blob: 425e146523cc95fce55fa924e87b5cb7f62f3201 [file] [log] [blame]
Googler9398cc32022-12-02 17:21:52 +08001// SPDX-License-Identifier: GPL-2.0-only
Googleraf606d22022-10-26 21:40:12 -07002/*
Googler9398cc32022-12-02 17:21:52 +08003 * Copyright (c) 2007-2017 Nicira, Inc.
Googleraf606d22022-10-26 21:40:12 -07004 */
5
6#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7
8#include <linux/skbuff.h>
9#include <linux/in.h>
10#include <linux/ip.h>
11#include <linux/openvswitch.h>
12#include <linux/netfilter_ipv6.h>
13#include <linux/sctp.h>
14#include <linux/tcp.h>
15#include <linux/udp.h>
16#include <linux/in6.h>
17#include <linux/if_arp.h>
18#include <linux/if_vlan.h>
19
20#include <net/dst.h>
21#include <net/ip.h>
22#include <net/ipv6.h>
23#include <net/ip6_fib.h>
24#include <net/checksum.h>
25#include <net/dsfield.h>
26#include <net/mpls.h>
27#include <net/sctp/checksum.h>
28
29#include "datapath.h"
30#include "flow.h"
31#include "conntrack.h"
32#include "vport.h"
Googler9398cc32022-12-02 17:21:52 +080033#include "flow_netlink.h"
Googleraf606d22022-10-26 21:40:12 -070034
35struct deferred_action {
36 struct sk_buff *skb;
37 const struct nlattr *actions;
Googler9398cc32022-12-02 17:21:52 +080038 int actions_len;
Googleraf606d22022-10-26 21:40:12 -070039
40 /* Store pkt_key clone when creating deferred action. */
41 struct sw_flow_key pkt_key;
42};
43
44#define MAX_L2_LEN (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
45struct ovs_frag_data {
46 unsigned long dst;
47 struct vport *vport;
48 struct ovs_skb_cb cb;
49 __be16 inner_protocol;
Googler9398cc32022-12-02 17:21:52 +080050 u16 network_offset; /* valid only for MPLS */
51 u16 vlan_tci;
Googleraf606d22022-10-26 21:40:12 -070052 __be16 vlan_proto;
53 unsigned int l2_len;
Googler9398cc32022-12-02 17:21:52 +080054 u8 mac_proto;
Googleraf606d22022-10-26 21:40:12 -070055 u8 l2_data[MAX_L2_LEN];
56};
57
58static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
59
60#define DEFERRED_ACTION_FIFO_SIZE 10
61#define OVS_RECURSION_LIMIT 5
62#define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
63struct action_fifo {
64 int head;
65 int tail;
66 /* Deferred action fifo queue storage. */
67 struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
68};
69
Googler9398cc32022-12-02 17:21:52 +080070struct action_flow_keys {
Googleraf606d22022-10-26 21:40:12 -070071 struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
72};
73
74static struct action_fifo __percpu *action_fifos;
Googler9398cc32022-12-02 17:21:52 +080075static struct action_flow_keys __percpu *flow_keys;
Googleraf606d22022-10-26 21:40:12 -070076static DEFINE_PER_CPU(int, exec_actions_level);
77
Googler9398cc32022-12-02 17:21:52 +080078/* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
79 * space. Return NULL if out of key spaces.
80 */
81static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
82{
83 struct action_flow_keys *keys = this_cpu_ptr(flow_keys);
84 int level = this_cpu_read(exec_actions_level);
85 struct sw_flow_key *key = NULL;
86
87 if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
88 key = &keys->key[level - 1];
89 *key = *key_;
90 }
91
92 return key;
93}
94
Googleraf606d22022-10-26 21:40:12 -070095static void action_fifo_init(struct action_fifo *fifo)
96{
97 fifo->head = 0;
98 fifo->tail = 0;
99}
100
101static bool action_fifo_is_empty(const struct action_fifo *fifo)
102{
103 return (fifo->head == fifo->tail);
104}
105
106static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
107{
108 if (action_fifo_is_empty(fifo))
109 return NULL;
110
111 return &fifo->fifo[fifo->tail++];
112}
113
114static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
115{
116 if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
117 return NULL;
118
119 return &fifo->fifo[fifo->head++];
120}
121
122/* Return true if fifo is not full */
123static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
Googler9398cc32022-12-02 17:21:52 +0800124 const struct sw_flow_key *key,
125 const struct nlattr *actions,
126 const int actions_len)
Googleraf606d22022-10-26 21:40:12 -0700127{
128 struct action_fifo *fifo;
129 struct deferred_action *da;
130
131 fifo = this_cpu_ptr(action_fifos);
132 da = action_fifo_put(fifo);
133 if (da) {
134 da->skb = skb;
Googler9398cc32022-12-02 17:21:52 +0800135 da->actions = actions;
136 da->actions_len = actions_len;
Googleraf606d22022-10-26 21:40:12 -0700137 da->pkt_key = *key;
138 }
139
140 return da;
141}
142
143static void invalidate_flow_key(struct sw_flow_key *key)
144{
Googler9398cc32022-12-02 17:21:52 +0800145 key->mac_proto |= SW_FLOW_KEY_INVALID;
Googleraf606d22022-10-26 21:40:12 -0700146}
147
148static bool is_flow_key_valid(const struct sw_flow_key *key)
149{
Googler9398cc32022-12-02 17:21:52 +0800150 return !(key->mac_proto & SW_FLOW_KEY_INVALID);
Googleraf606d22022-10-26 21:40:12 -0700151}
152
Googler9398cc32022-12-02 17:21:52 +0800153static int clone_execute(struct datapath *dp, struct sk_buff *skb,
154 struct sw_flow_key *key,
155 u32 recirc_id,
156 const struct nlattr *actions, int len,
157 bool last, bool clone_flow_key);
Googleraf606d22022-10-26 21:40:12 -0700158
Googler9398cc32022-12-02 17:21:52 +0800159static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
160 struct sw_flow_key *key,
161 const struct nlattr *attr, int len);
Googleraf606d22022-10-26 21:40:12 -0700162
163static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
164 const struct ovs_action_push_mpls *mpls)
165{
Googler9398cc32022-12-02 17:21:52 +0800166 int err;
Googleraf606d22022-10-26 21:40:12 -0700167
Googler9398cc32022-12-02 17:21:52 +0800168 err = skb_mpls_push(skb, mpls->mpls_lse, mpls->mpls_ethertype,
169 skb->mac_len,
170 ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET);
171 if (err)
172 return err;
Googleraf606d22022-10-26 21:40:12 -0700173
174 invalidate_flow_key(key);
175 return 0;
176}
177
178static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
179 const __be16 ethertype)
180{
181 int err;
182
Googler9398cc32022-12-02 17:21:52 +0800183 err = skb_mpls_pop(skb, ethertype, skb->mac_len,
184 ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET);
185 if (err)
Googleraf606d22022-10-26 21:40:12 -0700186 return err;
187
Googleraf606d22022-10-26 21:40:12 -0700188 invalidate_flow_key(key);
189 return 0;
190}
191
192static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
193 const __be32 *mpls_lse, const __be32 *mask)
194{
195 struct mpls_shim_hdr *stack;
196 __be32 lse;
197 int err;
198
Googler9398cc32022-12-02 17:21:52 +0800199 if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
200 return -ENOMEM;
Googleraf606d22022-10-26 21:40:12 -0700201
202 stack = mpls_hdr(skb);
203 lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
Googler9398cc32022-12-02 17:21:52 +0800204 err = skb_mpls_update_lse(skb, lse);
205 if (err)
206 return err;
Googleraf606d22022-10-26 21:40:12 -0700207
Googleraf606d22022-10-26 21:40:12 -0700208 flow_key->mpls.top_lse = lse;
209 return 0;
210}
211
212static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
213{
214 int err;
215
216 err = skb_vlan_pop(skb);
217 if (skb_vlan_tag_present(skb)) {
218 invalidate_flow_key(key);
219 } else {
220 key->eth.vlan.tci = 0;
221 key->eth.vlan.tpid = 0;
222 }
223 return err;
224}
225
226static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
227 const struct ovs_action_push_vlan *vlan)
228{
229 if (skb_vlan_tag_present(skb)) {
230 invalidate_flow_key(key);
231 } else {
232 key->eth.vlan.tci = vlan->vlan_tci;
233 key->eth.vlan.tpid = vlan->vlan_tpid;
234 }
235 return skb_vlan_push(skb, vlan->vlan_tpid,
Googler9398cc32022-12-02 17:21:52 +0800236 ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
Googleraf606d22022-10-26 21:40:12 -0700237}
238
239/* 'src' is already properly masked. */
240static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
241{
242 u16 *dst = (u16 *)dst_;
243 const u16 *src = (const u16 *)src_;
244 const u16 *mask = (const u16 *)mask_;
245
246 OVS_SET_MASKED(dst[0], src[0], mask[0]);
247 OVS_SET_MASKED(dst[1], src[1], mask[1]);
248 OVS_SET_MASKED(dst[2], src[2], mask[2]);
249}
250
251static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
252 const struct ovs_key_ethernet *key,
253 const struct ovs_key_ethernet *mask)
254{
255 int err;
256
257 err = skb_ensure_writable(skb, ETH_HLEN);
258 if (unlikely(err))
259 return err;
260
261 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
262
263 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
264 mask->eth_src);
265 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
266 mask->eth_dst);
267
268 skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
269
270 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
271 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
272 return 0;
273}
274
Googler9398cc32022-12-02 17:21:52 +0800275/* pop_eth does not support VLAN packets as this action is never called
276 * for them.
277 */
278static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
279{
280 skb_pull_rcsum(skb, ETH_HLEN);
281 skb_reset_mac_header(skb);
282 skb_reset_mac_len(skb);
283
284 /* safe right before invalidate_flow_key */
285 key->mac_proto = MAC_PROTO_NONE;
286 invalidate_flow_key(key);
287 return 0;
288}
289
290static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
291 const struct ovs_action_push_eth *ethh)
292{
293 struct ethhdr *hdr;
294
295 /* Add the new Ethernet header */
296 if (skb_cow_head(skb, ETH_HLEN) < 0)
297 return -ENOMEM;
298
299 skb_push(skb, ETH_HLEN);
300 skb_reset_mac_header(skb);
301 skb_reset_mac_len(skb);
302
303 hdr = eth_hdr(skb);
304 ether_addr_copy(hdr->h_source, ethh->addresses.eth_src);
305 ether_addr_copy(hdr->h_dest, ethh->addresses.eth_dst);
306 hdr->h_proto = skb->protocol;
307
308 skb_postpush_rcsum(skb, hdr, ETH_HLEN);
309
310 /* safe right before invalidate_flow_key */
311 key->mac_proto = MAC_PROTO_ETHERNET;
312 invalidate_flow_key(key);
313 return 0;
314}
315
316static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
317 const struct nshhdr *nh)
318{
319 int err;
320
321 err = nsh_push(skb, nh);
322 if (err)
323 return err;
324
325 /* safe right before invalidate_flow_key */
326 key->mac_proto = MAC_PROTO_NONE;
327 invalidate_flow_key(key);
328 return 0;
329}
330
331static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
332{
333 int err;
334
335 err = nsh_pop(skb);
336 if (err)
337 return err;
338
339 /* safe right before invalidate_flow_key */
340 if (skb->protocol == htons(ETH_P_TEB))
341 key->mac_proto = MAC_PROTO_ETHERNET;
342 else
343 key->mac_proto = MAC_PROTO_NONE;
344 invalidate_flow_key(key);
345 return 0;
346}
347
Googleraf606d22022-10-26 21:40:12 -0700348static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
349 __be32 addr, __be32 new_addr)
350{
351 int transport_len = skb->len - skb_transport_offset(skb);
352
353 if (nh->frag_off & htons(IP_OFFSET))
354 return;
355
356 if (nh->protocol == IPPROTO_TCP) {
357 if (likely(transport_len >= sizeof(struct tcphdr)))
358 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
359 addr, new_addr, true);
360 } else if (nh->protocol == IPPROTO_UDP) {
361 if (likely(transport_len >= sizeof(struct udphdr))) {
362 struct udphdr *uh = udp_hdr(skb);
363
364 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
365 inet_proto_csum_replace4(&uh->check, skb,
366 addr, new_addr, true);
367 if (!uh->check)
368 uh->check = CSUM_MANGLED_0;
369 }
370 }
371 }
372}
373
374static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
375 __be32 *addr, __be32 new_addr)
376{
377 update_ip_l4_checksum(skb, nh, *addr, new_addr);
378 csum_replace4(&nh->check, *addr, new_addr);
379 skb_clear_hash(skb);
Googleraf606d22022-10-26 21:40:12 -0700380 *addr = new_addr;
381}
382
383static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
384 __be32 addr[4], const __be32 new_addr[4])
385{
386 int transport_len = skb->len - skb_transport_offset(skb);
387
388 if (l4_proto == NEXTHDR_TCP) {
389 if (likely(transport_len >= sizeof(struct tcphdr)))
390 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
391 addr, new_addr, true);
392 } else if (l4_proto == NEXTHDR_UDP) {
393 if (likely(transport_len >= sizeof(struct udphdr))) {
394 struct udphdr *uh = udp_hdr(skb);
395
396 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
397 inet_proto_csum_replace16(&uh->check, skb,
398 addr, new_addr, true);
399 if (!uh->check)
400 uh->check = CSUM_MANGLED_0;
401 }
402 }
403 } else if (l4_proto == NEXTHDR_ICMP) {
404 if (likely(transport_len >= sizeof(struct icmp6hdr)))
405 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
406 skb, addr, new_addr, true);
407 }
408}
409
410static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
411 const __be32 mask[4], __be32 masked[4])
412{
413 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
414 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
415 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
416 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
417}
418
419static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
420 __be32 addr[4], const __be32 new_addr[4],
421 bool recalculate_csum)
422{
423 if (recalculate_csum)
424 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
425
426 skb_clear_hash(skb);
Googleraf606d22022-10-26 21:40:12 -0700427 memcpy(addr, new_addr, sizeof(__be32[4]));
428}
429
Googlerb48fa912023-03-17 12:40:29 +0530430static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
Googleraf606d22022-10-26 21:40:12 -0700431{
Googleraf606d22022-10-26 21:40:12 -0700432 /* Bits 21-24 are always unmasked, so this retains their values. */
Googlerb48fa912023-03-17 12:40:29 +0530433 OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
434 OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
435 OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
Googleraf606d22022-10-26 21:40:12 -0700436}
437
438static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
439 u8 mask)
440{
441 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
442
443 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
444 nh->ttl = new_ttl;
445}
446
447static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
448 const struct ovs_key_ipv4 *key,
449 const struct ovs_key_ipv4 *mask)
450{
451 struct iphdr *nh;
452 __be32 new_addr;
453 int err;
454
455 err = skb_ensure_writable(skb, skb_network_offset(skb) +
456 sizeof(struct iphdr));
457 if (unlikely(err))
458 return err;
459
460 nh = ip_hdr(skb);
461
462 /* Setting an IP addresses is typically only a side effect of
463 * matching on them in the current userspace implementation, so it
464 * makes sense to check if the value actually changed.
465 */
466 if (mask->ipv4_src) {
467 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
468
469 if (unlikely(new_addr != nh->saddr)) {
470 set_ip_addr(skb, nh, &nh->saddr, new_addr);
471 flow_key->ipv4.addr.src = new_addr;
472 }
473 }
474 if (mask->ipv4_dst) {
475 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
476
477 if (unlikely(new_addr != nh->daddr)) {
478 set_ip_addr(skb, nh, &nh->daddr, new_addr);
479 flow_key->ipv4.addr.dst = new_addr;
480 }
481 }
482 if (mask->ipv4_tos) {
483 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
484 flow_key->ip.tos = nh->tos;
485 }
486 if (mask->ipv4_ttl) {
487 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
488 flow_key->ip.ttl = nh->ttl;
489 }
490
491 return 0;
492}
493
494static bool is_ipv6_mask_nonzero(const __be32 addr[4])
495{
496 return !!(addr[0] | addr[1] | addr[2] | addr[3]);
497}
498
499static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
500 const struct ovs_key_ipv6 *key,
501 const struct ovs_key_ipv6 *mask)
502{
503 struct ipv6hdr *nh;
504 int err;
505
506 err = skb_ensure_writable(skb, skb_network_offset(skb) +
507 sizeof(struct ipv6hdr));
508 if (unlikely(err))
509 return err;
510
511 nh = ipv6_hdr(skb);
512
513 /* Setting an IP addresses is typically only a side effect of
514 * matching on them in the current userspace implementation, so it
515 * makes sense to check if the value actually changed.
516 */
517 if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
518 __be32 *saddr = (__be32 *)&nh->saddr;
519 __be32 masked[4];
520
521 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
522
523 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
524 set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
525 true);
526 memcpy(&flow_key->ipv6.addr.src, masked,
527 sizeof(flow_key->ipv6.addr.src));
528 }
529 }
530 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
531 unsigned int offset = 0;
532 int flags = IP6_FH_F_SKIP_RH;
533 bool recalc_csum = true;
534 __be32 *daddr = (__be32 *)&nh->daddr;
535 __be32 masked[4];
536
537 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
538
539 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
540 if (ipv6_ext_hdr(nh->nexthdr))
541 recalc_csum = (ipv6_find_hdr(skb, &offset,
542 NEXTHDR_ROUTING,
543 NULL, &flags)
544 != NEXTHDR_ROUTING);
545
546 set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
547 recalc_csum);
548 memcpy(&flow_key->ipv6.addr.dst, masked,
549 sizeof(flow_key->ipv6.addr.dst));
550 }
551 }
552 if (mask->ipv6_tclass) {
Googlerb48fa912023-03-17 12:40:29 +0530553 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
Googleraf606d22022-10-26 21:40:12 -0700554 flow_key->ip.tos = ipv6_get_dsfield(nh);
555 }
556 if (mask->ipv6_label) {
Googlerb48fa912023-03-17 12:40:29 +0530557 set_ipv6_fl(nh, ntohl(key->ipv6_label),
Googleraf606d22022-10-26 21:40:12 -0700558 ntohl(mask->ipv6_label));
559 flow_key->ipv6.label =
560 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
561 }
562 if (mask->ipv6_hlimit) {
Googlerb48fa912023-03-17 12:40:29 +0530563 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
564 mask->ipv6_hlimit);
Googleraf606d22022-10-26 21:40:12 -0700565 flow_key->ip.ttl = nh->hop_limit;
566 }
567 return 0;
568}
569
Googler9398cc32022-12-02 17:21:52 +0800570static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
571 const struct nlattr *a)
572{
573 struct nshhdr *nh;
574 size_t length;
575 int err;
576 u8 flags;
577 u8 ttl;
578 int i;
579
580 struct ovs_key_nsh key;
581 struct ovs_key_nsh mask;
582
583 err = nsh_key_from_nlattr(a, &key, &mask);
584 if (err)
585 return err;
586
587 /* Make sure the NSH base header is there */
588 if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
589 return -ENOMEM;
590
591 nh = nsh_hdr(skb);
592 length = nsh_hdr_len(nh);
593
594 /* Make sure the whole NSH header is there */
595 err = skb_ensure_writable(skb, skb_network_offset(skb) +
596 length);
597 if (unlikely(err))
598 return err;
599
600 nh = nsh_hdr(skb);
601 skb_postpull_rcsum(skb, nh, length);
602 flags = nsh_get_flags(nh);
603 flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
604 flow_key->nsh.base.flags = flags;
605 ttl = nsh_get_ttl(nh);
606 ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
607 flow_key->nsh.base.ttl = ttl;
608 nsh_set_flags_and_ttl(nh, flags, ttl);
609 nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
610 mask.base.path_hdr);
611 flow_key->nsh.base.path_hdr = nh->path_hdr;
612 switch (nh->mdtype) {
613 case NSH_M_TYPE1:
614 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
615 nh->md1.context[i] =
616 OVS_MASKED(nh->md1.context[i], key.context[i],
617 mask.context[i]);
618 }
619 memcpy(flow_key->nsh.context, nh->md1.context,
620 sizeof(nh->md1.context));
621 break;
622 case NSH_M_TYPE2:
623 memset(flow_key->nsh.context, 0,
624 sizeof(flow_key->nsh.context));
625 break;
626 default:
627 return -EINVAL;
628 }
629 skb_postpush_rcsum(skb, nh, length);
630 return 0;
631}
632
Googleraf606d22022-10-26 21:40:12 -0700633/* Must follow skb_ensure_writable() since that can move the skb data. */
634static void set_tp_port(struct sk_buff *skb, __be16 *port,
635 __be16 new_port, __sum16 *check)
636{
Googleraf606d22022-10-26 21:40:12 -0700637 inet_proto_csum_replace2(check, skb, *port, new_port, false);
638 *port = new_port;
639}
640
641static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
642 const struct ovs_key_udp *key,
643 const struct ovs_key_udp *mask)
644{
645 struct udphdr *uh;
646 __be16 src, dst;
647 int err;
648
649 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
650 sizeof(struct udphdr));
651 if (unlikely(err))
652 return err;
653
654 uh = udp_hdr(skb);
655 /* Either of the masks is non-zero, so do not bother checking them. */
656 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
657 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
658
659 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
660 if (likely(src != uh->source)) {
661 set_tp_port(skb, &uh->source, src, &uh->check);
662 flow_key->tp.src = src;
663 }
664 if (likely(dst != uh->dest)) {
665 set_tp_port(skb, &uh->dest, dst, &uh->check);
666 flow_key->tp.dst = dst;
667 }
668
669 if (unlikely(!uh->check))
670 uh->check = CSUM_MANGLED_0;
671 } else {
672 uh->source = src;
673 uh->dest = dst;
674 flow_key->tp.src = src;
675 flow_key->tp.dst = dst;
Googleraf606d22022-10-26 21:40:12 -0700676 }
677
678 skb_clear_hash(skb);
679
680 return 0;
681}
682
683static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
684 const struct ovs_key_tcp *key,
685 const struct ovs_key_tcp *mask)
686{
687 struct tcphdr *th;
688 __be16 src, dst;
689 int err;
690
691 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
692 sizeof(struct tcphdr));
693 if (unlikely(err))
694 return err;
695
696 th = tcp_hdr(skb);
697 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
698 if (likely(src != th->source)) {
699 set_tp_port(skb, &th->source, src, &th->check);
700 flow_key->tp.src = src;
701 }
702 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
703 if (likely(dst != th->dest)) {
704 set_tp_port(skb, &th->dest, dst, &th->check);
705 flow_key->tp.dst = dst;
706 }
707 skb_clear_hash(skb);
708
709 return 0;
710}
711
712static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
713 const struct ovs_key_sctp *key,
714 const struct ovs_key_sctp *mask)
715{
716 unsigned int sctphoff = skb_transport_offset(skb);
717 struct sctphdr *sh;
718 __le32 old_correct_csum, new_csum, old_csum;
719 int err;
720
721 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
722 if (unlikely(err))
723 return err;
724
725 sh = sctp_hdr(skb);
726 old_csum = sh->checksum;
727 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
728
729 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
730 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
731
732 new_csum = sctp_compute_cksum(skb, sctphoff);
733
734 /* Carry any checksum errors through. */
735 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
736
737 skb_clear_hash(skb);
Googleraf606d22022-10-26 21:40:12 -0700738 flow_key->tp.src = sh->source;
739 flow_key->tp.dst = sh->dest;
740
741 return 0;
742}
743
744static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb)
745{
746 struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
747 struct vport *vport = data->vport;
748
749 if (skb_cow_head(skb, data->l2_len) < 0) {
750 kfree_skb(skb);
751 return -ENOMEM;
752 }
753
754 __skb_dst_copy(skb, data->dst);
755 *OVS_CB(skb) = data->cb;
756 skb->inner_protocol = data->inner_protocol;
Googler9398cc32022-12-02 17:21:52 +0800757 if (data->vlan_tci & VLAN_CFI_MASK)
758 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
759 else
760 __vlan_hwaccel_clear_tag(skb);
Googleraf606d22022-10-26 21:40:12 -0700761
762 /* Reconstruct the MAC header. */
763 skb_push(skb, data->l2_len);
764 memcpy(skb->data, &data->l2_data, data->l2_len);
765 skb_postpush_rcsum(skb, skb->data, data->l2_len);
766 skb_reset_mac_header(skb);
767
Googler9398cc32022-12-02 17:21:52 +0800768 if (eth_p_mpls(skb->protocol)) {
769 skb->inner_network_header = skb->network_header;
770 skb_set_network_header(skb, data->network_offset);
771 skb_reset_mac_len(skb);
772 }
773
774 ovs_vport_send(vport, skb, data->mac_proto);
Googleraf606d22022-10-26 21:40:12 -0700775 return 0;
776}
777
778static unsigned int
779ovs_dst_get_mtu(const struct dst_entry *dst)
780{
781 return dst->dev->mtu;
782}
783
784static struct dst_ops ovs_dst_ops = {
785 .family = AF_UNSPEC,
786 .mtu = ovs_dst_get_mtu,
787};
788
789/* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
790 * ovs_vport_output(), which is called once per fragmented packet.
791 */
Googler9398cc32022-12-02 17:21:52 +0800792static void prepare_frag(struct vport *vport, struct sk_buff *skb,
793 u16 orig_network_offset, u8 mac_proto)
Googleraf606d22022-10-26 21:40:12 -0700794{
795 unsigned int hlen = skb_network_offset(skb);
796 struct ovs_frag_data *data;
797
798 data = this_cpu_ptr(&ovs_frag_data_storage);
799 data->dst = skb->_skb_refdst;
800 data->vport = vport;
801 data->cb = *OVS_CB(skb);
802 data->inner_protocol = skb->inner_protocol;
Googler9398cc32022-12-02 17:21:52 +0800803 data->network_offset = orig_network_offset;
804 if (skb_vlan_tag_present(skb))
805 data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
806 else
807 data->vlan_tci = 0;
Googleraf606d22022-10-26 21:40:12 -0700808 data->vlan_proto = skb->vlan_proto;
Googler9398cc32022-12-02 17:21:52 +0800809 data->mac_proto = mac_proto;
Googleraf606d22022-10-26 21:40:12 -0700810 data->l2_len = hlen;
811 memcpy(&data->l2_data, skb->data, hlen);
812
813 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
814 skb_pull(skb, hlen);
815}
816
817static void ovs_fragment(struct net *net, struct vport *vport,
Googler9398cc32022-12-02 17:21:52 +0800818 struct sk_buff *skb, u16 mru,
819 struct sw_flow_key *key)
Googleraf606d22022-10-26 21:40:12 -0700820{
Googler9398cc32022-12-02 17:21:52 +0800821 u16 orig_network_offset = 0;
822
823 if (eth_p_mpls(skb->protocol)) {
824 orig_network_offset = skb_network_offset(skb);
825 skb->network_header = skb->inner_network_header;
826 }
827
Googleraf606d22022-10-26 21:40:12 -0700828 if (skb_network_offset(skb) > MAX_L2_LEN) {
829 OVS_NLERR(1, "L2 header too long to fragment");
830 goto err;
831 }
832
Googler9398cc32022-12-02 17:21:52 +0800833 if (key->eth.type == htons(ETH_P_IP)) {
Googlerb48fa912023-03-17 12:40:29 +0530834 struct dst_entry ovs_dst;
Googleraf606d22022-10-26 21:40:12 -0700835 unsigned long orig_dst;
836
Googler9398cc32022-12-02 17:21:52 +0800837 prepare_frag(vport, skb, orig_network_offset,
838 ovs_key_mac_proto(key));
Googlerb48fa912023-03-17 12:40:29 +0530839 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
Googleraf606d22022-10-26 21:40:12 -0700840 DST_OBSOLETE_NONE, DST_NOCOUNT);
Googlerb48fa912023-03-17 12:40:29 +0530841 ovs_dst.dev = vport->dev;
Googleraf606d22022-10-26 21:40:12 -0700842
843 orig_dst = skb->_skb_refdst;
Googlerb48fa912023-03-17 12:40:29 +0530844 skb_dst_set_noref(skb, &ovs_dst);
Googleraf606d22022-10-26 21:40:12 -0700845 IPCB(skb)->frag_max_size = mru;
846
847 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
848 refdst_drop(orig_dst);
Googler9398cc32022-12-02 17:21:52 +0800849 } else if (key->eth.type == htons(ETH_P_IPV6)) {
Googleraf606d22022-10-26 21:40:12 -0700850 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
851 unsigned long orig_dst;
852 struct rt6_info ovs_rt;
853
Googler9398cc32022-12-02 17:21:52 +0800854 if (!v6ops)
Googleraf606d22022-10-26 21:40:12 -0700855 goto err;
856
Googler9398cc32022-12-02 17:21:52 +0800857 prepare_frag(vport, skb, orig_network_offset,
858 ovs_key_mac_proto(key));
Googleraf606d22022-10-26 21:40:12 -0700859 memset(&ovs_rt, 0, sizeof(ovs_rt));
860 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
861 DST_OBSOLETE_NONE, DST_NOCOUNT);
862 ovs_rt.dst.dev = vport->dev;
863
864 orig_dst = skb->_skb_refdst;
865 skb_dst_set_noref(skb, &ovs_rt.dst);
866 IP6CB(skb)->frag_max_size = mru;
867
868 v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
869 refdst_drop(orig_dst);
870 } else {
871 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
Googler9398cc32022-12-02 17:21:52 +0800872 ovs_vport_name(vport), ntohs(key->eth.type), mru,
Googleraf606d22022-10-26 21:40:12 -0700873 vport->dev->mtu);
874 goto err;
875 }
876
877 return;
878err:
879 kfree_skb(skb);
880}
881
882static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
883 struct sw_flow_key *key)
884{
885 struct vport *vport = ovs_vport_rcu(dp, out_port);
886
887 if (likely(vport)) {
888 u16 mru = OVS_CB(skb)->mru;
889 u32 cutlen = OVS_CB(skb)->cutlen;
890
891 if (unlikely(cutlen > 0)) {
Googler9398cc32022-12-02 17:21:52 +0800892 if (skb->len - cutlen > ovs_mac_header_len(key))
Googleraf606d22022-10-26 21:40:12 -0700893 pskb_trim(skb, skb->len - cutlen);
894 else
Googler9398cc32022-12-02 17:21:52 +0800895 pskb_trim(skb, ovs_mac_header_len(key));
Googleraf606d22022-10-26 21:40:12 -0700896 }
897
Googler9398cc32022-12-02 17:21:52 +0800898 if (likely(!mru ||
899 (skb->len <= mru + vport->dev->hard_header_len))) {
900 ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
Googleraf606d22022-10-26 21:40:12 -0700901 } else if (mru <= vport->dev->mtu) {
902 struct net *net = read_pnet(&dp->net);
903
Googler9398cc32022-12-02 17:21:52 +0800904 ovs_fragment(net, vport, skb, mru, key);
Googleraf606d22022-10-26 21:40:12 -0700905 } else {
906 kfree_skb(skb);
907 }
908 } else {
909 kfree_skb(skb);
910 }
911}
912
913static int output_userspace(struct datapath *dp, struct sk_buff *skb,
914 struct sw_flow_key *key, const struct nlattr *attr,
915 const struct nlattr *actions, int actions_len,
916 uint32_t cutlen)
917{
918 struct dp_upcall_info upcall;
919 const struct nlattr *a;
920 int rem;
921
922 memset(&upcall, 0, sizeof(upcall));
923 upcall.cmd = OVS_PACKET_CMD_ACTION;
924 upcall.mru = OVS_CB(skb)->mru;
925
926 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
927 a = nla_next(a, &rem)) {
928 switch (nla_type(a)) {
929 case OVS_USERSPACE_ATTR_USERDATA:
930 upcall.userdata = a;
931 break;
932
933 case OVS_USERSPACE_ATTR_PID:
934 upcall.portid = nla_get_u32(a);
935 break;
936
937 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
938 /* Get out tunnel info. */
939 struct vport *vport;
940
941 vport = ovs_vport_rcu(dp, nla_get_u32(a));
942 if (vport) {
943 int err;
944
945 err = dev_fill_metadata_dst(vport->dev, skb);
946 if (!err)
947 upcall.egress_tun_info = skb_tunnel_info(skb);
948 }
949
950 break;
951 }
952
953 case OVS_USERSPACE_ATTR_ACTIONS: {
954 /* Include actions. */
955 upcall.actions = actions;
956 upcall.actions_len = actions_len;
957 break;
958 }
959
960 } /* End of switch. */
961 }
962
963 return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
964}
965
Googler9398cc32022-12-02 17:21:52 +0800966/* When 'last' is true, sample() should always consume the 'skb'.
967 * Otherwise, sample() should keep 'skb' intact regardless what
968 * actions are executed within sample().
969 */
Googleraf606d22022-10-26 21:40:12 -0700970static int sample(struct datapath *dp, struct sk_buff *skb,
971 struct sw_flow_key *key, const struct nlattr *attr,
Googler9398cc32022-12-02 17:21:52 +0800972 bool last)
Googleraf606d22022-10-26 21:40:12 -0700973{
Googler9398cc32022-12-02 17:21:52 +0800974 struct nlattr *actions;
975 struct nlattr *sample_arg;
976 int rem = nla_len(attr);
977 const struct sample_arg *arg;
978 bool clone_flow_key;
Googleraf606d22022-10-26 21:40:12 -0700979
Googler9398cc32022-12-02 17:21:52 +0800980 /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
981 sample_arg = nla_data(attr);
982 arg = nla_data(sample_arg);
983 actions = nla_next(sample_arg, &rem);
Googleraf606d22022-10-26 21:40:12 -0700984
Googler9398cc32022-12-02 17:21:52 +0800985 if ((arg->probability != U32_MAX) &&
986 (!arg->probability || prandom_u32() > arg->probability)) {
987 if (last)
988 consume_skb(skb);
Googler9726be62022-12-14 05:53:31 +0000989 return 0;
Googler9726be62022-12-14 05:53:31 +0000990 }
991
Googler9398cc32022-12-02 17:21:52 +0800992 clone_flow_key = !arg->exec;
993 return clone_execute(dp, skb, key, 0, actions, rem, last,
994 clone_flow_key);
995}
Googler9726be62022-12-14 05:53:31 +0000996
Googler9398cc32022-12-02 17:21:52 +0800997/* When 'last' is true, clone() should always consume the 'skb'.
998 * Otherwise, clone() should keep 'skb' intact regardless what
999 * actions are executed within clone().
1000 */
1001static int clone(struct datapath *dp, struct sk_buff *skb,
1002 struct sw_flow_key *key, const struct nlattr *attr,
1003 bool last)
1004{
1005 struct nlattr *actions;
1006 struct nlattr *clone_arg;
1007 int rem = nla_len(attr);
1008 bool dont_clone_flow_key;
Googler9726be62022-12-14 05:53:31 +00001009
Googlerb48fa912023-03-17 12:40:29 +05301010 /* The first action is always 'OVS_CLONE_ATTR_ARG'. */
Googler9398cc32022-12-02 17:21:52 +08001011 clone_arg = nla_data(attr);
1012 dont_clone_flow_key = nla_get_u32(clone_arg);
1013 actions = nla_next(clone_arg, &rem);
Googler9726be62022-12-14 05:53:31 +00001014
Googler9398cc32022-12-02 17:21:52 +08001015 return clone_execute(dp, skb, key, 0, actions, rem, last,
1016 !dont_clone_flow_key);
Googleraf606d22022-10-26 21:40:12 -07001017}
1018
1019static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1020 const struct nlattr *attr)
1021{
1022 struct ovs_action_hash *hash_act = nla_data(attr);
1023 u32 hash = 0;
1024
1025 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
1026 hash = skb_get_hash(skb);
1027 hash = jhash_1word(hash, hash_act->hash_basis);
1028 if (!hash)
1029 hash = 0x1;
1030
1031 key->ovs_flow_hash = hash;
1032}
1033
1034static int execute_set_action(struct sk_buff *skb,
1035 struct sw_flow_key *flow_key,
1036 const struct nlattr *a)
1037{
1038 /* Only tunnel set execution is supported without a mask. */
1039 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1040 struct ovs_tunnel_info *tun = nla_data(a);
1041
1042 skb_dst_drop(skb);
1043 dst_hold((struct dst_entry *)tun->tun_dst);
1044 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1045 return 0;
1046 }
1047
1048 return -EINVAL;
1049}
1050
1051/* Mask is at the midpoint of the data. */
1052#define get_mask(a, type) ((const type)nla_data(a) + 1)
1053
1054static int execute_masked_set_action(struct sk_buff *skb,
1055 struct sw_flow_key *flow_key,
1056 const struct nlattr *a)
1057{
1058 int err = 0;
1059
1060 switch (nla_type(a)) {
1061 case OVS_KEY_ATTR_PRIORITY:
1062 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1063 *get_mask(a, u32 *));
1064 flow_key->phy.priority = skb->priority;
1065 break;
1066
1067 case OVS_KEY_ATTR_SKB_MARK:
1068 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1069 flow_key->phy.skb_mark = skb->mark;
1070 break;
1071
1072 case OVS_KEY_ATTR_TUNNEL_INFO:
1073 /* Masked data not supported for tunnel. */
1074 err = -EINVAL;
1075 break;
1076
1077 case OVS_KEY_ATTR_ETHERNET:
1078 err = set_eth_addr(skb, flow_key, nla_data(a),
1079 get_mask(a, struct ovs_key_ethernet *));
1080 break;
1081
Googler9398cc32022-12-02 17:21:52 +08001082 case OVS_KEY_ATTR_NSH:
1083 err = set_nsh(skb, flow_key, a);
1084 break;
1085
Googleraf606d22022-10-26 21:40:12 -07001086 case OVS_KEY_ATTR_IPV4:
1087 err = set_ipv4(skb, flow_key, nla_data(a),
1088 get_mask(a, struct ovs_key_ipv4 *));
1089 break;
1090
1091 case OVS_KEY_ATTR_IPV6:
1092 err = set_ipv6(skb, flow_key, nla_data(a),
1093 get_mask(a, struct ovs_key_ipv6 *));
1094 break;
1095
1096 case OVS_KEY_ATTR_TCP:
1097 err = set_tcp(skb, flow_key, nla_data(a),
1098 get_mask(a, struct ovs_key_tcp *));
1099 break;
1100
1101 case OVS_KEY_ATTR_UDP:
1102 err = set_udp(skb, flow_key, nla_data(a),
1103 get_mask(a, struct ovs_key_udp *));
1104 break;
1105
1106 case OVS_KEY_ATTR_SCTP:
1107 err = set_sctp(skb, flow_key, nla_data(a),
1108 get_mask(a, struct ovs_key_sctp *));
1109 break;
1110
1111 case OVS_KEY_ATTR_MPLS:
1112 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1113 __be32 *));
1114 break;
1115
1116 case OVS_KEY_ATTR_CT_STATE:
1117 case OVS_KEY_ATTR_CT_ZONE:
1118 case OVS_KEY_ATTR_CT_MARK:
1119 case OVS_KEY_ATTR_CT_LABELS:
Googler9398cc32022-12-02 17:21:52 +08001120 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1121 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
Googleraf606d22022-10-26 21:40:12 -07001122 err = -EINVAL;
1123 break;
1124 }
1125
1126 return err;
1127}
1128
1129static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1130 struct sw_flow_key *key,
Googler9398cc32022-12-02 17:21:52 +08001131 const struct nlattr *a, bool last)
Googleraf606d22022-10-26 21:40:12 -07001132{
Googler9398cc32022-12-02 17:21:52 +08001133 u32 recirc_id;
Googleraf606d22022-10-26 21:40:12 -07001134
1135 if (!is_flow_key_valid(key)) {
1136 int err;
1137
1138 err = ovs_flow_key_update(skb, key);
1139 if (err)
1140 return err;
1141 }
1142 BUG_ON(!is_flow_key_valid(key));
1143
Googler9398cc32022-12-02 17:21:52 +08001144 recirc_id = nla_get_u32(a);
1145 return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1146}
1147
1148static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1149 struct sw_flow_key *key,
1150 const struct nlattr *attr, bool last)
1151{
1152 struct ovs_skb_cb *ovs_cb = OVS_CB(skb);
1153 const struct nlattr *actions, *cpl_arg;
1154 int len, max_len, rem = nla_len(attr);
1155 const struct check_pkt_len_arg *arg;
1156 bool clone_flow_key;
1157
1158 /* The first netlink attribute in 'attr' is always
1159 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1160 */
1161 cpl_arg = nla_data(attr);
1162 arg = nla_data(cpl_arg);
1163
1164 len = ovs_cb->mru ? ovs_cb->mru + skb->mac_len : skb->len;
1165 max_len = arg->pkt_len;
1166
1167 if ((skb_is_gso(skb) && skb_gso_validate_mac_len(skb, max_len)) ||
1168 len <= max_len) {
1169 /* Second netlink attribute in 'attr' is always
1170 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
Googler9726be62022-12-14 05:53:31 +00001171 */
Googler9398cc32022-12-02 17:21:52 +08001172 actions = nla_next(cpl_arg, &rem);
1173 clone_flow_key = !arg->exec_for_lesser_equal;
Googler9726be62022-12-14 05:53:31 +00001174 } else {
Googler9398cc32022-12-02 17:21:52 +08001175 /* Third netlink attribute in 'attr' is always
1176 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1177 */
1178 actions = nla_next(cpl_arg, &rem);
1179 actions = nla_next(actions, &rem);
1180 clone_flow_key = !arg->exec_for_greater;
Googler9726be62022-12-14 05:53:31 +00001181 }
1182
Googler9398cc32022-12-02 17:21:52 +08001183 return clone_execute(dp, skb, key, 0, nla_data(actions),
1184 nla_len(actions), last, clone_flow_key);
Googleraf606d22022-10-26 21:40:12 -07001185}
1186
1187/* Execute a list of actions against 'skb'. */
1188static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1189 struct sw_flow_key *key,
1190 const struct nlattr *attr, int len)
1191{
1192 const struct nlattr *a;
1193 int rem;
1194
1195 for (a = attr, rem = len; rem > 0;
1196 a = nla_next(a, &rem)) {
1197 int err = 0;
1198
Googler9726be62022-12-14 05:53:31 +00001199 switch (nla_type(a)) {
Googler9398cc32022-12-02 17:21:52 +08001200 case OVS_ACTION_ATTR_OUTPUT: {
1201 int port = nla_get_u32(a);
1202 struct sk_buff *clone;
1203
1204 /* Every output action needs a separate clone
1205 * of 'skb', In case the output action is the
1206 * last action, cloning can be avoided.
1207 */
1208 if (nla_is_last(a, rem)) {
1209 do_output(dp, skb, port, key);
1210 /* 'skb' has been used for output.
1211 */
1212 return 0;
1213 }
1214
1215 clone = skb_clone(skb, GFP_ATOMIC);
1216 if (clone)
1217 do_output(dp, clone, port, key);
1218 OVS_CB(skb)->cutlen = 0;
Googler9726be62022-12-14 05:53:31 +00001219 break;
Googler9398cc32022-12-02 17:21:52 +08001220 }
Googler9726be62022-12-14 05:53:31 +00001221
Googleraf606d22022-10-26 21:40:12 -07001222 case OVS_ACTION_ATTR_TRUNC: {
1223 struct ovs_action_trunc *trunc = nla_data(a);
1224
1225 if (skb->len > trunc->max_len)
1226 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1227 break;
1228 }
1229
1230 case OVS_ACTION_ATTR_USERSPACE:
1231 output_userspace(dp, skb, key, a, attr,
1232 len, OVS_CB(skb)->cutlen);
1233 OVS_CB(skb)->cutlen = 0;
1234 break;
1235
1236 case OVS_ACTION_ATTR_HASH:
1237 execute_hash(skb, key, a);
1238 break;
1239
1240 case OVS_ACTION_ATTR_PUSH_MPLS:
1241 err = push_mpls(skb, key, nla_data(a));
1242 break;
1243
1244 case OVS_ACTION_ATTR_POP_MPLS:
1245 err = pop_mpls(skb, key, nla_get_be16(a));
1246 break;
1247
1248 case OVS_ACTION_ATTR_PUSH_VLAN:
1249 err = push_vlan(skb, key, nla_data(a));
1250 break;
1251
1252 case OVS_ACTION_ATTR_POP_VLAN:
1253 err = pop_vlan(skb, key);
1254 break;
1255
Googler9398cc32022-12-02 17:21:52 +08001256 case OVS_ACTION_ATTR_RECIRC: {
1257 bool last = nla_is_last(a, rem);
1258
1259 err = execute_recirc(dp, skb, key, a, last);
1260 if (last) {
Googleraf606d22022-10-26 21:40:12 -07001261 /* If this is the last action, the skb has
1262 * been consumed or freed.
1263 * Return immediately.
1264 */
1265 return err;
1266 }
1267 break;
Googler9398cc32022-12-02 17:21:52 +08001268 }
Googleraf606d22022-10-26 21:40:12 -07001269
1270 case OVS_ACTION_ATTR_SET:
1271 err = execute_set_action(skb, key, nla_data(a));
1272 break;
1273
1274 case OVS_ACTION_ATTR_SET_MASKED:
1275 case OVS_ACTION_ATTR_SET_TO_MASKED:
1276 err = execute_masked_set_action(skb, key, nla_data(a));
1277 break;
1278
Googler9398cc32022-12-02 17:21:52 +08001279 case OVS_ACTION_ATTR_SAMPLE: {
1280 bool last = nla_is_last(a, rem);
1281
1282 err = sample(dp, skb, key, a, last);
1283 if (last)
1284 return err;
1285
Googleraf606d22022-10-26 21:40:12 -07001286 break;
Googler9398cc32022-12-02 17:21:52 +08001287 }
Googleraf606d22022-10-26 21:40:12 -07001288
1289 case OVS_ACTION_ATTR_CT:
1290 if (!is_flow_key_valid(key)) {
1291 err = ovs_flow_key_update(skb, key);
1292 if (err)
1293 return err;
1294 }
1295
1296 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1297 nla_data(a));
1298
1299 /* Hide stolen IP fragments from user space. */
1300 if (err)
1301 return err == -EINPROGRESS ? 0 : err;
1302 break;
Googler9398cc32022-12-02 17:21:52 +08001303
1304 case OVS_ACTION_ATTR_CT_CLEAR:
1305 err = ovs_ct_clear(skb, key);
1306 break;
1307
1308 case OVS_ACTION_ATTR_PUSH_ETH:
1309 err = push_eth(skb, key, nla_data(a));
1310 break;
1311
1312 case OVS_ACTION_ATTR_POP_ETH:
1313 err = pop_eth(skb, key);
1314 break;
1315
1316 case OVS_ACTION_ATTR_PUSH_NSH: {
1317 u8 buffer[NSH_HDR_MAX_LEN];
1318 struct nshhdr *nh = (struct nshhdr *)buffer;
1319
1320 err = nsh_hdr_from_nlattr(nla_data(a), nh,
1321 NSH_HDR_MAX_LEN);
1322 if (unlikely(err))
1323 break;
1324 err = push_nsh(skb, key, nh);
1325 break;
1326 }
1327
1328 case OVS_ACTION_ATTR_POP_NSH:
1329 err = pop_nsh(skb, key);
1330 break;
1331
1332 case OVS_ACTION_ATTR_METER:
1333 if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1334 consume_skb(skb);
1335 return 0;
1336 }
1337 break;
1338
1339 case OVS_ACTION_ATTR_CLONE: {
1340 bool last = nla_is_last(a, rem);
1341
1342 err = clone(dp, skb, key, a, last);
1343 if (last)
1344 return err;
1345
1346 break;
1347 }
1348
1349 case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1350 bool last = nla_is_last(a, rem);
1351
1352 err = execute_check_pkt_len(dp, skb, key, a, last);
1353 if (last)
1354 return err;
1355
1356 break;
1357 }
Googleraf606d22022-10-26 21:40:12 -07001358 }
1359
1360 if (unlikely(err)) {
1361 kfree_skb(skb);
1362 return err;
1363 }
1364 }
1365
Googler9398cc32022-12-02 17:21:52 +08001366 consume_skb(skb);
1367 return 0;
1368}
Googleraf606d22022-10-26 21:40:12 -07001369
Googler9398cc32022-12-02 17:21:52 +08001370/* Execute the actions on the clone of the packet. The effect of the
1371 * execution does not affect the original 'skb' nor the original 'key'.
1372 *
1373 * The execution may be deferred in case the actions can not be executed
1374 * immediately.
1375 */
1376static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1377 struct sw_flow_key *key, u32 recirc_id,
1378 const struct nlattr *actions, int len,
1379 bool last, bool clone_flow_key)
1380{
1381 struct deferred_action *da;
1382 struct sw_flow_key *clone;
1383
1384 skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1385 if (!skb) {
1386 /* Out of memory, skip this action.
1387 */
1388 return 0;
1389 }
1390
1391 /* When clone_flow_key is false, the 'key' will not be change
1392 * by the actions, then the 'key' can be used directly.
1393 * Otherwise, try to clone key from the next recursion level of
1394 * 'flow_keys'. If clone is successful, execute the actions
1395 * without deferring.
1396 */
1397 clone = clone_flow_key ? clone_key(key) : key;
1398 if (clone) {
1399 int err = 0;
1400
1401 if (actions) { /* Sample action */
1402 if (clone_flow_key)
1403 __this_cpu_inc(exec_actions_level);
1404
1405 err = do_execute_actions(dp, skb, clone,
1406 actions, len);
1407
1408 if (clone_flow_key)
1409 __this_cpu_dec(exec_actions_level);
1410 } else { /* Recirc action */
1411 clone->recirc_id = recirc_id;
1412 ovs_dp_process_packet(skb, clone);
1413 }
1414 return err;
1415 }
1416
1417 /* Out of 'flow_keys' space. Defer actions */
1418 da = add_deferred_actions(skb, key, actions, len);
1419 if (da) {
1420 if (!actions) { /* Recirc action */
1421 key = &da->pkt_key;
1422 key->recirc_id = recirc_id;
1423 }
1424 } else {
1425 /* Out of per CPU action FIFO space. Drop the 'skb' and
1426 * log an error.
1427 */
1428 kfree_skb(skb);
1429
1430 if (net_ratelimit()) {
1431 if (actions) { /* Sample action */
1432 pr_warn("%s: deferred action limit reached, drop sample action\n",
1433 ovs_dp_name(dp));
1434 } else { /* Recirc action */
1435 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1436 ovs_dp_name(dp));
1437 }
1438 }
1439 }
Googleraf606d22022-10-26 21:40:12 -07001440 return 0;
1441}
1442
1443static void process_deferred_actions(struct datapath *dp)
1444{
1445 struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1446
1447 /* Do not touch the FIFO in case there is no deferred actions. */
1448 if (action_fifo_is_empty(fifo))
1449 return;
1450
1451 /* Finishing executing all deferred actions. */
1452 do {
1453 struct deferred_action *da = action_fifo_get(fifo);
1454 struct sk_buff *skb = da->skb;
1455 struct sw_flow_key *key = &da->pkt_key;
1456 const struct nlattr *actions = da->actions;
Googler9398cc32022-12-02 17:21:52 +08001457 int actions_len = da->actions_len;
Googleraf606d22022-10-26 21:40:12 -07001458
1459 if (actions)
Googler9398cc32022-12-02 17:21:52 +08001460 do_execute_actions(dp, skb, key, actions, actions_len);
Googleraf606d22022-10-26 21:40:12 -07001461 else
1462 ovs_dp_process_packet(skb, key);
1463 } while (!action_fifo_is_empty(fifo));
1464
1465 /* Reset FIFO for the next packet. */
1466 action_fifo_init(fifo);
1467}
1468
1469/* Execute a list of actions against 'skb'. */
1470int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1471 const struct sw_flow_actions *acts,
1472 struct sw_flow_key *key)
1473{
1474 int err, level;
1475
1476 level = __this_cpu_inc_return(exec_actions_level);
1477 if (unlikely(level > OVS_RECURSION_LIMIT)) {
1478 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1479 ovs_dp_name(dp));
1480 kfree_skb(skb);
1481 err = -ENETDOWN;
1482 goto out;
1483 }
1484
1485 OVS_CB(skb)->acts_origlen = acts->orig_len;
1486 err = do_execute_actions(dp, skb, key,
1487 acts->actions, acts->actions_len);
1488
1489 if (level == 1)
1490 process_deferred_actions(dp);
1491
1492out:
1493 __this_cpu_dec(exec_actions_level);
1494 return err;
1495}
1496
1497int action_fifos_init(void)
1498{
1499 action_fifos = alloc_percpu(struct action_fifo);
1500 if (!action_fifos)
1501 return -ENOMEM;
1502
Googler9398cc32022-12-02 17:21:52 +08001503 flow_keys = alloc_percpu(struct action_flow_keys);
1504 if (!flow_keys) {
Googleraf606d22022-10-26 21:40:12 -07001505 free_percpu(action_fifos);
1506 return -ENOMEM;
1507 }
1508
1509 return 0;
1510}
1511
1512void action_fifos_exit(void)
1513{
1514 free_percpu(action_fifos);
Googler9398cc32022-12-02 17:21:52 +08001515 free_percpu(flow_keys);
Googleraf606d22022-10-26 21:40:12 -07001516}