xref: /xnu-11417.121.6/bsd/skywalk/nexus/netif/nx_netif_gso.c (revision a1e26a70f38d1d7daa7b49b258e2f8538ad81650)
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 *__bidi_indexable baddr);
108 	void (*internal)(struct netif_gso_ip_tcp_state*, uint32_t partial,
109 	    uint16_t payload_len, uint32_t *csum_flags);
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 	boolean_t copy_data_sum;
126 };
127 
128 static inline uint8_t
netif_gso_get_frame_header_len(struct mbuf * m,uint8_t * hlen)129 netif_gso_get_frame_header_len(struct mbuf *m, uint8_t *hlen)
130 {
131 	uint64_t len;
132 	char *__single ph = m->m_pkthdr.pkt_hdr;
133 
134 	if (__improbable(m_pktlen(m) == 0 || ph == NULL ||
135 	    ph < (char *)m->m_data)) {
136 		return ERANGE;
137 	}
138 	len = (ph - m_mtod_current(m));
139 	if (__improbable(len > UINT8_MAX)) {
140 		return ERANGE;
141 	}
142 	*hlen = (uint8_t)len;
143 	return 0;
144 }
145 
146 static inline int
netif_gso_check_netif_active(struct ifnet * ifp,struct mbuf * m,struct kern_pbufpool ** pp)147 netif_gso_check_netif_active(struct ifnet *ifp, struct mbuf *m,
148     struct kern_pbufpool **pp)
149 {
150 	struct __kern_channel_ring *kring;
151 	struct nx_netif *nif = NA(ifp)->nifna_netif;
152 	struct netif_stats *nifs = &nif->nif_stats;
153 	struct kern_nexus *nx = nif->nif_nx;
154 	struct nexus_adapter *hwna = nx_port_get_na(nx, NEXUS_PORT_NET_IF_DEV);
155 	uint32_t sc_idx = MBUF_SCIDX(m_get_service_class(m));
156 
157 	if (__improbable(!NA_IS_ACTIVE(hwna))) {
158 		STATS_INC(nifs, NETIF_STATS_DROP_NA_INACTIVE);
159 		SK_DF(SK_VERB_NETIF,
160 		    "\"%s\" (0x%llx) not in skywalk mode anymore",
161 		    hwna->na_name, SK_KVA(hwna));
162 		return ENXIO;
163 	}
164 
165 	VERIFY(sc_idx < KPKT_SC_MAX_CLASSES);
166 	kring = &hwna->na_tx_rings[hwna->na_kring_svc_lut[sc_idx]];
167 	if (__improbable(KR_DROP(kring))) {
168 		STATS_INC(nifs, NETIF_STATS_DROP_KRDROP_MODE);
169 		SK_DF(SK_VERB_NETIF,
170 		    "kr \"%s\" (0x%llx) krflags 0x%b or %s in drop mode",
171 		    kring->ckr_name, SK_KVA(kring), kring->ckr_flags,
172 		    CKRF_BITS, ifp->if_xname);
173 		return ENXIO;
174 	}
175 	*pp = kring->ckr_pp;
176 	return 0;
177 }
178 
179 boolean_t
netif_chain_enqueue_enabled(struct ifnet * ifp)180 netif_chain_enqueue_enabled(struct ifnet *ifp)
181 {
182 	return netif_chain_enqueue != 0 && ifp->if_output_netem == NULL &&
183 	       (ifp->if_eflags & IFEF_ENQUEUE_MULTI) == 0;
184 }
185 
186 static inline int
netif_gso_send(struct ifnet * ifp,struct __kern_packet * head,struct __kern_packet * tail,uint32_t count,uint32_t bytes)187 netif_gso_send(struct ifnet *ifp, struct __kern_packet *head,
188     struct __kern_packet *tail, uint32_t count, uint32_t bytes)
189 {
190 	struct nx_netif *nif = NA(ifp)->nifna_netif;
191 	struct netif_stats *nifs = &nif->nif_stats;
192 	struct netif_qset *__single qset = NULL;
193 	uint64_t qset_id = 0;
194 	int error = 0;
195 	boolean_t dropped;
196 
197 	if (NX_LLINK_PROV(nif->nif_nx) &&
198 	    ifp->if_traffic_rule_count > 0 &&
199 	    nxctl_inet_traffic_rule_find_qset_id_with_pkt(ifp->if_xname,
200 	    head, &qset_id) == 0) {
201 		qset = nx_netif_find_qset(nif, qset_id);
202 		ASSERT(qset != NULL);
203 	}
204 	if (netif_chain_enqueue_enabled(ifp)) {
205 		dropped = false;
206 		if (qset != NULL) {
207 			head->pkt_qset_idx = qset->nqs_idx;
208 			error = ifnet_enqueue_ifcq_pkt_chain(ifp, qset->nqs_ifcq,
209 			    head, tail, count, bytes, false, &dropped);
210 		} else {
211 			error = ifnet_enqueue_pkt_chain(ifp, head, tail,
212 			    count, bytes, false, &dropped);
213 		}
214 		if (__improbable(dropped)) {
215 			STATS_ADD(nifs, NETIF_STATS_TX_DROP_ENQ_AQM, count);
216 			STATS_ADD(nifs, NETIF_STATS_DROP, count);
217 		}
218 	} else {
219 		struct __kern_packet *pkt = head, *next;
220 		uint32_t c = 0, b = 0;
221 
222 		while (pkt != NULL) {
223 			int err;
224 
225 			next = pkt->pkt_nextpkt;
226 			pkt->pkt_nextpkt = NULL;
227 			c++;
228 			b += pkt->pkt_length;
229 
230 			dropped = false;
231 			if (qset != NULL) {
232 				pkt->pkt_qset_idx = qset->nqs_idx;
233 				err = ifnet_enqueue_ifcq_pkt(ifp, qset->nqs_ifcq,
234 				    pkt, false, &dropped);
235 			} else {
236 				err = ifnet_enqueue_pkt(ifp, pkt, false, &dropped);
237 			}
238 			if (error == 0 && __improbable(err != 0)) {
239 				error = err;
240 			}
241 			if (__improbable(dropped)) {
242 				STATS_INC(nifs, NETIF_STATS_TX_DROP_ENQ_AQM);
243 				STATS_INC(nifs, NETIF_STATS_DROP);
244 			}
245 			pkt = next;
246 		}
247 		ASSERT(c == count);
248 		ASSERT(b == bytes);
249 	}
250 	if (qset != NULL) {
251 		nx_netif_qset_release(&qset);
252 	}
253 	netif_transmit(ifp, NETIF_XMIT_FLAG_HOST);
254 	return error;
255 }
256 
257 /*
258  * Segment and transmit a queue of packets which fit the given mss + hdr_len.
259  * m points to mbuf chain to be segmented.
260  * This function splits the payload (m-> m_pkthdr.len - hdr_len)
261  * into segments of length MSS bytes and then copy the first hdr_len bytes
262  * from m at the top of each segment.
263  */
264 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)265 netif_gso_tcp_segment_mbuf(struct mbuf *m, struct ifnet *ifp,
266     struct netif_gso_ip_tcp_state *state, struct kern_pbufpool *pp)
267 {
268 	uuid_t euuid;
269 	struct pktq pktq_alloc, pktq_seg;
270 	uint64_t timestamp = 0, m_tx_timestamp = 0;
271 	uint64_t pflags;
272 	int error = 0;
273 	uint32_t policy_id;
274 	uint32_t skip_policy_id;
275 	uint32_t svc_class;
276 	uint32_t n, n_pkts, n_bytes;
277 	int32_t off = 0, total_len = m->m_pkthdr.len;
278 	uint8_t tx_headroom = (uint8_t)ifp->if_tx_headroom;
279 	struct netif_stats *nifs = &NA(ifp)->nifna_netif->nif_stats;
280 	struct __kern_packet *pkt_chain_head, *pkt_chain_tail;
281 	struct m_tag *ts_tag = NULL;
282 	uint16_t mss = state->mss;
283 	bool skip_pktap;
284 
285 	VERIFY(total_len > state->hlen);
286 	VERIFY(((tx_headroom + state->mac_hlen) & 0x1) == 0);
287 	VERIFY((tx_headroom + state->hlen + mss) <= PP_BUF_SIZE_DEF(pp));
288 
289 	KPKTQ_INIT(&pktq_alloc);
290 	KPKTQ_INIT(&pktq_seg);
291 	/* batch allocate enough packets */
292 	n_pkts = (uint32_t)(SK_ROUNDUP((total_len - state->hlen), mss) / mss);
293 	error = pp_alloc_pktq(pp, 1, &pktq_alloc, n_pkts, NULL,
294 	    NULL, SKMEM_NOSLEEP);
295 	if (__improbable(error != 0)) {
296 		STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_NOMEM);
297 		SK_ERR("failed to alloc %u pkts", n_pkts);
298 		pp_free_pktq(&pktq_alloc);
299 		error = ENOBUFS;
300 		goto done;
301 	}
302 
303 	ASSERT(m->m_pkthdr.pkt_proto == IPPROTO_TCP);
304 	ASSERT((m->m_flags & M_BCAST) == 0);
305 	ASSERT((m->m_flags & M_MCAST) == 0);
306 	ASSERT(((m->m_pkthdr.pkt_flags & PKTF_TX_COMPL_TS_REQ) == 0));
307 	pflags = m->m_pkthdr.pkt_flags & PKT_F_COMMON_MASK;
308 	pflags |= PKTF_START_SEQ;
309 	pflags |= (m->m_pkthdr.pkt_ext_flags & PKTF_EXT_L4S) ? PKT_F_L4S : 0;
310 	(void) mbuf_get_timestamp(m, &timestamp, NULL);
311 	necp_get_app_uuid_from_packet(m, euuid);
312 	policy_id = necp_get_policy_id_from_packet(m);
313 	skip_policy_id = necp_get_skip_policy_id_from_packet(m);
314 	svc_class = m_get_service_class(m);
315 	skip_pktap = (m->m_pkthdr.pkt_flags & PKTF_SKIP_PKTAP) != 0 ||
316 	    pktap_total_tap_count == 0;
317 
318 	ts_tag = m_tag_locate(m, KERNEL_MODULE_TAG_ID, KERNEL_TAG_TYPE_AQM);
319 	if (ts_tag != NULL) {
320 		m_tx_timestamp = *(uint64_t *)(ts_tag->m_tag_data);
321 	}
322 
323 	for (n = 1, off = state->hlen; off < total_len; off += mss, n++) {
324 		uint8_t *baddr, *baddr0;
325 		uint32_t partial = 0;
326 		struct __kern_packet *pkt;
327 
328 		KPKTQ_DEQUEUE(&pktq_alloc, pkt);
329 		ASSERT(pkt != NULL);
330 
331 		/* get buffer address from packet */
332 		MD_BUFLET_ADDR_ABS(pkt, baddr0);
333 		baddr = baddr0;
334 		baddr += tx_headroom;
335 
336 		/*
337 		 * Copy the link-layer, IP and TCP header from the
338 		 * original packet.
339 		 */
340 		m_copydata(m, 0, state->hlen, baddr);
341 		baddr += state->hlen;
342 
343 		/*
344 		 * Copy the payload from original packet and
345 		 * compute partial checksum on the payload.
346 		 */
347 		if (off + mss > total_len) {
348 			/* if last segment is less than mss */
349 			mss = (uint16_t)(total_len - off);
350 		}
351 		if (state->copy_data_sum) {
352 			partial = m_copydata_sum(m, off, mss, baddr, 0, NULL);
353 		} else {
354 			m_copydata(m, off, mss, baddr);
355 		}
356 
357 		/*
358 		 * update packet metadata
359 		 */
360 		pkt->pkt_headroom = tx_headroom;
361 		pkt->pkt_l2_len = state->mac_hlen;
362 		pkt->pkt_link_flags = 0;
363 		pkt->pkt_csum_flags = 0;
364 		pkt->pkt_csum_tx_start_off = 0;
365 		pkt->pkt_csum_tx_stuff_off = 0;
366 		uuid_copy(pkt->pkt_policy_euuid, euuid);
367 		pkt->pkt_policy_id = policy_id;
368 		pkt->pkt_skip_policy_id = skip_policy_id;
369 		pkt->pkt_timestamp = timestamp;
370 		pkt->pkt_svc_class = svc_class;
371 		pkt->pkt_pflags |= pflags;
372 		pkt->pkt_flowsrc_type = m->m_pkthdr.pkt_flowsrc;
373 		pkt->pkt_flow_token = m->m_pkthdr.pkt_flowid;
374 		pkt->pkt_comp_gencnt = m->m_pkthdr.comp_gencnt;
375 		pkt->pkt_flow_ip_proto = IPPROTO_TCP;
376 		pkt->pkt_transport_protocol = IPPROTO_TCP;
377 		pkt->pkt_flow_tcp_seq = htonl(state->tcp_seq);
378 		__packet_set_tx_timestamp(SK_PKT2PH(pkt), m_tx_timestamp);
379 
380 		state->update(state, pkt, baddr0);
381 		/*
382 		 * FIN or PUSH flags if present will be set only on the last
383 		 * segment.
384 		 */
385 		if (n != n_pkts) {
386 			state->tcp->th_flags &= ~(TH_FIN | TH_PUSH);
387 		}
388 		/*
389 		 * CWR flag if present is set only on the first segment
390 		 * and cleared on the subsequent segments.
391 		 */
392 		if (n != 1) {
393 			state->tcp->th_flags &= ~TH_CWR;
394 			state->tcp->th_seq = htonl(state->tcp_seq);
395 		}
396 		ASSERT(state->tcp->th_seq == pkt->pkt_flow_tcp_seq);
397 		state->internal(state, partial, mss, &pkt->pkt_csum_flags);
398 		METADATA_ADJUST_LEN(pkt, state->hlen + mss, tx_headroom);
399 		VERIFY(__packet_finalize(SK_PKT2PH(pkt)) == 0);
400 		KPKTQ_ENQUEUE(&pktq_seg, pkt);
401 		if (!skip_pktap) {
402 			nx_netif_pktap_output(ifp, state->af, pkt);
403 		}
404 	}
405 	ASSERT(off == total_len);
406 	STATS_ADD(nifs, NETIF_STATS_GSO_SEG, n_pkts);
407 
408 	/* ifnet_enqueue_pkt_chain() consumes the packet chain */
409 	pkt_chain_head = KPKTQ_FIRST(&pktq_seg);
410 	pkt_chain_tail = KPKTQ_LAST(&pktq_seg);
411 	KPKTQ_INIT(&pktq_seg);
412 	n_bytes = total_len + (state->hlen * (n_pkts - 1));
413 
414 	error = netif_gso_send(ifp, pkt_chain_head, pkt_chain_tail,
415 	    n_pkts, n_bytes);
416 
417 done:
418 	KPKTQ_FINI(&pktq_alloc);
419 	return error;
420 }
421 
422 /*
423  * Update the pointers to TCP and IPv4 headers
424  */
425 static void
netif_gso_ipv4_tcp_update(struct netif_gso_ip_tcp_state * state,struct __kern_packet * pkt,uint8_t * __bidi_indexable baddr)426 netif_gso_ipv4_tcp_update(struct netif_gso_ip_tcp_state *state,
427     struct __kern_packet *pkt, uint8_t *__bidi_indexable baddr)
428 {
429 	state->hdr.ip = (struct ip *)(void *)(baddr + pkt->pkt_headroom +
430 	    pkt->pkt_l2_len);
431 	state->tcp = (struct tcphdr *)(void *)(baddr + pkt->pkt_headroom +
432 	    pkt->pkt_l2_len +  state->ip_hlen);
433 }
434 
435 /*
436  * Finalize the TCP and IPv4 headers
437  */
438 static void
netif_gso_ipv4_tcp_internal(struct netif_gso_ip_tcp_state * state,uint32_t partial,uint16_t payload_len,uint32_t * csum_flags __unused)439 netif_gso_ipv4_tcp_internal(struct netif_gso_ip_tcp_state *state,
440     uint32_t partial, uint16_t payload_len, uint32_t *csum_flags __unused)
441 {
442 	int hlen;
443 	uint8_t *__sized_by(hlen) buffer;
444 
445 	/*
446 	 * Update IP header
447 	 */
448 	state->hdr.ip->ip_id = htons((state->ip_id)++);
449 	state->hdr.ip->ip_len = htons(state->ip_hlen + state->tcp_hlen +
450 	    payload_len);
451 	/*
452 	 * IP header checksum
453 	 */
454 	state->hdr.ip->ip_sum = 0;
455 	buffer = (uint8_t *__bidi_indexable)(struct ip *__bidi_indexable)
456 	    state->hdr.ip;
457 	hlen = state->ip_hlen;
458 	state->hdr.ip->ip_sum = inet_cksum_buffer(buffer, 0, 0, hlen);
459 	/*
460 	 * TCP Checksum
461 	 */
462 	state->tcp->th_sum = 0;
463 	partial = __packet_cksum(state->tcp, state->tcp_hlen, partial);
464 	partial += htons(state->tcp_hlen + IPPROTO_TCP + payload_len);
465 	partial += state->psuedo_hdr_csum;
466 	ADDCARRY(partial);
467 	state->tcp->th_sum = ~(uint16_t)partial;
468 	/*
469 	 * Update tcp sequence number in gso state
470 	 */
471 	state->tcp_seq += payload_len;
472 }
473 
474 static void
netif_gso_ipv4_tcp_internal_nosum(struct netif_gso_ip_tcp_state * state,uint32_t partial __unused,uint16_t payload_len __unused,uint32_t * csum_flags)475 netif_gso_ipv4_tcp_internal_nosum(struct netif_gso_ip_tcp_state *state,
476     uint32_t partial __unused, uint16_t payload_len __unused,
477     uint32_t *csum_flags)
478 {
479 	/*
480 	 * Update IP header
481 	 */
482 	state->hdr.ip->ip_id = htons((state->ip_id)++);
483 	state->hdr.ip->ip_len = htons(state->ip_hlen + state->tcp_hlen +
484 	    payload_len);
485 	/*
486 	 * Update tcp sequence number in gso state
487 	 */
488 	state->tcp_seq += payload_len;
489 
490 	/* offload csum to hardware */
491 	*csum_flags |= PACKET_CSUM_IP | PACKET_CSUM_TCP;
492 }
493 
494 /*
495  * Updates the pointers to TCP and IPv6 headers
496  */
497 static void
netif_gso_ipv6_tcp_update(struct netif_gso_ip_tcp_state * state,struct __kern_packet * pkt,uint8_t * __bidi_indexable baddr)498 netif_gso_ipv6_tcp_update(struct netif_gso_ip_tcp_state *state,
499     struct __kern_packet *pkt, uint8_t *__bidi_indexable baddr)
500 {
501 	state->hdr.ip6 = (struct ip6_hdr *)(baddr + pkt->pkt_headroom +
502 	    pkt->pkt_l2_len);
503 	state->tcp = (struct tcphdr *)(void *)(baddr + pkt->pkt_headroom +
504 	    pkt->pkt_l2_len + state->ip_hlen);
505 }
506 
507 /*
508  * Finalize the TCP and IPv6 headers
509  */
510 static void
netif_gso_ipv6_tcp_internal_nosum(struct netif_gso_ip_tcp_state * state,uint32_t partial __unused,uint16_t payload_len __unused,uint32_t * csum_flags)511 netif_gso_ipv6_tcp_internal_nosum(struct netif_gso_ip_tcp_state *state,
512     uint32_t partial __unused, uint16_t payload_len __unused,
513     uint32_t *csum_flags)
514 {
515 	/*
516 	 * Update IP header
517 	 */
518 	state->hdr.ip6->ip6_plen = htons(state->tcp_hlen + payload_len);
519 
520 	/*
521 	 * Update tcp sequence number
522 	 */
523 	state->tcp_seq += payload_len;
524 
525 	/* offload csum to hardware */
526 	*csum_flags |= PACKET_CSUM_TCPIPV6;
527 }
528 
529 /*
530  * Finalize the TCP and IPv6 headers
531  */
532 static void
netif_gso_ipv6_tcp_internal(struct netif_gso_ip_tcp_state * state,uint32_t partial,uint16_t payload_len,uint32_t * csum_flags __unused)533 netif_gso_ipv6_tcp_internal(struct netif_gso_ip_tcp_state *state,
534     uint32_t partial, uint16_t payload_len, uint32_t *csum_flags __unused)
535 {
536 	/*
537 	 * Update IP header
538 	 */
539 	state->hdr.ip6->ip6_plen = htons(state->tcp_hlen + payload_len);
540 	/*
541 	 * TCP Checksum
542 	 */
543 	state->tcp->th_sum = 0;
544 	partial = __packet_cksum(state->tcp, state->tcp_hlen, partial);
545 	partial += htonl(state->tcp_hlen + IPPROTO_TCP + payload_len);
546 	partial += state->psuedo_hdr_csum;
547 	ADDCARRY(partial);
548 	state->tcp->th_sum = ~(uint16_t)partial;
549 	/*
550 	 * Update tcp sequence number
551 	 */
552 	state->tcp_seq += payload_len;
553 }
554 
555 /*
556  * Init the state during the TCP segmentation
557  */
558 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,ifnet_t ifp)559 netif_gso_ip_tcp_init_state(struct netif_gso_ip_tcp_state *state,
560     struct mbuf *m, uint8_t mac_hlen, uint8_t ip_hlen, bool isipv6, ifnet_t ifp)
561 {
562 	if (isipv6) {
563 		state->af = AF_INET6;
564 		state->hdr.ip6 = (struct ip6_hdr *)(mtod(m, uint8_t *) +
565 		    mac_hlen);
566 		/* should be atleast 16 bit aligned */
567 		VERIFY(((uintptr_t)state->hdr.ip6 & (uintptr_t)0x1) == 0);
568 		state->tcp = (struct tcphdr *)(void *)(m_mtod_current(m) +
569 		    mac_hlen + ip_hlen);
570 		state->update = netif_gso_ipv6_tcp_update;
571 		if (ifp->if_hwassist & IFNET_CSUM_TCPIPV6) {
572 			state->internal = netif_gso_ipv6_tcp_internal_nosum;
573 			state->copy_data_sum = false;
574 		} else {
575 			state->internal = netif_gso_ipv6_tcp_internal;
576 			state->copy_data_sum = true;
577 		}
578 		state->psuedo_hdr_csum = in6_pseudo(&state->hdr.ip6->ip6_src,
579 		    &state->hdr.ip6->ip6_dst, 0);
580 	} else {
581 		struct in_addr ip_src, ip_dst;
582 
583 		state->af = AF_INET;
584 		state->hdr.ip = (struct ip *)(void *)(mtod(m, uint8_t *) +
585 		    mac_hlen);
586 		/* should be atleast 16 bit aligned */
587 		VERIFY(((uintptr_t)state->hdr.ip & (uintptr_t)0x1) == 0);
588 		state->ip_id = ntohs(state->hdr.ip->ip_id);
589 		state->tcp = (struct tcphdr *)(void *)(m_mtod_current(m) +
590 		    mac_hlen + ip_hlen);
591 		state->update = netif_gso_ipv4_tcp_update;
592 		if ((ifp->if_hwassist & (IFNET_CSUM_IP | IFNET_CSUM_TCP)) ==
593 		    (IFNET_CSUM_IP | IFNET_CSUM_TCP)) {
594 			state->internal = netif_gso_ipv4_tcp_internal_nosum;
595 			state->copy_data_sum = false;
596 		} else {
597 			state->internal = netif_gso_ipv4_tcp_internal;
598 			state->copy_data_sum = true;
599 		}
600 		bcopy(&state->hdr.ip->ip_src, &ip_src, sizeof(ip_src));
601 		bcopy(&state->hdr.ip->ip_dst, &ip_dst, sizeof(ip_dst));
602 		state->psuedo_hdr_csum = in_pseudo(ip_src.s_addr,
603 		    ip_dst.s_addr, 0);
604 	}
605 
606 	state->mac_hlen = mac_hlen;
607 	state->ip_hlen = ip_hlen;
608 	state->tcp_hlen = (uint8_t)(state->tcp->th_off << 2);
609 	state->hlen = mac_hlen + ip_hlen + state->tcp_hlen;
610 	VERIFY(m->m_pkthdr.tso_segsz != 0);
611 	state->mss = (uint16_t)m->m_pkthdr.tso_segsz;
612 	state->tcp_seq = ntohl(state->tcp->th_seq);
613 }
614 
615 /*
616  * GSO on TCP/IPv4
617  */
618 static int
netif_gso_ipv4_tcp(struct ifnet * ifp,struct mbuf * m)619 netif_gso_ipv4_tcp(struct ifnet *ifp, struct mbuf *m)
620 {
621 	struct ip *ip;
622 	struct kern_pbufpool *__single pp = NULL;
623 	struct netif_gso_ip_tcp_state state;
624 	uint16_t hlen;
625 	uint8_t ip_hlen;
626 	uint8_t mac_hlen;
627 	struct netif_stats *nifs = &NA(ifp)->nifna_netif->nif_stats;
628 	boolean_t pkt_dropped = false;
629 	int error;
630 
631 	STATS_INC(nifs, NETIF_STATS_GSO_PKT);
632 	if (__improbable(m->m_pkthdr.pkt_proto != IPPROTO_TCP)) {
633 		STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_NONTCP);
634 		error = ENOTSUP;
635 		pkt_dropped = true;
636 		goto done;
637 	}
638 
639 	error = netif_gso_check_netif_active(ifp, m, &pp);
640 	if (__improbable(error != 0)) {
641 		STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_NA_INACTIVE);
642 		error = ENXIO;
643 		pkt_dropped = true;
644 		goto done;
645 	}
646 
647 	error = netif_gso_get_frame_header_len(m, &mac_hlen);
648 	if (__improbable(error != 0)) {
649 		STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_BADLEN);
650 		pkt_dropped = true;
651 		goto done;
652 	}
653 
654 	hlen = mac_hlen + sizeof(struct ip);
655 	if (__improbable(m->m_len < hlen)) {
656 		m = m_pullup(m, hlen);
657 		if (m == NULL) {
658 			STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_NOMEM);
659 			error = ENOBUFS;
660 			pkt_dropped = true;
661 			goto done;
662 		}
663 	}
664 	ip = (struct ip *)(void *)(mtod(m, uint8_t *) + mac_hlen);
665 	ip_hlen = (uint8_t)(ip->ip_hl << 2);
666 	hlen = mac_hlen + ip_hlen + sizeof(struct tcphdr);
667 	if (__improbable(m->m_len < hlen)) {
668 		m = m_pullup(m, hlen);
669 		if (m == NULL) {
670 			STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_NOMEM);
671 			error = ENOBUFS;
672 			pkt_dropped = true;
673 			goto done;
674 		}
675 	}
676 	netif_gso_ip_tcp_init_state(&state, m, mac_hlen, ip_hlen, false, ifp);
677 	error = netif_gso_tcp_segment_mbuf(m, ifp, &state, pp);
678 done:
679 	m_freem(m);
680 	if (__improbable(pkt_dropped)) {
681 		STATS_INC(nifs, NETIF_STATS_DROP);
682 	}
683 	return error;
684 }
685 
686 /*
687  * GSO on TCP/IPv6
688  */
689 static int
netif_gso_ipv6_tcp(struct ifnet * ifp,struct mbuf * m)690 netif_gso_ipv6_tcp(struct ifnet *ifp, struct mbuf *m)
691 {
692 	struct ip6_hdr *ip6;
693 	struct kern_pbufpool *__single pp = NULL;
694 	struct netif_gso_ip_tcp_state state;
695 	int lasthdr_off;
696 	uint16_t hlen;
697 	uint8_t ip_hlen;
698 	uint8_t mac_hlen;
699 	struct netif_stats *nifs = &NA(ifp)->nifna_netif->nif_stats;
700 	boolean_t pkt_dropped = false;
701 	int error;
702 
703 	STATS_INC(nifs, NETIF_STATS_GSO_PKT);
704 	if (__improbable(m->m_pkthdr.pkt_proto != IPPROTO_TCP)) {
705 		STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_NONTCP);
706 		error = ENOTSUP;
707 		pkt_dropped = true;
708 		goto done;
709 	}
710 
711 	error = netif_gso_check_netif_active(ifp, m, &pp);
712 	if (__improbable(error != 0)) {
713 		STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_NA_INACTIVE);
714 		error = ENXIO;
715 		pkt_dropped = true;
716 		goto done;
717 	}
718 
719 	error = netif_gso_get_frame_header_len(m, &mac_hlen);
720 	if (__improbable(error != 0)) {
721 		STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_BADLEN);
722 		pkt_dropped = true;
723 		goto done;
724 	}
725 
726 	hlen = mac_hlen + sizeof(struct ip6_hdr);
727 	if (__improbable(m->m_len < hlen)) {
728 		m = m_pullup(m, hlen);
729 		if (m == NULL) {
730 			STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_NOMEM);
731 			error = ENOBUFS;
732 			pkt_dropped = true;
733 			goto done;
734 		}
735 	}
736 	ip6 = (struct ip6_hdr *)(mtod(m, uint8_t *) + mac_hlen);
737 	lasthdr_off = ip6_lasthdr(m, mac_hlen, IPPROTO_IPV6, NULL) - mac_hlen;
738 	VERIFY(lasthdr_off <= UINT8_MAX);
739 	ip_hlen = (uint8_t)lasthdr_off;
740 	hlen = mac_hlen + ip_hlen + sizeof(struct tcphdr);
741 	if (__improbable(m->m_len < hlen)) {
742 		m = m_pullup(m, hlen);
743 		if (m == NULL) {
744 			STATS_INC(nifs, NETIF_STATS_GSO_PKT_DROP_NOMEM);
745 			error = ENOBUFS;
746 			pkt_dropped = true;
747 			goto done;
748 		}
749 	}
750 	netif_gso_ip_tcp_init_state(&state, m, mac_hlen, ip_hlen, true, ifp);
751 	error = netif_gso_tcp_segment_mbuf(m, ifp, &state, pp);
752 done:
753 	m_freem(m);
754 	if (__improbable(pkt_dropped)) {
755 		STATS_INC(nifs, NETIF_STATS_DROP);
756 	}
757 	return error;
758 }
759 
760 int
netif_gso_dispatch(struct ifnet * ifp,struct mbuf * m)761 netif_gso_dispatch(struct ifnet *ifp, struct mbuf *m)
762 {
763 	int gso_flags;
764 
765 	ASSERT(m->m_nextpkt == NULL);
766 	gso_flags = CSUM_TO_GSO(m->m_pkthdr.csum_flags);
767 	VERIFY(gso_flags < GSO_END_OF_TYPE);
768 	return netif_gso_functions[gso_flags](ifp, m);
769 }
770 
771 void
netif_gso_init(void)772 netif_gso_init(void)
773 {
774 	_CASSERT(CSUM_TO_GSO(~(CSUM_TSO_IPV4 | CSUM_TSO_IPV6)) == GSO_NONE);
775 	_CASSERT(CSUM_TO_GSO(CSUM_TSO_IPV4) == GSO_TCP4);
776 	_CASSERT(CSUM_TO_GSO(CSUM_TSO_IPV6) == GSO_TCP6);
777 	netif_gso_functions[GSO_NONE] = nx_netif_host_output;
778 	netif_gso_functions[GSO_TCP4] = netif_gso_ipv4_tcp;
779 	netif_gso_functions[GSO_TCP6] = netif_gso_ipv6_tcp;
780 }
781 
782 void
netif_gso_fini(void)783 netif_gso_fini(void)
784 {
785 	netif_gso_functions[GSO_NONE] = NULL;
786 	netif_gso_functions[GSO_TCP4] = NULL;
787 	netif_gso_functions[GSO_TCP6] = NULL;
788 }
789