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 #include "bpf_compiler.h"
10
11 /* Packet parsing state machine helpers. */
12 #define cursor_advance(_cursor, _len) \
13 ({ void *_tmp = _cursor; _cursor += _len; _tmp; })
14
15 #define SR6_FLAG_ALERT (1 << 4)
16
17 #define BPF_PACKET_HEADER __attribute__((packed))
18
19 struct ip6_t {
20 unsigned int ver:4;
21 unsigned int priority:8;
22 unsigned int flow_label:20;
23 unsigned short payload_len;
24 unsigned char next_header;
25 unsigned char hop_limit;
26 unsigned long long src_hi;
27 unsigned long long src_lo;
28 unsigned long long dst_hi;
29 unsigned long long dst_lo;
30 } BPF_PACKET_HEADER;
31
32 struct ip6_addr_t {
33 unsigned long long hi;
34 unsigned long long lo;
35 } BPF_PACKET_HEADER;
36
37 struct ip6_srh_t {
38 unsigned char nexthdr;
39 unsigned char hdrlen;
40 unsigned char type;
41 unsigned char segments_left;
42 unsigned char first_segment;
43 unsigned char flags;
44 unsigned short tag;
45
46 struct ip6_addr_t segments[0];
47 } BPF_PACKET_HEADER;
48
49 struct sr6_tlv_t {
50 unsigned char type;
51 unsigned char len;
52 unsigned char value[0];
53 } BPF_PACKET_HEADER;
54
get_srh(struct __sk_buff * skb)55 static __always_inline struct ip6_srh_t *get_srh(struct __sk_buff *skb)
56 {
57 void *cursor, *data_end;
58 struct ip6_srh_t *srh;
59 struct ip6_t *ip;
60 uint8_t *ipver;
61
62 data_end = (void *)(long)skb->data_end;
63 cursor = (void *)(long)skb->data;
64 ipver = (uint8_t *)cursor;
65
66 if ((void *)ipver + sizeof(*ipver) > data_end)
67 return NULL;
68
69 if ((*ipver >> 4) != 6)
70 return NULL;
71
72 ip = cursor_advance(cursor, sizeof(*ip));
73 if ((void *)ip + sizeof(*ip) > data_end)
74 return NULL;
75
76 if (ip->next_header != 43)
77 return NULL;
78
79 srh = cursor_advance(cursor, sizeof(*srh));
80 if ((void *)srh + sizeof(*srh) > data_end)
81 return NULL;
82
83 if (srh->type != 4)
84 return NULL;
85
86 return srh;
87 }
88
update_tlv_pad(struct __sk_buff * skb,uint32_t new_pad,uint32_t old_pad,uint32_t pad_off)89 static __always_inline int update_tlv_pad(struct __sk_buff *skb,
90 uint32_t new_pad, uint32_t old_pad,
91 uint32_t pad_off)
92 {
93 int err;
94
95 if (new_pad != old_pad) {
96 err = bpf_lwt_seg6_adjust_srh(skb, pad_off,
97 (int) new_pad - (int) old_pad);
98 if (err)
99 return err;
100 }
101
102 if (new_pad > 0) {
103 char pad_tlv_buf[16] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
104 0, 0, 0};
105 struct sr6_tlv_t *pad_tlv = (struct sr6_tlv_t *) pad_tlv_buf;
106
107 pad_tlv->type = SR6_TLV_PADDING;
108 pad_tlv->len = new_pad - 2;
109
110 err = bpf_lwt_seg6_store_bytes(skb, pad_off,
111 (void *)pad_tlv_buf, new_pad);
112 if (err)
113 return err;
114 }
115
116 return 0;
117 }
118
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)119 static __always_inline int is_valid_tlv_boundary(struct __sk_buff *skb,
120 struct ip6_srh_t *srh,
121 uint32_t *tlv_off,
122 uint32_t *pad_size,
123 uint32_t *pad_off)
124 {
125 uint32_t srh_off, cur_off;
126 int offset_valid = 0;
127 int err;
128
129 srh_off = (char *)srh - (char *)(long)skb->data;
130 // cur_off = end of segments, start of possible TLVs
131 cur_off = srh_off + sizeof(*srh) +
132 sizeof(struct ip6_addr_t) * (srh->first_segment + 1);
133
134 *pad_off = 0;
135
136 // we can only go as far as ~10 TLVs due to the BPF max stack size
137 // workaround: define induction variable "i" as "long" instead
138 // of "int" to prevent alu32 sub-register spilling.
139 __pragma_loop_no_unroll
140 for (long i = 0; i < 100; i++) {
141 struct sr6_tlv_t tlv;
142
143 if (cur_off == *tlv_off)
144 offset_valid = 1;
145
146 if (cur_off >= srh_off + ((srh->hdrlen + 1) << 3))
147 break;
148
149 err = bpf_skb_load_bytes(skb, cur_off, &tlv, sizeof(tlv));
150 if (err)
151 return err;
152
153 if (tlv.type == SR6_TLV_PADDING) {
154 *pad_size = tlv.len + sizeof(tlv);
155 *pad_off = cur_off;
156
157 if (*tlv_off == srh_off) {
158 *tlv_off = cur_off;
159 offset_valid = 1;
160 }
161 break;
162
163 } else if (tlv.type == SR6_TLV_HMAC) {
164 break;
165 }
166
167 cur_off += sizeof(tlv) + tlv.len;
168 } // we reached the padding or HMAC TLVs, or the end of the SRH
169
170 if (*pad_off == 0)
171 *pad_off = cur_off;
172
173 if (*tlv_off == -1)
174 *tlv_off = cur_off;
175 else if (!offset_valid)
176 return -EINVAL;
177
178 return 0;
179 }
180
add_tlv(struct __sk_buff * skb,struct ip6_srh_t * srh,uint32_t tlv_off,struct sr6_tlv_t * itlv,uint8_t tlv_size)181 static __always_inline int add_tlv(struct __sk_buff *skb,
182 struct ip6_srh_t *srh, uint32_t tlv_off,
183 struct sr6_tlv_t *itlv, uint8_t tlv_size)
184 {
185 uint32_t srh_off = (char *)srh - (char *)(long)skb->data;
186 uint8_t len_remaining, new_pad;
187 uint32_t pad_off = 0;
188 uint32_t pad_size = 0;
189 uint32_t partial_srh_len;
190 int err;
191
192 if (tlv_off != -1)
193 tlv_off += srh_off;
194
195 if (itlv->type == SR6_TLV_PADDING || itlv->type == SR6_TLV_HMAC)
196 return -EINVAL;
197
198 err = is_valid_tlv_boundary(skb, srh, &tlv_off, &pad_size, &pad_off);
199 if (err)
200 return err;
201
202 err = bpf_lwt_seg6_adjust_srh(skb, tlv_off, sizeof(*itlv) + itlv->len);
203 if (err)
204 return err;
205
206 err = bpf_lwt_seg6_store_bytes(skb, tlv_off, (void *)itlv, tlv_size);
207 if (err)
208 return err;
209
210 // the following can't be moved inside update_tlv_pad because the
211 // bpf verifier has some issues with it
212 pad_off += sizeof(*itlv) + itlv->len;
213 partial_srh_len = pad_off - srh_off;
214 len_remaining = partial_srh_len % 8;
215 new_pad = 8 - len_remaining;
216
217 if (new_pad == 1) // cannot pad for 1 byte only
218 new_pad = 9;
219 else if (new_pad == 8)
220 new_pad = 0;
221
222 return update_tlv_pad(skb, new_pad, pad_size, pad_off);
223 }
224
225 // Add an Egress TLV fc00::4, add the flag A,
226 // and apply End.X action to fc42::1
227 SEC("lwt_seg6local")
__add_egr_x(struct __sk_buff * skb)228 int __add_egr_x(struct __sk_buff *skb)
229 {
230 unsigned long long hi = 0xfc42000000000000;
231 unsigned long long lo = 0x1;
232 struct ip6_srh_t *srh = get_srh(skb);
233 uint8_t new_flags = SR6_FLAG_ALERT;
234 struct ip6_addr_t addr;
235 int err, offset;
236
237 if (srh == NULL)
238 return BPF_DROP;
239
240 uint8_t tlv[20] = {2, 18, 0, 0, 0xfd, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0,
241 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x4};
242
243 err = add_tlv(skb, srh, (srh->hdrlen+1) << 3,
244 (struct sr6_tlv_t *)&tlv, 20);
245 if (err)
246 return BPF_DROP;
247
248 offset = sizeof(struct ip6_t) + offsetof(struct ip6_srh_t, flags);
249 err = bpf_lwt_seg6_store_bytes(skb, offset,
250 (void *)&new_flags, sizeof(new_flags));
251 if (err)
252 return BPF_DROP;
253
254 addr.lo = bpf_cpu_to_be64(lo);
255 addr.hi = bpf_cpu_to_be64(hi);
256 err = bpf_lwt_seg6_action(skb, SEG6_LOCAL_ACTION_END_X,
257 (void *)&addr, sizeof(addr));
258 if (err)
259 return BPF_DROP;
260 return BPF_REDIRECT;
261 }
262 char __license[] SEC("license") = "GPL";
263