xref: /xnu-8792.61.2/bsd/skywalk/nexus/netif/nx_netif_gso.c (revision 42e220869062b56f8d7d0726fd4c88954f87902c)
1 /*
2  * Copyright (c) 2020-2022 Apple Inc. All rights reserved.
3  *
4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5  *
6  * This file contains Original Code and/or Modifications of Original Code
7  * as defined in and that are subject to the Apple Public Source License
8  * Version 2.0 (the 'License'). You may not use this file except in
9  * compliance with the License. The rights granted to you under the License
10  * may not be used to create, or enable the creation or redistribution of,
11  * unlawful or unlicensed copies of an Apple operating system, or to
12  * circumvent, violate, or enable the circumvention or violation of, any
13  * terms of an Apple operating system software license agreement.
14  *
15  * Please obtain a copy of the License at
16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
17  *
18  * The Original Code and all software distributed under the License are
19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23  * Please see the License for the specific language governing rights and
24  * limitations under the License.
25  *
26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27  */
28 
29 /*
30  * Copyright (C) 2014, Stefano Garzarella - Universita` di Pisa.
31  * All rights reserved.
32  *
33  * Redistribution and use in source and binary forms, with or without
34  * modification, are permitted provided that the following conditions
35  * are met:
36  *   1. Redistributions of source code must retain the above copyright
37  *      notice, this list of conditions and the following disclaimer.
38  *   2. Redistributions in binary form must reproduce the above copyright
39  *      notice, this list of conditions and the following disclaimer in the
40  *      documentation and/or other materials provided with the distribution.
41  *
42  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
43  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
44  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
45  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
46  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
47  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
48  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
49  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
50  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
51  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
52  * SUCH DAMAGE.
53  */
54 
55 #include <sys/param.h>
56 #include <sys/kernel.h>
57 #include <sys/types.h>
58 #include <sys/systm.h>
59 #include <sys/mbuf.h>
60 #include <sys/socket.h>
61 #include <sys/sysctl.h>
62 #include <sys/malloc.h>
63 
64 #include <netinet/in.h>
65 #include <netinet/ip_var.h>
66 #include <netinet/ip.h>
67 #include <netinet/tcp.h>
68 #include <netinet/tcpip.h>
69 #include <netinet/ip6.h>
70 #include <netinet6/ip6_var.h>
71 
72 #include <net/if.h>
73 #include <net/if_var.h>
74 #include <net/ethernet.h>
75 #include <net/pktap.h>
76 #include <skywalk/os_skywalk_private.h>
77 #include <skywalk/nexus/netif/nx_netif.h>
78 
79 #define CSUM_GSO_MASK    0x00300000
80 #define CSUM_GSO_OFFSET  20
81 #define CSUM_TO_GSO(x) ((x & CSUM_GSO_MASK) >> CSUM_GSO_OFFSET)
82 
83 enum netif_gso_type {
84 	GSO_NONE,
85 	GSO_TCP4,
86 	GSO_TCP6,
87 	GSO_END_OF_TYPE
88 };
89 
90 uint32_t netif_chain_enqueue = 1;
91 #if (DEVELOPMENT || DEBUG)
92 SYSCTL_UINT(_kern_skywalk_netif, OID_AUTO, chain_enqueue,
93     CTLFLAG_RW | CTLFLAG_LOCKED, &netif_chain_enqueue, 0,
94     "netif chain enqueue");
95 #endif /* (DEVELOPMENT || DEBUG) */
96 
97 /*
98  * Array of function pointers that execute GSO depending on packet type
99  */
100 int (*netif_gso_functions[GSO_END_OF_TYPE]) (struct ifnet*, struct mbuf*);
101 
102 /*
103  * Structure that contains the state during the TCP segmentation
104  */
105 struct netif_gso_ip_tcp_state {
106 	void (*update)(struct netif_gso_ip_tcp_state*,
107 	    struct __kern_packet *pkt, uint8_t *baddr);
108 	void (*internal)(struct netif_gso_ip_tcp_state*, uint32_t partial,
109 	    uint16_t payload_len);
110 	union {
111 		struct ip *ip;
112 		struct ip6_hdr *ip6;
113 	} hdr;
114 	int af;
115 	struct tcphdr *tcp;
116 	struct kern_pbufpool *pp;
117 	uint32_t psuedo_hdr_csum;
118 	uint32_t tcp_seq;
119 	uint16_t hlen;
120 	uint16_t mss;
121 	uint16_t ip_id;
122 	uint8_t mac_hlen;
123 	uint8_t ip_hlen;
124 	uint8_t tcp_hlen;
125 };
126 
127 static inline uint8_t
netif_gso_get_frame_header_len(struct mbuf * m,uint8_t * hlen)128 netif_gso_get_frame_header_len(struct mbuf *m, uint8_t *hlen)
129 {
130 	uint64_t len;
131 	char *ph = m->m_pkthdr.pkt_hdr;
132 
133 	if (__improbable(m_pktlen(m) == 0 || ph == NULL ||
134 	    ph < (char *)m->m_data)) {
135 		return ERANGE;
136 	}
137 	len = (ph - m->m_data);
138 	if (__improbable(len > UINT8_MAX)) {
139 		return ERANGE;
140 	}
141 	*hlen = (uint8_t)len;
142 	return 0;
143 }
144 
145 static inline int
netif_gso_check_netif_active(struct ifnet * ifp,struct mbuf * m,struct kern_pbufpool ** pp)146 netif_gso_check_netif_active(struct ifnet *ifp, struct mbuf *m,
147     struct kern_pbufpool **pp)
148 {
149 	struct __kern_channel_ring *kring;
150 	struct nx_netif *nif = NA(ifp)->nifna_netif;
151 	struct netif_stats *nifs = &nif->nif_stats;
152 	struct kern_nexus *nx = nif->nif_nx;
153 	struct nexus_adapter *hwna = nx_port_get_na(nx, NEXUS_PORT_NET_IF_DEV);
154 	uint32_t sc_idx = MBUF_SCIDX(m_get_service_class(m));
155 
156 	if (__improbable(!NA_IS_ACTIVE(hwna))) {
157 		STATS_INC(nifs, NETIF_STATS_DROP_NA_INACTIVE);
158 		SK_DF(SK_VERB_NETIF,
159 		    "\"%s\" (0x%llx) not in skywalk mode anymore",
160 		    hwna->na_name, SK_KVA(hwna));
161 		return ENXIO;
162 	}
163 
164 	VERIFY(sc_idx < KPKT_SC_MAX_CLASSES);
165 	kring = &hwna->na_tx_rings[hwna->na_kring_svc_lut[sc_idx]];
166 	if (__improbable(KR_DROP(kring))) {
167 		STATS_INC(nifs, NETIF_STATS_DROP_KRDROP_MODE);
168 		SK_DF(SK_VERB_NETIF,
169 		    "kr \"%s\" (0x%llx) krflags 0x%b or %s in drop mode",
170 		    kring->ckr_name, SK_KVA(kring), kring->ckr_flags,
171 		    CKRF_BITS, ifp->if_xname);
172 		return ENXIO;
173 	}
174 	*pp = kring->ckr_pp;
175 	return 0;
176 }
177 
178 static inline boolean_t
netif_chain_enqueue_enabled(struct ifnet * ifp)179 netif_chain_enqueue_enabled(struct ifnet *ifp)
180 {
181 	return netif_chain_enqueue != 0 && ifp->if_output_netem == NULL &&
182 	       (ifp->if_eflags & IFEF_ENQUEUE_MULTI) == 0;
183 }
184 
185 static inline int
netif_gso_send(struct ifnet * ifp,struct __kern_packet * head,struct __kern_packet * tail,uint32_t count,uint32_t bytes)186 netif_gso_send(struct ifnet *ifp, struct __kern_packet *head,
187     struct __kern_packet *tail, uint32_t count, uint32_t bytes)
188 {
189 	struct nx_netif *nif = NA(ifp)->nifna_netif;
190 	struct netif_stats *nifs = &nif->nif_stats;
191 	struct netif_qset *qset = NULL;
192 	uint64_t qset_id = 0;
193 	int error = 0;
194 	boolean_t dropped;
195 
196 	if (NX_LLINK_PROV(nif->nif_nx) &&
197 	    ifp->if_traffic_rule_count > 0 &&
198 	    nxctl_inet_traffic_rule_find_qset_id_with_pkt(ifp->if_xname,
199 	    head, &qset_id) == 0) {
200 		qset = nx_netif_find_qset(nif, qset_id);
201 		ASSERT(qset != NULL);
202 	}
203 	if (netif_chain_enqueue_enabled(ifp)) {
204 		dropped = false;
205 		if (qset != NULL) {
206 			head->pkt_qset_idx = qset->nqs_idx;
207 			error = ifnet_enqueue_ifcq_pkt_chain(ifp, qset->nqs_ifcq,
208 			    head, tail, count, bytes, false, &dropped);
209 		} else {
210 			error = ifnet_enqueue_pkt_chain(ifp, head, tail,
211 			    count, bytes, false, &dropped);
212 		}
213 		if (__improbable(dropped)) {
214 			STATS_ADD(nifs, NETIF_STATS_TX_DROP_ENQ_AQM, count);
215 			STATS_ADD(nifs, NETIF_STATS_DROP, count);
216 		}
217 	} else {
218 		struct __kern_packet *pkt = head, *next;
219 		uint32_t c = 0, b = 0;
220 
221 		while (pkt != NULL) {
222 			int err;
223 
224 			next = pkt->pkt_nextpkt;
225 			pkt->pkt_nextpkt = NULL;
226 			c++;
227 			b += pkt->pkt_length;
228 
229 			dropped = false;
230 			if (qset != NULL) {
231 				pkt->pkt_qset_idx = qset->nqs_idx;
232 				err = ifnet_enqueue_ifcq_pkt(ifp, qset->nqs_ifcq,
233 				    pkt, false, &dropped);
234 			} else {
235 				err = ifnet_enqueue_pkt(ifp, pkt, false, &dropped);
236 			}
237 			if (error == 0 && __improbable(err != 0)) {
238 				error = err;
239 			}
240 			if (__improbable(dropped)) {
241 				STATS_INC(nifs, NETIF_STATS_TX_DROP_ENQ_AQM);
242 				STATS_INC(nifs, NETIF_STATS_DROP);
243 			}
244 			pkt = next;
245 		}
246 		ASSERT(c == count);
247 		ASSERT(b == bytes);
248 	}
249 	if (qset != NULL) {
250 		nx_netif_qset_release(&qset);
251 	}
252 	netif_transmit(ifp, NETIF_XMIT_FLAG_HOST);
253 	return error;
254 }
255 
256 /*
257  * Segment and transmit a queue of packets which fit the given mss + hdr_len.
258  * m points to mbuf chain to be segmented.
259  * This function splits the payload (m-> m_pkthdr.len - hdr_len)
260  * into segments of length MSS bytes and then copy the first hdr_len bytes
261  * from m at the top of each segment.
262  */
263 static inline int
netif_gso_tcp_segment_mbuf(struct mbuf * m,struct ifnet * ifp,struct netif_gso_ip_tcp_state * state,struct kern_pbufpool * pp)264 netif_gso_tcp_segment_mbuf(struct mbuf *m, struct ifnet *ifp,
265     struct netif_gso_ip_tcp_state *state, struct kern_pbufpool *pp)
266 {
267 	uuid_t euuid;
268 	struct pktq pktq_alloc, pktq_seg;
269 	uint64_t timestamp = 0;
270 	uint64_t pflags;
271 	int error = 0;
272 	uint32_t policy_id;
273 	uint32_t svc_class;
274 	uint32_t n, n_pkts, n_bytes;
275 	int32_t off = 0, total_len = m->m_pkthdr.len;
276 	uint8_t tx_headroom = (uint8_t)ifp->if_tx_headroom;
277 	struct netif_stats *nifs = &NA(ifp)->nifna_netif->nif_stats;
278 	struct __kern_packet *pkt_chain_head, *pkt_chain_tail;
279 	uint16_t mss = state->mss;
280 	bool skip_pktap;
281 
282 	VERIFY(total_len > state->hlen);
283 	VERIFY(((tx_headroom + state->mac_hlen) & 0x1) == 0);
284 	VERIFY((tx_headroom + state->hlen + mss) <= PP_BUF_SIZE_DEF(pp));
285 
286 	KPKTQ_INIT(&pktq_alloc);
287 	KPKTQ_INIT(&pktq_seg);
288 	/* batch allocate enough packets */
289 	n_pkts = (uint32_t)(SK_ROUNDUP((total_len - state->hlen), mss) / mss);
290 	error = pp_alloc_pktq(pp, 1, &pktq_alloc, n_pkts, NULL,
291 	    NULL, SKMEM_NOSLEEP);
292 	if (__improbable(error != 0)) {
293 		STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_NOMEM);
294 		SK_ERR("failed to alloc %u pkts", n_pkts);
295 		pp_free_pktq(&pktq_alloc);
296 		error = ENOBUFS;
297 		goto done;
298 	}
299 
300 	ASSERT(m->m_pkthdr.pkt_proto == IPPROTO_TCP);
301 	ASSERT((m->m_flags & M_BCAST) == 0);
302 	ASSERT((m->m_flags & M_MCAST) == 0);
303 	ASSERT(((m->m_pkthdr.pkt_flags & PKTF_TX_COMPL_TS_REQ) == 0));
304 	pflags = m->m_pkthdr.pkt_flags & PKT_F_COMMON_MASK;
305 	pflags |= PKTF_START_SEQ;
306 	(void) mbuf_get_timestamp(m, &timestamp, NULL);
307 	necp_get_app_uuid_from_packet(m, euuid);
308 	policy_id = necp_get_policy_id_from_packet(m);
309 	svc_class = m_get_service_class(m);
310 	skip_pktap = (m->m_pkthdr.pkt_flags & PKTF_SKIP_PKTAP) != 0 ||
311 	    pktap_total_tap_count == 0;
312 
313 	for (n = 1, off = state->hlen; off < total_len; off += mss, n++) {
314 		uint8_t *baddr, *baddr0;
315 		uint32_t partial;
316 		struct __kern_packet *pkt;
317 
318 		KPKTQ_DEQUEUE(&pktq_alloc, pkt);
319 		ASSERT(pkt != NULL);
320 
321 		/* get buffer address from packet */
322 		MD_BUFLET_ADDR_ABS(pkt, baddr0);
323 		baddr = baddr0;
324 		baddr += tx_headroom;
325 
326 		/*
327 		 * Copy the link-layer, IP and TCP header from the
328 		 * original packet.
329 		 */
330 		m_copydata(m, 0, state->hlen, baddr);
331 		baddr += state->hlen;
332 
333 		/*
334 		 * Copy the payload from original packet and
335 		 * compute partial checksum on the payload.
336 		 */
337 		if (off + mss > total_len) {
338 			/* if last segment is less than mss */
339 			mss = (uint16_t)(total_len - off);
340 		}
341 		partial = m_copydata_sum(m, off, mss, baddr, 0, NULL);
342 
343 		/*
344 		 * update packet metadata
345 		 */
346 		pkt->pkt_headroom = tx_headroom;
347 		pkt->pkt_l2_len = state->mac_hlen;
348 		pkt->pkt_link_flags = 0;
349 		pkt->pkt_csum_flags = 0;
350 		pkt->pkt_csum_tx_start_off = 0;
351 		pkt->pkt_csum_tx_stuff_off = 0;
352 		uuid_copy(pkt->pkt_policy_euuid, euuid);
353 		pkt->pkt_policy_id = policy_id;
354 		pkt->pkt_timestamp = timestamp;
355 		pkt->pkt_svc_class = svc_class;
356 		pkt->pkt_pflags |= pflags;
357 		pkt->pkt_flowsrc_type = m->m_pkthdr.pkt_flowsrc;
358 		pkt->pkt_flow_token = m->m_pkthdr.pkt_flowid;
359 		pkt->pkt_comp_gencnt = m->m_pkthdr.comp_gencnt;
360 		pkt->pkt_flow_ip_proto = IPPROTO_TCP;
361 		pkt->pkt_transport_protocol = IPPROTO_TCP;
362 		pkt->pkt_flow_tcp_seq = htonl(state->tcp_seq);
363 
364 		state->update(state, pkt, baddr0);
365 		/*
366 		 * FIN or PUSH flags if present will be set only on the last
367 		 * segment.
368 		 */
369 		if (n != n_pkts) {
370 			state->tcp->th_flags &= ~(TH_FIN | TH_PUSH);
371 		}
372 		/*
373 		 * CWR flag if present is set only on the first segment
374 		 * and cleared on the subsequent segments.
375 		 */
376 		if (n != 1) {
377 			state->tcp->th_flags &= ~TH_CWR;
378 			state->tcp->th_seq = htonl(state->tcp_seq);
379 		}
380 		ASSERT(state->tcp->th_seq == pkt->pkt_flow_tcp_seq);
381 		state->internal(state, partial, mss);
382 		METADATA_ADJUST_LEN(pkt, state->hlen + mss, tx_headroom);
383 		VERIFY(__packet_finalize(SK_PKT2PH(pkt)) == 0);
384 		KPKTQ_ENQUEUE(&pktq_seg, pkt);
385 		if (!skip_pktap) {
386 			nx_netif_pktap_output(ifp, state->af, pkt);
387 		}
388 	}
389 	ASSERT(off == total_len);
390 	STATS_ADD(nifs, NETIF_STATS_GSO_SEG, n_pkts);
391 
392 	/* ifnet_enqueue_pkt_chain() consumes the packet chain */
393 	pkt_chain_head = KPKTQ_FIRST(&pktq_seg);
394 	pkt_chain_tail = KPKTQ_LAST(&pktq_seg);
395 	KPKTQ_INIT(&pktq_seg);
396 	n_bytes = total_len + (state->hlen * (n_pkts - 1));
397 
398 	error = netif_gso_send(ifp, pkt_chain_head, pkt_chain_tail,
399 	    n_pkts, n_bytes);
400 
401 done:
402 	KPKTQ_FINI(&pktq_alloc);
403 	return error;
404 }
405 
406 /*
407  * Update the pointers to TCP and IPv4 headers
408  */
409 static void
netif_gso_ipv4_tcp_update(struct netif_gso_ip_tcp_state * state,struct __kern_packet * pkt,uint8_t * baddr)410 netif_gso_ipv4_tcp_update(struct netif_gso_ip_tcp_state *state,
411     struct __kern_packet *pkt, uint8_t *baddr)
412 {
413 	state->hdr.ip = (struct ip *)(void *)(baddr + pkt->pkt_headroom +
414 	    pkt->pkt_l2_len);
415 	state->tcp = (struct tcphdr *)(void *)((caddr_t)(state->hdr.ip) +
416 	    state->ip_hlen);
417 }
418 
419 /*
420  * Finalize the TCP and IPv4 headers
421  */
422 static void
netif_gso_ipv4_tcp_internal(struct netif_gso_ip_tcp_state * state,uint32_t partial,uint16_t payload_len)423 netif_gso_ipv4_tcp_internal(struct netif_gso_ip_tcp_state *state,
424     uint32_t partial, uint16_t payload_len)
425 {
426 	/*
427 	 * Update IP header
428 	 */
429 	state->hdr.ip->ip_id = htons((state->ip_id)++);
430 	state->hdr.ip->ip_len = htons(state->ip_hlen + state->tcp_hlen +
431 	    payload_len);
432 	/*
433 	 * IP header checksum
434 	 */
435 	state->hdr.ip->ip_sum = 0;
436 	state->hdr.ip->ip_sum = inet_cksum_buffer(state->hdr.ip, 0, 0,
437 	    state->ip_hlen);
438 	/*
439 	 * TCP Checksum
440 	 */
441 	state->tcp->th_sum = 0;
442 	partial = __packet_cksum(state->tcp, state->tcp_hlen, partial);
443 	partial += htons(state->tcp_hlen + IPPROTO_TCP + payload_len);
444 	partial += state->psuedo_hdr_csum;
445 	ADDCARRY(partial);
446 	state->tcp->th_sum = ~(uint16_t)partial;
447 	/*
448 	 * Update tcp sequence number in gso state
449 	 */
450 	state->tcp_seq += payload_len;
451 }
452 
453 /*
454  * Updates the pointers to TCP and IPv6 headers
455  */
456 static void
netif_gso_ipv6_tcp_update(struct netif_gso_ip_tcp_state * state,struct __kern_packet * pkt,uint8_t * baddr)457 netif_gso_ipv6_tcp_update(struct netif_gso_ip_tcp_state *state,
458     struct __kern_packet *pkt, uint8_t *baddr)
459 {
460 	state->hdr.ip6 = (struct ip6_hdr *)(baddr + pkt->pkt_headroom +
461 	    pkt->pkt_l2_len);
462 	state->tcp = (struct tcphdr *)(void *)((caddr_t)(state->hdr.ip6) +
463 	    state->ip_hlen);
464 }
465 
466 /*
467  * Finalize the TCP and IPv6 headers
468  */
469 static void
netif_gso_ipv6_tcp_internal(struct netif_gso_ip_tcp_state * state,uint32_t partial,uint16_t payload_len)470 netif_gso_ipv6_tcp_internal(struct netif_gso_ip_tcp_state *state,
471     uint32_t partial, uint16_t payload_len)
472 {
473 	/*
474 	 * Update IP header
475 	 */
476 	state->hdr.ip6->ip6_plen = htons(state->tcp_hlen + payload_len);
477 	/*
478 	 * TCP Checksum
479 	 */
480 	state->tcp->th_sum = 0;
481 	partial = __packet_cksum(state->tcp, state->tcp_hlen, partial);
482 	partial += htonl(state->tcp_hlen + IPPROTO_TCP + payload_len);
483 	partial += state->psuedo_hdr_csum;
484 	ADDCARRY(partial);
485 	state->tcp->th_sum = ~(uint16_t)partial;
486 	/*
487 	 * Update tcp sequence number
488 	 */
489 	state->tcp_seq += payload_len;
490 }
491 
492 /*
493  * Init the state during the TCP segmentation
494  */
495 static inline void
netif_gso_ip_tcp_init_state(struct netif_gso_ip_tcp_state * state,struct mbuf * m,uint8_t mac_hlen,uint8_t ip_hlen,bool isipv6)496 netif_gso_ip_tcp_init_state(struct netif_gso_ip_tcp_state *state,
497     struct mbuf *m, uint8_t mac_hlen, uint8_t ip_hlen, bool isipv6)
498 {
499 	if (isipv6) {
500 		state->af = AF_INET6;
501 		state->hdr.ip6 = (struct ip6_hdr *)(mtod(m, uint8_t *) +
502 		    mac_hlen);
503 		/* should be atleast 16 bit aligned */
504 		VERIFY(((uintptr_t)state->hdr.ip6 & (uintptr_t)0x1) == 0);
505 		state->tcp = (struct tcphdr *)(void *)((caddr_t)
506 		    (state->hdr.ip6) + ip_hlen);
507 		state->update = netif_gso_ipv6_tcp_update;
508 		state->internal = netif_gso_ipv6_tcp_internal;
509 		state->psuedo_hdr_csum = in6_pseudo(&state->hdr.ip6->ip6_src,
510 		    &state->hdr.ip6->ip6_dst, 0);
511 	} else {
512 		struct in_addr ip_src, ip_dst;
513 
514 		state->af = AF_INET;
515 		state->hdr.ip = (struct ip *)(void *)(mtod(m, uint8_t *) +
516 		    mac_hlen);
517 		/* should be atleast 16 bit aligned */
518 		VERIFY(((uintptr_t)state->hdr.ip & (uintptr_t)0x1) == 0);
519 		state->ip_id = ntohs(state->hdr.ip->ip_id);
520 		state->tcp = (struct tcphdr *)(void *)((caddr_t)
521 		    (state->hdr.ip) + ip_hlen);
522 		state->update = netif_gso_ipv4_tcp_update;
523 		state->internal = netif_gso_ipv4_tcp_internal;
524 		bcopy(&state->hdr.ip->ip_src, &ip_src, sizeof(ip_src));
525 		bcopy(&state->hdr.ip->ip_dst, &ip_dst, sizeof(ip_dst));
526 		state->psuedo_hdr_csum = in_pseudo(ip_src.s_addr,
527 		    ip_dst.s_addr, 0);
528 	}
529 
530 	state->mac_hlen = mac_hlen;
531 	state->ip_hlen = ip_hlen;
532 	state->tcp_hlen = (uint8_t)(state->tcp->th_off << 2);
533 	state->hlen = mac_hlen + ip_hlen + state->tcp_hlen;
534 	VERIFY(m->m_pkthdr.tso_segsz != 0);
535 	state->mss = (uint16_t)m->m_pkthdr.tso_segsz;
536 	state->tcp_seq = ntohl(state->tcp->th_seq);
537 }
538 
539 /*
540  * GSO on TCP/IPv4
541  */
542 static int
netif_gso_ipv4_tcp(struct ifnet * ifp,struct mbuf * m)543 netif_gso_ipv4_tcp(struct ifnet *ifp, struct mbuf *m)
544 {
545 	struct ip *ip;
546 	struct kern_pbufpool *pp = NULL;
547 	struct netif_gso_ip_tcp_state state;
548 	uint16_t hlen;
549 	uint8_t ip_hlen;
550 	uint8_t mac_hlen;
551 	struct netif_stats *nifs = &NA(ifp)->nifna_netif->nif_stats;
552 	boolean_t pkt_dropped = false;
553 	int error;
554 
555 	STATS_INC(nifs, NETIF_STATS_GSO_PKT);
556 	if (__improbable(m->m_pkthdr.pkt_proto != IPPROTO_TCP)) {
557 		STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_NONTCP);
558 		error = ENOTSUP;
559 		pkt_dropped = true;
560 		goto done;
561 	}
562 
563 	error = netif_gso_check_netif_active(ifp, m, &pp);
564 	if (__improbable(error != 0)) {
565 		STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_NA_INACTIVE);
566 		error = ENXIO;
567 		pkt_dropped = true;
568 		goto done;
569 	}
570 
571 	error = netif_gso_get_frame_header_len(m, &mac_hlen);
572 	if (__improbable(error != 0)) {
573 		STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_BADLEN);
574 		pkt_dropped = true;
575 		goto done;
576 	}
577 
578 	hlen = mac_hlen + sizeof(struct ip);
579 	if (__improbable(m->m_len < hlen)) {
580 		m = m_pullup(m, hlen);
581 		if (m == NULL) {
582 			STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_NOMEM);
583 			error = ENOBUFS;
584 			pkt_dropped = true;
585 			goto done;
586 		}
587 	}
588 	ip = (struct ip *)(void *)(mtod(m, uint8_t *) + mac_hlen);
589 	ip_hlen = (uint8_t)(ip->ip_hl << 2);
590 	hlen = mac_hlen + ip_hlen + sizeof(struct tcphdr);
591 	if (__improbable(m->m_len < hlen)) {
592 		m = m_pullup(m, hlen);
593 		if (m == NULL) {
594 			STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_NOMEM);
595 			error = ENOBUFS;
596 			pkt_dropped = true;
597 			goto done;
598 		}
599 	}
600 	netif_gso_ip_tcp_init_state(&state, m, mac_hlen, ip_hlen, false);
601 	error = netif_gso_tcp_segment_mbuf(m, ifp, &state, pp);
602 done:
603 	m_freem(m);
604 	if (__improbable(pkt_dropped)) {
605 		STATS_INC(nifs, NETIF_STATS_DROP);
606 	}
607 	return error;
608 }
609 
610 /*
611  * GSO on TCP/IPv6
612  */
613 static int
netif_gso_ipv6_tcp(struct ifnet * ifp,struct mbuf * m)614 netif_gso_ipv6_tcp(struct ifnet *ifp, struct mbuf *m)
615 {
616 	struct ip6_hdr *ip6;
617 	struct kern_pbufpool *pp = NULL;
618 	struct netif_gso_ip_tcp_state state;
619 	int lasthdr_off;
620 	uint16_t hlen;
621 	uint8_t ip_hlen;
622 	uint8_t mac_hlen;
623 	struct netif_stats *nifs = &NA(ifp)->nifna_netif->nif_stats;
624 	boolean_t pkt_dropped = false;
625 	int error;
626 
627 	STATS_INC(nifs, NETIF_STATS_GSO_PKT);
628 	if (__improbable(m->m_pkthdr.pkt_proto != IPPROTO_TCP)) {
629 		STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_NONTCP);
630 		error = ENOTSUP;
631 		pkt_dropped = true;
632 		goto done;
633 	}
634 
635 	error = netif_gso_check_netif_active(ifp, m, &pp);
636 	if (__improbable(error != 0)) {
637 		STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_NA_INACTIVE);
638 		error = ENXIO;
639 		pkt_dropped = true;
640 		goto done;
641 	}
642 
643 	error = netif_gso_get_frame_header_len(m, &mac_hlen);
644 	if (__improbable(error != 0)) {
645 		STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_BADLEN);
646 		pkt_dropped = true;
647 		goto done;
648 	}
649 
650 	hlen = mac_hlen + sizeof(struct ip6_hdr);
651 	if (__improbable(m->m_len < hlen)) {
652 		m = m_pullup(m, hlen);
653 		if (m == NULL) {
654 			STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_NOMEM);
655 			error = ENOBUFS;
656 			pkt_dropped = true;
657 			goto done;
658 		}
659 	}
660 	ip6 = (struct ip6_hdr *)(mtod(m, uint8_t *) + mac_hlen);
661 	lasthdr_off = ip6_lasthdr(m, mac_hlen, IPPROTO_IPV6, NULL) - mac_hlen;
662 	VERIFY(lasthdr_off <= UINT8_MAX);
663 	ip_hlen = (uint8_t)lasthdr_off;
664 	hlen = mac_hlen + ip_hlen + sizeof(struct tcphdr);
665 	if (__improbable(m->m_len < hlen)) {
666 		m = m_pullup(m, hlen);
667 		if (m == NULL) {
668 			STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_NOMEM);
669 			error = ENOBUFS;
670 			pkt_dropped = true;
671 			goto done;
672 		}
673 	}
674 	netif_gso_ip_tcp_init_state(&state, m, mac_hlen, ip_hlen, true);
675 	error = netif_gso_tcp_segment_mbuf(m, ifp, &state, pp);
676 done:
677 	m_freem(m);
678 	if (__improbable(pkt_dropped)) {
679 		STATS_INC(nifs, NETIF_STATS_DROP);
680 	}
681 	return error;
682 }
683 
684 int
netif_gso_dispatch(struct ifnet * ifp,struct mbuf * m)685 netif_gso_dispatch(struct ifnet *ifp, struct mbuf *m)
686 {
687 	int gso_flags;
688 
689 	ASSERT(m->m_nextpkt == NULL);
690 	gso_flags = CSUM_TO_GSO(m->m_pkthdr.csum_flags);
691 	VERIFY(gso_flags < GSO_END_OF_TYPE);
692 	return netif_gso_functions[gso_flags](ifp, m);
693 }
694 
695 void
netif_gso_init(void)696 netif_gso_init(void)
697 {
698 	_CASSERT(CSUM_TO_GSO(~(CSUM_TSO_IPV4 | CSUM_TSO_IPV6)) == GSO_NONE);
699 	_CASSERT(CSUM_TO_GSO(CSUM_TSO_IPV4) == GSO_TCP4);
700 	_CASSERT(CSUM_TO_GSO(CSUM_TSO_IPV6) == GSO_TCP6);
701 	netif_gso_functions[GSO_NONE] = nx_netif_host_output;
702 	netif_gso_functions[GSO_TCP4] = netif_gso_ipv4_tcp;
703 	netif_gso_functions[GSO_TCP6] = netif_gso_ipv6_tcp;
704 }
705 
706 void
netif_gso_fini(void)707 netif_gso_fini(void)
708 {
709 	netif_gso_functions[GSO_NONE] = NULL;
710 	netif_gso_functions[GSO_TCP4] = NULL;
711 	netif_gso_functions[GSO_TCP6] = NULL;
712 }
713