xref: /openbmc/linux/net/core/sysctl_net_core.c (revision fe1e8381)
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 
27 #include "dev.h"
28 
29 static int int_3600 = 3600;
30 static int min_sndbuf = SOCK_MIN_SNDBUF;
31 static int min_rcvbuf = SOCK_MIN_RCVBUF;
32 static int max_skb_frags = MAX_SKB_FRAGS;
33 static int min_mem_pcpu_rsv = SK_MEMORY_PCPU_RESERVE;
34 
35 static int net_msg_warn;	/* Unused, but still a sysctl */
36 
37 int sysctl_fb_tunnels_only_for_init_net __read_mostly = 0;
38 EXPORT_SYMBOL(sysctl_fb_tunnels_only_for_init_net);
39 
40 /* 0 - Keep current behavior:
41  *     IPv4: inherit all current settings from init_net
42  *     IPv6: reset all settings to default
43  * 1 - Both inherit all current settings from init_net
44  * 2 - Both reset all settings to default
45  * 3 - Both inherit all settings from current netns
46  */
47 int sysctl_devconf_inherit_init_net __read_mostly;
48 EXPORT_SYMBOL(sysctl_devconf_inherit_init_net);
49 
50 #if IS_ENABLED(CONFIG_NET_FLOW_LIMIT) || IS_ENABLED(CONFIG_RPS)
dump_cpumask(void * buffer,size_t * lenp,loff_t * ppos,struct cpumask * mask)51 static void dump_cpumask(void *buffer, size_t *lenp, loff_t *ppos,
52 			 struct cpumask *mask)
53 {
54 	char kbuf[128];
55 	int len;
56 
57 	if (*ppos || !*lenp) {
58 		*lenp = 0;
59 		return;
60 	}
61 
62 	len = min(sizeof(kbuf) - 1, *lenp);
63 	len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
64 	if (!len) {
65 		*lenp = 0;
66 		return;
67 	}
68 
69 	if (len < *lenp)
70 		kbuf[len++] = '\n';
71 	memcpy(buffer, kbuf, len);
72 	*lenp = len;
73 	*ppos += len;
74 }
75 #endif
76 
77 #ifdef CONFIG_RPS
78 
rps_default_mask_cow_alloc(struct net * net)79 static struct cpumask *rps_default_mask_cow_alloc(struct net *net)
80 {
81 	struct cpumask *rps_default_mask;
82 
83 	if (net->core.rps_default_mask)
84 		return net->core.rps_default_mask;
85 
86 	rps_default_mask = kzalloc(cpumask_size(), GFP_KERNEL);
87 	if (!rps_default_mask)
88 		return NULL;
89 
90 	/* pairs with READ_ONCE in rx_queue_default_mask() */
91 	WRITE_ONCE(net->core.rps_default_mask, rps_default_mask);
92 	return rps_default_mask;
93 }
94 
rps_default_mask_sysctl(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)95 static int rps_default_mask_sysctl(struct ctl_table *table, int write,
96 				   void *buffer, size_t *lenp, loff_t *ppos)
97 {
98 	struct net *net = (struct net *)table->data;
99 	int err = 0;
100 
101 	rtnl_lock();
102 	if (write) {
103 		struct cpumask *rps_default_mask = rps_default_mask_cow_alloc(net);
104 
105 		err = -ENOMEM;
106 		if (!rps_default_mask)
107 			goto done;
108 
109 		err = cpumask_parse(buffer, rps_default_mask);
110 		if (err)
111 			goto done;
112 
113 		err = rps_cpumask_housekeeping(rps_default_mask);
114 		if (err)
115 			goto done;
116 	} else {
117 		dump_cpumask(buffer, lenp, ppos,
118 			     net->core.rps_default_mask ? : cpu_none_mask);
119 	}
120 
121 done:
122 	rtnl_unlock();
123 	return err;
124 }
125 
rps_sock_flow_sysctl(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)126 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
127 				void *buffer, size_t *lenp, loff_t *ppos)
128 {
129 	unsigned int orig_size, size;
130 	int ret, i;
131 	struct ctl_table tmp = {
132 		.data = &size,
133 		.maxlen = sizeof(size),
134 		.mode = table->mode
135 	};
136 	struct rps_sock_flow_table *orig_sock_table, *sock_table;
137 	static DEFINE_MUTEX(sock_flow_mutex);
138 
139 	mutex_lock(&sock_flow_mutex);
140 
141 	orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
142 					lockdep_is_held(&sock_flow_mutex));
143 	size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
144 
145 	ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
146 
147 	if (write) {
148 		if (size) {
149 			if (size > 1<<29) {
150 				/* Enforce limit to prevent overflow */
151 				mutex_unlock(&sock_flow_mutex);
152 				return -EINVAL;
153 			}
154 			size = roundup_pow_of_two(size);
155 			if (size != orig_size) {
156 				sock_table =
157 				    vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
158 				if (!sock_table) {
159 					mutex_unlock(&sock_flow_mutex);
160 					return -ENOMEM;
161 				}
162 				rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
163 				sock_table->mask = size - 1;
164 			} else
165 				sock_table = orig_sock_table;
166 
167 			for (i = 0; i < size; i++)
168 				sock_table->ents[i] = RPS_NO_CPU;
169 		} else
170 			sock_table = NULL;
171 
172 		if (sock_table != orig_sock_table) {
173 			rcu_assign_pointer(rps_sock_flow_table, sock_table);
174 			if (sock_table) {
175 				static_branch_inc(&rps_needed);
176 				static_branch_inc(&rfs_needed);
177 			}
178 			if (orig_sock_table) {
179 				static_branch_dec(&rps_needed);
180 				static_branch_dec(&rfs_needed);
181 				kvfree_rcu_mightsleep(orig_sock_table);
182 			}
183 		}
184 	}
185 
186 	mutex_unlock(&sock_flow_mutex);
187 
188 	return ret;
189 }
190 #endif /* CONFIG_RPS */
191 
192 #ifdef CONFIG_NET_FLOW_LIMIT
193 static DEFINE_MUTEX(flow_limit_update_mutex);
194 
flow_limit_cpu_sysctl(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)195 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
196 				 void *buffer, size_t *lenp, loff_t *ppos)
197 {
198 	struct sd_flow_limit *cur;
199 	struct softnet_data *sd;
200 	cpumask_var_t mask;
201 	int i, len, ret = 0;
202 
203 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
204 		return -ENOMEM;
205 
206 	if (write) {
207 		ret = cpumask_parse(buffer, mask);
208 		if (ret)
209 			goto done;
210 
211 		mutex_lock(&flow_limit_update_mutex);
212 		len = sizeof(*cur) + netdev_flow_limit_table_len;
213 		for_each_possible_cpu(i) {
214 			sd = &per_cpu(softnet_data, i);
215 			cur = rcu_dereference_protected(sd->flow_limit,
216 				     lockdep_is_held(&flow_limit_update_mutex));
217 			if (cur && !cpumask_test_cpu(i, mask)) {
218 				RCU_INIT_POINTER(sd->flow_limit, NULL);
219 				kfree_rcu_mightsleep(cur);
220 			} else if (!cur && cpumask_test_cpu(i, mask)) {
221 				cur = kzalloc_node(len, GFP_KERNEL,
222 						   cpu_to_node(i));
223 				if (!cur) {
224 					/* not unwinding previous changes */
225 					ret = -ENOMEM;
226 					goto write_unlock;
227 				}
228 				cur->num_buckets = netdev_flow_limit_table_len;
229 				rcu_assign_pointer(sd->flow_limit, cur);
230 			}
231 		}
232 write_unlock:
233 		mutex_unlock(&flow_limit_update_mutex);
234 	} else {
235 		cpumask_clear(mask);
236 		rcu_read_lock();
237 		for_each_possible_cpu(i) {
238 			sd = &per_cpu(softnet_data, i);
239 			if (rcu_dereference(sd->flow_limit))
240 				cpumask_set_cpu(i, mask);
241 		}
242 		rcu_read_unlock();
243 
244 		dump_cpumask(buffer, lenp, ppos, mask);
245 	}
246 
247 done:
248 	free_cpumask_var(mask);
249 	return ret;
250 }
251 
flow_limit_table_len_sysctl(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)252 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
253 				       void *buffer, size_t *lenp, loff_t *ppos)
254 {
255 	unsigned int old, *ptr;
256 	int ret;
257 
258 	mutex_lock(&flow_limit_update_mutex);
259 
260 	ptr = table->data;
261 	old = *ptr;
262 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
263 	if (!ret && write && !is_power_of_2(*ptr)) {
264 		*ptr = old;
265 		ret = -EINVAL;
266 	}
267 
268 	mutex_unlock(&flow_limit_update_mutex);
269 	return ret;
270 }
271 #endif /* CONFIG_NET_FLOW_LIMIT */
272 
273 #ifdef CONFIG_NET_SCHED
set_default_qdisc(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)274 static int set_default_qdisc(struct ctl_table *table, int write,
275 			     void *buffer, size_t *lenp, loff_t *ppos)
276 {
277 	char id[IFNAMSIZ];
278 	struct ctl_table tbl = {
279 		.data = id,
280 		.maxlen = IFNAMSIZ,
281 	};
282 	int ret;
283 
284 	qdisc_get_default(id, IFNAMSIZ);
285 
286 	ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
287 	if (write && ret == 0)
288 		ret = qdisc_set_default(id);
289 	return ret;
290 }
291 #endif
292 
proc_do_dev_weight(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)293 static int proc_do_dev_weight(struct ctl_table *table, int write,
294 			   void *buffer, size_t *lenp, loff_t *ppos)
295 {
296 	static DEFINE_MUTEX(dev_weight_mutex);
297 	int ret, weight;
298 
299 	mutex_lock(&dev_weight_mutex);
300 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
301 	if (!ret && write) {
302 		weight = READ_ONCE(weight_p);
303 		WRITE_ONCE(dev_rx_weight, weight * dev_weight_rx_bias);
304 		WRITE_ONCE(dev_tx_weight, weight * dev_weight_tx_bias);
305 	}
306 	mutex_unlock(&dev_weight_mutex);
307 
308 	return ret;
309 }
310 
proc_do_rss_key(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)311 static int proc_do_rss_key(struct ctl_table *table, int write,
312 			   void *buffer, size_t *lenp, loff_t *ppos)
313 {
314 	struct ctl_table fake_table;
315 	char buf[NETDEV_RSS_KEY_LEN * 3];
316 
317 	snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
318 	fake_table.data = buf;
319 	fake_table.maxlen = sizeof(buf);
320 	return proc_dostring(&fake_table, write, buffer, lenp, ppos);
321 }
322 
323 #ifdef CONFIG_BPF_JIT
proc_dointvec_minmax_bpf_enable(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)324 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write,
325 					   void *buffer, size_t *lenp,
326 					   loff_t *ppos)
327 {
328 	int ret, jit_enable = *(int *)table->data;
329 	int min = *(int *)table->extra1;
330 	int max = *(int *)table->extra2;
331 	struct ctl_table tmp = *table;
332 
333 	if (write && !capable(CAP_SYS_ADMIN))
334 		return -EPERM;
335 
336 	tmp.data = &jit_enable;
337 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
338 	if (write && !ret) {
339 		if (jit_enable < 2 ||
340 		    (jit_enable == 2 && bpf_dump_raw_ok(current_cred()))) {
341 			*(int *)table->data = jit_enable;
342 			if (jit_enable == 2)
343 				pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
344 		} else {
345 			ret = -EPERM;
346 		}
347 	}
348 
349 	if (write && ret && min == max)
350 		pr_info_once("CONFIG_BPF_JIT_ALWAYS_ON is enabled, bpf_jit_enable is permanently set to 1.\n");
351 
352 	return ret;
353 }
354 
355 # ifdef CONFIG_HAVE_EBPF_JIT
356 static int
proc_dointvec_minmax_bpf_restricted(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)357 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write,
358 				    void *buffer, size_t *lenp, loff_t *ppos)
359 {
360 	if (!capable(CAP_SYS_ADMIN))
361 		return -EPERM;
362 
363 	return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
364 }
365 # endif /* CONFIG_HAVE_EBPF_JIT */
366 
367 static int
proc_dolongvec_minmax_bpf_restricted(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)368 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
369 				     void *buffer, size_t *lenp, loff_t *ppos)
370 {
371 	if (!capable(CAP_SYS_ADMIN))
372 		return -EPERM;
373 
374 	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
375 }
376 #endif
377 
378 static struct ctl_table net_core_table[] = {
379 	{
380 		.procname	= "wmem_max",
381 		.data		= &sysctl_wmem_max,
382 		.maxlen		= sizeof(int),
383 		.mode		= 0644,
384 		.proc_handler	= proc_dointvec_minmax,
385 		.extra1		= &min_sndbuf,
386 	},
387 	{
388 		.procname	= "rmem_max",
389 		.data		= &sysctl_rmem_max,
390 		.maxlen		= sizeof(int),
391 		.mode		= 0644,
392 		.proc_handler	= proc_dointvec_minmax,
393 		.extra1		= &min_rcvbuf,
394 	},
395 	{
396 		.procname	= "wmem_default",
397 		.data		= &sysctl_wmem_default,
398 		.maxlen		= sizeof(int),
399 		.mode		= 0644,
400 		.proc_handler	= proc_dointvec_minmax,
401 		.extra1		= &min_sndbuf,
402 	},
403 	{
404 		.procname	= "rmem_default",
405 		.data		= &sysctl_rmem_default,
406 		.maxlen		= sizeof(int),
407 		.mode		= 0644,
408 		.proc_handler	= proc_dointvec_minmax,
409 		.extra1		= &min_rcvbuf,
410 	},
411 	{
412 		.procname	= "mem_pcpu_rsv",
413 		.data		= &sysctl_mem_pcpu_rsv,
414 		.maxlen		= sizeof(int),
415 		.mode		= 0644,
416 		.proc_handler	= proc_dointvec_minmax,
417 		.extra1		= &min_mem_pcpu_rsv,
418 	},
419 	{
420 		.procname	= "dev_weight",
421 		.data		= &weight_p,
422 		.maxlen		= sizeof(int),
423 		.mode		= 0644,
424 		.proc_handler	= proc_do_dev_weight,
425 	},
426 	{
427 		.procname	= "dev_weight_rx_bias",
428 		.data		= &dev_weight_rx_bias,
429 		.maxlen		= sizeof(int),
430 		.mode		= 0644,
431 		.proc_handler	= proc_do_dev_weight,
432 	},
433 	{
434 		.procname	= "dev_weight_tx_bias",
435 		.data		= &dev_weight_tx_bias,
436 		.maxlen		= sizeof(int),
437 		.mode		= 0644,
438 		.proc_handler	= proc_do_dev_weight,
439 	},
440 	{
441 		.procname	= "netdev_max_backlog",
442 		.data		= &netdev_max_backlog,
443 		.maxlen		= sizeof(int),
444 		.mode		= 0644,
445 		.proc_handler	= proc_dointvec
446 	},
447 	{
448 		.procname	= "netdev_rss_key",
449 		.data		= &netdev_rss_key,
450 		.maxlen		= sizeof(int),
451 		.mode		= 0444,
452 		.proc_handler	= proc_do_rss_key,
453 	},
454 #ifdef CONFIG_BPF_JIT
455 	{
456 		.procname	= "bpf_jit_enable",
457 		.data		= &bpf_jit_enable,
458 		.maxlen		= sizeof(int),
459 		.mode		= 0644,
460 		.proc_handler	= proc_dointvec_minmax_bpf_enable,
461 # ifdef CONFIG_BPF_JIT_ALWAYS_ON
462 		.extra1		= SYSCTL_ONE,
463 		.extra2		= SYSCTL_ONE,
464 # else
465 		.extra1		= SYSCTL_ZERO,
466 		.extra2		= SYSCTL_TWO,
467 # endif
468 	},
469 # ifdef CONFIG_HAVE_EBPF_JIT
470 	{
471 		.procname	= "bpf_jit_harden",
472 		.data		= &bpf_jit_harden,
473 		.maxlen		= sizeof(int),
474 		.mode		= 0600,
475 		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
476 		.extra1		= SYSCTL_ZERO,
477 		.extra2		= SYSCTL_TWO,
478 	},
479 	{
480 		.procname	= "bpf_jit_kallsyms",
481 		.data		= &bpf_jit_kallsyms,
482 		.maxlen		= sizeof(int),
483 		.mode		= 0600,
484 		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
485 		.extra1		= SYSCTL_ZERO,
486 		.extra2		= SYSCTL_ONE,
487 	},
488 # endif
489 	{
490 		.procname	= "bpf_jit_limit",
491 		.data		= &bpf_jit_limit,
492 		.maxlen		= sizeof(long),
493 		.mode		= 0600,
494 		.proc_handler	= proc_dolongvec_minmax_bpf_restricted,
495 		.extra1		= SYSCTL_LONG_ONE,
496 		.extra2		= &bpf_jit_limit_max,
497 	},
498 #endif
499 	{
500 		.procname	= "netdev_tstamp_prequeue",
501 		.data		= &netdev_tstamp_prequeue,
502 		.maxlen		= sizeof(int),
503 		.mode		= 0644,
504 		.proc_handler	= proc_dointvec
505 	},
506 	{
507 		.procname	= "message_cost",
508 		.data		= &net_ratelimit_state.interval,
509 		.maxlen		= sizeof(int),
510 		.mode		= 0644,
511 		.proc_handler	= proc_dointvec_jiffies,
512 	},
513 	{
514 		.procname	= "message_burst",
515 		.data		= &net_ratelimit_state.burst,
516 		.maxlen		= sizeof(int),
517 		.mode		= 0644,
518 		.proc_handler	= proc_dointvec,
519 	},
520 	{
521 		.procname	= "optmem_max",
522 		.data		= &sysctl_optmem_max,
523 		.maxlen		= sizeof(int),
524 		.mode		= 0644,
525 		.proc_handler	= proc_dointvec
526 	},
527 	{
528 		.procname	= "tstamp_allow_data",
529 		.data		= &sysctl_tstamp_allow_data,
530 		.maxlen		= sizeof(int),
531 		.mode		= 0644,
532 		.proc_handler	= proc_dointvec_minmax,
533 		.extra1		= SYSCTL_ZERO,
534 		.extra2		= SYSCTL_ONE
535 	},
536 #ifdef CONFIG_RPS
537 	{
538 		.procname	= "rps_sock_flow_entries",
539 		.maxlen		= sizeof(int),
540 		.mode		= 0644,
541 		.proc_handler	= rps_sock_flow_sysctl
542 	},
543 #endif
544 #ifdef CONFIG_NET_FLOW_LIMIT
545 	{
546 		.procname	= "flow_limit_cpu_bitmap",
547 		.mode		= 0644,
548 		.proc_handler	= flow_limit_cpu_sysctl
549 	},
550 	{
551 		.procname	= "flow_limit_table_len",
552 		.data		= &netdev_flow_limit_table_len,
553 		.maxlen		= sizeof(int),
554 		.mode		= 0644,
555 		.proc_handler	= flow_limit_table_len_sysctl
556 	},
557 #endif /* CONFIG_NET_FLOW_LIMIT */
558 #ifdef CONFIG_NET_RX_BUSY_POLL
559 	{
560 		.procname	= "busy_poll",
561 		.data		= &sysctl_net_busy_poll,
562 		.maxlen		= sizeof(unsigned int),
563 		.mode		= 0644,
564 		.proc_handler	= proc_dointvec_minmax,
565 		.extra1		= SYSCTL_ZERO,
566 	},
567 	{
568 		.procname	= "busy_read",
569 		.data		= &sysctl_net_busy_read,
570 		.maxlen		= sizeof(unsigned int),
571 		.mode		= 0644,
572 		.proc_handler	= proc_dointvec_minmax,
573 		.extra1		= SYSCTL_ZERO,
574 	},
575 #endif
576 #ifdef CONFIG_NET_SCHED
577 	{
578 		.procname	= "default_qdisc",
579 		.mode		= 0644,
580 		.maxlen		= IFNAMSIZ,
581 		.proc_handler	= set_default_qdisc
582 	},
583 #endif
584 	{
585 		.procname	= "netdev_budget",
586 		.data		= &netdev_budget,
587 		.maxlen		= sizeof(int),
588 		.mode		= 0644,
589 		.proc_handler	= proc_dointvec
590 	},
591 	{
592 		.procname	= "warnings",
593 		.data		= &net_msg_warn,
594 		.maxlen		= sizeof(int),
595 		.mode		= 0644,
596 		.proc_handler	= proc_dointvec
597 	},
598 	{
599 		.procname	= "max_skb_frags",
600 		.data		= &sysctl_max_skb_frags,
601 		.maxlen		= sizeof(int),
602 		.mode		= 0644,
603 		.proc_handler	= proc_dointvec_minmax,
604 		.extra1		= SYSCTL_ONE,
605 		.extra2		= &max_skb_frags,
606 	},
607 	{
608 		.procname	= "netdev_budget_usecs",
609 		.data		= &netdev_budget_usecs,
610 		.maxlen		= sizeof(unsigned int),
611 		.mode		= 0644,
612 		.proc_handler	= proc_dointvec_minmax,
613 		.extra1		= SYSCTL_ZERO,
614 	},
615 	{
616 		.procname	= "fb_tunnels_only_for_init_net",
617 		.data		= &sysctl_fb_tunnels_only_for_init_net,
618 		.maxlen		= sizeof(int),
619 		.mode		= 0644,
620 		.proc_handler	= proc_dointvec_minmax,
621 		.extra1		= SYSCTL_ZERO,
622 		.extra2		= SYSCTL_TWO,
623 	},
624 	{
625 		.procname	= "devconf_inherit_init_net",
626 		.data		= &sysctl_devconf_inherit_init_net,
627 		.maxlen		= sizeof(int),
628 		.mode		= 0644,
629 		.proc_handler	= proc_dointvec_minmax,
630 		.extra1		= SYSCTL_ZERO,
631 		.extra2		= SYSCTL_THREE,
632 	},
633 	{
634 		.procname	= "high_order_alloc_disable",
635 		.data		= &net_high_order_alloc_disable_key.key,
636 		.maxlen         = sizeof(net_high_order_alloc_disable_key),
637 		.mode		= 0644,
638 		.proc_handler	= proc_do_static_key,
639 	},
640 	{
641 		.procname	= "gro_normal_batch",
642 		.data		= &gro_normal_batch,
643 		.maxlen		= sizeof(unsigned int),
644 		.mode		= 0644,
645 		.proc_handler	= proc_dointvec_minmax,
646 		.extra1		= SYSCTL_ONE,
647 	},
648 	{
649 		.procname	= "netdev_unregister_timeout_secs",
650 		.data		= &netdev_unregister_timeout_secs,
651 		.maxlen		= sizeof(unsigned int),
652 		.mode		= 0644,
653 		.proc_handler	= proc_dointvec_minmax,
654 		.extra1		= SYSCTL_ONE,
655 		.extra2		= &int_3600,
656 	},
657 	{
658 		.procname	= "skb_defer_max",
659 		.data		= &sysctl_skb_defer_max,
660 		.maxlen		= sizeof(unsigned int),
661 		.mode		= 0644,
662 		.proc_handler	= proc_dointvec_minmax,
663 		.extra1		= SYSCTL_ZERO,
664 	},
665 	{ }
666 };
667 
668 static struct ctl_table netns_core_table[] = {
669 #if IS_ENABLED(CONFIG_RPS)
670 	{
671 		.procname	= "rps_default_mask",
672 		.data		= &init_net,
673 		.mode		= 0644,
674 		.proc_handler	= rps_default_mask_sysctl
675 	},
676 #endif
677 	{
678 		.procname	= "somaxconn",
679 		.data		= &init_net.core.sysctl_somaxconn,
680 		.maxlen		= sizeof(int),
681 		.mode		= 0644,
682 		.extra1		= SYSCTL_ZERO,
683 		.proc_handler	= proc_dointvec_minmax
684 	},
685 	{
686 		.procname	= "txrehash",
687 		.data		= &init_net.core.sysctl_txrehash,
688 		.maxlen		= sizeof(u8),
689 		.mode		= 0644,
690 		.extra1		= SYSCTL_ZERO,
691 		.extra2		= SYSCTL_ONE,
692 		.proc_handler	= proc_dou8vec_minmax,
693 	},
694 	{ }
695 };
696 
fb_tunnels_only_for_init_net_sysctl_setup(char * str)697 static int __init fb_tunnels_only_for_init_net_sysctl_setup(char *str)
698 {
699 	/* fallback tunnels for initns only */
700 	if (!strncmp(str, "initns", 6))
701 		sysctl_fb_tunnels_only_for_init_net = 1;
702 	/* no fallback tunnels anywhere */
703 	else if (!strncmp(str, "none", 4))
704 		sysctl_fb_tunnels_only_for_init_net = 2;
705 
706 	return 1;
707 }
708 __setup("fb_tunnels=", fb_tunnels_only_for_init_net_sysctl_setup);
709 
sysctl_core_net_init(struct net * net)710 static __net_init int sysctl_core_net_init(struct net *net)
711 {
712 	struct ctl_table *tbl, *tmp;
713 
714 	tbl = netns_core_table;
715 	if (!net_eq(net, &init_net)) {
716 		tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
717 		if (tbl == NULL)
718 			goto err_dup;
719 
720 		for (tmp = tbl; tmp->procname; tmp++)
721 			tmp->data += (char *)net - (char *)&init_net;
722 	}
723 
724 	net->core.sysctl_hdr = register_net_sysctl_sz(net, "net/core", tbl,
725 						      ARRAY_SIZE(netns_core_table));
726 	if (net->core.sysctl_hdr == NULL)
727 		goto err_reg;
728 
729 	return 0;
730 
731 err_reg:
732 	if (tbl != netns_core_table)
733 		kfree(tbl);
734 err_dup:
735 	return -ENOMEM;
736 }
737 
sysctl_core_net_exit(struct net * net)738 static __net_exit void sysctl_core_net_exit(struct net *net)
739 {
740 	struct ctl_table *tbl;
741 
742 	tbl = net->core.sysctl_hdr->ctl_table_arg;
743 	unregister_net_sysctl_table(net->core.sysctl_hdr);
744 	BUG_ON(tbl == netns_core_table);
745 #if IS_ENABLED(CONFIG_RPS)
746 	kfree(net->core.rps_default_mask);
747 #endif
748 	kfree(tbl);
749 }
750 
751 static __net_initdata struct pernet_operations sysctl_core_ops = {
752 	.init = sysctl_core_net_init,
753 	.exit = sysctl_core_net_exit,
754 };
755 
sysctl_core_init(void)756 static __init int sysctl_core_init(void)
757 {
758 	register_net_sysctl(&init_net, "net/core", net_core_table);
759 	return register_pernet_subsys(&sysctl_core_ops);
760 }
761 
762 fs_initcall(sysctl_core_init);
763