xref: /linux/net/core/sysctl_net_core.c (revision 140eb5227767c6754742020a16d2691222b9c19b)
1 // SPDX-License-Identifier: GPL-2.0
2 /* -*- linux-c -*-
3  * sysctl_net_core.c: sysctl interface to net core subsystem.
4  *
5  * Begun April 1, 1996, Mike Shaver.
6  * Added /proc/sys/net/core directory entry (empty =) ). [MS]
7  */
8 
9 #include <linux/mm.h>
10 #include <linux/sysctl.h>
11 #include <linux/module.h>
12 #include <linux/socket.h>
13 #include <linux/netdevice.h>
14 #include <linux/ratelimit.h>
15 #include <linux/vmalloc.h>
16 #include <linux/init.h>
17 #include <linux/slab.h>
18 #include <linux/kmemleak.h>
19 
20 #include <net/ip.h>
21 #include <net/sock.h>
22 #include <net/net_ratelimit.h>
23 #include <net/busy_poll.h>
24 #include <net/pkt_sched.h>
25 
26 static int zero = 0;
27 static int one = 1;
28 static int min_sndbuf = SOCK_MIN_SNDBUF;
29 static int min_rcvbuf = SOCK_MIN_RCVBUF;
30 static int max_skb_frags = MAX_SKB_FRAGS;
31 
32 static int net_msg_warn;	/* Unused, but still a sysctl */
33 
34 #ifdef CONFIG_RPS
35 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
36 				void __user *buffer, size_t *lenp, loff_t *ppos)
37 {
38 	unsigned int orig_size, size;
39 	int ret, i;
40 	struct ctl_table tmp = {
41 		.data = &size,
42 		.maxlen = sizeof(size),
43 		.mode = table->mode
44 	};
45 	struct rps_sock_flow_table *orig_sock_table, *sock_table;
46 	static DEFINE_MUTEX(sock_flow_mutex);
47 
48 	mutex_lock(&sock_flow_mutex);
49 
50 	orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
51 					lockdep_is_held(&sock_flow_mutex));
52 	size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
53 
54 	ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
55 
56 	if (write) {
57 		if (size) {
58 			if (size > 1<<29) {
59 				/* Enforce limit to prevent overflow */
60 				mutex_unlock(&sock_flow_mutex);
61 				return -EINVAL;
62 			}
63 			size = roundup_pow_of_two(size);
64 			if (size != orig_size) {
65 				sock_table =
66 				    vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
67 				if (!sock_table) {
68 					mutex_unlock(&sock_flow_mutex);
69 					return -ENOMEM;
70 				}
71 				rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
72 				sock_table->mask = size - 1;
73 			} else
74 				sock_table = orig_sock_table;
75 
76 			for (i = 0; i < size; i++)
77 				sock_table->ents[i] = RPS_NO_CPU;
78 		} else
79 			sock_table = NULL;
80 
81 		if (sock_table != orig_sock_table) {
82 			rcu_assign_pointer(rps_sock_flow_table, sock_table);
83 			if (sock_table) {
84 				static_key_slow_inc(&rps_needed);
85 				static_key_slow_inc(&rfs_needed);
86 			}
87 			if (orig_sock_table) {
88 				static_key_slow_dec(&rps_needed);
89 				static_key_slow_dec(&rfs_needed);
90 				synchronize_rcu();
91 				vfree(orig_sock_table);
92 			}
93 		}
94 	}
95 
96 	mutex_unlock(&sock_flow_mutex);
97 
98 	return ret;
99 }
100 #endif /* CONFIG_RPS */
101 
102 #ifdef CONFIG_NET_FLOW_LIMIT
103 static DEFINE_MUTEX(flow_limit_update_mutex);
104 
105 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
106 				 void __user *buffer, size_t *lenp,
107 				 loff_t *ppos)
108 {
109 	struct sd_flow_limit *cur;
110 	struct softnet_data *sd;
111 	cpumask_var_t mask;
112 	int i, len, ret = 0;
113 
114 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
115 		return -ENOMEM;
116 
117 	if (write) {
118 		ret = cpumask_parse_user(buffer, *lenp, mask);
119 		if (ret)
120 			goto done;
121 
122 		mutex_lock(&flow_limit_update_mutex);
123 		len = sizeof(*cur) + netdev_flow_limit_table_len;
124 		for_each_possible_cpu(i) {
125 			sd = &per_cpu(softnet_data, i);
126 			cur = rcu_dereference_protected(sd->flow_limit,
127 				     lockdep_is_held(&flow_limit_update_mutex));
128 			if (cur && !cpumask_test_cpu(i, mask)) {
129 				RCU_INIT_POINTER(sd->flow_limit, NULL);
130 				synchronize_rcu();
131 				kfree(cur);
132 			} else if (!cur && cpumask_test_cpu(i, mask)) {
133 				cur = kzalloc_node(len, GFP_KERNEL,
134 						   cpu_to_node(i));
135 				if (!cur) {
136 					/* not unwinding previous changes */
137 					ret = -ENOMEM;
138 					goto write_unlock;
139 				}
140 				cur->num_buckets = netdev_flow_limit_table_len;
141 				rcu_assign_pointer(sd->flow_limit, cur);
142 			}
143 		}
144 write_unlock:
145 		mutex_unlock(&flow_limit_update_mutex);
146 	} else {
147 		char kbuf[128];
148 
149 		if (*ppos || !*lenp) {
150 			*lenp = 0;
151 			goto done;
152 		}
153 
154 		cpumask_clear(mask);
155 		rcu_read_lock();
156 		for_each_possible_cpu(i) {
157 			sd = &per_cpu(softnet_data, i);
158 			if (rcu_dereference(sd->flow_limit))
159 				cpumask_set_cpu(i, mask);
160 		}
161 		rcu_read_unlock();
162 
163 		len = min(sizeof(kbuf) - 1, *lenp);
164 		len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
165 		if (!len) {
166 			*lenp = 0;
167 			goto done;
168 		}
169 		if (len < *lenp)
170 			kbuf[len++] = '\n';
171 		if (copy_to_user(buffer, kbuf, len)) {
172 			ret = -EFAULT;
173 			goto done;
174 		}
175 		*lenp = len;
176 		*ppos += len;
177 	}
178 
179 done:
180 	free_cpumask_var(mask);
181 	return ret;
182 }
183 
184 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
185 				       void __user *buffer, size_t *lenp,
186 				       loff_t *ppos)
187 {
188 	unsigned int old, *ptr;
189 	int ret;
190 
191 	mutex_lock(&flow_limit_update_mutex);
192 
193 	ptr = table->data;
194 	old = *ptr;
195 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
196 	if (!ret && write && !is_power_of_2(*ptr)) {
197 		*ptr = old;
198 		ret = -EINVAL;
199 	}
200 
201 	mutex_unlock(&flow_limit_update_mutex);
202 	return ret;
203 }
204 #endif /* CONFIG_NET_FLOW_LIMIT */
205 
206 #ifdef CONFIG_NET_SCHED
207 static int set_default_qdisc(struct ctl_table *table, int write,
208 			     void __user *buffer, size_t *lenp, loff_t *ppos)
209 {
210 	char id[IFNAMSIZ];
211 	struct ctl_table tbl = {
212 		.data = id,
213 		.maxlen = IFNAMSIZ,
214 	};
215 	int ret;
216 
217 	qdisc_get_default(id, IFNAMSIZ);
218 
219 	ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
220 	if (write && ret == 0)
221 		ret = qdisc_set_default(id);
222 	return ret;
223 }
224 #endif
225 
226 static int proc_do_dev_weight(struct ctl_table *table, int write,
227 			   void __user *buffer, size_t *lenp, loff_t *ppos)
228 {
229 	int ret;
230 
231 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
232 	if (ret != 0)
233 		return ret;
234 
235 	dev_rx_weight = weight_p * dev_weight_rx_bias;
236 	dev_tx_weight = weight_p * dev_weight_tx_bias;
237 
238 	return ret;
239 }
240 
241 static int proc_do_rss_key(struct ctl_table *table, int write,
242 			   void __user *buffer, size_t *lenp, loff_t *ppos)
243 {
244 	struct ctl_table fake_table;
245 	char buf[NETDEV_RSS_KEY_LEN * 3];
246 
247 	snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
248 	fake_table.data = buf;
249 	fake_table.maxlen = sizeof(buf);
250 	return proc_dostring(&fake_table, write, buffer, lenp, ppos);
251 }
252 
253 static struct ctl_table net_core_table[] = {
254 #ifdef CONFIG_NET
255 	{
256 		.procname	= "wmem_max",
257 		.data		= &sysctl_wmem_max,
258 		.maxlen		= sizeof(int),
259 		.mode		= 0644,
260 		.proc_handler	= proc_dointvec_minmax,
261 		.extra1		= &min_sndbuf,
262 	},
263 	{
264 		.procname	= "rmem_max",
265 		.data		= &sysctl_rmem_max,
266 		.maxlen		= sizeof(int),
267 		.mode		= 0644,
268 		.proc_handler	= proc_dointvec_minmax,
269 		.extra1		= &min_rcvbuf,
270 	},
271 	{
272 		.procname	= "wmem_default",
273 		.data		= &sysctl_wmem_default,
274 		.maxlen		= sizeof(int),
275 		.mode		= 0644,
276 		.proc_handler	= proc_dointvec_minmax,
277 		.extra1		= &min_sndbuf,
278 	},
279 	{
280 		.procname	= "rmem_default",
281 		.data		= &sysctl_rmem_default,
282 		.maxlen		= sizeof(int),
283 		.mode		= 0644,
284 		.proc_handler	= proc_dointvec_minmax,
285 		.extra1		= &min_rcvbuf,
286 	},
287 	{
288 		.procname	= "dev_weight",
289 		.data		= &weight_p,
290 		.maxlen		= sizeof(int),
291 		.mode		= 0644,
292 		.proc_handler	= proc_do_dev_weight,
293 	},
294 	{
295 		.procname	= "dev_weight_rx_bias",
296 		.data		= &dev_weight_rx_bias,
297 		.maxlen		= sizeof(int),
298 		.mode		= 0644,
299 		.proc_handler	= proc_do_dev_weight,
300 	},
301 	{
302 		.procname	= "dev_weight_tx_bias",
303 		.data		= &dev_weight_tx_bias,
304 		.maxlen		= sizeof(int),
305 		.mode		= 0644,
306 		.proc_handler	= proc_do_dev_weight,
307 	},
308 	{
309 		.procname	= "netdev_max_backlog",
310 		.data		= &netdev_max_backlog,
311 		.maxlen		= sizeof(int),
312 		.mode		= 0644,
313 		.proc_handler	= proc_dointvec
314 	},
315 	{
316 		.procname	= "netdev_rss_key",
317 		.data		= &netdev_rss_key,
318 		.maxlen		= sizeof(int),
319 		.mode		= 0444,
320 		.proc_handler	= proc_do_rss_key,
321 	},
322 #ifdef CONFIG_BPF_JIT
323 	{
324 		.procname	= "bpf_jit_enable",
325 		.data		= &bpf_jit_enable,
326 		.maxlen		= sizeof(int),
327 		.mode		= 0644,
328 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
329 		.proc_handler	= proc_dointvec
330 #else
331 		.proc_handler	= proc_dointvec_minmax,
332 		.extra1		= &one,
333 		.extra2		= &one,
334 #endif
335 	},
336 # ifdef CONFIG_HAVE_EBPF_JIT
337 	{
338 		.procname	= "bpf_jit_harden",
339 		.data		= &bpf_jit_harden,
340 		.maxlen		= sizeof(int),
341 		.mode		= 0600,
342 		.proc_handler	= proc_dointvec,
343 	},
344 	{
345 		.procname	= "bpf_jit_kallsyms",
346 		.data		= &bpf_jit_kallsyms,
347 		.maxlen		= sizeof(int),
348 		.mode		= 0600,
349 		.proc_handler	= proc_dointvec,
350 	},
351 # endif
352 #endif
353 	{
354 		.procname	= "netdev_tstamp_prequeue",
355 		.data		= &netdev_tstamp_prequeue,
356 		.maxlen		= sizeof(int),
357 		.mode		= 0644,
358 		.proc_handler	= proc_dointvec
359 	},
360 	{
361 		.procname	= "message_cost",
362 		.data		= &net_ratelimit_state.interval,
363 		.maxlen		= sizeof(int),
364 		.mode		= 0644,
365 		.proc_handler	= proc_dointvec_jiffies,
366 	},
367 	{
368 		.procname	= "message_burst",
369 		.data		= &net_ratelimit_state.burst,
370 		.maxlen		= sizeof(int),
371 		.mode		= 0644,
372 		.proc_handler	= proc_dointvec,
373 	},
374 	{
375 		.procname	= "optmem_max",
376 		.data		= &sysctl_optmem_max,
377 		.maxlen		= sizeof(int),
378 		.mode		= 0644,
379 		.proc_handler	= proc_dointvec
380 	},
381 	{
382 		.procname	= "tstamp_allow_data",
383 		.data		= &sysctl_tstamp_allow_data,
384 		.maxlen		= sizeof(int),
385 		.mode		= 0644,
386 		.proc_handler	= proc_dointvec_minmax,
387 		.extra1		= &zero,
388 		.extra2		= &one
389 	},
390 #ifdef CONFIG_RPS
391 	{
392 		.procname	= "rps_sock_flow_entries",
393 		.maxlen		= sizeof(int),
394 		.mode		= 0644,
395 		.proc_handler	= rps_sock_flow_sysctl
396 	},
397 #endif
398 #ifdef CONFIG_NET_FLOW_LIMIT
399 	{
400 		.procname	= "flow_limit_cpu_bitmap",
401 		.mode		= 0644,
402 		.proc_handler	= flow_limit_cpu_sysctl
403 	},
404 	{
405 		.procname	= "flow_limit_table_len",
406 		.data		= &netdev_flow_limit_table_len,
407 		.maxlen		= sizeof(int),
408 		.mode		= 0644,
409 		.proc_handler	= flow_limit_table_len_sysctl
410 	},
411 #endif /* CONFIG_NET_FLOW_LIMIT */
412 #ifdef CONFIG_NET_RX_BUSY_POLL
413 	{
414 		.procname	= "busy_poll",
415 		.data		= &sysctl_net_busy_poll,
416 		.maxlen		= sizeof(unsigned int),
417 		.mode		= 0644,
418 		.proc_handler	= proc_dointvec_minmax,
419 		.extra1		= &zero,
420 	},
421 	{
422 		.procname	= "busy_read",
423 		.data		= &sysctl_net_busy_read,
424 		.maxlen		= sizeof(unsigned int),
425 		.mode		= 0644,
426 		.proc_handler	= proc_dointvec_minmax,
427 		.extra1		= &zero,
428 	},
429 #endif
430 #ifdef CONFIG_NET_SCHED
431 	{
432 		.procname	= "default_qdisc",
433 		.mode		= 0644,
434 		.maxlen		= IFNAMSIZ,
435 		.proc_handler	= set_default_qdisc
436 	},
437 #endif
438 #endif /* CONFIG_NET */
439 	{
440 		.procname	= "netdev_budget",
441 		.data		= &netdev_budget,
442 		.maxlen		= sizeof(int),
443 		.mode		= 0644,
444 		.proc_handler	= proc_dointvec
445 	},
446 	{
447 		.procname	= "warnings",
448 		.data		= &net_msg_warn,
449 		.maxlen		= sizeof(int),
450 		.mode		= 0644,
451 		.proc_handler	= proc_dointvec
452 	},
453 	{
454 		.procname	= "max_skb_frags",
455 		.data		= &sysctl_max_skb_frags,
456 		.maxlen		= sizeof(int),
457 		.mode		= 0644,
458 		.proc_handler	= proc_dointvec_minmax,
459 		.extra1		= &one,
460 		.extra2		= &max_skb_frags,
461 	},
462 	{
463 		.procname	= "netdev_budget_usecs",
464 		.data		= &netdev_budget_usecs,
465 		.maxlen		= sizeof(unsigned int),
466 		.mode		= 0644,
467 		.proc_handler	= proc_dointvec_minmax,
468 		.extra1		= &zero,
469 	},
470 	{ }
471 };
472 
473 static struct ctl_table netns_core_table[] = {
474 	{
475 		.procname	= "somaxconn",
476 		.data		= &init_net.core.sysctl_somaxconn,
477 		.maxlen		= sizeof(int),
478 		.mode		= 0644,
479 		.extra1		= &zero,
480 		.proc_handler	= proc_dointvec_minmax
481 	},
482 	{ }
483 };
484 
485 static __net_init int sysctl_core_net_init(struct net *net)
486 {
487 	struct ctl_table *tbl;
488 
489 	tbl = netns_core_table;
490 	if (!net_eq(net, &init_net)) {
491 		tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
492 		if (tbl == NULL)
493 			goto err_dup;
494 
495 		tbl[0].data = &net->core.sysctl_somaxconn;
496 
497 		/* Don't export any sysctls to unprivileged users */
498 		if (net->user_ns != &init_user_ns) {
499 			tbl[0].procname = NULL;
500 		}
501 	}
502 
503 	net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
504 	if (net->core.sysctl_hdr == NULL)
505 		goto err_reg;
506 
507 	return 0;
508 
509 err_reg:
510 	if (tbl != netns_core_table)
511 		kfree(tbl);
512 err_dup:
513 	return -ENOMEM;
514 }
515 
516 static __net_exit void sysctl_core_net_exit(struct net *net)
517 {
518 	struct ctl_table *tbl;
519 
520 	tbl = net->core.sysctl_hdr->ctl_table_arg;
521 	unregister_net_sysctl_table(net->core.sysctl_hdr);
522 	BUG_ON(tbl == netns_core_table);
523 	kfree(tbl);
524 }
525 
526 static __net_initdata struct pernet_operations sysctl_core_ops = {
527 	.init = sysctl_core_net_init,
528 	.exit = sysctl_core_net_exit,
529 };
530 
531 static __init int sysctl_core_init(void)
532 {
533 	register_net_sysctl(&init_net, "net/core", net_core_table);
534 	return register_pernet_subsys(&sysctl_core_ops);
535 }
536 
537 fs_initcall(sysctl_core_init);
538