xref: /linux/net/core/sysctl_net_core.c (revision 3f41368fbfe1b3d5922d317fe1a0a0cab6846802)
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/filter.h>
10 #include <linux/mm.h>
11 #include <linux/sysctl.h>
12 #include <linux/module.h>
13 #include <linux/socket.h>
14 #include <linux/netdevice.h>
15 #include <linux/ratelimit.h>
16 #include <linux/vmalloc.h>
17 #include <linux/init.h>
18 #include <linux/slab.h>
19 #include <linux/sched/isolation.h>
20 
21 #include <net/ip.h>
22 #include <net/sock.h>
23 #include <net/net_ratelimit.h>
24 #include <net/busy_poll.h>
25 #include <net/pkt_sched.h>
26 #include <net/hotdata.h>
27 #include <net/proto_memory.h>
28 #include <net/rps.h>
29 
30 #include "dev.h"
31 
32 static int int_3600 = 3600;
33 static int min_sndbuf = SOCK_MIN_SNDBUF;
34 static int min_rcvbuf = SOCK_MIN_RCVBUF;
35 static int max_skb_frags = MAX_SKB_FRAGS;
36 static int min_mem_pcpu_rsv = SK_MEMORY_PCPU_RESERVE;
37 
38 static int net_msg_warn;	/* Unused, but still a sysctl */
39 
40 int sysctl_fb_tunnels_only_for_init_net __read_mostly = 0;
41 EXPORT_SYMBOL(sysctl_fb_tunnels_only_for_init_net);
42 
43 /* 0 - Keep current behavior:
44  *     IPv4: inherit all current settings from init_net
45  *     IPv6: reset all settings to default
46  * 1 - Both inherit all current settings from init_net
47  * 2 - Both reset all settings to default
48  * 3 - Both inherit all settings from current netns
49  */
50 int sysctl_devconf_inherit_init_net __read_mostly;
51 EXPORT_SYMBOL(sysctl_devconf_inherit_init_net);
52 
53 #if IS_ENABLED(CONFIG_NET_FLOW_LIMIT) || IS_ENABLED(CONFIG_RPS)
54 static void dump_cpumask(void *buffer, size_t *lenp, loff_t *ppos,
55 			 struct cpumask *mask)
56 {
57 	char kbuf[128];
58 	int len;
59 
60 	if (*ppos || !*lenp) {
61 		*lenp = 0;
62 		return;
63 	}
64 
65 	len = min(sizeof(kbuf) - 1, *lenp);
66 	len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
67 	if (!len) {
68 		*lenp = 0;
69 		return;
70 	}
71 
72 	if (len < *lenp)
73 		kbuf[len++] = '\n';
74 	memcpy(buffer, kbuf, len);
75 	*lenp = len;
76 	*ppos += len;
77 }
78 #endif
79 
80 #ifdef CONFIG_RPS
81 
82 static struct cpumask *rps_default_mask_cow_alloc(struct net *net)
83 {
84 	struct cpumask *rps_default_mask;
85 
86 	if (net->core.rps_default_mask)
87 		return net->core.rps_default_mask;
88 
89 	rps_default_mask = kzalloc(cpumask_size(), GFP_KERNEL);
90 	if (!rps_default_mask)
91 		return NULL;
92 
93 	/* pairs with READ_ONCE in rx_queue_default_mask() */
94 	WRITE_ONCE(net->core.rps_default_mask, rps_default_mask);
95 	return rps_default_mask;
96 }
97 
98 static int rps_default_mask_sysctl(struct ctl_table *table, int write,
99 				   void *buffer, size_t *lenp, loff_t *ppos)
100 {
101 	struct net *net = (struct net *)table->data;
102 	int err = 0;
103 
104 	rtnl_lock();
105 	if (write) {
106 		struct cpumask *rps_default_mask = rps_default_mask_cow_alloc(net);
107 
108 		err = -ENOMEM;
109 		if (!rps_default_mask)
110 			goto done;
111 
112 		err = cpumask_parse(buffer, rps_default_mask);
113 		if (err)
114 			goto done;
115 
116 		err = rps_cpumask_housekeeping(rps_default_mask);
117 		if (err)
118 			goto done;
119 	} else {
120 		dump_cpumask(buffer, lenp, ppos,
121 			     net->core.rps_default_mask ? : cpu_none_mask);
122 	}
123 
124 done:
125 	rtnl_unlock();
126 	return err;
127 }
128 
129 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
130 				void *buffer, size_t *lenp, loff_t *ppos)
131 {
132 	unsigned int orig_size, size;
133 	int ret, i;
134 	struct ctl_table tmp = {
135 		.data = &size,
136 		.maxlen = sizeof(size),
137 		.mode = table->mode
138 	};
139 	struct rps_sock_flow_table *orig_sock_table, *sock_table;
140 	static DEFINE_MUTEX(sock_flow_mutex);
141 
142 	mutex_lock(&sock_flow_mutex);
143 
144 	orig_sock_table = rcu_dereference_protected(
145 					net_hotdata.rps_sock_flow_table,
146 					lockdep_is_held(&sock_flow_mutex));
147 	size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
148 
149 	ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
150 
151 	if (write) {
152 		if (size) {
153 			if (size > 1<<29) {
154 				/* Enforce limit to prevent overflow */
155 				mutex_unlock(&sock_flow_mutex);
156 				return -EINVAL;
157 			}
158 			size = roundup_pow_of_two(size);
159 			if (size != orig_size) {
160 				sock_table =
161 				    vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
162 				if (!sock_table) {
163 					mutex_unlock(&sock_flow_mutex);
164 					return -ENOMEM;
165 				}
166 				net_hotdata.rps_cpu_mask =
167 					roundup_pow_of_two(nr_cpu_ids) - 1;
168 				sock_table->mask = size - 1;
169 			} else
170 				sock_table = orig_sock_table;
171 
172 			for (i = 0; i < size; i++)
173 				sock_table->ents[i] = RPS_NO_CPU;
174 		} else
175 			sock_table = NULL;
176 
177 		if (sock_table != orig_sock_table) {
178 			rcu_assign_pointer(net_hotdata.rps_sock_flow_table,
179 					   sock_table);
180 			if (sock_table) {
181 				static_branch_inc(&rps_needed);
182 				static_branch_inc(&rfs_needed);
183 			}
184 			if (orig_sock_table) {
185 				static_branch_dec(&rps_needed);
186 				static_branch_dec(&rfs_needed);
187 				kvfree_rcu_mightsleep(orig_sock_table);
188 			}
189 		}
190 	}
191 
192 	mutex_unlock(&sock_flow_mutex);
193 
194 	return ret;
195 }
196 #endif /* CONFIG_RPS */
197 
198 #ifdef CONFIG_NET_FLOW_LIMIT
199 static DEFINE_MUTEX(flow_limit_update_mutex);
200 
201 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
202 				 void *buffer, size_t *lenp, loff_t *ppos)
203 {
204 	struct sd_flow_limit *cur;
205 	struct softnet_data *sd;
206 	cpumask_var_t mask;
207 	int i, len, ret = 0;
208 
209 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
210 		return -ENOMEM;
211 
212 	if (write) {
213 		ret = cpumask_parse(buffer, mask);
214 		if (ret)
215 			goto done;
216 
217 		mutex_lock(&flow_limit_update_mutex);
218 		len = sizeof(*cur) + netdev_flow_limit_table_len;
219 		for_each_possible_cpu(i) {
220 			sd = &per_cpu(softnet_data, i);
221 			cur = rcu_dereference_protected(sd->flow_limit,
222 				     lockdep_is_held(&flow_limit_update_mutex));
223 			if (cur && !cpumask_test_cpu(i, mask)) {
224 				RCU_INIT_POINTER(sd->flow_limit, NULL);
225 				kfree_rcu_mightsleep(cur);
226 			} else if (!cur && cpumask_test_cpu(i, mask)) {
227 				cur = kzalloc_node(len, GFP_KERNEL,
228 						   cpu_to_node(i));
229 				if (!cur) {
230 					/* not unwinding previous changes */
231 					ret = -ENOMEM;
232 					goto write_unlock;
233 				}
234 				cur->num_buckets = netdev_flow_limit_table_len;
235 				rcu_assign_pointer(sd->flow_limit, cur);
236 			}
237 		}
238 write_unlock:
239 		mutex_unlock(&flow_limit_update_mutex);
240 	} else {
241 		cpumask_clear(mask);
242 		rcu_read_lock();
243 		for_each_possible_cpu(i) {
244 			sd = &per_cpu(softnet_data, i);
245 			if (rcu_dereference(sd->flow_limit))
246 				cpumask_set_cpu(i, mask);
247 		}
248 		rcu_read_unlock();
249 
250 		dump_cpumask(buffer, lenp, ppos, mask);
251 	}
252 
253 done:
254 	free_cpumask_var(mask);
255 	return ret;
256 }
257 
258 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
259 				       void *buffer, size_t *lenp, loff_t *ppos)
260 {
261 	unsigned int old, *ptr;
262 	int ret;
263 
264 	mutex_lock(&flow_limit_update_mutex);
265 
266 	ptr = table->data;
267 	old = *ptr;
268 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
269 	if (!ret && write && !is_power_of_2(*ptr)) {
270 		*ptr = old;
271 		ret = -EINVAL;
272 	}
273 
274 	mutex_unlock(&flow_limit_update_mutex);
275 	return ret;
276 }
277 #endif /* CONFIG_NET_FLOW_LIMIT */
278 
279 #ifdef CONFIG_NET_SCHED
280 static int set_default_qdisc(struct ctl_table *table, int write,
281 			     void *buffer, size_t *lenp, loff_t *ppos)
282 {
283 	char id[IFNAMSIZ];
284 	struct ctl_table tbl = {
285 		.data = id,
286 		.maxlen = IFNAMSIZ,
287 	};
288 	int ret;
289 
290 	qdisc_get_default(id, IFNAMSIZ);
291 
292 	ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
293 	if (write && ret == 0)
294 		ret = qdisc_set_default(id);
295 	return ret;
296 }
297 #endif
298 
299 static int proc_do_dev_weight(struct ctl_table *table, int write,
300 			   void *buffer, size_t *lenp, loff_t *ppos)
301 {
302 	static DEFINE_MUTEX(dev_weight_mutex);
303 	int ret, weight;
304 
305 	mutex_lock(&dev_weight_mutex);
306 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
307 	if (!ret && write) {
308 		weight = READ_ONCE(weight_p);
309 		WRITE_ONCE(net_hotdata.dev_rx_weight, weight * dev_weight_rx_bias);
310 		WRITE_ONCE(net_hotdata.dev_tx_weight, weight * dev_weight_tx_bias);
311 	}
312 	mutex_unlock(&dev_weight_mutex);
313 
314 	return ret;
315 }
316 
317 static int proc_do_rss_key(struct ctl_table *table, int write,
318 			   void *buffer, size_t *lenp, loff_t *ppos)
319 {
320 	struct ctl_table fake_table;
321 	char buf[NETDEV_RSS_KEY_LEN * 3];
322 
323 	snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
324 	fake_table.data = buf;
325 	fake_table.maxlen = sizeof(buf);
326 	return proc_dostring(&fake_table, write, buffer, lenp, ppos);
327 }
328 
329 #ifdef CONFIG_BPF_JIT
330 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write,
331 					   void *buffer, size_t *lenp,
332 					   loff_t *ppos)
333 {
334 	int ret, jit_enable = *(int *)table->data;
335 	int min = *(int *)table->extra1;
336 	int max = *(int *)table->extra2;
337 	struct ctl_table tmp = *table;
338 
339 	if (write && !capable(CAP_SYS_ADMIN))
340 		return -EPERM;
341 
342 	tmp.data = &jit_enable;
343 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
344 	if (write && !ret) {
345 		if (jit_enable < 2 ||
346 		    (jit_enable == 2 && bpf_dump_raw_ok(current_cred()))) {
347 			*(int *)table->data = jit_enable;
348 			if (jit_enable == 2)
349 				pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
350 		} else {
351 			ret = -EPERM;
352 		}
353 	}
354 
355 	if (write && ret && min == max)
356 		pr_info_once("CONFIG_BPF_JIT_ALWAYS_ON is enabled, bpf_jit_enable is permanently set to 1.\n");
357 
358 	return ret;
359 }
360 
361 # ifdef CONFIG_HAVE_EBPF_JIT
362 static int
363 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write,
364 				    void *buffer, size_t *lenp, loff_t *ppos)
365 {
366 	if (!capable(CAP_SYS_ADMIN))
367 		return -EPERM;
368 
369 	return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
370 }
371 # endif /* CONFIG_HAVE_EBPF_JIT */
372 
373 static int
374 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
375 				     void *buffer, size_t *lenp, loff_t *ppos)
376 {
377 	if (!capable(CAP_SYS_ADMIN))
378 		return -EPERM;
379 
380 	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
381 }
382 #endif
383 
384 static struct ctl_table net_core_table[] = {
385 	{
386 		.procname	= "wmem_max",
387 		.data		= &sysctl_wmem_max,
388 		.maxlen		= sizeof(int),
389 		.mode		= 0644,
390 		.proc_handler	= proc_dointvec_minmax,
391 		.extra1		= &min_sndbuf,
392 	},
393 	{
394 		.procname	= "rmem_max",
395 		.data		= &sysctl_rmem_max,
396 		.maxlen		= sizeof(int),
397 		.mode		= 0644,
398 		.proc_handler	= proc_dointvec_minmax,
399 		.extra1		= &min_rcvbuf,
400 	},
401 	{
402 		.procname	= "wmem_default",
403 		.data		= &sysctl_wmem_default,
404 		.maxlen		= sizeof(int),
405 		.mode		= 0644,
406 		.proc_handler	= proc_dointvec_minmax,
407 		.extra1		= &min_sndbuf,
408 	},
409 	{
410 		.procname	= "rmem_default",
411 		.data		= &sysctl_rmem_default,
412 		.maxlen		= sizeof(int),
413 		.mode		= 0644,
414 		.proc_handler	= proc_dointvec_minmax,
415 		.extra1		= &min_rcvbuf,
416 	},
417 	{
418 		.procname	= "mem_pcpu_rsv",
419 		.data		= &net_hotdata.sysctl_mem_pcpu_rsv,
420 		.maxlen		= sizeof(int),
421 		.mode		= 0644,
422 		.proc_handler	= proc_dointvec_minmax,
423 		.extra1		= &min_mem_pcpu_rsv,
424 	},
425 	{
426 		.procname	= "dev_weight",
427 		.data		= &weight_p,
428 		.maxlen		= sizeof(int),
429 		.mode		= 0644,
430 		.proc_handler	= proc_do_dev_weight,
431 	},
432 	{
433 		.procname	= "dev_weight_rx_bias",
434 		.data		= &dev_weight_rx_bias,
435 		.maxlen		= sizeof(int),
436 		.mode		= 0644,
437 		.proc_handler	= proc_do_dev_weight,
438 	},
439 	{
440 		.procname	= "dev_weight_tx_bias",
441 		.data		= &dev_weight_tx_bias,
442 		.maxlen		= sizeof(int),
443 		.mode		= 0644,
444 		.proc_handler	= proc_do_dev_weight,
445 	},
446 	{
447 		.procname	= "netdev_max_backlog",
448 		.data		= &net_hotdata.max_backlog,
449 		.maxlen		= sizeof(int),
450 		.mode		= 0644,
451 		.proc_handler	= proc_dointvec
452 	},
453 	{
454 		.procname	= "netdev_rss_key",
455 		.data		= &netdev_rss_key,
456 		.maxlen		= sizeof(int),
457 		.mode		= 0444,
458 		.proc_handler	= proc_do_rss_key,
459 	},
460 #ifdef CONFIG_BPF_JIT
461 	{
462 		.procname	= "bpf_jit_enable",
463 		.data		= &bpf_jit_enable,
464 		.maxlen		= sizeof(int),
465 		.mode		= 0644,
466 		.proc_handler	= proc_dointvec_minmax_bpf_enable,
467 # ifdef CONFIG_BPF_JIT_ALWAYS_ON
468 		.extra1		= SYSCTL_ONE,
469 		.extra2		= SYSCTL_ONE,
470 # else
471 		.extra1		= SYSCTL_ZERO,
472 		.extra2		= SYSCTL_TWO,
473 # endif
474 	},
475 # ifdef CONFIG_HAVE_EBPF_JIT
476 	{
477 		.procname	= "bpf_jit_harden",
478 		.data		= &bpf_jit_harden,
479 		.maxlen		= sizeof(int),
480 		.mode		= 0600,
481 		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
482 		.extra1		= SYSCTL_ZERO,
483 		.extra2		= SYSCTL_TWO,
484 	},
485 	{
486 		.procname	= "bpf_jit_kallsyms",
487 		.data		= &bpf_jit_kallsyms,
488 		.maxlen		= sizeof(int),
489 		.mode		= 0600,
490 		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
491 		.extra1		= SYSCTL_ZERO,
492 		.extra2		= SYSCTL_ONE,
493 	},
494 # endif
495 	{
496 		.procname	= "bpf_jit_limit",
497 		.data		= &bpf_jit_limit,
498 		.maxlen		= sizeof(long),
499 		.mode		= 0600,
500 		.proc_handler	= proc_dolongvec_minmax_bpf_restricted,
501 		.extra1		= SYSCTL_LONG_ONE,
502 		.extra2		= &bpf_jit_limit_max,
503 	},
504 #endif
505 	{
506 		.procname	= "netdev_tstamp_prequeue",
507 		.data		= &net_hotdata.tstamp_prequeue,
508 		.maxlen		= sizeof(int),
509 		.mode		= 0644,
510 		.proc_handler	= proc_dointvec
511 	},
512 	{
513 		.procname	= "message_cost",
514 		.data		= &net_ratelimit_state.interval,
515 		.maxlen		= sizeof(int),
516 		.mode		= 0644,
517 		.proc_handler	= proc_dointvec_jiffies,
518 	},
519 	{
520 		.procname	= "message_burst",
521 		.data		= &net_ratelimit_state.burst,
522 		.maxlen		= sizeof(int),
523 		.mode		= 0644,
524 		.proc_handler	= proc_dointvec,
525 	},
526 	{
527 		.procname	= "tstamp_allow_data",
528 		.data		= &sysctl_tstamp_allow_data,
529 		.maxlen		= sizeof(int),
530 		.mode		= 0644,
531 		.proc_handler	= proc_dointvec_minmax,
532 		.extra1		= SYSCTL_ZERO,
533 		.extra2		= SYSCTL_ONE
534 	},
535 #ifdef CONFIG_RPS
536 	{
537 		.procname	= "rps_sock_flow_entries",
538 		.maxlen		= sizeof(int),
539 		.mode		= 0644,
540 		.proc_handler	= rps_sock_flow_sysctl
541 	},
542 #endif
543 #ifdef CONFIG_NET_FLOW_LIMIT
544 	{
545 		.procname	= "flow_limit_cpu_bitmap",
546 		.mode		= 0644,
547 		.proc_handler	= flow_limit_cpu_sysctl
548 	},
549 	{
550 		.procname	= "flow_limit_table_len",
551 		.data		= &netdev_flow_limit_table_len,
552 		.maxlen		= sizeof(int),
553 		.mode		= 0644,
554 		.proc_handler	= flow_limit_table_len_sysctl
555 	},
556 #endif /* CONFIG_NET_FLOW_LIMIT */
557 #ifdef CONFIG_NET_RX_BUSY_POLL
558 	{
559 		.procname	= "busy_poll",
560 		.data		= &sysctl_net_busy_poll,
561 		.maxlen		= sizeof(unsigned int),
562 		.mode		= 0644,
563 		.proc_handler	= proc_dointvec_minmax,
564 		.extra1		= SYSCTL_ZERO,
565 	},
566 	{
567 		.procname	= "busy_read",
568 		.data		= &sysctl_net_busy_read,
569 		.maxlen		= sizeof(unsigned int),
570 		.mode		= 0644,
571 		.proc_handler	= proc_dointvec_minmax,
572 		.extra1		= SYSCTL_ZERO,
573 	},
574 #endif
575 #ifdef CONFIG_NET_SCHED
576 	{
577 		.procname	= "default_qdisc",
578 		.mode		= 0644,
579 		.maxlen		= IFNAMSIZ,
580 		.proc_handler	= set_default_qdisc
581 	},
582 #endif
583 	{
584 		.procname	= "netdev_budget",
585 		.data		= &net_hotdata.netdev_budget,
586 		.maxlen		= sizeof(int),
587 		.mode		= 0644,
588 		.proc_handler	= proc_dointvec
589 	},
590 	{
591 		.procname	= "warnings",
592 		.data		= &net_msg_warn,
593 		.maxlen		= sizeof(int),
594 		.mode		= 0644,
595 		.proc_handler	= proc_dointvec
596 	},
597 	{
598 		.procname	= "max_skb_frags",
599 		.data		= &net_hotdata.sysctl_max_skb_frags,
600 		.maxlen		= sizeof(int),
601 		.mode		= 0644,
602 		.proc_handler	= proc_dointvec_minmax,
603 		.extra1		= SYSCTL_ONE,
604 		.extra2		= &max_skb_frags,
605 	},
606 	{
607 		.procname	= "netdev_budget_usecs",
608 		.data		= &net_hotdata.netdev_budget_usecs,
609 		.maxlen		= sizeof(unsigned int),
610 		.mode		= 0644,
611 		.proc_handler	= proc_dointvec_minmax,
612 		.extra1		= SYSCTL_ZERO,
613 	},
614 	{
615 		.procname	= "fb_tunnels_only_for_init_net",
616 		.data		= &sysctl_fb_tunnels_only_for_init_net,
617 		.maxlen		= sizeof(int),
618 		.mode		= 0644,
619 		.proc_handler	= proc_dointvec_minmax,
620 		.extra1		= SYSCTL_ZERO,
621 		.extra2		= SYSCTL_TWO,
622 	},
623 	{
624 		.procname	= "devconf_inherit_init_net",
625 		.data		= &sysctl_devconf_inherit_init_net,
626 		.maxlen		= sizeof(int),
627 		.mode		= 0644,
628 		.proc_handler	= proc_dointvec_minmax,
629 		.extra1		= SYSCTL_ZERO,
630 		.extra2		= SYSCTL_THREE,
631 	},
632 	{
633 		.procname	= "high_order_alloc_disable",
634 		.data		= &net_high_order_alloc_disable_key.key,
635 		.maxlen         = sizeof(net_high_order_alloc_disable_key),
636 		.mode		= 0644,
637 		.proc_handler	= proc_do_static_key,
638 	},
639 	{
640 		.procname	= "gro_normal_batch",
641 		.data		= &net_hotdata.gro_normal_batch,
642 		.maxlen		= sizeof(unsigned int),
643 		.mode		= 0644,
644 		.proc_handler	= proc_dointvec_minmax,
645 		.extra1		= SYSCTL_ONE,
646 	},
647 	{
648 		.procname	= "netdev_unregister_timeout_secs",
649 		.data		= &netdev_unregister_timeout_secs,
650 		.maxlen		= sizeof(unsigned int),
651 		.mode		= 0644,
652 		.proc_handler	= proc_dointvec_minmax,
653 		.extra1		= SYSCTL_ONE,
654 		.extra2		= &int_3600,
655 	},
656 	{
657 		.procname	= "skb_defer_max",
658 		.data		= &net_hotdata.sysctl_skb_defer_max,
659 		.maxlen		= sizeof(unsigned int),
660 		.mode		= 0644,
661 		.proc_handler	= proc_dointvec_minmax,
662 		.extra1		= SYSCTL_ZERO,
663 	},
664 };
665 
666 static struct ctl_table netns_core_table[] = {
667 #if IS_ENABLED(CONFIG_RPS)
668 	{
669 		.procname	= "rps_default_mask",
670 		.data		= &init_net,
671 		.mode		= 0644,
672 		.proc_handler	= rps_default_mask_sysctl
673 	},
674 #endif
675 	{
676 		.procname	= "somaxconn",
677 		.data		= &init_net.core.sysctl_somaxconn,
678 		.maxlen		= sizeof(int),
679 		.mode		= 0644,
680 		.extra1		= SYSCTL_ZERO,
681 		.proc_handler	= proc_dointvec_minmax
682 	},
683 	{
684 		.procname	= "optmem_max",
685 		.data		= &init_net.core.sysctl_optmem_max,
686 		.maxlen		= sizeof(int),
687 		.mode		= 0644,
688 		.extra1		= SYSCTL_ZERO,
689 		.proc_handler	= proc_dointvec_minmax
690 	},
691 	{
692 		.procname	= "txrehash",
693 		.data		= &init_net.core.sysctl_txrehash,
694 		.maxlen		= sizeof(u8),
695 		.mode		= 0644,
696 		.extra1		= SYSCTL_ZERO,
697 		.extra2		= SYSCTL_ONE,
698 		.proc_handler	= proc_dou8vec_minmax,
699 	},
700 };
701 
702 static int __init fb_tunnels_only_for_init_net_sysctl_setup(char *str)
703 {
704 	/* fallback tunnels for initns only */
705 	if (!strncmp(str, "initns", 6))
706 		sysctl_fb_tunnels_only_for_init_net = 1;
707 	/* no fallback tunnels anywhere */
708 	else if (!strncmp(str, "none", 4))
709 		sysctl_fb_tunnels_only_for_init_net = 2;
710 
711 	return 1;
712 }
713 __setup("fb_tunnels=", fb_tunnels_only_for_init_net_sysctl_setup);
714 
715 static __net_init int sysctl_core_net_init(struct net *net)
716 {
717 	size_t table_size = ARRAY_SIZE(netns_core_table);
718 	struct ctl_table *tbl;
719 
720 	tbl = netns_core_table;
721 	if (!net_eq(net, &init_net)) {
722 		int i;
723 		tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
724 		if (tbl == NULL)
725 			goto err_dup;
726 
727 		for (i = 0; i < table_size; ++i)
728 			tbl[i].data += (char *)net - (char *)&init_net;
729 	}
730 
731 	net->core.sysctl_hdr = register_net_sysctl_sz(net, "net/core", tbl, table_size);
732 	if (net->core.sysctl_hdr == NULL)
733 		goto err_reg;
734 
735 	return 0;
736 
737 err_reg:
738 	if (tbl != netns_core_table)
739 		kfree(tbl);
740 err_dup:
741 	return -ENOMEM;
742 }
743 
744 static __net_exit void sysctl_core_net_exit(struct net *net)
745 {
746 	const struct ctl_table *tbl;
747 
748 	tbl = net->core.sysctl_hdr->ctl_table_arg;
749 	unregister_net_sysctl_table(net->core.sysctl_hdr);
750 	BUG_ON(tbl == netns_core_table);
751 #if IS_ENABLED(CONFIG_RPS)
752 	kfree(net->core.rps_default_mask);
753 #endif
754 	kfree(tbl);
755 }
756 
757 static __net_initdata struct pernet_operations sysctl_core_ops = {
758 	.init = sysctl_core_net_init,
759 	.exit = sysctl_core_net_exit,
760 };
761 
762 static __init int sysctl_core_init(void)
763 {
764 	register_net_sysctl(&init_net, "net/core", net_core_table);
765 	return register_pernet_subsys(&sysctl_core_ops);
766 }
767 
768 fs_initcall(sysctl_core_init);
769