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/init.h>
7 #include <linux/string.h>
8 #include <linux/errno.h>
9 #include <linux/skbuff.h>
10 #include <linux/atmdev.h>
11 #include <linux/atmclip.h>
12 #include <linux/rtnetlink.h>
13 #include <linux/file.h> /* for fput */
14 #include <net/netlink.h>
15 #include <net/pkt_sched.h>
17 extern struct socket
*sockfd_lookup(int fd
, int *err
); /* @@@ fix this */
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
*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 bstats
;
53 struct gnet_stats_queue qstats
;
54 struct atm_flow_data
*next
;
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 atm_flow_data
*flows
; /* NB: "link" is also on this
65 struct tasklet_struct task
; /* dequeue tasklet */
68 /* ------------------------- Class/flow operations ------------------------- */
70 static int find_flow(struct atm_qdisc_data
*qdisc
, struct atm_flow_data
*flow
)
72 struct atm_flow_data
*walk
;
74 pr_debug("find_flow(qdisc %p,flow %p)\n", qdisc
, flow
);
75 for (walk
= qdisc
->flows
; walk
; walk
= walk
->next
)
78 pr_debug("find_flow: not found\n");
82 static inline struct atm_flow_data
*lookup_flow(struct Qdisc
*sch
, u32 classid
)
84 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
85 struct atm_flow_data
*flow
;
87 for (flow
= p
->flows
; flow
; flow
= flow
->next
)
88 if (flow
->classid
== classid
)
93 static int atm_tc_graft(struct Qdisc
*sch
, unsigned long arg
,
94 struct Qdisc
*new, struct Qdisc
**old
)
96 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
97 struct atm_flow_data
*flow
= (struct atm_flow_data
*)arg
;
99 pr_debug("atm_tc_graft(sch %p,[qdisc %p],flow %p,new %p,old %p)\n",
100 sch
, p
, flow
, new, old
);
101 if (!find_flow(p
, flow
))
112 static struct Qdisc
*atm_tc_leaf(struct Qdisc
*sch
, unsigned long cl
)
114 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
116 pr_debug("atm_tc_leaf(sch %p,flow %p)\n", sch
, flow
);
117 return flow
? flow
->q
: NULL
;
120 static unsigned long atm_tc_get(struct Qdisc
*sch
, u32 classid
)
122 struct atm_qdisc_data
*p __maybe_unused
= qdisc_priv(sch
);
123 struct atm_flow_data
*flow
;
125 pr_debug("atm_tc_get(sch %p,[qdisc %p],classid %x)\n", sch
, p
, classid
);
126 flow
= lookup_flow(sch
, classid
);
129 pr_debug("atm_tc_get: flow %p\n", flow
);
130 return (unsigned long)flow
;
133 static unsigned long atm_tc_bind_filter(struct Qdisc
*sch
,
134 unsigned long parent
, u32 classid
)
136 return atm_tc_get(sch
, classid
);
140 * atm_tc_put handles all destructions, including the ones that are explicitly
141 * requested (atm_tc_destroy, etc.). The assumption here is that we never drop
142 * anything that still seems to be in use.
144 static void atm_tc_put(struct Qdisc
*sch
, unsigned long cl
)
146 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
147 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
148 struct atm_flow_data
**prev
;
150 pr_debug("atm_tc_put(sch %p,[qdisc %p],flow %p)\n", sch
, p
, flow
);
153 pr_debug("atm_tc_put: destroying\n");
154 for (prev
= &p
->flows
; *prev
; prev
= &(*prev
)->next
)
158 printk(KERN_CRIT
"atm_tc_put: class %p not found\n", flow
);
162 pr_debug("atm_tc_put: qdisc %p\n", flow
->q
);
163 qdisc_destroy(flow
->q
);
164 tcf_destroy_chain(&flow
->filter_list
);
166 pr_debug("atm_tc_put: f_count %ld\n",
167 file_count(flow
->sock
->file
));
168 flow
->vcc
->pop
= flow
->old_pop
;
169 sockfd_put(flow
->sock
);
172 atm_tc_put(sch
, (unsigned long)flow
->excess
);
173 if (flow
!= &p
->link
)
176 * If flow == &p->link, the qdisc no longer works at this point and
177 * needs to be removed. (By the caller of atm_tc_put.)
181 static void sch_atm_pop(struct atm_vcc
*vcc
, struct sk_buff
*skb
)
183 struct atm_qdisc_data
*p
= VCC2FLOW(vcc
)->parent
;
185 pr_debug("sch_atm_pop(vcc %p,skb %p,[qdisc %p])\n", vcc
, skb
, p
);
186 VCC2FLOW(vcc
)->old_pop(vcc
, skb
);
187 tasklet_schedule(&p
->task
);
190 static const u8 llc_oui_ip
[] = {
191 0xaa, /* DSAP: non-ISO */
192 0xaa, /* SSAP: non-ISO */
193 0x03, /* Ctrl: Unnumbered Information Command PDU */
194 0x00, /* OUI: EtherType */
197 }; /* Ethertype IP (0800) */
199 static const struct nla_policy atm_policy
[TCA_ATM_MAX
+ 1] = {
200 [TCA_ATM_FD
] = { .type
= NLA_U32
},
201 [TCA_ATM_EXCESS
] = { .type
= NLA_U32
},
204 static int atm_tc_change(struct Qdisc
*sch
, u32 classid
, u32 parent
,
205 struct nlattr
**tca
, unsigned long *arg
)
207 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
208 struct atm_flow_data
*flow
= (struct atm_flow_data
*)*arg
;
209 struct atm_flow_data
*excess
= NULL
;
210 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
211 struct nlattr
*tb
[TCA_ATM_MAX
+ 1];
213 int fd
, error
, hdr_len
;
216 pr_debug("atm_tc_change(sch %p,[qdisc %p],classid %x,parent %x,"
217 "flow %p,opt %p)\n", sch
, p
, classid
, parent
, flow
, opt
);
219 * The concept of parents doesn't apply for this qdisc.
221 if (parent
&& parent
!= TC_H_ROOT
&& parent
!= sch
->handle
)
224 * ATM classes cannot be changed. In order to change properties of the
225 * ATM connection, that socket needs to be modified directly (via the
226 * native ATM API. In order to send a flow to a different VC, the old
227 * class needs to be removed and a new one added. (This may be changed
235 error
= nla_parse_nested(tb
, TCA_ATM_MAX
, opt
, atm_policy
);
241 fd
= nla_get_u32(tb
[TCA_ATM_FD
]);
242 pr_debug("atm_tc_change: fd %d\n", fd
);
243 if (tb
[TCA_ATM_HDR
]) {
244 hdr_len
= nla_len(tb
[TCA_ATM_HDR
]);
245 hdr
= nla_data(tb
[TCA_ATM_HDR
]);
247 hdr_len
= RFC1483LLC_LEN
;
248 hdr
= NULL
; /* default LLC/SNAP for IP */
250 if (!tb
[TCA_ATM_EXCESS
])
253 excess
= (struct atm_flow_data
*)
254 atm_tc_get(sch
, nla_get_u32(tb
[TCA_ATM_EXCESS
]));
258 pr_debug("atm_tc_change: type %d, payload %d, hdr_len %d\n",
259 opt
->nla_type
, nla_len(opt
), hdr_len
);
260 sock
= sockfd_lookup(fd
, &error
);
262 return error
; /* f_count++ */
263 pr_debug("atm_tc_change: f_count %ld\n", file_count(sock
->file
));
264 if (sock
->ops
->family
!= PF_ATMSVC
&& sock
->ops
->family
!= PF_ATMPVC
) {
268 /* @@@ should check if the socket is really operational or we'll crash
271 if (TC_H_MAJ(classid
^ sch
->handle
)) {
272 pr_debug("atm_tc_change: classid mismatch\n");
276 if (find_flow(p
, flow
)) {
284 for (i
= 1; i
< 0x8000; i
++) {
285 classid
= TC_H_MAKE(sch
->handle
, 0x8000 | i
);
286 cl
= atm_tc_get(sch
, classid
);
292 pr_debug("atm_tc_change: new id %x\n", classid
);
293 flow
= kzalloc(sizeof(struct atm_flow_data
) + hdr_len
, GFP_KERNEL
);
294 pr_debug("atm_tc_change: flow %p\n", flow
);
299 flow
->filter_list
= NULL
;
300 flow
->q
= qdisc_create_dflt(qdisc_dev(sch
), sch
->dev_queue
,
301 &pfifo_qdisc_ops
, classid
);
303 flow
->q
= &noop_qdisc
;
304 pr_debug("atm_tc_change: qdisc %p\n", flow
->q
);
306 flow
->vcc
= ATM_SD(sock
); /* speedup */
307 flow
->vcc
->user_back
= flow
;
308 pr_debug("atm_tc_change: vcc %p\n", flow
->vcc
);
309 flow
->old_pop
= flow
->vcc
->pop
;
311 flow
->vcc
->pop
= sch_atm_pop
;
312 flow
->classid
= classid
;
314 flow
->excess
= excess
;
315 flow
->next
= p
->link
.next
;
317 flow
->hdr_len
= hdr_len
;
319 memcpy(flow
->hdr
, hdr
, hdr_len
);
321 memcpy(flow
->hdr
, llc_oui_ip
, sizeof(llc_oui_ip
));
322 *arg
= (unsigned long)flow
;
326 atm_tc_put(sch
, (unsigned long)excess
);
331 static int atm_tc_delete(struct Qdisc
*sch
, unsigned long arg
)
333 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
334 struct atm_flow_data
*flow
= (struct atm_flow_data
*)arg
;
336 pr_debug("atm_tc_delete(sch %p,[qdisc %p],flow %p)\n", sch
, p
, flow
);
337 if (!find_flow(qdisc_priv(sch
), flow
))
339 if (flow
->filter_list
|| flow
== &p
->link
)
342 * Reference count must be 2: one for "keepalive" (set at class
343 * creation), and one for the reference held when calling delete.
346 printk(KERN_ERR
"atm_tc_delete: flow->ref == %d\n", flow
->ref
);
350 return -EBUSY
; /* catch references via excess, etc. */
351 atm_tc_put(sch
, arg
);
355 static void atm_tc_walk(struct Qdisc
*sch
, struct qdisc_walker
*walker
)
357 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
358 struct atm_flow_data
*flow
;
360 pr_debug("atm_tc_walk(sch %p,[qdisc %p],walker %p)\n", sch
, p
, walker
);
363 for (flow
= p
->flows
; flow
; flow
= flow
->next
) {
364 if (walker
->count
>= walker
->skip
)
365 if (walker
->fn(sch
, (unsigned long)flow
, walker
) < 0) {
373 static struct tcf_proto
**atm_tc_find_tcf(struct Qdisc
*sch
, unsigned long cl
)
375 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
376 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
378 pr_debug("atm_tc_find_tcf(sch %p,[qdisc %p],flow %p)\n", sch
, p
, flow
);
379 return flow
? &flow
->filter_list
: &p
->link
.filter_list
;
382 /* --------------------------- Qdisc operations ---------------------------- */
384 static int atm_tc_enqueue(struct sk_buff
*skb
, struct Qdisc
*sch
)
386 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
387 struct atm_flow_data
*flow
= NULL
; /* @@@ */
388 struct tcf_result res
;
390 int ret
= NET_XMIT_POLICED
;
392 pr_debug("atm_tc_enqueue(skb %p,sch %p,[qdisc %p])\n", skb
, sch
, p
);
393 result
= TC_POLICE_OK
; /* be nice to gcc */
394 if (TC_H_MAJ(skb
->priority
) != sch
->handle
||
395 !(flow
= (struct atm_flow_data
*)atm_tc_get(sch
, skb
->priority
)))
396 for (flow
= p
->flows
; flow
; flow
= flow
->next
)
397 if (flow
->filter_list
) {
398 result
= tc_classify_compat(skb
,
403 flow
= (struct atm_flow_data
*)res
.class;
405 flow
= lookup_flow(sch
, res
.classid
);
412 ATM_SKB(skb
)->atm_options
= flow
->vcc
->atm_options
;
413 /*@@@ looks good ... but it's not supposed to work :-) */
414 #ifdef CONFIG_NET_CLS_ACT
419 return NET_XMIT_SUCCESS
| __NET_XMIT_STOLEN
;
423 case TC_POLICE_RECLASSIFY
:
427 ATM_SKB(skb
)->atm_options
|= ATM_ATMOPT_CLP
;
433 ret
= qdisc_enqueue(skb
, flow
->q
);
436 if (net_xmit_drop_count(ret
)) {
439 flow
->qstats
.drops
++;
443 sch
->bstats
.bytes
+= qdisc_pkt_len(skb
);
444 sch
->bstats
.packets
++;
445 flow
->bstats
.bytes
+= qdisc_pkt_len(skb
);
446 flow
->bstats
.packets
++;
448 * Okay, this may seem weird. We pretend we've dropped the packet if
449 * it goes via ATM. The reason for this is that the outer qdisc
450 * expects to be able to q->dequeue the packet later on if we return
451 * success at this place. Also, sch->q.qdisc needs to reflect whether
452 * there is a packet egligible for dequeuing or not. Note that the
453 * statistics of the outer qdisc are necessarily wrong because of all
454 * this. There's currently no correct solution for this.
456 if (flow
== &p
->link
) {
460 tasklet_schedule(&p
->task
);
461 return NET_XMIT_SUCCESS
| __NET_XMIT_BYPASS
;
465 * Dequeue packets and send them over ATM. Note that we quite deliberately
466 * avoid checking net_device's flow control here, simply because sch_atm
467 * uses its own channels, which have nothing to do with any CLIP/LANE/or
468 * non-ATM interfaces.
471 static void sch_atm_dequeue(unsigned long data
)
473 struct Qdisc
*sch
= (struct Qdisc
*)data
;
474 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
475 struct atm_flow_data
*flow
;
478 pr_debug("sch_atm_dequeue(sch %p,[qdisc %p])\n", sch
, p
);
479 for (flow
= p
->link
.next
; flow
; flow
= flow
->next
)
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 pr_debug("atm_tc_dequeue: sending on class %p\n", flow
);
493 /* remove any LL header somebody else has attached */
494 skb_pull(skb
, skb_network_offset(skb
));
495 if (skb_headroom(skb
) < flow
->hdr_len
) {
498 new = skb_realloc_headroom(skb
, flow
->hdr_len
);
504 pr_debug("sch_atm_dequeue: ip %p, data %p\n",
505 skb_network_header(skb
), skb
->data
);
506 ATM_SKB(skb
)->vcc
= flow
->vcc
;
507 memcpy(skb_push(skb
, flow
->hdr_len
), flow
->hdr
,
509 atomic_add(skb
->truesize
,
510 &sk_atm(flow
->vcc
)->sk_wmem_alloc
);
511 /* atm.atm_options are already set by atm_tc_enqueue */
512 flow
->vcc
->send(flow
->vcc
, skb
);
516 static struct sk_buff
*atm_tc_dequeue(struct Qdisc
*sch
)
518 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
521 pr_debug("atm_tc_dequeue(sch %p,[qdisc %p])\n", sch
, p
);
522 tasklet_schedule(&p
->task
);
523 skb
= qdisc_dequeue_peeked(p
->link
.q
);
529 static struct sk_buff
*atm_tc_peek(struct Qdisc
*sch
)
531 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
533 pr_debug("atm_tc_peek(sch %p,[qdisc %p])\n", sch
, p
);
535 return p
->link
.q
->ops
->peek(p
->link
.q
);
538 static unsigned int atm_tc_drop(struct Qdisc
*sch
)
540 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
541 struct atm_flow_data
*flow
;
544 pr_debug("atm_tc_drop(sch %p,[qdisc %p])\n", sch
, p
);
545 for (flow
= p
->flows
; flow
; flow
= flow
->next
)
546 if (flow
->q
->ops
->drop
&& (len
= flow
->q
->ops
->drop(flow
->q
)))
551 static int atm_tc_init(struct Qdisc
*sch
, struct nlattr
*opt
)
553 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
555 pr_debug("atm_tc_init(sch %p,[qdisc %p],opt %p)\n", sch
, p
, opt
);
557 p
->link
.q
= qdisc_create_dflt(qdisc_dev(sch
), sch
->dev_queue
,
558 &pfifo_qdisc_ops
, sch
->handle
);
560 p
->link
.q
= &noop_qdisc
;
561 pr_debug("atm_tc_init: link (%p) qdisc %p\n", &p
->link
, p
->link
.q
);
562 p
->link
.filter_list
= NULL
;
565 p
->link
.classid
= sch
->handle
;
568 tasklet_init(&p
->task
, sch_atm_dequeue
, (unsigned long)sch
);
572 static void atm_tc_reset(struct Qdisc
*sch
)
574 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
575 struct atm_flow_data
*flow
;
577 pr_debug("atm_tc_reset(sch %p,[qdisc %p])\n", sch
, p
);
578 for (flow
= p
->flows
; flow
; flow
= flow
->next
)
579 qdisc_reset(flow
->q
);
583 static void atm_tc_destroy(struct Qdisc
*sch
)
585 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
586 struct atm_flow_data
*flow
;
588 pr_debug("atm_tc_destroy(sch %p,[qdisc %p])\n", sch
, p
);
589 for (flow
= p
->flows
; flow
; flow
= flow
->next
)
590 tcf_destroy_chain(&flow
->filter_list
);
593 while ((flow
= p
->flows
)) {
595 printk(KERN_ERR
"atm_destroy: %p->ref = %d\n", flow
,
597 atm_tc_put(sch
, (unsigned long)flow
);
598 if (p
->flows
== flow
) {
599 printk(KERN_ERR
"atm_destroy: putting flow %p didn't "
601 p
->flows
= flow
->next
; /* brute force */
605 tasklet_kill(&p
->task
);
608 static int atm_tc_dump_class(struct Qdisc
*sch
, unsigned long cl
,
609 struct sk_buff
*skb
, struct tcmsg
*tcm
)
611 struct atm_qdisc_data
*p
= qdisc_priv(sch
);
612 struct atm_flow_data
*flow
= (struct atm_flow_data
*)cl
;
615 pr_debug("atm_tc_dump_class(sch %p,[qdisc %p],flow %p,skb %p,tcm %p)\n",
616 sch
, p
, flow
, skb
, tcm
);
617 if (!find_flow(p
, flow
))
619 tcm
->tcm_handle
= flow
->classid
;
620 tcm
->tcm_info
= flow
->q
->handle
;
622 nest
= nla_nest_start(skb
, TCA_OPTIONS
);
624 goto nla_put_failure
;
626 NLA_PUT(skb
, TCA_ATM_HDR
, flow
->hdr_len
, flow
->hdr
);
628 struct sockaddr_atmpvc pvc
;
631 pvc
.sap_family
= AF_ATMPVC
;
632 pvc
.sap_addr
.itf
= flow
->vcc
->dev
? flow
->vcc
->dev
->number
: -1;
633 pvc
.sap_addr
.vpi
= flow
->vcc
->vpi
;
634 pvc
.sap_addr
.vci
= flow
->vcc
->vci
;
635 NLA_PUT(skb
, TCA_ATM_ADDR
, sizeof(pvc
), &pvc
);
636 state
= ATM_VF2VS(flow
->vcc
->flags
);
637 NLA_PUT_U32(skb
, TCA_ATM_STATE
, state
);
640 NLA_PUT_U32(skb
, TCA_ATM_EXCESS
, flow
->classid
);
642 NLA_PUT_U32(skb
, TCA_ATM_EXCESS
, 0);
645 nla_nest_end(skb
, nest
);
649 nla_nest_cancel(skb
, nest
);
653 atm_tc_dump_class_stats(struct Qdisc
*sch
, unsigned long arg
,
656 struct atm_flow_data
*flow
= (struct atm_flow_data
*)arg
;
658 flow
->qstats
.qlen
= flow
->q
->q
.qlen
;
660 if (gnet_stats_copy_basic(d
, &flow
->bstats
) < 0 ||
661 gnet_stats_copy_queue(d
, &flow
->qstats
) < 0)
667 static int atm_tc_dump(struct Qdisc
*sch
, struct sk_buff
*skb
)
672 static const struct Qdisc_class_ops atm_class_ops
= {
673 .graft
= atm_tc_graft
,
677 .change
= atm_tc_change
,
678 .delete = atm_tc_delete
,
680 .tcf_chain
= atm_tc_find_tcf
,
681 .bind_tcf
= atm_tc_bind_filter
,
682 .unbind_tcf
= atm_tc_put
,
683 .dump
= atm_tc_dump_class
,
684 .dump_stats
= atm_tc_dump_class_stats
,
687 static struct Qdisc_ops atm_qdisc_ops __read_mostly
= {
688 .cl_ops
= &atm_class_ops
,
690 .priv_size
= sizeof(struct atm_qdisc_data
),
691 .enqueue
= atm_tc_enqueue
,
692 .dequeue
= atm_tc_dequeue
,
696 .reset
= atm_tc_reset
,
697 .destroy
= atm_tc_destroy
,
699 .owner
= THIS_MODULE
,
702 static int __init
atm_init(void)
704 return register_qdisc(&atm_qdisc_ops
);
707 static void __exit
atm_exit(void)
709 unregister_qdisc(&atm_qdisc_ops
);
712 module_init(atm_init
)
713 module_exit(atm_exit
)
714 MODULE_LICENSE("GPL");