change system file

hc

2024-10-09 244b2c5ca8b14627e4a17755e5922221e121c771

 kernel/net/ipv4/ip_output.c
..
..
@@ -1,3 +1,4 @@
1
+// SPDX-License-Identifier: GPL-2.0-only
1
2
 /*
2
3
  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3
4
  *		operating system.  INET is implemented using the  BSD Socket
..
..
@@ -142,7 +143,8 @@
142
143
  *
143
144
  */
144
145
 int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
145
-			  __be32 saddr, __be32 daddr, struct ip_options_rcu *opt)
146
+			  __be32 saddr, __be32 daddr, struct ip_options_rcu *opt,
147
+			  u8 tos)
146
148
 {
147
149
 	struct inet_sock *inet = inet_sk(sk);
148
150
 	struct rtable *rt = skb_rtable(skb);
..
..
@@ -155,7 +157,7 @@
155
157
 	iph = ip_hdr(skb);
156
158
 	iph->version  = 4;
157
159
 	iph->ihl      = 5;
158
-	iph->tos      = inet->tos;
160
+	iph->tos      = tos;
159
161
 	iph->ttl      = ip_select_ttl(inet, &rt->dst);
160
162
 	iph->daddr    = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
161
163
 	iph->saddr    = saddr;
..
..
@@ -196,7 +198,7 @@
196
198
 	struct net_device *dev = dst->dev;
197
199
 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
198
200
 	struct neighbour *neigh;
199
-	u32 nexthop;
201
+	bool is_v6gw = false;
200
202
 
201
203
 	if (rt->rt_type == RTN_MULTICAST) {
202
204
 		IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len);
..
..
@@ -221,21 +223,18 @@
221
223
 	if (lwtunnel_xmit_redirect(dst->lwtstate)) {
222
224
 		int res = lwtunnel_xmit(skb);
223
225
 
224
-		if (res < 0 || res == LWTUNNEL_XMIT_DONE)
226
+		if (res != LWTUNNEL_XMIT_CONTINUE)
225
227
 			return res;
226
228
 	}
227
229
 
228
230
 	rcu_read_lock_bh();
229
-	nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr);
230
-	neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
231
-	if (unlikely(!neigh))
232
-		neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
231
+	neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
233
232
 	if (!IS_ERR(neigh)) {
234
233
 		int res;
235
234
 
236
235
 		sock_confirm_neigh(skb, neigh);
237
-		res = neigh_output(neigh, skb);
238
-
236
+		/* if crossing protocols, can not use the cached header */
237
+		res = neigh_output(neigh, skb, is_v6gw);
239
238
 		rcu_read_unlock_bh();
240
239
 		return res;
241
240
 	}
..
..
@@ -250,8 +249,8 @@
250
249
 static int ip_finish_output_gso(struct net *net, struct sock *sk,
251
250
 				struct sk_buff *skb, unsigned int mtu)
252
251
 {
252
+	struct sk_buff *segs, *nskb;
253
253
 	netdev_features_t features;
254
-	struct sk_buff *segs;
255
254
 	int ret = 0;
256
255
 
257
256
 	/* common case: seglen is <= mtu
..
..
@@ -273,7 +272,7 @@
273
272
 	 *    insufficent MTU.
274
273
 	 */
275
274
 	features = netif_skb_features(skb);
276
-	BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_SGO_CB_OFFSET);
275
+	BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_GSO_CB_OFFSET);
277
276
 	segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
278
277
 	if (IS_ERR_OR_NULL(segs)) {
279
278
 		kfree_skb(skb);
..
..
@@ -282,31 +281,22 @@
282
281
 
283
282
 	consume_skb(skb);
284
283
 
285
-	do {
286
-		struct sk_buff *nskb = segs->next;
284
+	skb_list_walk_safe(segs, segs, nskb) {
287
285
 		int err;
288
286
 
289
-		segs->next = NULL;
287
+		skb_mark_not_on_list(segs);
290
288
 		err = ip_fragment(net, sk, segs, mtu, ip_finish_output2);
291
289
 
292
290
 		if (err && ret == 0)
293
291
 			ret = err;
294
-		segs = nskb;
295
-	} while (segs);
292
+	}
296
293
 
297
294
 	return ret;
298
295
 }
299
296
 
300
-static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
297
+static int __ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
301
298
 {
302
299
 	unsigned int mtu;
303
-	int ret;
304
-
305
-	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
306
-	if (ret) {
307
-		kfree_skb(skb);
308
-		return ret;
309
-	}
310
300
 
311
301
 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
312
302
 	/* Policy lookup after SNAT yielded a new policy */
..
..
@@ -325,18 +315,54 @@
325
315
 	return ip_finish_output2(net, sk, skb);
326
316
 }
327
317
 
328
-static int ip_mc_finish_output(struct net *net, struct sock *sk,
329
-			       struct sk_buff *skb)
318
+static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
330
319
 {
331
320
 	int ret;
332
321
 
333
322
 	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
334
-	if (ret) {
323
+	switch (ret) {
324
+	case NET_XMIT_SUCCESS:
325
+		return __ip_finish_output(net, sk, skb);
326
+	case NET_XMIT_CN:
327
+		return __ip_finish_output(net, sk, skb) ? : ret;
328
+	default:
329
+		kfree_skb(skb);
330
+		return ret;
331
+	}
332
+}
333
+
334
+static int ip_mc_finish_output(struct net *net, struct sock *sk,
335
+			       struct sk_buff *skb)
336
+{
337
+	struct rtable *new_rt;
338
+	bool do_cn = false;
339
+	int ret, err;
340
+
341
+	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
342
+	switch (ret) {
343
+	case NET_XMIT_CN:
344
+		do_cn = true;
345
+		fallthrough;
346
+	case NET_XMIT_SUCCESS:
347
+		break;
348
+	default:
335
349
 		kfree_skb(skb);
336
350
 		return ret;
337
351
 	}
338
352
 
339
-	return dev_loopback_xmit(net, sk, skb);
353
+	/* Reset rt_iif so that inet_iif() will return skb->skb_iif. Setting
354
+	 * this to non-zero causes ipi_ifindex in in_pktinfo to be overwritten,
355
+	 * see ipv4_pktinfo_prepare().
356
+	 */
357
+	new_rt = rt_dst_clone(net->loopback_dev, skb_rtable(skb));
358
+	if (new_rt) {
359
+		new_rt->rt_iif = 0;
360
+		skb_dst_drop(skb);
361
+		skb_dst_set(skb, &new_rt->dst);
362
+	}
363
+
364
+	err = dev_loopback_xmit(net, sk, skb);
365
+	return (do_cn && err) ? ret : err;
340
366
 }
341
367
 
342
368
 int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb)
..
..
@@ -403,7 +429,7 @@
403
429
 
404
430
 int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb)
405
431
 {
406
-	struct net_device *dev = skb_dst(skb)->dev;
432
+	struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
407
433
 
408
434
 	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
409
435
 
..
..
@@ -411,7 +437,7 @@
411
437
 	skb->protocol = htons(ETH_P_IP);
412
438
 
413
439
 	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
414
-			    net, sk, skb, NULL, dev,
440
+			    net, sk, skb, indev, dev,
415
441
 			    ip_finish_output,
416
442
 			    !(IPCB(skb)->flags & IPSKB_REROUTED));
417
443
 }
..
..
@@ -523,6 +549,12 @@
523
549
 }
524
550
 EXPORT_SYMBOL(__ip_queue_xmit);
525
551
 
552
+int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
553
+{
554
+	return __ip_queue_xmit(sk, skb, fl, inet_sk(sk)->tos);
555
+}
556
+EXPORT_SYMBOL(ip_queue_xmit);
557
+
526
558
 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
527
559
 {
528
560
 	to->pkt_type = from->pkt_type;
..
..
@@ -536,13 +568,11 @@
536
568
 
537
569
 	skb_copy_hash(to, from);
538
570
 
539
-	/* Copy the flags to each fragment. */
540
-	IPCB(to)->flags = IPCB(from)->flags;
541
-
542
571
 #ifdef CONFIG_NET_SCHED
543
572
 	to->tc_index = from->tc_index;
544
573
 #endif
545
574
 	nf_copy(to, from);
575
+	skb_ext_copy(to, from);
546
576
 #if IS_ENABLED(CONFIG_IP_VS)
547
577
 	to->ipvs_property = from->ipvs_property;
548
578
 #endif
..
..
@@ -571,6 +601,163 @@
571
601
 	return ip_do_fragment(net, sk, skb, output);
572
602
 }
573
603
 
604
+void ip_fraglist_init(struct sk_buff *skb, struct iphdr *iph,
605
+		      unsigned int hlen, struct ip_fraglist_iter *iter)
606
+{
607
+	unsigned int first_len = skb_pagelen(skb);
608
+
609
+	iter->frag = skb_shinfo(skb)->frag_list;
610
+	skb_frag_list_init(skb);
611
+
612
+	iter->offset = 0;
613
+	iter->iph = iph;
614
+	iter->hlen = hlen;
615
+
616
+	skb->data_len = first_len - skb_headlen(skb);
617
+	skb->len = first_len;
618
+	iph->tot_len = htons(first_len);
619
+	iph->frag_off = htons(IP_MF);
620
+	ip_send_check(iph);
621
+}
622
+EXPORT_SYMBOL(ip_fraglist_init);
623
+
624
+void ip_fraglist_prepare(struct sk_buff *skb, struct ip_fraglist_iter *iter)
625
+{
626
+	unsigned int hlen = iter->hlen;
627
+	struct iphdr *iph = iter->iph;
628
+	struct sk_buff *frag;
629
+
630
+	frag = iter->frag;
631
+	frag->ip_summed = CHECKSUM_NONE;
632
+	skb_reset_transport_header(frag);
633
+	__skb_push(frag, hlen);
634
+	skb_reset_network_header(frag);
635
+	memcpy(skb_network_header(frag), iph, hlen);
636
+	iter->iph = ip_hdr(frag);
637
+	iph = iter->iph;
638
+	iph->tot_len = htons(frag->len);
639
+	ip_copy_metadata(frag, skb);
640
+	iter->offset += skb->len - hlen;
641
+	iph->frag_off = htons(iter->offset >> 3);
642
+	if (frag->next)
643
+		iph->frag_off |= htons(IP_MF);
644
+	/* Ready, complete checksum */
645
+	ip_send_check(iph);
646
+}
647
+EXPORT_SYMBOL(ip_fraglist_prepare);
648
+
649
+void ip_frag_init(struct sk_buff *skb, unsigned int hlen,
650
+		  unsigned int ll_rs, unsigned int mtu, bool DF,
651
+		  struct ip_frag_state *state)
652
+{
653
+	struct iphdr *iph = ip_hdr(skb);
654
+
655
+	state->DF = DF;
656
+	state->hlen = hlen;
657
+	state->ll_rs = ll_rs;
658
+	state->mtu = mtu;
659
+
660
+	state->left = skb->len - hlen;	/* Space per frame */
661
+	state->ptr = hlen;		/* Where to start from */
662
+
663
+	state->offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
664
+	state->not_last_frag = iph->frag_off & htons(IP_MF);
665
+}
666
+EXPORT_SYMBOL(ip_frag_init);
667
+
668
+static void ip_frag_ipcb(struct sk_buff *from, struct sk_buff *to,
669
+			 bool first_frag)
670
+{
671
+	/* Copy the flags to each fragment. */
672
+	IPCB(to)->flags = IPCB(from)->flags;
673
+
674
+	/* ANK: dirty, but effective trick. Upgrade options only if
675
+	 * the segment to be fragmented was THE FIRST (otherwise,
676
+	 * options are already fixed) and make it ONCE
677
+	 * on the initial skb, so that all the following fragments
678
+	 * will inherit fixed options.
679
+	 */
680
+	if (first_frag)
681
+		ip_options_fragment(from);
682
+}
683
+
684
+struct sk_buff *ip_frag_next(struct sk_buff *skb, struct ip_frag_state *state)
685
+{
686
+	unsigned int len = state->left;
687
+	struct sk_buff *skb2;
688
+	struct iphdr *iph;
689
+
690
+	len = state->left;
691
+	/* IF: it doesn't fit, use 'mtu' - the data space left */
692
+	if (len > state->mtu)
693
+		len = state->mtu;
694
+	/* IF: we are not sending up to and including the packet end
695
+	   then align the next start on an eight byte boundary */
696
+	if (len < state->left)	{
697
+		len &= ~7;
698
+	}
699
+
700
+	/* Allocate buffer */
701
+	skb2 = alloc_skb(len + state->hlen + state->ll_rs, GFP_ATOMIC);
702
+	if (!skb2)
703
+		return ERR_PTR(-ENOMEM);
704
+
705
+	/*
706
+	 *	Set up data on packet
707
+	 */
708
+
709
+	ip_copy_metadata(skb2, skb);
710
+	skb_reserve(skb2, state->ll_rs);
711
+	skb_put(skb2, len + state->hlen);
712
+	skb_reset_network_header(skb2);
713
+	skb2->transport_header = skb2->network_header + state->hlen;
714
+
715
+	/*
716
+	 *	Charge the memory for the fragment to any owner
717
+	 *	it might possess
718
+	 */
719
+
720
+	if (skb->sk)
721
+		skb_set_owner_w(skb2, skb->sk);
722
+
723
+	/*
724
+	 *	Copy the packet header into the new buffer.
725
+	 */
726
+
727
+	skb_copy_from_linear_data(skb, skb_network_header(skb2), state->hlen);
728
+
729
+	/*
730
+	 *	Copy a block of the IP datagram.
731
+	 */
732
+	if (skb_copy_bits(skb, state->ptr, skb_transport_header(skb2), len))
733
+		BUG();
734
+	state->left -= len;
735
+
736
+	/*
737
+	 *	Fill in the new header fields.
738
+	 */
739
+	iph = ip_hdr(skb2);
740
+	iph->frag_off = htons((state->offset >> 3));
741
+	if (state->DF)
742
+		iph->frag_off |= htons(IP_DF);
743
+
744
+	/*
745
+	 *	Added AC : If we are fragmenting a fragment that's not the
746
+	 *		   last fragment then keep MF on each bit
747
+	 */
748
+	if (state->left > 0 || state->not_last_frag)
749
+		iph->frag_off |= htons(IP_MF);
750
+	state->ptr += len;
751
+	state->offset += len;
752
+
753
+	iph->tot_len = htons(len + state->hlen);
754
+
755
+	ip_send_check(iph);
756
+
757
+	return skb2;
758
+}
759
+EXPORT_SYMBOL(ip_frag_next);
760
+
574
761
 /*
575
762
  *	This IP datagram is too large to be sent in one piece.  Break it up into
576
763
  *	smaller pieces (each of size equal to IP header plus
..
..
@@ -582,12 +769,12 @@
582
769
 		   int (*output)(struct net *, struct sock *, struct sk_buff *))
583
770
 {
584
771
 	struct iphdr *iph;
585
-	int ptr;
586
772
 	struct sk_buff *skb2;
587
-	unsigned int mtu, hlen, left, len, ll_rs;
588
-	int offset;
589
-	__be16 not_last_frag;
590
773
 	struct rtable *rt = skb_rtable(skb);
774
+	unsigned int mtu, hlen, ll_rs;
775
+	struct ip_fraglist_iter iter;
776
+	ktime_t tstamp = skb->tstamp;
777
+	struct ip_frag_state state;
591
778
 	int err = 0;
592
779
 
593
780
 	/* for offloaded checksums cleanup checksum before fragmentation */
..
..
@@ -652,49 +839,37 @@
652
839
 		}
653
840
 
654
841
 		/* Everything is OK. Generate! */
655
-
656
-		err = 0;
657
-		offset = 0;
658
-		frag = skb_shinfo(skb)->frag_list;
659
-		skb_frag_list_init(skb);
660
-		skb->data_len = first_len - skb_headlen(skb);
661
-		skb->len = first_len;
662
-		iph->tot_len = htons(first_len);
663
-		iph->frag_off = htons(IP_MF);
664
-		ip_send_check(iph);
842
+		ip_fraglist_init(skb, iph, hlen, &iter);
665
843
 
666
844
 		for (;;) {
667
845
 			/* Prepare header of the next frame,
668
846
 			 * before previous one went down. */
669
-			if (frag) {
670
-				frag->ip_summed = CHECKSUM_NONE;
671
-				skb_reset_transport_header(frag);
672
-				__skb_push(frag, hlen);
673
-				skb_reset_network_header(frag);
674
-				memcpy(skb_network_header(frag), iph, hlen);
675
-				iph = ip_hdr(frag);
676
-				iph->tot_len = htons(frag->len);
677
-				ip_copy_metadata(frag, skb);
678
-				if (offset == 0)
679
-					ip_options_fragment(frag);
680
-				offset += skb->len - hlen;
681
-				iph->frag_off = htons(offset>>3);
682
-				if (frag->next)
683
-					iph->frag_off |= htons(IP_MF);
684
-				/* Ready, complete checksum */
685
-				ip_send_check(iph);
847
+			if (iter.frag) {
848
+				bool first_frag = (iter.offset == 0);
849
+
850
+				IPCB(iter.frag)->flags = IPCB(skb)->flags;
851
+				ip_fraglist_prepare(skb, &iter);
852
+				if (first_frag && IPCB(skb)->opt.optlen) {
853
+					/* ipcb->opt is not populated for frags
854
+					 * coming from __ip_make_skb(),
855
+					 * ip_options_fragment() needs optlen
856
+					 */
857
+					IPCB(iter.frag)->opt.optlen =
858
+						IPCB(skb)->opt.optlen;
859
+					ip_options_fragment(iter.frag);
860
+					ip_send_check(iter.iph);
861
+				}
686
862
 			}
687
863
 
864
+			skb->tstamp = tstamp;
688
865
 			err = output(net, sk, skb);
689
866
 
690
867
 			if (!err)
691
868
 				IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
692
-			if (err || !frag)
869
+			if (err || !iter.frag)
693
870
 				break;
694
871
 
695
-			skb = frag;
696
-			frag = skb->next;
697
-			skb->next = NULL;
872
+			skb = ip_fraglist_next(&iter);
698
873
 		}
699
874
 
700
875
 		if (err == 0) {
..
..
@@ -702,11 +877,8 @@
702
877
 			return 0;
703
878
 		}
704
879
 
705
-		while (frag) {
706
-			skb = frag->next;
707
-			kfree_skb(frag);
708
-			frag = skb;
709
-		}
880
+		kfree_skb_list(iter.frag);
881
+
710
882
 		IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
711
883
 		return err;
712
884
 
..
..
@@ -721,105 +893,31 @@
721
893
 	}
722
894
 
723
895
 slow_path:
724
-	iph = ip_hdr(skb);
725
-
726
-	left = skb->len - hlen;		/* Space per frame */
727
-	ptr = hlen;		/* Where to start from */
728
-
729
896
 	/*
730
897
 	 *	Fragment the datagram.
731
898
 	 */
732
899
 
733
-	offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
734
-	not_last_frag = iph->frag_off & htons(IP_MF);
900
+	ip_frag_init(skb, hlen, ll_rs, mtu, IPCB(skb)->flags & IPSKB_FRAG_PMTU,
901
+		     &state);
735
902
 
736
903
 	/*
737
904
 	 *	Keep copying data until we run out.
738
905
 	 */
739
906
 
740
-	while (left > 0) {
741
-		len = left;
742
-		/* IF: it doesn't fit, use 'mtu' - the data space left */
743
-		if (len > mtu)
744
-			len = mtu;
745
-		/* IF: we are not sending up to and including the packet end
746
-		   then align the next start on an eight byte boundary */
747
-		if (len < left)	{
748
-			len &= ~7;
749
-		}
907
+	while (state.left > 0) {
908
+		bool first_frag = (state.offset == 0);
750
909
 
751
-		/* Allocate buffer */
752
-		skb2 = alloc_skb(len + hlen + ll_rs, GFP_ATOMIC);
753
-		if (!skb2) {
754
-			err = -ENOMEM;
910
+		skb2 = ip_frag_next(skb, &state);
911
+		if (IS_ERR(skb2)) {
912
+			err = PTR_ERR(skb2);
755
913
 			goto fail;
756
914
 		}
757
-
758
-		/*
759
-		 *	Set up data on packet
760
-		 */
761
-
762
-		ip_copy_metadata(skb2, skb);
763
-		skb_reserve(skb2, ll_rs);
764
-		skb_put(skb2, len + hlen);
765
-		skb_reset_network_header(skb2);
766
-		skb2->transport_header = skb2->network_header + hlen;
767
-
768
-		/*
769
-		 *	Charge the memory for the fragment to any owner
770
-		 *	it might possess
771
-		 */
772
-
773
-		if (skb->sk)
774
-			skb_set_owner_w(skb2, skb->sk);
775
-
776
-		/*
777
-		 *	Copy the packet header into the new buffer.
778
-		 */
779
-
780
-		skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
781
-
782
-		/*
783
-		 *	Copy a block of the IP datagram.
784
-		 */
785
-		if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
786
-			BUG();
787
-		left -= len;
788
-
789
-		/*
790
-		 *	Fill in the new header fields.
791
-		 */
792
-		iph = ip_hdr(skb2);
793
-		iph->frag_off = htons((offset >> 3));
794
-
795
-		if (IPCB(skb)->flags & IPSKB_FRAG_PMTU)
796
-			iph->frag_off |= htons(IP_DF);
797
-
798
-		/* ANK: dirty, but effective trick. Upgrade options only if
799
-		 * the segment to be fragmented was THE FIRST (otherwise,
800
-		 * options are already fixed) and make it ONCE
801
-		 * on the initial skb, so that all the following fragments
802
-		 * will inherit fixed options.
803
-		 */
804
-		if (offset == 0)
805
-			ip_options_fragment(skb);
806
-
807
-		/*
808
-		 *	Added AC : If we are fragmenting a fragment that's not the
809
-		 *		   last fragment then keep MF on each bit
810
-		 */
811
-		if (left > 0 || not_last_frag)
812
-			iph->frag_off |= htons(IP_MF);
813
-		ptr += len;
814
-		offset += len;
915
+		ip_frag_ipcb(skb, skb2, first_frag);
815
916
 
816
917
 		/*
817
918
 		 *	Put this fragment into the sending queue.
818
919
 		 */
819
-		iph->tot_len = htons(len + hlen);
820
-
821
-		ip_send_check(iph);
822
-
920
+		skb2->tstamp = tstamp;
823
921
 		err = output(net, sk, skb2);
824
922
 		if (err)
825
923
 			goto fail;
..
..
@@ -877,6 +975,7 @@
877
975
 			    unsigned int flags)
878
976
 {
879
977
 	struct inet_sock *inet = inet_sk(sk);
978
+	struct ubuf_info *uarg = NULL;
880
979
 	struct sk_buff *skb;
881
980
 
882
981
 	struct ip_options *opt = cork->opt;
..
..
@@ -890,8 +989,8 @@
890
989
 	int csummode = CHECKSUM_NONE;
891
990
 	struct rtable *rt = (struct rtable *)cork->dst;
892
991
 	unsigned int wmem_alloc_delta = 0;
992
+	bool paged, extra_uref = false;
893
993
 	u32 tskey = 0;
894
-	bool paged;
895
994
 
896
995
 	skb = skb_peek_tail(queue);
897
996
 
..
..
@@ -907,7 +1006,7 @@
907
1006
 
908
1007
 	fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
909
1008
 	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
910
-	maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
1009
+	maxnonfragsize = ip_sk_ignore_df(sk) ? IP_MAX_MTU : mtu;
911
1010
 
912
1011
 	if (cork->length + length > maxnonfragsize - fragheaderlen) {
913
1012
 		ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
..
..
@@ -925,6 +1024,20 @@
925
1024
 	    (!(flags & MSG_MORE) || cork->gso_size) &&
926
1025
 	    (!exthdrlen || (rt->dst.dev->features & NETIF_F_HW_ESP_TX_CSUM)))
927
1026
 		csummode = CHECKSUM_PARTIAL;
1027
+
1028
+	if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) {
1029
+		uarg = sock_zerocopy_realloc(sk, length, skb_zcopy(skb));
1030
+		if (!uarg)
1031
+			return -ENOBUFS;
1032
+		extra_uref = !skb_zcopy(skb);	/* only ref on new uarg */
1033
+		if (rt->dst.dev->features & NETIF_F_SG &&
1034
+		    csummode == CHECKSUM_PARTIAL) {
1035
+			paged = true;
1036
+		} else {
1037
+			uarg->zerocopy = 0;
1038
+			skb_zcopy_set(skb, uarg, &extra_uref);
1039
+		}
1040
+	}
928
1041
 
929
1042
 	cork->length += length;
930
1043
 
..
..
@@ -1015,12 +1128,6 @@
1015
1128
 			skb->csum = 0;
1016
1129
 			skb_reserve(skb, hh_len);
1017
1130
 
1018
-			/* only the initial fragment is time stamped */
1019
-			skb_shinfo(skb)->tx_flags = cork->tx_flags;
1020
-			cork->tx_flags = 0;
1021
-			skb_shinfo(skb)->tskey = tskey;
1022
-			tskey = 0;
1023
-
1024
1131
 			/*
1025
1132
 			 *	Find where to start putting bytes.
1026
1133
 			 */
..
..
@@ -1033,7 +1140,7 @@
1033
1140
 			if (fraggap) {
1034
1141
 				skb->csum = skb_copy_and_csum_bits(
1035
1142
 					skb_prev, maxfraglen,
1036
-					data + transhdrlen, fraggap, 0);
1143
+					data + transhdrlen, fraggap);
1037
1144
 				skb_prev->csum = csum_sub(skb_prev->csum,
1038
1145
 							  skb->csum);
1039
1146
 				data += fraggap;
..
..
@@ -1052,6 +1159,13 @@
1052
1159
 			transhdrlen = 0;
1053
1160
 			exthdrlen = 0;
1054
1161
 			csummode = CHECKSUM_NONE;
1162
+
1163
+			/* only the initial fragment is time stamped */
1164
+			skb_shinfo(skb)->tx_flags = cork->tx_flags;
1165
+			cork->tx_flags = 0;
1166
+			skb_shinfo(skb)->tskey = tskey;
1167
+			tskey = 0;
1168
+			skb_zcopy_set(skb, uarg, &extra_uref);
1055
1169
 
1056
1170
 			if ((flags & MSG_CONFIRM) && !skb_prev)
1057
1171
 				skb_set_dst_pending_confirm(skb, 1);
..
..
@@ -1082,7 +1196,7 @@
1082
1196
 				err = -EFAULT;
1083
1197
 				goto error;
1084
1198
 			}
1085
-		} else {
1199
+		} else if (!uarg || !uarg->zerocopy) {
1086
1200
 			int i = skb_shinfo(skb)->nr_frags;
1087
1201
 
1088
1202
 			err = -ENOMEM;
..
..
@@ -1112,6 +1226,10 @@
1112
1226
 			skb->data_len += copy;
1113
1227
 			skb->truesize += copy;
1114
1228
 			wmem_alloc_delta += copy;
1229
+		} else {
1230
+			err = skb_zerocopy_iter_dgram(skb, from, copy);
1231
+			if (err < 0)
1232
+				goto error;
1115
1233
 		}
1116
1234
 		offset += copy;
1117
1235
 		length -= copy;
..
..
@@ -1124,6 +1242,8 @@
1124
1242
 error_efault:
1125
1243
 	err = -EFAULT;
1126
1244
 error:
1245
+	if (uarg)
1246
+		sock_zerocopy_put_abort(uarg, extra_uref);
1127
1247
 	cork->length -= length;
1128
1248
 	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1129
1249
 	refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
..
..
@@ -1171,6 +1291,7 @@
1171
1291
 	cork->length = 0;
1172
1292
 	cork->ttl = ipc->ttl;
1173
1293
 	cork->tos = ipc->tos;
1294
+	cork->mark = ipc->sockc.mark;
1174
1295
 	cork->priority = ipc->priority;
1175
1296
 	cork->transmit_time = ipc->sockc.transmit_time;
1176
1297
 	cork->tx_flags = 0;
..
..
@@ -1244,7 +1365,7 @@
1244
1365
 	if (cork->flags & IPCORK_OPT)
1245
1366
 		opt = cork->opt;
1246
1367
 
1247
-	if (!(rt->dst.dev->features&NETIF_F_SG))
1368
+	if (!(rt->dst.dev->features & NETIF_F_SG))
1248
1369
 		return -EOPNOTSUPP;
1249
1370
 
1250
1371
 	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
..
..
@@ -1304,7 +1425,7 @@
1304
1425
 				skb->csum = skb_copy_and_csum_bits(skb_prev,
1305
1426
 								   maxfraglen,
1306
1427
 						    skb_transport_header(skb),
1307
-								   fraggap, 0);
1428
+								   fraggap);
1308
1429
 				skb_prev->csum = csum_sub(skb_prev->csum,
1309
1430
 							  skb->csum);
1310
1431
 				pskb_trim_unique(skb_prev, maxfraglen);
..
..
@@ -1429,12 +1550,12 @@
1429
1550
 	ip_select_ident(net, skb, sk);
1430
1551
 
1431
1552
 	if (opt) {
1432
-		iph->ihl += opt->optlen>>2;
1553
+		iph->ihl += opt->optlen >> 2;
1433
1554
 		ip_options_build(skb, opt, cork->addr, rt, 0);
1434
1555
 	}
1435
1556
 
1436
1557
 	skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
1437
-	skb->mark = sk->sk_mark;
1558
+	skb->mark = cork->mark;
1438
1559
 	skb->tstamp = cork->transmit_time;
1439
1560
 	/*
1440
1561
 	 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
..
..
@@ -1443,9 +1564,19 @@
1443
1564
 	cork->dst = NULL;
1444
1565
 	skb_dst_set(skb, &rt->dst);
1445
1566
 
1446
-	if (iph->protocol == IPPROTO_ICMP)
1447
-		icmp_out_count(net, ((struct icmphdr *)
1448
-			skb_transport_header(skb))->type);
1567
+	if (iph->protocol == IPPROTO_ICMP) {
1568
+		u8 icmp_type;
1569
+
1570
+		/* For such sockets, transhdrlen is zero when do ip_append_data(),
1571
+		 * so icmphdr does not in skb linear region and can not get icmp_type
1572
+		 * by icmp_hdr(skb)->type.
1573
+		 */
1574
+		if (sk->sk_type == SOCK_RAW && !inet_sk(sk)->hdrincl)
1575
+			icmp_type = fl4->fl4_icmp_type;
1576
+		else
1577
+			icmp_type = icmp_hdr(skb)->type;
1578
+		icmp_out_count(net, icmp_type);
1579
+	}
1449
1580
 
1450
1581
 	ip_cork_release(cork);
1451
1582
 out:
..
..
@@ -1541,7 +1672,7 @@
1541
1672
 {
1542
1673
 	__wsum csum;
1543
1674
 
1544
-	csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1675
+	csum = csum_partial_copy_nocheck(dptr+offset, to, len);
1545
1676
 	skb->csum = csum_block_add(skb->csum, csum, odd);
1546
1677
 	return 0;
1547
1678
 }
..
..
@@ -1554,7 +1685,7 @@
1554
1685
 			   const struct ip_options *sopt,
1555
1686
 			   __be32 daddr, __be32 saddr,
1556
1687
 			   const struct ip_reply_arg *arg,
1557
-			   unsigned int len)
1688
+			   unsigned int len, u64 transmit_time)
1558
1689
 {
1559
1690
 	struct ip_options_data replyopts;
1560
1691
 	struct ipcm_cookie ipc;
..
..
@@ -1570,6 +1701,7 @@
1570
1701
 
1571
1702
 	ipcm_init(&ipc);
1572
1703
 	ipc.addr = daddr;
1704
+	ipc.sockc.transmit_time = transmit_time;
1573
1705
 
1574
1706
 	if (replyopts.opt.opt.optlen) {
1575
1707
 		ipc.opt = &replyopts.opt;
..
..
@@ -1590,18 +1722,17 @@
1590
1722
 			   daddr, saddr,
1591
1723
 			   tcp_hdr(skb)->source, tcp_hdr(skb)->dest,
1592
1724
 			   arg->uid);
1593
-	security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
1725
+	security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
1594
1726
 	rt = ip_route_output_key(net, &fl4);
1595
1727
 	if (IS_ERR(rt))
1596
1728
 		return;
1597
1729
 
1598
1730
 	inet_sk(sk)->tos = arg->tos & ~INET_ECN_MASK;
1599
1731
 
1600
-	sk->sk_priority = skb->priority;
1601
1732
 	sk->sk_protocol = ip_hdr(skb)->protocol;
1602
1733
 	sk->sk_bound_dev_if = arg->bound_dev_if;
1603
-	sk->sk_sndbuf = sysctl_wmem_default;
1604
-	sk->sk_mark = fl4.flowi4_mark;
1734
+	sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
1735
+	ipc.sockc.mark = fl4.flowi4_mark;
1605
1736
 	err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base,
1606
1737
 			     len, 0, &ipc, &rt, MSG_DONTWAIT);
1607
1738
 	if (unlikely(err)) {