xref: /linux/net/core/sysctl_net_core.c (revision 08dc30de1a402fe88fd80592cf6c72c4c2ebbcbc)
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 #include <linux/hex.h>
21 
22 #include <net/ip.h>
23 #include <net/sock.h>
24 #include <net/net_ratelimit.h>
25 #include <net/busy_poll.h>
26 #include <net/pkt_sched.h>
27 #include <net/hotdata.h>
28 #include <net/proto_memory.h>
29 #include <net/rps.h>
30 
31 #include "dev.h"
32 #include "net-sysfs.h"
33 
34 static int int_3600 = 3600;
35 static int min_sndbuf = SOCK_MIN_SNDBUF;
36 static int min_rcvbuf = SOCK_MIN_RCVBUF;
37 static int max_skb_frags = MAX_SKB_FRAGS;
38 static int min_mem_pcpu_rsv = SK_MEMORY_PCPU_RESERVE;
39 static int netdev_budget_usecs_min = 2 * USEC_PER_SEC / HZ;
40 
41 static int net_msg_warn;	/* Unused, but still a sysctl */
42 
43 int sysctl_fb_tunnels_only_for_init_net __read_mostly = 0;
44 EXPORT_SYMBOL(sysctl_fb_tunnels_only_for_init_net);
45 
46 /* 0 - Keep current behavior:
47  *     IPv4: inherit all current settings from init_net
48  *     IPv6: reset all settings to default
49  * 1 - Both inherit all current settings from init_net
50  * 2 - Both reset all settings to default
51  * 3 - Both inherit all settings from current netns
52  */
53 int sysctl_devconf_inherit_init_net __read_mostly;
54 EXPORT_SYMBOL(sysctl_devconf_inherit_init_net);
55 
56 #if IS_ENABLED(CONFIG_NET_FLOW_LIMIT) || IS_ENABLED(CONFIG_RPS)
57 static int dump_cpumask(void *buffer, size_t *lenp, loff_t *ppos,
58 			struct cpumask *mask)
59 {
60 	char *kbuf;
61 	int len;
62 
63 	if (*ppos || !*lenp) {
64 		*lenp = 0;
65 		return 0;
66 	}
67 
68 	/* CPUs are displayed as a hex bitmap + a comma between each groups of 8
69 	 * nibbles (except the last one which has a newline instead).
70 	 * Guesstimate the buffer size at the group granularity level.
71 	 */
72 	len = min(DIV_ROUND_UP(nr_cpumask_bits, 32) * (8 + 1), *lenp);
73 	kbuf = kmalloc(len, GFP_KERNEL);
74 	if (!kbuf) {
75 		*lenp = 0;
76 		return -ENOMEM;
77 	}
78 
79 	len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
80 	if (!len) {
81 		*lenp = 0;
82 		goto free_buf;
83 	}
84 
85 	/* scnprintf writes a trailing null char not counted in the returned
86 	 * length, override it with a newline.
87 	 */
88 	kbuf[len++] = '\n';
89 	memcpy(buffer, kbuf, len);
90 	*lenp = len;
91 	*ppos += len;
92 
93 free_buf:
94 	kfree(kbuf);
95 	return 0;
96 }
97 #endif
98 
99 #ifdef CONFIG_RPS
100 
101 DEFINE_MUTEX(rps_default_mask_mutex);
102 
103 static int rps_default_mask_sysctl(const struct ctl_table *table, int write,
104 				   void *buffer, size_t *lenp, loff_t *ppos)
105 {
106 	struct net *net = (struct net *)table->data;
107 	struct cpumask *mask;
108 	int err = 0;
109 
110 	mutex_lock(&rps_default_mask_mutex);
111 	mask = net->core.rps_default_mask;
112 	if (write) {
113 		if (!mask) {
114 			mask = kzalloc(cpumask_size(), GFP_KERNEL);
115 			net->core.rps_default_mask = mask;
116 		}
117 		err = -ENOMEM;
118 		if (!mask)
119 			goto done;
120 
121 		err = cpumask_parse(buffer, mask);
122 		if (err)
123 			goto done;
124 
125 		err = rps_cpumask_housekeeping(mask);
126 		if (err)
127 			goto done;
128 	} else {
129 		err = dump_cpumask(buffer, lenp, ppos,
130 				   mask ?: cpu_none_mask);
131 	}
132 
133 done:
134 	mutex_unlock(&rps_default_mask_mutex);
135 	return err;
136 }
137 
138 static int rps_sock_flow_sysctl(const struct ctl_table *table, int write,
139 				void *buffer, size_t *lenp, loff_t *ppos)
140 {
141 	struct rps_sock_flow_table *o_sock_table, *sock_table;
142 	static DEFINE_MUTEX(sock_flow_mutex);
143 	rps_tag_ptr o_tag_ptr, tag_ptr;
144 	unsigned int orig_size, size;
145 	struct ctl_table tmp = {
146 		.data = &size,
147 		.maxlen = sizeof(size),
148 		.mode = table->mode
149 	};
150 	void *tofree = NULL;
151 	int ret, i;
152 	u8 log;
153 
154 	mutex_lock(&sock_flow_mutex);
155 
156 	o_tag_ptr = tag_ptr = net_hotdata.rps_sock_flow_table;
157 
158 	size = o_tag_ptr ? rps_tag_to_mask(o_tag_ptr) + 1 : 0;
159 	o_sock_table = rps_tag_to_table(o_tag_ptr);
160 	orig_size = size;
161 
162 	ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
163 
164 	if (!write)
165 		goto unlock;
166 
167 	if (size) {
168 		if (size > 1<<29) {
169 			/* Enforce limit to prevent overflow */
170 			mutex_unlock(&sock_flow_mutex);
171 			return -EINVAL;
172 		}
173 		sock_table = o_sock_table;
174 		size = roundup_pow_of_two(size);
175 		if (size != orig_size) {
176 			sock_table = vmalloc_huge(size * sizeof(*sock_table),
177 						  GFP_KERNEL);
178 			if (!sock_table) {
179 				mutex_unlock(&sock_flow_mutex);
180 				return -ENOMEM;
181 			}
182 			net_hotdata.rps_cpu_mask =
183 				roundup_pow_of_two(nr_cpu_ids) - 1;
184 			log = ilog2(size);
185 			tag_ptr = (rps_tag_ptr)sock_table | log;
186 		}
187 
188 		for (i = 0; i < size; i++)
189 			sock_table[i].ent = RPS_NO_CPU;
190 	} else {
191 		sock_table = NULL;
192 		tag_ptr = 0UL;
193 	}
194 	if (tag_ptr != o_tag_ptr) {
195 		smp_store_release(&net_hotdata.rps_sock_flow_table, tag_ptr);
196 		if (sock_table) {
197 			static_branch_inc(&rps_needed);
198 			static_branch_inc(&rfs_needed);
199 		}
200 		if (o_sock_table) {
201 			static_branch_dec(&rps_needed);
202 			static_branch_dec(&rfs_needed);
203 			tofree = o_sock_table;
204 		}
205 	}
206 
207 unlock:
208 	mutex_unlock(&sock_flow_mutex);
209 
210 	kvfree_rcu_mightsleep(tofree);
211 	return ret;
212 }
213 #endif /* CONFIG_RPS */
214 
215 #ifdef CONFIG_NET_FLOW_LIMIT
216 static DEFINE_MUTEX(flow_limit_update_mutex);
217 
218 static int flow_limit_cpu_sysctl(const struct ctl_table *table, int write,
219 				 void *buffer, size_t *lenp, loff_t *ppos)
220 {
221 	struct sd_flow_limit *cur;
222 	struct softnet_data *sd;
223 	cpumask_var_t mask;
224 	int i, len, ret = 0;
225 
226 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
227 		return -ENOMEM;
228 
229 	if (write) {
230 		ret = cpumask_parse(buffer, mask);
231 		if (ret)
232 			goto done;
233 
234 		mutex_lock(&flow_limit_update_mutex);
235 		len = sizeof(*cur) + netdev_flow_limit_table_len;
236 		for_each_possible_cpu(i) {
237 			sd = &per_cpu(softnet_data, i);
238 			cur = rcu_dereference_protected(sd->flow_limit,
239 				     lockdep_is_held(&flow_limit_update_mutex));
240 			if (cur && !cpumask_test_cpu(i, mask)) {
241 				RCU_INIT_POINTER(sd->flow_limit, NULL);
242 				kfree_rcu(cur, rcu);
243 			} else if (!cur && cpumask_test_cpu(i, mask)) {
244 				cur = kzalloc_node(len, GFP_KERNEL,
245 						   cpu_to_node(i));
246 				if (!cur) {
247 					/* not unwinding previous changes */
248 					ret = -ENOMEM;
249 					goto write_unlock;
250 				}
251 				cur->log_buckets = ilog2(netdev_flow_limit_table_len);
252 				rcu_assign_pointer(sd->flow_limit, cur);
253 			}
254 		}
255 write_unlock:
256 		mutex_unlock(&flow_limit_update_mutex);
257 	} else {
258 		cpumask_clear(mask);
259 		rcu_read_lock();
260 		for_each_possible_cpu(i) {
261 			sd = &per_cpu(softnet_data, i);
262 			if (rcu_dereference(sd->flow_limit))
263 				cpumask_set_cpu(i, mask);
264 		}
265 		rcu_read_unlock();
266 
267 		ret = dump_cpumask(buffer, lenp, ppos, mask);
268 	}
269 
270 done:
271 	free_cpumask_var(mask);
272 	return ret;
273 }
274 
275 static int flow_limit_table_len_sysctl(const struct ctl_table *table, int write,
276 				       void *buffer, size_t *lenp, loff_t *ppos)
277 {
278 	unsigned int old, *ptr;
279 	int ret;
280 
281 	mutex_lock(&flow_limit_update_mutex);
282 
283 	ptr = table->data;
284 	old = *ptr;
285 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
286 	if (!ret && write && !is_power_of_2(*ptr)) {
287 		*ptr = old;
288 		ret = -EINVAL;
289 	}
290 
291 	mutex_unlock(&flow_limit_update_mutex);
292 	return ret;
293 }
294 #endif /* CONFIG_NET_FLOW_LIMIT */
295 
296 #ifdef CONFIG_NET_SCHED
297 static int set_default_qdisc(const struct ctl_table *table, int write,
298 			     void *buffer, size_t *lenp, loff_t *ppos)
299 {
300 	char id[IFNAMSIZ];
301 	struct ctl_table tbl = {
302 		.data = id,
303 		.maxlen = IFNAMSIZ,
304 	};
305 	int ret;
306 
307 	qdisc_get_default(id, IFNAMSIZ);
308 
309 	ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
310 	if (write && ret == 0)
311 		ret = qdisc_set_default(id);
312 	return ret;
313 }
314 #endif
315 
316 static int proc_do_dev_weight(const struct ctl_table *table, int write,
317 			   void *buffer, size_t *lenp, loff_t *ppos)
318 {
319 	static DEFINE_MUTEX(dev_weight_mutex);
320 	int ret, weight;
321 
322 	mutex_lock(&dev_weight_mutex);
323 	ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
324 	if (!ret && write) {
325 		weight = READ_ONCE(weight_p);
326 		WRITE_ONCE(net_hotdata.dev_rx_weight, weight * dev_weight_rx_bias);
327 		WRITE_ONCE(net_hotdata.dev_tx_weight, weight * dev_weight_tx_bias);
328 	}
329 	mutex_unlock(&dev_weight_mutex);
330 
331 	return ret;
332 }
333 
334 static int proc_do_rss_key(const struct ctl_table *table, int write,
335 			   void *buffer, size_t *lenp, loff_t *ppos)
336 {
337 	char buf[NETDEV_RSS_KEY_LEN * 3];
338 	struct ctl_table fake_table;
339 	char *pos = buf;
340 
341 	for (int i = 0; i < NETDEV_RSS_KEY_LEN; i++) {
342 		pos = hex_byte_pack(pos, netdev_rss_key[i]);
343 		*pos++ = ':';
344 	}
345 	*(--pos) = 0;
346 
347 	fake_table.data = buf;
348 	fake_table.maxlen = sizeof(buf);
349 	return proc_dostring(&fake_table, write, buffer, lenp, ppos);
350 }
351 
352 static int proc_do_skb_defer_max(const struct ctl_table *table, int write,
353 		 void *buffer, size_t *lenp, loff_t *ppos)
354 {
355 	static DEFINE_MUTEX(skb_defer_max_mutex);
356 	int ret, oval, nval;
357 
358 	mutex_lock(&skb_defer_max_mutex);
359 
360 	oval = !net_hotdata.sysctl_skb_defer_max;
361 	ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
362 	nval = !net_hotdata.sysctl_skb_defer_max;
363 
364 	if (nval != oval) {
365 		if (nval)
366 			static_branch_enable(&skb_defer_disable_key);
367 		else
368 			static_branch_disable(&skb_defer_disable_key);
369 	}
370 
371 	mutex_unlock(&skb_defer_max_mutex);
372 	return ret;
373 }
374 
375 #ifdef CONFIG_BPF_JIT
376 static int proc_dointvec_minmax_bpf_enable(const struct ctl_table *table, int write,
377 					   void *buffer, size_t *lenp,
378 					   loff_t *ppos)
379 {
380 	int ret, jit_enable = *(int *)table->data;
381 	int min = *(int *)table->extra1;
382 	int max = *(int *)table->extra2;
383 	struct ctl_table tmp = *table;
384 
385 	if (write && !capable(CAP_SYS_ADMIN))
386 		return -EPERM;
387 
388 	tmp.data = &jit_enable;
389 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
390 	if (write && !ret) {
391 		if (jit_enable < 2 ||
392 		    (jit_enable == 2 && bpf_dump_raw_ok(current_cred()))) {
393 			*(int *)table->data = jit_enable;
394 			if (jit_enable == 2)
395 				pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
396 		} else {
397 			ret = -EPERM;
398 		}
399 	}
400 
401 	if (write && ret && min == max)
402 		pr_info_once("CONFIG_BPF_JIT_ALWAYS_ON is enabled, bpf_jit_enable is permanently set to 1.\n");
403 
404 	return ret;
405 }
406 
407 # ifdef CONFIG_HAVE_EBPF_JIT
408 static int
409 proc_dointvec_minmax_bpf_restricted(const struct ctl_table *table, int write,
410 				    void *buffer, size_t *lenp, loff_t *ppos)
411 {
412 	if (!capable(CAP_SYS_ADMIN))
413 		return -EPERM;
414 
415 	return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
416 }
417 # endif /* CONFIG_HAVE_EBPF_JIT */
418 
419 static int
420 proc_dolongvec_minmax_bpf_restricted(const struct ctl_table *table, int write,
421 				     void *buffer, size_t *lenp, loff_t *ppos)
422 {
423 	if (!capable(CAP_SYS_ADMIN))
424 		return -EPERM;
425 
426 	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
427 }
428 #endif
429 
430 static struct ctl_table net_core_table[] = {
431 	{
432 		.procname	= "mem_pcpu_rsv",
433 		.data		= &net_hotdata.sysctl_mem_pcpu_rsv,
434 		.maxlen		= sizeof(int),
435 		.mode		= 0644,
436 		.proc_handler	= proc_dointvec_minmax,
437 		.extra1		= &min_mem_pcpu_rsv,
438 	},
439 	{
440 		.procname	= "dev_weight",
441 		.data		= &weight_p,
442 		.maxlen		= sizeof(int),
443 		.mode		= 0644,
444 		.proc_handler	= proc_do_dev_weight,
445 		.extra1         = SYSCTL_ONE,
446 	},
447 	{
448 		.procname	= "dev_weight_rx_bias",
449 		.data		= &dev_weight_rx_bias,
450 		.maxlen		= sizeof(int),
451 		.mode		= 0644,
452 		.proc_handler	= proc_do_dev_weight,
453 		.extra1         = SYSCTL_ONE,
454 	},
455 	{
456 		.procname	= "dev_weight_tx_bias",
457 		.data		= &dev_weight_tx_bias,
458 		.maxlen		= sizeof(int),
459 		.mode		= 0644,
460 		.proc_handler	= proc_do_dev_weight,
461 		.extra1         = SYSCTL_ONE,
462 	},
463 	{
464 		.procname	= "netdev_max_backlog",
465 		.data		= &net_hotdata.max_backlog,
466 		.maxlen		= sizeof(int),
467 		.mode		= 0644,
468 		.proc_handler	= proc_dointvec
469 	},
470 	{
471 		.procname	= "qdisc_max_burst",
472 		.data		= &net_hotdata.qdisc_max_burst,
473 		.maxlen		= sizeof(int),
474 		.mode		= 0644,
475 		.proc_handler	= proc_dointvec
476 	},
477 	{
478 		.procname	= "netdev_rss_key",
479 		.data		= &netdev_rss_key,
480 		.maxlen		= sizeof(int),
481 		.mode		= 0444,
482 		.proc_handler	= proc_do_rss_key,
483 	},
484 #ifdef CONFIG_BPF_JIT
485 	{
486 		.procname	= "bpf_jit_enable",
487 		.data		= &bpf_jit_enable,
488 		.maxlen		= sizeof(int),
489 		.mode		= 0644,
490 		.proc_handler	= proc_dointvec_minmax_bpf_enable,
491 # ifdef CONFIG_BPF_JIT_ALWAYS_ON
492 		.extra1		= SYSCTL_ONE,
493 		.extra2		= SYSCTL_ONE,
494 # else
495 		.extra1		= SYSCTL_ZERO,
496 		.extra2		= SYSCTL_TWO,
497 # endif
498 	},
499 # ifdef CONFIG_HAVE_EBPF_JIT
500 	{
501 		.procname	= "bpf_jit_harden",
502 		.data		= &bpf_jit_harden,
503 		.maxlen		= sizeof(int),
504 		.mode		= 0600,
505 		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
506 		.extra1		= SYSCTL_ZERO,
507 		.extra2		= SYSCTL_TWO,
508 	},
509 	{
510 		.procname	= "bpf_jit_kallsyms",
511 		.data		= &bpf_jit_kallsyms,
512 		.maxlen		= sizeof(int),
513 		.mode		= 0600,
514 		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
515 		.extra1		= SYSCTL_ZERO,
516 		.extra2		= SYSCTL_ONE,
517 	},
518 # endif
519 	{
520 		.procname	= "bpf_jit_limit",
521 		.data		= &bpf_jit_limit,
522 		.maxlen		= sizeof(long),
523 		.mode		= 0600,
524 		.proc_handler	= proc_dolongvec_minmax_bpf_restricted,
525 		.extra1		= SYSCTL_LONG_ONE,
526 		.extra2		= &bpf_jit_limit_max,
527 	},
528 #endif
529 	{
530 		.procname	= "netdev_tstamp_prequeue",
531 		.data		= &net_hotdata.tstamp_prequeue,
532 		.maxlen		= sizeof(int),
533 		.mode		= 0644,
534 		.proc_handler	= proc_dointvec
535 	},
536 	{
537 		.procname	= "message_cost",
538 		.data		= &net_ratelimit_state.interval,
539 		.maxlen		= sizeof(int),
540 		.mode		= 0644,
541 		.proc_handler	= proc_dointvec_jiffies,
542 	},
543 	{
544 		.procname	= "message_burst",
545 		.data		= &net_ratelimit_state.burst,
546 		.maxlen		= sizeof(int),
547 		.mode		= 0644,
548 		.proc_handler	= proc_dointvec,
549 	},
550 #ifdef CONFIG_RPS
551 	{
552 		.procname	= "rps_sock_flow_entries",
553 		.maxlen		= sizeof(int),
554 		.mode		= 0644,
555 		.proc_handler	= rps_sock_flow_sysctl
556 	},
557 #endif
558 #ifdef CONFIG_NET_FLOW_LIMIT
559 	{
560 		.procname	= "flow_limit_cpu_bitmap",
561 		.mode		= 0644,
562 		.proc_handler	= flow_limit_cpu_sysctl
563 	},
564 	{
565 		.procname	= "flow_limit_table_len",
566 		.data		= &netdev_flow_limit_table_len,
567 		.maxlen		= sizeof(int),
568 		.mode		= 0644,
569 		.proc_handler	= flow_limit_table_len_sysctl
570 	},
571 #endif /* CONFIG_NET_FLOW_LIMIT */
572 #ifdef CONFIG_NET_RX_BUSY_POLL
573 	{
574 		.procname	= "busy_poll",
575 		.data		= &sysctl_net_busy_poll,
576 		.maxlen		= sizeof(unsigned int),
577 		.mode		= 0644,
578 		.proc_handler	= proc_dointvec_minmax,
579 		.extra1		= SYSCTL_ZERO,
580 	},
581 	{
582 		.procname	= "busy_read",
583 		.data		= &sysctl_net_busy_read,
584 		.maxlen		= sizeof(unsigned int),
585 		.mode		= 0644,
586 		.proc_handler	= proc_dointvec_minmax,
587 		.extra1		= SYSCTL_ZERO,
588 	},
589 #endif
590 #ifdef CONFIG_NET_SCHED
591 	{
592 		.procname	= "default_qdisc",
593 		.mode		= 0644,
594 		.maxlen		= IFNAMSIZ,
595 		.proc_handler	= set_default_qdisc
596 	},
597 #endif
598 	{
599 		.procname	= "netdev_budget",
600 		.data		= &net_hotdata.netdev_budget,
601 		.maxlen		= sizeof(int),
602 		.mode		= 0644,
603 		.proc_handler	= proc_dointvec
604 	},
605 	{
606 		.procname	= "warnings",
607 		.data		= &net_msg_warn,
608 		.maxlen		= sizeof(int),
609 		.mode		= 0644,
610 		.proc_handler	= proc_dointvec
611 	},
612 	{
613 		.procname	= "max_skb_frags",
614 		.data		= &net_hotdata.sysctl_max_skb_frags,
615 		.maxlen		= sizeof(int),
616 		.mode		= 0644,
617 		.proc_handler	= proc_dointvec_minmax,
618 		.extra1		= SYSCTL_ONE,
619 		.extra2		= &max_skb_frags,
620 	},
621 	{
622 		.procname	= "netdev_budget_usecs",
623 		.data		= &net_hotdata.netdev_budget_usecs,
624 		.maxlen		= sizeof(unsigned int),
625 		.mode		= 0644,
626 		.proc_handler	= proc_dointvec_minmax,
627 		.extra1		= &netdev_budget_usecs_min,
628 	},
629 	{
630 		.procname	= "fb_tunnels_only_for_init_net",
631 		.data		= &sysctl_fb_tunnels_only_for_init_net,
632 		.maxlen		= sizeof(int),
633 		.mode		= 0644,
634 		.proc_handler	= proc_dointvec_minmax,
635 		.extra1		= SYSCTL_ZERO,
636 		.extra2		= SYSCTL_TWO,
637 	},
638 	{
639 		.procname	= "devconf_inherit_init_net",
640 		.data		= &sysctl_devconf_inherit_init_net,
641 		.maxlen		= sizeof(int),
642 		.mode		= 0644,
643 		.proc_handler	= proc_dointvec_minmax,
644 		.extra1		= SYSCTL_ZERO,
645 		.extra2		= SYSCTL_THREE,
646 	},
647 	{
648 		.procname	= "high_order_alloc_disable",
649 		.data		= &net_high_order_alloc_disable_key.key,
650 		.maxlen         = sizeof(net_high_order_alloc_disable_key),
651 		.mode		= 0644,
652 		.proc_handler	= proc_do_static_key,
653 	},
654 	{
655 		.procname	= "gro_normal_batch",
656 		.data		= &net_hotdata.gro_normal_batch,
657 		.maxlen		= sizeof(unsigned int),
658 		.mode		= 0644,
659 		.proc_handler	= proc_dointvec_minmax,
660 		.extra1		= SYSCTL_ONE,
661 	},
662 	{
663 		.procname	= "netdev_unregister_timeout_secs",
664 		.data		= &netdev_unregister_timeout_secs,
665 		.maxlen		= sizeof(unsigned int),
666 		.mode		= 0644,
667 		.proc_handler	= proc_dointvec_minmax,
668 		.extra1		= SYSCTL_ONE,
669 		.extra2		= &int_3600,
670 	},
671 	{
672 		.procname	= "skb_defer_max",
673 		.data		= &net_hotdata.sysctl_skb_defer_max,
674 		.maxlen		= sizeof(unsigned int),
675 		.mode		= 0644,
676 		.proc_handler	= proc_do_skb_defer_max,
677 		.extra1		= SYSCTL_ZERO,
678 	},
679 };
680 
681 static struct ctl_table netns_core_table[] = {
682 #if IS_ENABLED(CONFIG_RPS)
683 	{
684 		.procname	= "rps_default_mask",
685 		.data		= &init_net,
686 		.mode		= 0644,
687 		.proc_handler	= rps_default_mask_sysctl
688 	},
689 #endif
690 	{
691 		.procname	= "somaxconn",
692 		.data		= &init_net.core.sysctl_somaxconn,
693 		.maxlen		= sizeof(int),
694 		.mode		= 0644,
695 		.extra1		= SYSCTL_ZERO,
696 		.proc_handler	= proc_dointvec_minmax
697 	},
698 	{
699 		.procname	= "optmem_max",
700 		.data		= &init_net.core.sysctl_optmem_max,
701 		.maxlen		= sizeof(int),
702 		.mode		= 0644,
703 		.extra1		= SYSCTL_ZERO,
704 		.proc_handler	= proc_dointvec_minmax
705 	},
706 	{
707 		.procname	= "txrehash",
708 		.data		= &init_net.core.sysctl_txrehash,
709 		.maxlen		= sizeof(u8),
710 		.mode		= 0644,
711 		.extra1		= SYSCTL_ZERO,
712 		.extra2		= SYSCTL_ONE,
713 		.proc_handler	= proc_dou8vec_minmax,
714 	},
715 	{
716 		.procname	= "txq_reselection_ms",
717 		.data		= &init_net.core.sysctl_txq_reselection,
718 		.maxlen		= sizeof(int),
719 		.mode		= 0644,
720 		.proc_handler	= proc_dointvec_ms_jiffies,
721 	},
722 	{
723 		.procname	= "tstamp_allow_data",
724 		.data		= &init_net.core.sysctl_tstamp_allow_data,
725 		.maxlen		= sizeof(u8),
726 		.mode		= 0644,
727 		.proc_handler	= proc_dou8vec_minmax,
728 		.extra1		= SYSCTL_ZERO,
729 		.extra2		= SYSCTL_ONE
730 	},
731 	{
732 		.procname	= "bypass_prot_mem",
733 		.data		= &init_net.core.sysctl_bypass_prot_mem,
734 		.maxlen		= sizeof(u8),
735 		.mode		= 0644,
736 		.proc_handler	= proc_dou8vec_minmax,
737 		.extra1		= SYSCTL_ZERO,
738 		.extra2		= SYSCTL_ONE
739 	},
740 	/* sysctl_core_net_init() will set the values after this
741 	 * to readonly in network namespaces
742 	 */
743 	{
744 		.procname	= "wmem_max",
745 		.data		= &sysctl_wmem_max,
746 		.maxlen		= sizeof(int),
747 		.mode		= 0644,
748 		.proc_handler	= proc_dointvec_minmax,
749 		.extra1		= &min_sndbuf,
750 	},
751 	{
752 		.procname	= "rmem_max",
753 		.data		= &sysctl_rmem_max,
754 		.maxlen		= sizeof(int),
755 		.mode		= 0644,
756 		.proc_handler	= proc_dointvec_minmax,
757 		.extra1		= &min_rcvbuf,
758 	},
759 	{
760 		.procname	= "wmem_default",
761 		.data		= &sysctl_wmem_default,
762 		.maxlen		= sizeof(int),
763 		.mode		= 0644,
764 		.proc_handler	= proc_dointvec_minmax,
765 		.extra1		= &min_sndbuf,
766 	},
767 	{
768 		.procname	= "rmem_default",
769 		.data		= &sysctl_rmem_default,
770 		.maxlen		= sizeof(int),
771 		.mode		= 0644,
772 		.proc_handler	= proc_dointvec_minmax,
773 		.extra1		= &min_rcvbuf,
774 	},
775 };
776 
777 static int __init fb_tunnels_only_for_init_net_sysctl_setup(char *str)
778 {
779 	/* fallback tunnels for initns only */
780 	if (!strncmp(str, "initns", 6))
781 		sysctl_fb_tunnels_only_for_init_net = 1;
782 	/* no fallback tunnels anywhere */
783 	else if (!strncmp(str, "none", 4))
784 		sysctl_fb_tunnels_only_for_init_net = 2;
785 
786 	return 1;
787 }
788 __setup("fb_tunnels=", fb_tunnels_only_for_init_net_sysctl_setup);
789 
790 static __net_init int sysctl_core_net_init(struct net *net)
791 {
792 	size_t table_size = ARRAY_SIZE(netns_core_table);
793 	struct ctl_table *tbl;
794 
795 	tbl = netns_core_table;
796 	if (!net_eq(net, &init_net)) {
797 		int i;
798 		tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
799 		if (tbl == NULL)
800 			goto err_dup;
801 
802 		for (i = 0; i < table_size; ++i) {
803 			if (tbl[i].data == &sysctl_wmem_max)
804 				break;
805 
806 			tbl[i].data += (char *)net - (char *)&init_net;
807 		}
808 		for (; i < table_size; ++i)
809 			tbl[i].mode &= ~0222;
810 	}
811 
812 	net->core.sysctl_hdr = register_net_sysctl_sz(net, "net/core", tbl, table_size);
813 	if (net->core.sysctl_hdr == NULL)
814 		goto err_reg;
815 
816 	return 0;
817 
818 err_reg:
819 	if (tbl != netns_core_table)
820 		kfree(tbl);
821 err_dup:
822 	return -ENOMEM;
823 }
824 
825 static __net_exit void sysctl_core_net_exit(struct net *net)
826 {
827 	const struct ctl_table *tbl;
828 
829 	tbl = net->core.sysctl_hdr->ctl_table_arg;
830 	unregister_net_sysctl_table(net->core.sysctl_hdr);
831 	BUG_ON(tbl == netns_core_table);
832 #if IS_ENABLED(CONFIG_RPS)
833 	kfree(net->core.rps_default_mask);
834 #endif
835 	kfree(tbl);
836 }
837 
838 static __net_initdata struct pernet_operations sysctl_core_ops = {
839 	.init = sysctl_core_net_init,
840 	.exit = sysctl_core_net_exit,
841 };
842 
843 static __init int sysctl_core_init(void)
844 {
845 	register_net_sysctl(&init_net, "net/core", net_core_table);
846 	return register_pernet_subsys(&sysctl_core_ops);
847 }
848 
849 fs_initcall(sysctl_core_init);
850