tcp_lro.c 22.4 KB
Newer Older
1
2
3
/*-
 * Copyright (c) 2007, Myricom Inc.
 * Copyright (c) 2008, Intel Corporation.
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
4
 * Copyright (c) 2012 The FreeBSD Foundation
5
 * Copyright (c) 2016 Mellanox Technologies.
6
7
 * All rights reserved.
 *
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
8
9
10
 * Portions of this software were developed by Bjoern Zeeb
 * under sponsorship from the FreeBSD Foundation.
 *
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 */
32

Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
33
34
35
36
37
38
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");

#include "opt_inet.h"
#include "opt_inet6.h"

39
40
41
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
42
43
#include <sys/malloc.h>
#include <sys/mbuf.h>
44
#include <sys/socket.h>
45
#include <sys/sysctl.h>
46
47

#include <net/if.h>
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
48
#include <net/if_var.h>
49
#include <net/ethernet.h>
50
#include <net/vnet.h>
51
52
53

#include <netinet/in_systm.h>
#include <netinet/in.h>
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
54
#include <netinet/ip6.h>
55
#include <netinet/ip.h>
56
#include <netinet/ip_var.h>
57
58
#include <netinet/tcp.h>
#include <netinet/tcp_lro.h>
59
#include <netinet/tcp_var.h>
60

61
62
#include <netinet6/ip6_var.h>

63
64
#include <machine/in_cksum.h>

65
static MALLOC_DEFINE(M_LRO, "LRO", "LRO control structures");
66

Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
67
68
69
70
#define	TCP_LRO_UPDATE_CSUM	1
#ifndef	TCP_LRO_UPDATE_CSUM
#define	TCP_LRO_INVALID_CSUM	0x0000
#endif
71

72
static void	tcp_lro_rx_done(struct lro_ctrl *lc);
73
74
static int	tcp_lro_rx2(struct lro_ctrl *lc, struct mbuf *m,
		    uint32_t csum, int use_hash);
75

76
77
78
79
80
81
82
83
SYSCTL_NODE(_net_inet_tcp, OID_AUTO, lro,  CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
    "TCP LRO");

static unsigned	tcp_lro_entries = TCP_LRO_ENTRIES;
SYSCTL_UINT(_net_inet_tcp_lro, OID_AUTO, entries,
    CTLFLAG_RDTUN | CTLFLAG_MPSAFE, &tcp_lro_entries, 0,
    "default number of LRO entries");

84
static __inline void
85
86
tcp_lro_active_insert(struct lro_ctrl *lc, struct lro_head *bucket,
    struct lro_entry *le)
87
88
89
{

	LIST_INSERT_HEAD(&lc->lro_active, le, next);
90
	LIST_INSERT_HEAD(bucket, le, hash_next);
91
92
93
94
95
96
}

static __inline void
tcp_lro_active_remove(struct lro_entry *le)
{

97
98
	LIST_REMOVE(le, next);		/* active list */
	LIST_REMOVE(le, hash_next);	/* hash bucket */
99
100
}

101
int
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
102
tcp_lro_init(struct lro_ctrl *lc)
103
{
104
	return (tcp_lro_init_args(lc, NULL, tcp_lro_entries, 0));
105
106
107
108
109
}

int
tcp_lro_init_args(struct lro_ctrl *lc, struct ifnet *ifp,
    unsigned lro_entries, unsigned lro_mbufs)
110
{
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
111
	struct lro_entry *le;
112
	size_t size;
113
	unsigned i, elements;
114

Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
115
116
117
118
	lc->lro_bad_csum = 0;
	lc->lro_queued = 0;
	lc->lro_flushed = 0;
	lc->lro_cnt = 0;
119
120
121
	lc->lro_mbuf_count = 0;
	lc->lro_mbuf_max = lro_mbufs;
	lc->lro_cnt = lro_entries;
122
123
	lc->lro_ackcnt_lim = TCP_LRO_ACKCNT_MAX;
	lc->lro_length_lim = TCP_LRO_LENGTH_MAX;
124
	lc->ifp = ifp;
125
126
	LIST_INIT(&lc->lro_free);
	LIST_INIT(&lc->lro_active);
127

128
129
130
131
132
133
134
135
136
137
138
139
	/* create hash table to accelerate entry lookup */
	if (lro_entries > lro_mbufs)
		elements = lro_entries;
	else
		elements = lro_mbufs;
	lc->lro_hash = phashinit_flags(elements, M_LRO, &lc->lro_hashsz,
	    HASH_NOWAIT);
	if (lc->lro_hash == NULL) {
		memset(lc, 0, sizeof(*lc));
		return (ENOMEM);
	}

140
	/* compute size to allocate */
141
	size = (lro_mbufs * sizeof(struct lro_mbuf_sort)) +
142
	    (lro_entries * sizeof(*le));
143
	lc->lro_mbuf_data = (struct lro_mbuf_sort *)
144
145
146
147
148
149
150
151
152
153
154
155
156
	    malloc(size, M_LRO, M_NOWAIT | M_ZERO);

	/* check for out of memory */
	if (lc->lro_mbuf_data == NULL) {
		memset(lc, 0, sizeof(*lc));
		return (ENOMEM);
	}
	/* compute offset for LRO entries */
	le = (struct lro_entry *)
	    (lc->lro_mbuf_data + lro_mbufs);

	/* setup linked list */
	for (i = 0; i != lro_entries; i++)
157
		LIST_INSERT_HEAD(&lc->lro_free, le + i, next);
158
159

	return (0);
160
161
162
}

void
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
163
tcp_lro_free(struct lro_ctrl *lc)
164
{
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
165
	struct lro_entry *le;
166
167
168
	unsigned x;

	/* reset LRO free list */
169
	LIST_INIT(&lc->lro_free);
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
170

171
	/* free active mbufs, if any */
172
	while ((le = LIST_FIRST(&lc->lro_active)) != NULL) {
173
		tcp_lro_active_remove(le);
174
		m_freem(le->m_head);
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
175
	}
176

177
178
179
180
181
182
183
	/* free hash table */
	if (lc->lro_hash != NULL) {
		free(lc->lro_hash, M_LRO);
		lc->lro_hash = NULL;
	}
	lc->lro_hashsz = 0;

184
185
	/* free mbuf array, if any */
	for (x = 0; x != lc->lro_mbuf_count; x++)
186
		m_freem(lc->lro_mbuf_data[x].mb);
187
188
189
190
191
	lc->lro_mbuf_count = 0;
	
	/* free allocated memory, if any */
	free(lc->lro_mbuf_data, M_LRO);
	lc->lro_mbuf_data = NULL;
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
192
}
193

Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
#ifdef TCP_LRO_UPDATE_CSUM
static uint16_t
tcp_lro_csum_th(struct tcphdr *th)
{
	uint32_t ch;
	uint16_t *p, l;

	ch = th->th_sum = 0x0000;
	l = th->th_off;
	p = (uint16_t *)th;
	while (l > 0) {
		ch += *p;
		p++;
		ch += *p;
		p++;
		l--;
	}
	while (ch > 0xffff)
		ch = (ch >> 16) + (ch & 0xffff);

	return (ch & 0xffff);
}

static uint16_t
tcp_lro_rx_csum_fixup(struct lro_entry *le, void *l3hdr, struct tcphdr *th,
    uint16_t tcp_data_len, uint16_t csum)
{
	uint32_t c;
	uint16_t cs;

	c = csum;

	/* Remove length from checksum. */
	switch (le->eh_type) {
#ifdef INET6
	case ETHERTYPE_IPV6:
	{
		struct ip6_hdr *ip6;

		ip6 = (struct ip6_hdr *)l3hdr;
		if (le->append_cnt == 0)
			cs = ip6->ip6_plen;
		else {
			uint32_t cx;

			cx = ntohs(ip6->ip6_plen);
			cs = in6_cksum_pseudo(ip6, cx, ip6->ip6_nxt, 0);
		}
		break;
	}
#endif
#ifdef INET
	case ETHERTYPE_IP:
	{
		struct ip *ip4;

		ip4 = (struct ip *)l3hdr;
		if (le->append_cnt == 0)
			cs = ip4->ip_len;
		else {
			cs = in_addword(ntohs(ip4->ip_len) - sizeof(*ip4),
			    IPPROTO_TCP);
			cs = in_pseudo(ip4->ip_src.s_addr, ip4->ip_dst.s_addr,
			    htons(cs));
		}
		break;
	}
#endif
	default:
		cs = 0;		/* Keep compiler happy. */
264
	}
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
265
266
267
268
269
270
271
272
273
274
275

	cs = ~cs;
	c += cs;

	/* Remove TCP header csum. */
	cs = ~tcp_lro_csum_th(th);
	c += cs;
	while (c > 0xffff)
		c = (c >> 16) + (c & 0xffff);

	return (c & 0xffff);
276
}
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
277
#endif
278

279
280
281
282
283
static void
tcp_lro_rx_done(struct lro_ctrl *lc)
{
	struct lro_entry *le;

284
	while ((le = LIST_FIRST(&lc->lro_active)) != NULL) {
285
		tcp_lro_active_remove(le);
286
287
288
289
		tcp_lro_flush(lc, le);
	}
}

290
291
292
293
294
295
void
tcp_lro_flush_inactive(struct lro_ctrl *lc, const struct timeval *timeout)
{
	struct lro_entry *le, *le_tmp;
	struct timeval tv;

296
	if (LIST_EMPTY(&lc->lro_active))
297
298
299
300
		return;

	getmicrotime(&tv);
	timevalsub(&tv, timeout);
301
	LIST_FOREACH_SAFE(le, &lc->lro_active, next, le_tmp) {
302
		if (timevalcmp(&tv, &le->mtime, >=)) {
303
			tcp_lro_active_remove(le);
304
305
306
307
308
			tcp_lro_flush(lc, le);
		}
	}
}

309
void
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
310
tcp_lro_flush(struct lro_ctrl *lc, struct lro_entry *le)
311
{
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330

	if (le->append_cnt > 0) {
		struct tcphdr *th;
		uint16_t p_len;

		p_len = htons(le->p_len);
		switch (le->eh_type) {
#ifdef INET6
		case ETHERTYPE_IPV6:
		{
			struct ip6_hdr *ip6;

			ip6 = le->le_ip6;
			ip6->ip6_plen = p_len;
			th = (struct tcphdr *)(ip6 + 1);
			le->m_head->m_pkthdr.csum_flags = CSUM_DATA_VALID |
			    CSUM_PSEUDO_HDR;
			le->p_len += ETHER_HDR_LEN + sizeof(*ip6);
			break;
331
		}
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
#endif
#ifdef INET
		case ETHERTYPE_IP:
		{
			struct ip *ip4;
#ifdef TCP_LRO_UPDATE_CSUM
			uint32_t cl;
			uint16_t c;
#endif

			ip4 = le->le_ip4;
#ifdef TCP_LRO_UPDATE_CSUM
			/* Fix IP header checksum for new length. */
			c = ~ip4->ip_sum;
			cl = c;
			c = ~ip4->ip_len;
			cl += c + p_len;
			while (cl > 0xffff)
				cl = (cl >> 16) + (cl & 0xffff);
			c = cl;
			ip4->ip_sum = ~c;
#else
			ip4->ip_sum = TCP_LRO_INVALID_CSUM;
#endif
			ip4->ip_len = p_len;
			th = (struct tcphdr *)(ip4 + 1);
			le->m_head->m_pkthdr.csum_flags = CSUM_DATA_VALID |
			    CSUM_PSEUDO_HDR | CSUM_IP_CHECKED | CSUM_IP_VALID;
			le->p_len += ETHER_HDR_LEN;
			break;
		}
#endif
		default:
			th = NULL;	/* Keep compiler happy. */
		}
		le->m_head->m_pkthdr.csum_data = 0xffff;
		le->m_head->m_pkthdr.len = le->p_len;

		/* Incorporate the latest ACK into the TCP header. */
		th->th_ack = le->ack_seq;
		th->th_win = le->window;
		/* Incorporate latest timestamp into the TCP header. */
		if (le->timestamp != 0) {
			uint32_t *ts_ptr;

			ts_ptr = (uint32_t *)(th + 1);
			ts_ptr[1] = htonl(le->tsval);
			ts_ptr[2] = le->tsecr;
		}
#ifdef TCP_LRO_UPDATE_CSUM
		/* Update the TCP header checksum. */
		le->ulp_csum += p_len;
		le->ulp_csum += tcp_lro_csum_th(th);
		while (le->ulp_csum > 0xffff)
			le->ulp_csum = (le->ulp_csum >> 16) +
			    (le->ulp_csum & 0xffff);
		th->th_sum = (le->ulp_csum & 0xffff);
		th->th_sum = ~th->th_sum;
#else
		th->th_sum = TCP_LRO_INVALID_CSUM;
#endif
393
	}
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
394

395
	le->m_head->m_pkthdr.lro_nsegs = le->append_cnt + 1;
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
396
397
398
399
	(*lc->ifp->if_input)(lc->ifp, le->m_head);
	lc->lro_queued += le->append_cnt + 1;
	lc->lro_flushed++;
	bzero(le, sizeof(*le));
400
	LIST_INSERT_HEAD(&lc->lro_free, le, next);
401
402
}

403
404
405
406
407
#ifdef HAVE_INLINE_FLSLL
#define	tcp_lro_msb_64(x) (1ULL << (flsll(x) - 1))
#else
static inline uint64_t
tcp_lro_msb_64(uint64_t x)
408
{
409
410
411
412
413
414
415
416
417
	x |= (x >> 1);
	x |= (x >> 2);
	x |= (x >> 4);
	x |= (x >> 8);
	x |= (x >> 16);
	x |= (x >> 32);
	return (x & ~(x >> 1));
}
#endif
418

419
420
421
422
423
424
/*
 * The tcp_lro_sort() routine is comparable to qsort(), except it has
 * a worst case complexity limit of O(MIN(N,64)*N), where N is the
 * number of elements to sort and 64 is the number of sequence bits
 * available. The algorithm is bit-slicing the 64-bit sequence number,
 * sorting one bit at a time from the most significant bit until the
425
426
 * least significant one, skipping the constant bits. This is
 * typically called a radix sort.
427
428
429
430
431
432
433
434
435
436
437
 */
static void
tcp_lro_sort(struct lro_mbuf_sort *parray, uint32_t size)
{
	struct lro_mbuf_sort temp;
	uint64_t ones;
	uint64_t zeros;
	uint32_t x;
	uint32_t y;

repeat:
438
	/* for small arrays insertion sort is faster */
439
	if (size <= 12) {
440
441
442
443
444
		for (x = 1; x < size; x++) {
			temp = parray[x];
			for (y = x; y > 0 && temp.seq < parray[y - 1].seq; y--)
				parray[y] = parray[y - 1];
			parray[y] = temp;
445
446
447
		}
		return;
	}
448

449
450
451
452
453
454
455
	/* compute sequence bits which are constant */
	ones = 0;
	zeros = 0;
	for (x = 0; x != size; x++) {
		ones |= parray[x].seq;
		zeros |= ~parray[x].seq;
	}
456

457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
	/* compute bits which are not constant into "ones" */
	ones &= zeros;
	if (ones == 0)
		return;

	/* pick the most significant bit which is not constant */
	ones = tcp_lro_msb_64(ones);

	/*
	 * Move entries having cleared sequence bits to the beginning
	 * of the array:
	 */
	for (x = y = 0; y != size; y++) {
		/* skip set bits */
		if (parray[y].seq & ones)
			continue;
		/* swap entries */
		temp = parray[x];
		parray[x] = parray[y];
		parray[y] = temp;
		x++;
	}

	KASSERT(x != 0 && x != size, ("Memory is corrupted\n"));

	/* sort zeros */
	tcp_lro_sort(parray, x);

	/* sort ones */
	parray += x;
	size -= x;
	goto repeat;
489
490
491
492
493
}

void
tcp_lro_flush_all(struct lro_ctrl *lc)
{
494
495
	uint64_t seq;
	uint64_t nseq;
496
497
498
	unsigned x;

	/* check if no mbufs to flush */
499
	if (lc->lro_mbuf_count == 0)
500
501
502
		goto done;

	/* sort all mbufs according to stream */
503
	tcp_lro_sort(lc->lro_mbuf_data, lc->lro_mbuf_count);
504
505

	/* input data into LRO engine, stream by stream */
506
	seq = 0;
507
508
509
	for (x = 0; x != lc->lro_mbuf_count; x++) {
		struct mbuf *mb;

510
511
512
513
514
		/* get mbuf */
		mb = lc->lro_mbuf_data[x].mb;

		/* get sequence number, masking away the packet index */
		nseq = lc->lro_mbuf_data[x].seq & (-1ULL << 24);
515
516

		/* check for new stream */
517
518
		if (seq != nseq) {
			seq = nseq;
519
520

			/* flush active streams */
521
			tcp_lro_rx_done(lc);
522
		}
523

524
		/* add packet to LRO engine */
525
		if (tcp_lro_rx2(lc, mb, 0, 0) != 0) {
526
527
528
529
530
531
532
533
			/* input packet to network layer */
			(*lc->ifp->if_input)(lc->ifp, mb);
			lc->lro_queued++;
			lc->lro_flushed++;
		}
	}
done:
	/* flush active streams */
534
535
	tcp_lro_rx_done(lc);

536
537
538
	lc->lro_mbuf_count = 0;
}

Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
539
540
541
542
#ifdef INET6
static int
tcp_lro_rx_ipv6(struct lro_ctrl *lc, struct mbuf *m, struct ip6_hdr *ip6,
    struct tcphdr **th)
543
{
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578

	/* XXX-BZ we should check the flow-label. */

	/* XXX-BZ We do not yet support ext. hdrs. */
	if (ip6->ip6_nxt != IPPROTO_TCP)
		return (TCP_LRO_NOT_SUPPORTED);

	/* Find the TCP header. */
	*th = (struct tcphdr *)(ip6 + 1);

	return (0);
}
#endif

#ifdef INET
static int
tcp_lro_rx_ipv4(struct lro_ctrl *lc, struct mbuf *m, struct ip *ip4,
    struct tcphdr **th)
{
	int csum_flags;
	uint16_t csum;

	if (ip4->ip_p != IPPROTO_TCP)
		return (TCP_LRO_NOT_SUPPORTED);

	/* Ensure there are no options. */
	if ((ip4->ip_hl << 2) != sizeof (*ip4))
		return (TCP_LRO_CANNOT);

	/* .. and the packet is not fragmented. */
	if (ip4->ip_off & htons(IP_MF|IP_OFFMASK))
		return (TCP_LRO_CANNOT);

	/* Legacy IP has a header checksum that needs to be correct. */
	csum_flags = m->m_pkthdr.csum_flags;
579
580
	if (csum_flags & CSUM_IP_CHECKED) {
		if (__predict_false((csum_flags & CSUM_IP_VALID) == 0)) {
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
581
582
			lc->lro_bad_csum++;
			return (TCP_LRO_CANNOT);
583
584
		}
	} else {
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
585
		csum = in_cksum_hdr(ip4);
586
		if (__predict_false((csum) != 0)) {
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
587
588
			lc->lro_bad_csum++;
			return (TCP_LRO_CANNOT);
589
		}
590
	}
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
591
592
593
594
595
596
597
598

	/* Find the TCP header (we assured there are no IP options). */
	*th = (struct tcphdr *)(ip4 + 1);

	return (0);
}
#endif

599
600
static int
tcp_lro_rx2(struct lro_ctrl *lc, struct mbuf *m, uint32_t csum, int use_hash)
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
{
	struct lro_entry *le;
	struct ether_header *eh;
#ifdef INET6
	struct ip6_hdr *ip6 = NULL;	/* Keep compiler happy. */
#endif
#ifdef INET
	struct ip *ip4 = NULL;		/* Keep compiler happy. */
#endif
	struct tcphdr *th;
	void *l3hdr = NULL;		/* Keep compiler happy. */
	uint32_t *ts_ptr;
	tcp_seq seq;
	int error, ip_len, l;
	uint16_t eh_type, tcp_data_len;
616
	struct lro_head *bucket;
617
	int force_flush = 0;
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
618
619
620
621
622
623
624
625

	/* We expect a contiguous header [eh, ip, tcp]. */

	eh = mtod(m, struct ether_header *);
	eh_type = ntohs(eh->ether_type);
	switch (eh_type) {
#ifdef INET6
	case ETHERTYPE_IPV6:
626
627
	{
		CURVNET_SET(lc->ifp->if_vnet);
628
629
		if (V_ip6_forwarding != 0) {
			/* XXX-BZ stats but changing lro_ctrl is a problem. */
630
			CURVNET_RESTORE();
631
632
			return (TCP_LRO_CANNOT);
		}
633
		CURVNET_RESTORE();
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
634
635
636
637
638
639
640
		l3hdr = ip6 = (struct ip6_hdr *)(eh + 1);
		error = tcp_lro_rx_ipv6(lc, m, ip6, &th);
		if (error != 0)
			return (error);
		tcp_data_len = ntohs(ip6->ip6_plen);
		ip_len = sizeof(*ip6) + tcp_data_len;
		break;
641
	}
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
642
643
644
#endif
#ifdef INET
	case ETHERTYPE_IP:
645
646
	{
		CURVNET_SET(lc->ifp->if_vnet);
647
648
		if (V_ipforwarding != 0) {
			/* XXX-BZ stats but changing lro_ctrl is a problem. */
649
			CURVNET_RESTORE();
650
651
			return (TCP_LRO_CANNOT);
		}
652
		CURVNET_RESTORE();
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
653
654
655
656
657
658
659
		l3hdr = ip4 = (struct ip *)(eh + 1);
		error = tcp_lro_rx_ipv4(lc, m, ip4, &th);
		if (error != 0)
			return (error);
		ip_len = ntohs(ip4->ip_len);
		tcp_data_len = ip_len - sizeof(*ip4);
		break;
660
	}
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
661
662
663
664
#endif
	/* XXX-BZ what happens in case of VLAN(s)? */
	default:
		return (TCP_LRO_NOT_SUPPORTED);
665
666
	}

Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
	/*
	 * If the frame is padded beyond the end of the IP packet, then we must
	 * trim the extra bytes off.
	 */
	l = m->m_pkthdr.len - (ETHER_HDR_LEN + ip_len);
	if (l != 0) {
		if (l < 0)
			/* Truncated packet. */
			return (TCP_LRO_CANNOT);

		m_adj(m, -l);
	}

	/*
	 * Check TCP header constraints.
	 */
	/* Ensure no bits set besides ACK or PSH. */
684
685
686
687
688
689
690
691
692
	if ((th->th_flags & ~(TH_ACK | TH_PUSH)) != 0) {
		if (th->th_flags & TH_SYN)
			return (TCP_LRO_CANNOT);
		/*
		 * Make sure that previously seen segements/ACKs are delivered
		 * before this segement, e.g. FIN.
		 */
		force_flush = 1;
	}
693

Sepherosa Ziehau's avatar
Sepherosa Ziehau committed
694
	/* XXX-BZ We lose a ACK|PUSH flag concatenating multiple segments. */
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
695
696
697
698
699
700
	/* XXX-BZ Ideally we'd flush on PUSH? */

	/*
	 * Check for timestamps.
	 * Since the only option we handle are timestamps, we only have to
	 * handle the simple case of aligned timestamps.
701
	 */
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
702
703
704
705
706
707
	l = (th->th_off << 2);
	tcp_data_len -= l;
	l -= sizeof(*th);
	ts_ptr = (uint32_t *)(th + 1);
	if (l != 0 && (__predict_false(l != TCPOLEN_TSTAMP_APPA) ||
	    (*ts_ptr != ntohl(TCPOPT_NOP<<24|TCPOPT_NOP<<16|
708
709
710
711
712
713
714
	    TCPOPT_TIMESTAMP<<8|TCPOLEN_TIMESTAMP)))) {
		/*
		 * Make sure that previously seen segements/ACKs are delivered
		 * before this segement.
		 */
		force_flush = 1;
	}
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
715
716
717
718
719
720
721

	/* If the driver did not pass in the checksum, set it now. */
	if (csum == 0x0000)
		csum = th->th_sum;

	seq = ntohl(th->th_seq);

722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
	if (!use_hash) {
		bucket = &lc->lro_hash[0];
	} else if (M_HASHTYPE_ISHASH(m)) {
		bucket = &lc->lro_hash[m->m_pkthdr.flowid % lc->lro_hashsz];
	} else {
		uint32_t hash;

		switch (eh_type) {
#ifdef INET
		case ETHERTYPE_IP:
			hash = ip4->ip_src.s_addr + ip4->ip_dst.s_addr;
			break;
#endif
#ifdef INET6
		case ETHERTYPE_IPV6:
			hash = ip6->ip6_src.s6_addr32[0] +
			    ip6->ip6_dst.s6_addr32[0];
			hash += ip6->ip6_src.s6_addr32[1] +
			    ip6->ip6_dst.s6_addr32[1];
			hash += ip6->ip6_src.s6_addr32[2] +
			    ip6->ip6_dst.s6_addr32[2];
			hash += ip6->ip6_src.s6_addr32[3] +
			    ip6->ip6_dst.s6_addr32[3];
			break;
#endif
		default:
			hash = 0;
			break;
		}
		hash += th->th_sport + th->th_dport;
		bucket = &lc->lro_hash[hash % lc->lro_hashsz];
	}

Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
755
	/* Try to find a matching previous segment. */
756
	LIST_FOREACH(le, bucket, hash_next) {
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
		if (le->eh_type != eh_type)
			continue;
		if (le->source_port != th->th_sport ||
		    le->dest_port != th->th_dport)
			continue;
		switch (eh_type) {
#ifdef INET6
		case ETHERTYPE_IPV6:
			if (bcmp(&le->source_ip6, &ip6->ip6_src,
			    sizeof(struct in6_addr)) != 0 ||
			    bcmp(&le->dest_ip6, &ip6->ip6_dst,
			    sizeof(struct in6_addr)) != 0)
				continue;
			break;
#endif
#ifdef INET
		case ETHERTYPE_IP:
			if (le->source_ip4 != ip4->ip_src.s_addr ||
			    le->dest_ip4 != ip4->ip_dst.s_addr)
				continue;
			break;
#endif
779
780
		}

781
782
783
784
785
786
787
		if (force_flush) {
			/* Timestamps mismatch; this is a FIN, etc */
			tcp_lro_active_remove(le);
			tcp_lro_flush(lc, le);
			return (TCP_LRO_CANNOT);
		}

Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
788
		/* Flush now if appending will result in overflow. */
789
		if (le->p_len > (lc->lro_length_lim - tcp_data_len)) {
790
			tcp_lro_active_remove(le);
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
791
792
793
794
795
796
797
798
			tcp_lro_flush(lc, le);
			break;
		}

		/* Try to append the new segment. */
		if (__predict_false(seq != le->next_seq ||
		    (tcp_data_len == 0 && le->ack_seq == th->th_ack))) {
			/* Out of order packet or duplicate ACK. */
799
			tcp_lro_active_remove(le);
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
800
801
802
803
804
805
806
807
808
809
810
811
812
813
			tcp_lro_flush(lc, le);
			return (TCP_LRO_CANNOT);
		}

		if (l != 0) {
			uint32_t tsval = ntohl(*(ts_ptr + 1));
			/* Make sure timestamp values are increasing. */
			/* XXX-BZ flip and use TSTMP_GEQ macro for this? */
			if (__predict_false(le->tsval > tsval ||
			    *(ts_ptr + 2) == 0))
				return (TCP_LRO_CANNOT);
			le->tsval = tsval;
			le->tsecr = *(ts_ptr + 2);
		}
814

Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
815
816
817
818
819
820
821
822
823
824
825
826
		le->next_seq += tcp_data_len;
		le->ack_seq = th->th_ack;
		le->window = th->th_win;
		le->append_cnt++;

#ifdef TCP_LRO_UPDATE_CSUM
		le->ulp_csum += tcp_lro_rx_csum_fixup(le, l3hdr, th,
		    tcp_data_len, ~csum);
#endif

		if (tcp_data_len == 0) {
			m_freem(m);
827
828
829
830
831
			/*
			 * Flush this LRO entry, if this ACK should not
			 * be further delayed.
			 */
			if (le->append_cnt >= lc->lro_ackcnt_lim) {
832
				tcp_lro_active_remove(le);
833
834
				tcp_lro_flush(lc, le);
			}
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
835
			return (0);
836
		}
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
837
838
839
840
841
842
843
844
845

		le->p_len += tcp_data_len;

		/*
		 * Adjust the mbuf so that m_data points to the first byte of
		 * the ULP payload.  Adjust the mbuf to avoid complications and
		 * append new segment to existing mbuf chain.
		 */
		m_adj(m, m->m_pkthdr.len - tcp_data_len);
846
		m_demote_pkthdr(m);
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
847
848
849
850
851
852
853
854

		le->m_tail->m_next = m;
		le->m_tail = m_last(m);

		/*
		 * If a possible next full length packet would cause an
		 * overflow, pro-actively flush now.
		 */
855
		if (le->p_len > (lc->lro_length_lim - lc->ifp->if_mtu)) {
856
			tcp_lro_active_remove(le);
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
857
			tcp_lro_flush(lc, le);
858
859
		} else
			getmicrotime(&le->mtime);
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
860
861

		return (0);
862
863
	}

864
865
866
867
868
869
870
871
	if (force_flush) {
		/*
		 * Nothing to flush, but this segment can not be further
		 * aggregated/delayed.
		 */
		return (TCP_LRO_CANNOT);
	}

Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
872
	/* Try to find an empty slot. */
873
	if (LIST_EMPTY(&lc->lro_free))
874
		return (TCP_LRO_NO_ENTRIES);
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
875
876

	/* Start a new segment chain. */
877
878
	le = LIST_FIRST(&lc->lro_free);
	LIST_REMOVE(le, next);
879
	tcp_lro_active_insert(lc, bucket, le);
880
	getmicrotime(&le->mtime);
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901

	/* Start filling in details. */
	switch (eh_type) {
#ifdef INET6
	case ETHERTYPE_IPV6:
		le->le_ip6 = ip6;
		le->source_ip6 = ip6->ip6_src;
		le->dest_ip6 = ip6->ip6_dst;
		le->eh_type = eh_type;
		le->p_len = m->m_pkthdr.len - ETHER_HDR_LEN - sizeof(*ip6);
		break;
#endif
#ifdef INET
	case ETHERTYPE_IP:
		le->le_ip4 = ip4;
		le->source_ip4 = ip4->ip_src.s_addr;
		le->dest_ip4 = ip4->ip_dst.s_addr;
		le->eh_type = eh_type;
		le->p_len = m->m_pkthdr.len - ETHER_HDR_LEN;
		break;
#endif
902
	}
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
	le->source_port = th->th_sport;
	le->dest_port = th->th_dport;

	le->next_seq = seq + tcp_data_len;
	le->ack_seq = th->th_ack;
	le->window = th->th_win;
	if (l != 0) {
		le->timestamp = 1;
		le->tsval = ntohl(*(ts_ptr + 1));
		le->tsecr = *(ts_ptr + 2);
	}

#ifdef TCP_LRO_UPDATE_CSUM
	/*
	 * Do not touch the csum of the first packet.  However save the
	 * "adjusted" checksum of just the source and destination addresses,
	 * the next header and the TCP payload.  The length and TCP header
	 * parts may change, so we remove those from the saved checksum and
	 * re-add with final values on tcp_lro_flush() if needed.
	 */
	KASSERT(le->ulp_csum == 0, ("%s: le=%p le->ulp_csum=0x%04x\n",
	    __func__, le, le->ulp_csum));

	le->ulp_csum = tcp_lro_rx_csum_fixup(le, l3hdr, th, tcp_data_len,
	    ~csum);
	th->th_sum = csum;	/* Restore checksum on first packet. */
#endif

	le->m_head = m;
	le->m_tail = m_last(m);

	return (0);
935
}
Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
936

937
938
939
940
941
942
943
int
tcp_lro_rx(struct lro_ctrl *lc, struct mbuf *m, uint32_t csum)
{

	return tcp_lro_rx2(lc, m, csum, 1);
}

944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
void
tcp_lro_queue_mbuf(struct lro_ctrl *lc, struct mbuf *mb)
{
	/* sanity checks */
	if (__predict_false(lc->ifp == NULL || lc->lro_mbuf_data == NULL ||
	    lc->lro_mbuf_max == 0)) {
		/* packet drop */
		m_freem(mb);
		return;
	}

	/* check if packet is not LRO capable */
	if (__predict_false(mb->m_pkthdr.csum_flags == 0 ||
	    (lc->ifp->if_capenable & IFCAP_LRO) == 0)) {
		lc->lro_flushed++;
		lc->lro_queued++;

		/* input packet to network layer */
		(*lc->ifp->if_input) (lc->ifp, mb);
		return;
	}

	/* check if array is full */
	if (__predict_false(lc->lro_mbuf_count == lc->lro_mbuf_max))
		tcp_lro_flush_all(lc);

970
971
972
973
974
	/* create sequence number */
	lc->lro_mbuf_data[lc->lro_mbuf_count].seq =
	    (((uint64_t)M_HASHTYPE_GET(mb)) << 56) |
	    (((uint64_t)mb->m_pkthdr.flowid) << 24) |
	    ((uint64_t)lc->lro_mbuf_count);
975
976

	/* enter mbuf */
977
	lc->lro_mbuf_data[lc->lro_mbuf_count++].mb = mb;
978
979
}

Bjoern A. Zeeb's avatar
Bjoern A. Zeeb committed
980
/* end */