1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <linux/kernel.h>
3 #include <linux/init.h>
4 #include <linux/module.h>
5 #include <linux/netfilter.h>
6 #include <linux/rhashtable.h>
7 #include <linux/netdevice.h>
9 #include <net/ip6_route.h>
10 #include <net/netfilter/nf_tables.h>
11 #include <net/netfilter/nf_flow_table.h>
12 #include <net/netfilter/nf_conntrack.h>
13 #include <net/netfilter/nf_conntrack_core.h>
14 #include <net/netfilter/nf_conntrack_l4proto.h>
15 #include <net/netfilter/nf_conntrack_tuple.h>
17 static DEFINE_MUTEX(flowtable_lock
);
18 static LIST_HEAD(flowtables
);
21 flow_offload_fill_dir(struct flow_offload
*flow
, struct nf_conn
*ct
,
22 enum flow_offload_tuple_dir dir
)
24 struct flow_offload_tuple
*ft
= &flow
->tuplehash
[dir
].tuple
;
25 struct nf_conntrack_tuple
*ctt
= &ct
->tuplehash
[dir
].tuple
;
29 switch (ctt
->src
.l3num
) {
31 ft
->src_v4
= ctt
->src
.u3
.in
;
32 ft
->dst_v4
= ctt
->dst
.u3
.in
;
35 ft
->src_v6
= ctt
->src
.u3
.in6
;
36 ft
->dst_v6
= ctt
->dst
.u3
.in6
;
40 ft
->l3proto
= ctt
->src
.l3num
;
41 ft
->l4proto
= ctt
->dst
.protonum
;
42 ft
->src_port
= ctt
->src
.u
.tcp
.port
;
43 ft
->dst_port
= ctt
->dst
.u
.tcp
.port
;
46 struct flow_offload
*flow_offload_alloc(struct nf_conn
*ct
)
48 struct flow_offload
*flow
;
50 if (unlikely(nf_ct_is_dying(ct
) ||
51 !atomic_inc_not_zero(&ct
->ct_general
.use
)))
54 flow
= kzalloc(sizeof(*flow
), GFP_ATOMIC
);
60 flow_offload_fill_dir(flow
, ct
, FLOW_OFFLOAD_DIR_ORIGINAL
);
61 flow_offload_fill_dir(flow
, ct
, FLOW_OFFLOAD_DIR_REPLY
);
63 if (ct
->status
& IPS_SRC_NAT
)
64 flow
->flags
|= FLOW_OFFLOAD_SNAT
;
65 if (ct
->status
& IPS_DST_NAT
)
66 flow
->flags
|= FLOW_OFFLOAD_DNAT
;
75 EXPORT_SYMBOL_GPL(flow_offload_alloc
);
77 static int flow_offload_fill_route(struct flow_offload
*flow
,
78 const struct nf_flow_route
*route
,
79 enum flow_offload_tuple_dir dir
)
81 struct flow_offload_tuple
*flow_tuple
= &flow
->tuplehash
[dir
].tuple
;
82 struct dst_entry
*other_dst
= route
->tuple
[!dir
].dst
;
83 struct dst_entry
*dst
= route
->tuple
[dir
].dst
;
85 if (!dst_hold_safe(route
->tuple
[dir
].dst
))
88 switch (flow_tuple
->l3proto
) {
90 flow_tuple
->mtu
= ip_dst_mtu_maybe_forward(dst
, true);
93 flow_tuple
->mtu
= ip6_dst_mtu_forward(dst
);
97 flow_tuple
->iifidx
= other_dst
->dev
->ifindex
;
98 flow_tuple
->dst_cache
= dst
;
103 int flow_offload_route_init(struct flow_offload
*flow
,
104 const struct nf_flow_route
*route
)
108 err
= flow_offload_fill_route(flow
, route
, FLOW_OFFLOAD_DIR_ORIGINAL
);
112 err
= flow_offload_fill_route(flow
, route
, FLOW_OFFLOAD_DIR_REPLY
);
114 goto err_route_reply
;
116 flow
->type
= NF_FLOW_OFFLOAD_ROUTE
;
121 dst_release(route
->tuple
[FLOW_OFFLOAD_DIR_ORIGINAL
].dst
);
125 EXPORT_SYMBOL_GPL(flow_offload_route_init
);
127 static void flow_offload_fixup_tcp(struct ip_ct_tcp
*tcp
)
129 tcp
->state
= TCP_CONNTRACK_ESTABLISHED
;
130 tcp
->seen
[0].td_maxwin
= 0;
131 tcp
->seen
[1].td_maxwin
= 0;
134 #define NF_FLOWTABLE_TCP_PICKUP_TIMEOUT (120 * HZ)
135 #define NF_FLOWTABLE_UDP_PICKUP_TIMEOUT (30 * HZ)
137 static inline __s32
nf_flow_timeout_delta(unsigned int timeout
)
139 return (__s32
)(timeout
- (u32
)jiffies
);
142 static void flow_offload_fixup_ct_timeout(struct nf_conn
*ct
)
144 const struct nf_conntrack_l4proto
*l4proto
;
145 int l4num
= nf_ct_protonum(ct
);
146 unsigned int timeout
;
148 l4proto
= nf_ct_l4proto_find(l4num
);
152 if (l4num
== IPPROTO_TCP
)
153 timeout
= NF_FLOWTABLE_TCP_PICKUP_TIMEOUT
;
154 else if (l4num
== IPPROTO_UDP
)
155 timeout
= NF_FLOWTABLE_UDP_PICKUP_TIMEOUT
;
159 if (nf_flow_timeout_delta(ct
->timeout
) > (__s32
)timeout
)
160 ct
->timeout
= nfct_time_stamp
+ timeout
;
163 static void flow_offload_fixup_ct_state(struct nf_conn
*ct
)
165 if (nf_ct_protonum(ct
) == IPPROTO_TCP
)
166 flow_offload_fixup_tcp(&ct
->proto
.tcp
);
169 static void flow_offload_fixup_ct(struct nf_conn
*ct
)
171 flow_offload_fixup_ct_state(ct
);
172 flow_offload_fixup_ct_timeout(ct
);
175 static void flow_offload_route_release(struct flow_offload
*flow
)
177 dst_release(flow
->tuplehash
[FLOW_OFFLOAD_DIR_ORIGINAL
].tuple
.dst_cache
);
178 dst_release(flow
->tuplehash
[FLOW_OFFLOAD_DIR_REPLY
].tuple
.dst_cache
);
181 void flow_offload_free(struct flow_offload
*flow
)
183 switch (flow
->type
) {
184 case NF_FLOW_OFFLOAD_ROUTE
:
185 flow_offload_route_release(flow
);
190 if (flow
->flags
& FLOW_OFFLOAD_DYING
)
191 nf_ct_delete(flow
->ct
, 0, 0);
193 kfree_rcu(flow
, rcu_head
);
195 EXPORT_SYMBOL_GPL(flow_offload_free
);
197 static u32
flow_offload_hash(const void *data
, u32 len
, u32 seed
)
199 const struct flow_offload_tuple
*tuple
= data
;
201 return jhash(tuple
, offsetof(struct flow_offload_tuple
, dir
), seed
);
204 static u32
flow_offload_hash_obj(const void *data
, u32 len
, u32 seed
)
206 const struct flow_offload_tuple_rhash
*tuplehash
= data
;
208 return jhash(&tuplehash
->tuple
, offsetof(struct flow_offload_tuple
, dir
), seed
);
211 static int flow_offload_hash_cmp(struct rhashtable_compare_arg
*arg
,
214 const struct flow_offload_tuple
*tuple
= arg
->key
;
215 const struct flow_offload_tuple_rhash
*x
= ptr
;
217 if (memcmp(&x
->tuple
, tuple
, offsetof(struct flow_offload_tuple
, dir
)))
223 static const struct rhashtable_params nf_flow_offload_rhash_params
= {
224 .head_offset
= offsetof(struct flow_offload_tuple_rhash
, node
),
225 .hashfn
= flow_offload_hash
,
226 .obj_hashfn
= flow_offload_hash_obj
,
227 .obj_cmpfn
= flow_offload_hash_cmp
,
228 .automatic_shrinking
= true,
231 int flow_offload_add(struct nf_flowtable
*flow_table
, struct flow_offload
*flow
)
235 flow
->timeout
= (u32
)jiffies
+ NF_FLOW_TIMEOUT
;
237 err
= rhashtable_insert_fast(&flow_table
->rhashtable
,
238 &flow
->tuplehash
[0].node
,
239 nf_flow_offload_rhash_params
);
243 err
= rhashtable_insert_fast(&flow_table
->rhashtable
,
244 &flow
->tuplehash
[1].node
,
245 nf_flow_offload_rhash_params
);
247 rhashtable_remove_fast(&flow_table
->rhashtable
,
248 &flow
->tuplehash
[0].node
,
249 nf_flow_offload_rhash_params
);
253 if (flow_table
->flags
& NF_FLOWTABLE_HW_OFFLOAD
)
254 nf_flow_offload_add(flow_table
, flow
);
258 EXPORT_SYMBOL_GPL(flow_offload_add
);
260 static inline bool nf_flow_has_expired(const struct flow_offload
*flow
)
262 return nf_flow_timeout_delta(flow
->timeout
) <= 0;
265 static void flow_offload_del(struct nf_flowtable
*flow_table
,
266 struct flow_offload
*flow
)
268 rhashtable_remove_fast(&flow_table
->rhashtable
,
269 &flow
->tuplehash
[FLOW_OFFLOAD_DIR_ORIGINAL
].node
,
270 nf_flow_offload_rhash_params
);
271 rhashtable_remove_fast(&flow_table
->rhashtable
,
272 &flow
->tuplehash
[FLOW_OFFLOAD_DIR_REPLY
].node
,
273 nf_flow_offload_rhash_params
);
275 clear_bit(IPS_OFFLOAD_BIT
, &flow
->ct
->status
);
277 if (nf_flow_has_expired(flow
))
278 flow_offload_fixup_ct(flow
->ct
);
279 else if (flow
->flags
& FLOW_OFFLOAD_TEARDOWN
)
280 flow_offload_fixup_ct_timeout(flow
->ct
);
282 flow_offload_free(flow
);
285 void flow_offload_teardown(struct flow_offload
*flow
)
287 flow
->flags
|= FLOW_OFFLOAD_TEARDOWN
;
289 flow_offload_fixup_ct_state(flow
->ct
);
291 EXPORT_SYMBOL_GPL(flow_offload_teardown
);
293 struct flow_offload_tuple_rhash
*
294 flow_offload_lookup(struct nf_flowtable
*flow_table
,
295 struct flow_offload_tuple
*tuple
)
297 struct flow_offload_tuple_rhash
*tuplehash
;
298 struct flow_offload
*flow
;
301 tuplehash
= rhashtable_lookup(&flow_table
->rhashtable
, tuple
,
302 nf_flow_offload_rhash_params
);
306 dir
= tuplehash
->tuple
.dir
;
307 flow
= container_of(tuplehash
, struct flow_offload
, tuplehash
[dir
]);
308 if (flow
->flags
& (FLOW_OFFLOAD_DYING
| FLOW_OFFLOAD_TEARDOWN
))
311 if (unlikely(nf_ct_is_dying(flow
->ct
)))
316 EXPORT_SYMBOL_GPL(flow_offload_lookup
);
319 nf_flow_table_iterate(struct nf_flowtable
*flow_table
,
320 void (*iter
)(struct flow_offload
*flow
, void *data
),
323 struct flow_offload_tuple_rhash
*tuplehash
;
324 struct rhashtable_iter hti
;
325 struct flow_offload
*flow
;
328 rhashtable_walk_enter(&flow_table
->rhashtable
, &hti
);
329 rhashtable_walk_start(&hti
);
331 while ((tuplehash
= rhashtable_walk_next(&hti
))) {
332 if (IS_ERR(tuplehash
)) {
333 if (PTR_ERR(tuplehash
) != -EAGAIN
) {
334 err
= PTR_ERR(tuplehash
);
339 if (tuplehash
->tuple
.dir
)
342 flow
= container_of(tuplehash
, struct flow_offload
, tuplehash
[0]);
346 rhashtable_walk_stop(&hti
);
347 rhashtable_walk_exit(&hti
);
352 static void nf_flow_offload_gc_step(struct flow_offload
*flow
, void *data
)
354 struct nf_flowtable
*flow_table
= data
;
356 if (flow
->flags
& FLOW_OFFLOAD_HW
)
357 nf_flow_offload_stats(flow_table
, flow
);
359 if (nf_flow_has_expired(flow
) || nf_ct_is_dying(flow
->ct
) ||
360 (flow
->flags
& (FLOW_OFFLOAD_DYING
| FLOW_OFFLOAD_TEARDOWN
))) {
361 if (flow
->flags
& FLOW_OFFLOAD_HW
) {
362 if (!(flow
->flags
& FLOW_OFFLOAD_HW_DYING
))
363 nf_flow_offload_del(flow_table
, flow
);
364 else if (flow
->flags
& FLOW_OFFLOAD_HW_DEAD
)
365 flow_offload_del(flow_table
, flow
);
367 flow_offload_del(flow_table
, flow
);
372 static void nf_flow_offload_work_gc(struct work_struct
*work
)
374 struct nf_flowtable
*flow_table
;
376 flow_table
= container_of(work
, struct nf_flowtable
, gc_work
.work
);
377 nf_flow_table_iterate(flow_table
, nf_flow_offload_gc_step
, flow_table
);
378 queue_delayed_work(system_power_efficient_wq
, &flow_table
->gc_work
, HZ
);
381 static int nf_flow_nat_port_tcp(struct sk_buff
*skb
, unsigned int thoff
,
382 __be16 port
, __be16 new_port
)
386 if (!pskb_may_pull(skb
, thoff
+ sizeof(*tcph
)) ||
387 skb_try_make_writable(skb
, thoff
+ sizeof(*tcph
)))
390 tcph
= (void *)(skb_network_header(skb
) + thoff
);
391 inet_proto_csum_replace2(&tcph
->check
, skb
, port
, new_port
, true);
396 static int nf_flow_nat_port_udp(struct sk_buff
*skb
, unsigned int thoff
,
397 __be16 port
, __be16 new_port
)
401 if (!pskb_may_pull(skb
, thoff
+ sizeof(*udph
)) ||
402 skb_try_make_writable(skb
, thoff
+ sizeof(*udph
)))
405 udph
= (void *)(skb_network_header(skb
) + thoff
);
406 if (udph
->check
|| skb
->ip_summed
== CHECKSUM_PARTIAL
) {
407 inet_proto_csum_replace2(&udph
->check
, skb
, port
,
410 udph
->check
= CSUM_MANGLED_0
;
416 static int nf_flow_nat_port(struct sk_buff
*skb
, unsigned int thoff
,
417 u8 protocol
, __be16 port
, __be16 new_port
)
421 if (nf_flow_nat_port_tcp(skb
, thoff
, port
, new_port
) < 0)
425 if (nf_flow_nat_port_udp(skb
, thoff
, port
, new_port
) < 0)
433 int nf_flow_snat_port(const struct flow_offload
*flow
,
434 struct sk_buff
*skb
, unsigned int thoff
,
435 u8 protocol
, enum flow_offload_tuple_dir dir
)
437 struct flow_ports
*hdr
;
438 __be16 port
, new_port
;
440 if (!pskb_may_pull(skb
, thoff
+ sizeof(*hdr
)) ||
441 skb_try_make_writable(skb
, thoff
+ sizeof(*hdr
)))
444 hdr
= (void *)(skb_network_header(skb
) + thoff
);
447 case FLOW_OFFLOAD_DIR_ORIGINAL
:
449 new_port
= flow
->tuplehash
[FLOW_OFFLOAD_DIR_REPLY
].tuple
.dst_port
;
450 hdr
->source
= new_port
;
452 case FLOW_OFFLOAD_DIR_REPLY
:
454 new_port
= flow
->tuplehash
[FLOW_OFFLOAD_DIR_ORIGINAL
].tuple
.src_port
;
455 hdr
->dest
= new_port
;
461 return nf_flow_nat_port(skb
, thoff
, protocol
, port
, new_port
);
463 EXPORT_SYMBOL_GPL(nf_flow_snat_port
);
465 int nf_flow_dnat_port(const struct flow_offload
*flow
,
466 struct sk_buff
*skb
, unsigned int thoff
,
467 u8 protocol
, enum flow_offload_tuple_dir dir
)
469 struct flow_ports
*hdr
;
470 __be16 port
, new_port
;
472 if (!pskb_may_pull(skb
, thoff
+ sizeof(*hdr
)) ||
473 skb_try_make_writable(skb
, thoff
+ sizeof(*hdr
)))
476 hdr
= (void *)(skb_network_header(skb
) + thoff
);
479 case FLOW_OFFLOAD_DIR_ORIGINAL
:
481 new_port
= flow
->tuplehash
[FLOW_OFFLOAD_DIR_REPLY
].tuple
.src_port
;
482 hdr
->dest
= new_port
;
484 case FLOW_OFFLOAD_DIR_REPLY
:
486 new_port
= flow
->tuplehash
[FLOW_OFFLOAD_DIR_ORIGINAL
].tuple
.dst_port
;
487 hdr
->source
= new_port
;
493 return nf_flow_nat_port(skb
, thoff
, protocol
, port
, new_port
);
495 EXPORT_SYMBOL_GPL(nf_flow_dnat_port
);
497 int nf_flow_table_init(struct nf_flowtable
*flowtable
)
501 INIT_DEFERRABLE_WORK(&flowtable
->gc_work
, nf_flow_offload_work_gc
);
502 flow_block_init(&flowtable
->flow_block
);
504 err
= rhashtable_init(&flowtable
->rhashtable
,
505 &nf_flow_offload_rhash_params
);
509 queue_delayed_work(system_power_efficient_wq
,
510 &flowtable
->gc_work
, HZ
);
512 mutex_lock(&flowtable_lock
);
513 list_add(&flowtable
->list
, &flowtables
);
514 mutex_unlock(&flowtable_lock
);
518 EXPORT_SYMBOL_GPL(nf_flow_table_init
);
520 static void nf_flow_table_do_cleanup(struct flow_offload
*flow
, void *data
)
522 struct net_device
*dev
= data
;
525 flow_offload_teardown(flow
);
529 if (net_eq(nf_ct_net(flow
->ct
), dev_net(dev
)) &&
530 (flow
->tuplehash
[0].tuple
.iifidx
== dev
->ifindex
||
531 flow
->tuplehash
[1].tuple
.iifidx
== dev
->ifindex
))
532 flow_offload_dead(flow
);
535 static void nf_flow_table_iterate_cleanup(struct nf_flowtable
*flowtable
,
536 struct net_device
*dev
)
538 nf_flow_table_offload_flush(flowtable
);
539 nf_flow_table_iterate(flowtable
, nf_flow_table_do_cleanup
, dev
);
540 flush_delayed_work(&flowtable
->gc_work
);
543 void nf_flow_table_cleanup(struct net_device
*dev
)
545 struct nf_flowtable
*flowtable
;
547 mutex_lock(&flowtable_lock
);
548 list_for_each_entry(flowtable
, &flowtables
, list
)
549 nf_flow_table_iterate_cleanup(flowtable
, dev
);
550 mutex_unlock(&flowtable_lock
);
552 EXPORT_SYMBOL_GPL(nf_flow_table_cleanup
);
554 void nf_flow_table_free(struct nf_flowtable
*flow_table
)
556 mutex_lock(&flowtable_lock
);
557 list_del(&flow_table
->list
);
558 mutex_unlock(&flowtable_lock
);
559 cancel_delayed_work_sync(&flow_table
->gc_work
);
560 nf_flow_table_iterate(flow_table
, nf_flow_table_do_cleanup
, NULL
);
561 nf_flow_table_iterate(flow_table
, nf_flow_offload_gc_step
, flow_table
);
562 rhashtable_destroy(&flow_table
->rhashtable
);
564 EXPORT_SYMBOL_GPL(nf_flow_table_free
);
566 static int __init
nf_flow_table_module_init(void)
568 return nf_flow_table_offload_init();
571 static void __exit
nf_flow_table_module_exit(void)
573 nf_flow_table_offload_exit();
576 module_init(nf_flow_table_module_init
);
577 module_exit(nf_flow_table_module_exit
);
579 MODULE_LICENSE("GPL");
580 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");