1 #include <linux/types.h>
2 #include <linux/netfilter.h>
3 #include <net/tcp.h>
4 
5 #include <net/netfilter/nf_conntrack.h>
6 #include <net/netfilter/nf_conntrack_extend.h>
7 #include <net/netfilter/nf_conntrack_seqadj.h>
8 
9 int nf_ct_seqadj_init(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
10 		      s32 off)
11 {
12 	enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
13 	struct nf_conn_seqadj *seqadj;
14 	struct nf_ct_seqadj *this_way;
15 
16 	if (off == 0)
17 		return 0;
18 
19 	set_bit(IPS_SEQ_ADJUST_BIT, &ct->status);
20 
21 	seqadj = nfct_seqadj(ct);
22 	this_way = &seqadj->seq[dir];
23 	this_way->offset_before	 = off;
24 	this_way->offset_after	 = off;
25 	return 0;
26 }
27 EXPORT_SYMBOL_GPL(nf_ct_seqadj_init);
28 
29 int nf_ct_seqadj_set(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
30 		     __be32 seq, s32 off)
31 {
32 	struct nf_conn_seqadj *seqadj = nfct_seqadj(ct);
33 	enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
34 	struct nf_ct_seqadj *this_way;
35 
36 	if (off == 0)
37 		return 0;
38 
39 	if (unlikely(!seqadj)) {
40 		WARN_ONCE(1, "Missing nfct_seqadj_ext_add() setup call\n");
41 		return 0;
42 	}
43 
44 	set_bit(IPS_SEQ_ADJUST_BIT, &ct->status);
45 
46 	spin_lock_bh(&ct->lock);
47 	this_way = &seqadj->seq[dir];
48 	if (this_way->offset_before == this_way->offset_after ||
49 	    before(this_way->correction_pos, ntohl(seq))) {
50 		this_way->correction_pos = ntohl(seq);
51 		this_way->offset_before	 = this_way->offset_after;
52 		this_way->offset_after	+= off;
53 	}
54 	spin_unlock_bh(&ct->lock);
55 	return 0;
56 }
57 EXPORT_SYMBOL_GPL(nf_ct_seqadj_set);
58 
59 void nf_ct_tcp_seqadj_set(struct sk_buff *skb,
60 			  struct nf_conn *ct, enum ip_conntrack_info ctinfo,
61 			  s32 off)
62 {
63 	const struct tcphdr *th;
64 
65 	if (nf_ct_protonum(ct) != IPPROTO_TCP)
66 		return;
67 
68 	th = (struct tcphdr *)(skb_network_header(skb) + ip_hdrlen(skb));
69 	nf_ct_seqadj_set(ct, ctinfo, th->seq, off);
70 }
71 EXPORT_SYMBOL_GPL(nf_ct_tcp_seqadj_set);
72 
73 /* Adjust one found SACK option including checksum correction */
74 static void nf_ct_sack_block_adjust(struct sk_buff *skb,
75 				    struct tcphdr *tcph,
76 				    unsigned int sackoff,
77 				    unsigned int sackend,
78 				    struct nf_ct_seqadj *seq)
79 {
80 	while (sackoff < sackend) {
81 		struct tcp_sack_block_wire *sack;
82 		__be32 new_start_seq, new_end_seq;
83 
84 		sack = (void *)skb->data + sackoff;
85 		if (after(ntohl(sack->start_seq) - seq->offset_before,
86 			  seq->correction_pos))
87 			new_start_seq = htonl(ntohl(sack->start_seq) -
88 					seq->offset_after);
89 		else
90 			new_start_seq = htonl(ntohl(sack->start_seq) -
91 					seq->offset_before);
92 
93 		if (after(ntohl(sack->end_seq) - seq->offset_before,
94 			  seq->correction_pos))
95 			new_end_seq = htonl(ntohl(sack->end_seq) -
96 				      seq->offset_after);
97 		else
98 			new_end_seq = htonl(ntohl(sack->end_seq) -
99 				      seq->offset_before);
100 
101 		pr_debug("sack_adjust: start_seq: %u->%u, end_seq: %u->%u\n",
102 			 ntohl(sack->start_seq), ntohl(new_start_seq),
103 			 ntohl(sack->end_seq), ntohl(new_end_seq));
104 
105 		inet_proto_csum_replace4(&tcph->check, skb,
106 					 sack->start_seq, new_start_seq, false);
107 		inet_proto_csum_replace4(&tcph->check, skb,
108 					 sack->end_seq, new_end_seq, false);
109 		sack->start_seq = new_start_seq;
110 		sack->end_seq = new_end_seq;
111 		sackoff += sizeof(*sack);
112 	}
113 }
114 
115 /* TCP SACK sequence number adjustment */
116 static unsigned int nf_ct_sack_adjust(struct sk_buff *skb,
117 				      unsigned int protoff,
118 				      struct tcphdr *tcph,
119 				      struct nf_conn *ct,
120 				      enum ip_conntrack_info ctinfo)
121 {
122 	unsigned int dir, optoff, optend;
123 	struct nf_conn_seqadj *seqadj = nfct_seqadj(ct);
124 
125 	optoff = protoff + sizeof(struct tcphdr);
126 	optend = protoff + tcph->doff * 4;
127 
128 	if (!skb_make_writable(skb, optend))
129 		return 0;
130 
131 	dir = CTINFO2DIR(ctinfo);
132 
133 	while (optoff < optend) {
134 		/* Usually: option, length. */
135 		unsigned char *op = skb->data + optoff;
136 
137 		switch (op[0]) {
138 		case TCPOPT_EOL:
139 			return 1;
140 		case TCPOPT_NOP:
141 			optoff++;
142 			continue;
143 		default:
144 			/* no partial options */
145 			if (optoff + 1 == optend ||
146 			    optoff + op[1] > optend ||
147 			    op[1] < 2)
148 				return 0;
149 			if (op[0] == TCPOPT_SACK &&
150 			    op[1] >= 2+TCPOLEN_SACK_PERBLOCK &&
151 			    ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0)
152 				nf_ct_sack_block_adjust(skb, tcph, optoff + 2,
153 							optoff+op[1],
154 							&seqadj->seq[!dir]);
155 			optoff += op[1];
156 		}
157 	}
158 	return 1;
159 }
160 
161 /* TCP sequence number adjustment.  Returns 1 on success, 0 on failure */
162 int nf_ct_seq_adjust(struct sk_buff *skb,
163 		     struct nf_conn *ct, enum ip_conntrack_info ctinfo,
164 		     unsigned int protoff)
165 {
166 	enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
167 	struct tcphdr *tcph;
168 	__be32 newseq, newack;
169 	s32 seqoff, ackoff;
170 	struct nf_conn_seqadj *seqadj = nfct_seqadj(ct);
171 	struct nf_ct_seqadj *this_way, *other_way;
172 	int res;
173 
174 	this_way  = &seqadj->seq[dir];
175 	other_way = &seqadj->seq[!dir];
176 
177 	if (!skb_make_writable(skb, protoff + sizeof(*tcph)))
178 		return 0;
179 
180 	tcph = (void *)skb->data + protoff;
181 	spin_lock_bh(&ct->lock);
182 	if (after(ntohl(tcph->seq), this_way->correction_pos))
183 		seqoff = this_way->offset_after;
184 	else
185 		seqoff = this_way->offset_before;
186 
187 	if (after(ntohl(tcph->ack_seq) - other_way->offset_before,
188 		  other_way->correction_pos))
189 		ackoff = other_way->offset_after;
190 	else
191 		ackoff = other_way->offset_before;
192 
193 	newseq = htonl(ntohl(tcph->seq) + seqoff);
194 	newack = htonl(ntohl(tcph->ack_seq) - ackoff);
195 
196 	inet_proto_csum_replace4(&tcph->check, skb, tcph->seq, newseq, false);
197 	inet_proto_csum_replace4(&tcph->check, skb, tcph->ack_seq, newack,
198 				 false);
199 
200 	pr_debug("Adjusting sequence number from %u->%u, ack from %u->%u\n",
201 		 ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq),
202 		 ntohl(newack));
203 
204 	tcph->seq = newseq;
205 	tcph->ack_seq = newack;
206 
207 	res = nf_ct_sack_adjust(skb, protoff, tcph, ct, ctinfo);
208 	spin_unlock_bh(&ct->lock);
209 
210 	return res;
211 }
212 EXPORT_SYMBOL_GPL(nf_ct_seq_adjust);
213 
214 s32 nf_ct_seq_offset(const struct nf_conn *ct,
215 		     enum ip_conntrack_dir dir,
216 		     u32 seq)
217 {
218 	struct nf_conn_seqadj *seqadj = nfct_seqadj(ct);
219 	struct nf_ct_seqadj *this_way;
220 
221 	if (!seqadj)
222 		return 0;
223 
224 	this_way = &seqadj->seq[dir];
225 	return after(seq, this_way->correction_pos) ?
226 		 this_way->offset_after : this_way->offset_before;
227 }
228 EXPORT_SYMBOL_GPL(nf_ct_seq_offset);
229 
230 static struct nf_ct_ext_type nf_ct_seqadj_extend __read_mostly = {
231 	.len	= sizeof(struct nf_conn_seqadj),
232 	.align	= __alignof__(struct nf_conn_seqadj),
233 	.id	= NF_CT_EXT_SEQADJ,
234 };
235 
236 int nf_conntrack_seqadj_init(void)
237 {
238 	return nf_ct_extend_register(&nf_ct_seqadj_extend);
239 }
240 
241 void nf_conntrack_seqadj_fini(void)
242 {
243 	nf_ct_extend_unregister(&nf_ct_seqadj_extend);
244 }
245