1 /* -*- linux-c -*-
2 * sysctl_net_core.c: sysctl interface to net core subsystem.
3 *
4 * Begun April 1, 1996, Mike Shaver.
5 * Added /proc/sys/net/core directory entry (empty =) ). [MS]
6 */
7
8 #include <linux/mm.h>
9 #include <linux/sysctl.h>
10 #include <linux/module.h>
11 #include <linux/socket.h>
12 #include <linux/netdevice.h>
13 #include <linux/ratelimit.h>
14 #include <linux/vmalloc.h>
15 #include <linux/init.h>
16 #include <linux/slab.h>
17 #include <linux/kmemleak.h>
18
19 #include <net/ip.h>
20 #include <net/sock.h>
21 #include <net/net_ratelimit.h>
22 #include <net/busy_poll.h>
23 #include <net/pkt_sched.h>
24
25 static int zero = 0;
26 static int one = 1;
27 static int min_sndbuf = SOCK_MIN_SNDBUF;
28 static int min_rcvbuf = SOCK_MIN_RCVBUF;
29 static int max_skb_frags = MAX_SKB_FRAGS;
30
31 static int net_msg_warn; /* Unused, but still a sysctl */
32
33 #ifdef CONFIG_RPS
rps_sock_flow_sysctl(struct ctl_table * table,int write,void __user * buffer,size_t * lenp,loff_t * ppos)34 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
35 void __user *buffer, size_t *lenp, loff_t *ppos)
36 {
37 unsigned int orig_size, size;
38 int ret, i;
39 struct ctl_table tmp = {
40 .data = &size,
41 .maxlen = sizeof(size),
42 .mode = table->mode
43 };
44 struct rps_sock_flow_table *orig_sock_table, *sock_table;
45 static DEFINE_MUTEX(sock_flow_mutex);
46
47 mutex_lock(&sock_flow_mutex);
48
49 orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
50 lockdep_is_held(&sock_flow_mutex));
51 size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
52
53 ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
54
55 if (write) {
56 if (size) {
57 if (size > 1<<29) {
58 /* Enforce limit to prevent overflow */
59 mutex_unlock(&sock_flow_mutex);
60 return -EINVAL;
61 }
62 size = roundup_pow_of_two(size);
63 if (size != orig_size) {
64 sock_table =
65 vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
66 if (!sock_table) {
67 mutex_unlock(&sock_flow_mutex);
68 return -ENOMEM;
69 }
70 rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
71 sock_table->mask = size - 1;
72 } else
73 sock_table = orig_sock_table;
74
75 for (i = 0; i < size; i++)
76 sock_table->ents[i] = RPS_NO_CPU;
77 } else
78 sock_table = NULL;
79
80 if (sock_table != orig_sock_table) {
81 rcu_assign_pointer(rps_sock_flow_table, sock_table);
82 if (sock_table)
83 static_key_slow_inc(&rps_needed);
84 if (orig_sock_table) {
85 static_key_slow_dec(&rps_needed);
86 synchronize_rcu();
87 vfree(orig_sock_table);
88 }
89 }
90 }
91
92 mutex_unlock(&sock_flow_mutex);
93
94 return ret;
95 }
96 #endif /* CONFIG_RPS */
97
98 #ifdef CONFIG_NET_FLOW_LIMIT
99 static DEFINE_MUTEX(flow_limit_update_mutex);
100
flow_limit_cpu_sysctl(struct ctl_table * table,int write,void __user * buffer,size_t * lenp,loff_t * ppos)101 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
102 void __user *buffer, size_t *lenp,
103 loff_t *ppos)
104 {
105 struct sd_flow_limit *cur;
106 struct softnet_data *sd;
107 cpumask_var_t mask;
108 int i, len, ret = 0;
109
110 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
111 return -ENOMEM;
112
113 if (write) {
114 ret = cpumask_parse_user(buffer, *lenp, mask);
115 if (ret)
116 goto done;
117
118 mutex_lock(&flow_limit_update_mutex);
119 len = sizeof(*cur) + netdev_flow_limit_table_len;
120 for_each_possible_cpu(i) {
121 sd = &per_cpu(softnet_data, i);
122 cur = rcu_dereference_protected(sd->flow_limit,
123 lockdep_is_held(&flow_limit_update_mutex));
124 if (cur && !cpumask_test_cpu(i, mask)) {
125 RCU_INIT_POINTER(sd->flow_limit, NULL);
126 synchronize_rcu();
127 kfree(cur);
128 } else if (!cur && cpumask_test_cpu(i, mask)) {
129 cur = kzalloc_node(len, GFP_KERNEL,
130 cpu_to_node(i));
131 if (!cur) {
132 /* not unwinding previous changes */
133 ret = -ENOMEM;
134 goto write_unlock;
135 }
136 cur->num_buckets = netdev_flow_limit_table_len;
137 rcu_assign_pointer(sd->flow_limit, cur);
138 }
139 }
140 write_unlock:
141 mutex_unlock(&flow_limit_update_mutex);
142 } else {
143 char kbuf[128];
144
145 if (*ppos || !*lenp) {
146 *lenp = 0;
147 goto done;
148 }
149
150 cpumask_clear(mask);
151 rcu_read_lock();
152 for_each_possible_cpu(i) {
153 sd = &per_cpu(softnet_data, i);
154 if (rcu_dereference(sd->flow_limit))
155 cpumask_set_cpu(i, mask);
156 }
157 rcu_read_unlock();
158
159 len = min(sizeof(kbuf) - 1, *lenp);
160 len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
161 if (!len) {
162 *lenp = 0;
163 goto done;
164 }
165 if (len < *lenp)
166 kbuf[len++] = '\n';
167 if (copy_to_user(buffer, kbuf, len)) {
168 ret = -EFAULT;
169 goto done;
170 }
171 *lenp = len;
172 *ppos += len;
173 }
174
175 done:
176 free_cpumask_var(mask);
177 return ret;
178 }
179
flow_limit_table_len_sysctl(struct ctl_table * table,int write,void __user * buffer,size_t * lenp,loff_t * ppos)180 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
181 void __user *buffer, size_t *lenp,
182 loff_t *ppos)
183 {
184 unsigned int old, *ptr;
185 int ret;
186
187 mutex_lock(&flow_limit_update_mutex);
188
189 ptr = table->data;
190 old = *ptr;
191 ret = proc_dointvec(table, write, buffer, lenp, ppos);
192 if (!ret && write && !is_power_of_2(*ptr)) {
193 *ptr = old;
194 ret = -EINVAL;
195 }
196
197 mutex_unlock(&flow_limit_update_mutex);
198 return ret;
199 }
200 #endif /* CONFIG_NET_FLOW_LIMIT */
201
202 #ifdef CONFIG_NET_SCHED
set_default_qdisc(struct ctl_table * table,int write,void __user * buffer,size_t * lenp,loff_t * ppos)203 static int set_default_qdisc(struct ctl_table *table, int write,
204 void __user *buffer, size_t *lenp, loff_t *ppos)
205 {
206 char id[IFNAMSIZ];
207 struct ctl_table tbl = {
208 .data = id,
209 .maxlen = IFNAMSIZ,
210 };
211 int ret;
212
213 qdisc_get_default(id, IFNAMSIZ);
214
215 ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
216 if (write && ret == 0)
217 ret = qdisc_set_default(id);
218 return ret;
219 }
220 #endif
221
proc_do_rss_key(struct ctl_table * table,int write,void __user * buffer,size_t * lenp,loff_t * ppos)222 static int proc_do_rss_key(struct ctl_table *table, int write,
223 void __user *buffer, size_t *lenp, loff_t *ppos)
224 {
225 struct ctl_table fake_table;
226 char buf[NETDEV_RSS_KEY_LEN * 3];
227
228 snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
229 fake_table.data = buf;
230 fake_table.maxlen = sizeof(buf);
231 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
232 }
233
234 static struct ctl_table net_core_table[] = {
235 #ifdef CONFIG_NET
236 {
237 .procname = "wmem_max",
238 .data = &sysctl_wmem_max,
239 .maxlen = sizeof(int),
240 .mode = 0644,
241 .proc_handler = proc_dointvec_minmax,
242 .extra1 = &min_sndbuf,
243 },
244 {
245 .procname = "rmem_max",
246 .data = &sysctl_rmem_max,
247 .maxlen = sizeof(int),
248 .mode = 0644,
249 .proc_handler = proc_dointvec_minmax,
250 .extra1 = &min_rcvbuf,
251 },
252 {
253 .procname = "wmem_default",
254 .data = &sysctl_wmem_default,
255 .maxlen = sizeof(int),
256 .mode = 0644,
257 .proc_handler = proc_dointvec_minmax,
258 .extra1 = &min_sndbuf,
259 },
260 {
261 .procname = "rmem_default",
262 .data = &sysctl_rmem_default,
263 .maxlen = sizeof(int),
264 .mode = 0644,
265 .proc_handler = proc_dointvec_minmax,
266 .extra1 = &min_rcvbuf,
267 },
268 {
269 .procname = "dev_weight",
270 .data = &weight_p,
271 .maxlen = sizeof(int),
272 .mode = 0644,
273 .proc_handler = proc_dointvec
274 },
275 {
276 .procname = "netdev_max_backlog",
277 .data = &netdev_max_backlog,
278 .maxlen = sizeof(int),
279 .mode = 0644,
280 .proc_handler = proc_dointvec
281 },
282 {
283 .procname = "netdev_rss_key",
284 .data = &netdev_rss_key,
285 .maxlen = sizeof(int),
286 .mode = 0444,
287 .proc_handler = proc_do_rss_key,
288 },
289 #ifdef CONFIG_BPF_JIT
290 {
291 .procname = "bpf_jit_enable",
292 .data = &bpf_jit_enable,
293 .maxlen = sizeof(int),
294 .mode = 0644,
295 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
296 .proc_handler = proc_dointvec
297 #else
298 .proc_handler = proc_dointvec_minmax,
299 .extra1 = &one,
300 .extra2 = &one,
301 #endif
302 },
303 # ifdef CONFIG_HAVE_EBPF_JIT
304 {
305 .procname = "bpf_jit_harden",
306 .data = &bpf_jit_harden,
307 .maxlen = sizeof(int),
308 .mode = 0600,
309 .proc_handler = proc_dointvec,
310 },
311 # endif
312 #endif
313 {
314 .procname = "netdev_tstamp_prequeue",
315 .data = &netdev_tstamp_prequeue,
316 .maxlen = sizeof(int),
317 .mode = 0644,
318 .proc_handler = proc_dointvec
319 },
320 {
321 .procname = "message_cost",
322 .data = &net_ratelimit_state.interval,
323 .maxlen = sizeof(int),
324 .mode = 0644,
325 .proc_handler = proc_dointvec_jiffies,
326 },
327 {
328 .procname = "message_burst",
329 .data = &net_ratelimit_state.burst,
330 .maxlen = sizeof(int),
331 .mode = 0644,
332 .proc_handler = proc_dointvec,
333 },
334 {
335 .procname = "optmem_max",
336 .data = &sysctl_optmem_max,
337 .maxlen = sizeof(int),
338 .mode = 0644,
339 .proc_handler = proc_dointvec
340 },
341 {
342 .procname = "tstamp_allow_data",
343 .data = &sysctl_tstamp_allow_data,
344 .maxlen = sizeof(int),
345 .mode = 0644,
346 .proc_handler = proc_dointvec_minmax,
347 .extra1 = &zero,
348 .extra2 = &one
349 },
350 #ifdef CONFIG_RPS
351 {
352 .procname = "rps_sock_flow_entries",
353 .maxlen = sizeof(int),
354 .mode = 0644,
355 .proc_handler = rps_sock_flow_sysctl
356 },
357 #endif
358 #ifdef CONFIG_NET_FLOW_LIMIT
359 {
360 .procname = "flow_limit_cpu_bitmap",
361 .mode = 0644,
362 .proc_handler = flow_limit_cpu_sysctl
363 },
364 {
365 .procname = "flow_limit_table_len",
366 .data = &netdev_flow_limit_table_len,
367 .maxlen = sizeof(int),
368 .mode = 0644,
369 .proc_handler = flow_limit_table_len_sysctl
370 },
371 #endif /* CONFIG_NET_FLOW_LIMIT */
372 #ifdef CONFIG_NET_RX_BUSY_POLL
373 {
374 .procname = "busy_poll",
375 .data = &sysctl_net_busy_poll,
376 .maxlen = sizeof(unsigned int),
377 .mode = 0644,
378 .proc_handler = proc_dointvec_minmax,
379 .extra1 = &zero,
380 },
381 {
382 .procname = "busy_read",
383 .data = &sysctl_net_busy_read,
384 .maxlen = sizeof(unsigned int),
385 .mode = 0644,
386 .proc_handler = proc_dointvec_minmax,
387 .extra1 = &zero,
388 },
389 #endif
390 #ifdef CONFIG_NET_SCHED
391 {
392 .procname = "default_qdisc",
393 .mode = 0644,
394 .maxlen = IFNAMSIZ,
395 .proc_handler = set_default_qdisc
396 },
397 #endif
398 #endif /* CONFIG_NET */
399 {
400 .procname = "netdev_budget",
401 .data = &netdev_budget,
402 .maxlen = sizeof(int),
403 .mode = 0644,
404 .proc_handler = proc_dointvec
405 },
406 {
407 .procname = "warnings",
408 .data = &net_msg_warn,
409 .maxlen = sizeof(int),
410 .mode = 0644,
411 .proc_handler = proc_dointvec
412 },
413 {
414 .procname = "max_skb_frags",
415 .data = &sysctl_max_skb_frags,
416 .maxlen = sizeof(int),
417 .mode = 0644,
418 .proc_handler = proc_dointvec_minmax,
419 .extra1 = &one,
420 .extra2 = &max_skb_frags,
421 },
422 { }
423 };
424
425 static struct ctl_table netns_core_table[] = {
426 {
427 .procname = "somaxconn",
428 .data = &init_net.core.sysctl_somaxconn,
429 .maxlen = sizeof(int),
430 .mode = 0644,
431 .extra1 = &zero,
432 .proc_handler = proc_dointvec_minmax
433 },
434 { }
435 };
436
sysctl_core_net_init(struct net * net)437 static __net_init int sysctl_core_net_init(struct net *net)
438 {
439 struct ctl_table *tbl;
440
441 tbl = netns_core_table;
442 if (!net_eq(net, &init_net)) {
443 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
444 if (tbl == NULL)
445 goto err_dup;
446
447 tbl[0].data = &net->core.sysctl_somaxconn;
448
449 /* Don't export any sysctls to unprivileged users */
450 if (net->user_ns != &init_user_ns) {
451 tbl[0].procname = NULL;
452 }
453 }
454
455 net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
456 if (net->core.sysctl_hdr == NULL)
457 goto err_reg;
458
459 return 0;
460
461 err_reg:
462 if (tbl != netns_core_table)
463 kfree(tbl);
464 err_dup:
465 return -ENOMEM;
466 }
467
sysctl_core_net_exit(struct net * net)468 static __net_exit void sysctl_core_net_exit(struct net *net)
469 {
470 struct ctl_table *tbl;
471
472 tbl = net->core.sysctl_hdr->ctl_table_arg;
473 unregister_net_sysctl_table(net->core.sysctl_hdr);
474 BUG_ON(tbl == netns_core_table);
475 kfree(tbl);
476 }
477
478 static __net_initdata struct pernet_operations sysctl_core_ops = {
479 .init = sysctl_core_net_init,
480 .exit = sysctl_core_net_exit,
481 };
482
sysctl_core_init(void)483 static __init int sysctl_core_init(void)
484 {
485 register_net_sysctl(&init_net, "net/core", net_core_table);
486 return register_pernet_subsys(&sysctl_core_ops);
487 }
488
489 fs_initcall(sysctl_core_init);
490