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
 32static int int_3600 = 3600;
 33static int min_sndbuf = SOCK_MIN_SNDBUF;
 34static int min_rcvbuf = SOCK_MIN_RCVBUF;
 35static int max_skb_frags = MAX_SKB_FRAGS;
 36static int min_mem_pcpu_rsv = SK_MEMORY_PCPU_RESERVE;
 37static int netdev_budget_usecs_min = 2 * USEC_PER_SEC / HZ;
 38
 39static int net_msg_warn;	/* Unused, but still a sysctl */
 40
 41int sysctl_fb_tunnels_only_for_init_net __read_mostly = 0;
 42EXPORT_SYMBOL(sysctl_fb_tunnels_only_for_init_net);
 43
 44/* 0 - Keep current behavior:
 45 *     IPv4: inherit all current settings from init_net
 46 *     IPv6: reset all settings to default
 47 * 1 - Both inherit all current settings from init_net
 48 * 2 - Both reset all settings to default
 49 * 3 - Both inherit all settings from current netns
 50 */
 51int sysctl_devconf_inherit_init_net __read_mostly;
 52EXPORT_SYMBOL(sysctl_devconf_inherit_init_net);
 53
 54#if IS_ENABLED(CONFIG_NET_FLOW_LIMIT) || IS_ENABLED(CONFIG_RPS)
 55static int dump_cpumask(void *buffer, size_t *lenp, loff_t *ppos,
 56			struct cpumask *mask)
 57{
 58	char *kbuf;
 59	int len;
 60
 61	if (*ppos || !*lenp) {
 62		*lenp = 0;
 63		return 0;
 64	}
 65
 66	/* CPUs are displayed as a hex bitmap + a comma between each groups of 8
 67	 * nibbles (except the last one which has a newline instead).
 68	 * Guesstimate the buffer size at the group granularity level.
 69	 */
 70	len = min(DIV_ROUND_UP(nr_cpumask_bits, 32) * (8 + 1), *lenp);
 71	kbuf = kmalloc(len, GFP_KERNEL);
 72	if (!kbuf) {
 73		*lenp = 0;
 74		return -ENOMEM;
 75	}
 76
 77	len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
 78	if (!len) {
 79		*lenp = 0;
 80		goto free_buf;
 81	}
 82
 83	/* scnprintf writes a trailing null char not counted in the returned
 84	 * length, override it with a newline.
 85	 */
 86	kbuf[len++] = '\n';
 87	memcpy(buffer, kbuf, len);
 88	*lenp = len;
 89	*ppos += len;
 90
 91free_buf:
 92	kfree(kbuf);
 93	return 0;
 94}
 95#endif
 96
 97#ifdef CONFIG_RPS
 98
 99static struct cpumask *rps_default_mask_cow_alloc(struct net *net)
100{
101	struct cpumask *rps_default_mask;
102
103	if (net->core.rps_default_mask)
104		return net->core.rps_default_mask;
105
106	rps_default_mask = kzalloc(cpumask_size(), GFP_KERNEL);
107	if (!rps_default_mask)
108		return NULL;
109
110	/* pairs with READ_ONCE in rx_queue_default_mask() */
111	WRITE_ONCE(net->core.rps_default_mask, rps_default_mask);
112	return rps_default_mask;
113}
114
115static int rps_default_mask_sysctl(const struct ctl_table *table, int write,
116				   void *buffer, size_t *lenp, loff_t *ppos)
117{
118	struct net *net = (struct net *)table->data;
119	int err = 0;
120
121	rtnl_lock();
122	if (write) {
123		struct cpumask *rps_default_mask = rps_default_mask_cow_alloc(net);
124
125		err = -ENOMEM;
126		if (!rps_default_mask)
127			goto done;
128
129		err = cpumask_parse(buffer, rps_default_mask);
130		if (err)
131			goto done;
132
133		err = rps_cpumask_housekeeping(rps_default_mask);
134		if (err)
135			goto done;
136	} else {
137		err = dump_cpumask(buffer, lenp, ppos,
138				   net->core.rps_default_mask ? : cpu_none_mask);
139	}
140
141done:
142	rtnl_unlock();
143	return err;
144}
145
146static int rps_sock_flow_sysctl(const struct ctl_table *table, int write,
147				void *buffer, size_t *lenp, loff_t *ppos)
148{
149	unsigned int orig_size, size;
150	int ret, i;
151	struct ctl_table tmp = {
152		.data = &size,
153		.maxlen = sizeof(size),
154		.mode = table->mode
155	};
156	struct rps_sock_flow_table *orig_sock_table, *sock_table;
157	static DEFINE_MUTEX(sock_flow_mutex);
158
159	mutex_lock(&sock_flow_mutex);
160
161	orig_sock_table = rcu_dereference_protected(
162					net_hotdata.rps_sock_flow_table,
163					lockdep_is_held(&sock_flow_mutex));
164	size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
165
166	ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
167
168	if (write) {
169		if (size) {
170			if (size > 1<<29) {
171				/* Enforce limit to prevent overflow */
172				mutex_unlock(&sock_flow_mutex);
173				return -EINVAL;
174			}
175			size = roundup_pow_of_two(size);
176			if (size != orig_size) {
177				sock_table =
178				    vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
179				if (!sock_table) {
180					mutex_unlock(&sock_flow_mutex);
181					return -ENOMEM;
182				}
183				net_hotdata.rps_cpu_mask =
184					roundup_pow_of_two(nr_cpu_ids) - 1;
185				sock_table->mask = size - 1;
186			} else
187				sock_table = orig_sock_table;
188
189			for (i = 0; i < size; i++)
190				sock_table->ents[i] = RPS_NO_CPU;
191		} else
192			sock_table = NULL;
193
194		if (sock_table != orig_sock_table) {
195			rcu_assign_pointer(net_hotdata.rps_sock_flow_table,
196					   sock_table);
197			if (sock_table) {
198				static_branch_inc(&rps_needed);
199				static_branch_inc(&rfs_needed);
200			}
201			if (orig_sock_table) {
202				static_branch_dec(&rps_needed);
203				static_branch_dec(&rfs_needed);
204				kvfree_rcu_mightsleep(orig_sock_table);
 
205			}
206		}
207	}
208
209	mutex_unlock(&sock_flow_mutex);
210
211	return ret;
212}
213#endif /* CONFIG_RPS */
214
215#ifdef CONFIG_NET_FLOW_LIMIT
216static DEFINE_MUTEX(flow_limit_update_mutex);
217
218static 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_mightsleep(cur);
 
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->num_buckets = netdev_flow_limit_table_len;
252				rcu_assign_pointer(sd->flow_limit, cur);
253			}
254		}
255write_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
270done:
271	free_cpumask_var(mask);
272	return ret;
273}
274
275static 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
297static 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
316static 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
334static int proc_do_rss_key(const struct ctl_table *table, int write,
335			   void *buffer, size_t *lenp, loff_t *ppos)
336{
337	struct ctl_table fake_table;
338	char buf[NETDEV_RSS_KEY_LEN * 3];
339
340	snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
341	fake_table.data = buf;
342	fake_table.maxlen = sizeof(buf);
343	return proc_dostring(&fake_table, write, buffer, lenp, ppos);
344}
345
346#ifdef CONFIG_BPF_JIT
347static int proc_dointvec_minmax_bpf_enable(const struct ctl_table *table, int write,
348					   void *buffer, size_t *lenp,
349					   loff_t *ppos)
350{
351	int ret, jit_enable = *(int *)table->data;
352	int min = *(int *)table->extra1;
353	int max = *(int *)table->extra2;
354	struct ctl_table tmp = *table;
355
356	if (write && !capable(CAP_SYS_ADMIN))
357		return -EPERM;
358
359	tmp.data = &jit_enable;
360	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
361	if (write && !ret) {
362		if (jit_enable < 2 ||
363		    (jit_enable == 2 && bpf_dump_raw_ok(current_cred()))) {
364			*(int *)table->data = jit_enable;
365			if (jit_enable == 2)
366				pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
367		} else {
368			ret = -EPERM;
369		}
370	}
371
372	if (write && ret && min == max)
373		pr_info_once("CONFIG_BPF_JIT_ALWAYS_ON is enabled, bpf_jit_enable is permanently set to 1.\n");
374
375	return ret;
376}
377
378# ifdef CONFIG_HAVE_EBPF_JIT
379static int
380proc_dointvec_minmax_bpf_restricted(const struct ctl_table *table, int write,
381				    void *buffer, size_t *lenp, loff_t *ppos)
382{
383	if (!capable(CAP_SYS_ADMIN))
384		return -EPERM;
385
386	return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
387}
388# endif /* CONFIG_HAVE_EBPF_JIT */
389
390static int
391proc_dolongvec_minmax_bpf_restricted(const struct ctl_table *table, int write,
392				     void *buffer, size_t *lenp, loff_t *ppos)
393{
394	if (!capable(CAP_SYS_ADMIN))
395		return -EPERM;
396
397	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
398}
399#endif
400
401static struct ctl_table net_core_table[] = {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
402	{
403		.procname	= "mem_pcpu_rsv",
404		.data		= &net_hotdata.sysctl_mem_pcpu_rsv,
405		.maxlen		= sizeof(int),
406		.mode		= 0644,
407		.proc_handler	= proc_dointvec_minmax,
408		.extra1		= &min_mem_pcpu_rsv,
409	},
410	{
411		.procname	= "dev_weight",
412		.data		= &weight_p,
413		.maxlen		= sizeof(int),
414		.mode		= 0644,
415		.proc_handler	= proc_do_dev_weight,
416		.extra1         = SYSCTL_ONE,
417	},
418	{
419		.procname	= "dev_weight_rx_bias",
420		.data		= &dev_weight_rx_bias,
421		.maxlen		= sizeof(int),
422		.mode		= 0644,
423		.proc_handler	= proc_do_dev_weight,
424		.extra1         = SYSCTL_ONE,
425	},
426	{
427		.procname	= "dev_weight_tx_bias",
428		.data		= &dev_weight_tx_bias,
429		.maxlen		= sizeof(int),
430		.mode		= 0644,
431		.proc_handler	= proc_do_dev_weight,
432		.extra1         = SYSCTL_ONE,
433	},
434	{
435		.procname	= "netdev_max_backlog",
436		.data		= &net_hotdata.max_backlog,
437		.maxlen		= sizeof(int),
438		.mode		= 0644,
439		.proc_handler	= proc_dointvec
440	},
441	{
442		.procname	= "netdev_rss_key",
443		.data		= &netdev_rss_key,
444		.maxlen		= sizeof(int),
445		.mode		= 0444,
446		.proc_handler	= proc_do_rss_key,
447	},
448#ifdef CONFIG_BPF_JIT
449	{
450		.procname	= "bpf_jit_enable",
451		.data		= &bpf_jit_enable,
452		.maxlen		= sizeof(int),
453		.mode		= 0644,
454		.proc_handler	= proc_dointvec_minmax_bpf_enable,
455# ifdef CONFIG_BPF_JIT_ALWAYS_ON
456		.extra1		= SYSCTL_ONE,
457		.extra2		= SYSCTL_ONE,
458# else
459		.extra1		= SYSCTL_ZERO,
460		.extra2		= SYSCTL_TWO,
461# endif
462	},
463# ifdef CONFIG_HAVE_EBPF_JIT
464	{
465		.procname	= "bpf_jit_harden",
466		.data		= &bpf_jit_harden,
467		.maxlen		= sizeof(int),
468		.mode		= 0600,
469		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
470		.extra1		= SYSCTL_ZERO,
471		.extra2		= SYSCTL_TWO,
472	},
473	{
474		.procname	= "bpf_jit_kallsyms",
475		.data		= &bpf_jit_kallsyms,
476		.maxlen		= sizeof(int),
477		.mode		= 0600,
478		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
479		.extra1		= SYSCTL_ZERO,
480		.extra2		= SYSCTL_ONE,
481	},
482# endif
483	{
484		.procname	= "bpf_jit_limit",
485		.data		= &bpf_jit_limit,
486		.maxlen		= sizeof(long),
487		.mode		= 0600,
488		.proc_handler	= proc_dolongvec_minmax_bpf_restricted,
489		.extra1		= SYSCTL_LONG_ONE,
490		.extra2		= &bpf_jit_limit_max,
491	},
492#endif
493	{
494		.procname	= "netdev_tstamp_prequeue",
495		.data		= &net_hotdata.tstamp_prequeue,
496		.maxlen		= sizeof(int),
497		.mode		= 0644,
498		.proc_handler	= proc_dointvec
499	},
500	{
501		.procname	= "message_cost",
502		.data		= &net_ratelimit_state.interval,
503		.maxlen		= sizeof(int),
504		.mode		= 0644,
505		.proc_handler	= proc_dointvec_jiffies,
506	},
507	{
508		.procname	= "message_burst",
509		.data		= &net_ratelimit_state.burst,
510		.maxlen		= sizeof(int),
511		.mode		= 0644,
512		.proc_handler	= proc_dointvec,
513	},
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
514#ifdef CONFIG_RPS
515	{
516		.procname	= "rps_sock_flow_entries",
517		.maxlen		= sizeof(int),
518		.mode		= 0644,
519		.proc_handler	= rps_sock_flow_sysctl
520	},
521#endif
522#ifdef CONFIG_NET_FLOW_LIMIT
523	{
524		.procname	= "flow_limit_cpu_bitmap",
525		.mode		= 0644,
526		.proc_handler	= flow_limit_cpu_sysctl
527	},
528	{
529		.procname	= "flow_limit_table_len",
530		.data		= &netdev_flow_limit_table_len,
531		.maxlen		= sizeof(int),
532		.mode		= 0644,
533		.proc_handler	= flow_limit_table_len_sysctl
534	},
535#endif /* CONFIG_NET_FLOW_LIMIT */
536#ifdef CONFIG_NET_RX_BUSY_POLL
537	{
538		.procname	= "busy_poll",
539		.data		= &sysctl_net_busy_poll,
540		.maxlen		= sizeof(unsigned int),
541		.mode		= 0644,
542		.proc_handler	= proc_dointvec_minmax,
543		.extra1		= SYSCTL_ZERO,
544	},
545	{
546		.procname	= "busy_read",
547		.data		= &sysctl_net_busy_read,
548		.maxlen		= sizeof(unsigned int),
549		.mode		= 0644,
550		.proc_handler	= proc_dointvec_minmax,
551		.extra1		= SYSCTL_ZERO,
552	},
553#endif
554#ifdef CONFIG_NET_SCHED
555	{
556		.procname	= "default_qdisc",
557		.mode		= 0644,
558		.maxlen		= IFNAMSIZ,
559		.proc_handler	= set_default_qdisc
560	},
561#endif
 
562	{
563		.procname	= "netdev_budget",
564		.data		= &net_hotdata.netdev_budget,
565		.maxlen		= sizeof(int),
566		.mode		= 0644,
567		.proc_handler	= proc_dointvec
568	},
569	{
570		.procname	= "warnings",
571		.data		= &net_msg_warn,
572		.maxlen		= sizeof(int),
573		.mode		= 0644,
574		.proc_handler	= proc_dointvec
575	},
576	{
577		.procname	= "max_skb_frags",
578		.data		= &net_hotdata.sysctl_max_skb_frags,
579		.maxlen		= sizeof(int),
580		.mode		= 0644,
581		.proc_handler	= proc_dointvec_minmax,
582		.extra1		= SYSCTL_ONE,
583		.extra2		= &max_skb_frags,
584	},
585	{
586		.procname	= "netdev_budget_usecs",
587		.data		= &net_hotdata.netdev_budget_usecs,
588		.maxlen		= sizeof(unsigned int),
589		.mode		= 0644,
590		.proc_handler	= proc_dointvec_minmax,
591		.extra1		= &netdev_budget_usecs_min,
592	},
593	{
594		.procname	= "fb_tunnels_only_for_init_net",
595		.data		= &sysctl_fb_tunnels_only_for_init_net,
596		.maxlen		= sizeof(int),
597		.mode		= 0644,
598		.proc_handler	= proc_dointvec_minmax,
599		.extra1		= SYSCTL_ZERO,
600		.extra2		= SYSCTL_TWO,
601	},
602	{
603		.procname	= "devconf_inherit_init_net",
604		.data		= &sysctl_devconf_inherit_init_net,
605		.maxlen		= sizeof(int),
606		.mode		= 0644,
607		.proc_handler	= proc_dointvec_minmax,
608		.extra1		= SYSCTL_ZERO,
609		.extra2		= SYSCTL_THREE,
610	},
611	{
612		.procname	= "high_order_alloc_disable",
613		.data		= &net_high_order_alloc_disable_key.key,
614		.maxlen         = sizeof(net_high_order_alloc_disable_key),
615		.mode		= 0644,
616		.proc_handler	= proc_do_static_key,
617	},
618	{
619		.procname	= "gro_normal_batch",
620		.data		= &net_hotdata.gro_normal_batch,
621		.maxlen		= sizeof(unsigned int),
622		.mode		= 0644,
623		.proc_handler	= proc_dointvec_minmax,
624		.extra1		= SYSCTL_ONE,
625	},
626	{
627		.procname	= "netdev_unregister_timeout_secs",
628		.data		= &netdev_unregister_timeout_secs,
629		.maxlen		= sizeof(unsigned int),
630		.mode		= 0644,
631		.proc_handler	= proc_dointvec_minmax,
632		.extra1		= SYSCTL_ONE,
633		.extra2		= &int_3600,
634	},
635	{
636		.procname	= "skb_defer_max",
637		.data		= &net_hotdata.sysctl_skb_defer_max,
638		.maxlen		= sizeof(unsigned int),
639		.mode		= 0644,
640		.proc_handler	= proc_dointvec_minmax,
641		.extra1		= SYSCTL_ZERO,
642	},
643};
644
645static struct ctl_table netns_core_table[] = {
646#if IS_ENABLED(CONFIG_RPS)
647	{
648		.procname	= "rps_default_mask",
649		.data		= &init_net,
650		.mode		= 0644,
651		.proc_handler	= rps_default_mask_sysctl
652	},
653#endif
654	{
655		.procname	= "somaxconn",
656		.data		= &init_net.core.sysctl_somaxconn,
657		.maxlen		= sizeof(int),
658		.mode		= 0644,
659		.extra1		= SYSCTL_ZERO,
660		.proc_handler	= proc_dointvec_minmax
661	},
662	{
663		.procname	= "optmem_max",
664		.data		= &init_net.core.sysctl_optmem_max,
665		.maxlen		= sizeof(int),
666		.mode		= 0644,
667		.extra1		= SYSCTL_ZERO,
668		.proc_handler	= proc_dointvec_minmax
669	},
670	{
671		.procname	= "txrehash",
672		.data		= &init_net.core.sysctl_txrehash,
673		.maxlen		= sizeof(u8),
674		.mode		= 0644,
675		.extra1		= SYSCTL_ZERO,
676		.extra2		= SYSCTL_ONE,
677		.proc_handler	= proc_dou8vec_minmax,
678	},
679	{
680		.procname	= "tstamp_allow_data",
681		.data		= &init_net.core.sysctl_tstamp_allow_data,
682		.maxlen		= sizeof(u8),
683		.mode		= 0644,
684		.proc_handler	= proc_dou8vec_minmax,
685		.extra1		= SYSCTL_ZERO,
686		.extra2		= SYSCTL_ONE
687	},
688	/* sysctl_core_net_init() will set the values after this
689	 * to readonly in network namespaces
690	 */
691	{
692		.procname	= "wmem_max",
693		.data		= &sysctl_wmem_max,
694		.maxlen		= sizeof(int),
695		.mode		= 0644,
696		.proc_handler	= proc_dointvec_minmax,
697		.extra1		= &min_sndbuf,
698	},
699	{
700		.procname	= "rmem_max",
701		.data		= &sysctl_rmem_max,
702		.maxlen		= sizeof(int),
703		.mode		= 0644,
704		.proc_handler	= proc_dointvec_minmax,
705		.extra1		= &min_rcvbuf,
706	},
707	{
708		.procname	= "wmem_default",
709		.data		= &sysctl_wmem_default,
710		.maxlen		= sizeof(int),
711		.mode		= 0644,
712		.proc_handler	= proc_dointvec_minmax,
713		.extra1		= &min_sndbuf,
714	},
715	{
716		.procname	= "rmem_default",
717		.data		= &sysctl_rmem_default,
718		.maxlen		= sizeof(int),
719		.mode		= 0644,
720		.proc_handler	= proc_dointvec_minmax,
721		.extra1		= &min_rcvbuf,
722	},
723};
724
725static int __init fb_tunnels_only_for_init_net_sysctl_setup(char *str)
726{
727	/* fallback tunnels for initns only */
728	if (!strncmp(str, "initns", 6))
729		sysctl_fb_tunnels_only_for_init_net = 1;
730	/* no fallback tunnels anywhere */
731	else if (!strncmp(str, "none", 4))
732		sysctl_fb_tunnels_only_for_init_net = 2;
733
734	return 1;
735}
736__setup("fb_tunnels=", fb_tunnels_only_for_init_net_sysctl_setup);
737
738static __net_init int sysctl_core_net_init(struct net *net)
739{
740	size_t table_size = ARRAY_SIZE(netns_core_table);
741	struct ctl_table *tbl;
742
743	tbl = netns_core_table;
744	if (!net_eq(net, &init_net)) {
745		int i;
746		tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
747		if (tbl == NULL)
748			goto err_dup;
749
750		for (i = 0; i < table_size; ++i) {
751			if (tbl[i].data == &sysctl_wmem_max)
752				break;
753
754			tbl[i].data += (char *)net - (char *)&init_net;
 
 
755		}
756		for (; i < table_size; ++i)
757			tbl[i].mode &= ~0222;
758	}
759
760	net->core.sysctl_hdr = register_net_sysctl_sz(net, "net/core", tbl, table_size);
761	if (net->core.sysctl_hdr == NULL)
762		goto err_reg;
763
764	return 0;
765
766err_reg:
767	if (tbl != netns_core_table)
768		kfree(tbl);
769err_dup:
770	return -ENOMEM;
771}
772
773static __net_exit void sysctl_core_net_exit(struct net *net)
774{
775	const struct ctl_table *tbl;
776
777	tbl = net->core.sysctl_hdr->ctl_table_arg;
778	unregister_net_sysctl_table(net->core.sysctl_hdr);
779	BUG_ON(tbl == netns_core_table);
780#if IS_ENABLED(CONFIG_RPS)
781	kfree(net->core.rps_default_mask);
782#endif
783	kfree(tbl);
784}
785
786static __net_initdata struct pernet_operations sysctl_core_ops = {
787	.init = sysctl_core_net_init,
788	.exit = sysctl_core_net_exit,
789};
790
791static __init int sysctl_core_init(void)
792{
793	register_net_sysctl(&init_net, "net/core", net_core_table);
794	return register_pernet_subsys(&sysctl_core_ops);
795}
796
797fs_initcall(sysctl_core_init);