Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-or-later
2 : /*
3 : * IPV4 GSO/GRO offload support
4 : * Linux INET implementation
5 : *
6 : * TCPv4 GSO/GRO support
7 : */
8 :
9 : #include <linux/indirect_call_wrapper.h>
10 : #include <linux/skbuff.h>
11 : #include <net/tcp.h>
12 : #include <net/protocol.h>
13 :
14 0 : static void tcp_gso_tstamp(struct sk_buff *skb, unsigned int ts_seq,
15 : unsigned int seq, unsigned int mss)
16 : {
17 0 : while (skb) {
18 0 : if (before(ts_seq, seq + mss)) {
19 0 : skb_shinfo(skb)->tx_flags |= SKBTX_SW_TSTAMP;
20 0 : skb_shinfo(skb)->tskey = ts_seq;
21 0 : return;
22 : }
23 :
24 0 : skb = skb->next;
25 0 : seq += mss;
26 : }
27 : }
28 :
29 0 : static struct sk_buff *tcp4_gso_segment(struct sk_buff *skb,
30 : netdev_features_t features)
31 : {
32 0 : if (!(skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4))
33 0 : return ERR_PTR(-EINVAL);
34 :
35 0 : if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
36 0 : return ERR_PTR(-EINVAL);
37 :
38 0 : if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
39 0 : const struct iphdr *iph = ip_hdr(skb);
40 0 : struct tcphdr *th = tcp_hdr(skb);
41 :
42 : /* Set up checksum pseudo header, usually expect stack to
43 : * have done this already.
44 : */
45 :
46 0 : th->check = 0;
47 0 : skb->ip_summed = CHECKSUM_PARTIAL;
48 0 : __tcp_v4_send_check(skb, iph->saddr, iph->daddr);
49 : }
50 :
51 0 : return tcp_gso_segment(skb, features);
52 : }
53 :
54 0 : struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
55 : netdev_features_t features)
56 : {
57 0 : struct sk_buff *segs = ERR_PTR(-EINVAL);
58 0 : unsigned int sum_truesize = 0;
59 0 : struct tcphdr *th;
60 0 : unsigned int thlen;
61 0 : unsigned int seq;
62 0 : __be32 delta;
63 0 : unsigned int oldlen;
64 0 : unsigned int mss;
65 0 : struct sk_buff *gso_skb = skb;
66 0 : __sum16 newcheck;
67 0 : bool ooo_okay, copy_destructor;
68 :
69 0 : th = tcp_hdr(skb);
70 0 : thlen = th->doff * 4;
71 0 : if (thlen < sizeof(*th))
72 0 : goto out;
73 :
74 0 : if (!pskb_may_pull(skb, thlen))
75 0 : goto out;
76 :
77 0 : oldlen = (u16)~skb->len;
78 0 : __skb_pull(skb, thlen);
79 :
80 0 : mss = skb_shinfo(skb)->gso_size;
81 0 : if (unlikely(skb->len <= mss))
82 0 : goto out;
83 :
84 0 : if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
85 : /* Packet is from an untrusted source, reset gso_segs. */
86 :
87 0 : skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
88 :
89 0 : segs = NULL;
90 0 : goto out;
91 : }
92 :
93 0 : copy_destructor = gso_skb->destructor == tcp_wfree;
94 0 : ooo_okay = gso_skb->ooo_okay;
95 : /* All segments but the first should have ooo_okay cleared */
96 0 : skb->ooo_okay = 0;
97 :
98 0 : segs = skb_segment(skb, features);
99 0 : if (IS_ERR(segs))
100 0 : goto out;
101 :
102 : /* Only first segment might have ooo_okay set */
103 0 : segs->ooo_okay = ooo_okay;
104 :
105 : /* GSO partial and frag_list segmentation only requires splitting
106 : * the frame into an MSS multiple and possibly a remainder, both
107 : * cases return a GSO skb. So update the mss now.
108 : */
109 0 : if (skb_is_gso(segs))
110 0 : mss *= skb_shinfo(segs)->gso_segs;
111 :
112 0 : delta = htonl(oldlen + (thlen + mss));
113 :
114 0 : skb = segs;
115 0 : th = tcp_hdr(skb);
116 0 : seq = ntohl(th->seq);
117 :
118 0 : if (unlikely(skb_shinfo(gso_skb)->tx_flags & SKBTX_SW_TSTAMP))
119 0 : tcp_gso_tstamp(segs, skb_shinfo(gso_skb)->tskey, seq, mss);
120 :
121 0 : newcheck = ~csum_fold((__force __wsum)((__force u32)th->check +
122 : (__force u32)delta));
123 :
124 0 : while (skb->next) {
125 0 : th->fin = th->psh = 0;
126 0 : th->check = newcheck;
127 :
128 0 : if (skb->ip_summed == CHECKSUM_PARTIAL)
129 0 : gso_reset_checksum(skb, ~th->check);
130 : else
131 0 : th->check = gso_make_checksum(skb, ~th->check);
132 :
133 0 : seq += mss;
134 0 : if (copy_destructor) {
135 0 : skb->destructor = gso_skb->destructor;
136 0 : skb->sk = gso_skb->sk;
137 0 : sum_truesize += skb->truesize;
138 : }
139 0 : skb = skb->next;
140 0 : th = tcp_hdr(skb);
141 :
142 0 : th->seq = htonl(seq);
143 0 : th->cwr = 0;
144 : }
145 :
146 : /* Following permits TCP Small Queues to work well with GSO :
147 : * The callback to TCP stack will be called at the time last frag
148 : * is freed at TX completion, and not right now when gso_skb
149 : * is freed by GSO engine
150 : */
151 0 : if (copy_destructor) {
152 0 : int delta;
153 :
154 0 : swap(gso_skb->sk, skb->sk);
155 0 : swap(gso_skb->destructor, skb->destructor);
156 0 : sum_truesize += skb->truesize;
157 0 : delta = sum_truesize - gso_skb->truesize;
158 : /* In some pathological cases, delta can be negative.
159 : * We need to either use refcount_add() or refcount_sub_and_test()
160 : */
161 0 : if (likely(delta >= 0))
162 0 : refcount_add(delta, &skb->sk->sk_wmem_alloc);
163 : else
164 0 : WARN_ON_ONCE(refcount_sub_and_test(-delta, &skb->sk->sk_wmem_alloc));
165 : }
166 :
167 0 : delta = htonl(oldlen + (skb_tail_pointer(skb) -
168 : skb_transport_header(skb)) +
169 : skb->data_len);
170 0 : th->check = ~csum_fold((__force __wsum)((__force u32)th->check +
171 : (__force u32)delta));
172 0 : if (skb->ip_summed == CHECKSUM_PARTIAL)
173 0 : gso_reset_checksum(skb, ~th->check);
174 : else
175 0 : th->check = gso_make_checksum(skb, ~th->check);
176 0 : out:
177 0 : return segs;
178 : }
179 :
180 705 : struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb)
181 : {
182 705 : struct sk_buff *pp = NULL;
183 705 : struct sk_buff *p;
184 705 : struct tcphdr *th;
185 705 : struct tcphdr *th2;
186 705 : unsigned int len;
187 705 : unsigned int thlen;
188 705 : __be32 flags;
189 705 : unsigned int mss = 1;
190 705 : unsigned int hlen;
191 705 : unsigned int off;
192 705 : int flush = 1;
193 705 : int i;
194 :
195 705 : off = skb_gro_offset(skb);
196 705 : hlen = off + sizeof(*th);
197 705 : th = skb_gro_header_fast(skb, off);
198 705 : if (skb_gro_header_hard(skb, hlen)) {
199 705 : th = skb_gro_header_slow(skb, hlen, off);
200 705 : if (unlikely(!th))
201 0 : goto out;
202 : }
203 :
204 705 : thlen = th->doff * 4;
205 705 : if (thlen < sizeof(*th))
206 0 : goto out;
207 :
208 705 : hlen = off + thlen;
209 705 : if (skb_gro_header_hard(skb, hlen)) {
210 705 : th = skb_gro_header_slow(skb, hlen, off);
211 705 : if (unlikely(!th))
212 0 : goto out;
213 : }
214 :
215 705 : skb_gro_pull(skb, thlen);
216 :
217 705 : len = skb_gro_len(skb);
218 705 : flags = tcp_flag_word(th);
219 :
220 705 : list_for_each_entry(p, head, list) {
221 269 : if (!NAPI_GRO_CB(p)->same_flow)
222 0 : continue;
223 :
224 269 : th2 = tcp_hdr(p);
225 :
226 269 : if (*(u32 *)&th->source ^ *(u32 *)&th2->source) {
227 0 : NAPI_GRO_CB(p)->same_flow = 0;
228 0 : continue;
229 : }
230 :
231 269 : goto found;
232 : }
233 436 : p = NULL;
234 436 : goto out_check_final;
235 :
236 269 : found:
237 : /* Include the IP ID check below from the inner most IP hdr */
238 269 : flush = NAPI_GRO_CB(p)->flush;
239 269 : flush |= (__force int)(flags & TCP_FLAG_CWR);
240 269 : flush |= (__force int)((flags ^ tcp_flag_word(th2)) &
241 : ~(TCP_FLAG_CWR | TCP_FLAG_FIN | TCP_FLAG_PSH));
242 269 : flush |= (__force int)(th->ack_seq ^ th2->ack_seq);
243 269 : for (i = sizeof(*th); i < thlen; i += 4)
244 0 : flush |= *(u32 *)((u8 *)th + i) ^
245 0 : *(u32 *)((u8 *)th2 + i);
246 :
247 : /* When we receive our second frame we can made a decision on if we
248 : * continue this flow as an atomic flow with a fixed ID or if we use
249 : * an incrementing ID.
250 : */
251 269 : if (NAPI_GRO_CB(p)->flush_id != 1 ||
252 0 : NAPI_GRO_CB(p)->count != 1 ||
253 0 : !NAPI_GRO_CB(p)->is_atomic)
254 269 : flush |= NAPI_GRO_CB(p)->flush_id;
255 : else
256 0 : NAPI_GRO_CB(p)->is_atomic = false;
257 :
258 269 : mss = skb_shinfo(p)->gso_size;
259 :
260 269 : flush |= (len - 1) >= mss;
261 269 : flush |= (ntohl(th2->seq) + skb_gro_len(p)) ^ ntohl(th->seq);
262 : #ifdef CONFIG_TLS_DEVICE
263 : flush |= p->decrypted ^ skb->decrypted;
264 : #endif
265 :
266 269 : if (flush || skb_gro_receive(p, skb)) {
267 2 : mss = 1;
268 2 : goto out_check_final;
269 : }
270 :
271 267 : tcp_flag_word(th2) |= flags & (TCP_FLAG_FIN | TCP_FLAG_PSH);
272 :
273 705 : out_check_final:
274 705 : flush = len < mss;
275 705 : flush |= (__force int)(flags & (TCP_FLAG_URG | TCP_FLAG_PSH |
276 : TCP_FLAG_RST | TCP_FLAG_SYN |
277 : TCP_FLAG_FIN));
278 :
279 705 : if (p && (!NAPI_GRO_CB(skb)->same_flow || flush))
280 5 : pp = p;
281 :
282 700 : out:
283 705 : NAPI_GRO_CB(skb)->flush |= (flush != 0);
284 :
285 705 : return pp;
286 : }
287 :
288 15 : int tcp_gro_complete(struct sk_buff *skb)
289 : {
290 15 : struct tcphdr *th = tcp_hdr(skb);
291 :
292 15 : skb->csum_start = (unsigned char *)th - skb->head;
293 15 : skb->csum_offset = offsetof(struct tcphdr, check);
294 15 : skb->ip_summed = CHECKSUM_PARTIAL;
295 :
296 15 : skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
297 :
298 15 : if (th->cwr)
299 0 : skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
300 :
301 15 : return 0;
302 : }
303 : EXPORT_SYMBOL(tcp_gro_complete);
304 :
305 : INDIRECT_CALLABLE_SCOPE
306 705 : struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)
307 : {
308 : /* Don't bother verifying checksum if we're going to flush anyway. */
309 705 : if (!NAPI_GRO_CB(skb)->flush &&
310 705 : skb_gro_checksum_validate(skb, IPPROTO_TCP,
311 : inet_gro_compute_pseudo)) {
312 0 : NAPI_GRO_CB(skb)->flush = 1;
313 0 : return NULL;
314 : }
315 :
316 705 : return tcp_gro_receive(head, skb);
317 : }
318 :
319 15 : INDIRECT_CALLABLE_SCOPE int tcp4_gro_complete(struct sk_buff *skb, int thoff)
320 : {
321 15 : const struct iphdr *iph = ip_hdr(skb);
322 15 : struct tcphdr *th = tcp_hdr(skb);
323 :
324 15 : th->check = ~tcp_v4_check(skb->len - thoff, iph->saddr,
325 : iph->daddr, 0);
326 15 : skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV4;
327 :
328 15 : if (NAPI_GRO_CB(skb)->is_atomic)
329 0 : skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_FIXEDID;
330 :
331 15 : return tcp_gro_complete(skb);
332 : }
333 :
334 : static const struct net_offload tcpv4_offload = {
335 : .callbacks = {
336 : .gso_segment = tcp4_gso_segment,
337 : .gro_receive = tcp4_gro_receive,
338 : .gro_complete = tcp4_gro_complete,
339 : },
340 : };
341 :
342 1 : int __init tcpv4_offload_init(void)
343 : {
344 1 : return inet_add_offload(&tcpv4_offload, IPPROTO_TCP);
345 : }
|
