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]
10 #include <linux/sysctl.h>
11 #include <linux/module.h>
12 #include <linux/socket.h>
13 #include <linux/netdevice.h>
14 #include <linux/ratelimit.h>
15 #include <linux/vmalloc.h>
16 #include <linux/init.h>
17 #include <linux/slab.h>
21 #include <net/net_ratelimit.h>
22 #include <net/busy_poll.h>
23 #include <net/pkt_sched.h>
25 static int two __maybe_unused = 2;
26 static int min_sndbuf = SOCK_MIN_SNDBUF;
27 static int min_rcvbuf = SOCK_MIN_RCVBUF;
28 static int max_skb_frags = MAX_SKB_FRAGS;
29 static long long_one __maybe_unused = 1;
30 static long long_max __maybe_unused = LONG_MAX;
32 static int net_msg_warn; /* Unused, but still a sysctl */
34 int sysctl_fb_tunnels_only_for_init_net __read_mostly = 0;
35 EXPORT_SYMBOL(sysctl_fb_tunnels_only_for_init_net);
37 /* 0 - Keep current behavior:
38 * IPv4: inherit all current settings from init_net
39 * IPv6: reset all settings to default
40 * 1 - Both inherit all current settings from init_net
41 * 2 - Both reset all settings to default
43 int sysctl_devconf_inherit_init_net __read_mostly;
44 EXPORT_SYMBOL(sysctl_devconf_inherit_init_net);
47 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
48 void __user *buffer, size_t *lenp, loff_t *ppos)
50 unsigned int orig_size, size;
52 struct ctl_table tmp = {
54 .maxlen = sizeof(size),
57 struct rps_sock_flow_table *orig_sock_table, *sock_table;
58 static DEFINE_MUTEX(sock_flow_mutex);
60 mutex_lock(&sock_flow_mutex);
62 orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
63 lockdep_is_held(&sock_flow_mutex));
64 size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
66 ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
71 /* Enforce limit to prevent overflow */
72 mutex_unlock(&sock_flow_mutex);
75 size = roundup_pow_of_two(size);
76 if (size != orig_size) {
78 vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
80 mutex_unlock(&sock_flow_mutex);
83 rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
84 sock_table->mask = size - 1;
86 sock_table = orig_sock_table;
88 for (i = 0; i < size; i++)
89 sock_table->ents[i] = RPS_NO_CPU;
93 if (sock_table != orig_sock_table) {
94 rcu_assign_pointer(rps_sock_flow_table, sock_table);
96 static_branch_inc(&rps_needed);
97 static_branch_inc(&rfs_needed);
99 if (orig_sock_table) {
100 static_branch_dec(&rps_needed);
101 static_branch_dec(&rfs_needed);
103 vfree(orig_sock_table);
108 mutex_unlock(&sock_flow_mutex);
112 #endif /* CONFIG_RPS */
114 #ifdef CONFIG_NET_FLOW_LIMIT
115 static DEFINE_MUTEX(flow_limit_update_mutex);
117 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
118 void __user *buffer, size_t *lenp,
121 struct sd_flow_limit *cur;
122 struct softnet_data *sd;
126 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
130 ret = cpumask_parse_user(buffer, *lenp, mask);
134 mutex_lock(&flow_limit_update_mutex);
135 len = sizeof(*cur) + netdev_flow_limit_table_len;
136 for_each_possible_cpu(i) {
137 sd = &per_cpu(softnet_data, i);
138 cur = rcu_dereference_protected(sd->flow_limit,
139 lockdep_is_held(&flow_limit_update_mutex));
140 if (cur && !cpumask_test_cpu(i, mask)) {
141 RCU_INIT_POINTER(sd->flow_limit, NULL);
144 } else if (!cur && cpumask_test_cpu(i, mask)) {
145 cur = kzalloc_node(len, GFP_KERNEL,
148 /* not unwinding previous changes */
152 cur->num_buckets = netdev_flow_limit_table_len;
153 rcu_assign_pointer(sd->flow_limit, cur);
157 mutex_unlock(&flow_limit_update_mutex);
161 if (*ppos || !*lenp) {
168 for_each_possible_cpu(i) {
169 sd = &per_cpu(softnet_data, i);
170 if (rcu_dereference(sd->flow_limit))
171 cpumask_set_cpu(i, mask);
175 len = min(sizeof(kbuf) - 1, *lenp);
176 len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
183 if (copy_to_user(buffer, kbuf, len)) {
192 free_cpumask_var(mask);
196 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
197 void __user *buffer, size_t *lenp,
200 unsigned int old, *ptr;
203 mutex_lock(&flow_limit_update_mutex);
207 ret = proc_dointvec(table, write, buffer, lenp, ppos);
208 if (!ret && write && !is_power_of_2(*ptr)) {
213 mutex_unlock(&flow_limit_update_mutex);
216 #endif /* CONFIG_NET_FLOW_LIMIT */
218 #ifdef CONFIG_NET_SCHED
219 static int set_default_qdisc(struct ctl_table *table, int write,
220 void __user *buffer, size_t *lenp, loff_t *ppos)
223 struct ctl_table tbl = {
229 qdisc_get_default(id, IFNAMSIZ);
231 ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
232 if (write && ret == 0)
233 ret = qdisc_set_default(id);
238 static int proc_do_dev_weight(struct ctl_table *table, int write,
239 void __user *buffer, size_t *lenp, loff_t *ppos)
243 ret = proc_dointvec(table, write, buffer, lenp, ppos);
247 dev_rx_weight = weight_p * dev_weight_rx_bias;
248 dev_tx_weight = weight_p * dev_weight_tx_bias;
253 static int proc_do_rss_key(struct ctl_table *table, int write,
254 void __user *buffer, size_t *lenp, loff_t *ppos)
256 struct ctl_table fake_table;
257 char buf[NETDEV_RSS_KEY_LEN * 3];
259 snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
260 fake_table.data = buf;
261 fake_table.maxlen = sizeof(buf);
262 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
265 #ifdef CONFIG_BPF_JIT
266 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write,
267 void __user *buffer, size_t *lenp,
270 int ret, jit_enable = *(int *)table->data;
271 struct ctl_table tmp = *table;
273 if (write && !capable(CAP_SYS_ADMIN))
276 tmp.data = &jit_enable;
277 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
279 if (jit_enable < 2 ||
280 (jit_enable == 2 && bpf_dump_raw_ok())) {
281 *(int *)table->data = jit_enable;
283 pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
291 # ifdef CONFIG_HAVE_EBPF_JIT
293 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write,
294 void __user *buffer, size_t *lenp,
297 if (!capable(CAP_SYS_ADMIN))
300 return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
302 # endif /* CONFIG_HAVE_EBPF_JIT */
305 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
306 void __user *buffer, size_t *lenp,
309 if (!capable(CAP_SYS_ADMIN))
312 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
316 static struct ctl_table net_core_table[] = {
319 .procname = "wmem_max",
320 .data = &sysctl_wmem_max,
321 .maxlen = sizeof(int),
323 .proc_handler = proc_dointvec_minmax,
324 .extra1 = &min_sndbuf,
327 .procname = "rmem_max",
328 .data = &sysctl_rmem_max,
329 .maxlen = sizeof(int),
331 .proc_handler = proc_dointvec_minmax,
332 .extra1 = &min_rcvbuf,
335 .procname = "wmem_default",
336 .data = &sysctl_wmem_default,
337 .maxlen = sizeof(int),
339 .proc_handler = proc_dointvec_minmax,
340 .extra1 = &min_sndbuf,
343 .procname = "rmem_default",
344 .data = &sysctl_rmem_default,
345 .maxlen = sizeof(int),
347 .proc_handler = proc_dointvec_minmax,
348 .extra1 = &min_rcvbuf,
351 .procname = "dev_weight",
353 .maxlen = sizeof(int),
355 .proc_handler = proc_do_dev_weight,
358 .procname = "dev_weight_rx_bias",
359 .data = &dev_weight_rx_bias,
360 .maxlen = sizeof(int),
362 .proc_handler = proc_do_dev_weight,
365 .procname = "dev_weight_tx_bias",
366 .data = &dev_weight_tx_bias,
367 .maxlen = sizeof(int),
369 .proc_handler = proc_do_dev_weight,
372 .procname = "netdev_max_backlog",
373 .data = &netdev_max_backlog,
374 .maxlen = sizeof(int),
376 .proc_handler = proc_dointvec
379 .procname = "netdev_rss_key",
380 .data = &netdev_rss_key,
381 .maxlen = sizeof(int),
383 .proc_handler = proc_do_rss_key,
385 #ifdef CONFIG_BPF_JIT
387 .procname = "bpf_jit_enable",
388 .data = &bpf_jit_enable,
389 .maxlen = sizeof(int),
391 .proc_handler = proc_dointvec_minmax_bpf_enable,
392 # ifdef CONFIG_BPF_JIT_ALWAYS_ON
393 .extra1 = SYSCTL_ONE,
394 .extra2 = SYSCTL_ONE,
396 .extra1 = SYSCTL_ZERO,
400 # ifdef CONFIG_HAVE_EBPF_JIT
402 .procname = "bpf_jit_harden",
403 .data = &bpf_jit_harden,
404 .maxlen = sizeof(int),
406 .proc_handler = proc_dointvec_minmax_bpf_restricted,
407 .extra1 = SYSCTL_ZERO,
411 .procname = "bpf_jit_kallsyms",
412 .data = &bpf_jit_kallsyms,
413 .maxlen = sizeof(int),
415 .proc_handler = proc_dointvec_minmax_bpf_restricted,
416 .extra1 = SYSCTL_ZERO,
417 .extra2 = SYSCTL_ONE,
421 .procname = "bpf_jit_limit",
422 .data = &bpf_jit_limit,
423 .maxlen = sizeof(long),
425 .proc_handler = proc_dolongvec_minmax_bpf_restricted,
431 .procname = "netdev_tstamp_prequeue",
432 .data = &netdev_tstamp_prequeue,
433 .maxlen = sizeof(int),
435 .proc_handler = proc_dointvec
438 .procname = "message_cost",
439 .data = &net_ratelimit_state.interval,
440 .maxlen = sizeof(int),
442 .proc_handler = proc_dointvec_jiffies,
445 .procname = "message_burst",
446 .data = &net_ratelimit_state.burst,
447 .maxlen = sizeof(int),
449 .proc_handler = proc_dointvec,
452 .procname = "optmem_max",
453 .data = &sysctl_optmem_max,
454 .maxlen = sizeof(int),
456 .proc_handler = proc_dointvec
459 .procname = "tstamp_allow_data",
460 .data = &sysctl_tstamp_allow_data,
461 .maxlen = sizeof(int),
463 .proc_handler = proc_dointvec_minmax,
464 .extra1 = SYSCTL_ZERO,
469 .procname = "rps_sock_flow_entries",
470 .maxlen = sizeof(int),
472 .proc_handler = rps_sock_flow_sysctl
475 #ifdef CONFIG_NET_FLOW_LIMIT
477 .procname = "flow_limit_cpu_bitmap",
479 .proc_handler = flow_limit_cpu_sysctl
482 .procname = "flow_limit_table_len",
483 .data = &netdev_flow_limit_table_len,
484 .maxlen = sizeof(int),
486 .proc_handler = flow_limit_table_len_sysctl
488 #endif /* CONFIG_NET_FLOW_LIMIT */
489 #ifdef CONFIG_NET_RX_BUSY_POLL
491 .procname = "busy_poll",
492 .data = &sysctl_net_busy_poll,
493 .maxlen = sizeof(unsigned int),
495 .proc_handler = proc_dointvec_minmax,
496 .extra1 = SYSCTL_ZERO,
499 .procname = "busy_read",
500 .data = &sysctl_net_busy_read,
501 .maxlen = sizeof(unsigned int),
503 .proc_handler = proc_dointvec_minmax,
504 .extra1 = SYSCTL_ZERO,
507 #ifdef CONFIG_NET_SCHED
509 .procname = "default_qdisc",
512 .proc_handler = set_default_qdisc
515 #endif /* CONFIG_NET */
517 .procname = "netdev_budget",
518 .data = &netdev_budget,
519 .maxlen = sizeof(int),
521 .proc_handler = proc_dointvec
524 .procname = "warnings",
525 .data = &net_msg_warn,
526 .maxlen = sizeof(int),
528 .proc_handler = proc_dointvec
531 .procname = "max_skb_frags",
532 .data = &sysctl_max_skb_frags,
533 .maxlen = sizeof(int),
535 .proc_handler = proc_dointvec_minmax,
536 .extra1 = SYSCTL_ONE,
537 .extra2 = &max_skb_frags,
540 .procname = "netdev_budget_usecs",
541 .data = &netdev_budget_usecs,
542 .maxlen = sizeof(unsigned int),
544 .proc_handler = proc_dointvec_minmax,
545 .extra1 = SYSCTL_ZERO,
548 .procname = "fb_tunnels_only_for_init_net",
549 .data = &sysctl_fb_tunnels_only_for_init_net,
550 .maxlen = sizeof(int),
552 .proc_handler = proc_dointvec_minmax,
553 .extra1 = SYSCTL_ZERO,
554 .extra2 = SYSCTL_ONE,
557 .procname = "devconf_inherit_init_net",
558 .data = &sysctl_devconf_inherit_init_net,
559 .maxlen = sizeof(int),
561 .proc_handler = proc_dointvec_minmax,
562 .extra1 = SYSCTL_ZERO,
566 .procname = "high_order_alloc_disable",
567 .data = &net_high_order_alloc_disable_key.key,
568 .maxlen = sizeof(net_high_order_alloc_disable_key),
570 .proc_handler = proc_do_static_key,
573 .procname = "gro_normal_batch",
574 .data = &gro_normal_batch,
575 .maxlen = sizeof(unsigned int),
577 .proc_handler = proc_dointvec_minmax,
578 .extra1 = SYSCTL_ONE,
583 static struct ctl_table netns_core_table[] = {
585 .procname = "somaxconn",
586 .data = &init_net.core.sysctl_somaxconn,
587 .maxlen = sizeof(int),
589 .extra1 = SYSCTL_ZERO,
590 .proc_handler = proc_dointvec_minmax
595 static __net_init int sysctl_core_net_init(struct net *net)
597 struct ctl_table *tbl;
599 tbl = netns_core_table;
600 if (!net_eq(net, &init_net)) {
601 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
605 tbl[0].data = &net->core.sysctl_somaxconn;
607 /* Don't export any sysctls to unprivileged users */
608 if (net->user_ns != &init_user_ns) {
609 tbl[0].procname = NULL;
613 net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
614 if (net->core.sysctl_hdr == NULL)
620 if (tbl != netns_core_table)
626 static __net_exit void sysctl_core_net_exit(struct net *net)
628 struct ctl_table *tbl;
630 tbl = net->core.sysctl_hdr->ctl_table_arg;
631 unregister_net_sysctl_table(net->core.sysctl_hdr);
632 BUG_ON(tbl == netns_core_table);
636 static __net_initdata struct pernet_operations sysctl_core_ops = {
637 .init = sysctl_core_net_init,
638 .exit = sysctl_core_net_exit,
641 static __init int sysctl_core_init(void)
643 register_net_sysctl(&init_net, "net/core", net_core_table);
644 return register_pernet_subsys(&sysctl_core_ops);
647 fs_initcall(sysctl_core_init);