xref: /openbmc/linux/net/core/sysctl_net_core.c (revision 0f9b4c3ca5fdf3e177266ef994071b1a03f07318)
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 static int netdev_budget_usecs_min = 2 * USEC_PER_SEC / HZ;
35 
36 static int net_msg_warn;	/* Unused, but still a sysctl */
37 
38 int sysctl_fb_tunnels_only_for_init_net __read_mostly = 0;
39 EXPORT_SYMBOL(sysctl_fb_tunnels_only_for_init_net);
40 
41 /* 0 - Keep current behavior:
42  *     IPv4: inherit all current settings from init_net
43  *     IPv6: reset all settings to default
44  * 1 - Both inherit all current settings from init_net
45  * 2 - Both reset all settings to default
46  * 3 - Both inherit all settings from current netns
47  */
48 int sysctl_devconf_inherit_init_net __read_mostly;
49 EXPORT_SYMBOL(sysctl_devconf_inherit_init_net);
50 
51 #if IS_ENABLED(CONFIG_NET_FLOW_LIMIT) || IS_ENABLED(CONFIG_RPS)
dump_cpumask(void * buffer,size_t * lenp,loff_t * ppos,struct cpumask * mask)52 static void dump_cpumask(void *buffer, size_t *lenp, loff_t *ppos,
53 			 struct cpumask *mask)
54 {
55 	char kbuf[128];
56 	int len;
57 
58 	if (*ppos || !*lenp) {
59 		*lenp = 0;
60 		return;
61 	}
62 
63 	len = min(sizeof(kbuf) - 1, *lenp);
64 	len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
65 	if (!len) {
66 		*lenp = 0;
67 		return;
68 	}
69 
70 	if (len < *lenp)
71 		kbuf[len++] = '\n';
72 	memcpy(buffer, kbuf, len);
73 	*lenp = len;
74 	*ppos += len;
75 }
76 #endif
77 
78 #ifdef CONFIG_RPS
79 
rps_default_mask_cow_alloc(struct net * net)80 static struct cpumask *rps_default_mask_cow_alloc(struct net *net)
81 {
82 	struct cpumask *rps_default_mask;
83 
84 	if (net->core.rps_default_mask)
85 		return net->core.rps_default_mask;
86 
87 	rps_default_mask = kzalloc(cpumask_size(), GFP_KERNEL);
88 	if (!rps_default_mask)
89 		return NULL;
90 
91 	/* pairs with READ_ONCE in rx_queue_default_mask() */
92 	WRITE_ONCE(net->core.rps_default_mask, rps_default_mask);
93 	return rps_default_mask;
94 }
95 
rps_default_mask_sysctl(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)96 static int rps_default_mask_sysctl(struct ctl_table *table, int write,
97 				   void *buffer, size_t *lenp, loff_t *ppos)
98 {
99 	struct net *net = (struct net *)table->data;
100 	int err = 0;
101 
102 	rtnl_lock();
103 	if (write) {
104 		struct cpumask *rps_default_mask = rps_default_mask_cow_alloc(net);
105 
106 		err = -ENOMEM;
107 		if (!rps_default_mask)
108 			goto done;
109 
110 		err = cpumask_parse(buffer, rps_default_mask);
111 		if (err)
112 			goto done;
113 
114 		err = rps_cpumask_housekeeping(rps_default_mask);
115 		if (err)
116 			goto done;
117 	} else {
118 		dump_cpumask(buffer, lenp, ppos,
119 			     net->core.rps_default_mask ? : cpu_none_mask);
120 	}
121 
122 done:
123 	rtnl_unlock();
124 	return err;
125 }
126 
rps_sock_flow_sysctl(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)127 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
128 				void *buffer, size_t *lenp, loff_t *ppos)
129 {
130 	unsigned int orig_size, size;
131 	int ret, i;
132 	struct ctl_table tmp = {
133 		.data = &size,
134 		.maxlen = sizeof(size),
135 		.mode = table->mode
136 	};
137 	struct rps_sock_flow_table *orig_sock_table, *sock_table;
138 	static DEFINE_MUTEX(sock_flow_mutex);
139 
140 	mutex_lock(&sock_flow_mutex);
141 
142 	orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
143 					lockdep_is_held(&sock_flow_mutex));
144 	size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
145 
146 	ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
147 
148 	if (write) {
149 		if (size) {
150 			if (size > 1<<29) {
151 				/* Enforce limit to prevent overflow */
152 				mutex_unlock(&sock_flow_mutex);
153 				return -EINVAL;
154 			}
155 			size = roundup_pow_of_two(size);
156 			if (size != orig_size) {
157 				sock_table =
158 				    vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
159 				if (!sock_table) {
160 					mutex_unlock(&sock_flow_mutex);
161 					return -ENOMEM;
162 				}
163 				rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
164 				sock_table->mask = size - 1;
165 			} else
166 				sock_table = orig_sock_table;
167 
168 			for (i = 0; i < size; i++)
169 				sock_table->ents[i] = RPS_NO_CPU;
170 		} else
171 			sock_table = NULL;
172 
173 		if (sock_table != orig_sock_table) {
174 			rcu_assign_pointer(rps_sock_flow_table, sock_table);
175 			if (sock_table) {
176 				static_branch_inc(&rps_needed);
177 				static_branch_inc(&rfs_needed);
178 			}
179 			if (orig_sock_table) {
180 				static_branch_dec(&rps_needed);
181 				static_branch_dec(&rfs_needed);
182 				kvfree_rcu_mightsleep(orig_sock_table);
183 			}
184 		}
185 	}
186 
187 	mutex_unlock(&sock_flow_mutex);
188 
189 	return ret;
190 }
191 #endif /* CONFIG_RPS */
192 
193 #ifdef CONFIG_NET_FLOW_LIMIT
194 static DEFINE_MUTEX(flow_limit_update_mutex);
195 
flow_limit_cpu_sysctl(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)196 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
197 				 void *buffer, size_t *lenp, loff_t *ppos)
198 {
199 	struct sd_flow_limit *cur;
200 	struct softnet_data *sd;
201 	cpumask_var_t mask;
202 	int i, len, ret = 0;
203 
204 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
205 		return -ENOMEM;
206 
207 	if (write) {
208 		ret = cpumask_parse(buffer, mask);
209 		if (ret)
210 			goto done;
211 
212 		mutex_lock(&flow_limit_update_mutex);
213 		len = sizeof(*cur) + netdev_flow_limit_table_len;
214 		for_each_possible_cpu(i) {
215 			sd = &per_cpu(softnet_data, i);
216 			cur = rcu_dereference_protected(sd->flow_limit,
217 				     lockdep_is_held(&flow_limit_update_mutex));
218 			if (cur && !cpumask_test_cpu(i, mask)) {
219 				RCU_INIT_POINTER(sd->flow_limit, NULL);
220 				kfree_rcu_mightsleep(cur);
221 			} else if (!cur && cpumask_test_cpu(i, mask)) {
222 				cur = kzalloc_node(len, GFP_KERNEL,
223 						   cpu_to_node(i));
224 				if (!cur) {
225 					/* not unwinding previous changes */
226 					ret = -ENOMEM;
227 					goto write_unlock;
228 				}
229 				cur->num_buckets = netdev_flow_limit_table_len;
230 				rcu_assign_pointer(sd->flow_limit, cur);
231 			}
232 		}
233 write_unlock:
234 		mutex_unlock(&flow_limit_update_mutex);
235 	} else {
236 		cpumask_clear(mask);
237 		rcu_read_lock();
238 		for_each_possible_cpu(i) {
239 			sd = &per_cpu(softnet_data, i);
240 			if (rcu_dereference(sd->flow_limit))
241 				cpumask_set_cpu(i, mask);
242 		}
243 		rcu_read_unlock();
244 
245 		dump_cpumask(buffer, lenp, ppos, mask);
246 	}
247 
248 done:
249 	free_cpumask_var(mask);
250 	return ret;
251 }
252 
flow_limit_table_len_sysctl(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)253 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
254 				       void *buffer, size_t *lenp, loff_t *ppos)
255 {
256 	unsigned int old, *ptr;
257 	int ret;
258 
259 	mutex_lock(&flow_limit_update_mutex);
260 
261 	ptr = table->data;
262 	old = *ptr;
263 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
264 	if (!ret && write && !is_power_of_2(*ptr)) {
265 		*ptr = old;
266 		ret = -EINVAL;
267 	}
268 
269 	mutex_unlock(&flow_limit_update_mutex);
270 	return ret;
271 }
272 #endif /* CONFIG_NET_FLOW_LIMIT */
273 
274 #ifdef CONFIG_NET_SCHED
set_default_qdisc(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)275 static int set_default_qdisc(struct ctl_table *table, int write,
276 			     void *buffer, size_t *lenp, loff_t *ppos)
277 {
278 	char id[IFNAMSIZ];
279 	struct ctl_table tbl = {
280 		.data = id,
281 		.maxlen = IFNAMSIZ,
282 	};
283 	int ret;
284 
285 	qdisc_get_default(id, IFNAMSIZ);
286 
287 	ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
288 	if (write && ret == 0)
289 		ret = qdisc_set_default(id);
290 	return ret;
291 }
292 #endif
293 
proc_do_dev_weight(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)294 static int proc_do_dev_weight(struct ctl_table *table, int write,
295 			   void *buffer, size_t *lenp, loff_t *ppos)
296 {
297 	static DEFINE_MUTEX(dev_weight_mutex);
298 	int ret, weight;
299 
300 	mutex_lock(&dev_weight_mutex);
301 	ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
302 	if (!ret && write) {
303 		weight = READ_ONCE(weight_p);
304 		WRITE_ONCE(dev_rx_weight, weight * dev_weight_rx_bias);
305 		WRITE_ONCE(dev_tx_weight, weight * dev_weight_tx_bias);
306 	}
307 	mutex_unlock(&dev_weight_mutex);
308 
309 	return ret;
310 }
311 
proc_do_rss_key(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)312 static int proc_do_rss_key(struct ctl_table *table, int write,
313 			   void *buffer, size_t *lenp, loff_t *ppos)
314 {
315 	struct ctl_table fake_table;
316 	char buf[NETDEV_RSS_KEY_LEN * 3];
317 
318 	snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
319 	fake_table.data = buf;
320 	fake_table.maxlen = sizeof(buf);
321 	return proc_dostring(&fake_table, write, buffer, lenp, ppos);
322 }
323 
324 #ifdef CONFIG_BPF_JIT
proc_dointvec_minmax_bpf_enable(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)325 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write,
326 					   void *buffer, size_t *lenp,
327 					   loff_t *ppos)
328 {
329 	int ret, jit_enable = *(int *)table->data;
330 	int min = *(int *)table->extra1;
331 	int max = *(int *)table->extra2;
332 	struct ctl_table tmp = *table;
333 
334 	if (write && !capable(CAP_SYS_ADMIN))
335 		return -EPERM;
336 
337 	tmp.data = &jit_enable;
338 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
339 	if (write && !ret) {
340 		if (jit_enable < 2 ||
341 		    (jit_enable == 2 && bpf_dump_raw_ok(current_cred()))) {
342 			*(int *)table->data = jit_enable;
343 			if (jit_enable == 2)
344 				pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
345 		} else {
346 			ret = -EPERM;
347 		}
348 	}
349 
350 	if (write && ret && min == max)
351 		pr_info_once("CONFIG_BPF_JIT_ALWAYS_ON is enabled, bpf_jit_enable is permanently set to 1.\n");
352 
353 	return ret;
354 }
355 
356 # ifdef CONFIG_HAVE_EBPF_JIT
357 static int
proc_dointvec_minmax_bpf_restricted(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)358 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write,
359 				    void *buffer, size_t *lenp, loff_t *ppos)
360 {
361 	if (!capable(CAP_SYS_ADMIN))
362 		return -EPERM;
363 
364 	return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
365 }
366 # endif /* CONFIG_HAVE_EBPF_JIT */
367 
368 static int
proc_dolongvec_minmax_bpf_restricted(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)369 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
370 				     void *buffer, size_t *lenp, loff_t *ppos)
371 {
372 	if (!capable(CAP_SYS_ADMIN))
373 		return -EPERM;
374 
375 	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
376 }
377 #endif
378 
379 static struct ctl_table net_core_table[] = {
380 	{
381 		.procname	= "wmem_max",
382 		.data		= &sysctl_wmem_max,
383 		.maxlen		= sizeof(int),
384 		.mode		= 0644,
385 		.proc_handler	= proc_dointvec_minmax,
386 		.extra1		= &min_sndbuf,
387 	},
388 	{
389 		.procname	= "rmem_max",
390 		.data		= &sysctl_rmem_max,
391 		.maxlen		= sizeof(int),
392 		.mode		= 0644,
393 		.proc_handler	= proc_dointvec_minmax,
394 		.extra1		= &min_rcvbuf,
395 	},
396 	{
397 		.procname	= "wmem_default",
398 		.data		= &sysctl_wmem_default,
399 		.maxlen		= sizeof(int),
400 		.mode		= 0644,
401 		.proc_handler	= proc_dointvec_minmax,
402 		.extra1		= &min_sndbuf,
403 	},
404 	{
405 		.procname	= "rmem_default",
406 		.data		= &sysctl_rmem_default,
407 		.maxlen		= sizeof(int),
408 		.mode		= 0644,
409 		.proc_handler	= proc_dointvec_minmax,
410 		.extra1		= &min_rcvbuf,
411 	},
412 	{
413 		.procname	= "mem_pcpu_rsv",
414 		.data		= &sysctl_mem_pcpu_rsv,
415 		.maxlen		= sizeof(int),
416 		.mode		= 0644,
417 		.proc_handler	= proc_dointvec_minmax,
418 		.extra1		= &min_mem_pcpu_rsv,
419 	},
420 	{
421 		.procname	= "dev_weight",
422 		.data		= &weight_p,
423 		.maxlen		= sizeof(int),
424 		.mode		= 0644,
425 		.proc_handler	= proc_do_dev_weight,
426 		.extra1         = SYSCTL_ONE,
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 		.extra1         = SYSCTL_ONE,
435 	},
436 	{
437 		.procname	= "dev_weight_tx_bias",
438 		.data		= &dev_weight_tx_bias,
439 		.maxlen		= sizeof(int),
440 		.mode		= 0644,
441 		.proc_handler	= proc_do_dev_weight,
442 		.extra1         = SYSCTL_ONE,
443 	},
444 	{
445 		.procname	= "netdev_max_backlog",
446 		.data		= &netdev_max_backlog,
447 		.maxlen		= sizeof(int),
448 		.mode		= 0644,
449 		.proc_handler	= proc_dointvec
450 	},
451 	{
452 		.procname	= "netdev_rss_key",
453 		.data		= &netdev_rss_key,
454 		.maxlen		= sizeof(int),
455 		.mode		= 0444,
456 		.proc_handler	= proc_do_rss_key,
457 	},
458 #ifdef CONFIG_BPF_JIT
459 	{
460 		.procname	= "bpf_jit_enable",
461 		.data		= &bpf_jit_enable,
462 		.maxlen		= sizeof(int),
463 		.mode		= 0644,
464 		.proc_handler	= proc_dointvec_minmax_bpf_enable,
465 # ifdef CONFIG_BPF_JIT_ALWAYS_ON
466 		.extra1		= SYSCTL_ONE,
467 		.extra2		= SYSCTL_ONE,
468 # else
469 		.extra1		= SYSCTL_ZERO,
470 		.extra2		= SYSCTL_TWO,
471 # endif
472 	},
473 # ifdef CONFIG_HAVE_EBPF_JIT
474 	{
475 		.procname	= "bpf_jit_harden",
476 		.data		= &bpf_jit_harden,
477 		.maxlen		= sizeof(int),
478 		.mode		= 0600,
479 		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
480 		.extra1		= SYSCTL_ZERO,
481 		.extra2		= SYSCTL_TWO,
482 	},
483 	{
484 		.procname	= "bpf_jit_kallsyms",
485 		.data		= &bpf_jit_kallsyms,
486 		.maxlen		= sizeof(int),
487 		.mode		= 0600,
488 		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
489 		.extra1		= SYSCTL_ZERO,
490 		.extra2		= SYSCTL_ONE,
491 	},
492 # endif
493 	{
494 		.procname	= "bpf_jit_limit",
495 		.data		= &bpf_jit_limit,
496 		.maxlen		= sizeof(long),
497 		.mode		= 0600,
498 		.proc_handler	= proc_dolongvec_minmax_bpf_restricted,
499 		.extra1		= SYSCTL_LONG_ONE,
500 		.extra2		= &bpf_jit_limit_max,
501 	},
502 #endif
503 	{
504 		.procname	= "netdev_tstamp_prequeue",
505 		.data		= &netdev_tstamp_prequeue,
506 		.maxlen		= sizeof(int),
507 		.mode		= 0644,
508 		.proc_handler	= proc_dointvec
509 	},
510 	{
511 		.procname	= "message_cost",
512 		.data		= &net_ratelimit_state.interval,
513 		.maxlen		= sizeof(int),
514 		.mode		= 0644,
515 		.proc_handler	= proc_dointvec_jiffies,
516 	},
517 	{
518 		.procname	= "message_burst",
519 		.data		= &net_ratelimit_state.burst,
520 		.maxlen		= sizeof(int),
521 		.mode		= 0644,
522 		.proc_handler	= proc_dointvec,
523 	},
524 	{
525 		.procname	= "optmem_max",
526 		.data		= &sysctl_optmem_max,
527 		.maxlen		= sizeof(int),
528 		.mode		= 0644,
529 		.proc_handler	= proc_dointvec
530 	},
531 	{
532 		.procname	= "tstamp_allow_data",
533 		.data		= &sysctl_tstamp_allow_data,
534 		.maxlen		= sizeof(int),
535 		.mode		= 0644,
536 		.proc_handler	= proc_dointvec_minmax,
537 		.extra1		= SYSCTL_ZERO,
538 		.extra2		= SYSCTL_ONE
539 	},
540 #ifdef CONFIG_RPS
541 	{
542 		.procname	= "rps_sock_flow_entries",
543 		.maxlen		= sizeof(int),
544 		.mode		= 0644,
545 		.proc_handler	= rps_sock_flow_sysctl
546 	},
547 #endif
548 #ifdef CONFIG_NET_FLOW_LIMIT
549 	{
550 		.procname	= "flow_limit_cpu_bitmap",
551 		.mode		= 0644,
552 		.proc_handler	= flow_limit_cpu_sysctl
553 	},
554 	{
555 		.procname	= "flow_limit_table_len",
556 		.data		= &netdev_flow_limit_table_len,
557 		.maxlen		= sizeof(int),
558 		.mode		= 0644,
559 		.proc_handler	= flow_limit_table_len_sysctl
560 	},
561 #endif /* CONFIG_NET_FLOW_LIMIT */
562 #ifdef CONFIG_NET_RX_BUSY_POLL
563 	{
564 		.procname	= "busy_poll",
565 		.data		= &sysctl_net_busy_poll,
566 		.maxlen		= sizeof(unsigned int),
567 		.mode		= 0644,
568 		.proc_handler	= proc_dointvec_minmax,
569 		.extra1		= SYSCTL_ZERO,
570 	},
571 	{
572 		.procname	= "busy_read",
573 		.data		= &sysctl_net_busy_read,
574 		.maxlen		= sizeof(unsigned int),
575 		.mode		= 0644,
576 		.proc_handler	= proc_dointvec_minmax,
577 		.extra1		= SYSCTL_ZERO,
578 	},
579 #endif
580 #ifdef CONFIG_NET_SCHED
581 	{
582 		.procname	= "default_qdisc",
583 		.mode		= 0644,
584 		.maxlen		= IFNAMSIZ,
585 		.proc_handler	= set_default_qdisc
586 	},
587 #endif
588 	{
589 		.procname	= "netdev_budget",
590 		.data		= &netdev_budget,
591 		.maxlen		= sizeof(int),
592 		.mode		= 0644,
593 		.proc_handler	= proc_dointvec
594 	},
595 	{
596 		.procname	= "warnings",
597 		.data		= &net_msg_warn,
598 		.maxlen		= sizeof(int),
599 		.mode		= 0644,
600 		.proc_handler	= proc_dointvec
601 	},
602 	{
603 		.procname	= "max_skb_frags",
604 		.data		= &sysctl_max_skb_frags,
605 		.maxlen		= sizeof(int),
606 		.mode		= 0644,
607 		.proc_handler	= proc_dointvec_minmax,
608 		.extra1		= SYSCTL_ONE,
609 		.extra2		= &max_skb_frags,
610 	},
611 	{
612 		.procname	= "netdev_budget_usecs",
613 		.data		= &netdev_budget_usecs,
614 		.maxlen		= sizeof(unsigned int),
615 		.mode		= 0644,
616 		.proc_handler	= proc_dointvec_minmax,
617 		.extra1		= &netdev_budget_usecs_min,
618 	},
619 	{
620 		.procname	= "fb_tunnels_only_for_init_net",
621 		.data		= &sysctl_fb_tunnels_only_for_init_net,
622 		.maxlen		= sizeof(int),
623 		.mode		= 0644,
624 		.proc_handler	= proc_dointvec_minmax,
625 		.extra1		= SYSCTL_ZERO,
626 		.extra2		= SYSCTL_TWO,
627 	},
628 	{
629 		.procname	= "devconf_inherit_init_net",
630 		.data		= &sysctl_devconf_inherit_init_net,
631 		.maxlen		= sizeof(int),
632 		.mode		= 0644,
633 		.proc_handler	= proc_dointvec_minmax,
634 		.extra1		= SYSCTL_ZERO,
635 		.extra2		= SYSCTL_THREE,
636 	},
637 	{
638 		.procname	= "high_order_alloc_disable",
639 		.data		= &net_high_order_alloc_disable_key.key,
640 		.maxlen         = sizeof(net_high_order_alloc_disable_key),
641 		.mode		= 0644,
642 		.proc_handler	= proc_do_static_key,
643 	},
644 	{
645 		.procname	= "gro_normal_batch",
646 		.data		= &gro_normal_batch,
647 		.maxlen		= sizeof(unsigned int),
648 		.mode		= 0644,
649 		.proc_handler	= proc_dointvec_minmax,
650 		.extra1		= SYSCTL_ONE,
651 	},
652 	{
653 		.procname	= "netdev_unregister_timeout_secs",
654 		.data		= &netdev_unregister_timeout_secs,
655 		.maxlen		= sizeof(unsigned int),
656 		.mode		= 0644,
657 		.proc_handler	= proc_dointvec_minmax,
658 		.extra1		= SYSCTL_ONE,
659 		.extra2		= &int_3600,
660 	},
661 	{
662 		.procname	= "skb_defer_max",
663 		.data		= &sysctl_skb_defer_max,
664 		.maxlen		= sizeof(unsigned int),
665 		.mode		= 0644,
666 		.proc_handler	= proc_dointvec_minmax,
667 		.extra1		= SYSCTL_ZERO,
668 	},
669 	{ }
670 };
671 
672 static struct ctl_table netns_core_table[] = {
673 #if IS_ENABLED(CONFIG_RPS)
674 	{
675 		.procname	= "rps_default_mask",
676 		.data		= &init_net,
677 		.mode		= 0644,
678 		.proc_handler	= rps_default_mask_sysctl
679 	},
680 #endif
681 	{
682 		.procname	= "somaxconn",
683 		.data		= &init_net.core.sysctl_somaxconn,
684 		.maxlen		= sizeof(int),
685 		.mode		= 0644,
686 		.extra1		= SYSCTL_ZERO,
687 		.proc_handler	= proc_dointvec_minmax
688 	},
689 	{
690 		.procname	= "txrehash",
691 		.data		= &init_net.core.sysctl_txrehash,
692 		.maxlen		= sizeof(u8),
693 		.mode		= 0644,
694 		.extra1		= SYSCTL_ZERO,
695 		.extra2		= SYSCTL_ONE,
696 		.proc_handler	= proc_dou8vec_minmax,
697 	},
698 	{ }
699 };
700 
fb_tunnels_only_for_init_net_sysctl_setup(char * str)701 static int __init fb_tunnels_only_for_init_net_sysctl_setup(char *str)
702 {
703 	/* fallback tunnels for initns only */
704 	if (!strncmp(str, "initns", 6))
705 		sysctl_fb_tunnels_only_for_init_net = 1;
706 	/* no fallback tunnels anywhere */
707 	else if (!strncmp(str, "none", 4))
708 		sysctl_fb_tunnels_only_for_init_net = 2;
709 
710 	return 1;
711 }
712 __setup("fb_tunnels=", fb_tunnels_only_for_init_net_sysctl_setup);
713 
sysctl_core_net_init(struct net * net)714 static __net_init int sysctl_core_net_init(struct net *net)
715 {
716 	struct ctl_table *tbl, *tmp;
717 
718 	tbl = netns_core_table;
719 	if (!net_eq(net, &init_net)) {
720 		tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
721 		if (tbl == NULL)
722 			goto err_dup;
723 
724 		for (tmp = tbl; tmp->procname; tmp++)
725 			tmp->data += (char *)net - (char *)&init_net;
726 	}
727 
728 	net->core.sysctl_hdr = register_net_sysctl_sz(net, "net/core", tbl,
729 						      ARRAY_SIZE(netns_core_table));
730 	if (net->core.sysctl_hdr == NULL)
731 		goto err_reg;
732 
733 	return 0;
734 
735 err_reg:
736 	if (tbl != netns_core_table)
737 		kfree(tbl);
738 err_dup:
739 	return -ENOMEM;
740 }
741 
sysctl_core_net_exit(struct net * net)742 static __net_exit void sysctl_core_net_exit(struct net *net)
743 {
744 	struct ctl_table *tbl;
745 
746 	tbl = net->core.sysctl_hdr->ctl_table_arg;
747 	unregister_net_sysctl_table(net->core.sysctl_hdr);
748 	BUG_ON(tbl == netns_core_table);
749 #if IS_ENABLED(CONFIG_RPS)
750 	kfree(net->core.rps_default_mask);
751 #endif
752 	kfree(tbl);
753 }
754 
755 static __net_initdata struct pernet_operations sysctl_core_ops = {
756 	.init = sysctl_core_net_init,
757 	.exit = sysctl_core_net_exit,
758 };
759 
sysctl_core_init(void)760 static __init int sysctl_core_init(void)
761 {
762 	register_net_sysctl(&init_net, "net/core", net_core_table);
763 	return register_pernet_subsys(&sysctl_core_ops);
764 }
765 
766 fs_initcall(sysctl_core_init);
767