projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
net/sched: cls_flower: Allow setting encapsulation fields as used key
[linux-2.6-block.git] / net / sched / cls_flower.c
CommitLineData
77b9900e
JP		 1/*
2 * net/sched/cls_flower.c Flower classifier
3 *
4 * Copyright (c) 2015 Jiri Pirko <jiri@resnulli.us>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 */
11
12#include <linux/kernel.h>
13#include <linux/init.h>
14#include <linux/module.h>
15#include <linux/rhashtable.h>
16
17#include <linux/if_ether.h>
18#include <linux/in6.h>
19#include <linux/ip.h>
20
21#include <net/sch_generic.h>
22#include <net/pkt_cls.h>
23#include <net/ip.h>
24#include <net/flow_dissector.h>
25
bc3103f1
AV		 26#include <net/dst.h>
27#include <net/dst_metadata.h>
28
77b9900e
JP		 29struct fl_flow_key {
30 int indev_ifindex;
42aecaa9 31 struct flow_dissector_key_control control;
bc3103f1 32 struct flow_dissector_key_control enc_control;
77b9900e
JP		 33 struct flow_dissector_key_basic basic;
34 struct flow_dissector_key_eth_addrs eth;
9399ae9a 35 struct flow_dissector_key_vlan vlan;
77b9900e 36 union {
c3f83241 37 struct flow_dissector_key_ipv4_addrs ipv4;
77b9900e
JP		 38 struct flow_dissector_key_ipv6_addrs ipv6;
39 };
40 struct flow_dissector_key_ports tp;
bc3103f1
AV		 41 struct flow_dissector_key_keyid enc_key_id;
42 union {
43 struct flow_dissector_key_ipv4_addrs enc_ipv4;
44 struct flow_dissector_key_ipv6_addrs enc_ipv6;
45 };
77b9900e
JP		 46} __aligned(BITS_PER_LONG / 8); /* Ensure that we can do comparisons as longs. */
47
48struct fl_flow_mask_range {
49 unsigned short int start;
50 unsigned short int end;
51};
52
53struct fl_flow_mask {
54 struct fl_flow_key key;
55 struct fl_flow_mask_range range;
56 struct rcu_head rcu;
57};
58
59struct cls_fl_head {
60 struct rhashtable ht;
61 struct fl_flow_mask mask;
62 struct flow_dissector dissector;
63 u32 hgen;
64 bool mask_assigned;
65 struct list_head filters;
66 struct rhashtable_params ht_params;
67 struct rcu_head rcu;
68};
69
70struct cls_fl_filter {
71 struct rhash_head ht_node;
72 struct fl_flow_key mkey;
73 struct tcf_exts exts;
74 struct tcf_result res;
75 struct fl_flow_key key;
76 struct list_head list;
77 u32 handle;
e69985c6 78 u32 flags;
77b9900e
JP		 79 struct rcu_head rcu;
80};
81
82static unsigned short int fl_mask_range(const struct fl_flow_mask *mask)
83{
84 return mask->range.end - mask->range.start;
85}
86
87static void fl_mask_update_range(struct fl_flow_mask *mask)
88{
89 const u8 *bytes = (const u8 *) &mask->key;
90 size_t size = sizeof(mask->key);
91 size_t i, first = 0, last = size - 1;
92
93 for (i = 0; i < sizeof(mask->key); i++) {
94 if (bytes[i]) {
95 if (!first && i)
96 first = i;
97 last = i;
98 }
99 }
100 mask->range.start = rounddown(first, sizeof(long));
101 mask->range.end = roundup(last + 1, sizeof(long));
102}
103
104static void *fl_key_get_start(struct fl_flow_key *key,
105 const struct fl_flow_mask *mask)
106{
107 return (u8 *) key + mask->range.start;
108}
109
110static void fl_set_masked_key(struct fl_flow_key *mkey, struct fl_flow_key *key,
111 struct fl_flow_mask *mask)
112{
113 const long *lkey = fl_key_get_start(key, mask);
114 const long *lmask = fl_key_get_start(&mask->key, mask);
115 long *lmkey = fl_key_get_start(mkey, mask);
116 int i;
117
118 for (i = 0; i < fl_mask_range(mask); i += sizeof(long))
119 *lmkey++ = *lkey++ & *lmask++;
120}
121
122static void fl_clear_masked_range(struct fl_flow_key *key,
123 struct fl_flow_mask *mask)
124{
125 memset(fl_key_get_start(key, mask), 0, fl_mask_range(mask));
126}
127
128static int fl_classify(struct sk_buff *skb, const struct tcf_proto *tp,
129 struct tcf_result *res)
130{
131 struct cls_fl_head *head = rcu_dereference_bh(tp->root);
132 struct cls_fl_filter *f;
133 struct fl_flow_key skb_key;
134 struct fl_flow_key skb_mkey;
bc3103f1 135 struct ip_tunnel_info *info;
77b9900e 136
e69985c6
AV		 137 if (!atomic_read(&head->ht.nelems))
138 return -1;
139
77b9900e 140 fl_clear_masked_range(&skb_key, &head->mask);
bc3103f1
AV		 141
142 info = skb_tunnel_info(skb);
143 if (info) {
144 struct ip_tunnel_key *key = &info->key;
145
146 switch (ip_tunnel_info_af(info)) {
147 case AF_INET:
148 skb_key.enc_ipv4.src = key->u.ipv4.src;
149 skb_key.enc_ipv4.dst = key->u.ipv4.dst;
150 break;
151 case AF_INET6:
152 skb_key.enc_ipv6.src = key->u.ipv6.src;
153 skb_key.enc_ipv6.dst = key->u.ipv6.dst;
154 break;
155 }
156
157 skb_key.enc_key_id.keyid = tunnel_id_to_key32(key->tun_id);
158 }
159
77b9900e
JP		 160 skb_key.indev_ifindex = skb->skb_iif;
161 /* skb_flow_dissect() does not set n_proto in case an unknown protocol,
162 * so do it rather here.
163 */
164 skb_key.basic.n_proto = skb->protocol;
cd79a238 165 skb_flow_dissect(skb, &head->dissector, &skb_key, 0);
77b9900e
JP		 166
167 fl_set_masked_key(&skb_mkey, &skb_key, &head->mask);
168
169 f = rhashtable_lookup_fast(&head->ht,
170 fl_key_get_start(&skb_mkey, &head->mask),
171 head->ht_params);
e8eb36cd 172 if (f && !tc_skip_sw(f->flags)) {
77b9900e
JP		 173 *res = f->res;
174 return tcf_exts_exec(skb, &f->exts, res);
175 }
176 return -1;
177}
178
179static int fl_init(struct tcf_proto *tp)
180{
181 struct cls_fl_head *head;
182
183 head = kzalloc(sizeof(*head), GFP_KERNEL);
184 if (!head)
185 return -ENOBUFS;
186
187 INIT_LIST_HEAD_RCU(&head->filters);
188 rcu_assign_pointer(tp->root, head);
189
190 return 0;
191}
192
193static void fl_destroy_filter(struct rcu_head *head)
194{
195 struct cls_fl_filter *f = container_of(head, struct cls_fl_filter, rcu);
196
197 tcf_exts_destroy(&f->exts);
198 kfree(f);
199}
200
8208d21b 201static void fl_hw_destroy_filter(struct tcf_proto *tp, unsigned long cookie)
5b33f488
AV		 202{
203 struct net_device *dev = tp->q->dev_queue->dev;
204 struct tc_cls_flower_offload offload = {0};
205 struct tc_to_netdev tc;
206
92c075db 207 if (!tc_should_offload(dev, tp, 0))
5b33f488
AV		 208 return;
209
210 offload.command = TC_CLSFLOWER_DESTROY;
211 offload.cookie = cookie;
212
213 tc.type = TC_SETUP_CLSFLOWER;
214 tc.cls_flower = &offload;
215
216 dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle, tp->protocol, &tc);
217}
218
e8eb36cd
AV		 219static int fl_hw_replace_filter(struct tcf_proto *tp,
220 struct flow_dissector *dissector,
221 struct fl_flow_key *mask,
222 struct fl_flow_key *key,
223 struct tcf_exts *actions,
224 unsigned long cookie, u32 flags)
5b33f488
AV		 225{
226 struct net_device *dev = tp->q->dev_queue->dev;
227 struct tc_cls_flower_offload offload = {0};
228 struct tc_to_netdev tc;
e8eb36cd 229 int err;
5b33f488 230
92c075db 231 if (!tc_should_offload(dev, tp, flags))
e8eb36cd 232 return tc_skip_sw(flags) ? -EINVAL : 0;
5b33f488
AV		 233
234 offload.command = TC_CLSFLOWER_REPLACE;
235 offload.cookie = cookie;
236 offload.dissector = dissector;
237 offload.mask = mask;
238 offload.key = key;
239 offload.exts = actions;
240
241 tc.type = TC_SETUP_CLSFLOWER;
242 tc.cls_flower = &offload;
243
5a7a5555
JHS		 244 err = dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle, tp->protocol,
245 &tc);
e8eb36cd
AV		 246
247 if (tc_skip_sw(flags))
248 return err;
249
250 return 0;
5b33f488
AV		 251}
252
10cbc684
AV		 253static void fl_hw_update_stats(struct tcf_proto *tp, struct cls_fl_filter *f)
254{
255 struct net_device *dev = tp->q->dev_queue->dev;
256 struct tc_cls_flower_offload offload = {0};
257 struct tc_to_netdev tc;
258
92c075db 259 if (!tc_should_offload(dev, tp, 0))
10cbc684
AV		 260 return;
261
262 offload.command = TC_CLSFLOWER_STATS;
263 offload.cookie = (unsigned long)f;
264 offload.exts = &f->exts;
265
266 tc.type = TC_SETUP_CLSFLOWER;
267 tc.cls_flower = &offload;
268
269 dev->netdev_ops->ndo_setup_tc(dev, tp->q->handle, tp->protocol, &tc);
270}
271
13fa876e
RD		 272static void __fl_delete(struct tcf_proto *tp, struct cls_fl_filter *f)
273{
274 list_del_rcu(&f->list);
275 fl_hw_destroy_filter(tp, (unsigned long)f);
276 tcf_unbind_filter(tp, &f->res);
277 call_rcu(&f->rcu, fl_destroy_filter);
278}
279
77b9900e
JP		 280static bool fl_destroy(struct tcf_proto *tp, bool force)
281{
282 struct cls_fl_head *head = rtnl_dereference(tp->root);
283 struct cls_fl_filter *f, *next;
284
285 if (!force && !list_empty(&head->filters))
286 return false;
287
13fa876e
RD		 288 list_for_each_entry_safe(f, next, &head->filters, list)
289 __fl_delete(tp, f);
77b9900e
JP		 290 RCU_INIT_POINTER(tp->root, NULL);
291 if (head->mask_assigned)
292 rhashtable_destroy(&head->ht);
293 kfree_rcu(head, rcu);
294 return true;
295}
296
297static unsigned long fl_get(struct tcf_proto *tp, u32 handle)
298{
299 struct cls_fl_head *head = rtnl_dereference(tp->root);
300 struct cls_fl_filter *f;
301
302 list_for_each_entry(f, &head->filters, list)
303 if (f->handle == handle)
304 return (unsigned long) f;
305 return 0;
306}
307
308static const struct nla_policy fl_policy[TCA_FLOWER_MAX + 1] = {
309 [TCA_FLOWER_UNSPEC] = { .type = NLA_UNSPEC },
310 [TCA_FLOWER_CLASSID] = { .type = NLA_U32 },
311 [TCA_FLOWER_INDEV] = { .type = NLA_STRING,
312 .len = IFNAMSIZ },
313 [TCA_FLOWER_KEY_ETH_DST] = { .len = ETH_ALEN },
314 [TCA_FLOWER_KEY_ETH_DST_MASK] = { .len = ETH_ALEN },
315 [TCA_FLOWER_KEY_ETH_SRC] = { .len = ETH_ALEN },
316 [TCA_FLOWER_KEY_ETH_SRC_MASK] = { .len = ETH_ALEN },
317 [TCA_FLOWER_KEY_ETH_TYPE] = { .type = NLA_U16 },
318 [TCA_FLOWER_KEY_IP_PROTO] = { .type = NLA_U8 },
319 [TCA_FLOWER_KEY_IPV4_SRC] = { .type = NLA_U32 },
320 [TCA_FLOWER_KEY_IPV4_SRC_MASK] = { .type = NLA_U32 },
321 [TCA_FLOWER_KEY_IPV4_DST] = { .type = NLA_U32 },
322 [TCA_FLOWER_KEY_IPV4_DST_MASK] = { .type = NLA_U32 },
323 [TCA_FLOWER_KEY_IPV6_SRC] = { .len = sizeof(struct in6_addr) },
324 [TCA_FLOWER_KEY_IPV6_SRC_MASK] = { .len = sizeof(struct in6_addr) },
325 [TCA_FLOWER_KEY_IPV6_DST] = { .len = sizeof(struct in6_addr) },
326 [TCA_FLOWER_KEY_IPV6_DST_MASK] = { .len = sizeof(struct in6_addr) },
327 [TCA_FLOWER_KEY_TCP_SRC] = { .type = NLA_U16 },
328 [TCA_FLOWER_KEY_TCP_DST] = { .type = NLA_U16 },
b175c3a4
JHS		 329 [TCA_FLOWER_KEY_UDP_SRC] = { .type = NLA_U16 },
330 [TCA_FLOWER_KEY_UDP_DST] = { .type = NLA_U16 },
9399ae9a
HHZ		 331 [TCA_FLOWER_KEY_VLAN_ID] = { .type = NLA_U16 },
332 [TCA_FLOWER_KEY_VLAN_PRIO] = { .type = NLA_U8 },
333 [TCA_FLOWER_KEY_VLAN_ETH_TYPE] = { .type = NLA_U16 },
bc3103f1
AV		 334 [TCA_FLOWER_KEY_ENC_KEY_ID] = { .type = NLA_U32 },
335 [TCA_FLOWER_KEY_ENC_IPV4_SRC] = { .type = NLA_U32 },
336 [TCA_FLOWER_KEY_ENC_IPV4_SRC_MASK] = { .type = NLA_U32 },
337 [TCA_FLOWER_KEY_ENC_IPV4_DST] = { .type = NLA_U32 },
338 [TCA_FLOWER_KEY_ENC_IPV4_DST_MASK] = { .type = NLA_U32 },
339 [TCA_FLOWER_KEY_ENC_IPV6_SRC] = { .len = sizeof(struct in6_addr) },
340 [TCA_FLOWER_KEY_ENC_IPV6_SRC_MASK] = { .len = sizeof(struct in6_addr) },
341 [TCA_FLOWER_KEY_ENC_IPV6_DST] = { .len = sizeof(struct in6_addr) },
342 [TCA_FLOWER_KEY_ENC_IPV6_DST_MASK] = { .len = sizeof(struct in6_addr) },
aa72d708
OG		 343 [TCA_FLOWER_KEY_TCP_SRC_MASK] = { .type = NLA_U16 },
344 [TCA_FLOWER_KEY_TCP_DST_MASK] = { .type = NLA_U16 },
345 [TCA_FLOWER_KEY_UDP_SRC_MASK] = { .type = NLA_U16 },
346 [TCA_FLOWER_KEY_UDP_DST_MASK] = { .type = NLA_U16 },
5976c5f4
SH		 347 [TCA_FLOWER_KEY_SCTP_SRC_MASK] = { .type = NLA_U16 },
348 [TCA_FLOWER_KEY_SCTP_DST_MASK] = { .type = NLA_U16 },
349 [TCA_FLOWER_KEY_SCTP_SRC] = { .type = NLA_U16 },
350 [TCA_FLOWER_KEY_SCTP_DST] = { .type = NLA_U16 },
77b9900e
JP		 351};
352
353static void fl_set_key_val(struct nlattr **tb,
354 void *val, int val_type,
355 void *mask, int mask_type, int len)
356{
357 if (!tb[val_type])
358 return;
359 memcpy(val, nla_data(tb[val_type]), len);
360 if (mask_type == TCA_FLOWER_UNSPEC || !tb[mask_type])
361 memset(mask, 0xff, len);
362 else
363 memcpy(mask, nla_data(tb[mask_type]), len);
364}
365
9399ae9a
HHZ		 366static void fl_set_key_vlan(struct nlattr **tb,
367 struct flow_dissector_key_vlan *key_val,
368 struct flow_dissector_key_vlan *key_mask)
369{
370#define VLAN_PRIORITY_MASK 0x7
371
372 if (tb[TCA_FLOWER_KEY_VLAN_ID]) {
373 key_val->vlan_id =
374 nla_get_u16(tb[TCA_FLOWER_KEY_VLAN_ID]) & VLAN_VID_MASK;
375 key_mask->vlan_id = VLAN_VID_MASK;
376 }
377 if (tb[TCA_FLOWER_KEY_VLAN_PRIO]) {
378 key_val->vlan_priority =
379 nla_get_u8(tb[TCA_FLOWER_KEY_VLAN_PRIO]) &
380 VLAN_PRIORITY_MASK;
381 key_mask->vlan_priority = VLAN_PRIORITY_MASK;
382 }
383}
384
77b9900e
JP		 385static int fl_set_key(struct net *net, struct nlattr **tb,
386 struct fl_flow_key *key, struct fl_flow_key *mask)
387{
9399ae9a 388 __be16 ethertype;
dd3aa3b5 389#ifdef CONFIG_NET_CLS_IND
77b9900e 390 if (tb[TCA_FLOWER_INDEV]) {
dd3aa3b5 391 int err = tcf_change_indev(net, tb[TCA_FLOWER_INDEV]);
77b9900e
JP		 392 if (err < 0)
393 return err;
394 key->indev_ifindex = err;
395 mask->indev_ifindex = 0xffffffff;
396 }
dd3aa3b5 397#endif
77b9900e
JP		 398
399 fl_set_key_val(tb, key->eth.dst, TCA_FLOWER_KEY_ETH_DST,
400 mask->eth.dst, TCA_FLOWER_KEY_ETH_DST_MASK,
401 sizeof(key->eth.dst));
402 fl_set_key_val(tb, key->eth.src, TCA_FLOWER_KEY_ETH_SRC,
403 mask->eth.src, TCA_FLOWER_KEY_ETH_SRC_MASK,
404 sizeof(key->eth.src));
66530bdf 405
0b498a52 406 if (tb[TCA_FLOWER_KEY_ETH_TYPE]) {
9399ae9a
HHZ		 407 ethertype = nla_get_be16(tb[TCA_FLOWER_KEY_ETH_TYPE]);
408
0b498a52
AB		 409 if (ethertype == htons(ETH_P_8021Q)) {
410 fl_set_key_vlan(tb, &key->vlan, &mask->vlan);
411 fl_set_key_val(tb, &key->basic.n_proto,
412 TCA_FLOWER_KEY_VLAN_ETH_TYPE,
413 &mask->basic.n_proto, TCA_FLOWER_UNSPEC,
414 sizeof(key->basic.n_proto));
415 } else {
416 key->basic.n_proto = ethertype;
417 mask->basic.n_proto = cpu_to_be16(~0);
418 }
9399ae9a 419 }
66530bdf 420
77b9900e
JP		 421 if (key->basic.n_proto == htons(ETH_P_IP) ||
422 key->basic.n_proto == htons(ETH_P_IPV6)) {
423 fl_set_key_val(tb, &key->basic.ip_proto, TCA_FLOWER_KEY_IP_PROTO,
424 &mask->basic.ip_proto, TCA_FLOWER_UNSPEC,
425 sizeof(key->basic.ip_proto));
426 }
66530bdf
JHS		 427
428 if (tb[TCA_FLOWER_KEY_IPV4_SRC] || tb[TCA_FLOWER_KEY_IPV4_DST]) {
429 key->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
77b9900e
JP		 430 fl_set_key_val(tb, &key->ipv4.src, TCA_FLOWER_KEY_IPV4_SRC,
431 &mask->ipv4.src, TCA_FLOWER_KEY_IPV4_SRC_MASK,
432 sizeof(key->ipv4.src));
433 fl_set_key_val(tb, &key->ipv4.dst, TCA_FLOWER_KEY_IPV4_DST,
434 &mask->ipv4.dst, TCA_FLOWER_KEY_IPV4_DST_MASK,
435 sizeof(key->ipv4.dst));
66530bdf
JHS		 436 } else if (tb[TCA_FLOWER_KEY_IPV6_SRC] || tb[TCA_FLOWER_KEY_IPV6_DST]) {
437 key->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
77b9900e
JP		 438 fl_set_key_val(tb, &key->ipv6.src, TCA_FLOWER_KEY_IPV6_SRC,
439 &mask->ipv6.src, TCA_FLOWER_KEY_IPV6_SRC_MASK,
440 sizeof(key->ipv6.src));
441 fl_set_key_val(tb, &key->ipv6.dst, TCA_FLOWER_KEY_IPV6_DST,
442 &mask->ipv6.dst, TCA_FLOWER_KEY_IPV6_DST_MASK,
443 sizeof(key->ipv6.dst));
444 }
66530bdf 445
77b9900e
JP		 446 if (key->basic.ip_proto == IPPROTO_TCP) {
447 fl_set_key_val(tb, &key->tp.src, TCA_FLOWER_KEY_TCP_SRC,
aa72d708 448 &mask->tp.src, TCA_FLOWER_KEY_TCP_SRC_MASK,
77b9900e
JP		 449 sizeof(key->tp.src));
450 fl_set_key_val(tb, &key->tp.dst, TCA_FLOWER_KEY_TCP_DST,
aa72d708 451 &mask->tp.dst, TCA_FLOWER_KEY_TCP_DST_MASK,
77b9900e
JP		 452 sizeof(key->tp.dst));
453 } else if (key->basic.ip_proto == IPPROTO_UDP) {
454 fl_set_key_val(tb, &key->tp.src, TCA_FLOWER_KEY_UDP_SRC,
aa72d708 455 &mask->tp.src, TCA_FLOWER_KEY_UDP_SRC_MASK,
77b9900e
JP		 456 sizeof(key->tp.src));
457 fl_set_key_val(tb, &key->tp.dst, TCA_FLOWER_KEY_UDP_DST,
aa72d708 458 &mask->tp.dst, TCA_FLOWER_KEY_UDP_DST_MASK,
77b9900e 459 sizeof(key->tp.dst));
5976c5f4
SH		 460 } else if (key->basic.ip_proto == IPPROTO_SCTP) {
461 fl_set_key_val(tb, &key->tp.src, TCA_FLOWER_KEY_SCTP_SRC,
462 &mask->tp.src, TCA_FLOWER_KEY_SCTP_SRC_MASK,
463 sizeof(key->tp.src));
464 fl_set_key_val(tb, &key->tp.dst, TCA_FLOWER_KEY_SCTP_DST,
465 &mask->tp.dst, TCA_FLOWER_KEY_SCTP_DST_MASK,
466 sizeof(key->tp.dst));
77b9900e
JP		 467 }
468
bc3103f1
AV		 469 if (tb[TCA_FLOWER_KEY_ENC_IPV4_SRC] ||
470 tb[TCA_FLOWER_KEY_ENC_IPV4_DST]) {
471 key->enc_control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
472 fl_set_key_val(tb, &key->enc_ipv4.src,
473 TCA_FLOWER_KEY_ENC_IPV4_SRC,
474 &mask->enc_ipv4.src,
475 TCA_FLOWER_KEY_ENC_IPV4_SRC_MASK,
476 sizeof(key->enc_ipv4.src));
477 fl_set_key_val(tb, &key->enc_ipv4.dst,
478 TCA_FLOWER_KEY_ENC_IPV4_DST,
479 &mask->enc_ipv4.dst,
480 TCA_FLOWER_KEY_ENC_IPV4_DST_MASK,
481 sizeof(key->enc_ipv4.dst));
482 }
483
484 if (tb[TCA_FLOWER_KEY_ENC_IPV6_SRC] ||
485 tb[TCA_FLOWER_KEY_ENC_IPV6_DST]) {
486 key->enc_control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
487 fl_set_key_val(tb, &key->enc_ipv6.src,
488 TCA_FLOWER_KEY_ENC_IPV6_SRC,
489 &mask->enc_ipv6.src,
490 TCA_FLOWER_KEY_ENC_IPV6_SRC_MASK,
491 sizeof(key->enc_ipv6.src));
492 fl_set_key_val(tb, &key->enc_ipv6.dst,
493 TCA_FLOWER_KEY_ENC_IPV6_DST,
494 &mask->enc_ipv6.dst,
495 TCA_FLOWER_KEY_ENC_IPV6_DST_MASK,
496 sizeof(key->enc_ipv6.dst));
497 }
498
499 fl_set_key_val(tb, &key->enc_key_id.keyid, TCA_FLOWER_KEY_ENC_KEY_ID,
eb523f42 500 &mask->enc_key_id.keyid, TCA_FLOWER_UNSPEC,
bc3103f1
AV		 501 sizeof(key->enc_key_id.keyid));
502
77b9900e
JP		 503 return 0;
504}
505
506static bool fl_mask_eq(struct fl_flow_mask *mask1,
507 struct fl_flow_mask *mask2)
508{
509 const long *lmask1 = fl_key_get_start(&mask1->key, mask1);
510 const long *lmask2 = fl_key_get_start(&mask2->key, mask2);
511
512 return !memcmp(&mask1->range, &mask2->range, sizeof(mask1->range)) &&
513 !memcmp(lmask1, lmask2, fl_mask_range(mask1));
514}
515
516static const struct rhashtable_params fl_ht_params = {
517 .key_offset = offsetof(struct cls_fl_filter, mkey), /* base offset */
518 .head_offset = offsetof(struct cls_fl_filter, ht_node),
519 .automatic_shrinking = true,
520};
521
522static int fl_init_hashtable(struct cls_fl_head *head,
523 struct fl_flow_mask *mask)
524{
525 head->ht_params = fl_ht_params;
526 head->ht_params.key_len = fl_mask_range(mask);
527 head->ht_params.key_offset += mask->range.start;
528
529 return rhashtable_init(&head->ht, &head->ht_params);
530}
531
532#define FL_KEY_MEMBER_OFFSET(member) offsetof(struct fl_flow_key, member)
533#define FL_KEY_MEMBER_SIZE(member) (sizeof(((struct fl_flow_key *) 0)->member))
77b9900e 534
339ba878
HHZ		 535#define FL_KEY_IS_MASKED(mask, member) \
536 memchr_inv(((char *)mask) + FL_KEY_MEMBER_OFFSET(member), \
537 0, FL_KEY_MEMBER_SIZE(member)) \
77b9900e
JP		 538
539#define FL_KEY_SET(keys, cnt, id, member) \
540 do { \
541 keys[cnt].key_id = id; \
542 keys[cnt].offset = FL_KEY_MEMBER_OFFSET(member); \
543 cnt++; \
544 } while(0);
545
339ba878 546#define FL_KEY_SET_IF_MASKED(mask, keys, cnt, id, member) \
77b9900e 547 do { \
339ba878 548 if (FL_KEY_IS_MASKED(mask, member)) \
77b9900e
JP		 549 FL_KEY_SET(keys, cnt, id, member); \
550 } while(0);
551
552static void fl_init_dissector(struct cls_fl_head *head,
553 struct fl_flow_mask *mask)
554{
555 struct flow_dissector_key keys[FLOW_DISSECTOR_KEY_MAX];
556 size_t cnt = 0;
557
42aecaa9 558 FL_KEY_SET(keys, cnt, FLOW_DISSECTOR_KEY_CONTROL, control);
77b9900e 559 FL_KEY_SET(keys, cnt, FLOW_DISSECTOR_KEY_BASIC, basic);
339ba878
HHZ		 560 FL_KEY_SET_IF_MASKED(&mask->key, keys, cnt,
561 FLOW_DISSECTOR_KEY_ETH_ADDRS, eth);
562 FL_KEY_SET_IF_MASKED(&mask->key, keys, cnt,
563 FLOW_DISSECTOR_KEY_IPV4_ADDRS, ipv4);
564 FL_KEY_SET_IF_MASKED(&mask->key, keys, cnt,
565 FLOW_DISSECTOR_KEY_IPV6_ADDRS, ipv6);
566 FL_KEY_SET_IF_MASKED(&mask->key, keys, cnt,
567 FLOW_DISSECTOR_KEY_PORTS, tp);
9399ae9a
HHZ		 568 FL_KEY_SET_IF_MASKED(&mask->key, keys, cnt,
569 FLOW_DISSECTOR_KEY_VLAN, vlan);
519d1052
HHZ		 570 FL_KEY_SET_IF_MASKED(&mask->key, keys, cnt,
571 FLOW_DISSECTOR_KEY_ENC_KEYID, enc_key_id);
572 FL_KEY_SET_IF_MASKED(&mask->key, keys, cnt,
573 FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS, enc_ipv4);
574 FL_KEY_SET_IF_MASKED(&mask->key, keys, cnt,
575 FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS, enc_ipv6);
576 if (FL_KEY_IS_MASKED(&mask->key, enc_ipv4) ||
577 FL_KEY_IS_MASKED(&mask->key, enc_ipv6))
578 FL_KEY_SET(keys, cnt, FLOW_DISSECTOR_KEY_ENC_CONTROL,
579 enc_control);
77b9900e
JP		 580
581 skb_flow_dissector_init(&head->dissector, keys, cnt);
582}
583
584static int fl_check_assign_mask(struct cls_fl_head *head,
585 struct fl_flow_mask *mask)
586{
587 int err;
588
589 if (head->mask_assigned) {
590 if (!fl_mask_eq(&head->mask, mask))
591 return -EINVAL;
592 else
593 return 0;
594 }
595
596 /* Mask is not assigned yet. So assign it and init hashtable
597 * according to that.
598 */
599 err = fl_init_hashtable(head, mask);
600 if (err)
601 return err;
602 memcpy(&head->mask, mask, sizeof(head->mask));
603 head->mask_assigned = true;
604
605 fl_init_dissector(head, mask);
606
607 return 0;
608}
609
610static int fl_set_parms(struct net *net, struct tcf_proto *tp,
611 struct cls_fl_filter *f, struct fl_flow_mask *mask,
612 unsigned long base, struct nlattr **tb,
613 struct nlattr *est, bool ovr)
614{
615 struct tcf_exts e;
616 int err;
617
b9a24bb7 618 err = tcf_exts_init(&e, TCA_FLOWER_ACT, 0);
77b9900e
JP		 619 if (err < 0)
620 return err;
b9a24bb7
WC		 621 err = tcf_exts_validate(net, tp, tb, est, &e, ovr);
622 if (err < 0)
623 goto errout;
77b9900e
JP		 624
625 if (tb[TCA_FLOWER_CLASSID]) {
626 f->res.classid = nla_get_u32(tb[TCA_FLOWER_CLASSID]);
627 tcf_bind_filter(tp, &f->res, base);
628 }
629
630 err = fl_set_key(net, tb, &f->key, &mask->key);
631 if (err)
632 goto errout;
633
634 fl_mask_update_range(mask);
635 fl_set_masked_key(&f->mkey, &f->key, mask);
636
637 tcf_exts_change(tp, &f->exts, &e);
638
639 return 0;
640errout:
641 tcf_exts_destroy(&e);
642 return err;
643}
644
645static u32 fl_grab_new_handle(struct tcf_proto *tp,
646 struct cls_fl_head *head)
647{
648 unsigned int i = 0x80000000;
649 u32 handle;
650
651 do {
652 if (++head->hgen == 0x7FFFFFFF)
653 head->hgen = 1;
654 } while (--i > 0 && fl_get(tp, head->hgen));
655
656 if (unlikely(i == 0)) {
657 pr_err("Insufficient number of handles\n");
658 handle = 0;
659 } else {
660 handle = head->hgen;
661 }
662
663 return handle;
664}
665
666static int fl_change(struct net *net, struct sk_buff *in_skb,
667 struct tcf_proto *tp, unsigned long base,
668 u32 handle, struct nlattr **tca,
669 unsigned long *arg, bool ovr)
670{
671 struct cls_fl_head *head = rtnl_dereference(tp->root);
672 struct cls_fl_filter *fold = (struct cls_fl_filter *) *arg;
673 struct cls_fl_filter *fnew;
674 struct nlattr *tb[TCA_FLOWER_MAX + 1];
675 struct fl_flow_mask mask = {};
676 int err;
677
678 if (!tca[TCA_OPTIONS])
679 return -EINVAL;
680
681 err = nla_parse_nested(tb, TCA_FLOWER_MAX, tca[TCA_OPTIONS], fl_policy);
682 if (err < 0)
683 return err;
684
685 if (fold && handle && fold->handle != handle)
686 return -EINVAL;
687
688 fnew = kzalloc(sizeof(*fnew), GFP_KERNEL);
689 if (!fnew)
690 return -ENOBUFS;
691
b9a24bb7
WC		 692 err = tcf_exts_init(&fnew->exts, TCA_FLOWER_ACT, 0);
693 if (err < 0)
694 goto errout;
77b9900e
JP		 695
696 if (!handle) {
697 handle = fl_grab_new_handle(tp, head);
698 if (!handle) {
699 err = -EINVAL;
700 goto errout;
701 }
702 }
703 fnew->handle = handle;
704
e69985c6
AV		 705 if (tb[TCA_FLOWER_FLAGS]) {
706 fnew->flags = nla_get_u32(tb[TCA_FLOWER_FLAGS]);
707
708 if (!tc_flags_valid(fnew->flags)) {
709 err = -EINVAL;
710 goto errout;
711 }
712 }
5b33f488 713
77b9900e
JP		 714 err = fl_set_parms(net, tp, fnew, &mask, base, tb, tca[TCA_RATE], ovr);
715 if (err)
716 goto errout;
717
718 err = fl_check_assign_mask(head, &mask);
719 if (err)
720 goto errout;
721
e8eb36cd 722 if (!tc_skip_sw(fnew->flags)) {
e69985c6
AV		 723 err = rhashtable_insert_fast(&head->ht, &fnew->ht_node,
724 head->ht_params);
725 if (err)
726 goto errout;
727 }
5b33f488 728
e8eb36cd
AV		 729 err = fl_hw_replace_filter(tp,
730 &head->dissector,
731 &mask.key,
732 &fnew->key,
733 &fnew->exts,
734 (unsigned long)fnew,
735 fnew->flags);
736 if (err)
737 goto errout;
5b33f488
AV		 738
739 if (fold) {
77b9900e
JP		 740 rhashtable_remove_fast(&head->ht, &fold->ht_node,
741 head->ht_params);
8208d21b 742 fl_hw_destroy_filter(tp, (unsigned long)fold);
5b33f488 743 }
77b9900e
JP		 744
745 *arg = (unsigned long) fnew;
746
747 if (fold) {
ff3532f2 748 list_replace_rcu(&fold->list, &fnew->list);
77b9900e
JP		 749 tcf_unbind_filter(tp, &fold->res);
750 call_rcu(&fold->rcu, fl_destroy_filter);
751 } else {
752 list_add_tail_rcu(&fnew->list, &head->filters);
753 }
754
755 return 0;
756
757errout:
b9a24bb7 758 tcf_exts_destroy(&fnew->exts);
77b9900e
JP		 759 kfree(fnew);
760 return err;
761}
762
763static int fl_delete(struct tcf_proto *tp, unsigned long arg)
764{
765 struct cls_fl_head *head = rtnl_dereference(tp->root);
766 struct cls_fl_filter *f = (struct cls_fl_filter *) arg;
767
768 rhashtable_remove_fast(&head->ht, &f->ht_node,
769 head->ht_params);
13fa876e 770 __fl_delete(tp, f);
77b9900e
JP		 771 return 0;
772}
773
774static void fl_walk(struct tcf_proto *tp, struct tcf_walker *arg)
775{
776 struct cls_fl_head *head = rtnl_dereference(tp->root);
777 struct cls_fl_filter *f;
778
779 list_for_each_entry_rcu(f, &head->filters, list) {
780 if (arg->count < arg->skip)
781 goto skip;
782 if (arg->fn(tp, (unsigned long) f, arg) < 0) {
783 arg->stop = 1;
784 break;
785 }
786skip:
787 arg->count++;
788 }
789}
790
791static int fl_dump_key_val(struct sk_buff *skb,
792 void *val, int val_type,
793 void *mask, int mask_type, int len)
794{
795 int err;
796
797 if (!memchr_inv(mask, 0, len))
798 return 0;
799 err = nla_put(skb, val_type, len, val);
800 if (err)
801 return err;
802 if (mask_type != TCA_FLOWER_UNSPEC) {
803 err = nla_put(skb, mask_type, len, mask);
804 if (err)
805 return err;
806 }
807 return 0;
808}
809
9399ae9a
HHZ		 810static int fl_dump_key_vlan(struct sk_buff *skb,
811 struct flow_dissector_key_vlan *vlan_key,
812 struct flow_dissector_key_vlan *vlan_mask)
813{
814 int err;
815
816 if (!memchr_inv(vlan_mask, 0, sizeof(*vlan_mask)))
817 return 0;
818 if (vlan_mask->vlan_id) {
819 err = nla_put_u16(skb, TCA_FLOWER_KEY_VLAN_ID,
820 vlan_key->vlan_id);
821 if (err)
822 return err;
823 }
824 if (vlan_mask->vlan_priority) {
825 err = nla_put_u8(skb, TCA_FLOWER_KEY_VLAN_PRIO,
826 vlan_key->vlan_priority);
827 if (err)
828 return err;
829 }
830 return 0;
831}
832
77b9900e
JP		 833static int fl_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
834 struct sk_buff *skb, struct tcmsg *t)
835{
836 struct cls_fl_head *head = rtnl_dereference(tp->root);
837 struct cls_fl_filter *f = (struct cls_fl_filter *) fh;
838 struct nlattr *nest;
839 struct fl_flow_key *key, *mask;
840
841 if (!f)
842 return skb->len;
843
844 t->tcm_handle = f->handle;
845
846 nest = nla_nest_start(skb, TCA_OPTIONS);
847 if (!nest)
848 goto nla_put_failure;
849
850 if (f->res.classid &&
851 nla_put_u32(skb, TCA_FLOWER_CLASSID, f->res.classid))
852 goto nla_put_failure;
853
854 key = &f->key;
855 mask = &head->mask.key;
856
857 if (mask->indev_ifindex) {
858 struct net_device *dev;
859
860 dev = __dev_get_by_index(net, key->indev_ifindex);
861 if (dev && nla_put_string(skb, TCA_FLOWER_INDEV, dev->name))
862 goto nla_put_failure;
863 }
864
10cbc684
AV		 865 fl_hw_update_stats(tp, f);
866
77b9900e
JP		 867 if (fl_dump_key_val(skb, key->eth.dst, TCA_FLOWER_KEY_ETH_DST,
868 mask->eth.dst, TCA_FLOWER_KEY_ETH_DST_MASK,
869 sizeof(key->eth.dst)) ||
870 fl_dump_key_val(skb, key->eth.src, TCA_FLOWER_KEY_ETH_SRC,
871 mask->eth.src, TCA_FLOWER_KEY_ETH_SRC_MASK,
872 sizeof(key->eth.src)) ||
873 fl_dump_key_val(skb, &key->basic.n_proto, TCA_FLOWER_KEY_ETH_TYPE,
874 &mask->basic.n_proto, TCA_FLOWER_UNSPEC,
875 sizeof(key->basic.n_proto)))
876 goto nla_put_failure;
9399ae9a
HHZ		 877
878 if (fl_dump_key_vlan(skb, &key->vlan, &mask->vlan))
879 goto nla_put_failure;
880
77b9900e
JP		 881 if ((key->basic.n_proto == htons(ETH_P_IP) ||
882 key->basic.n_proto == htons(ETH_P_IPV6)) &&
883 fl_dump_key_val(skb, &key->basic.ip_proto, TCA_FLOWER_KEY_IP_PROTO,
884 &mask->basic.ip_proto, TCA_FLOWER_UNSPEC,
885 sizeof(key->basic.ip_proto)))
886 goto nla_put_failure;
887
c3f83241 888 if (key->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS &&
77b9900e
JP		 889 (fl_dump_key_val(skb, &key->ipv4.src, TCA_FLOWER_KEY_IPV4_SRC,
890 &mask->ipv4.src, TCA_FLOWER_KEY_IPV4_SRC_MASK,
891 sizeof(key->ipv4.src)) ||
892 fl_dump_key_val(skb, &key->ipv4.dst, TCA_FLOWER_KEY_IPV4_DST,
893 &mask->ipv4.dst, TCA_FLOWER_KEY_IPV4_DST_MASK,
894 sizeof(key->ipv4.dst))))
895 goto nla_put_failure;
c3f83241 896 else if (key->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS &&
77b9900e
JP		 897 (fl_dump_key_val(skb, &key->ipv6.src, TCA_FLOWER_KEY_IPV6_SRC,
898 &mask->ipv6.src, TCA_FLOWER_KEY_IPV6_SRC_MASK,
899 sizeof(key->ipv6.src)) ||
900 fl_dump_key_val(skb, &key->ipv6.dst, TCA_FLOWER_KEY_IPV6_DST,
901 &mask->ipv6.dst, TCA_FLOWER_KEY_IPV6_DST_MASK,
902 sizeof(key->ipv6.dst))))
903 goto nla_put_failure;
904
905 if (key->basic.ip_proto == IPPROTO_TCP &&
906 (fl_dump_key_val(skb, &key->tp.src, TCA_FLOWER_KEY_TCP_SRC,
aa72d708 907 &mask->tp.src, TCA_FLOWER_KEY_TCP_SRC_MASK,
77b9900e
JP		 908 sizeof(key->tp.src)) ||
909 fl_dump_key_val(skb, &key->tp.dst, TCA_FLOWER_KEY_TCP_DST,
aa72d708 910 &mask->tp.dst, TCA_FLOWER_KEY_TCP_DST_MASK,
77b9900e
JP		 911 sizeof(key->tp.dst))))
912 goto nla_put_failure;
913 else if (key->basic.ip_proto == IPPROTO_UDP &&
914 (fl_dump_key_val(skb, &key->tp.src, TCA_FLOWER_KEY_UDP_SRC,
aa72d708 915 &mask->tp.src, TCA_FLOWER_KEY_UDP_SRC_MASK,
77b9900e
JP		 916 sizeof(key->tp.src)) ||
917 fl_dump_key_val(skb, &key->tp.dst, TCA_FLOWER_KEY_UDP_DST,
aa72d708 918 &mask->tp.dst, TCA_FLOWER_KEY_UDP_DST_MASK,
5976c5f4
SH		 919 sizeof(key->tp.dst))))
920 goto nla_put_failure;
921 else if (key->basic.ip_proto == IPPROTO_SCTP &&
922 (fl_dump_key_val(skb, &key->tp.src, TCA_FLOWER_KEY_SCTP_SRC,
923 &mask->tp.src, TCA_FLOWER_KEY_SCTP_SRC_MASK,
924 sizeof(key->tp.src)) ||
925 fl_dump_key_val(skb, &key->tp.dst, TCA_FLOWER_KEY_SCTP_DST,
926 &mask->tp.dst, TCA_FLOWER_KEY_SCTP_DST_MASK,
77b9900e
JP		 927 sizeof(key->tp.dst))))
928 goto nla_put_failure;
929
bc3103f1
AV		 930 if (key->enc_control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS &&
931 (fl_dump_key_val(skb, &key->enc_ipv4.src,
932 TCA_FLOWER_KEY_ENC_IPV4_SRC, &mask->enc_ipv4.src,
933 TCA_FLOWER_KEY_ENC_IPV4_SRC_MASK,
934 sizeof(key->enc_ipv4.src)) ||
935 fl_dump_key_val(skb, &key->enc_ipv4.dst,
936 TCA_FLOWER_KEY_ENC_IPV4_DST, &mask->enc_ipv4.dst,
937 TCA_FLOWER_KEY_ENC_IPV4_DST_MASK,
938 sizeof(key->enc_ipv4.dst))))
939 goto nla_put_failure;
940 else if (key->enc_control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS &&
941 (fl_dump_key_val(skb, &key->enc_ipv6.src,
942 TCA_FLOWER_KEY_ENC_IPV6_SRC, &mask->enc_ipv6.src,
943 TCA_FLOWER_KEY_ENC_IPV6_SRC_MASK,
944 sizeof(key->enc_ipv6.src)) ||
945 fl_dump_key_val(skb, &key->enc_ipv6.dst,
946 TCA_FLOWER_KEY_ENC_IPV6_DST,
947 &mask->enc_ipv6.dst,
948 TCA_FLOWER_KEY_ENC_IPV6_DST_MASK,
949 sizeof(key->enc_ipv6.dst))))
950 goto nla_put_failure;
951
952 if (fl_dump_key_val(skb, &key->enc_key_id, TCA_FLOWER_KEY_ENC_KEY_ID,
eb523f42 953 &mask->enc_key_id, TCA_FLOWER_UNSPEC,
bc3103f1
AV		 954 sizeof(key->enc_key_id)))
955 goto nla_put_failure;
956
e69985c6
AV		 957 nla_put_u32(skb, TCA_FLOWER_FLAGS, f->flags);
958
77b9900e
JP		 959 if (tcf_exts_dump(skb, &f->exts))
960 goto nla_put_failure;
961
962 nla_nest_end(skb, nest);
963
964 if (tcf_exts_dump_stats(skb, &f->exts) < 0)
965 goto nla_put_failure;
966
967 return skb->len;
968
969nla_put_failure:
970 nla_nest_cancel(skb, nest);
971 return -1;
972}
973
974static struct tcf_proto_ops cls_fl_ops __read_mostly = {
975 .kind = "flower",
976 .classify = fl_classify,
977 .init = fl_init,
978 .destroy = fl_destroy,
979 .get = fl_get,
980 .change = fl_change,
981 .delete = fl_delete,
982 .walk = fl_walk,
983 .dump = fl_dump,
984 .owner = THIS_MODULE,
985};
986
987static int __init cls_fl_init(void)
988{
989 return register_tcf_proto_ops(&cls_fl_ops);
990}
991
992static void __exit cls_fl_exit(void)
993{
994 unregister_tcf_proto_ops(&cls_fl_ops);
995}
996
997module_init(cls_fl_init);
998module_exit(cls_fl_exit);
999
1000MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>");
1001MODULE_DESCRIPTION("Flower classifier");
1002MODULE_LICENSE("GPL v2");
Linux 6.x block layer and io_uring tree(s)