Skip to content

Commit c129412

Browse files
wenxukuba-moo
wenxu
authored and
kuba-moo
committed
net/sched: sch_frag: add generic packet fragment support.
Currently kernel tc subsystem can do conntrack in cat_ct. But when several fragment packets go through the act_ct, function tcf_ct_handle_fragments will defrag the packets to a big one. But the last action will redirect mirred to a device which maybe lead the reassembly big packet over the mtu of target device. This patch add support for a xmit hook to mirred, that gets executed before xmiting the packet. Then, when act_ct gets loaded, it configs that hook. The frag xmit hook maybe reused by other modules. Signed-off-by: wenxu <[email protected]> Acked-by: Cong Wang <[email protected]> Acked-by: Jamal Hadi Salim <[email protected]> Signed-off-by: Jakub Kicinski <[email protected]>
1 parent fa6d639 commit c129412

File tree

7 files changed

+179
-1
lines changed

7 files changed

+179
-1
lines changed

‎include/net/act_api.h‎

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -239,6 +239,12 @@ int tcf_action_check_ctrlact(int action, struct tcf_proto *tp,
239239
struct netlink_ext_ack *newchain);
240240
struct tcf_chain *tcf_action_set_ctrlact(struct tc_action *a, int action,
241241
struct tcf_chain *newchain);
242+
243+
#ifdef CONFIG_INET
244+
DECLARE_STATIC_KEY_FALSE(tcf_frag_xmit_count);
245+
#endif
246+
247+
int tcf_dev_queue_xmit(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
242248
#endif /* CONFIG_NET_CLS_ACT */
243249

244250
static inline void tcf_action_stats_update(struct tc_action *a, u64 bytes,

‎include/net/sch_generic.h‎

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1281,4 +1281,6 @@ void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
12811281
void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
12821282
struct tcf_block *block);
12831283

1284+
int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb));
1285+
12841286
#endif

‎net/sched/Makefile‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -5,6 +5,7 @@
55

66
obj-y := sch_generic.o sch_mq.o
77

8+
obj-$(CONFIG_INET) += sch_frag.o
89
obj-$(CONFIG_NET_SCHED) += sch_api.o sch_blackhole.o
910
obj-$(CONFIG_NET_CLS) += cls_api.o
1011
obj-$(CONFIG_NET_CLS_ACT) += act_api.o

‎net/sched/act_api.c‎

Lines changed: 16 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -22,6 +22,22 @@
2222
#include <net/act_api.h>
2323
#include <net/netlink.h>
2424

25+
#ifdef CONFIG_INET
26+
DEFINE_STATIC_KEY_FALSE(tcf_frag_xmit_count);
27+
EXPORT_SYMBOL_GPL(tcf_frag_xmit_count);
28+
#endif
29+
30+
int tcf_dev_queue_xmit(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb))
31+
{
32+
#ifdef CONFIG_INET
33+
if (static_branch_unlikely(&tcf_frag_xmit_count))
34+
return sch_frag_xmit_hook(skb, xmit);
35+
#endif
36+
37+
return xmit(skb);
38+
}
39+
EXPORT_SYMBOL_GPL(tcf_dev_queue_xmit);
40+
2541
static void tcf_action_goto_chain_exec(const struct tc_action *a,
2642
struct tcf_result *res)
2743
{

‎net/sched/act_ct.c‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1541,6 +1541,8 @@ static int __init ct_init_module(void)
15411541
if (err)
15421542
goto err_register;
15431543

1544+
static_branch_inc(&tcf_frag_xmit_count);
1545+
15441546
return 0;
15451547

15461548
err_register:
@@ -1552,6 +1554,7 @@ static int __init ct_init_module(void)
15521554

15531555
static void __exit ct_cleanup_module(void)
15541556
{
1557+
static_branch_dec(&tcf_frag_xmit_count);
15551558
tcf_unregister_action(&act_ct_ops, &ct_net_ops);
15561559
tcf_ct_flow_tables_uninit();
15571560
destroy_workqueue(act_ct_wq);

‎net/sched/act_mirred.c‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -210,7 +210,7 @@ static int tcf_mirred_forward(bool want_ingress, struct sk_buff *skb)
210210
int err;
211211

212212
if (!want_ingress)
213-
err = dev_queue_xmit(skb);
213+
err = tcf_dev_queue_xmit(skb, dev_queue_xmit);
214214
else
215215
err = netif_receive_skb(skb);
216216

‎net/sched/sch_frag.c‎

Lines changed: 150 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,150 @@
1+
// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
2+
#include <net/netlink.h>
3+
#include <net/sch_generic.h>
4+
#include <net/dst.h>
5+
#include <net/ip.h>
6+
#include <net/ip6_fib.h>
7+
8+
struct sch_frag_data {
9+
unsigned long dst;
10+
struct qdisc_skb_cb cb;
11+
__be16 inner_protocol;
12+
u16 vlan_tci;
13+
__be16 vlan_proto;
14+
unsigned int l2_len;
15+
u8 l2_data[VLAN_ETH_HLEN];
16+
int (*xmit)(struct sk_buff *skb);
17+
};
18+
19+
static DEFINE_PER_CPU(struct sch_frag_data, sch_frag_data_storage);
20+
21+
static int sch_frag_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
22+
{
23+
struct sch_frag_data *data = this_cpu_ptr(&sch_frag_data_storage);
24+
25+
if (skb_cow_head(skb, data->l2_len) < 0) {
26+
kfree_skb(skb);
27+
return -ENOMEM;
28+
}
29+
30+
__skb_dst_copy(skb, data->dst);
31+
*qdisc_skb_cb(skb) = data->cb;
32+
skb->inner_protocol = data->inner_protocol;
33+
if (data->vlan_tci & VLAN_CFI_MASK)
34+
__vlan_hwaccel_put_tag(skb, data->vlan_proto,
35+
data->vlan_tci & ~VLAN_CFI_MASK);
36+
else
37+
__vlan_hwaccel_clear_tag(skb);
38+
39+
/* Reconstruct the MAC header. */
40+
skb_push(skb, data->l2_len);
41+
memcpy(skb->data, &data->l2_data, data->l2_len);
42+
skb_postpush_rcsum(skb, skb->data, data->l2_len);
43+
skb_reset_mac_header(skb);
44+
45+
return data->xmit(skb);
46+
}
47+
48+
static void sch_frag_prepare_frag(struct sk_buff *skb,
49+
int (*xmit)(struct sk_buff *skb))
50+
{
51+
unsigned int hlen = skb_network_offset(skb);
52+
struct sch_frag_data *data;
53+
54+
data = this_cpu_ptr(&sch_frag_data_storage);
55+
data->dst = skb->_skb_refdst;
56+
data->cb = *qdisc_skb_cb(skb);
57+
data->xmit = xmit;
58+
data->inner_protocol = skb->inner_protocol;
59+
if (skb_vlan_tag_present(skb))
60+
data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
61+
else
62+
data->vlan_tci = 0;
63+
data->vlan_proto = skb->vlan_proto;
64+
data->l2_len = hlen;
65+
memcpy(&data->l2_data, skb->data, hlen);
66+
67+
memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
68+
skb_pull(skb, hlen);
69+
}
70+
71+
static unsigned int
72+
sch_frag_dst_get_mtu(const struct dst_entry *dst)
73+
{
74+
return dst->dev->mtu;
75+
}
76+
77+
static struct dst_ops sch_frag_dst_ops = {
78+
.family = AF_UNSPEC,
79+
.mtu = sch_frag_dst_get_mtu,
80+
};
81+
82+
static int sch_fragment(struct net *net, struct sk_buff *skb,
83+
u16 mru, int (*xmit)(struct sk_buff *skb))
84+
{
85+
int ret = -1;
86+
87+
if (skb_network_offset(skb) > VLAN_ETH_HLEN) {
88+
net_warn_ratelimited("L2 header too long to fragment\n");
89+
goto err;
90+
}
91+
92+
if (skb_protocol(skb, true) == htons(ETH_P_IP)) {
93+
struct dst_entry sch_frag_dst;
94+
unsigned long orig_dst;
95+
96+
sch_frag_prepare_frag(skb, xmit);
97+
dst_init(&sch_frag_dst, &sch_frag_dst_ops, NULL, 1,
98+
DST_OBSOLETE_NONE, DST_NOCOUNT);
99+
sch_frag_dst.dev = skb->dev;
100+
101+
orig_dst = skb->_skb_refdst;
102+
skb_dst_set_noref(skb, &sch_frag_dst);
103+
IPCB(skb)->frag_max_size = mru;
104+
105+
ret = ip_do_fragment(net, skb->sk, skb, sch_frag_xmit);
106+
refdst_drop(orig_dst);
107+
} else if (skb_protocol(skb, true) == htons(ETH_P_IPV6)) {
108+
unsigned long orig_dst;
109+
struct rt6_info sch_frag_rt;
110+
111+
sch_frag_prepare_frag(skb, xmit);
112+
memset(&sch_frag_rt, 0, sizeof(sch_frag_rt));
113+
dst_init(&sch_frag_rt.dst, &sch_frag_dst_ops, NULL, 1,
114+
DST_OBSOLETE_NONE, DST_NOCOUNT);
115+
sch_frag_rt.dst.dev = skb->dev;
116+
117+
orig_dst = skb->_skb_refdst;
118+
skb_dst_set_noref(skb, &sch_frag_rt.dst);
119+
IP6CB(skb)->frag_max_size = mru;
120+
121+
ret = ipv6_stub->ipv6_fragment(net, skb->sk, skb,
122+
sch_frag_xmit);
123+
refdst_drop(orig_dst);
124+
} else {
125+
net_warn_ratelimited("Fail frag %s: eth=%x, MRU=%d, MTU=%d\n",
126+
netdev_name(skb->dev),
127+
ntohs(skb_protocol(skb, true)), mru,
128+
skb->dev->mtu);
129+
goto err;
130+
}
131+
132+
return ret;
133+
err:
134+
kfree_skb(skb);
135+
return ret;
136+
}
137+
138+
int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb))
139+
{
140+
u16 mru = qdisc_skb_cb(skb)->mru;
141+
int err;
142+
143+
if (mru && skb->len > mru + skb->dev->hard_header_len)
144+
err = sch_fragment(dev_net(skb->dev), skb, mru, xmit);
145+
else
146+
err = xmit(skb);
147+
148+
return err;
149+
}
150+
EXPORT_SYMBOL_GPL(sch_frag_xmit_hook);

0 commit comments

Comments
 (0)