• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2007-2017 Nicira, Inc.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of version 2 of the GNU General Public
6  * License as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16  * 02110-1301, USA
17  */
18 
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20 
21 #include <linux/skbuff.h>
22 #include <linux/in.h>
23 #include <linux/ip.h>
24 #include <linux/openvswitch.h>
25 #include <linux/netfilter_ipv6.h>
26 #include <linux/sctp.h>
27 #include <linux/tcp.h>
28 #include <linux/udp.h>
29 #include <linux/in6.h>
30 #include <linux/if_arp.h>
31 #include <linux/if_vlan.h>
32 
33 #include <net/dst.h>
34 #include <net/ip.h>
35 #include <net/ipv6.h>
36 #include <net/ip6_fib.h>
37 #include <net/checksum.h>
38 #include <net/dsfield.h>
39 #include <net/mpls.h>
40 #include <net/sctp/checksum.h>
41 
42 #include "datapath.h"
43 #include "flow.h"
44 #include "conntrack.h"
45 #include "vport.h"
46 
47 struct deferred_action {
48 	struct sk_buff *skb;
49 	const struct nlattr *actions;
50 	int actions_len;
51 
52 	/* Store pkt_key clone when creating deferred action. */
53 	struct sw_flow_key pkt_key;
54 };
55 
56 #define MAX_L2_LEN	(VLAN_ETH_HLEN + 3 * MPLS_HLEN)
57 struct ovs_frag_data {
58 	unsigned long dst;
59 	struct vport *vport;
60 	struct ovs_skb_cb cb;
61 	__be16 inner_protocol;
62 	u16 network_offset;	/* valid only for MPLS */
63 	u16 vlan_tci;
64 	__be16 vlan_proto;
65 	unsigned int l2_len;
66 	u8 mac_proto;
67 	u8 l2_data[MAX_L2_LEN];
68 };
69 
70 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
71 
72 #define DEFERRED_ACTION_FIFO_SIZE 10
73 #define OVS_RECURSION_LIMIT 5
74 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
75 struct action_fifo {
76 	int head;
77 	int tail;
78 	/* Deferred action fifo queue storage. */
79 	struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
80 };
81 
82 struct action_flow_keys {
83 	struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
84 };
85 
86 static struct action_fifo __percpu *action_fifos;
87 static struct action_flow_keys __percpu *flow_keys;
88 static DEFINE_PER_CPU(int, exec_actions_level);
89 
90 /* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
91  * space. Return NULL if out of key spaces.
92  */
clone_key(const struct sw_flow_key * key_)93 static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
94 {
95 	struct action_flow_keys *keys = this_cpu_ptr(flow_keys);
96 	int level = this_cpu_read(exec_actions_level);
97 	struct sw_flow_key *key = NULL;
98 
99 	if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
100 		key = &keys->key[level - 1];
101 		*key = *key_;
102 	}
103 
104 	return key;
105 }
106 
action_fifo_init(struct action_fifo * fifo)107 static void action_fifo_init(struct action_fifo *fifo)
108 {
109 	fifo->head = 0;
110 	fifo->tail = 0;
111 }
112 
action_fifo_is_empty(const struct action_fifo * fifo)113 static bool action_fifo_is_empty(const struct action_fifo *fifo)
114 {
115 	return (fifo->head == fifo->tail);
116 }
117 
action_fifo_get(struct action_fifo * fifo)118 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
119 {
120 	if (action_fifo_is_empty(fifo))
121 		return NULL;
122 
123 	return &fifo->fifo[fifo->tail++];
124 }
125 
action_fifo_put(struct action_fifo * fifo)126 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
127 {
128 	if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
129 		return NULL;
130 
131 	return &fifo->fifo[fifo->head++];
132 }
133 
134 /* Return true if fifo is not full */
add_deferred_actions(struct sk_buff * skb,const struct sw_flow_key * key,const struct nlattr * actions,const int actions_len)135 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
136 				    const struct sw_flow_key *key,
137 				    const struct nlattr *actions,
138 				    const int actions_len)
139 {
140 	struct action_fifo *fifo;
141 	struct deferred_action *da;
142 
143 	fifo = this_cpu_ptr(action_fifos);
144 	da = action_fifo_put(fifo);
145 	if (da) {
146 		da->skb = skb;
147 		da->actions = actions;
148 		da->actions_len = actions_len;
149 		da->pkt_key = *key;
150 	}
151 
152 	return da;
153 }
154 
invalidate_flow_key(struct sw_flow_key * key)155 static void invalidate_flow_key(struct sw_flow_key *key)
156 {
157 	key->mac_proto |= SW_FLOW_KEY_INVALID;
158 }
159 
is_flow_key_valid(const struct sw_flow_key * key)160 static bool is_flow_key_valid(const struct sw_flow_key *key)
161 {
162 	return !(key->mac_proto & SW_FLOW_KEY_INVALID);
163 }
164 
165 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
166 			 struct sw_flow_key *key,
167 			 u32 recirc_id,
168 			 const struct nlattr *actions, int len,
169 			 bool last, bool clone_flow_key);
170 
update_ethertype(struct sk_buff * skb,struct ethhdr * hdr,__be16 ethertype)171 static void update_ethertype(struct sk_buff *skb, struct ethhdr *hdr,
172 			     __be16 ethertype)
173 {
174 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
175 		__be16 diff[] = { ~(hdr->h_proto), ethertype };
176 
177 		skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
178 	}
179 
180 	hdr->h_proto = ethertype;
181 }
182 
push_mpls(struct sk_buff * skb,struct sw_flow_key * key,const struct ovs_action_push_mpls * mpls)183 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
184 		     const struct ovs_action_push_mpls *mpls)
185 {
186 	struct mpls_shim_hdr *new_mpls_lse;
187 
188 	/* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
189 	if (skb->encapsulation)
190 		return -ENOTSUPP;
191 
192 	if (skb_cow_head(skb, MPLS_HLEN) < 0)
193 		return -ENOMEM;
194 
195 	if (!skb->inner_protocol) {
196 		skb_set_inner_network_header(skb, skb->mac_len);
197 		skb_set_inner_protocol(skb, skb->protocol);
198 	}
199 
200 	skb_push(skb, MPLS_HLEN);
201 	memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
202 		skb->mac_len);
203 	skb_reset_mac_header(skb);
204 	skb_set_network_header(skb, skb->mac_len);
205 
206 	new_mpls_lse = mpls_hdr(skb);
207 	new_mpls_lse->label_stack_entry = mpls->mpls_lse;
208 
209 	skb_postpush_rcsum(skb, new_mpls_lse, MPLS_HLEN);
210 
211 	if (ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET)
212 		update_ethertype(skb, eth_hdr(skb), mpls->mpls_ethertype);
213 	skb->protocol = mpls->mpls_ethertype;
214 
215 	invalidate_flow_key(key);
216 	return 0;
217 }
218 
pop_mpls(struct sk_buff * skb,struct sw_flow_key * key,const __be16 ethertype)219 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
220 		    const __be16 ethertype)
221 {
222 	int err;
223 
224 	err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
225 	if (unlikely(err))
226 		return err;
227 
228 	skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
229 
230 	memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
231 		skb->mac_len);
232 
233 	__skb_pull(skb, MPLS_HLEN);
234 	skb_reset_mac_header(skb);
235 	skb_set_network_header(skb, skb->mac_len);
236 
237 	if (ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET) {
238 		struct ethhdr *hdr;
239 
240 		/* mpls_hdr() is used to locate the ethertype field correctly in the
241 		 * presence of VLAN tags.
242 		 */
243 		hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
244 		update_ethertype(skb, hdr, ethertype);
245 	}
246 	if (eth_p_mpls(skb->protocol))
247 		skb->protocol = ethertype;
248 
249 	invalidate_flow_key(key);
250 	return 0;
251 }
252 
set_mpls(struct sk_buff * skb,struct sw_flow_key * flow_key,const __be32 * mpls_lse,const __be32 * mask)253 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
254 		    const __be32 *mpls_lse, const __be32 *mask)
255 {
256 	struct mpls_shim_hdr *stack;
257 	__be32 lse;
258 	int err;
259 
260 	err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
261 	if (unlikely(err))
262 		return err;
263 
264 	stack = mpls_hdr(skb);
265 	lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
266 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
267 		__be32 diff[] = { ~(stack->label_stack_entry), lse };
268 
269 		skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
270 	}
271 
272 	stack->label_stack_entry = lse;
273 	flow_key->mpls.top_lse = lse;
274 	return 0;
275 }
276 
pop_vlan(struct sk_buff * skb,struct sw_flow_key * key)277 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
278 {
279 	int err;
280 
281 	err = skb_vlan_pop(skb);
282 	if (skb_vlan_tag_present(skb)) {
283 		invalidate_flow_key(key);
284 	} else {
285 		key->eth.vlan.tci = 0;
286 		key->eth.vlan.tpid = 0;
287 	}
288 	return err;
289 }
290 
push_vlan(struct sk_buff * skb,struct sw_flow_key * key,const struct ovs_action_push_vlan * vlan)291 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
292 		     const struct ovs_action_push_vlan *vlan)
293 {
294 	if (skb_vlan_tag_present(skb)) {
295 		invalidate_flow_key(key);
296 	} else {
297 		key->eth.vlan.tci = vlan->vlan_tci;
298 		key->eth.vlan.tpid = vlan->vlan_tpid;
299 	}
300 	return skb_vlan_push(skb, vlan->vlan_tpid,
301 			     ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
302 }
303 
304 /* 'src' is already properly masked. */
ether_addr_copy_masked(u8 * dst_,const u8 * src_,const u8 * mask_)305 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
306 {
307 	u16 *dst = (u16 *)dst_;
308 	const u16 *src = (const u16 *)src_;
309 	const u16 *mask = (const u16 *)mask_;
310 
311 	OVS_SET_MASKED(dst[0], src[0], mask[0]);
312 	OVS_SET_MASKED(dst[1], src[1], mask[1]);
313 	OVS_SET_MASKED(dst[2], src[2], mask[2]);
314 }
315 
set_eth_addr(struct sk_buff * skb,struct sw_flow_key * flow_key,const struct ovs_key_ethernet * key,const struct ovs_key_ethernet * mask)316 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
317 			const struct ovs_key_ethernet *key,
318 			const struct ovs_key_ethernet *mask)
319 {
320 	int err;
321 
322 	err = skb_ensure_writable(skb, ETH_HLEN);
323 	if (unlikely(err))
324 		return err;
325 
326 	skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
327 
328 	ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
329 			       mask->eth_src);
330 	ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
331 			       mask->eth_dst);
332 
333 	skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
334 
335 	ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
336 	ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
337 	return 0;
338 }
339 
340 /* pop_eth does not support VLAN packets as this action is never called
341  * for them.
342  */
pop_eth(struct sk_buff * skb,struct sw_flow_key * key)343 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
344 {
345 	skb_pull_rcsum(skb, ETH_HLEN);
346 	skb_reset_mac_header(skb);
347 	skb_reset_mac_len(skb);
348 
349 	/* safe right before invalidate_flow_key */
350 	key->mac_proto = MAC_PROTO_NONE;
351 	invalidate_flow_key(key);
352 	return 0;
353 }
354 
push_eth(struct sk_buff * skb,struct sw_flow_key * key,const struct ovs_action_push_eth * ethh)355 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
356 		    const struct ovs_action_push_eth *ethh)
357 {
358 	struct ethhdr *hdr;
359 
360 	/* Add the new Ethernet header */
361 	if (skb_cow_head(skb, ETH_HLEN) < 0)
362 		return -ENOMEM;
363 
364 	skb_push(skb, ETH_HLEN);
365 	skb_reset_mac_header(skb);
366 	skb_reset_mac_len(skb);
367 
368 	hdr = eth_hdr(skb);
369 	ether_addr_copy(hdr->h_source, ethh->addresses.eth_src);
370 	ether_addr_copy(hdr->h_dest, ethh->addresses.eth_dst);
371 	hdr->h_proto = skb->protocol;
372 
373 	skb_postpush_rcsum(skb, hdr, ETH_HLEN);
374 
375 	/* safe right before invalidate_flow_key */
376 	key->mac_proto = MAC_PROTO_ETHERNET;
377 	invalidate_flow_key(key);
378 	return 0;
379 }
380 
update_ip_l4_checksum(struct sk_buff * skb,struct iphdr * nh,__be32 addr,__be32 new_addr)381 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
382 				  __be32 addr, __be32 new_addr)
383 {
384 	int transport_len = skb->len - skb_transport_offset(skb);
385 
386 	if (nh->frag_off & htons(IP_OFFSET))
387 		return;
388 
389 	if (nh->protocol == IPPROTO_TCP) {
390 		if (likely(transport_len >= sizeof(struct tcphdr)))
391 			inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
392 						 addr, new_addr, true);
393 	} else if (nh->protocol == IPPROTO_UDP) {
394 		if (likely(transport_len >= sizeof(struct udphdr))) {
395 			struct udphdr *uh = udp_hdr(skb);
396 
397 			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
398 				inet_proto_csum_replace4(&uh->check, skb,
399 							 addr, new_addr, true);
400 				if (!uh->check)
401 					uh->check = CSUM_MANGLED_0;
402 			}
403 		}
404 	}
405 }
406 
set_ip_addr(struct sk_buff * skb,struct iphdr * nh,__be32 * addr,__be32 new_addr)407 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
408 			__be32 *addr, __be32 new_addr)
409 {
410 	update_ip_l4_checksum(skb, nh, *addr, new_addr);
411 	csum_replace4(&nh->check, *addr, new_addr);
412 	skb_clear_hash(skb);
413 	*addr = new_addr;
414 }
415 
update_ipv6_checksum(struct sk_buff * skb,u8 l4_proto,__be32 addr[4],const __be32 new_addr[4])416 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
417 				 __be32 addr[4], const __be32 new_addr[4])
418 {
419 	int transport_len = skb->len - skb_transport_offset(skb);
420 
421 	if (l4_proto == NEXTHDR_TCP) {
422 		if (likely(transport_len >= sizeof(struct tcphdr)))
423 			inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
424 						  addr, new_addr, true);
425 	} else if (l4_proto == NEXTHDR_UDP) {
426 		if (likely(transport_len >= sizeof(struct udphdr))) {
427 			struct udphdr *uh = udp_hdr(skb);
428 
429 			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
430 				inet_proto_csum_replace16(&uh->check, skb,
431 							  addr, new_addr, true);
432 				if (!uh->check)
433 					uh->check = CSUM_MANGLED_0;
434 			}
435 		}
436 	} else if (l4_proto == NEXTHDR_ICMP) {
437 		if (likely(transport_len >= sizeof(struct icmp6hdr)))
438 			inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
439 						  skb, addr, new_addr, true);
440 	}
441 }
442 
mask_ipv6_addr(const __be32 old[4],const __be32 addr[4],const __be32 mask[4],__be32 masked[4])443 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
444 			   const __be32 mask[4], __be32 masked[4])
445 {
446 	masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
447 	masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
448 	masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
449 	masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
450 }
451 
set_ipv6_addr(struct sk_buff * skb,u8 l4_proto,__be32 addr[4],const __be32 new_addr[4],bool recalculate_csum)452 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
453 			  __be32 addr[4], const __be32 new_addr[4],
454 			  bool recalculate_csum)
455 {
456 	if (recalculate_csum)
457 		update_ipv6_checksum(skb, l4_proto, addr, new_addr);
458 
459 	skb_clear_hash(skb);
460 	memcpy(addr, new_addr, sizeof(__be32[4]));
461 }
462 
set_ipv6_fl(struct ipv6hdr * nh,u32 fl,u32 mask)463 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
464 {
465 	/* Bits 21-24 are always unmasked, so this retains their values. */
466 	OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
467 	OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
468 	OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
469 }
470 
set_ip_ttl(struct sk_buff * skb,struct iphdr * nh,u8 new_ttl,u8 mask)471 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
472 		       u8 mask)
473 {
474 	new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
475 
476 	csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
477 	nh->ttl = new_ttl;
478 }
479 
set_ipv4(struct sk_buff * skb,struct sw_flow_key * flow_key,const struct ovs_key_ipv4 * key,const struct ovs_key_ipv4 * mask)480 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
481 		    const struct ovs_key_ipv4 *key,
482 		    const struct ovs_key_ipv4 *mask)
483 {
484 	struct iphdr *nh;
485 	__be32 new_addr;
486 	int err;
487 
488 	err = skb_ensure_writable(skb, skb_network_offset(skb) +
489 				  sizeof(struct iphdr));
490 	if (unlikely(err))
491 		return err;
492 
493 	nh = ip_hdr(skb);
494 
495 	/* Setting an IP addresses is typically only a side effect of
496 	 * matching on them in the current userspace implementation, so it
497 	 * makes sense to check if the value actually changed.
498 	 */
499 	if (mask->ipv4_src) {
500 		new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
501 
502 		if (unlikely(new_addr != nh->saddr)) {
503 			set_ip_addr(skb, nh, &nh->saddr, new_addr);
504 			flow_key->ipv4.addr.src = new_addr;
505 		}
506 	}
507 	if (mask->ipv4_dst) {
508 		new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
509 
510 		if (unlikely(new_addr != nh->daddr)) {
511 			set_ip_addr(skb, nh, &nh->daddr, new_addr);
512 			flow_key->ipv4.addr.dst = new_addr;
513 		}
514 	}
515 	if (mask->ipv4_tos) {
516 		ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
517 		flow_key->ip.tos = nh->tos;
518 	}
519 	if (mask->ipv4_ttl) {
520 		set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
521 		flow_key->ip.ttl = nh->ttl;
522 	}
523 
524 	return 0;
525 }
526 
is_ipv6_mask_nonzero(const __be32 addr[4])527 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
528 {
529 	return !!(addr[0] | addr[1] | addr[2] | addr[3]);
530 }
531 
set_ipv6(struct sk_buff * skb,struct sw_flow_key * flow_key,const struct ovs_key_ipv6 * key,const struct ovs_key_ipv6 * mask)532 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
533 		    const struct ovs_key_ipv6 *key,
534 		    const struct ovs_key_ipv6 *mask)
535 {
536 	struct ipv6hdr *nh;
537 	int err;
538 
539 	err = skb_ensure_writable(skb, skb_network_offset(skb) +
540 				  sizeof(struct ipv6hdr));
541 	if (unlikely(err))
542 		return err;
543 
544 	nh = ipv6_hdr(skb);
545 
546 	/* Setting an IP addresses is typically only a side effect of
547 	 * matching on them in the current userspace implementation, so it
548 	 * makes sense to check if the value actually changed.
549 	 */
550 	if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
551 		__be32 *saddr = (__be32 *)&nh->saddr;
552 		__be32 masked[4];
553 
554 		mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
555 
556 		if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
557 			set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
558 				      true);
559 			memcpy(&flow_key->ipv6.addr.src, masked,
560 			       sizeof(flow_key->ipv6.addr.src));
561 		}
562 	}
563 	if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
564 		unsigned int offset = 0;
565 		int flags = IP6_FH_F_SKIP_RH;
566 		bool recalc_csum = true;
567 		__be32 *daddr = (__be32 *)&nh->daddr;
568 		__be32 masked[4];
569 
570 		mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
571 
572 		if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
573 			if (ipv6_ext_hdr(nh->nexthdr))
574 				recalc_csum = (ipv6_find_hdr(skb, &offset,
575 							     NEXTHDR_ROUTING,
576 							     NULL, &flags)
577 					       != NEXTHDR_ROUTING);
578 
579 			set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
580 				      recalc_csum);
581 			memcpy(&flow_key->ipv6.addr.dst, masked,
582 			       sizeof(flow_key->ipv6.addr.dst));
583 		}
584 	}
585 	if (mask->ipv6_tclass) {
586 		ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
587 		flow_key->ip.tos = ipv6_get_dsfield(nh);
588 	}
589 	if (mask->ipv6_label) {
590 		set_ipv6_fl(nh, ntohl(key->ipv6_label),
591 			    ntohl(mask->ipv6_label));
592 		flow_key->ipv6.label =
593 		    *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
594 	}
595 	if (mask->ipv6_hlimit) {
596 		OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
597 			       mask->ipv6_hlimit);
598 		flow_key->ip.ttl = nh->hop_limit;
599 	}
600 	return 0;
601 }
602 
603 /* Must follow skb_ensure_writable() since that can move the skb data. */
set_tp_port(struct sk_buff * skb,__be16 * port,__be16 new_port,__sum16 * check)604 static void set_tp_port(struct sk_buff *skb, __be16 *port,
605 			__be16 new_port, __sum16 *check)
606 {
607 	inet_proto_csum_replace2(check, skb, *port, new_port, false);
608 	*port = new_port;
609 }
610 
set_udp(struct sk_buff * skb,struct sw_flow_key * flow_key,const struct ovs_key_udp * key,const struct ovs_key_udp * mask)611 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
612 		   const struct ovs_key_udp *key,
613 		   const struct ovs_key_udp *mask)
614 {
615 	struct udphdr *uh;
616 	__be16 src, dst;
617 	int err;
618 
619 	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
620 				  sizeof(struct udphdr));
621 	if (unlikely(err))
622 		return err;
623 
624 	uh = udp_hdr(skb);
625 	/* Either of the masks is non-zero, so do not bother checking them. */
626 	src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
627 	dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
628 
629 	if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
630 		if (likely(src != uh->source)) {
631 			set_tp_port(skb, &uh->source, src, &uh->check);
632 			flow_key->tp.src = src;
633 		}
634 		if (likely(dst != uh->dest)) {
635 			set_tp_port(skb, &uh->dest, dst, &uh->check);
636 			flow_key->tp.dst = dst;
637 		}
638 
639 		if (unlikely(!uh->check))
640 			uh->check = CSUM_MANGLED_0;
641 	} else {
642 		uh->source = src;
643 		uh->dest = dst;
644 		flow_key->tp.src = src;
645 		flow_key->tp.dst = dst;
646 	}
647 
648 	skb_clear_hash(skb);
649 
650 	return 0;
651 }
652 
set_tcp(struct sk_buff * skb,struct sw_flow_key * flow_key,const struct ovs_key_tcp * key,const struct ovs_key_tcp * mask)653 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
654 		   const struct ovs_key_tcp *key,
655 		   const struct ovs_key_tcp *mask)
656 {
657 	struct tcphdr *th;
658 	__be16 src, dst;
659 	int err;
660 
661 	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
662 				  sizeof(struct tcphdr));
663 	if (unlikely(err))
664 		return err;
665 
666 	th = tcp_hdr(skb);
667 	src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
668 	if (likely(src != th->source)) {
669 		set_tp_port(skb, &th->source, src, &th->check);
670 		flow_key->tp.src = src;
671 	}
672 	dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
673 	if (likely(dst != th->dest)) {
674 		set_tp_port(skb, &th->dest, dst, &th->check);
675 		flow_key->tp.dst = dst;
676 	}
677 	skb_clear_hash(skb);
678 
679 	return 0;
680 }
681 
set_sctp(struct sk_buff * skb,struct sw_flow_key * flow_key,const struct ovs_key_sctp * key,const struct ovs_key_sctp * mask)682 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
683 		    const struct ovs_key_sctp *key,
684 		    const struct ovs_key_sctp *mask)
685 {
686 	unsigned int sctphoff = skb_transport_offset(skb);
687 	struct sctphdr *sh;
688 	__le32 old_correct_csum, new_csum, old_csum;
689 	int err;
690 
691 	err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
692 	if (unlikely(err))
693 		return err;
694 
695 	sh = sctp_hdr(skb);
696 	old_csum = sh->checksum;
697 	old_correct_csum = sctp_compute_cksum(skb, sctphoff);
698 
699 	sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
700 	sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
701 
702 	new_csum = sctp_compute_cksum(skb, sctphoff);
703 
704 	/* Carry any checksum errors through. */
705 	sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
706 
707 	skb_clear_hash(skb);
708 	flow_key->tp.src = sh->source;
709 	flow_key->tp.dst = sh->dest;
710 
711 	return 0;
712 }
713 
ovs_vport_output(struct net * net,struct sock * sk,struct sk_buff * skb)714 static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb)
715 {
716 	struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
717 	struct vport *vport = data->vport;
718 
719 	if (skb_cow_head(skb, data->l2_len) < 0) {
720 		kfree_skb(skb);
721 		return -ENOMEM;
722 	}
723 
724 	__skb_dst_copy(skb, data->dst);
725 	*OVS_CB(skb) = data->cb;
726 	skb->inner_protocol = data->inner_protocol;
727 	skb->vlan_tci = data->vlan_tci;
728 	skb->vlan_proto = data->vlan_proto;
729 
730 	/* Reconstruct the MAC header.  */
731 	skb_push(skb, data->l2_len);
732 	memcpy(skb->data, &data->l2_data, data->l2_len);
733 	skb_postpush_rcsum(skb, skb->data, data->l2_len);
734 	skb_reset_mac_header(skb);
735 
736 	if (eth_p_mpls(skb->protocol)) {
737 		skb->inner_network_header = skb->network_header;
738 		skb_set_network_header(skb, data->network_offset);
739 		skb_reset_mac_len(skb);
740 	}
741 
742 	ovs_vport_send(vport, skb, data->mac_proto);
743 	return 0;
744 }
745 
746 static unsigned int
ovs_dst_get_mtu(const struct dst_entry * dst)747 ovs_dst_get_mtu(const struct dst_entry *dst)
748 {
749 	return dst->dev->mtu;
750 }
751 
752 static struct dst_ops ovs_dst_ops = {
753 	.family = AF_UNSPEC,
754 	.mtu = ovs_dst_get_mtu,
755 };
756 
757 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
758  * ovs_vport_output(), which is called once per fragmented packet.
759  */
prepare_frag(struct vport * vport,struct sk_buff * skb,u16 orig_network_offset,u8 mac_proto)760 static void prepare_frag(struct vport *vport, struct sk_buff *skb,
761 			 u16 orig_network_offset, u8 mac_proto)
762 {
763 	unsigned int hlen = skb_network_offset(skb);
764 	struct ovs_frag_data *data;
765 
766 	data = this_cpu_ptr(&ovs_frag_data_storage);
767 	data->dst = skb->_skb_refdst;
768 	data->vport = vport;
769 	data->cb = *OVS_CB(skb);
770 	data->inner_protocol = skb->inner_protocol;
771 	data->network_offset = orig_network_offset;
772 	data->vlan_tci = skb->vlan_tci;
773 	data->vlan_proto = skb->vlan_proto;
774 	data->mac_proto = mac_proto;
775 	data->l2_len = hlen;
776 	memcpy(&data->l2_data, skb->data, hlen);
777 
778 	memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
779 	skb_pull(skb, hlen);
780 }
781 
ovs_fragment(struct net * net,struct vport * vport,struct sk_buff * skb,u16 mru,struct sw_flow_key * key)782 static void ovs_fragment(struct net *net, struct vport *vport,
783 			 struct sk_buff *skb, u16 mru,
784 			 struct sw_flow_key *key)
785 {
786 	u16 orig_network_offset = 0;
787 
788 	if (eth_p_mpls(skb->protocol)) {
789 		orig_network_offset = skb_network_offset(skb);
790 		skb->network_header = skb->inner_network_header;
791 	}
792 
793 	if (skb_network_offset(skb) > MAX_L2_LEN) {
794 		OVS_NLERR(1, "L2 header too long to fragment");
795 		goto err;
796 	}
797 
798 	if (key->eth.type == htons(ETH_P_IP)) {
799 		struct dst_entry ovs_dst;
800 		unsigned long orig_dst;
801 
802 		prepare_frag(vport, skb, orig_network_offset,
803 			     ovs_key_mac_proto(key));
804 		dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
805 			 DST_OBSOLETE_NONE, DST_NOCOUNT);
806 		ovs_dst.dev = vport->dev;
807 
808 		orig_dst = skb->_skb_refdst;
809 		skb_dst_set_noref(skb, &ovs_dst);
810 		IPCB(skb)->frag_max_size = mru;
811 
812 		ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
813 		refdst_drop(orig_dst);
814 	} else if (key->eth.type == htons(ETH_P_IPV6)) {
815 		const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
816 		unsigned long orig_dst;
817 		struct rt6_info ovs_rt;
818 
819 		if (!v6ops)
820 			goto err;
821 
822 		prepare_frag(vport, skb, orig_network_offset,
823 			     ovs_key_mac_proto(key));
824 		memset(&ovs_rt, 0, sizeof(ovs_rt));
825 		dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
826 			 DST_OBSOLETE_NONE, DST_NOCOUNT);
827 		ovs_rt.dst.dev = vport->dev;
828 
829 		orig_dst = skb->_skb_refdst;
830 		skb_dst_set_noref(skb, &ovs_rt.dst);
831 		IP6CB(skb)->frag_max_size = mru;
832 
833 		v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
834 		refdst_drop(orig_dst);
835 	} else {
836 		WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
837 			  ovs_vport_name(vport), ntohs(key->eth.type), mru,
838 			  vport->dev->mtu);
839 		goto err;
840 	}
841 
842 	return;
843 err:
844 	kfree_skb(skb);
845 }
846 
do_output(struct datapath * dp,struct sk_buff * skb,int out_port,struct sw_flow_key * key)847 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
848 		      struct sw_flow_key *key)
849 {
850 	struct vport *vport = ovs_vport_rcu(dp, out_port);
851 
852 	if (likely(vport)) {
853 		u16 mru = OVS_CB(skb)->mru;
854 		u32 cutlen = OVS_CB(skb)->cutlen;
855 
856 		if (unlikely(cutlen > 0)) {
857 			if (skb->len - cutlen > ovs_mac_header_len(key))
858 				pskb_trim(skb, skb->len - cutlen);
859 			else
860 				pskb_trim(skb, ovs_mac_header_len(key));
861 		}
862 
863 		if (likely(!mru ||
864 		           (skb->len <= mru + vport->dev->hard_header_len))) {
865 			ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
866 		} else if (mru <= vport->dev->mtu) {
867 			struct net *net = read_pnet(&dp->net);
868 
869 			ovs_fragment(net, vport, skb, mru, key);
870 		} else {
871 			kfree_skb(skb);
872 		}
873 	} else {
874 		kfree_skb(skb);
875 	}
876 }
877 
output_userspace(struct datapath * dp,struct sk_buff * skb,struct sw_flow_key * key,const struct nlattr * attr,const struct nlattr * actions,int actions_len,uint32_t cutlen)878 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
879 			    struct sw_flow_key *key, const struct nlattr *attr,
880 			    const struct nlattr *actions, int actions_len,
881 			    uint32_t cutlen)
882 {
883 	struct dp_upcall_info upcall;
884 	const struct nlattr *a;
885 	int rem;
886 
887 	memset(&upcall, 0, sizeof(upcall));
888 	upcall.cmd = OVS_PACKET_CMD_ACTION;
889 	upcall.mru = OVS_CB(skb)->mru;
890 
891 	for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
892 		 a = nla_next(a, &rem)) {
893 		switch (nla_type(a)) {
894 		case OVS_USERSPACE_ATTR_USERDATA:
895 			upcall.userdata = a;
896 			break;
897 
898 		case OVS_USERSPACE_ATTR_PID:
899 			upcall.portid = nla_get_u32(a);
900 			break;
901 
902 		case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
903 			/* Get out tunnel info. */
904 			struct vport *vport;
905 
906 			vport = ovs_vport_rcu(dp, nla_get_u32(a));
907 			if (vport) {
908 				int err;
909 
910 				err = dev_fill_metadata_dst(vport->dev, skb);
911 				if (!err)
912 					upcall.egress_tun_info = skb_tunnel_info(skb);
913 			}
914 
915 			break;
916 		}
917 
918 		case OVS_USERSPACE_ATTR_ACTIONS: {
919 			/* Include actions. */
920 			upcall.actions = actions;
921 			upcall.actions_len = actions_len;
922 			break;
923 		}
924 
925 		} /* End of switch. */
926 	}
927 
928 	return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
929 }
930 
931 /* When 'last' is true, sample() should always consume the 'skb'.
932  * Otherwise, sample() should keep 'skb' intact regardless what
933  * actions are executed within sample().
934  */
sample(struct datapath * dp,struct sk_buff * skb,struct sw_flow_key * key,const struct nlattr * attr,bool last)935 static int sample(struct datapath *dp, struct sk_buff *skb,
936 		  struct sw_flow_key *key, const struct nlattr *attr,
937 		  bool last)
938 {
939 	struct nlattr *actions;
940 	struct nlattr *sample_arg;
941 	int rem = nla_len(attr);
942 	const struct sample_arg *arg;
943 	bool clone_flow_key;
944 
945 	/* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
946 	sample_arg = nla_data(attr);
947 	arg = nla_data(sample_arg);
948 	actions = nla_next(sample_arg, &rem);
949 
950 	if ((arg->probability != U32_MAX) &&
951 	    (!arg->probability || prandom_u32() > arg->probability)) {
952 		if (last)
953 			consume_skb(skb);
954 		return 0;
955 	}
956 
957 	clone_flow_key = !arg->exec;
958 	return clone_execute(dp, skb, key, 0, actions, rem, last,
959 			     clone_flow_key);
960 }
961 
execute_hash(struct sk_buff * skb,struct sw_flow_key * key,const struct nlattr * attr)962 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
963 			 const struct nlattr *attr)
964 {
965 	struct ovs_action_hash *hash_act = nla_data(attr);
966 	u32 hash = 0;
967 
968 	/* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
969 	hash = skb_get_hash(skb);
970 	hash = jhash_1word(hash, hash_act->hash_basis);
971 	if (!hash)
972 		hash = 0x1;
973 
974 	key->ovs_flow_hash = hash;
975 }
976 
execute_set_action(struct sk_buff * skb,struct sw_flow_key * flow_key,const struct nlattr * a)977 static int execute_set_action(struct sk_buff *skb,
978 			      struct sw_flow_key *flow_key,
979 			      const struct nlattr *a)
980 {
981 	/* Only tunnel set execution is supported without a mask. */
982 	if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
983 		struct ovs_tunnel_info *tun = nla_data(a);
984 
985 		skb_dst_drop(skb);
986 		dst_hold((struct dst_entry *)tun->tun_dst);
987 		skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
988 		return 0;
989 	}
990 
991 	return -EINVAL;
992 }
993 
994 /* Mask is at the midpoint of the data. */
995 #define get_mask(a, type) ((const type)nla_data(a) + 1)
996 
execute_masked_set_action(struct sk_buff * skb,struct sw_flow_key * flow_key,const struct nlattr * a)997 static int execute_masked_set_action(struct sk_buff *skb,
998 				     struct sw_flow_key *flow_key,
999 				     const struct nlattr *a)
1000 {
1001 	int err = 0;
1002 
1003 	switch (nla_type(a)) {
1004 	case OVS_KEY_ATTR_PRIORITY:
1005 		OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1006 			       *get_mask(a, u32 *));
1007 		flow_key->phy.priority = skb->priority;
1008 		break;
1009 
1010 	case OVS_KEY_ATTR_SKB_MARK:
1011 		OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1012 		flow_key->phy.skb_mark = skb->mark;
1013 		break;
1014 
1015 	case OVS_KEY_ATTR_TUNNEL_INFO:
1016 		/* Masked data not supported for tunnel. */
1017 		err = -EINVAL;
1018 		break;
1019 
1020 	case OVS_KEY_ATTR_ETHERNET:
1021 		err = set_eth_addr(skb, flow_key, nla_data(a),
1022 				   get_mask(a, struct ovs_key_ethernet *));
1023 		break;
1024 
1025 	case OVS_KEY_ATTR_IPV4:
1026 		err = set_ipv4(skb, flow_key, nla_data(a),
1027 			       get_mask(a, struct ovs_key_ipv4 *));
1028 		break;
1029 
1030 	case OVS_KEY_ATTR_IPV6:
1031 		err = set_ipv6(skb, flow_key, nla_data(a),
1032 			       get_mask(a, struct ovs_key_ipv6 *));
1033 		break;
1034 
1035 	case OVS_KEY_ATTR_TCP:
1036 		err = set_tcp(skb, flow_key, nla_data(a),
1037 			      get_mask(a, struct ovs_key_tcp *));
1038 		break;
1039 
1040 	case OVS_KEY_ATTR_UDP:
1041 		err = set_udp(skb, flow_key, nla_data(a),
1042 			      get_mask(a, struct ovs_key_udp *));
1043 		break;
1044 
1045 	case OVS_KEY_ATTR_SCTP:
1046 		err = set_sctp(skb, flow_key, nla_data(a),
1047 			       get_mask(a, struct ovs_key_sctp *));
1048 		break;
1049 
1050 	case OVS_KEY_ATTR_MPLS:
1051 		err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1052 								    __be32 *));
1053 		break;
1054 
1055 	case OVS_KEY_ATTR_CT_STATE:
1056 	case OVS_KEY_ATTR_CT_ZONE:
1057 	case OVS_KEY_ATTR_CT_MARK:
1058 	case OVS_KEY_ATTR_CT_LABELS:
1059 	case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1060 	case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1061 		err = -EINVAL;
1062 		break;
1063 	}
1064 
1065 	return err;
1066 }
1067 
execute_recirc(struct datapath * dp,struct sk_buff * skb,struct sw_flow_key * key,const struct nlattr * a,bool last)1068 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1069 			  struct sw_flow_key *key,
1070 			  const struct nlattr *a, bool last)
1071 {
1072 	u32 recirc_id;
1073 
1074 	if (!is_flow_key_valid(key)) {
1075 		int err;
1076 
1077 		err = ovs_flow_key_update(skb, key);
1078 		if (err)
1079 			return err;
1080 	}
1081 	BUG_ON(!is_flow_key_valid(key));
1082 
1083 	recirc_id = nla_get_u32(a);
1084 	return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1085 }
1086 
1087 /* Execute a list of actions against 'skb'. */
do_execute_actions(struct datapath * dp,struct sk_buff * skb,struct sw_flow_key * key,const struct nlattr * attr,int len)1088 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1089 			      struct sw_flow_key *key,
1090 			      const struct nlattr *attr, int len)
1091 {
1092 	const struct nlattr *a;
1093 	int rem;
1094 
1095 	for (a = attr, rem = len; rem > 0;
1096 	     a = nla_next(a, &rem)) {
1097 		int err = 0;
1098 
1099 		switch (nla_type(a)) {
1100 		case OVS_ACTION_ATTR_OUTPUT: {
1101 			int port = nla_get_u32(a);
1102 			struct sk_buff *clone;
1103 
1104 			/* Every output action needs a separate clone
1105 			 * of 'skb', In case the output action is the
1106 			 * last action, cloning can be avoided.
1107 			 */
1108 			if (nla_is_last(a, rem)) {
1109 				do_output(dp, skb, port, key);
1110 				/* 'skb' has been used for output.
1111 				 */
1112 				return 0;
1113 			}
1114 
1115 			clone = skb_clone(skb, GFP_ATOMIC);
1116 			if (clone)
1117 				do_output(dp, clone, port, key);
1118 			OVS_CB(skb)->cutlen = 0;
1119 			break;
1120 		}
1121 
1122 		case OVS_ACTION_ATTR_TRUNC: {
1123 			struct ovs_action_trunc *trunc = nla_data(a);
1124 
1125 			if (skb->len > trunc->max_len)
1126 				OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1127 			break;
1128 		}
1129 
1130 		case OVS_ACTION_ATTR_USERSPACE:
1131 			output_userspace(dp, skb, key, a, attr,
1132 						     len, OVS_CB(skb)->cutlen);
1133 			OVS_CB(skb)->cutlen = 0;
1134 			break;
1135 
1136 		case OVS_ACTION_ATTR_HASH:
1137 			execute_hash(skb, key, a);
1138 			break;
1139 
1140 		case OVS_ACTION_ATTR_PUSH_MPLS:
1141 			err = push_mpls(skb, key, nla_data(a));
1142 			break;
1143 
1144 		case OVS_ACTION_ATTR_POP_MPLS:
1145 			err = pop_mpls(skb, key, nla_get_be16(a));
1146 			break;
1147 
1148 		case OVS_ACTION_ATTR_PUSH_VLAN:
1149 			err = push_vlan(skb, key, nla_data(a));
1150 			break;
1151 
1152 		case OVS_ACTION_ATTR_POP_VLAN:
1153 			err = pop_vlan(skb, key);
1154 			break;
1155 
1156 		case OVS_ACTION_ATTR_RECIRC: {
1157 			bool last = nla_is_last(a, rem);
1158 
1159 			err = execute_recirc(dp, skb, key, a, last);
1160 			if (last) {
1161 				/* If this is the last action, the skb has
1162 				 * been consumed or freed.
1163 				 * Return immediately.
1164 				 */
1165 				return err;
1166 			}
1167 			break;
1168 		}
1169 
1170 		case OVS_ACTION_ATTR_SET:
1171 			err = execute_set_action(skb, key, nla_data(a));
1172 			break;
1173 
1174 		case OVS_ACTION_ATTR_SET_MASKED:
1175 		case OVS_ACTION_ATTR_SET_TO_MASKED:
1176 			err = execute_masked_set_action(skb, key, nla_data(a));
1177 			break;
1178 
1179 		case OVS_ACTION_ATTR_SAMPLE: {
1180 			bool last = nla_is_last(a, rem);
1181 
1182 			err = sample(dp, skb, key, a, last);
1183 			if (last)
1184 				return err;
1185 
1186 			break;
1187 		}
1188 
1189 		case OVS_ACTION_ATTR_CT:
1190 			if (!is_flow_key_valid(key)) {
1191 				err = ovs_flow_key_update(skb, key);
1192 				if (err)
1193 					return err;
1194 			}
1195 
1196 			err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1197 					     nla_data(a));
1198 
1199 			/* Hide stolen IP fragments from user space. */
1200 			if (err)
1201 				return err == -EINPROGRESS ? 0 : err;
1202 			break;
1203 
1204 		case OVS_ACTION_ATTR_PUSH_ETH:
1205 			err = push_eth(skb, key, nla_data(a));
1206 			break;
1207 
1208 		case OVS_ACTION_ATTR_POP_ETH:
1209 			err = pop_eth(skb, key);
1210 			break;
1211 		}
1212 
1213 		if (unlikely(err)) {
1214 			kfree_skb(skb);
1215 			return err;
1216 		}
1217 	}
1218 
1219 	consume_skb(skb);
1220 	return 0;
1221 }
1222 
1223 /* Execute the actions on the clone of the packet. The effect of the
1224  * execution does not affect the original 'skb' nor the original 'key'.
1225  *
1226  * The execution may be deferred in case the actions can not be executed
1227  * immediately.
1228  */
clone_execute(struct datapath * dp,struct sk_buff * skb,struct sw_flow_key * key,u32 recirc_id,const struct nlattr * actions,int len,bool last,bool clone_flow_key)1229 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1230 			 struct sw_flow_key *key, u32 recirc_id,
1231 			 const struct nlattr *actions, int len,
1232 			 bool last, bool clone_flow_key)
1233 {
1234 	struct deferred_action *da;
1235 	struct sw_flow_key *clone;
1236 
1237 	skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1238 	if (!skb) {
1239 		/* Out of memory, skip this action.
1240 		 */
1241 		return 0;
1242 	}
1243 
1244 	/* When clone_flow_key is false, the 'key' will not be change
1245 	 * by the actions, then the 'key' can be used directly.
1246 	 * Otherwise, try to clone key from the next recursion level of
1247 	 * 'flow_keys'. If clone is successful, execute the actions
1248 	 * without deferring.
1249 	 */
1250 	clone = clone_flow_key ? clone_key(key) : key;
1251 	if (clone) {
1252 		int err = 0;
1253 
1254 		if (actions) { /* Sample action */
1255 			if (clone_flow_key)
1256 				__this_cpu_inc(exec_actions_level);
1257 
1258 			err = do_execute_actions(dp, skb, clone,
1259 						 actions, len);
1260 
1261 			if (clone_flow_key)
1262 				__this_cpu_dec(exec_actions_level);
1263 		} else { /* Recirc action */
1264 			clone->recirc_id = recirc_id;
1265 			ovs_dp_process_packet(skb, clone);
1266 		}
1267 		return err;
1268 	}
1269 
1270 	/* Out of 'flow_keys' space. Defer actions */
1271 	da = add_deferred_actions(skb, key, actions, len);
1272 	if (da) {
1273 		if (!actions) { /* Recirc action */
1274 			key = &da->pkt_key;
1275 			key->recirc_id = recirc_id;
1276 		}
1277 	} else {
1278 		/* Out of per CPU action FIFO space. Drop the 'skb' and
1279 		 * log an error.
1280 		 */
1281 		kfree_skb(skb);
1282 
1283 		if (net_ratelimit()) {
1284 			if (actions) { /* Sample action */
1285 				pr_warn("%s: deferred action limit reached, drop sample action\n",
1286 					ovs_dp_name(dp));
1287 			} else {  /* Recirc action */
1288 				pr_warn("%s: deferred action limit reached, drop recirc action\n",
1289 					ovs_dp_name(dp));
1290 			}
1291 		}
1292 	}
1293 	return 0;
1294 }
1295 
process_deferred_actions(struct datapath * dp)1296 static void process_deferred_actions(struct datapath *dp)
1297 {
1298 	struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1299 
1300 	/* Do not touch the FIFO in case there is no deferred actions. */
1301 	if (action_fifo_is_empty(fifo))
1302 		return;
1303 
1304 	/* Finishing executing all deferred actions. */
1305 	do {
1306 		struct deferred_action *da = action_fifo_get(fifo);
1307 		struct sk_buff *skb = da->skb;
1308 		struct sw_flow_key *key = &da->pkt_key;
1309 		const struct nlattr *actions = da->actions;
1310 		int actions_len = da->actions_len;
1311 
1312 		if (actions)
1313 			do_execute_actions(dp, skb, key, actions, actions_len);
1314 		else
1315 			ovs_dp_process_packet(skb, key);
1316 	} while (!action_fifo_is_empty(fifo));
1317 
1318 	/* Reset FIFO for the next packet.  */
1319 	action_fifo_init(fifo);
1320 }
1321 
1322 /* Execute a list of actions against 'skb'. */
ovs_execute_actions(struct datapath * dp,struct sk_buff * skb,const struct sw_flow_actions * acts,struct sw_flow_key * key)1323 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1324 			const struct sw_flow_actions *acts,
1325 			struct sw_flow_key *key)
1326 {
1327 	int err, level;
1328 
1329 	level = __this_cpu_inc_return(exec_actions_level);
1330 	if (unlikely(level > OVS_RECURSION_LIMIT)) {
1331 		net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1332 				     ovs_dp_name(dp));
1333 		kfree_skb(skb);
1334 		err = -ENETDOWN;
1335 		goto out;
1336 	}
1337 
1338 	OVS_CB(skb)->acts_origlen = acts->orig_len;
1339 	err = do_execute_actions(dp, skb, key,
1340 				 acts->actions, acts->actions_len);
1341 
1342 	if (level == 1)
1343 		process_deferred_actions(dp);
1344 
1345 out:
1346 	__this_cpu_dec(exec_actions_level);
1347 	return err;
1348 }
1349 
action_fifos_init(void)1350 int action_fifos_init(void)
1351 {
1352 	action_fifos = alloc_percpu(struct action_fifo);
1353 	if (!action_fifos)
1354 		return -ENOMEM;
1355 
1356 	flow_keys = alloc_percpu(struct action_flow_keys);
1357 	if (!flow_keys) {
1358 		free_percpu(action_fifos);
1359 		return -ENOMEM;
1360 	}
1361 
1362 	return 0;
1363 }
1364 
action_fifos_exit(void)1365 void action_fifos_exit(void)
1366 {
1367 	free_percpu(action_fifos);
1368 	free_percpu(flow_keys);
1369 }
1370