fq_codel: fix return value of fq_codel_drop()
[linux-2.6-block.git] / net / sched / act_bpf.c
CommitLineData
d23b8ad8
JP
1/*
2 * Copyright (c) 2015 Jiri Pirko <jiri@resnulli.us>
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 */
9
10#include <linux/module.h>
11#include <linux/init.h>
12#include <linux/kernel.h>
13#include <linux/skbuff.h>
14#include <linux/rtnetlink.h>
15#include <linux/filter.h>
a8cb5f55
DB
16#include <linux/bpf.h>
17
d23b8ad8
JP
18#include <net/netlink.h>
19#include <net/pkt_sched.h>
20
21#include <linux/tc_act/tc_bpf.h>
22#include <net/tc_act/tc_bpf.h>
23
a8cb5f55
DB
24#define BPF_TAB_MASK 15
25#define ACT_BPF_NAME_LEN 256
26
27struct tcf_bpf_cfg {
28 struct bpf_prog *filter;
29 struct sock_filter *bpf_ops;
30 char *bpf_name;
31 u32 bpf_fd;
32 u16 bpf_num_ops;
33};
d23b8ad8 34
a8cb5f55 35static int tcf_bpf(struct sk_buff *skb, const struct tc_action *act,
d23b8ad8
JP
36 struct tcf_result *res)
37{
a8cb5f55 38 struct tcf_bpf *prog = act->priv;
ced585c8 39 int action, filter_res;
3431205e 40 bool at_ingress = G_TC_AT(skb->tc_verd) & AT_INGRESS;
d23b8ad8 41
a166151c
AS
42 if (unlikely(!skb_mac_header_was_set(skb)))
43 return TC_ACT_UNSPEC;
44
a8cb5f55 45 spin_lock(&prog->tcf_lock);
ced585c8 46
a8cb5f55
DB
47 prog->tcf_tm.lastuse = jiffies;
48 bstats_update(&prog->tcf_bstats, skb);
d23b8ad8 49
a8cb5f55
DB
50 /* Needed here for accessing maps. */
51 rcu_read_lock();
3431205e
AS
52 if (at_ingress) {
53 __skb_push(skb, skb->mac_len);
54 filter_res = BPF_PROG_RUN(prog->filter, skb);
55 __skb_pull(skb, skb->mac_len);
56 } else {
57 filter_res = BPF_PROG_RUN(prog->filter, skb);
58 }
a8cb5f55 59 rcu_read_unlock();
ced585c8
DB
60
61 /* A BPF program may overwrite the default action opcode.
62 * Similarly as in cls_bpf, if filter_res == -1 we use the
63 * default action specified from tc.
64 *
65 * In case a different well-known TC_ACT opcode has been
66 * returned, it will overwrite the default one.
67 *
68 * For everything else that is unkown, TC_ACT_UNSPEC is
69 * returned.
70 */
71 switch (filter_res) {
72 case TC_ACT_PIPE:
73 case TC_ACT_RECLASSIFY:
74 case TC_ACT_OK:
75 action = filter_res;
76 break;
77 case TC_ACT_SHOT:
78 action = filter_res;
a8cb5f55 79 prog->tcf_qstats.drops++;
ced585c8
DB
80 break;
81 case TC_ACT_UNSPEC:
a8cb5f55 82 action = prog->tcf_action;
ced585c8
DB
83 break;
84 default:
85 action = TC_ACT_UNSPEC;
86 break;
d23b8ad8
JP
87 }
88
a8cb5f55 89 spin_unlock(&prog->tcf_lock);
d23b8ad8
JP
90 return action;
91}
92
a8cb5f55
DB
93static bool tcf_bpf_is_ebpf(const struct tcf_bpf *prog)
94{
95 return !prog->bpf_ops;
96}
97
98static int tcf_bpf_dump_bpf_info(const struct tcf_bpf *prog,
99 struct sk_buff *skb)
100{
101 struct nlattr *nla;
102
103 if (nla_put_u16(skb, TCA_ACT_BPF_OPS_LEN, prog->bpf_num_ops))
104 return -EMSGSIZE;
105
106 nla = nla_reserve(skb, TCA_ACT_BPF_OPS, prog->bpf_num_ops *
107 sizeof(struct sock_filter));
108 if (nla == NULL)
109 return -EMSGSIZE;
110
111 memcpy(nla_data(nla), prog->bpf_ops, nla_len(nla));
112
113 return 0;
114}
115
116static int tcf_bpf_dump_ebpf_info(const struct tcf_bpf *prog,
117 struct sk_buff *skb)
118{
119 if (nla_put_u32(skb, TCA_ACT_BPF_FD, prog->bpf_fd))
120 return -EMSGSIZE;
121
122 if (prog->bpf_name &&
123 nla_put_string(skb, TCA_ACT_BPF_NAME, prog->bpf_name))
124 return -EMSGSIZE;
125
126 return 0;
127}
128
129static int tcf_bpf_dump(struct sk_buff *skb, struct tc_action *act,
d23b8ad8
JP
130 int bind, int ref)
131{
132 unsigned char *tp = skb_tail_pointer(skb);
a8cb5f55 133 struct tcf_bpf *prog = act->priv;
d23b8ad8 134 struct tc_act_bpf opt = {
a8cb5f55
DB
135 .index = prog->tcf_index,
136 .refcnt = prog->tcf_refcnt - ref,
137 .bindcnt = prog->tcf_bindcnt - bind,
138 .action = prog->tcf_action,
d23b8ad8 139 };
a8cb5f55
DB
140 struct tcf_t tm;
141 int ret;
d23b8ad8
JP
142
143 if (nla_put(skb, TCA_ACT_BPF_PARMS, sizeof(opt), &opt))
144 goto nla_put_failure;
145
a8cb5f55
DB
146 if (tcf_bpf_is_ebpf(prog))
147 ret = tcf_bpf_dump_ebpf_info(prog, skb);
148 else
149 ret = tcf_bpf_dump_bpf_info(prog, skb);
150 if (ret)
d23b8ad8
JP
151 goto nla_put_failure;
152
a8cb5f55
DB
153 tm.install = jiffies_to_clock_t(jiffies - prog->tcf_tm.install);
154 tm.lastuse = jiffies_to_clock_t(jiffies - prog->tcf_tm.lastuse);
155 tm.expires = jiffies_to_clock_t(prog->tcf_tm.expires);
d23b8ad8 156
a8cb5f55 157 if (nla_put(skb, TCA_ACT_BPF_TM, sizeof(tm), &tm))
d23b8ad8 158 goto nla_put_failure;
a8cb5f55 159
d23b8ad8
JP
160 return skb->len;
161
162nla_put_failure:
163 nlmsg_trim(skb, tp);
164 return -1;
165}
166
167static const struct nla_policy act_bpf_policy[TCA_ACT_BPF_MAX + 1] = {
168 [TCA_ACT_BPF_PARMS] = { .len = sizeof(struct tc_act_bpf) },
a8cb5f55
DB
169 [TCA_ACT_BPF_FD] = { .type = NLA_U32 },
170 [TCA_ACT_BPF_NAME] = { .type = NLA_NUL_STRING, .len = ACT_BPF_NAME_LEN },
d23b8ad8
JP
171 [TCA_ACT_BPF_OPS_LEN] = { .type = NLA_U16 },
172 [TCA_ACT_BPF_OPS] = { .type = NLA_BINARY,
173 .len = sizeof(struct sock_filter) * BPF_MAXINSNS },
174};
175
a8cb5f55 176static int tcf_bpf_init_from_ops(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
d23b8ad8 177{
d23b8ad8 178 struct sock_filter *bpf_ops;
a8cb5f55 179 struct sock_fprog_kern fprog_tmp;
d23b8ad8 180 struct bpf_prog *fp;
a8cb5f55 181 u16 bpf_size, bpf_num_ops;
d23b8ad8
JP
182 int ret;
183
d23b8ad8
JP
184 bpf_num_ops = nla_get_u16(tb[TCA_ACT_BPF_OPS_LEN]);
185 if (bpf_num_ops > BPF_MAXINSNS || bpf_num_ops == 0)
186 return -EINVAL;
187
188 bpf_size = bpf_num_ops * sizeof(*bpf_ops);
fd3e646c
DB
189 if (bpf_size != nla_len(tb[TCA_ACT_BPF_OPS]))
190 return -EINVAL;
191
d23b8ad8 192 bpf_ops = kzalloc(bpf_size, GFP_KERNEL);
a8cb5f55 193 if (bpf_ops == NULL)
d23b8ad8
JP
194 return -ENOMEM;
195
196 memcpy(bpf_ops, nla_data(tb[TCA_ACT_BPF_OPS]), bpf_size);
197
a8cb5f55
DB
198 fprog_tmp.len = bpf_num_ops;
199 fprog_tmp.filter = bpf_ops;
d23b8ad8 200
a8cb5f55
DB
201 ret = bpf_prog_create(&fp, &fprog_tmp);
202 if (ret < 0) {
203 kfree(bpf_ops);
204 return ret;
205 }
d23b8ad8 206
a8cb5f55
DB
207 cfg->bpf_ops = bpf_ops;
208 cfg->bpf_num_ops = bpf_num_ops;
209 cfg->filter = fp;
210
211 return 0;
212}
213
214static int tcf_bpf_init_from_efd(struct nlattr **tb, struct tcf_bpf_cfg *cfg)
215{
216 struct bpf_prog *fp;
217 char *name = NULL;
218 u32 bpf_fd;
219
220 bpf_fd = nla_get_u32(tb[TCA_ACT_BPF_FD]);
221
222 fp = bpf_prog_get(bpf_fd);
223 if (IS_ERR(fp))
224 return PTR_ERR(fp);
225
226 if (fp->type != BPF_PROG_TYPE_SCHED_ACT) {
227 bpf_prog_put(fp);
228 return -EINVAL;
229 }
230
231 if (tb[TCA_ACT_BPF_NAME]) {
232 name = kmemdup(nla_data(tb[TCA_ACT_BPF_NAME]),
233 nla_len(tb[TCA_ACT_BPF_NAME]),
234 GFP_KERNEL);
235 if (!name) {
236 bpf_prog_put(fp);
237 return -ENOMEM;
238 }
239 }
240
241 cfg->bpf_fd = bpf_fd;
242 cfg->bpf_name = name;
243 cfg->filter = fp;
244
245 return 0;
246}
247
248static int tcf_bpf_init(struct net *net, struct nlattr *nla,
249 struct nlattr *est, struct tc_action *act,
250 int replace, int bind)
251{
252 struct nlattr *tb[TCA_ACT_BPF_MAX + 1];
253 struct tc_act_bpf *parm;
254 struct tcf_bpf *prog;
255 struct tcf_bpf_cfg cfg;
256 bool is_bpf, is_ebpf;
257 int ret;
258
259 if (!nla)
260 return -EINVAL;
261
262 ret = nla_parse_nested(tb, TCA_ACT_BPF_MAX, nla, act_bpf_policy);
263 if (ret < 0)
264 return ret;
265
266 is_bpf = tb[TCA_ACT_BPF_OPS_LEN] && tb[TCA_ACT_BPF_OPS];
267 is_ebpf = tb[TCA_ACT_BPF_FD];
268
269 if ((!is_bpf && !is_ebpf) || (is_bpf && is_ebpf) ||
270 !tb[TCA_ACT_BPF_PARMS])
271 return -EINVAL;
272
273 parm = nla_data(tb[TCA_ACT_BPF_PARMS]);
274
275 memset(&cfg, 0, sizeof(cfg));
276
277 ret = is_bpf ? tcf_bpf_init_from_ops(tb, &cfg) :
278 tcf_bpf_init_from_efd(tb, &cfg);
279 if (ret < 0)
280 return ret;
281
282 if (!tcf_hash_check(parm->index, act, bind)) {
283 ret = tcf_hash_create(parm->index, est, act,
284 sizeof(*prog), bind);
285 if (ret < 0)
d23b8ad8
JP
286 goto destroy_fp;
287
288 ret = ACT_P_CREATED;
289 } else {
a8cb5f55 290 /* Don't override defaults. */
d23b8ad8
JP
291 if (bind)
292 goto destroy_fp;
a8cb5f55
DB
293
294 tcf_hash_release(act, bind);
295 if (!replace) {
d23b8ad8
JP
296 ret = -EEXIST;
297 goto destroy_fp;
298 }
299 }
300
a8cb5f55
DB
301 prog = to_bpf(act);
302 spin_lock_bh(&prog->tcf_lock);
303
304 prog->bpf_ops = cfg.bpf_ops;
305 prog->bpf_name = cfg.bpf_name;
306
307 if (cfg.bpf_num_ops)
308 prog->bpf_num_ops = cfg.bpf_num_ops;
309 if (cfg.bpf_fd)
310 prog->bpf_fd = cfg.bpf_fd;
311
312 prog->tcf_action = parm->action;
313 prog->filter = cfg.filter;
314
315 spin_unlock_bh(&prog->tcf_lock);
d23b8ad8
JP
316
317 if (ret == ACT_P_CREATED)
a8cb5f55
DB
318 tcf_hash_insert(act);
319
d23b8ad8
JP
320 return ret;
321
322destroy_fp:
a8cb5f55
DB
323 if (is_ebpf)
324 bpf_prog_put(cfg.filter);
325 else
326 bpf_prog_destroy(cfg.filter);
327
328 kfree(cfg.bpf_ops);
329 kfree(cfg.bpf_name);
330
d23b8ad8
JP
331 return ret;
332}
333
a8cb5f55 334static void tcf_bpf_cleanup(struct tc_action *act, int bind)
d23b8ad8 335{
a8cb5f55 336 const struct tcf_bpf *prog = act->priv;
d23b8ad8 337
a8cb5f55
DB
338 if (tcf_bpf_is_ebpf(prog))
339 bpf_prog_put(prog->filter);
340 else
341 bpf_prog_destroy(prog->filter);
d23b8ad8
JP
342}
343
a8cb5f55
DB
344static struct tc_action_ops act_bpf_ops __read_mostly = {
345 .kind = "bpf",
346 .type = TCA_ACT_BPF,
347 .owner = THIS_MODULE,
348 .act = tcf_bpf,
349 .dump = tcf_bpf_dump,
350 .cleanup = tcf_bpf_cleanup,
351 .init = tcf_bpf_init,
d23b8ad8
JP
352};
353
354static int __init bpf_init_module(void)
355{
356 return tcf_register_action(&act_bpf_ops, BPF_TAB_MASK);
357}
358
359static void __exit bpf_cleanup_module(void)
360{
361 tcf_unregister_action(&act_bpf_ops);
362}
363
364module_init(bpf_init_module);
365module_exit(bpf_cleanup_module);
366
367MODULE_AUTHOR("Jiri Pirko <jiri@resnulli.us>");
368MODULE_DESCRIPTION("TC BPF based action");
369MODULE_LICENSE("GPL v2");