1 // SPDX-License-Identifier: GPL-2.0
3 * sysctl_net_core.c: sysctl interface to net core subsystem.
5 * Begun April 1, 1996, Mike Shaver.
6 * Added /proc/sys/net/core directory entry (empty =) ). [MS]
9 #include <linux/filter.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>
23 #include <net/net_ratelimit.h>
24 #include <net/busy_poll.h>
25 #include <net/pkt_sched.h>
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;
34 static int net_msg_warn; /* Unused, but still a sysctl */
36 int sysctl_fb_tunnels_only_for_init_net __read_mostly = 0;
37 EXPORT_SYMBOL(sysctl_fb_tunnels_only_for_init_net);
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
46 int sysctl_devconf_inherit_init_net __read_mostly;
47 EXPORT_SYMBOL(sysctl_devconf_inherit_init_net);
49 #if IS_ENABLED(CONFIG_NET_FLOW_LIMIT) || IS_ENABLED(CONFIG_RPS)
50 static void dump_cpumask(void *buffer, size_t *lenp, loff_t *ppos,
56 if (*ppos || !*lenp) {
61 len = min(sizeof(kbuf) - 1, *lenp);
62 len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
70 memcpy(buffer, kbuf, len);
78 static struct cpumask *rps_default_mask_cow_alloc(struct net *net)
80 struct cpumask *rps_default_mask;
82 if (net->core.rps_default_mask)
83 return net->core.rps_default_mask;
85 rps_default_mask = kzalloc(cpumask_size(), GFP_KERNEL);
86 if (!rps_default_mask)
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;
94 static int rps_default_mask_sysctl(struct ctl_table *table, int write,
95 void *buffer, size_t *lenp, loff_t *ppos)
97 struct net *net = (struct net *)table->data;
102 struct cpumask *rps_default_mask = rps_default_mask_cow_alloc(net);
105 if (!rps_default_mask)
108 err = cpumask_parse(buffer, rps_default_mask);
112 err = rps_cpumask_housekeeping(rps_default_mask);
116 dump_cpumask(buffer, lenp, ppos,
117 net->core.rps_default_mask ? : cpu_none_mask);
125 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
126 void *buffer, size_t *lenp, loff_t *ppos)
128 unsigned int orig_size, size;
130 struct ctl_table tmp = {
132 .maxlen = sizeof(size),
135 struct rps_sock_flow_table *orig_sock_table, *sock_table;
136 static DEFINE_MUTEX(sock_flow_mutex);
138 mutex_lock(&sock_flow_mutex);
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;
144 ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
149 /* Enforce limit to prevent overflow */
150 mutex_unlock(&sock_flow_mutex);
153 size = roundup_pow_of_two(size);
154 if (size != orig_size) {
156 vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
158 mutex_unlock(&sock_flow_mutex);
161 rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
162 sock_table->mask = size - 1;
164 sock_table = orig_sock_table;
166 for (i = 0; i < size; i++)
167 sock_table->ents[i] = RPS_NO_CPU;
171 if (sock_table != orig_sock_table) {
172 rcu_assign_pointer(rps_sock_flow_table, sock_table);
174 static_branch_inc(&rps_needed);
175 static_branch_inc(&rfs_needed);
177 if (orig_sock_table) {
178 static_branch_dec(&rps_needed);
179 static_branch_dec(&rfs_needed);
180 kvfree_rcu_mightsleep(orig_sock_table);
185 mutex_unlock(&sock_flow_mutex);
189 #endif /* CONFIG_RPS */
191 #ifdef CONFIG_NET_FLOW_LIMIT
192 static DEFINE_MUTEX(flow_limit_update_mutex);
194 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
195 void *buffer, size_t *lenp, loff_t *ppos)
197 struct sd_flow_limit *cur;
198 struct softnet_data *sd;
202 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
206 ret = cpumask_parse(buffer, mask);
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,
223 /* not unwinding previous changes */
227 cur->num_buckets = netdev_flow_limit_table_len;
228 rcu_assign_pointer(sd->flow_limit, cur);
232 mutex_unlock(&flow_limit_update_mutex);
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);
243 dump_cpumask(buffer, lenp, ppos, mask);
247 free_cpumask_var(mask);
251 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
252 void *buffer, size_t *lenp, loff_t *ppos)
254 unsigned int old, *ptr;
257 mutex_lock(&flow_limit_update_mutex);
261 ret = proc_dointvec(table, write, buffer, lenp, ppos);
262 if (!ret && write && !is_power_of_2(*ptr)) {
267 mutex_unlock(&flow_limit_update_mutex);
270 #endif /* CONFIG_NET_FLOW_LIMIT */
272 #ifdef CONFIG_NET_SCHED
273 static int set_default_qdisc(struct ctl_table *table, int write,
274 void *buffer, size_t *lenp, loff_t *ppos)
277 struct ctl_table tbl = {
283 qdisc_get_default(id, IFNAMSIZ);
285 ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
286 if (write && ret == 0)
287 ret = qdisc_set_default(id);
292 static int proc_do_dev_weight(struct ctl_table *table, int write,
293 void *buffer, size_t *lenp, loff_t *ppos)
295 static DEFINE_MUTEX(dev_weight_mutex);
298 mutex_lock(&dev_weight_mutex);
299 ret = proc_dointvec(table, write, buffer, lenp, ppos);
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);
305 mutex_unlock(&dev_weight_mutex);
310 static int proc_do_rss_key(struct ctl_table *table, int write,
311 void *buffer, size_t *lenp, loff_t *ppos)
313 struct ctl_table fake_table;
314 char buf[NETDEV_RSS_KEY_LEN * 3];
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);
322 #ifdef CONFIG_BPF_JIT
323 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write,
324 void *buffer, size_t *lenp,
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;
332 if (write && !capable(CAP_SYS_ADMIN))
335 tmp.data = &jit_enable;
336 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
338 if (jit_enable < 2 ||
339 (jit_enable == 2 && bpf_dump_raw_ok(current_cred()))) {
340 *(int *)table->data = jit_enable;
342 pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
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");
354 # ifdef CONFIG_HAVE_EBPF_JIT
356 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write,
357 void *buffer, size_t *lenp, loff_t *ppos)
359 if (!capable(CAP_SYS_ADMIN))
362 return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
364 # endif /* CONFIG_HAVE_EBPF_JIT */
367 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
368 void *buffer, size_t *lenp, loff_t *ppos)
370 if (!capable(CAP_SYS_ADMIN))
373 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
377 static struct ctl_table net_core_table[] = {
379 .procname = "wmem_max",
380 .data = &sysctl_wmem_max,
381 .maxlen = sizeof(int),
383 .proc_handler = proc_dointvec_minmax,
384 .extra1 = &min_sndbuf,
387 .procname = "rmem_max",
388 .data = &sysctl_rmem_max,
389 .maxlen = sizeof(int),
391 .proc_handler = proc_dointvec_minmax,
392 .extra1 = &min_rcvbuf,
395 .procname = "wmem_default",
396 .data = &sysctl_wmem_default,
397 .maxlen = sizeof(int),
399 .proc_handler = proc_dointvec_minmax,
400 .extra1 = &min_sndbuf,
403 .procname = "rmem_default",
404 .data = &sysctl_rmem_default,
405 .maxlen = sizeof(int),
407 .proc_handler = proc_dointvec_minmax,
408 .extra1 = &min_rcvbuf,
411 .procname = "dev_weight",
413 .maxlen = sizeof(int),
415 .proc_handler = proc_do_dev_weight,
418 .procname = "dev_weight_rx_bias",
419 .data = &dev_weight_rx_bias,
420 .maxlen = sizeof(int),
422 .proc_handler = proc_do_dev_weight,
425 .procname = "dev_weight_tx_bias",
426 .data = &dev_weight_tx_bias,
427 .maxlen = sizeof(int),
429 .proc_handler = proc_do_dev_weight,
432 .procname = "netdev_max_backlog",
433 .data = &netdev_max_backlog,
434 .maxlen = sizeof(int),
436 .proc_handler = proc_dointvec
439 .procname = "netdev_rss_key",
440 .data = &netdev_rss_key,
441 .maxlen = sizeof(int),
443 .proc_handler = proc_do_rss_key,
445 #ifdef CONFIG_BPF_JIT
447 .procname = "bpf_jit_enable",
448 .data = &bpf_jit_enable,
449 .maxlen = sizeof(int),
451 .proc_handler = proc_dointvec_minmax_bpf_enable,
452 # ifdef CONFIG_BPF_JIT_ALWAYS_ON
453 .extra1 = SYSCTL_ONE,
454 .extra2 = SYSCTL_ONE,
456 .extra1 = SYSCTL_ZERO,
457 .extra2 = SYSCTL_TWO,
460 # ifdef CONFIG_HAVE_EBPF_JIT
462 .procname = "bpf_jit_harden",
463 .data = &bpf_jit_harden,
464 .maxlen = sizeof(int),
466 .proc_handler = proc_dointvec_minmax_bpf_restricted,
467 .extra1 = SYSCTL_ZERO,
468 .extra2 = SYSCTL_TWO,
471 .procname = "bpf_jit_kallsyms",
472 .data = &bpf_jit_kallsyms,
473 .maxlen = sizeof(int),
475 .proc_handler = proc_dointvec_minmax_bpf_restricted,
476 .extra1 = SYSCTL_ZERO,
477 .extra2 = SYSCTL_ONE,
481 .procname = "bpf_jit_limit",
482 .data = &bpf_jit_limit,
483 .maxlen = sizeof(long),
485 .proc_handler = proc_dolongvec_minmax_bpf_restricted,
486 .extra1 = SYSCTL_LONG_ONE,
487 .extra2 = &bpf_jit_limit_max,
491 .procname = "netdev_tstamp_prequeue",
492 .data = &netdev_tstamp_prequeue,
493 .maxlen = sizeof(int),
495 .proc_handler = proc_dointvec
498 .procname = "message_cost",
499 .data = &net_ratelimit_state.interval,
500 .maxlen = sizeof(int),
502 .proc_handler = proc_dointvec_jiffies,
505 .procname = "message_burst",
506 .data = &net_ratelimit_state.burst,
507 .maxlen = sizeof(int),
509 .proc_handler = proc_dointvec,
512 .procname = "optmem_max",
513 .data = &sysctl_optmem_max,
514 .maxlen = sizeof(int),
516 .proc_handler = proc_dointvec
519 .procname = "tstamp_allow_data",
520 .data = &sysctl_tstamp_allow_data,
521 .maxlen = sizeof(int),
523 .proc_handler = proc_dointvec_minmax,
524 .extra1 = SYSCTL_ZERO,
529 .procname = "rps_sock_flow_entries",
530 .maxlen = sizeof(int),
532 .proc_handler = rps_sock_flow_sysctl
535 #ifdef CONFIG_NET_FLOW_LIMIT
537 .procname = "flow_limit_cpu_bitmap",
539 .proc_handler = flow_limit_cpu_sysctl
542 .procname = "flow_limit_table_len",
543 .data = &netdev_flow_limit_table_len,
544 .maxlen = sizeof(int),
546 .proc_handler = flow_limit_table_len_sysctl
548 #endif /* CONFIG_NET_FLOW_LIMIT */
549 #ifdef CONFIG_NET_RX_BUSY_POLL
551 .procname = "busy_poll",
552 .data = &sysctl_net_busy_poll,
553 .maxlen = sizeof(unsigned int),
555 .proc_handler = proc_dointvec_minmax,
556 .extra1 = SYSCTL_ZERO,
559 .procname = "busy_read",
560 .data = &sysctl_net_busy_read,
561 .maxlen = sizeof(unsigned int),
563 .proc_handler = proc_dointvec_minmax,
564 .extra1 = SYSCTL_ZERO,
567 #ifdef CONFIG_NET_SCHED
569 .procname = "default_qdisc",
572 .proc_handler = set_default_qdisc
576 .procname = "netdev_budget",
577 .data = &netdev_budget,
578 .maxlen = sizeof(int),
580 .proc_handler = proc_dointvec
583 .procname = "warnings",
584 .data = &net_msg_warn,
585 .maxlen = sizeof(int),
587 .proc_handler = proc_dointvec
590 .procname = "max_skb_frags",
591 .data = &sysctl_max_skb_frags,
592 .maxlen = sizeof(int),
594 .proc_handler = proc_dointvec_minmax,
595 .extra1 = SYSCTL_ONE,
596 .extra2 = &max_skb_frags,
599 .procname = "netdev_budget_usecs",
600 .data = &netdev_budget_usecs,
601 .maxlen = sizeof(unsigned int),
603 .proc_handler = proc_dointvec_minmax,
604 .extra1 = SYSCTL_ZERO,
607 .procname = "fb_tunnels_only_for_init_net",
608 .data = &sysctl_fb_tunnels_only_for_init_net,
609 .maxlen = sizeof(int),
611 .proc_handler = proc_dointvec_minmax,
612 .extra1 = SYSCTL_ZERO,
613 .extra2 = SYSCTL_TWO,
616 .procname = "devconf_inherit_init_net",
617 .data = &sysctl_devconf_inherit_init_net,
618 .maxlen = sizeof(int),
620 .proc_handler = proc_dointvec_minmax,
621 .extra1 = SYSCTL_ZERO,
622 .extra2 = SYSCTL_THREE,
625 .procname = "high_order_alloc_disable",
626 .data = &net_high_order_alloc_disable_key.key,
627 .maxlen = sizeof(net_high_order_alloc_disable_key),
629 .proc_handler = proc_do_static_key,
632 .procname = "gro_normal_batch",
633 .data = &gro_normal_batch,
634 .maxlen = sizeof(unsigned int),
636 .proc_handler = proc_dointvec_minmax,
637 .extra1 = SYSCTL_ONE,
640 .procname = "netdev_unregister_timeout_secs",
641 .data = &netdev_unregister_timeout_secs,
642 .maxlen = sizeof(unsigned int),
644 .proc_handler = proc_dointvec_minmax,
645 .extra1 = SYSCTL_ONE,
649 .procname = "skb_defer_max",
650 .data = &sysctl_skb_defer_max,
651 .maxlen = sizeof(unsigned int),
653 .proc_handler = proc_dointvec_minmax,
654 .extra1 = SYSCTL_ZERO,
659 static struct ctl_table netns_core_table[] = {
660 #if IS_ENABLED(CONFIG_RPS)
662 .procname = "rps_default_mask",
665 .proc_handler = rps_default_mask_sysctl
669 .procname = "somaxconn",
670 .data = &init_net.core.sysctl_somaxconn,
671 .maxlen = sizeof(int),
673 .extra1 = SYSCTL_ZERO,
674 .proc_handler = proc_dointvec_minmax
677 .procname = "txrehash",
678 .data = &init_net.core.sysctl_txrehash,
679 .maxlen = sizeof(u8),
681 .extra1 = SYSCTL_ZERO,
682 .extra2 = SYSCTL_ONE,
683 .proc_handler = proc_dou8vec_minmax,
688 static int __init fb_tunnels_only_for_init_net_sysctl_setup(char *str)
690 /* fallback tunnels for initns only */
691 if (!strncmp(str, "initns", 6))
692 sysctl_fb_tunnels_only_for_init_net = 1;
693 /* no fallback tunnels anywhere */
694 else if (!strncmp(str, "none", 4))
695 sysctl_fb_tunnels_only_for_init_net = 2;
699 __setup("fb_tunnels=", fb_tunnels_only_for_init_net_sysctl_setup);
701 static __net_init int sysctl_core_net_init(struct net *net)
703 struct ctl_table *tbl, *tmp;
705 tbl = netns_core_table;
706 if (!net_eq(net, &init_net)) {
707 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
711 for (tmp = tbl; tmp->procname; tmp++)
712 tmp->data += (char *)net - (char *)&init_net;
715 net->core.sysctl_hdr = register_net_sysctl_sz(net, "net/core", tbl,
716 ARRAY_SIZE(netns_core_table));
717 if (net->core.sysctl_hdr == NULL)
723 if (tbl != netns_core_table)
729 static __net_exit void sysctl_core_net_exit(struct net *net)
731 struct ctl_table *tbl;
733 tbl = net->core.sysctl_hdr->ctl_table_arg;
734 unregister_net_sysctl_table(net->core.sysctl_hdr);
735 BUG_ON(tbl == netns_core_table);
736 #if IS_ENABLED(CONFIG_RPS)
737 kfree(net->core.rps_default_mask);
742 static __net_initdata struct pernet_operations sysctl_core_ops = {
743 .init = sysctl_core_net_init,
744 .exit = sysctl_core_net_exit,
747 static __init int sysctl_core_init(void)
749 register_net_sysctl(&init_net, "net/core", net_core_table);
750 return register_pernet_subsys(&sysctl_core_ops);
753 fs_initcall(sysctl_core_init);