1 #include <stddef.h>
2 #include <inttypes.h>
3 #include <errno.h>
4 #include <linux/seg6_local.h>
5 #include <linux/bpf.h>
6 #include <bpf/bpf_helpers.h>
7 #include <bpf/bpf_endian.h>
8
9 /* Packet parsing state machine helpers. */
10 #define cursor_advance(_cursor, _len) \
11 ({ void *_tmp = _cursor; _cursor += _len; _tmp; })
12
13 #define SR6_FLAG_ALERT (1 << 4)
14
15 #define BPF_PACKET_HEADER __attribute__((packed))
16
17 struct ip6_t {
18 unsigned int ver:4;
19 unsigned int priority:8;
20 unsigned int flow_label:20;
21 unsigned short payload_len;
22 unsigned char next_header;
23 unsigned char hop_limit;
24 unsigned long long src_hi;
25 unsigned long long src_lo;
26 unsigned long long dst_hi;
27 unsigned long long dst_lo;
28 } BPF_PACKET_HEADER;
29
30 struct ip6_addr_t {
31 unsigned long long hi;
32 unsigned long long lo;
33 } BPF_PACKET_HEADER;
34
35 struct ip6_srh_t {
36 unsigned char nexthdr;
37 unsigned char hdrlen;
38 unsigned char type;
39 unsigned char segments_left;
40 unsigned char first_segment;
41 unsigned char flags;
42 unsigned short tag;
43
44 struct ip6_addr_t segments[0];
45 } BPF_PACKET_HEADER;
46
47 struct sr6_tlv_t {
48 unsigned char type;
49 unsigned char len;
50 unsigned char value[0];
51 } BPF_PACKET_HEADER;
52
get_srh(struct __sk_buff * skb)53 static __always_inline struct ip6_srh_t *get_srh(struct __sk_buff *skb)
54 {
55 void *cursor, *data_end;
56 struct ip6_srh_t *srh;
57 struct ip6_t *ip;
58 uint8_t *ipver;
59
60 data_end = (void *)(long)skb->data_end;
61 cursor = (void *)(long)skb->data;
62 ipver = (uint8_t *)cursor;
63
64 if ((void *)ipver + sizeof(*ipver) > data_end)
65 return NULL;
66
67 if ((*ipver >> 4) != 6)
68 return NULL;
69
70 ip = cursor_advance(cursor, sizeof(*ip));
71 if ((void *)ip + sizeof(*ip) > data_end)
72 return NULL;
73
74 if (ip->next_header != 43)
75 return NULL;
76
77 srh = cursor_advance(cursor, sizeof(*srh));
78 if ((void *)srh + sizeof(*srh) > data_end)
79 return NULL;
80
81 if (srh->type != 4)
82 return NULL;
83
84 return srh;
85 }
86
update_tlv_pad(struct __sk_buff * skb,uint32_t new_pad,uint32_t old_pad,uint32_t pad_off)87 static __always_inline int update_tlv_pad(struct __sk_buff *skb,
88 uint32_t new_pad, uint32_t old_pad,
89 uint32_t pad_off)
90 {
91 int err;
92
93 if (new_pad != old_pad) {
94 err = bpf_lwt_seg6_adjust_srh(skb, pad_off,
95 (int) new_pad - (int) old_pad);
96 if (err)
97 return err;
98 }
99
100 if (new_pad > 0) {
101 char pad_tlv_buf[16] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
102 0, 0, 0};
103 struct sr6_tlv_t *pad_tlv = (struct sr6_tlv_t *) pad_tlv_buf;
104
105 pad_tlv->type = SR6_TLV_PADDING;
106 pad_tlv->len = new_pad - 2;
107
108 err = bpf_lwt_seg6_store_bytes(skb, pad_off,
109 (void *)pad_tlv_buf, new_pad);
110 if (err)
111 return err;
112 }
113
114 return 0;
115 }
116
is_valid_tlv_boundary(struct __sk_buff * skb,struct ip6_srh_t * srh,uint32_t * tlv_off,uint32_t * pad_size,uint32_t * pad_off)117 static __always_inline int is_valid_tlv_boundary(struct __sk_buff *skb,
118 struct ip6_srh_t *srh,
119 uint32_t *tlv_off,
120 uint32_t *pad_size,
121 uint32_t *pad_off)
122 {
123 uint32_t srh_off, cur_off;
124 int offset_valid = 0;
125 int err;
126
127 srh_off = (char *)srh - (char *)(long)skb->data;
128 // cur_off = end of segments, start of possible TLVs
129 cur_off = srh_off + sizeof(*srh) +
130 sizeof(struct ip6_addr_t) * (srh->first_segment + 1);
131
132 *pad_off = 0;
133
134 // we can only go as far as ~10 TLVs due to the BPF max stack size
135 // workaround: define induction variable "i" as "long" instead
136 // of "int" to prevent alu32 sub-register spilling.
137 #pragma clang loop unroll(disable)
138 for (long i = 0; i < 100; i++) {
139 struct sr6_tlv_t tlv;
140
141 if (cur_off == *tlv_off)
142 offset_valid = 1;
143
144 if (cur_off >= srh_off + ((srh->hdrlen + 1) << 3))
145 break;
146
147 err = bpf_skb_load_bytes(skb, cur_off, &tlv, sizeof(tlv));
148 if (err)
149 return err;
150
151 if (tlv.type == SR6_TLV_PADDING) {
152 *pad_size = tlv.len + sizeof(tlv);
153 *pad_off = cur_off;
154
155 if (*tlv_off == srh_off) {
156 *tlv_off = cur_off;
157 offset_valid = 1;
158 }
159 break;
160
161 } else if (tlv.type == SR6_TLV_HMAC) {
162 break;
163 }
164
165 cur_off += sizeof(tlv) + tlv.len;
166 } // we reached the padding or HMAC TLVs, or the end of the SRH
167
168 if (*pad_off == 0)
169 *pad_off = cur_off;
170
171 if (*tlv_off == -1)
172 *tlv_off = cur_off;
173 else if (!offset_valid)
174 return -EINVAL;
175
176 return 0;
177 }
178
add_tlv(struct __sk_buff * skb,struct ip6_srh_t * srh,uint32_t tlv_off,struct sr6_tlv_t * itlv,uint8_t tlv_size)179 static __always_inline int add_tlv(struct __sk_buff *skb,
180 struct ip6_srh_t *srh, uint32_t tlv_off,
181 struct sr6_tlv_t *itlv, uint8_t tlv_size)
182 {
183 uint32_t srh_off = (char *)srh - (char *)(long)skb->data;
184 uint8_t len_remaining, new_pad;
185 uint32_t pad_off = 0;
186 uint32_t pad_size = 0;
187 uint32_t partial_srh_len;
188 int err;
189
190 if (tlv_off != -1)
191 tlv_off += srh_off;
192
193 if (itlv->type == SR6_TLV_PADDING || itlv->type == SR6_TLV_HMAC)
194 return -EINVAL;
195
196 err = is_valid_tlv_boundary(skb, srh, &tlv_off, &pad_size, &pad_off);
197 if (err)
198 return err;
199
200 err = bpf_lwt_seg6_adjust_srh(skb, tlv_off, sizeof(*itlv) + itlv->len);
201 if (err)
202 return err;
203
204 err = bpf_lwt_seg6_store_bytes(skb, tlv_off, (void *)itlv, tlv_size);
205 if (err)
206 return err;
207
208 // the following can't be moved inside update_tlv_pad because the
209 // bpf verifier has some issues with it
210 pad_off += sizeof(*itlv) + itlv->len;
211 partial_srh_len = pad_off - srh_off;
212 len_remaining = partial_srh_len % 8;
213 new_pad = 8 - len_remaining;
214
215 if (new_pad == 1) // cannot pad for 1 byte only
216 new_pad = 9;
217 else if (new_pad == 8)
218 new_pad = 0;
219
220 return update_tlv_pad(skb, new_pad, pad_size, pad_off);
221 }
222
223 // Add an Egress TLV fc00::4, add the flag A,
224 // and apply End.X action to fc42::1
225 SEC("lwt_seg6local")
__add_egr_x(struct __sk_buff * skb)226 int __add_egr_x(struct __sk_buff *skb)
227 {
228 unsigned long long hi = 0xfc42000000000000;
229 unsigned long long lo = 0x1;
230 struct ip6_srh_t *srh = get_srh(skb);
231 uint8_t new_flags = SR6_FLAG_ALERT;
232 struct ip6_addr_t addr;
233 int err, offset;
234
235 if (srh == NULL)
236 return BPF_DROP;
237
238 uint8_t tlv[20] = {2, 18, 0, 0, 0xfd, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0,
239 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x4};
240
241 err = add_tlv(skb, srh, (srh->hdrlen+1) << 3,
242 (struct sr6_tlv_t *)&tlv, 20);
243 if (err)
244 return BPF_DROP;
245
246 offset = sizeof(struct ip6_t) + offsetof(struct ip6_srh_t, flags);
247 err = bpf_lwt_seg6_store_bytes(skb, offset,
248 (void *)&new_flags, sizeof(new_flags));
249 if (err)
250 return BPF_DROP;
251
252 addr.lo = bpf_cpu_to_be64(lo);
253 addr.hi = bpf_cpu_to_be64(hi);
254 err = bpf_lwt_seg6_action(skb, SEG6_LOCAL_ACTION_END_X,
255 (void *)&addr, sizeof(addr));
256 if (err)
257 return BPF_DROP;
258 return BPF_REDIRECT;
259 }
260 char __license[] SEC("license") = "GPL";
261