xref: /linux-6.15/net/core/sysctl_net_core.c (revision 490a79fa) 
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/rps.h>
28 
29 #include "dev.h"
30 
31 static int int_3600 = 3600;
32 static int min_sndbuf = SOCK_MIN_SNDBUF;
33 static int min_rcvbuf = SOCK_MIN_RCVBUF;
34 static int max_skb_frags = MAX_SKB_FRAGS;
35 static int min_mem_pcpu_rsv = SK_MEMORY_PCPU_RESERVE;
36 
37 static int net_msg_warn;	/* Unused, but still a sysctl */
38 
39 int sysctl_fb_tunnels_only_for_init_net __read_mostly = 0;
40 EXPORT_SYMBOL(sysctl_fb_tunnels_only_for_init_net);
41 
42 /* 0 - Keep current behavior:
43  *     IPv4: inherit all current settings from init_net
44  *     IPv6: reset all settings to default
45  * 1 - Both inherit all current settings from init_net
46  * 2 - Both reset all settings to default
47  * 3 - Both inherit all settings from current netns
48  */
49 int sysctl_devconf_inherit_init_net __read_mostly;
50 EXPORT_SYMBOL(sysctl_devconf_inherit_init_net);
51 
52 #if IS_ENABLED(CONFIG_NET_FLOW_LIMIT) || IS_ENABLED(CONFIG_RPS)
53 static void dump_cpumask(void *buffer, size_t *lenp, loff_t *ppos,
54 			 struct cpumask *mask)
55 {
56 	char kbuf[128];
57 	int len;
58 
59 	if (*ppos || !*lenp) {
60 		*lenp = 0;
61 		return;
62 	}
63 
64 	len = min(sizeof(kbuf) - 1, *lenp);
65 	len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
66 	if (!len) {
67 		*lenp = 0;
68 		return;
69 	}
70 
71 	if (len < *lenp)
72 		kbuf[len++] = '\n';
73 	memcpy(buffer, kbuf, len);
74 	*lenp = len;
75 	*ppos += len;
76 }
77 #endif
78 
79 #ifdef CONFIG_RPS
80 
81 static struct cpumask *rps_default_mask_cow_alloc(struct net *net)
82 {
83 	struct cpumask *rps_default_mask;
84 
85 	if (net->core.rps_default_mask)
86 		return net->core.rps_default_mask;
87 
88 	rps_default_mask = kzalloc(cpumask_size(), GFP_KERNEL);
89 	if (!rps_default_mask)
90 		return NULL;
91 
92 	/* pairs with READ_ONCE in rx_queue_default_mask() */
93 	WRITE_ONCE(net->core.rps_default_mask, rps_default_mask);
94 	return rps_default_mask;
95 }
96 
97 static int rps_default_mask_sysctl(struct ctl_table *table, int write,
98 				   void *buffer, size_t *lenp, loff_t *ppos)
99 {
100 	struct net *net = (struct net *)table->data;
101 	int err = 0;
102 
103 	rtnl_lock();
104 	if (write) {
105 		struct cpumask *rps_default_mask = rps_default_mask_cow_alloc(net);
106 
107 		err = -ENOMEM;
108 		if (!rps_default_mask)
109 			goto done;
110 
111 		err = cpumask_parse(buffer, rps_default_mask);
112 		if (err)
113 			goto done;
114 
115 		err = rps_cpumask_housekeeping(rps_default_mask);
116 		if (err)
117 			goto done;
118 	} else {
119 		dump_cpumask(buffer, lenp, ppos,
120 			     net->core.rps_default_mask ? : cpu_none_mask);
121 	}
122 
123 done:
124 	rtnl_unlock();
125 	return err;
126 }
127 
128 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
129 				void *buffer, size_t *lenp, loff_t *ppos)
130 {
131 	unsigned int orig_size, size;
132 	int ret, i;
133 	struct ctl_table tmp = {
134 		.data = &size,
135 		.maxlen = sizeof(size),
136 		.mode = table->mode
137 	};
138 	struct rps_sock_flow_table *orig_sock_table, *sock_table;
139 	static DEFINE_MUTEX(sock_flow_mutex);
140 
141 	mutex_lock(&sock_flow_mutex);
142 
143 	orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
144 					lockdep_is_held(&sock_flow_mutex));
145 	size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
146 
147 	ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
148 
149 	if (write) {
150 		if (size) {
151 			if (size > 1<<29) {
152 				/* Enforce limit to prevent overflow */
153 				mutex_unlock(&sock_flow_mutex);
154 				return -EINVAL;
155 			}
156 			size = roundup_pow_of_two(size);
157 			if (size != orig_size) {
158 				sock_table =
159 				    vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
160 				if (!sock_table) {
161 					mutex_unlock(&sock_flow_mutex);
162 					return -ENOMEM;
163 				}
164 				rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
165 				sock_table->mask = size - 1;
166 			} else
167 				sock_table = orig_sock_table;
168 
169 			for (i = 0; i < size; i++)
170 				sock_table->ents[i] = RPS_NO_CPU;
171 		} else
172 			sock_table = NULL;
173 
174 		if (sock_table != orig_sock_table) {
175 			rcu_assign_pointer(rps_sock_flow_table, sock_table);
176 			if (sock_table) {
177 				static_branch_inc(&rps_needed);
178 				static_branch_inc(&rfs_needed);
179 			}
180 			if (orig_sock_table) {
181 				static_branch_dec(&rps_needed);
182 				static_branch_dec(&rfs_needed);
183 				kvfree_rcu_mightsleep(orig_sock_table);
184 			}
185 		}
186 	}
187 
188 	mutex_unlock(&sock_flow_mutex);
189 
190 	return ret;
191 }
192 #endif /* CONFIG_RPS */
193 
194 #ifdef CONFIG_NET_FLOW_LIMIT
195 static DEFINE_MUTEX(flow_limit_update_mutex);
196 
197 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
198 				 void *buffer, size_t *lenp, loff_t *ppos)
199 {
200 	struct sd_flow_limit *cur;
201 	struct softnet_data *sd;
202 	cpumask_var_t mask;
203 	int i, len, ret = 0;
204 
205 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
206 		return -ENOMEM;
207 
208 	if (write) {
209 		ret = cpumask_parse(buffer, mask);
210 		if (ret)
211 			goto done;
212 
213 		mutex_lock(&flow_limit_update_mutex);
214 		len = sizeof(*cur) + netdev_flow_limit_table_len;
215 		for_each_possible_cpu(i) {
216 			sd = &per_cpu(softnet_data, i);
217 			cur = rcu_dereference_protected(sd->flow_limit,
218 				     lockdep_is_held(&flow_limit_update_mutex));
219 			if (cur && !cpumask_test_cpu(i, mask)) {
220 				RCU_INIT_POINTER(sd->flow_limit, NULL);
221 				kfree_rcu_mightsleep(cur);
222 			} else if (!cur && cpumask_test_cpu(i, mask)) {
223 				cur = kzalloc_node(len, GFP_KERNEL,
224 						   cpu_to_node(i));
225 				if (!cur) {
226 					/* not unwinding previous changes */
227 					ret = -ENOMEM;
228 					goto write_unlock;
229 				}
230 				cur->num_buckets = netdev_flow_limit_table_len;
231 				rcu_assign_pointer(sd->flow_limit, cur);
232 			}
233 		}
234 write_unlock:
235 		mutex_unlock(&flow_limit_update_mutex);
236 	} else {
237 		cpumask_clear(mask);
238 		rcu_read_lock();
239 		for_each_possible_cpu(i) {
240 			sd = &per_cpu(softnet_data, i);
241 			if (rcu_dereference(sd->flow_limit))
242 				cpumask_set_cpu(i, mask);
243 		}
244 		rcu_read_unlock();
245 
246 		dump_cpumask(buffer, lenp, ppos, mask);
247 	}
248 
249 done:
250 	free_cpumask_var(mask);
251 	return ret;
252 }
253 
254 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
255 				       void *buffer, size_t *lenp, loff_t *ppos)
256 {
257 	unsigned int old, *ptr;
258 	int ret;
259 
260 	mutex_lock(&flow_limit_update_mutex);
261 
262 	ptr = table->data;
263 	old = *ptr;
264 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
265 	if (!ret && write && !is_power_of_2(*ptr)) {
266 		*ptr = old;
267 		ret = -EINVAL;
268 	}
269 
270 	mutex_unlock(&flow_limit_update_mutex);
271 	return ret;
272 }
273 #endif /* CONFIG_NET_FLOW_LIMIT */
274 
275 #ifdef CONFIG_NET_SCHED
276 static int set_default_qdisc(struct ctl_table *table, int write,
277 			     void *buffer, size_t *lenp, loff_t *ppos)
278 {
279 	char id[IFNAMSIZ];
280 	struct ctl_table tbl = {
281 		.data = id,
282 		.maxlen = IFNAMSIZ,
283 	};
284 	int ret;
285 
286 	qdisc_get_default(id, IFNAMSIZ);
287 
288 	ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
289 	if (write && ret == 0)
290 		ret = qdisc_set_default(id);
291 	return ret;
292 }
293 #endif
294 
295 static int proc_do_dev_weight(struct ctl_table *table, int write,
296 			   void *buffer, size_t *lenp, loff_t *ppos)
297 {
298 	static DEFINE_MUTEX(dev_weight_mutex);
299 	int ret, weight;
300 
301 	mutex_lock(&dev_weight_mutex);
302 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
303 	if (!ret && write) {
304 		weight = READ_ONCE(weight_p);
305 		WRITE_ONCE(net_hotdata.dev_rx_weight, weight * dev_weight_rx_bias);
306 		WRITE_ONCE(net_hotdata.dev_tx_weight, weight * dev_weight_tx_bias);
307 	}
308 	mutex_unlock(&dev_weight_mutex);
309 
310 	return ret;
311 }
312 
313 static int proc_do_rss_key(struct ctl_table *table, int write,
314 			   void *buffer, size_t *lenp, loff_t *ppos)
315 {
316 	struct ctl_table fake_table;
317 	char buf[NETDEV_RSS_KEY_LEN * 3];
318 
319 	snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
320 	fake_table.data = buf;
321 	fake_table.maxlen = sizeof(buf);
322 	return proc_dostring(&fake_table, write, buffer, lenp, ppos);
323 }
324 
325 #ifdef CONFIG_BPF_JIT
326 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write,
327 					   void *buffer, size_t *lenp,
328 					   loff_t *ppos)
329 {
330 	int ret, jit_enable = *(int *)table->data;
331 	int min = *(int *)table->extra1;
332 	int max = *(int *)table->extra2;
333 	struct ctl_table tmp = *table;
334 
335 	if (write && !capable(CAP_SYS_ADMIN))
336 		return -EPERM;
337 
338 	tmp.data = &jit_enable;
339 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
340 	if (write && !ret) {
341 		if (jit_enable < 2 ||
342 		    (jit_enable == 2 && bpf_dump_raw_ok(current_cred()))) {
343 			*(int *)table->data = jit_enable;
344 			if (jit_enable == 2)
345 				pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
346 		} else {
347 			ret = -EPERM;
348 		}
349 	}
350 
351 	if (write && ret && min == max)
352 		pr_info_once("CONFIG_BPF_JIT_ALWAYS_ON is enabled, bpf_jit_enable is permanently set to 1.\n");
353 
354 	return ret;
355 }
356 
357 # ifdef CONFIG_HAVE_EBPF_JIT
358 static int
359 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write,
360 				    void *buffer, size_t *lenp, loff_t *ppos)
361 {
362 	if (!capable(CAP_SYS_ADMIN))
363 		return -EPERM;
364 
365 	return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
366 }
367 # endif /* CONFIG_HAVE_EBPF_JIT */
368 
369 static int
370 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
371 				     void *buffer, size_t *lenp, loff_t *ppos)
372 {
373 	if (!capable(CAP_SYS_ADMIN))
374 		return -EPERM;
375 
376 	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
377 }
378 #endif
379 
380 static struct ctl_table net_core_table[] = {
381 	{
382 		.procname	= "wmem_max",
383 		.data		= &sysctl_wmem_max,
384 		.maxlen		= sizeof(int),
385 		.mode		= 0644,
386 		.proc_handler	= proc_dointvec_minmax,
387 		.extra1		= &min_sndbuf,
388 	},
389 	{
390 		.procname	= "rmem_max",
391 		.data		= &sysctl_rmem_max,
392 		.maxlen		= sizeof(int),
393 		.mode		= 0644,
394 		.proc_handler	= proc_dointvec_minmax,
395 		.extra1		= &min_rcvbuf,
396 	},
397 	{
398 		.procname	= "wmem_default",
399 		.data		= &sysctl_wmem_default,
400 		.maxlen		= sizeof(int),
401 		.mode		= 0644,
402 		.proc_handler	= proc_dointvec_minmax,
403 		.extra1		= &min_sndbuf,
404 	},
405 	{
406 		.procname	= "rmem_default",
407 		.data		= &sysctl_rmem_default,
408 		.maxlen		= sizeof(int),
409 		.mode		= 0644,
410 		.proc_handler	= proc_dointvec_minmax,
411 		.extra1		= &min_rcvbuf,
412 	},
413 	{
414 		.procname	= "mem_pcpu_rsv",
415 		.data		= &sysctl_mem_pcpu_rsv,
416 		.maxlen		= sizeof(int),
417 		.mode		= 0644,
418 		.proc_handler	= proc_dointvec_minmax,
419 		.extra1		= &min_mem_pcpu_rsv,
420 	},
421 	{
422 		.procname	= "dev_weight",
423 		.data		= &weight_p,
424 		.maxlen		= sizeof(int),
425 		.mode		= 0644,
426 		.proc_handler	= proc_do_dev_weight,
427 	},
428 	{
429 		.procname	= "dev_weight_rx_bias",
430 		.data		= &dev_weight_rx_bias,
431 		.maxlen		= sizeof(int),
432 		.mode		= 0644,
433 		.proc_handler	= proc_do_dev_weight,
434 	},
435 	{
436 		.procname	= "dev_weight_tx_bias",
437 		.data		= &dev_weight_tx_bias,
438 		.maxlen		= sizeof(int),
439 		.mode		= 0644,
440 		.proc_handler	= proc_do_dev_weight,
441 	},
442 	{
443 		.procname	= "netdev_max_backlog",
444 		.data		= &net_hotdata.max_backlog,
445 		.maxlen		= sizeof(int),
446 		.mode		= 0644,
447 		.proc_handler	= proc_dointvec
448 	},
449 	{
450 		.procname	= "netdev_rss_key",
451 		.data		= &netdev_rss_key,
452 		.maxlen		= sizeof(int),
453 		.mode		= 0444,
454 		.proc_handler	= proc_do_rss_key,
455 	},
456 #ifdef CONFIG_BPF_JIT
457 	{
458 		.procname	= "bpf_jit_enable",
459 		.data		= &bpf_jit_enable,
460 		.maxlen		= sizeof(int),
461 		.mode		= 0644,
462 		.proc_handler	= proc_dointvec_minmax_bpf_enable,
463 # ifdef CONFIG_BPF_JIT_ALWAYS_ON
464 		.extra1		= SYSCTL_ONE,
465 		.extra2		= SYSCTL_ONE,
466 # else
467 		.extra1		= SYSCTL_ZERO,
468 		.extra2		= SYSCTL_TWO,
469 # endif
470 	},
471 # ifdef CONFIG_HAVE_EBPF_JIT
472 	{
473 		.procname	= "bpf_jit_harden",
474 		.data		= &bpf_jit_harden,
475 		.maxlen		= sizeof(int),
476 		.mode		= 0600,
477 		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
478 		.extra1		= SYSCTL_ZERO,
479 		.extra2		= SYSCTL_TWO,
480 	},
481 	{
482 		.procname	= "bpf_jit_kallsyms",
483 		.data		= &bpf_jit_kallsyms,
484 		.maxlen		= sizeof(int),
485 		.mode		= 0600,
486 		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
487 		.extra1		= SYSCTL_ZERO,
488 		.extra2		= SYSCTL_ONE,
489 	},
490 # endif
491 	{
492 		.procname	= "bpf_jit_limit",
493 		.data		= &bpf_jit_limit,
494 		.maxlen		= sizeof(long),
495 		.mode		= 0600,
496 		.proc_handler	= proc_dolongvec_minmax_bpf_restricted,
497 		.extra1		= SYSCTL_LONG_ONE,
498 		.extra2		= &bpf_jit_limit_max,
499 	},
500 #endif
501 	{
502 		.procname	= "netdev_tstamp_prequeue",
503 		.data		= &net_hotdata.tstamp_prequeue,
504 		.maxlen		= sizeof(int),
505 		.mode		= 0644,
506 		.proc_handler	= proc_dointvec
507 	},
508 	{
509 		.procname	= "message_cost",
510 		.data		= &net_ratelimit_state.interval,
511 		.maxlen		= sizeof(int),
512 		.mode		= 0644,
513 		.proc_handler	= proc_dointvec_jiffies,
514 	},
515 	{
516 		.procname	= "message_burst",
517 		.data		= &net_ratelimit_state.burst,
518 		.maxlen		= sizeof(int),
519 		.mode		= 0644,
520 		.proc_handler	= proc_dointvec,
521 	},
522 	{
523 		.procname	= "tstamp_allow_data",
524 		.data		= &sysctl_tstamp_allow_data,
525 		.maxlen		= sizeof(int),
526 		.mode		= 0644,
527 		.proc_handler	= proc_dointvec_minmax,
528 		.extra1		= SYSCTL_ZERO,
529 		.extra2		= SYSCTL_ONE
530 	},
531 #ifdef CONFIG_RPS
532 	{
533 		.procname	= "rps_sock_flow_entries",
534 		.maxlen		= sizeof(int),
535 		.mode		= 0644,
536 		.proc_handler	= rps_sock_flow_sysctl
537 	},
538 #endif
539 #ifdef CONFIG_NET_FLOW_LIMIT
540 	{
541 		.procname	= "flow_limit_cpu_bitmap",
542 		.mode		= 0644,
543 		.proc_handler	= flow_limit_cpu_sysctl
544 	},
545 	{
546 		.procname	= "flow_limit_table_len",
547 		.data		= &netdev_flow_limit_table_len,
548 		.maxlen		= sizeof(int),
549 		.mode		= 0644,
550 		.proc_handler	= flow_limit_table_len_sysctl
551 	},
552 #endif /* CONFIG_NET_FLOW_LIMIT */
553 #ifdef CONFIG_NET_RX_BUSY_POLL
554 	{
555 		.procname	= "busy_poll",
556 		.data		= &sysctl_net_busy_poll,
557 		.maxlen		= sizeof(unsigned int),
558 		.mode		= 0644,
559 		.proc_handler	= proc_dointvec_minmax,
560 		.extra1		= SYSCTL_ZERO,
561 	},
562 	{
563 		.procname	= "busy_read",
564 		.data		= &sysctl_net_busy_read,
565 		.maxlen		= sizeof(unsigned int),
566 		.mode		= 0644,
567 		.proc_handler	= proc_dointvec_minmax,
568 		.extra1		= SYSCTL_ZERO,
569 	},
570 #endif
571 #ifdef CONFIG_NET_SCHED
572 	{
573 		.procname	= "default_qdisc",
574 		.mode		= 0644,
575 		.maxlen		= IFNAMSIZ,
576 		.proc_handler	= set_default_qdisc
577 	},
578 #endif
579 	{
580 		.procname	= "netdev_budget",
581 		.data		= &net_hotdata.netdev_budget,
582 		.maxlen		= sizeof(int),
583 		.mode		= 0644,
584 		.proc_handler	= proc_dointvec
585 	},
586 	{
587 		.procname	= "warnings",
588 		.data		= &net_msg_warn,
589 		.maxlen		= sizeof(int),
590 		.mode		= 0644,
591 		.proc_handler	= proc_dointvec
592 	},
593 	{
594 		.procname	= "max_skb_frags",
595 		.data		= &sysctl_max_skb_frags,
596 		.maxlen		= sizeof(int),
597 		.mode		= 0644,
598 		.proc_handler	= proc_dointvec_minmax,
599 		.extra1		= SYSCTL_ONE,
600 		.extra2		= &max_skb_frags,
601 	},
602 	{
603 		.procname	= "netdev_budget_usecs",
604 		.data		= &net_hotdata.netdev_budget_usecs,
605 		.maxlen		= sizeof(unsigned int),
606 		.mode		= 0644,
607 		.proc_handler	= proc_dointvec_minmax,
608 		.extra1		= SYSCTL_ZERO,
609 	},
610 	{
611 		.procname	= "fb_tunnels_only_for_init_net",
612 		.data		= &sysctl_fb_tunnels_only_for_init_net,
613 		.maxlen		= sizeof(int),
614 		.mode		= 0644,
615 		.proc_handler	= proc_dointvec_minmax,
616 		.extra1		= SYSCTL_ZERO,
617 		.extra2		= SYSCTL_TWO,
618 	},
619 	{
620 		.procname	= "devconf_inherit_init_net",
621 		.data		= &sysctl_devconf_inherit_init_net,
622 		.maxlen		= sizeof(int),
623 		.mode		= 0644,
624 		.proc_handler	= proc_dointvec_minmax,
625 		.extra1		= SYSCTL_ZERO,
626 		.extra2		= SYSCTL_THREE,
627 	},
628 	{
629 		.procname	= "high_order_alloc_disable",
630 		.data		= &net_high_order_alloc_disable_key.key,
631 		.maxlen         = sizeof(net_high_order_alloc_disable_key),
632 		.mode		= 0644,
633 		.proc_handler	= proc_do_static_key,
634 	},
635 	{
636 		.procname	= "gro_normal_batch",
637 		.data		= &net_hotdata.gro_normal_batch,
638 		.maxlen		= sizeof(unsigned int),
639 		.mode		= 0644,
640 		.proc_handler	= proc_dointvec_minmax,
641 		.extra1		= SYSCTL_ONE,
642 	},
643 	{
644 		.procname	= "netdev_unregister_timeout_secs",
645 		.data		= &netdev_unregister_timeout_secs,
646 		.maxlen		= sizeof(unsigned int),
647 		.mode		= 0644,
648 		.proc_handler	= proc_dointvec_minmax,
649 		.extra1		= SYSCTL_ONE,
650 		.extra2		= &int_3600,
651 	},
652 	{
653 		.procname	= "skb_defer_max",
654 		.data		= &sysctl_skb_defer_max,
655 		.maxlen		= sizeof(unsigned int),
656 		.mode		= 0644,
657 		.proc_handler	= proc_dointvec_minmax,
658 		.extra1		= SYSCTL_ZERO,
659 	},
660 	{ }
661 };
662 
663 static struct ctl_table netns_core_table[] = {
664 #if IS_ENABLED(CONFIG_RPS)
665 	{
666 		.procname	= "rps_default_mask",
667 		.data		= &init_net,
668 		.mode		= 0644,
669 		.proc_handler	= rps_default_mask_sysctl
670 	},
671 #endif
672 	{
673 		.procname	= "somaxconn",
674 		.data		= &init_net.core.sysctl_somaxconn,
675 		.maxlen		= sizeof(int),
676 		.mode		= 0644,
677 		.extra1		= SYSCTL_ZERO,
678 		.proc_handler	= proc_dointvec_minmax
679 	},
680 	{
681 		.procname	= "optmem_max",
682 		.data		= &init_net.core.sysctl_optmem_max,
683 		.maxlen		= sizeof(int),
684 		.mode		= 0644,
685 		.extra1		= SYSCTL_ZERO,
686 		.proc_handler	= proc_dointvec_minmax
687 	},
688 	{
689 		.procname	= "txrehash",
690 		.data		= &init_net.core.sysctl_txrehash,
691 		.maxlen		= sizeof(u8),
692 		.mode		= 0644,
693 		.extra1		= SYSCTL_ZERO,
694 		.extra2		= SYSCTL_ONE,
695 		.proc_handler	= proc_dou8vec_minmax,
696 	},
697 	{ }
698 };
699 
700 static int __init fb_tunnels_only_for_init_net_sysctl_setup(char *str)
701 {
702 	/* fallback tunnels for initns only */
703 	if (!strncmp(str, "initns", 6))
704 		sysctl_fb_tunnels_only_for_init_net = 1;
705 	/* no fallback tunnels anywhere */
706 	else if (!strncmp(str, "none", 4))
707 		sysctl_fb_tunnels_only_for_init_net = 2;
708 
709 	return 1;
710 }
711 __setup("fb_tunnels=", fb_tunnels_only_for_init_net_sysctl_setup);
712 
713 static __net_init int sysctl_core_net_init(struct net *net)
714 {
715 	struct ctl_table *tbl, *tmp;
716 
717 	tbl = netns_core_table;
718 	if (!net_eq(net, &init_net)) {
719 		tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
720 		if (tbl == NULL)
721 			goto err_dup;
722 
723 		for (tmp = tbl; tmp->procname; tmp++)
724 			tmp->data += (char *)net - (char *)&init_net;
725 	}
726 
727 	net->core.sysctl_hdr = register_net_sysctl_sz(net, "net/core", tbl,
728 						      ARRAY_SIZE(netns_core_table));
729 	if (net->core.sysctl_hdr == NULL)
730 		goto err_reg;
731 
732 	return 0;
733 
734 err_reg:
735 	if (tbl != netns_core_table)
736 		kfree(tbl);
737 err_dup:
738 	return -ENOMEM;
739 }
740 
741 static __net_exit void sysctl_core_net_exit(struct net *net)
742 {
743 	struct ctl_table *tbl;
744 
745 	tbl = net->core.sysctl_hdr->ctl_table_arg;
746 	unregister_net_sysctl_table(net->core.sysctl_hdr);
747 	BUG_ON(tbl == netns_core_table);
748 #if IS_ENABLED(CONFIG_RPS)
749 	kfree(net->core.rps_default_mask);
750 #endif
751 	kfree(tbl);
752 }
753 
754 static __net_initdata struct pernet_operations sysctl_core_ops = {
755 	.init = sysctl_core_net_init,
756 	.exit = sysctl_core_net_exit,
757 };
758 
759 static __init int sysctl_core_init(void)
760 {
761 	register_net_sysctl(&init_net, "net/core", net_core_table);
762 	return register_pernet_subsys(&sysctl_core_ops);
763 }
764 
765 fs_initcall(sysctl_core_init);
766