xref: /linux/net/core/sysctl_net_core.c (revision 2dbc0838bcf24ca59cabc3130cf3b1d6809cdcd4)
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 
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 two __maybe_unused = 2;
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 static long long_one __maybe_unused = 1;
32 static long long_max __maybe_unused = LONG_MAX;
33 
34 static int net_msg_warn;	/* Unused, but still a sysctl */
35 
36 int sysctl_fb_tunnels_only_for_init_net __read_mostly = 0;
37 EXPORT_SYMBOL(sysctl_fb_tunnels_only_for_init_net);
38 
39 /* 0 - Keep current behavior:
40  *     IPv4: inherit all current settings from init_net
41  *     IPv6: reset all settings to default
42  * 1 - Both inherit all current settings from init_net
43  * 2 - Both reset all settings to default
44  */
45 int sysctl_devconf_inherit_init_net __read_mostly;
46 EXPORT_SYMBOL(sysctl_devconf_inherit_init_net);
47 
48 #ifdef CONFIG_RPS
49 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
50 				void __user *buffer, size_t *lenp, loff_t *ppos)
51 {
52 	unsigned int orig_size, size;
53 	int ret, i;
54 	struct ctl_table tmp = {
55 		.data = &size,
56 		.maxlen = sizeof(size),
57 		.mode = table->mode
58 	};
59 	struct rps_sock_flow_table *orig_sock_table, *sock_table;
60 	static DEFINE_MUTEX(sock_flow_mutex);
61 
62 	mutex_lock(&sock_flow_mutex);
63 
64 	orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
65 					lockdep_is_held(&sock_flow_mutex));
66 	size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
67 
68 	ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
69 
70 	if (write) {
71 		if (size) {
72 			if (size > 1<<29) {
73 				/* Enforce limit to prevent overflow */
74 				mutex_unlock(&sock_flow_mutex);
75 				return -EINVAL;
76 			}
77 			size = roundup_pow_of_two(size);
78 			if (size != orig_size) {
79 				sock_table =
80 				    vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
81 				if (!sock_table) {
82 					mutex_unlock(&sock_flow_mutex);
83 					return -ENOMEM;
84 				}
85 				rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
86 				sock_table->mask = size - 1;
87 			} else
88 				sock_table = orig_sock_table;
89 
90 			for (i = 0; i < size; i++)
91 				sock_table->ents[i] = RPS_NO_CPU;
92 		} else
93 			sock_table = NULL;
94 
95 		if (sock_table != orig_sock_table) {
96 			rcu_assign_pointer(rps_sock_flow_table, sock_table);
97 			if (sock_table) {
98 				static_branch_inc(&rps_needed);
99 				static_branch_inc(&rfs_needed);
100 			}
101 			if (orig_sock_table) {
102 				static_branch_dec(&rps_needed);
103 				static_branch_dec(&rfs_needed);
104 				synchronize_rcu();
105 				vfree(orig_sock_table);
106 			}
107 		}
108 	}
109 
110 	mutex_unlock(&sock_flow_mutex);
111 
112 	return ret;
113 }
114 #endif /* CONFIG_RPS */
115 
116 #ifdef CONFIG_NET_FLOW_LIMIT
117 static DEFINE_MUTEX(flow_limit_update_mutex);
118 
119 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
120 				 void __user *buffer, size_t *lenp,
121 				 loff_t *ppos)
122 {
123 	struct sd_flow_limit *cur;
124 	struct softnet_data *sd;
125 	cpumask_var_t mask;
126 	int i, len, ret = 0;
127 
128 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
129 		return -ENOMEM;
130 
131 	if (write) {
132 		ret = cpumask_parse_user(buffer, *lenp, mask);
133 		if (ret)
134 			goto done;
135 
136 		mutex_lock(&flow_limit_update_mutex);
137 		len = sizeof(*cur) + netdev_flow_limit_table_len;
138 		for_each_possible_cpu(i) {
139 			sd = &per_cpu(softnet_data, i);
140 			cur = rcu_dereference_protected(sd->flow_limit,
141 				     lockdep_is_held(&flow_limit_update_mutex));
142 			if (cur && !cpumask_test_cpu(i, mask)) {
143 				RCU_INIT_POINTER(sd->flow_limit, NULL);
144 				synchronize_rcu();
145 				kfree(cur);
146 			} else if (!cur && cpumask_test_cpu(i, mask)) {
147 				cur = kzalloc_node(len, GFP_KERNEL,
148 						   cpu_to_node(i));
149 				if (!cur) {
150 					/* not unwinding previous changes */
151 					ret = -ENOMEM;
152 					goto write_unlock;
153 				}
154 				cur->num_buckets = netdev_flow_limit_table_len;
155 				rcu_assign_pointer(sd->flow_limit, cur);
156 			}
157 		}
158 write_unlock:
159 		mutex_unlock(&flow_limit_update_mutex);
160 	} else {
161 		char kbuf[128];
162 
163 		if (*ppos || !*lenp) {
164 			*lenp = 0;
165 			goto done;
166 		}
167 
168 		cpumask_clear(mask);
169 		rcu_read_lock();
170 		for_each_possible_cpu(i) {
171 			sd = &per_cpu(softnet_data, i);
172 			if (rcu_dereference(sd->flow_limit))
173 				cpumask_set_cpu(i, mask);
174 		}
175 		rcu_read_unlock();
176 
177 		len = min(sizeof(kbuf) - 1, *lenp);
178 		len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
179 		if (!len) {
180 			*lenp = 0;
181 			goto done;
182 		}
183 		if (len < *lenp)
184 			kbuf[len++] = '\n';
185 		if (copy_to_user(buffer, kbuf, len)) {
186 			ret = -EFAULT;
187 			goto done;
188 		}
189 		*lenp = len;
190 		*ppos += len;
191 	}
192 
193 done:
194 	free_cpumask_var(mask);
195 	return ret;
196 }
197 
198 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
199 				       void __user *buffer, size_t *lenp,
200 				       loff_t *ppos)
201 {
202 	unsigned int old, *ptr;
203 	int ret;
204 
205 	mutex_lock(&flow_limit_update_mutex);
206 
207 	ptr = table->data;
208 	old = *ptr;
209 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
210 	if (!ret && write && !is_power_of_2(*ptr)) {
211 		*ptr = old;
212 		ret = -EINVAL;
213 	}
214 
215 	mutex_unlock(&flow_limit_update_mutex);
216 	return ret;
217 }
218 #endif /* CONFIG_NET_FLOW_LIMIT */
219 
220 #ifdef CONFIG_NET_SCHED
221 static int set_default_qdisc(struct ctl_table *table, int write,
222 			     void __user *buffer, size_t *lenp, loff_t *ppos)
223 {
224 	char id[IFNAMSIZ];
225 	struct ctl_table tbl = {
226 		.data = id,
227 		.maxlen = IFNAMSIZ,
228 	};
229 	int ret;
230 
231 	qdisc_get_default(id, IFNAMSIZ);
232 
233 	ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
234 	if (write && ret == 0)
235 		ret = qdisc_set_default(id);
236 	return ret;
237 }
238 #endif
239 
240 static int proc_do_dev_weight(struct ctl_table *table, int write,
241 			   void __user *buffer, size_t *lenp, loff_t *ppos)
242 {
243 	int ret;
244 
245 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
246 	if (ret != 0)
247 		return ret;
248 
249 	dev_rx_weight = weight_p * dev_weight_rx_bias;
250 	dev_tx_weight = weight_p * dev_weight_tx_bias;
251 
252 	return ret;
253 }
254 
255 static int proc_do_rss_key(struct ctl_table *table, int write,
256 			   void __user *buffer, size_t *lenp, loff_t *ppos)
257 {
258 	struct ctl_table fake_table;
259 	char buf[NETDEV_RSS_KEY_LEN * 3];
260 
261 	snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
262 	fake_table.data = buf;
263 	fake_table.maxlen = sizeof(buf);
264 	return proc_dostring(&fake_table, write, buffer, lenp, ppos);
265 }
266 
267 #ifdef CONFIG_BPF_JIT
268 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write,
269 					   void __user *buffer, size_t *lenp,
270 					   loff_t *ppos)
271 {
272 	int ret, jit_enable = *(int *)table->data;
273 	struct ctl_table tmp = *table;
274 
275 	if (write && !capable(CAP_SYS_ADMIN))
276 		return -EPERM;
277 
278 	tmp.data = &jit_enable;
279 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
280 	if (write && !ret) {
281 		if (jit_enable < 2 ||
282 		    (jit_enable == 2 && bpf_dump_raw_ok())) {
283 			*(int *)table->data = jit_enable;
284 			if (jit_enable == 2)
285 				pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
286 		} else {
287 			ret = -EPERM;
288 		}
289 	}
290 	return ret;
291 }
292 
293 static int
294 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write,
295 				    void __user *buffer, size_t *lenp,
296 				    loff_t *ppos)
297 {
298 	if (!capable(CAP_SYS_ADMIN))
299 		return -EPERM;
300 
301 	return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
302 }
303 
304 static int
305 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
306 				     void __user *buffer, size_t *lenp,
307 				     loff_t *ppos)
308 {
309 	if (!capable(CAP_SYS_ADMIN))
310 		return -EPERM;
311 
312 	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
313 }
314 #endif
315 
316 static struct ctl_table net_core_table[] = {
317 #ifdef CONFIG_NET
318 	{
319 		.procname	= "wmem_max",
320 		.data		= &sysctl_wmem_max,
321 		.maxlen		= sizeof(int),
322 		.mode		= 0644,
323 		.proc_handler	= proc_dointvec_minmax,
324 		.extra1		= &min_sndbuf,
325 	},
326 	{
327 		.procname	= "rmem_max",
328 		.data		= &sysctl_rmem_max,
329 		.maxlen		= sizeof(int),
330 		.mode		= 0644,
331 		.proc_handler	= proc_dointvec_minmax,
332 		.extra1		= &min_rcvbuf,
333 	},
334 	{
335 		.procname	= "wmem_default",
336 		.data		= &sysctl_wmem_default,
337 		.maxlen		= sizeof(int),
338 		.mode		= 0644,
339 		.proc_handler	= proc_dointvec_minmax,
340 		.extra1		= &min_sndbuf,
341 	},
342 	{
343 		.procname	= "rmem_default",
344 		.data		= &sysctl_rmem_default,
345 		.maxlen		= sizeof(int),
346 		.mode		= 0644,
347 		.proc_handler	= proc_dointvec_minmax,
348 		.extra1		= &min_rcvbuf,
349 	},
350 	{
351 		.procname	= "dev_weight",
352 		.data		= &weight_p,
353 		.maxlen		= sizeof(int),
354 		.mode		= 0644,
355 		.proc_handler	= proc_do_dev_weight,
356 	},
357 	{
358 		.procname	= "dev_weight_rx_bias",
359 		.data		= &dev_weight_rx_bias,
360 		.maxlen		= sizeof(int),
361 		.mode		= 0644,
362 		.proc_handler	= proc_do_dev_weight,
363 	},
364 	{
365 		.procname	= "dev_weight_tx_bias",
366 		.data		= &dev_weight_tx_bias,
367 		.maxlen		= sizeof(int),
368 		.mode		= 0644,
369 		.proc_handler	= proc_do_dev_weight,
370 	},
371 	{
372 		.procname	= "netdev_max_backlog",
373 		.data		= &netdev_max_backlog,
374 		.maxlen		= sizeof(int),
375 		.mode		= 0644,
376 		.proc_handler	= proc_dointvec
377 	},
378 	{
379 		.procname	= "netdev_rss_key",
380 		.data		= &netdev_rss_key,
381 		.maxlen		= sizeof(int),
382 		.mode		= 0444,
383 		.proc_handler	= proc_do_rss_key,
384 	},
385 #ifdef CONFIG_BPF_JIT
386 	{
387 		.procname	= "bpf_jit_enable",
388 		.data		= &bpf_jit_enable,
389 		.maxlen		= sizeof(int),
390 		.mode		= 0644,
391 		.proc_handler	= proc_dointvec_minmax_bpf_enable,
392 # ifdef CONFIG_BPF_JIT_ALWAYS_ON
393 		.extra1		= &one,
394 		.extra2		= &one,
395 # else
396 		.extra1		= &zero,
397 		.extra2		= &two,
398 # endif
399 	},
400 # ifdef CONFIG_HAVE_EBPF_JIT
401 	{
402 		.procname	= "bpf_jit_harden",
403 		.data		= &bpf_jit_harden,
404 		.maxlen		= sizeof(int),
405 		.mode		= 0600,
406 		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
407 		.extra1		= &zero,
408 		.extra2		= &two,
409 	},
410 	{
411 		.procname	= "bpf_jit_kallsyms",
412 		.data		= &bpf_jit_kallsyms,
413 		.maxlen		= sizeof(int),
414 		.mode		= 0600,
415 		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
416 		.extra1		= &zero,
417 		.extra2		= &one,
418 	},
419 # endif
420 	{
421 		.procname	= "bpf_jit_limit",
422 		.data		= &bpf_jit_limit,
423 		.maxlen		= sizeof(long),
424 		.mode		= 0600,
425 		.proc_handler	= proc_dolongvec_minmax_bpf_restricted,
426 		.extra1		= &long_one,
427 		.extra2		= &long_max,
428 	},
429 #endif
430 	{
431 		.procname	= "netdev_tstamp_prequeue",
432 		.data		= &netdev_tstamp_prequeue,
433 		.maxlen		= sizeof(int),
434 		.mode		= 0644,
435 		.proc_handler	= proc_dointvec
436 	},
437 	{
438 		.procname	= "message_cost",
439 		.data		= &net_ratelimit_state.interval,
440 		.maxlen		= sizeof(int),
441 		.mode		= 0644,
442 		.proc_handler	= proc_dointvec_jiffies,
443 	},
444 	{
445 		.procname	= "message_burst",
446 		.data		= &net_ratelimit_state.burst,
447 		.maxlen		= sizeof(int),
448 		.mode		= 0644,
449 		.proc_handler	= proc_dointvec,
450 	},
451 	{
452 		.procname	= "optmem_max",
453 		.data		= &sysctl_optmem_max,
454 		.maxlen		= sizeof(int),
455 		.mode		= 0644,
456 		.proc_handler	= proc_dointvec
457 	},
458 	{
459 		.procname	= "tstamp_allow_data",
460 		.data		= &sysctl_tstamp_allow_data,
461 		.maxlen		= sizeof(int),
462 		.mode		= 0644,
463 		.proc_handler	= proc_dointvec_minmax,
464 		.extra1		= &zero,
465 		.extra2		= &one
466 	},
467 #ifdef CONFIG_RPS
468 	{
469 		.procname	= "rps_sock_flow_entries",
470 		.maxlen		= sizeof(int),
471 		.mode		= 0644,
472 		.proc_handler	= rps_sock_flow_sysctl
473 	},
474 #endif
475 #ifdef CONFIG_NET_FLOW_LIMIT
476 	{
477 		.procname	= "flow_limit_cpu_bitmap",
478 		.mode		= 0644,
479 		.proc_handler	= flow_limit_cpu_sysctl
480 	},
481 	{
482 		.procname	= "flow_limit_table_len",
483 		.data		= &netdev_flow_limit_table_len,
484 		.maxlen		= sizeof(int),
485 		.mode		= 0644,
486 		.proc_handler	= flow_limit_table_len_sysctl
487 	},
488 #endif /* CONFIG_NET_FLOW_LIMIT */
489 #ifdef CONFIG_NET_RX_BUSY_POLL
490 	{
491 		.procname	= "busy_poll",
492 		.data		= &sysctl_net_busy_poll,
493 		.maxlen		= sizeof(unsigned int),
494 		.mode		= 0644,
495 		.proc_handler	= proc_dointvec_minmax,
496 		.extra1		= &zero,
497 	},
498 	{
499 		.procname	= "busy_read",
500 		.data		= &sysctl_net_busy_read,
501 		.maxlen		= sizeof(unsigned int),
502 		.mode		= 0644,
503 		.proc_handler	= proc_dointvec_minmax,
504 		.extra1		= &zero,
505 	},
506 #endif
507 #ifdef CONFIG_NET_SCHED
508 	{
509 		.procname	= "default_qdisc",
510 		.mode		= 0644,
511 		.maxlen		= IFNAMSIZ,
512 		.proc_handler	= set_default_qdisc
513 	},
514 #endif
515 #endif /* CONFIG_NET */
516 	{
517 		.procname	= "netdev_budget",
518 		.data		= &netdev_budget,
519 		.maxlen		= sizeof(int),
520 		.mode		= 0644,
521 		.proc_handler	= proc_dointvec
522 	},
523 	{
524 		.procname	= "warnings",
525 		.data		= &net_msg_warn,
526 		.maxlen		= sizeof(int),
527 		.mode		= 0644,
528 		.proc_handler	= proc_dointvec
529 	},
530 	{
531 		.procname	= "max_skb_frags",
532 		.data		= &sysctl_max_skb_frags,
533 		.maxlen		= sizeof(int),
534 		.mode		= 0644,
535 		.proc_handler	= proc_dointvec_minmax,
536 		.extra1		= &one,
537 		.extra2		= &max_skb_frags,
538 	},
539 	{
540 		.procname	= "netdev_budget_usecs",
541 		.data		= &netdev_budget_usecs,
542 		.maxlen		= sizeof(unsigned int),
543 		.mode		= 0644,
544 		.proc_handler	= proc_dointvec_minmax,
545 		.extra1		= &zero,
546 	},
547 	{
548 		.procname	= "fb_tunnels_only_for_init_net",
549 		.data		= &sysctl_fb_tunnels_only_for_init_net,
550 		.maxlen		= sizeof(int),
551 		.mode		= 0644,
552 		.proc_handler	= proc_dointvec_minmax,
553 		.extra1		= &zero,
554 		.extra2		= &one,
555 	},
556 	{
557 		.procname	= "devconf_inherit_init_net",
558 		.data		= &sysctl_devconf_inherit_init_net,
559 		.maxlen		= sizeof(int),
560 		.mode		= 0644,
561 		.proc_handler	= proc_dointvec_minmax,
562 		.extra1		= &zero,
563 		.extra2		= &two,
564 	},
565 	{
566 		.procname	= "high_order_alloc_disable",
567 		.data		= &net_high_order_alloc_disable_key.key,
568 		.maxlen         = sizeof(net_high_order_alloc_disable_key),
569 		.mode		= 0644,
570 		.proc_handler	= proc_do_static_key,
571 	},
572 	{ }
573 };
574 
575 static struct ctl_table netns_core_table[] = {
576 	{
577 		.procname	= "somaxconn",
578 		.data		= &init_net.core.sysctl_somaxconn,
579 		.maxlen		= sizeof(int),
580 		.mode		= 0644,
581 		.extra1		= &zero,
582 		.proc_handler	= proc_dointvec_minmax
583 	},
584 	{ }
585 };
586 
587 static __net_init int sysctl_core_net_init(struct net *net)
588 {
589 	struct ctl_table *tbl;
590 
591 	tbl = netns_core_table;
592 	if (!net_eq(net, &init_net)) {
593 		tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
594 		if (tbl == NULL)
595 			goto err_dup;
596 
597 		tbl[0].data = &net->core.sysctl_somaxconn;
598 
599 		/* Don't export any sysctls to unprivileged users */
600 		if (net->user_ns != &init_user_ns) {
601 			tbl[0].procname = NULL;
602 		}
603 	}
604 
605 	net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
606 	if (net->core.sysctl_hdr == NULL)
607 		goto err_reg;
608 
609 	return 0;
610 
611 err_reg:
612 	if (tbl != netns_core_table)
613 		kfree(tbl);
614 err_dup:
615 	return -ENOMEM;
616 }
617 
618 static __net_exit void sysctl_core_net_exit(struct net *net)
619 {
620 	struct ctl_table *tbl;
621 
622 	tbl = net->core.sysctl_hdr->ctl_table_arg;
623 	unregister_net_sysctl_table(net->core.sysctl_hdr);
624 	BUG_ON(tbl == netns_core_table);
625 	kfree(tbl);
626 }
627 
628 static __net_initdata struct pernet_operations sysctl_core_ops = {
629 	.init = sysctl_core_net_init,
630 	.exit = sysctl_core_net_exit,
631 };
632 
633 static __init int sysctl_core_init(void)
634 {
635 	register_net_sysctl(&init_net, "net/core", net_core_table);
636 	return register_pernet_subsys(&sysctl_core_ops);
637 }
638 
639 fs_initcall(sysctl_core_init);
640