next
)
89 if (flow
->classid
== classid
)
94 static int atm_tc_graft(struct Qdisc
*sch
, unsigned long arg
,
95 struct Qdisc
*new, struct Qdisc
**old
)
97 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
98 struct atm_flow_data
*flow
= (struct atm_flow_data
*)arg
;
100 pr_debug("atm_tc_graft(sch %p,[qdisc %p],flow %p,new %p,old %p)\n",
101 sch
, p
, flow
, new, old
);
102 if (!find_flow(p
, flow
))
113 static struct Qdisc
*atm_tc_leaf(struct Qdisc
*sch
, unsigned long cl
)
115 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
117 pr_debug("atm_tc_leaf(sch %p,flow %p)\n", sch
, flow
);
118 return flow
? flow
->q
: NULL
;
121 static unsigned long atm_tc_get(struct Qdisc
*sch
, u32 classid
)
123 struct atm_qdisc_data
*p __maybe_unused
= qdisc_priv(sch
);
124 struct atm_flow_data
*flow
;
126 pr_debug("atm_tc_get(sch %p,[qdisc %p],classid %x)\n", sch
, p
, classid
);
127 flow
= lookup_flow(sch
, classid
);
130 pr_debug("atm_tc_get: flow %p\n", flow
);
131 return (unsigned long)flow
;
134 static unsigned long atm_tc_bind_filter(struct Qdisc
*sch
,
135 unsigned long parent
, u32 classid
)
137 return atm_tc_get(sch
, classid
);
141 * atm_tc_put handles all destructions, including the ones that are explicitly
142 * requested (atm_tc_destroy, etc.). The assumption here is that we never drop
143 * anything that still seems to be in use.
145 static void atm_tc_put(struct Qdisc
*sch
, unsigned long cl
)
147 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
148 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
149 struct atm_flow_data
**prev
;
151 pr_debug("atm_tc_put(sch %p,[qdisc %p],flow %p)\n", sch
, p
, flow
);
154 pr_debug("atm_tc_put: destroying\n");
155 for (prev
= &p
->flows
; *prev
; prev
= &(*prev
)->next
)
159 printk(KERN_CRIT
"atm_tc_put: class %p not found\n", flow
);
163 pr_debug("atm_tc_put: qdisc %p\n", flow
->q
);
164 qdisc_destroy(flow
->q
);
165 tcf_destroy_chain(&flow
->filter_list
);
167 pr_debug("atm_tc_put: f_count %ld\n",
168 file_count(flow
->sock
->file
));
169 flow
->vcc
->pop
= flow
->old_pop
;
170 sockfd_put(flow
->sock
);
173 atm_tc_put(sch
, (unsigned long)flow
->excess
);
174 if (flow
!= &p
->link
)
177 * If flow == &p->link, the qdisc no longer works at this point and
178 * needs to be removed. (By the caller of atm_tc_put.)
182 static void sch_atm_pop(struct atm_vcc
*vcc
, struct sk_buff
*skb
)
184 struct atm_qdisc_data
*p
= VCC2FLOW(vcc
)->parent
;
186 pr_debug("sch_atm_pop(vcc %p,skb %p,[qdisc %p])\n", vcc
, skb
, p
);
187 VCC2FLOW(vcc
)->old_pop(vcc
, skb
);
188 tasklet_schedule(&p
->task
);
191 static const u8 llc_oui_ip
[] = {
192 0xaa, /* DSAP: non-ISO */
193 0xaa, /* SSAP: non-ISO */
194 0x03, /* Ctrl: Unnumbered Information Command PDU */
195 0x00, /* OUI: EtherType */
198 }; /* Ethertype IP (0800) */
200 static const struct nla_policy atm_policy
[TCA_ATM_MAX
+ 1] = {
201 [TCA_ATM_FD
] = { .type
= NLA_U32
},
202 [TCA_ATM_EXCESS
] = { .type
= NLA_U32
},
205 static int atm_tc_change(struct Qdisc
*sch
, u32 classid
, u32 parent
,
206 struct nlattr
**tca
, unsigned long *arg
)
208 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
209 struct atm_flow_data
*flow
= (struct atm_flow_data
*)*arg
;
210 struct atm_flow_data
*excess
= NULL
;
211 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
212 struct nlattr
*tb
[TCA_ATM_MAX
+ 1];
214 int fd
, error
, hdr_len
;
217 pr_debug("atm_tc_change(sch %p,[qdisc %p],classid %x,parent %x,"
218 "flow %p,opt %p)\n", sch
, p
, classid
, parent
, flow
, opt
);
220 * The concept of parents doesn't apply for this qdisc.
222 if (parent
&& parent
!= TC_H_ROOT
&& parent
!= sch
->handle
)
225 * ATM classes cannot be changed. In order to change properties of the
226 * ATM connection, that socket needs to be modified directly (via the
227 * native ATM API. In order to send a flow to a different VC, the old
228 * class needs to be removed and a new one added. (This may be changed
236 error
= nla_parse_nested(tb
, TCA_ATM_MAX
, opt
, atm_policy
);
242 fd
= nla_get_u32(tb
[TCA_ATM_FD
]);
243 pr_debug("atm_tc_change: fd %d\n", fd
);
244 if (tb
[TCA_ATM_HDR
]) {
245 hdr_len
= nla_len(tb
[TCA_ATM_HDR
]);
246 hdr
= nla_data(tb
[TCA_ATM_HDR
]);
248 hdr_len
= RFC1483LLC_LEN
;
249 hdr
= NULL
; /* default LLC/SNAP for IP */
251 if (!tb
[TCA_ATM_EXCESS
])
254 excess
= (struct atm_flow_data
*)
255 atm_tc_get(sch
, nla_get_u32(tb
[TCA_ATM_EXCESS
]));
259 pr_debug("atm_tc_change: type %d, payload %d, hdr_len %d\n",
260 opt
->nla_type
, nla_len(opt
), hdr_len
);
261 sock
= sockfd_lookup(fd
, &error
);
263 return error
; /* f_count++ */
264 pr_debug("atm_tc_change: f_count %ld\n", file_count(sock
->file
));
265 if (sock
->ops
->family
!= PF_ATMSVC
&& sock
->ops
->family
!= PF_ATMPVC
) {
269 /* @@@ should check if the socket is really operational or we'll crash
272 if (TC_H_MAJ(classid
^ sch
->handle
)) {
273 pr_debug("atm_tc_change: classid mismatch\n");
277 if (find_flow(p
, flow
)) {
285 for (i
= 1; i
< 0x8000; i
++) {
286 classid
= TC_H_MAKE(sch
->handle
, 0x8000 | i
);
287 cl
= atm_tc_get(sch
, classid
);
293 pr_debug("atm_tc_change: new id %x\n", classid
);
294 flow
= kzalloc(sizeof(struct atm_flow_data
) + hdr_len
, GFP_KERNEL
);
295 pr_debug("atm_tc_change: flow %p\n", flow
);
300 flow
->filter_list
= NULL
;
301 flow
->q
= qdisc_create_dflt(qdisc_dev(sch
), sch
->dev_queue
,
302 &pfifo_qdisc_ops
, classid
);
304 flow
->q
= &noop_qdisc
;
305 pr_debug("atm_tc_change: qdisc %p\n", flow
->q
);
307 flow
->vcc
= ATM_SD(sock
); /* speedup */
308 flow
->vcc
->user_back
= flow
;
309 pr_debug("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
= qdisc_priv(sch
);
335 struct atm_flow_data
*flow
= (struct atm_flow_data
*)arg
;
337 pr_debug("atm_tc_delete(sch %p,[qdisc %p],flow %p)\n", sch
, p
, flow
);
338 if (!find_flow(qdisc_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
= qdisc_priv(sch
);
359 struct atm_flow_data
*flow
;
361 pr_debug("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
= qdisc_priv(sch
);
377 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
379 pr_debug("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
= qdisc_priv(sch
);
388 struct atm_flow_data
*flow
= NULL
; /* @@@ */
389 struct tcf_result res
;
391 int ret
= NET_XMIT_POLICED
;
393 pr_debug("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
| __NET_XMIT_STOLEN
;
424 case TC_POLICE_RECLASSIFY
:
428 ATM_SKB(skb
)->atm_options
|= ATM_ATMOPT_CLP
;
434 ret
= qdisc_enqueue(skb
, flow
->q
);
437 if (net_xmit_drop_count(ret
)) {
440 flow
->qstats
.drops
++;
444 sch
->bstats
.bytes
+= qdisc_pkt_len(skb
);
445 sch
->bstats
.packets
++;
446 flow
->bstats
.bytes
+= qdisc_pkt_len(skb
);
447 flow
->bstats
.packets
++;
449 * Okay, this may seem weird. We pretend we've dropped the packet if
450 * it goes via ATM. The reason for this is that the outer qdisc
451 * expects to be able to q->dequeue the packet later on if we return
452 * success at this place. Also, sch->q.qdisc needs to reflect whether
453 * there is a packet egligible for dequeuing or not. Note that the
454 * statistics of the outer qdisc are necessarily wrong because of all
455 * this. There's currently no correct solution for this.
457 if (flow
== &p
->link
) {
461 tasklet_schedule(&p
->task
);
462 return NET_XMIT_SUCCESS
| __NET_XMIT_BYPASS
;
466 * Dequeue packets and send them over ATM. Note that we quite deliberately
467 * avoid checking net_device's flow control here, simply because sch_atm
468 * uses its own channels, which have nothing to do with any CLIP/LANE/or
469 * non-ATM interfaces.
472 static void sch_atm_dequeue(unsigned long data
)
474 struct Qdisc
*sch
= (struct Qdisc
*)data
;
475 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
476 struct atm_flow_data
*flow
;
479 pr_debug("sch_atm_dequeue(sch %p,[qdisc %p])\n", sch
, p
);
480 for (flow
= p
->link
.next
; flow
; flow
= flow
->next
)
482 * If traffic is properly shaped, this won't generate nasty
483 * little bursts. Otherwise, it may ... (but that's okay)
485 while ((skb
= flow
->q
->ops
->peek(flow
->q
))) {
486 if (!atm_may_send(flow
->vcc
, skb
->truesize
))
489 skb
= qdisc_dequeue_peeked(flow
->q
);
493 pr_debug("atm_tc_dequeue: sending on class %p\n", flow
);
494 /* remove any LL header somebody else has attached */
495 skb_pull(skb
, skb_network_offset(skb
));
496 if (skb_headroom(skb
) < flow
->hdr_len
) {
499 new = skb_realloc_headroom(skb
, flow
->hdr_len
);
505 pr_debug("sch_atm_dequeue: ip %p, data %p\n",
506 skb_network_header(skb
), skb
->data
);
507 ATM_SKB(skb
)->vcc
= flow
->vcc
;
508 memcpy(skb_push(skb
, flow
->hdr_len
), flow
->hdr
,
510 atomic_add(skb
->truesize
,
511 &sk_atm(flow
->vcc
)->sk_wmem_alloc
);
512 /* atm.atm_options are already set by atm_tc_enqueue */
513 flow
->vcc
->send(flow
->vcc
, skb
);
517 static struct sk_buff
*atm_tc_dequeue(struct Qdisc
*sch
)
519 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
522 pr_debug("atm_tc_dequeue(sch %p,[qdisc %p])\n", sch
, p
);
523 tasklet_schedule(&p
->task
);
524 skb
= qdisc_dequeue_peeked(p
->link
.q
);
530 static struct sk_buff
*atm_tc_peek(struct Qdisc
*sch
)
532 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
534 pr_debug("atm_tc_peek(sch %p,[qdisc %p])\n", sch
, p
);
536 return p
->link
.q
->ops
->peek(p
->link
.q
);
539 static unsigned int atm_tc_drop(struct Qdisc
*sch
)
541 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
542 struct atm_flow_data
*flow
;
545 pr_debug("atm_tc_drop(sch %p,[qdisc %p])\n", sch
, p
);
546 for (flow
= p
->flows
; flow
; flow
= flow
->next
)
547 if (flow
->q
->ops
->drop
&& (len
= flow
->q
->ops
->drop(flow
->q
)))
552 static int atm_tc_init(struct Qdisc
*sch
, struct nlattr
*opt
)
554 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
556 pr_debug("atm_tc_init(sch %p,[qdisc %p],opt %p)\n", sch
, p
, opt
);
558 p
->link
.q
= qdisc_create_dflt(qdisc_dev(sch
), sch
->dev_queue
,
559 &pfifo_qdisc_ops
, sch
->handle
);
561 p
->link
.q
= &noop_qdisc
;
562 pr_debug("atm_tc_init: link (%p) qdisc %p\n", &p
->link
, p
->link
.q
);
563 p
->link
.filter_list
= NULL
;
566 p
->link
.classid
= sch
->handle
;
569 tasklet_init(&p
->task
, sch_atm_dequeue
, (unsigned long)sch
);
573 static void atm_tc_reset(struct Qdisc
*sch
)
575 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
576 struct atm_flow_data
*flow
;
578 pr_debug("atm_tc_reset(sch %p,[qdisc %p])\n", sch
, p
);
579 for (flow
= p
->flows
; flow
; flow
= flow
->next
)
580 qdisc_reset(flow
->q
);
584 static void atm_tc_destroy(struct Qdisc
*sch
)
586 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
587 struct atm_flow_data
*flow
;
589 pr_debug("atm_tc_destroy(sch %p,[qdisc %p])\n", sch
, p
);
590 for (flow
= p
->flows
; flow
; flow
= flow
->next
)
591 tcf_destroy_chain(&flow
->filter_list
);
594 while ((flow
= p
->flows
)) {
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
= qdisc_priv(sch
);
613 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
616 pr_debug("atm_tc_dump_class(sch %p,[qdisc %p],flow %p,skb %p,tcm %p)\n",
617 sch
, p
, flow
, skb
, tcm
);
618 if (!find_flow(p
, flow
))
620 tcm
->tcm_handle
= flow
->classid
;
621 tcm
->tcm_info
= flow
->q
->handle
;
623 nest
= nla_nest_start(skb
, TCA_OPTIONS
);
625 goto nla_put_failure
;
627 NLA_PUT(skb
, TCA_ATM_HDR
, flow
->hdr_len
, flow
->hdr
);
629 struct sockaddr_atmpvc pvc
;
632 pvc
.sap_family
= AF_ATMPVC
;
633 pvc
.sap_addr
.itf
= flow
->vcc
->dev
? flow
->vcc
->dev
->number
: -1;
634 pvc
.sap_addr
.vpi
= flow
->vcc
->vpi
;
635 pvc
.sap_addr
.vci
= flow
->vcc
->vci
;
636 NLA_PUT(skb
, TCA_ATM_ADDR
, sizeof(pvc
), &pvc
);
637 state
= ATM_VF2VS(flow
->vcc
->flags
);
638 NLA_PUT_U32(skb
, TCA_ATM_STATE
, state
);
641 NLA_PUT_U32(skb
, TCA_ATM_EXCESS
, flow
->classid
);
643 NLA_PUT_U32(skb
, TCA_ATM_EXCESS
, 0);
646 nla_nest_end(skb
, nest
);
650 nla_nest_cancel(skb
, nest
);
654 atm_tc_dump_class_stats(struct Qdisc
*sch
, unsigned long arg
,
657 struct atm_flow_data
*flow
= (struct atm_flow_data
*)arg
;
659 flow
->qstats
.qlen
= flow
->q
->q
.qlen
;
661 if (gnet_stats_copy_basic(d
, &flow
->bstats
) < 0 ||
662 gnet_stats_copy_queue(d
, &flow
->qstats
) < 0)
668 static int atm_tc_dump(struct Qdisc
*sch
, struct sk_buff
*skb
)
673 static const struct Qdisc_class_ops atm_class_ops
= {
674 .graft
= atm_tc_graft
,
678 .change
= atm_tc_change
,
679 .delete = atm_tc_delete
,
681 .tcf_chain
= atm_tc_find_tcf
,
682 .bind_tcf
= atm_tc_bind_filter
,
683 .unbind_tcf
= atm_tc_put
,
684 .dump
= atm_tc_dump_class
,
685 .dump_stats
= atm_tc_dump_class_stats
,
688 static struct Qdisc_ops atm_qdisc_ops __read_mostly
= {
689 .cl_ops
= &atm_class_ops
,
691 .priv_size
= sizeof(struct atm_qdisc_data
),
692 .enqueue
= atm_tc_enqueue
,
693 .dequeue
= atm_tc_dequeue
,
697 .reset
= atm_tc_reset
,
698 .destroy
= atm_tc_destroy
,
700 .owner
= THIS_MODULE
,
703 static int __init
atm_init(void)
705 return register_qdisc(&atm_qdisc_ops
);
708 static void __exit
atm_exit(void)
710 unregister_qdisc(&atm_qdisc_ops
);
713 module_init(atm_init
)
714 module_exit(atm_exit
)
715 MODULE_LICENSE("GPL");