Linux Audio

Check our new training course

Loading...
v6.8
  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
 29static int int_3600 = 3600;
 30static int min_sndbuf = SOCK_MIN_SNDBUF;
 31static int min_rcvbuf = SOCK_MIN_RCVBUF;
 32static int max_skb_frags = MAX_SKB_FRAGS;
 33
 34static int net_msg_warn;	/* Unused, but still a sysctl */
 35
 36int sysctl_fb_tunnels_only_for_init_net __read_mostly = 0;
 37EXPORT_SYMBOL(sysctl_fb_tunnels_only_for_init_net);
 38
 39/* 0 - Keep current behavior:
 40 *     IPv4: inherit all current settings from init_net
 41 *     IPv6: reset all settings to default
 42 * 1 - Both inherit all current settings from init_net
 43 * 2 - Both reset all settings to default
 44 * 3 - Both inherit all settings from current netns
 45 */
 46int sysctl_devconf_inherit_init_net __read_mostly;
 47EXPORT_SYMBOL(sysctl_devconf_inherit_init_net);
 48
 49#if IS_ENABLED(CONFIG_NET_FLOW_LIMIT) || IS_ENABLED(CONFIG_RPS)
 50static void dump_cpumask(void *buffer, size_t *lenp, loff_t *ppos,
 51			 struct cpumask *mask)
 52{
 53	char kbuf[128];
 54	int len;
 55
 56	if (*ppos || !*lenp) {
 57		*lenp = 0;
 58		return;
 59	}
 60
 61	len = min(sizeof(kbuf) - 1, *lenp);
 62	len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
 63	if (!len) {
 64		*lenp = 0;
 65		return;
 66	}
 67
 68	if (len < *lenp)
 69		kbuf[len++] = '\n';
 70	memcpy(buffer, kbuf, len);
 71	*lenp = len;
 72	*ppos += len;
 73}
 74#endif
 75
 76#ifdef CONFIG_RPS
 77
 78static struct cpumask *rps_default_mask_cow_alloc(struct net *net)
 79{
 80	struct cpumask *rps_default_mask;
 81
 82	if (net->core.rps_default_mask)
 83		return net->core.rps_default_mask;
 84
 85	rps_default_mask = kzalloc(cpumask_size(), GFP_KERNEL);
 86	if (!rps_default_mask)
 87		return NULL;
 88
 89	/* pairs with READ_ONCE in rx_queue_default_mask() */
 90	WRITE_ONCE(net->core.rps_default_mask, rps_default_mask);
 91	return rps_default_mask;
 92}
 93
 94static int rps_default_mask_sysctl(struct ctl_table *table, int write,
 95				   void *buffer, size_t *lenp, loff_t *ppos)
 96{
 97	struct net *net = (struct net *)table->data;
 98	int err = 0;
 99
100	rtnl_lock();
101	if (write) {
102		struct cpumask *rps_default_mask = rps_default_mask_cow_alloc(net);
103
104		err = -ENOMEM;
105		if (!rps_default_mask)
106			goto done;
107
108		err = cpumask_parse(buffer, rps_default_mask);
109		if (err)
110			goto done;
111
112		err = rps_cpumask_housekeeping(rps_default_mask);
113		if (err)
114			goto done;
115	} else {
116		dump_cpumask(buffer, lenp, ppos,
117			     net->core.rps_default_mask ? : cpu_none_mask);
118	}
119
120done:
121	rtnl_unlock();
122	return err;
123}
124
125static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
126				void *buffer, size_t *lenp, loff_t *ppos)
127{
128	unsigned int orig_size, size;
129	int ret, i;
130	struct ctl_table tmp = {
131		.data = &size,
132		.maxlen = sizeof(size),
133		.mode = table->mode
134	};
135	struct rps_sock_flow_table *orig_sock_table, *sock_table;
136	static DEFINE_MUTEX(sock_flow_mutex);
137
138	mutex_lock(&sock_flow_mutex);
139
140	orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
141					lockdep_is_held(&sock_flow_mutex));
142	size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
143
144	ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
145
146	if (write) {
147		if (size) {
148			if (size > 1<<29) {
149				/* Enforce limit to prevent overflow */
150				mutex_unlock(&sock_flow_mutex);
151				return -EINVAL;
152			}
153			size = roundup_pow_of_two(size);
154			if (size != orig_size) {
155				sock_table =
156				    vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
157				if (!sock_table) {
158					mutex_unlock(&sock_flow_mutex);
159					return -ENOMEM;
160				}
161				rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
162				sock_table->mask = size - 1;
163			} else
164				sock_table = orig_sock_table;
165
166			for (i = 0; i < size; i++)
167				sock_table->ents[i] = RPS_NO_CPU;
168		} else
169			sock_table = NULL;
170
171		if (sock_table != orig_sock_table) {
172			rcu_assign_pointer(rps_sock_flow_table, sock_table);
173			if (sock_table) {
174				static_branch_inc(&rps_needed);
175				static_branch_inc(&rfs_needed);
176			}
177			if (orig_sock_table) {
178				static_branch_dec(&rps_needed);
179				static_branch_dec(&rfs_needed);
180				kvfree_rcu_mightsleep(orig_sock_table);
181			}
182		}
183	}
184
185	mutex_unlock(&sock_flow_mutex);
186
187	return ret;
188}
189#endif /* CONFIG_RPS */
190
191#ifdef CONFIG_NET_FLOW_LIMIT
192static DEFINE_MUTEX(flow_limit_update_mutex);
193
194static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
195				 void *buffer, size_t *lenp, loff_t *ppos)
 
196{
197	struct sd_flow_limit *cur;
198	struct softnet_data *sd;
199	cpumask_var_t mask;
200	int i, len, ret = 0;
201
202	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
203		return -ENOMEM;
204
205	if (write) {
206		ret = cpumask_parse(buffer, mask);
207		if (ret)
208			goto done;
209
210		mutex_lock(&flow_limit_update_mutex);
211		len = sizeof(*cur) + netdev_flow_limit_table_len;
212		for_each_possible_cpu(i) {
213			sd = &per_cpu(softnet_data, i);
214			cur = rcu_dereference_protected(sd->flow_limit,
215				     lockdep_is_held(&flow_limit_update_mutex));
216			if (cur && !cpumask_test_cpu(i, mask)) {
217				RCU_INIT_POINTER(sd->flow_limit, NULL);
218				kfree_rcu_mightsleep(cur);
 
219			} else if (!cur && cpumask_test_cpu(i, mask)) {
220				cur = kzalloc_node(len, GFP_KERNEL,
221						   cpu_to_node(i));
222				if (!cur) {
223					/* not unwinding previous changes */
224					ret = -ENOMEM;
225					goto write_unlock;
226				}
227				cur->num_buckets = netdev_flow_limit_table_len;
228				rcu_assign_pointer(sd->flow_limit, cur);
229			}
230		}
231write_unlock:
232		mutex_unlock(&flow_limit_update_mutex);
233	} else {
 
 
 
 
 
 
 
234		cpumask_clear(mask);
235		rcu_read_lock();
236		for_each_possible_cpu(i) {
237			sd = &per_cpu(softnet_data, i);
238			if (rcu_dereference(sd->flow_limit))
239				cpumask_set_cpu(i, mask);
240		}
241		rcu_read_unlock();
242
243		dump_cpumask(buffer, lenp, ppos, mask);
 
 
 
 
 
 
 
 
 
 
 
 
 
244	}
245
246done:
247	free_cpumask_var(mask);
248	return ret;
249}
250
251static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
252				       void *buffer, size_t *lenp, loff_t *ppos)
 
253{
254	unsigned int old, *ptr;
255	int ret;
256
257	mutex_lock(&flow_limit_update_mutex);
258
259	ptr = table->data;
260	old = *ptr;
261	ret = proc_dointvec(table, write, buffer, lenp, ppos);
262	if (!ret && write && !is_power_of_2(*ptr)) {
263		*ptr = old;
264		ret = -EINVAL;
265	}
266
267	mutex_unlock(&flow_limit_update_mutex);
268	return ret;
269}
270#endif /* CONFIG_NET_FLOW_LIMIT */
271
272#ifdef CONFIG_NET_SCHED
273static int set_default_qdisc(struct ctl_table *table, int write,
274			     void *buffer, size_t *lenp, loff_t *ppos)
275{
276	char id[IFNAMSIZ];
277	struct ctl_table tbl = {
278		.data = id,
279		.maxlen = IFNAMSIZ,
280	};
281	int ret;
282
283	qdisc_get_default(id, IFNAMSIZ);
284
285	ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
286	if (write && ret == 0)
287		ret = qdisc_set_default(id);
288	return ret;
289}
290#endif
291
292static int proc_do_dev_weight(struct ctl_table *table, int write,
293			   void *buffer, size_t *lenp, loff_t *ppos)
294{
295	static DEFINE_MUTEX(dev_weight_mutex);
296	int ret, weight;
297
298	mutex_lock(&dev_weight_mutex);
299	ret = proc_dointvec(table, write, buffer, lenp, ppos);
300	if (!ret && write) {
301		weight = READ_ONCE(weight_p);
302		WRITE_ONCE(dev_rx_weight, weight * dev_weight_rx_bias);
303		WRITE_ONCE(dev_tx_weight, weight * dev_weight_tx_bias);
304	}
305	mutex_unlock(&dev_weight_mutex);
306
307	return ret;
308}
309
310static int proc_do_rss_key(struct ctl_table *table, int write,
311			   void *buffer, size_t *lenp, loff_t *ppos)
312{
313	struct ctl_table fake_table;
314	char buf[NETDEV_RSS_KEY_LEN * 3];
315
316	snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
317	fake_table.data = buf;
318	fake_table.maxlen = sizeof(buf);
319	return proc_dostring(&fake_table, write, buffer, lenp, ppos);
320}
321
322#ifdef CONFIG_BPF_JIT
323static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write,
324					   void *buffer, size_t *lenp,
325					   loff_t *ppos)
326{
327	int ret, jit_enable = *(int *)table->data;
328	int min = *(int *)table->extra1;
329	int max = *(int *)table->extra2;
330	struct ctl_table tmp = *table;
331
332	if (write && !capable(CAP_SYS_ADMIN))
333		return -EPERM;
334
335	tmp.data = &jit_enable;
336	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
337	if (write && !ret) {
338		if (jit_enable < 2 ||
339		    (jit_enable == 2 && bpf_dump_raw_ok(current_cred()))) {
340			*(int *)table->data = jit_enable;
341			if (jit_enable == 2)
342				pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
343		} else {
344			ret = -EPERM;
345		}
346	}
347
348	if (write && ret && min == max)
349		pr_info_once("CONFIG_BPF_JIT_ALWAYS_ON is enabled, bpf_jit_enable is permanently set to 1.\n");
350
351	return ret;
352}
353
354# ifdef CONFIG_HAVE_EBPF_JIT
355static int
356proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write,
357				    void *buffer, size_t *lenp, loff_t *ppos)
358{
359	if (!capable(CAP_SYS_ADMIN))
360		return -EPERM;
361
362	return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
363}
364# endif /* CONFIG_HAVE_EBPF_JIT */
365
366static int
367proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
368				     void *buffer, size_t *lenp, loff_t *ppos)
369{
370	if (!capable(CAP_SYS_ADMIN))
371		return -EPERM;
372
373	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
374}
375#endif
376
377static struct ctl_table net_core_table[] = {
 
378	{
379		.procname	= "wmem_max",
380		.data		= &sysctl_wmem_max,
381		.maxlen		= sizeof(int),
382		.mode		= 0644,
383		.proc_handler	= proc_dointvec_minmax,
384		.extra1		= &min_sndbuf,
385	},
386	{
387		.procname	= "rmem_max",
388		.data		= &sysctl_rmem_max,
389		.maxlen		= sizeof(int),
390		.mode		= 0644,
391		.proc_handler	= proc_dointvec_minmax,
392		.extra1		= &min_rcvbuf,
393	},
394	{
395		.procname	= "wmem_default",
396		.data		= &sysctl_wmem_default,
397		.maxlen		= sizeof(int),
398		.mode		= 0644,
399		.proc_handler	= proc_dointvec_minmax,
400		.extra1		= &min_sndbuf,
401	},
402	{
403		.procname	= "rmem_default",
404		.data		= &sysctl_rmem_default,
405		.maxlen		= sizeof(int),
406		.mode		= 0644,
407		.proc_handler	= proc_dointvec_minmax,
408		.extra1		= &min_rcvbuf,
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	},
417	{
418		.procname	= "dev_weight_rx_bias",
419		.data		= &dev_weight_rx_bias,
420		.maxlen		= sizeof(int),
421		.mode		= 0644,
422		.proc_handler	= proc_do_dev_weight,
423	},
424	{
425		.procname	= "dev_weight_tx_bias",
426		.data		= &dev_weight_tx_bias,
427		.maxlen		= sizeof(int),
428		.mode		= 0644,
429		.proc_handler	= proc_do_dev_weight,
430	},
431	{
432		.procname	= "netdev_max_backlog",
433		.data		= &netdev_max_backlog,
434		.maxlen		= sizeof(int),
435		.mode		= 0644,
436		.proc_handler	= proc_dointvec
437	},
438	{
439		.procname	= "netdev_rss_key",
440		.data		= &netdev_rss_key,
441		.maxlen		= sizeof(int),
442		.mode		= 0444,
443		.proc_handler	= proc_do_rss_key,
444	},
445#ifdef CONFIG_BPF_JIT
446	{
447		.procname	= "bpf_jit_enable",
448		.data		= &bpf_jit_enable,
449		.maxlen		= sizeof(int),
450		.mode		= 0644,
451		.proc_handler	= proc_dointvec_minmax_bpf_enable,
452# ifdef CONFIG_BPF_JIT_ALWAYS_ON
453		.extra1		= SYSCTL_ONE,
454		.extra2		= SYSCTL_ONE,
455# else
456		.extra1		= SYSCTL_ZERO,
457		.extra2		= SYSCTL_TWO,
458# endif
459	},
460# ifdef CONFIG_HAVE_EBPF_JIT
461	{
462		.procname	= "bpf_jit_harden",
463		.data		= &bpf_jit_harden,
464		.maxlen		= sizeof(int),
465		.mode		= 0600,
466		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
467		.extra1		= SYSCTL_ZERO,
468		.extra2		= SYSCTL_TWO,
469	},
470	{
471		.procname	= "bpf_jit_kallsyms",
472		.data		= &bpf_jit_kallsyms,
473		.maxlen		= sizeof(int),
474		.mode		= 0600,
475		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
476		.extra1		= SYSCTL_ZERO,
477		.extra2		= SYSCTL_ONE,
478	},
479# endif
480	{
481		.procname	= "bpf_jit_limit",
482		.data		= &bpf_jit_limit,
483		.maxlen		= sizeof(long),
484		.mode		= 0600,
485		.proc_handler	= proc_dolongvec_minmax_bpf_restricted,
486		.extra1		= SYSCTL_LONG_ONE,
487		.extra2		= &bpf_jit_limit_max,
488	},
489#endif
490	{
491		.procname	= "netdev_tstamp_prequeue",
492		.data		= &netdev_tstamp_prequeue,
493		.maxlen		= sizeof(int),
494		.mode		= 0644,
495		.proc_handler	= proc_dointvec
496	},
497	{
498		.procname	= "message_cost",
499		.data		= &net_ratelimit_state.interval,
500		.maxlen		= sizeof(int),
501		.mode		= 0644,
502		.proc_handler	= proc_dointvec_jiffies,
503	},
504	{
505		.procname	= "message_burst",
506		.data		= &net_ratelimit_state.burst,
507		.maxlen		= sizeof(int),
508		.mode		= 0644,
509		.proc_handler	= proc_dointvec,
510	},
511	{
 
 
 
 
 
 
 
512		.procname	= "tstamp_allow_data",
513		.data		= &sysctl_tstamp_allow_data,
514		.maxlen		= sizeof(int),
515		.mode		= 0644,
516		.proc_handler	= proc_dointvec_minmax,
517		.extra1		= SYSCTL_ZERO,
518		.extra2		= SYSCTL_ONE
519	},
520#ifdef CONFIG_RPS
521	{
522		.procname	= "rps_sock_flow_entries",
523		.maxlen		= sizeof(int),
524		.mode		= 0644,
525		.proc_handler	= rps_sock_flow_sysctl
526	},
527#endif
528#ifdef CONFIG_NET_FLOW_LIMIT
529	{
530		.procname	= "flow_limit_cpu_bitmap",
531		.mode		= 0644,
532		.proc_handler	= flow_limit_cpu_sysctl
533	},
534	{
535		.procname	= "flow_limit_table_len",
536		.data		= &netdev_flow_limit_table_len,
537		.maxlen		= sizeof(int),
538		.mode		= 0644,
539		.proc_handler	= flow_limit_table_len_sysctl
540	},
541#endif /* CONFIG_NET_FLOW_LIMIT */
542#ifdef CONFIG_NET_RX_BUSY_POLL
543	{
544		.procname	= "busy_poll",
545		.data		= &sysctl_net_busy_poll,
546		.maxlen		= sizeof(unsigned int),
547		.mode		= 0644,
548		.proc_handler	= proc_dointvec_minmax,
549		.extra1		= SYSCTL_ZERO,
550	},
551	{
552		.procname	= "busy_read",
553		.data		= &sysctl_net_busy_read,
554		.maxlen		= sizeof(unsigned int),
555		.mode		= 0644,
556		.proc_handler	= proc_dointvec_minmax,
557		.extra1		= SYSCTL_ZERO,
558	},
559#endif
560#ifdef CONFIG_NET_SCHED
561	{
562		.procname	= "default_qdisc",
563		.mode		= 0644,
564		.maxlen		= IFNAMSIZ,
565		.proc_handler	= set_default_qdisc
566	},
567#endif
 
568	{
569		.procname	= "netdev_budget",
570		.data		= &netdev_budget,
571		.maxlen		= sizeof(int),
572		.mode		= 0644,
573		.proc_handler	= proc_dointvec
574	},
575	{
576		.procname	= "warnings",
577		.data		= &net_msg_warn,
578		.maxlen		= sizeof(int),
579		.mode		= 0644,
580		.proc_handler	= proc_dointvec
581	},
582	{
583		.procname	= "max_skb_frags",
584		.data		= &sysctl_max_skb_frags,
585		.maxlen		= sizeof(int),
586		.mode		= 0644,
587		.proc_handler	= proc_dointvec_minmax,
588		.extra1		= SYSCTL_ONE,
589		.extra2		= &max_skb_frags,
590	},
591	{
592		.procname	= "netdev_budget_usecs",
593		.data		= &netdev_budget_usecs,
594		.maxlen		= sizeof(unsigned int),
595		.mode		= 0644,
596		.proc_handler	= proc_dointvec_minmax,
597		.extra1		= SYSCTL_ZERO,
598	},
599	{
600		.procname	= "fb_tunnels_only_for_init_net",
601		.data		= &sysctl_fb_tunnels_only_for_init_net,
602		.maxlen		= sizeof(int),
603		.mode		= 0644,
604		.proc_handler	= proc_dointvec_minmax,
605		.extra1		= SYSCTL_ZERO,
606		.extra2		= SYSCTL_TWO,
607	},
608	{
609		.procname	= "devconf_inherit_init_net",
610		.data		= &sysctl_devconf_inherit_init_net,
611		.maxlen		= sizeof(int),
612		.mode		= 0644,
613		.proc_handler	= proc_dointvec_minmax,
614		.extra1		= SYSCTL_ZERO,
615		.extra2		= SYSCTL_THREE,
616	},
617	{
618		.procname	= "high_order_alloc_disable",
619		.data		= &net_high_order_alloc_disable_key.key,
620		.maxlen         = sizeof(net_high_order_alloc_disable_key),
621		.mode		= 0644,
622		.proc_handler	= proc_do_static_key,
623	},
624	{
625		.procname	= "gro_normal_batch",
626		.data		= &gro_normal_batch,
627		.maxlen		= sizeof(unsigned int),
628		.mode		= 0644,
629		.proc_handler	= proc_dointvec_minmax,
630		.extra1		= SYSCTL_ONE,
631	},
632	{
633		.procname	= "netdev_unregister_timeout_secs",
634		.data		= &netdev_unregister_timeout_secs,
635		.maxlen		= sizeof(unsigned int),
636		.mode		= 0644,
637		.proc_handler	= proc_dointvec_minmax,
638		.extra1		= SYSCTL_ONE,
639		.extra2		= &int_3600,
640	},
641	{
642		.procname	= "skb_defer_max",
643		.data		= &sysctl_skb_defer_max,
644		.maxlen		= sizeof(unsigned int),
645		.mode		= 0644,
646		.proc_handler	= proc_dointvec_minmax,
647		.extra1		= SYSCTL_ZERO,
648	},
649	{ }
650};
651
652static struct ctl_table netns_core_table[] = {
653#if IS_ENABLED(CONFIG_RPS)
654	{
655		.procname	= "rps_default_mask",
656		.data		= &init_net,
657		.mode		= 0644,
658		.proc_handler	= rps_default_mask_sysctl
659	},
660#endif
661	{
662		.procname	= "somaxconn",
663		.data		= &init_net.core.sysctl_somaxconn,
664		.maxlen		= sizeof(int),
665		.mode		= 0644,
666		.extra1		= SYSCTL_ZERO,
667		.proc_handler	= proc_dointvec_minmax
668	},
669	{
670		.procname	= "optmem_max",
671		.data		= &init_net.core.sysctl_optmem_max,
672		.maxlen		= sizeof(int),
673		.mode		= 0644,
674		.extra1		= SYSCTL_ZERO,
675		.proc_handler	= proc_dointvec_minmax
676	},
677	{
678		.procname	= "txrehash",
679		.data		= &init_net.core.sysctl_txrehash,
680		.maxlen		= sizeof(u8),
681		.mode		= 0644,
682		.extra1		= SYSCTL_ZERO,
683		.extra2		= SYSCTL_ONE,
684		.proc_handler	= proc_dou8vec_minmax,
685	},
686	{ }
687};
688
689static int __init fb_tunnels_only_for_init_net_sysctl_setup(char *str)
690{
691	/* fallback tunnels for initns only */
692	if (!strncmp(str, "initns", 6))
693		sysctl_fb_tunnels_only_for_init_net = 1;
694	/* no fallback tunnels anywhere */
695	else if (!strncmp(str, "none", 4))
696		sysctl_fb_tunnels_only_for_init_net = 2;
697
698	return 1;
699}
700__setup("fb_tunnels=", fb_tunnels_only_for_init_net_sysctl_setup);
701
702static __net_init int sysctl_core_net_init(struct net *net)
703{
704	struct ctl_table *tbl, *tmp;
 
 
705
706	tbl = netns_core_table;
707	if (!net_eq(net, &init_net)) {
708		tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
709		if (tbl == NULL)
710			goto err_dup;
711
712		for (tmp = tbl; tmp->procname; tmp++)
713			tmp->data += (char *)net - (char *)&init_net;
 
 
 
 
714	}
715
716	net->core.sysctl_hdr = register_net_sysctl_sz(net, "net/core", tbl,
717						      ARRAY_SIZE(netns_core_table));
718	if (net->core.sysctl_hdr == NULL)
719		goto err_reg;
720
721	return 0;
722
723err_reg:
724	if (tbl != netns_core_table)
725		kfree(tbl);
726err_dup:
727	return -ENOMEM;
728}
729
730static __net_exit void sysctl_core_net_exit(struct net *net)
731{
732	struct ctl_table *tbl;
733
734	tbl = net->core.sysctl_hdr->ctl_table_arg;
735	unregister_net_sysctl_table(net->core.sysctl_hdr);
736	BUG_ON(tbl == netns_core_table);
737#if IS_ENABLED(CONFIG_RPS)
738	kfree(net->core.rps_default_mask);
739#endif
740	kfree(tbl);
741}
742
743static __net_initdata struct pernet_operations sysctl_core_ops = {
744	.init = sysctl_core_net_init,
745	.exit = sysctl_core_net_exit,
746};
747
748static __init int sysctl_core_init(void)
749{
750	register_net_sysctl(&init_net, "net/core", net_core_table);
751	return register_pernet_subsys(&sysctl_core_ops);
752}
753
754fs_initcall(sysctl_core_init);
v4.6
 
  1/* -*- linux-c -*-
  2 * sysctl_net_core.c: sysctl interface to net core subsystem.
  3 *
  4 * Begun April 1, 1996, Mike Shaver.
  5 * Added /proc/sys/net/core directory entry (empty =) ). [MS]
  6 */
  7
 
  8#include <linux/mm.h>
  9#include <linux/sysctl.h>
 10#include <linux/module.h>
 11#include <linux/socket.h>
 12#include <linux/netdevice.h>
 13#include <linux/ratelimit.h>
 14#include <linux/vmalloc.h>
 15#include <linux/init.h>
 16#include <linux/slab.h>
 17#include <linux/kmemleak.h>
 18
 19#include <net/ip.h>
 20#include <net/sock.h>
 21#include <net/net_ratelimit.h>
 22#include <net/busy_poll.h>
 23#include <net/pkt_sched.h>
 24
 25static int zero = 0;
 26static int one = 1;
 
 27static int min_sndbuf = SOCK_MIN_SNDBUF;
 28static int min_rcvbuf = SOCK_MIN_RCVBUF;
 29static int max_skb_frags = MAX_SKB_FRAGS;
 30
 31static int net_msg_warn;	/* Unused, but still a sysctl */
 32
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 33#ifdef CONFIG_RPS
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 34static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
 35				void __user *buffer, size_t *lenp, loff_t *ppos)
 36{
 37	unsigned int orig_size, size;
 38	int ret, i;
 39	struct ctl_table tmp = {
 40		.data = &size,
 41		.maxlen = sizeof(size),
 42		.mode = table->mode
 43	};
 44	struct rps_sock_flow_table *orig_sock_table, *sock_table;
 45	static DEFINE_MUTEX(sock_flow_mutex);
 46
 47	mutex_lock(&sock_flow_mutex);
 48
 49	orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
 50					lockdep_is_held(&sock_flow_mutex));
 51	size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
 52
 53	ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
 54
 55	if (write) {
 56		if (size) {
 57			if (size > 1<<29) {
 58				/* Enforce limit to prevent overflow */
 59				mutex_unlock(&sock_flow_mutex);
 60				return -EINVAL;
 61			}
 62			size = roundup_pow_of_two(size);
 63			if (size != orig_size) {
 64				sock_table =
 65				    vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
 66				if (!sock_table) {
 67					mutex_unlock(&sock_flow_mutex);
 68					return -ENOMEM;
 69				}
 70				rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
 71				sock_table->mask = size - 1;
 72			} else
 73				sock_table = orig_sock_table;
 74
 75			for (i = 0; i < size; i++)
 76				sock_table->ents[i] = RPS_NO_CPU;
 77		} else
 78			sock_table = NULL;
 79
 80		if (sock_table != orig_sock_table) {
 81			rcu_assign_pointer(rps_sock_flow_table, sock_table);
 82			if (sock_table)
 83				static_key_slow_inc(&rps_needed);
 
 
 84			if (orig_sock_table) {
 85				static_key_slow_dec(&rps_needed);
 86				synchronize_rcu();
 87				vfree(orig_sock_table);
 88			}
 89		}
 90	}
 91
 92	mutex_unlock(&sock_flow_mutex);
 93
 94	return ret;
 95}
 96#endif /* CONFIG_RPS */
 97
 98#ifdef CONFIG_NET_FLOW_LIMIT
 99static DEFINE_MUTEX(flow_limit_update_mutex);
100
101static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
102				 void __user *buffer, size_t *lenp,
103				 loff_t *ppos)
104{
105	struct sd_flow_limit *cur;
106	struct softnet_data *sd;
107	cpumask_var_t mask;
108	int i, len, ret = 0;
109
110	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
111		return -ENOMEM;
112
113	if (write) {
114		ret = cpumask_parse_user(buffer, *lenp, mask);
115		if (ret)
116			goto done;
117
118		mutex_lock(&flow_limit_update_mutex);
119		len = sizeof(*cur) + netdev_flow_limit_table_len;
120		for_each_possible_cpu(i) {
121			sd = &per_cpu(softnet_data, i);
122			cur = rcu_dereference_protected(sd->flow_limit,
123				     lockdep_is_held(&flow_limit_update_mutex));
124			if (cur && !cpumask_test_cpu(i, mask)) {
125				RCU_INIT_POINTER(sd->flow_limit, NULL);
126				synchronize_rcu();
127				kfree(cur);
128			} else if (!cur && cpumask_test_cpu(i, mask)) {
129				cur = kzalloc_node(len, GFP_KERNEL,
130						   cpu_to_node(i));
131				if (!cur) {
132					/* not unwinding previous changes */
133					ret = -ENOMEM;
134					goto write_unlock;
135				}
136				cur->num_buckets = netdev_flow_limit_table_len;
137				rcu_assign_pointer(sd->flow_limit, cur);
138			}
139		}
140write_unlock:
141		mutex_unlock(&flow_limit_update_mutex);
142	} else {
143		char kbuf[128];
144
145		if (*ppos || !*lenp) {
146			*lenp = 0;
147			goto done;
148		}
149
150		cpumask_clear(mask);
151		rcu_read_lock();
152		for_each_possible_cpu(i) {
153			sd = &per_cpu(softnet_data, i);
154			if (rcu_dereference(sd->flow_limit))
155				cpumask_set_cpu(i, mask);
156		}
157		rcu_read_unlock();
158
159		len = min(sizeof(kbuf) - 1, *lenp);
160		len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
161		if (!len) {
162			*lenp = 0;
163			goto done;
164		}
165		if (len < *lenp)
166			kbuf[len++] = '\n';
167		if (copy_to_user(buffer, kbuf, len)) {
168			ret = -EFAULT;
169			goto done;
170		}
171		*lenp = len;
172		*ppos += len;
173	}
174
175done:
176	free_cpumask_var(mask);
177	return ret;
178}
179
180static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
181				       void __user *buffer, size_t *lenp,
182				       loff_t *ppos)
183{
184	unsigned int old, *ptr;
185	int ret;
186
187	mutex_lock(&flow_limit_update_mutex);
188
189	ptr = table->data;
190	old = *ptr;
191	ret = proc_dointvec(table, write, buffer, lenp, ppos);
192	if (!ret && write && !is_power_of_2(*ptr)) {
193		*ptr = old;
194		ret = -EINVAL;
195	}
196
197	mutex_unlock(&flow_limit_update_mutex);
198	return ret;
199}
200#endif /* CONFIG_NET_FLOW_LIMIT */
201
202#ifdef CONFIG_NET_SCHED
203static int set_default_qdisc(struct ctl_table *table, int write,
204			     void __user *buffer, size_t *lenp, loff_t *ppos)
205{
206	char id[IFNAMSIZ];
207	struct ctl_table tbl = {
208		.data = id,
209		.maxlen = IFNAMSIZ,
210	};
211	int ret;
212
213	qdisc_get_default(id, IFNAMSIZ);
214
215	ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
216	if (write && ret == 0)
217		ret = qdisc_set_default(id);
218	return ret;
219}
220#endif
221
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
222static int proc_do_rss_key(struct ctl_table *table, int write,
223			   void __user *buffer, size_t *lenp, loff_t *ppos)
224{
225	struct ctl_table fake_table;
226	char buf[NETDEV_RSS_KEY_LEN * 3];
227
228	snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
229	fake_table.data = buf;
230	fake_table.maxlen = sizeof(buf);
231	return proc_dostring(&fake_table, write, buffer, lenp, ppos);
232}
233
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
234static struct ctl_table net_core_table[] = {
235#ifdef CONFIG_NET
236	{
237		.procname	= "wmem_max",
238		.data		= &sysctl_wmem_max,
239		.maxlen		= sizeof(int),
240		.mode		= 0644,
241		.proc_handler	= proc_dointvec_minmax,
242		.extra1		= &min_sndbuf,
243	},
244	{
245		.procname	= "rmem_max",
246		.data		= &sysctl_rmem_max,
247		.maxlen		= sizeof(int),
248		.mode		= 0644,
249		.proc_handler	= proc_dointvec_minmax,
250		.extra1		= &min_rcvbuf,
251	},
252	{
253		.procname	= "wmem_default",
254		.data		= &sysctl_wmem_default,
255		.maxlen		= sizeof(int),
256		.mode		= 0644,
257		.proc_handler	= proc_dointvec_minmax,
258		.extra1		= &min_sndbuf,
259	},
260	{
261		.procname	= "rmem_default",
262		.data		= &sysctl_rmem_default,
263		.maxlen		= sizeof(int),
264		.mode		= 0644,
265		.proc_handler	= proc_dointvec_minmax,
266		.extra1		= &min_rcvbuf,
267	},
268	{
269		.procname	= "dev_weight",
270		.data		= &weight_p,
271		.maxlen		= sizeof(int),
272		.mode		= 0644,
273		.proc_handler	= proc_dointvec
 
 
 
 
 
 
 
 
 
 
 
 
 
 
274	},
275	{
276		.procname	= "netdev_max_backlog",
277		.data		= &netdev_max_backlog,
278		.maxlen		= sizeof(int),
279		.mode		= 0644,
280		.proc_handler	= proc_dointvec
281	},
282	{
283		.procname	= "netdev_rss_key",
284		.data		= &netdev_rss_key,
285		.maxlen		= sizeof(int),
286		.mode		= 0444,
287		.proc_handler	= proc_do_rss_key,
288	},
289#ifdef CONFIG_BPF_JIT
290	{
291		.procname	= "bpf_jit_enable",
292		.data		= &bpf_jit_enable,
293		.maxlen		= sizeof(int),
294		.mode		= 0644,
295		.proc_handler	= proc_dointvec
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
296	},
297#endif
298	{
299		.procname	= "netdev_tstamp_prequeue",
300		.data		= &netdev_tstamp_prequeue,
301		.maxlen		= sizeof(int),
302		.mode		= 0644,
303		.proc_handler	= proc_dointvec
304	},
305	{
306		.procname	= "message_cost",
307		.data		= &net_ratelimit_state.interval,
308		.maxlen		= sizeof(int),
309		.mode		= 0644,
310		.proc_handler	= proc_dointvec_jiffies,
311	},
312	{
313		.procname	= "message_burst",
314		.data		= &net_ratelimit_state.burst,
315		.maxlen		= sizeof(int),
316		.mode		= 0644,
317		.proc_handler	= proc_dointvec,
318	},
319	{
320		.procname	= "optmem_max",
321		.data		= &sysctl_optmem_max,
322		.maxlen		= sizeof(int),
323		.mode		= 0644,
324		.proc_handler	= proc_dointvec
325	},
326	{
327		.procname	= "tstamp_allow_data",
328		.data		= &sysctl_tstamp_allow_data,
329		.maxlen		= sizeof(int),
330		.mode		= 0644,
331		.proc_handler	= proc_dointvec_minmax,
332		.extra1		= &zero,
333		.extra2		= &one
334	},
335#ifdef CONFIG_RPS
336	{
337		.procname	= "rps_sock_flow_entries",
338		.maxlen		= sizeof(int),
339		.mode		= 0644,
340		.proc_handler	= rps_sock_flow_sysctl
341	},
342#endif
343#ifdef CONFIG_NET_FLOW_LIMIT
344	{
345		.procname	= "flow_limit_cpu_bitmap",
346		.mode		= 0644,
347		.proc_handler	= flow_limit_cpu_sysctl
348	},
349	{
350		.procname	= "flow_limit_table_len",
351		.data		= &netdev_flow_limit_table_len,
352		.maxlen		= sizeof(int),
353		.mode		= 0644,
354		.proc_handler	= flow_limit_table_len_sysctl
355	},
356#endif /* CONFIG_NET_FLOW_LIMIT */
357#ifdef CONFIG_NET_RX_BUSY_POLL
358	{
359		.procname	= "busy_poll",
360		.data		= &sysctl_net_busy_poll,
361		.maxlen		= sizeof(unsigned int),
362		.mode		= 0644,
363		.proc_handler	= proc_dointvec
 
364	},
365	{
366		.procname	= "busy_read",
367		.data		= &sysctl_net_busy_read,
368		.maxlen		= sizeof(unsigned int),
369		.mode		= 0644,
370		.proc_handler	= proc_dointvec
 
371	},
372#endif
373#ifdef CONFIG_NET_SCHED
374	{
375		.procname	= "default_qdisc",
376		.mode		= 0644,
377		.maxlen		= IFNAMSIZ,
378		.proc_handler	= set_default_qdisc
379	},
380#endif
381#endif /* CONFIG_NET */
382	{
383		.procname	= "netdev_budget",
384		.data		= &netdev_budget,
385		.maxlen		= sizeof(int),
386		.mode		= 0644,
387		.proc_handler	= proc_dointvec
388	},
389	{
390		.procname	= "warnings",
391		.data		= &net_msg_warn,
392		.maxlen		= sizeof(int),
393		.mode		= 0644,
394		.proc_handler	= proc_dointvec
395	},
396	{
397		.procname	= "max_skb_frags",
398		.data		= &sysctl_max_skb_frags,
399		.maxlen		= sizeof(int),
400		.mode		= 0644,
401		.proc_handler	= proc_dointvec_minmax,
402		.extra1		= &one,
403		.extra2		= &max_skb_frags,
404	},
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
405	{ }
406};
407
408static struct ctl_table netns_core_table[] = {
 
 
 
 
 
 
 
 
409	{
410		.procname	= "somaxconn",
411		.data		= &init_net.core.sysctl_somaxconn,
412		.maxlen		= sizeof(int),
413		.mode		= 0644,
414		.extra1		= &zero,
415		.proc_handler	= proc_dointvec_minmax
416	},
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
417	{ }
418};
419
 
 
 
 
 
 
 
 
 
 
 
 
 
420static __net_init int sysctl_core_net_init(struct net *net)
421{
422	struct ctl_table *tbl;
423
424	net->core.sysctl_somaxconn = SOMAXCONN;
425
426	tbl = netns_core_table;
427	if (!net_eq(net, &init_net)) {
428		tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
429		if (tbl == NULL)
430			goto err_dup;
431
432		tbl[0].data = &net->core.sysctl_somaxconn;
433
434		/* Don't export any sysctls to unprivileged users */
435		if (net->user_ns != &init_user_ns) {
436			tbl[0].procname = NULL;
437		}
438	}
439
440	net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
 
441	if (net->core.sysctl_hdr == NULL)
442		goto err_reg;
443
444	return 0;
445
446err_reg:
447	if (tbl != netns_core_table)
448		kfree(tbl);
449err_dup:
450	return -ENOMEM;
451}
452
453static __net_exit void sysctl_core_net_exit(struct net *net)
454{
455	struct ctl_table *tbl;
456
457	tbl = net->core.sysctl_hdr->ctl_table_arg;
458	unregister_net_sysctl_table(net->core.sysctl_hdr);
459	BUG_ON(tbl == netns_core_table);
 
 
 
460	kfree(tbl);
461}
462
463static __net_initdata struct pernet_operations sysctl_core_ops = {
464	.init = sysctl_core_net_init,
465	.exit = sysctl_core_net_exit,
466};
467
468static __init int sysctl_core_init(void)
469{
470	register_net_sysctl(&init_net, "net/core", net_core_table);
471	return register_pernet_subsys(&sysctl_core_ops);
472}
473
474fs_initcall(sysctl_core_init);