~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/ipv4/udp_offload.c

Version: ~ [ linux-6.11.5 ] ~ [ linux-6.10.14 ] ~ [ linux-6.9.12 ] ~ [ linux-6.8.12 ] ~ [ linux-6.7.12 ] ~ [ linux-6.6.58 ] ~ [ linux-6.5.13 ] ~ [ linux-6.4.16 ] ~ [ linux-6.3.13 ] ~ [ linux-6.2.16 ] ~ [ linux-6.1.114 ] ~ [ linux-6.0.19 ] ~ [ linux-5.19.17 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.169 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.228 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.284 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.322 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.336 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.337 ] ~ [ linux-4.4.302 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0-or-later
  2 /*
  3  *      IPV4 GSO/GRO offload support
  4  *      Linux INET implementation
  5  *
  6  *      UDPv4 GSO support
  7  */
  8 
  9 #include <linux/skbuff.h>
 10 #include <net/gro.h>
 11 #include <net/gso.h>
 12 #include <net/udp.h>
 13 #include <net/protocol.h>
 14 #include <net/inet_common.h>
 15 
 16 static struct sk_buff *__skb_udp_tunnel_segment(struct sk_buff *skb,
 17         netdev_features_t features,
 18         struct sk_buff *(*gso_inner_segment)(struct sk_buff *skb,
 19                                              netdev_features_t features),
 20         __be16 new_protocol, bool is_ipv6)
 21 {
 22         int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
 23         bool remcsum, need_csum, offload_csum, gso_partial;
 24         struct sk_buff *segs = ERR_PTR(-EINVAL);
 25         struct udphdr *uh = udp_hdr(skb);
 26         u16 mac_offset = skb->mac_header;
 27         __be16 protocol = skb->protocol;
 28         u16 mac_len = skb->mac_len;
 29         int udp_offset, outer_hlen;
 30         __wsum partial;
 31         bool need_ipsec;
 32 
 33         if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
 34                 goto out;
 35 
 36         /* Adjust partial header checksum to negate old length.
 37          * We cannot rely on the value contained in uh->len as it is
 38          * possible that the actual value exceeds the boundaries of the
 39          * 16 bit length field due to the header being added outside of an
 40          * IP or IPv6 frame that was already limited to 64K - 1.
 41          */
 42         if (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)
 43                 partial = (__force __wsum)uh->len;
 44         else
 45                 partial = (__force __wsum)htonl(skb->len);
 46         partial = csum_sub(csum_unfold(uh->check), partial);
 47 
 48         /* setup inner skb. */
 49         skb->encapsulation = 0;
 50         SKB_GSO_CB(skb)->encap_level = 0;
 51         __skb_pull(skb, tnl_hlen);
 52         skb_reset_mac_header(skb);
 53         skb_set_network_header(skb, skb_inner_network_offset(skb));
 54         skb_set_transport_header(skb, skb_inner_transport_offset(skb));
 55         skb->mac_len = skb_inner_network_offset(skb);
 56         skb->protocol = new_protocol;
 57 
 58         need_csum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM);
 59         skb->encap_hdr_csum = need_csum;
 60 
 61         remcsum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TUNNEL_REMCSUM);
 62         skb->remcsum_offload = remcsum;
 63 
 64         need_ipsec = skb_dst(skb) && dst_xfrm(skb_dst(skb));
 65         /* Try to offload checksum if possible */
 66         offload_csum = !!(need_csum &&
 67                           !need_ipsec &&
 68                           (skb->dev->features &
 69                            (is_ipv6 ? (NETIF_F_HW_CSUM | NETIF_F_IPV6_CSUM) :
 70                                       (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM))));
 71 
 72         features &= skb->dev->hw_enc_features;
 73         if (need_csum)
 74                 features &= ~NETIF_F_SCTP_CRC;
 75 
 76         /* The only checksum offload we care about from here on out is the
 77          * outer one so strip the existing checksum feature flags and
 78          * instead set the flag based on our outer checksum offload value.
 79          */
 80         if (remcsum) {
 81                 features &= ~NETIF_F_CSUM_MASK;
 82                 if (!need_csum || offload_csum)
 83                         features |= NETIF_F_HW_CSUM;
 84         }
 85 
 86         /* segment inner packet. */
 87         segs = gso_inner_segment(skb, features);
 88         if (IS_ERR_OR_NULL(segs)) {
 89                 skb_gso_error_unwind(skb, protocol, tnl_hlen, mac_offset,
 90                                      mac_len);
 91                 goto out;
 92         }
 93 
 94         gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
 95 
 96         outer_hlen = skb_tnl_header_len(skb);
 97         udp_offset = outer_hlen - tnl_hlen;
 98         skb = segs;
 99         do {
100                 unsigned int len;
101 
102                 if (remcsum)
103                         skb->ip_summed = CHECKSUM_NONE;
104 
105                 /* Set up inner headers if we are offloading inner checksum */
106                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
107                         skb_reset_inner_headers(skb);
108                         skb->encapsulation = 1;
109                 }
110 
111                 skb->mac_len = mac_len;
112                 skb->protocol = protocol;
113 
114                 __skb_push(skb, outer_hlen);
115                 skb_reset_mac_header(skb);
116                 skb_set_network_header(skb, mac_len);
117                 skb_set_transport_header(skb, udp_offset);
118                 len = skb->len - udp_offset;
119                 uh = udp_hdr(skb);
120 
121                 /* If we are only performing partial GSO the inner header
122                  * will be using a length value equal to only one MSS sized
123                  * segment instead of the entire frame.
124                  */
125                 if (gso_partial && skb_is_gso(skb)) {
126                         uh->len = htons(skb_shinfo(skb)->gso_size +
127                                         SKB_GSO_CB(skb)->data_offset +
128                                         skb->head - (unsigned char *)uh);
129                 } else {
130                         uh->len = htons(len);
131                 }
132 
133                 if (!need_csum)
134                         continue;
135 
136                 uh->check = ~csum_fold(csum_add(partial,
137                                        (__force __wsum)htonl(len)));
138 
139                 if (skb->encapsulation || !offload_csum) {
140                         uh->check = gso_make_checksum(skb, ~uh->check);
141                         if (uh->check == 0)
142                                 uh->check = CSUM_MANGLED_0;
143                 } else {
144                         skb->ip_summed = CHECKSUM_PARTIAL;
145                         skb->csum_start = skb_transport_header(skb) - skb->head;
146                         skb->csum_offset = offsetof(struct udphdr, check);
147                 }
148         } while ((skb = skb->next));
149 out:
150         return segs;
151 }
152 
153 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
154                                        netdev_features_t features,
155                                        bool is_ipv6)
156 {
157         const struct net_offload __rcu **offloads;
158         __be16 protocol = skb->protocol;
159         const struct net_offload *ops;
160         struct sk_buff *segs = ERR_PTR(-EINVAL);
161         struct sk_buff *(*gso_inner_segment)(struct sk_buff *skb,
162                                              netdev_features_t features);
163 
164         rcu_read_lock();
165 
166         switch (skb->inner_protocol_type) {
167         case ENCAP_TYPE_ETHER:
168                 protocol = skb->inner_protocol;
169                 gso_inner_segment = skb_mac_gso_segment;
170                 break;
171         case ENCAP_TYPE_IPPROTO:
172                 offloads = is_ipv6 ? inet6_offloads : inet_offloads;
173                 ops = rcu_dereference(offloads[skb->inner_ipproto]);
174                 if (!ops || !ops->callbacks.gso_segment)
175                         goto out_unlock;
176                 gso_inner_segment = ops->callbacks.gso_segment;
177                 break;
178         default:
179                 goto out_unlock;
180         }
181 
182         segs = __skb_udp_tunnel_segment(skb, features, gso_inner_segment,
183                                         protocol, is_ipv6);
184 
185 out_unlock:
186         rcu_read_unlock();
187 
188         return segs;
189 }
190 EXPORT_SYMBOL(skb_udp_tunnel_segment);
191 
192 static void __udpv4_gso_segment_csum(struct sk_buff *seg,
193                                      __be32 *oldip, __be32 *newip,
194                                      __be16 *oldport, __be16 *newport)
195 {
196         struct udphdr *uh;
197         struct iphdr *iph;
198 
199         if (*oldip == *newip && *oldport == *newport)
200                 return;
201 
202         uh = udp_hdr(seg);
203         iph = ip_hdr(seg);
204 
205         if (uh->check) {
206                 inet_proto_csum_replace4(&uh->check, seg, *oldip, *newip,
207                                          true);
208                 inet_proto_csum_replace2(&uh->check, seg, *oldport, *newport,
209                                          false);
210                 if (!uh->check)
211                         uh->check = CSUM_MANGLED_0;
212         }
213         *oldport = *newport;
214 
215         csum_replace4(&iph->check, *oldip, *newip);
216         *oldip = *newip;
217 }
218 
219 static struct sk_buff *__udpv4_gso_segment_list_csum(struct sk_buff *segs)
220 {
221         struct sk_buff *seg;
222         struct udphdr *uh, *uh2;
223         struct iphdr *iph, *iph2;
224 
225         seg = segs;
226         uh = udp_hdr(seg);
227         iph = ip_hdr(seg);
228 
229         if ((udp_hdr(seg)->dest == udp_hdr(seg->next)->dest) &&
230             (udp_hdr(seg)->source == udp_hdr(seg->next)->source) &&
231             (ip_hdr(seg)->daddr == ip_hdr(seg->next)->daddr) &&
232             (ip_hdr(seg)->saddr == ip_hdr(seg->next)->saddr))
233                 return segs;
234 
235         while ((seg = seg->next)) {
236                 uh2 = udp_hdr(seg);
237                 iph2 = ip_hdr(seg);
238 
239                 __udpv4_gso_segment_csum(seg,
240                                          &iph2->saddr, &iph->saddr,
241                                          &uh2->source, &uh->source);
242                 __udpv4_gso_segment_csum(seg,
243                                          &iph2->daddr, &iph->daddr,
244                                          &uh2->dest, &uh->dest);
245         }
246 
247         return segs;
248 }
249 
250 static struct sk_buff *__udp_gso_segment_list(struct sk_buff *skb,
251                                               netdev_features_t features,
252                                               bool is_ipv6)
253 {
254         unsigned int mss = skb_shinfo(skb)->gso_size;
255 
256         skb = skb_segment_list(skb, features, skb_mac_header_len(skb));
257         if (IS_ERR(skb))
258                 return skb;
259 
260         udp_hdr(skb)->len = htons(sizeof(struct udphdr) + mss);
261 
262         return is_ipv6 ? skb : __udpv4_gso_segment_list_csum(skb);
263 }
264 
265 struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
266                                   netdev_features_t features, bool is_ipv6)
267 {
268         struct sock *sk = gso_skb->sk;
269         unsigned int sum_truesize = 0;
270         struct sk_buff *segs, *seg;
271         struct udphdr *uh;
272         unsigned int mss;
273         bool copy_dtor;
274         __sum16 check;
275         __be16 newlen;
276 
277         mss = skb_shinfo(gso_skb)->gso_size;
278         if (gso_skb->len <= sizeof(*uh) + mss)
279                 return ERR_PTR(-EINVAL);
280 
281         if (unlikely(skb_checksum_start(gso_skb) !=
282                      skb_transport_header(gso_skb) &&
283                      !(skb_shinfo(gso_skb)->gso_type & SKB_GSO_FRAGLIST)))
284                 return ERR_PTR(-EINVAL);
285 
286         /* We don't know if egress device can segment and checksum the packet
287          * when IPv6 extension headers are present. Fall back to software GSO.
288          */
289         if (gso_skb->ip_summed != CHECKSUM_PARTIAL)
290                 features &= ~(NETIF_F_GSO_UDP_L4 | NETIF_F_CSUM_MASK);
291 
292         if (skb_gso_ok(gso_skb, features | NETIF_F_GSO_ROBUST)) {
293                 /* Packet is from an untrusted source, reset gso_segs. */
294                 skb_shinfo(gso_skb)->gso_segs = DIV_ROUND_UP(gso_skb->len - sizeof(*uh),
295                                                              mss);
296                 return NULL;
297         }
298 
299         if (skb_shinfo(gso_skb)->gso_type & SKB_GSO_FRAGLIST) {
300                  /* Detect modified geometry and pass those to skb_segment. */
301                 if (skb_pagelen(gso_skb) - sizeof(*uh) == skb_shinfo(gso_skb)->gso_size)
302                         return __udp_gso_segment_list(gso_skb, features, is_ipv6);
303 
304                  /* Setup csum, as fraglist skips this in udp4_gro_receive. */
305                 gso_skb->csum_start = skb_transport_header(gso_skb) - gso_skb->head;
306                 gso_skb->csum_offset = offsetof(struct udphdr, check);
307                 gso_skb->ip_summed = CHECKSUM_PARTIAL;
308 
309                 uh = udp_hdr(gso_skb);
310                 if (is_ipv6)
311                         uh->check = ~udp_v6_check(gso_skb->len,
312                                                   &ipv6_hdr(gso_skb)->saddr,
313                                                   &ipv6_hdr(gso_skb)->daddr, 0);
314                 else
315                         uh->check = ~udp_v4_check(gso_skb->len,
316                                                   ip_hdr(gso_skb)->saddr,
317                                                   ip_hdr(gso_skb)->daddr, 0);
318         }
319 
320         skb_pull(gso_skb, sizeof(*uh));
321 
322         /* clear destructor to avoid skb_segment assigning it to tail */
323         copy_dtor = gso_skb->destructor == sock_wfree;
324         if (copy_dtor)
325                 gso_skb->destructor = NULL;
326 
327         segs = skb_segment(gso_skb, features);
328         if (IS_ERR_OR_NULL(segs)) {
329                 if (copy_dtor)
330                         gso_skb->destructor = sock_wfree;
331                 return segs;
332         }
333 
334         /* GSO partial and frag_list segmentation only requires splitting
335          * the frame into an MSS multiple and possibly a remainder, both
336          * cases return a GSO skb. So update the mss now.
337          */
338         if (skb_is_gso(segs))
339                 mss *= skb_shinfo(segs)->gso_segs;
340 
341         seg = segs;
342         uh = udp_hdr(seg);
343 
344         /* preserve TX timestamp flags and TS key for first segment */
345         skb_shinfo(seg)->tskey = skb_shinfo(gso_skb)->tskey;
346         skb_shinfo(seg)->tx_flags |=
347                         (skb_shinfo(gso_skb)->tx_flags & SKBTX_ANY_TSTAMP);
348 
349         /* compute checksum adjustment based on old length versus new */
350         newlen = htons(sizeof(*uh) + mss);
351         check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
352 
353         for (;;) {
354                 if (copy_dtor) {
355                         seg->destructor = sock_wfree;
356                         seg->sk = sk;
357                         sum_truesize += seg->truesize;
358                 }
359 
360                 if (!seg->next)
361                         break;
362 
363                 uh->len = newlen;
364                 uh->check = check;
365 
366                 if (seg->ip_summed == CHECKSUM_PARTIAL)
367                         gso_reset_checksum(seg, ~check);
368                 else
369                         uh->check = gso_make_checksum(seg, ~check) ? :
370                                     CSUM_MANGLED_0;
371 
372                 seg = seg->next;
373                 uh = udp_hdr(seg);
374         }
375 
376         /* last packet can be partial gso_size, account for that in checksum */
377         newlen = htons(skb_tail_pointer(seg) - skb_transport_header(seg) +
378                        seg->data_len);
379         check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
380 
381         uh->len = newlen;
382         uh->check = check;
383 
384         if (seg->ip_summed == CHECKSUM_PARTIAL)
385                 gso_reset_checksum(seg, ~check);
386         else
387                 uh->check = gso_make_checksum(seg, ~check) ? : CSUM_MANGLED_0;
388 
389         /* On the TX path, CHECKSUM_NONE and CHECKSUM_UNNECESSARY have the same
390          * meaning. However, check for bad offloads in the GSO stack expects the
391          * latter, if the checksum was calculated in software. To vouch for the
392          * segment skbs we actually need to set it on the gso_skb.
393          */
394         if (gso_skb->ip_summed == CHECKSUM_NONE)
395                 gso_skb->ip_summed = CHECKSUM_UNNECESSARY;
396 
397         /* update refcount for the packet */
398         if (copy_dtor) {
399                 int delta = sum_truesize - gso_skb->truesize;
400 
401                 /* In some pathological cases, delta can be negative.
402                  * We need to either use refcount_add() or refcount_sub_and_test()
403                  */
404                 if (likely(delta >= 0))
405                         refcount_add(delta, &sk->sk_wmem_alloc);
406                 else
407                         WARN_ON_ONCE(refcount_sub_and_test(-delta, &sk->sk_wmem_alloc));
408         }
409         return segs;
410 }
411 EXPORT_SYMBOL_GPL(__udp_gso_segment);
412 
413 static struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
414                                          netdev_features_t features)
415 {
416         struct sk_buff *segs = ERR_PTR(-EINVAL);
417         unsigned int mss;
418         __wsum csum;
419         struct udphdr *uh;
420         struct iphdr *iph;
421 
422         if (skb->encapsulation &&
423             (skb_shinfo(skb)->gso_type &
424              (SKB_GSO_UDP_TUNNEL|SKB_GSO_UDP_TUNNEL_CSUM))) {
425                 segs = skb_udp_tunnel_segment(skb, features, false);
426                 goto out;
427         }
428 
429         if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_UDP | SKB_GSO_UDP_L4)))
430                 goto out;
431 
432         if (!pskb_may_pull(skb, sizeof(struct udphdr)))
433                 goto out;
434 
435         if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4)
436                 return __udp_gso_segment(skb, features, false);
437 
438         mss = skb_shinfo(skb)->gso_size;
439         if (unlikely(skb->len <= mss))
440                 goto out;
441 
442         /* Do software UFO. Complete and fill in the UDP checksum as
443          * HW cannot do checksum of UDP packets sent as multiple
444          * IP fragments.
445          */
446 
447         uh = udp_hdr(skb);
448         iph = ip_hdr(skb);
449 
450         uh->check = 0;
451         csum = skb_checksum(skb, 0, skb->len, 0);
452         uh->check = udp_v4_check(skb->len, iph->saddr, iph->daddr, csum);
453         if (uh->check == 0)
454                 uh->check = CSUM_MANGLED_0;
455 
456         skb->ip_summed = CHECKSUM_UNNECESSARY;
457 
458         /* If there is no outer header we can fake a checksum offload
459          * due to the fact that we have already done the checksum in
460          * software prior to segmenting the frame.
461          */
462         if (!skb->encap_hdr_csum)
463                 features |= NETIF_F_HW_CSUM;
464 
465         /* Fragment the skb. IP headers of the fragments are updated in
466          * inet_gso_segment()
467          */
468         segs = skb_segment(skb, features);
469 out:
470         return segs;
471 }
472 
473 
474 #define UDP_GRO_CNT_MAX 64
475 static struct sk_buff *udp_gro_receive_segment(struct list_head *head,
476                                                struct sk_buff *skb)
477 {
478         struct udphdr *uh = udp_gro_udphdr(skb);
479         struct sk_buff *pp = NULL;
480         struct udphdr *uh2;
481         struct sk_buff *p;
482         unsigned int ulen;
483         int ret = 0;
484         int flush;
485 
486         /* requires non zero csum, for symmetry with GSO */
487         if (!uh->check) {
488                 NAPI_GRO_CB(skb)->flush = 1;
489                 return NULL;
490         }
491 
492         /* Do not deal with padded or malicious packets, sorry ! */
493         ulen = ntohs(uh->len);
494         if (ulen <= sizeof(*uh) || ulen != skb_gro_len(skb)) {
495                 NAPI_GRO_CB(skb)->flush = 1;
496                 return NULL;
497         }
498         /* pull encapsulating udp header */
499         skb_gro_pull(skb, sizeof(struct udphdr));
500 
501         list_for_each_entry(p, head, list) {
502                 if (!NAPI_GRO_CB(p)->same_flow)
503                         continue;
504 
505                 uh2 = udp_hdr(p);
506 
507                 /* Match ports only, as csum is always non zero */
508                 if ((*(u32 *)&uh->source != *(u32 *)&uh2->source)) {
509                         NAPI_GRO_CB(p)->same_flow = 0;
510                         continue;
511                 }
512 
513                 if (NAPI_GRO_CB(skb)->is_flist != NAPI_GRO_CB(p)->is_flist) {
514                         NAPI_GRO_CB(skb)->flush = 1;
515                         return p;
516                 }
517 
518                 flush = gro_receive_network_flush(uh, uh2, p);
519 
520                 /* Terminate the flow on len mismatch or if it grow "too much".
521                  * Under small packet flood GRO count could elsewhere grow a lot
522                  * leading to excessive truesize values.
523                  * On len mismatch merge the first packet shorter than gso_size,
524                  * otherwise complete the GRO packet.
525                  */
526                 if (ulen > ntohs(uh2->len) || flush) {
527                         pp = p;
528                 } else {
529                         if (NAPI_GRO_CB(skb)->is_flist) {
530                                 if (!pskb_may_pull(skb, skb_gro_offset(skb))) {
531                                         NAPI_GRO_CB(skb)->flush = 1;
532                                         return NULL;
533                                 }
534                                 if ((skb->ip_summed != p->ip_summed) ||
535                                     (skb->csum_level != p->csum_level)) {
536                                         NAPI_GRO_CB(skb)->flush = 1;
537                                         return NULL;
538                                 }
539                                 ret = skb_gro_receive_list(p, skb);
540                         } else {
541                                 skb_gro_postpull_rcsum(skb, uh,
542                                                        sizeof(struct udphdr));
543 
544                                 ret = skb_gro_receive(p, skb);
545                         }
546                 }
547 
548                 if (ret || ulen != ntohs(uh2->len) ||
549                     NAPI_GRO_CB(p)->count >= UDP_GRO_CNT_MAX)
550                         pp = p;
551 
552                 return pp;
553         }
554 
555         /* mismatch, but we never need to flush */
556         return NULL;
557 }
558 
559 struct sk_buff *udp_gro_receive(struct list_head *head, struct sk_buff *skb,
560                                 struct udphdr *uh, struct sock *sk)
561 {
562         struct sk_buff *pp = NULL;
563         struct sk_buff *p;
564         struct udphdr *uh2;
565         unsigned int off = skb_gro_offset(skb);
566         int flush = 1;
567 
568         /* We can do L4 aggregation only if the packet can't land in a tunnel
569          * otherwise we could corrupt the inner stream. Detecting such packets
570          * cannot be foolproof and the aggregation might still happen in some
571          * cases. Such packets should be caught in udp_unexpected_gso later.
572          */
573         NAPI_GRO_CB(skb)->is_flist = 0;
574         if (!sk || !udp_sk(sk)->gro_receive) {
575                 /* If the packet was locally encapsulated in a UDP tunnel that
576                  * wasn't detected above, do not GRO.
577                  */
578                 if (skb->encapsulation)
579                         goto out;
580 
581                 if (skb->dev->features & NETIF_F_GRO_FRAGLIST)
582                         NAPI_GRO_CB(skb)->is_flist = sk ? !udp_test_bit(GRO_ENABLED, sk) : 1;
583 
584                 if ((!sk && (skb->dev->features & NETIF_F_GRO_UDP_FWD)) ||
585                     (sk && udp_test_bit(GRO_ENABLED, sk)) || NAPI_GRO_CB(skb)->is_flist)
586                         return call_gro_receive(udp_gro_receive_segment, head, skb);
587 
588                 /* no GRO, be sure flush the current packet */
589                 goto out;
590         }
591 
592         if (NAPI_GRO_CB(skb)->encap_mark ||
593             (uh->check && skb->ip_summed != CHECKSUM_PARTIAL &&
594              NAPI_GRO_CB(skb)->csum_cnt == 0 &&
595              !NAPI_GRO_CB(skb)->csum_valid))
596                 goto out;
597 
598         /* mark that this skb passed once through the tunnel gro layer */
599         NAPI_GRO_CB(skb)->encap_mark = 1;
600 
601         flush = 0;
602 
603         list_for_each_entry(p, head, list) {
604                 if (!NAPI_GRO_CB(p)->same_flow)
605                         continue;
606 
607                 uh2 = (struct udphdr   *)(p->data + off);
608 
609                 /* Match ports and either checksums are either both zero
610                  * or nonzero.
611                  */
612                 if ((*(u32 *)&uh->source != *(u32 *)&uh2->source) ||
613                     (!uh->check ^ !uh2->check)) {
614                         NAPI_GRO_CB(p)->same_flow = 0;
615                         continue;
616                 }
617         }
618 
619         skb_gro_pull(skb, sizeof(struct udphdr)); /* pull encapsulating udp header */
620         skb_gro_postpull_rcsum(skb, uh, sizeof(struct udphdr));
621         pp = call_gro_receive_sk(udp_sk(sk)->gro_receive, sk, head, skb);
622 
623 out:
624         skb_gro_flush_final(skb, pp, flush);
625         return pp;
626 }
627 EXPORT_SYMBOL(udp_gro_receive);
628 
629 static struct sock *udp4_gro_lookup_skb(struct sk_buff *skb, __be16 sport,
630                                         __be16 dport)
631 {
632         const struct iphdr *iph = skb_gro_network_header(skb);
633         struct net *net = dev_net(skb->dev);
634         int iif, sdif;
635 
636         inet_get_iif_sdif(skb, &iif, &sdif);
637 
638         return __udp4_lib_lookup(net, iph->saddr, sport,
639                                  iph->daddr, dport, iif,
640                                  sdif, net->ipv4.udp_table, NULL);
641 }
642 
643 INDIRECT_CALLABLE_SCOPE
644 struct sk_buff *udp4_gro_receive(struct list_head *head, struct sk_buff *skb)
645 {
646         struct udphdr *uh = udp_gro_udphdr(skb);
647         struct sock *sk = NULL;
648         struct sk_buff *pp;
649 
650         if (unlikely(!uh))
651                 goto flush;
652 
653         /* Don't bother verifying checksum if we're going to flush anyway. */
654         if (NAPI_GRO_CB(skb)->flush)
655                 goto skip;
656 
657         if (skb_gro_checksum_validate_zero_check(skb, IPPROTO_UDP, uh->check,
658                                                  inet_gro_compute_pseudo))
659                 goto flush;
660         else if (uh->check)
661                 skb_gro_checksum_try_convert(skb, IPPROTO_UDP,
662                                              inet_gro_compute_pseudo);
663 skip:
664         NAPI_GRO_CB(skb)->is_ipv6 = 0;
665 
666         if (static_branch_unlikely(&udp_encap_needed_key))
667                 sk = udp4_gro_lookup_skb(skb, uh->source, uh->dest);
668 
669         pp = udp_gro_receive(head, skb, uh, sk);
670         return pp;
671 
672 flush:
673         NAPI_GRO_CB(skb)->flush = 1;
674         return NULL;
675 }
676 
677 static int udp_gro_complete_segment(struct sk_buff *skb)
678 {
679         struct udphdr *uh = udp_hdr(skb);
680 
681         skb->csum_start = (unsigned char *)uh - skb->head;
682         skb->csum_offset = offsetof(struct udphdr, check);
683         skb->ip_summed = CHECKSUM_PARTIAL;
684 
685         skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
686         skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_L4;
687 
688         if (skb->encapsulation)
689                 skb->inner_transport_header = skb->transport_header;
690 
691         return 0;
692 }
693 
694 int udp_gro_complete(struct sk_buff *skb, int nhoff,
695                      udp_lookup_t lookup)
696 {
697         __be16 newlen = htons(skb->len - nhoff);
698         struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
699         struct sock *sk;
700         int err;
701 
702         uh->len = newlen;
703 
704         sk = INDIRECT_CALL_INET(lookup, udp6_lib_lookup_skb,
705                                 udp4_lib_lookup_skb, skb, uh->source, uh->dest);
706         if (sk && udp_sk(sk)->gro_complete) {
707                 skb_shinfo(skb)->gso_type = uh->check ? SKB_GSO_UDP_TUNNEL_CSUM
708                                         : SKB_GSO_UDP_TUNNEL;
709 
710                 /* clear the encap mark, so that inner frag_list gro_complete
711                  * can take place
712                  */
713                 NAPI_GRO_CB(skb)->encap_mark = 0;
714 
715                 /* Set encapsulation before calling into inner gro_complete()
716                  * functions to make them set up the inner offsets.
717                  */
718                 skb->encapsulation = 1;
719                 err = udp_sk(sk)->gro_complete(sk, skb,
720                                 nhoff + sizeof(struct udphdr));
721         } else {
722                 err = udp_gro_complete_segment(skb);
723         }
724 
725         if (skb->remcsum_offload)
726                 skb_shinfo(skb)->gso_type |= SKB_GSO_TUNNEL_REMCSUM;
727 
728         return err;
729 }
730 EXPORT_SYMBOL(udp_gro_complete);
731 
732 INDIRECT_CALLABLE_SCOPE int udp4_gro_complete(struct sk_buff *skb, int nhoff)
733 {
734         const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
735         const struct iphdr *iph = (struct iphdr *)(skb->data + offset);
736         struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
737 
738         /* do fraglist only if there is no outer UDP encap (or we already processed it) */
739         if (NAPI_GRO_CB(skb)->is_flist && !NAPI_GRO_CB(skb)->encap_mark) {
740                 uh->len = htons(skb->len - nhoff);
741 
742                 skb_shinfo(skb)->gso_type |= (SKB_GSO_FRAGLIST|SKB_GSO_UDP_L4);
743                 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
744 
745                 __skb_incr_checksum_unnecessary(skb);
746 
747                 return 0;
748         }
749 
750         if (uh->check)
751                 uh->check = ~udp_v4_check(skb->len - nhoff, iph->saddr,
752                                           iph->daddr, 0);
753 
754         return udp_gro_complete(skb, nhoff, udp4_lib_lookup_skb);
755 }
756 
757 int __init udpv4_offload_init(void)
758 {
759         net_hotdata.udpv4_offload = (struct net_offload) {
760                 .callbacks = {
761                         .gso_segment = udp4_ufo_fragment,
762                         .gro_receive  = udp4_gro_receive,
763                         .gro_complete = udp4_gro_complete,
764                 },
765         };
766         return inet_add_offload(&net_hotdata.udpv4_offload, IPPROTO_UDP);
767 }
768 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

sflogo.php