1 // SPDX-License-Identifier: GPL-2.0-only
2 /* net/sched/sch_dsmark.c - Differentiated Services field marker */
4 /* Written 1998-2000 by Werner Almesberger, EPFL ICA */
7 #include <linux/module.h>
8 #include <linux/init.h>
9 #include <linux/slab.h>
10 #include <linux/types.h>
11 #include <linux/string.h>
12 #include <linux/errno.h>
13 #include <linux/skbuff.h>
14 #include <linux/rtnetlink.h>
15 #include <linux/bitops.h>
16 #include <net/pkt_sched.h>
17 #include <net/pkt_cls.h>
18 #include <net/dsfield.h>
19 #include <net/inet_ecn.h>
20 #include <asm/byteorder.h>
23 * classid class marking
24 * ------- ----- -------
28 * x:y y>0 y+1 use entry [y]
30 * x:indices-1 indices use entry [indices-1]
32 * x:y y+1 use entry [y & (indices-1)]
34 * 0xffff 0x10000 use entry [indices-1]
38 #define NO_DEFAULT_INDEX (1 << 16)
45 struct dsmark_qdisc_data
{
47 struct tcf_proto __rcu
*filter_list
;
48 struct tcf_block
*block
;
49 struct mask_value
*mv
;
52 u32 default_index
; /* index range is 0...0xffff */
53 #define DSMARK_EMBEDDED_SZ 16
54 struct mask_value embedded
[DSMARK_EMBEDDED_SZ
];
57 static inline int dsmark_valid_index(struct dsmark_qdisc_data
*p
, u16 index
)
59 return index
<= p
->indices
&& index
> 0;
62 /* ------------------------- Class/flow operations ------------------------- */
64 static int dsmark_graft(struct Qdisc
*sch
, unsigned long arg
,
65 struct Qdisc
*new, struct Qdisc
**old
,
66 struct netlink_ext_ack
*extack
)
68 struct dsmark_qdisc_data
*p
= qdisc_priv(sch
);
70 pr_debug("%s(sch %p,[qdisc %p],new %p,old %p)\n",
71 __func__
, sch
, p
, new, old
);
74 new = qdisc_create_dflt(sch
->dev_queue
, &pfifo_qdisc_ops
,
80 *old
= qdisc_replace(sch
, new, &p
->q
);
84 static struct Qdisc
*dsmark_leaf(struct Qdisc
*sch
, unsigned long arg
)
86 struct dsmark_qdisc_data
*p
= qdisc_priv(sch
);
90 static unsigned long dsmark_find(struct Qdisc
*sch
, u32 classid
)
92 return TC_H_MIN(classid
) + 1;
95 static unsigned long dsmark_bind_filter(struct Qdisc
*sch
,
96 unsigned long parent
, u32 classid
)
98 pr_debug("%s(sch %p,[qdisc %p],classid %x)\n",
99 __func__
, sch
, qdisc_priv(sch
), classid
);
101 return dsmark_find(sch
, classid
);
104 static void dsmark_unbind_filter(struct Qdisc
*sch
, unsigned long cl
)
108 static const struct nla_policy dsmark_policy
[TCA_DSMARK_MAX
+ 1] = {
109 [TCA_DSMARK_INDICES
] = { .type
= NLA_U16
},
110 [TCA_DSMARK_DEFAULT_INDEX
] = { .type
= NLA_U16
},
111 [TCA_DSMARK_SET_TC_INDEX
] = { .type
= NLA_FLAG
},
112 [TCA_DSMARK_MASK
] = { .type
= NLA_U8
},
113 [TCA_DSMARK_VALUE
] = { .type
= NLA_U8
},
116 static int dsmark_change(struct Qdisc
*sch
, u32 classid
, u32 parent
,
117 struct nlattr
**tca
, unsigned long *arg
,
118 struct netlink_ext_ack
*extack
)
120 struct dsmark_qdisc_data
*p
= qdisc_priv(sch
);
121 struct nlattr
*opt
= tca
[TCA_OPTIONS
];
122 struct nlattr
*tb
[TCA_DSMARK_MAX
+ 1];
125 pr_debug("%s(sch %p,[qdisc %p],classid %x,parent %x), arg 0x%lx\n",
126 __func__
, sch
, p
, classid
, parent
, *arg
);
128 if (!dsmark_valid_index(p
, *arg
)) {
136 err
= nla_parse_nested_deprecated(tb
, TCA_DSMARK_MAX
, opt
,
137 dsmark_policy
, NULL
);
141 if (tb
[TCA_DSMARK_VALUE
])
142 p
->mv
[*arg
- 1].value
= nla_get_u8(tb
[TCA_DSMARK_VALUE
]);
144 if (tb
[TCA_DSMARK_MASK
])
145 p
->mv
[*arg
- 1].mask
= nla_get_u8(tb
[TCA_DSMARK_MASK
]);
153 static int dsmark_delete(struct Qdisc
*sch
, unsigned long arg
)
155 struct dsmark_qdisc_data
*p
= qdisc_priv(sch
);
157 if (!dsmark_valid_index(p
, arg
))
160 p
->mv
[arg
- 1].mask
= 0xff;
161 p
->mv
[arg
- 1].value
= 0;
166 static void dsmark_walk(struct Qdisc
*sch
, struct qdisc_walker
*walker
)
168 struct dsmark_qdisc_data
*p
= qdisc_priv(sch
);
171 pr_debug("%s(sch %p,[qdisc %p],walker %p)\n",
172 __func__
, sch
, p
, walker
);
177 for (i
= 0; i
< p
->indices
; i
++) {
178 if (p
->mv
[i
].mask
== 0xff && !p
->mv
[i
].value
)
180 if (walker
->count
>= walker
->skip
) {
181 if (walker
->fn(sch
, i
+ 1, walker
) < 0) {
191 static struct tcf_block
*dsmark_tcf_block(struct Qdisc
*sch
, unsigned long cl
,
192 struct netlink_ext_ack
*extack
)
194 struct dsmark_qdisc_data
*p
= qdisc_priv(sch
);
199 /* --------------------------- Qdisc operations ---------------------------- */
201 static int dsmark_enqueue(struct sk_buff
*skb
, struct Qdisc
*sch
,
202 struct sk_buff
**to_free
)
204 unsigned int len
= qdisc_pkt_len(skb
);
205 struct dsmark_qdisc_data
*p
= qdisc_priv(sch
);
208 pr_debug("%s(skb %p,sch %p,[qdisc %p])\n", __func__
, skb
, sch
, p
);
210 if (p
->set_tc_index
) {
211 int wlen
= skb_network_offset(skb
);
213 switch (skb_protocol(skb
, true)) {
214 case htons(ETH_P_IP
):
215 wlen
+= sizeof(struct iphdr
);
216 if (!pskb_may_pull(skb
, wlen
) ||
217 skb_try_make_writable(skb
, wlen
))
220 skb
->tc_index
= ipv4_get_dsfield(ip_hdr(skb
))
224 case htons(ETH_P_IPV6
):
225 wlen
+= sizeof(struct ipv6hdr
);
226 if (!pskb_may_pull(skb
, wlen
) ||
227 skb_try_make_writable(skb
, wlen
))
230 skb
->tc_index
= ipv6_get_dsfield(ipv6_hdr(skb
))
239 if (TC_H_MAJ(skb
->priority
) == sch
->handle
)
240 skb
->tc_index
= TC_H_MIN(skb
->priority
);
242 struct tcf_result res
;
243 struct tcf_proto
*fl
= rcu_dereference_bh(p
->filter_list
);
244 int result
= tcf_classify(skb
, fl
, &res
, false);
246 pr_debug("result %d class 0x%04x\n", result
, res
.classid
);
249 #ifdef CONFIG_NET_CLS_ACT
253 __qdisc_drop(skb
, to_free
);
254 return NET_XMIT_SUCCESS
| __NET_XMIT_STOLEN
;
260 skb
->tc_index
= TC_H_MIN(res
.classid
);
264 if (p
->default_index
!= NO_DEFAULT_INDEX
)
265 skb
->tc_index
= p
->default_index
;
270 err
= qdisc_enqueue(skb
, p
->q
, to_free
);
271 if (err
!= NET_XMIT_SUCCESS
) {
272 if (net_xmit_drop_count(err
))
273 qdisc_qstats_drop(sch
);
277 sch
->qstats
.backlog
+= len
;
280 return NET_XMIT_SUCCESS
;
283 qdisc_drop(skb
, sch
, to_free
);
284 return NET_XMIT_SUCCESS
| __NET_XMIT_BYPASS
;
287 static struct sk_buff
*dsmark_dequeue(struct Qdisc
*sch
)
289 struct dsmark_qdisc_data
*p
= qdisc_priv(sch
);
293 pr_debug("%s(sch %p,[qdisc %p])\n", __func__
, sch
, p
);
295 skb
= qdisc_dequeue_peeked(p
->q
);
299 qdisc_bstats_update(sch
, skb
);
300 qdisc_qstats_backlog_dec(sch
, skb
);
303 index
= skb
->tc_index
& (p
->indices
- 1);
304 pr_debug("index %d->%d\n", skb
->tc_index
, index
);
306 switch (skb_protocol(skb
, true)) {
307 case htons(ETH_P_IP
):
308 ipv4_change_dsfield(ip_hdr(skb
), p
->mv
[index
].mask
,
311 case htons(ETH_P_IPV6
):
312 ipv6_change_dsfield(ipv6_hdr(skb
), p
->mv
[index
].mask
,
317 * Only complain if a change was actually attempted.
318 * This way, we can send non-IP traffic through dsmark
319 * and don't need yet another qdisc as a bypass.
321 if (p
->mv
[index
].mask
!= 0xff || p
->mv
[index
].value
)
322 pr_warn("%s: unsupported protocol %d\n",
323 __func__
, ntohs(skb_protocol(skb
, true)));
330 static struct sk_buff
*dsmark_peek(struct Qdisc
*sch
)
332 struct dsmark_qdisc_data
*p
= qdisc_priv(sch
);
334 pr_debug("%s(sch %p,[qdisc %p])\n", __func__
, sch
, p
);
336 return p
->q
->ops
->peek(p
->q
);
339 static int dsmark_init(struct Qdisc
*sch
, struct nlattr
*opt
,
340 struct netlink_ext_ack
*extack
)
342 struct dsmark_qdisc_data
*p
= qdisc_priv(sch
);
343 struct nlattr
*tb
[TCA_DSMARK_MAX
+ 1];
345 u32 default_index
= NO_DEFAULT_INDEX
;
349 pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__
, sch
, p
, opt
);
354 err
= tcf_block_get(&p
->block
, &p
->filter_list
, sch
, extack
);
358 err
= nla_parse_nested_deprecated(tb
, TCA_DSMARK_MAX
, opt
,
359 dsmark_policy
, NULL
);
364 if (!tb
[TCA_DSMARK_INDICES
])
366 indices
= nla_get_u16(tb
[TCA_DSMARK_INDICES
]);
368 if (hweight32(indices
) != 1)
371 if (tb
[TCA_DSMARK_DEFAULT_INDEX
])
372 default_index
= nla_get_u16(tb
[TCA_DSMARK_DEFAULT_INDEX
]);
374 if (indices
<= DSMARK_EMBEDDED_SZ
)
377 p
->mv
= kmalloc_array(indices
, sizeof(*p
->mv
), GFP_KERNEL
);
382 for (i
= 0; i
< indices
; i
++) {
383 p
->mv
[i
].mask
= 0xff;
386 p
->indices
= indices
;
387 p
->default_index
= default_index
;
388 p
->set_tc_index
= nla_get_flag(tb
[TCA_DSMARK_SET_TC_INDEX
]);
390 p
->q
= qdisc_create_dflt(sch
->dev_queue
, &pfifo_qdisc_ops
, sch
->handle
,
395 qdisc_hash_add(p
->q
, true);
397 pr_debug("%s: qdisc %p\n", __func__
, p
->q
);
404 static void dsmark_reset(struct Qdisc
*sch
)
406 struct dsmark_qdisc_data
*p
= qdisc_priv(sch
);
408 pr_debug("%s(sch %p,[qdisc %p])\n", __func__
, sch
, p
);
410 sch
->qstats
.backlog
= 0;
414 static void dsmark_destroy(struct Qdisc
*sch
)
416 struct dsmark_qdisc_data
*p
= qdisc_priv(sch
);
418 pr_debug("%s(sch %p,[qdisc %p])\n", __func__
, sch
, p
);
420 tcf_block_put(p
->block
);
422 if (p
->mv
!= p
->embedded
)
426 static int dsmark_dump_class(struct Qdisc
*sch
, unsigned long cl
,
427 struct sk_buff
*skb
, struct tcmsg
*tcm
)
429 struct dsmark_qdisc_data
*p
= qdisc_priv(sch
);
430 struct nlattr
*opts
= NULL
;
432 pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__
, sch
, p
, cl
);
434 if (!dsmark_valid_index(p
, cl
))
437 tcm
->tcm_handle
= TC_H_MAKE(TC_H_MAJ(sch
->handle
), cl
- 1);
438 tcm
->tcm_info
= p
->q
->handle
;
440 opts
= nla_nest_start_noflag(skb
, TCA_OPTIONS
);
442 goto nla_put_failure
;
443 if (nla_put_u8(skb
, TCA_DSMARK_MASK
, p
->mv
[cl
- 1].mask
) ||
444 nla_put_u8(skb
, TCA_DSMARK_VALUE
, p
->mv
[cl
- 1].value
))
445 goto nla_put_failure
;
447 return nla_nest_end(skb
, opts
);
450 nla_nest_cancel(skb
, opts
);
454 static int dsmark_dump(struct Qdisc
*sch
, struct sk_buff
*skb
)
456 struct dsmark_qdisc_data
*p
= qdisc_priv(sch
);
457 struct nlattr
*opts
= NULL
;
459 opts
= nla_nest_start_noflag(skb
, TCA_OPTIONS
);
461 goto nla_put_failure
;
462 if (nla_put_u16(skb
, TCA_DSMARK_INDICES
, p
->indices
))
463 goto nla_put_failure
;
465 if (p
->default_index
!= NO_DEFAULT_INDEX
&&
466 nla_put_u16(skb
, TCA_DSMARK_DEFAULT_INDEX
, p
->default_index
))
467 goto nla_put_failure
;
469 if (p
->set_tc_index
&&
470 nla_put_flag(skb
, TCA_DSMARK_SET_TC_INDEX
))
471 goto nla_put_failure
;
473 return nla_nest_end(skb
, opts
);
476 nla_nest_cancel(skb
, opts
);
480 static const struct Qdisc_class_ops dsmark_class_ops
= {
481 .graft
= dsmark_graft
,
484 .change
= dsmark_change
,
485 .delete = dsmark_delete
,
487 .tcf_block
= dsmark_tcf_block
,
488 .bind_tcf
= dsmark_bind_filter
,
489 .unbind_tcf
= dsmark_unbind_filter
,
490 .dump
= dsmark_dump_class
,
493 static struct Qdisc_ops dsmark_qdisc_ops __read_mostly
= {
495 .cl_ops
= &dsmark_class_ops
,
497 .priv_size
= sizeof(struct dsmark_qdisc_data
),
498 .enqueue
= dsmark_enqueue
,
499 .dequeue
= dsmark_dequeue
,
502 .reset
= dsmark_reset
,
503 .destroy
= dsmark_destroy
,
506 .owner
= THIS_MODULE
,
509 static int __init
dsmark_module_init(void)
511 return register_qdisc(&dsmark_qdisc_ops
);
514 static void __exit
dsmark_module_exit(void)
516 unregister_qdisc(&dsmark_qdisc_ops
);
519 module_init(dsmark_module_init
)
520 module_exit(dsmark_module_exit
)
522 MODULE_LICENSE("GPL");