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 *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 *buffer, size_t *lenp, loff_t *ppos)
120 struct sd_flow_limit *cur;
121 struct softnet_data *sd;
125 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
129 ret = cpumask_parse_user(buffer, *lenp, mask);
133 mutex_lock(&flow_limit_update_mutex);
134 len = sizeof(*cur) + netdev_flow_limit_table_len;
135 for_each_possible_cpu(i) {
136 sd = &per_cpu(softnet_data, i);
137 cur = rcu_dereference_protected(sd->flow_limit,
138 lockdep_is_held(&flow_limit_update_mutex));
139 if (cur && !cpumask_test_cpu(i, mask)) {
140 RCU_INIT_POINTER(sd->flow_limit, NULL);
143 } else if (!cur && cpumask_test_cpu(i, mask)) {
144 cur = kzalloc_node(len, GFP_KERNEL,
147 /* not unwinding previous changes */
151 cur->num_buckets = netdev_flow_limit_table_len;
152 rcu_assign_pointer(sd->flow_limit, cur);
156 mutex_unlock(&flow_limit_update_mutex);
160 if (*ppos || !*lenp) {
167 for_each_possible_cpu(i) {
168 sd = &per_cpu(softnet_data, i);
169 if (rcu_dereference(sd->flow_limit))
170 cpumask_set_cpu(i, mask);
174 len = min(sizeof(kbuf) - 1, *lenp);
175 len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
182 memcpy(buffer, kbuf, len);
188 free_cpumask_var(mask);
192 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
193 void *buffer, size_t *lenp, loff_t *ppos)
195 unsigned int old, *ptr;
198 mutex_lock(&flow_limit_update_mutex);
202 ret = proc_dointvec(table, write, buffer, lenp, ppos);
203 if (!ret && write && !is_power_of_2(*ptr)) {
208 mutex_unlock(&flow_limit_update_mutex);
211 #endif /* CONFIG_NET_FLOW_LIMIT */
213 #ifdef CONFIG_NET_SCHED
214 static int set_default_qdisc(struct ctl_table *table, int write,
215 void *buffer, size_t *lenp, loff_t *ppos)
218 struct ctl_table tbl = {
224 qdisc_get_default(id, IFNAMSIZ);
226 ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
227 if (write && ret == 0)
228 ret = qdisc_set_default(id);
233 static int proc_do_dev_weight(struct ctl_table *table, int write,
234 void *buffer, size_t *lenp, loff_t *ppos)
238 ret = proc_dointvec(table, write, buffer, lenp, ppos);
242 dev_rx_weight = weight_p * dev_weight_rx_bias;
243 dev_tx_weight = weight_p * dev_weight_tx_bias;
248 static int proc_do_rss_key(struct ctl_table *table, int write,
249 void *buffer, size_t *lenp, loff_t *ppos)
251 struct ctl_table fake_table;
252 char buf[NETDEV_RSS_KEY_LEN * 3];
254 snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
255 fake_table.data = buf;
256 fake_table.maxlen = sizeof(buf);
257 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
260 #ifdef CONFIG_BPF_JIT
261 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write,
262 void *buffer, size_t *lenp,
265 int ret, jit_enable = *(int *)table->data;
266 struct ctl_table tmp = *table;
268 if (write && !capable(CAP_SYS_ADMIN))
271 tmp.data = &jit_enable;
272 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
274 if (jit_enable < 2 ||
275 (jit_enable == 2 && bpf_dump_raw_ok())) {
276 *(int *)table->data = jit_enable;
278 pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
286 # ifdef CONFIG_HAVE_EBPF_JIT
288 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write,
289 void *buffer, size_t *lenp, loff_t *ppos)
291 if (!capable(CAP_SYS_ADMIN))
294 return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
296 # endif /* CONFIG_HAVE_EBPF_JIT */
299 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
300 void *buffer, size_t *lenp, loff_t *ppos)
302 if (!capable(CAP_SYS_ADMIN))
305 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
309 static struct ctl_table net_core_table[] = {
312 .procname = "wmem_max",
313 .data = &sysctl_wmem_max,
314 .maxlen = sizeof(int),
316 .proc_handler = proc_dointvec_minmax,
317 .extra1 = &min_sndbuf,
320 .procname = "rmem_max",
321 .data = &sysctl_rmem_max,
322 .maxlen = sizeof(int),
324 .proc_handler = proc_dointvec_minmax,
325 .extra1 = &min_rcvbuf,
328 .procname = "wmem_default",
329 .data = &sysctl_wmem_default,
330 .maxlen = sizeof(int),
332 .proc_handler = proc_dointvec_minmax,
333 .extra1 = &min_sndbuf,
336 .procname = "rmem_default",
337 .data = &sysctl_rmem_default,
338 .maxlen = sizeof(int),
340 .proc_handler = proc_dointvec_minmax,
341 .extra1 = &min_rcvbuf,
344 .procname = "dev_weight",
346 .maxlen = sizeof(int),
348 .proc_handler = proc_do_dev_weight,
351 .procname = "dev_weight_rx_bias",
352 .data = &dev_weight_rx_bias,
353 .maxlen = sizeof(int),
355 .proc_handler = proc_do_dev_weight,
358 .procname = "dev_weight_tx_bias",
359 .data = &dev_weight_tx_bias,
360 .maxlen = sizeof(int),
362 .proc_handler = proc_do_dev_weight,
365 .procname = "netdev_max_backlog",
366 .data = &netdev_max_backlog,
367 .maxlen = sizeof(int),
369 .proc_handler = proc_dointvec
372 .procname = "netdev_rss_key",
373 .data = &netdev_rss_key,
374 .maxlen = sizeof(int),
376 .proc_handler = proc_do_rss_key,
378 #ifdef CONFIG_BPF_JIT
380 .procname = "bpf_jit_enable",
381 .data = &bpf_jit_enable,
382 .maxlen = sizeof(int),
384 .proc_handler = proc_dointvec_minmax_bpf_enable,
385 # ifdef CONFIG_BPF_JIT_ALWAYS_ON
386 .extra1 = SYSCTL_ONE,
387 .extra2 = SYSCTL_ONE,
389 .extra1 = SYSCTL_ZERO,
393 # ifdef CONFIG_HAVE_EBPF_JIT
395 .procname = "bpf_jit_harden",
396 .data = &bpf_jit_harden,
397 .maxlen = sizeof(int),
399 .proc_handler = proc_dointvec_minmax_bpf_restricted,
400 .extra1 = SYSCTL_ZERO,
404 .procname = "bpf_jit_kallsyms",
405 .data = &bpf_jit_kallsyms,
406 .maxlen = sizeof(int),
408 .proc_handler = proc_dointvec_minmax_bpf_restricted,
409 .extra1 = SYSCTL_ZERO,
410 .extra2 = SYSCTL_ONE,
414 .procname = "bpf_jit_limit",
415 .data = &bpf_jit_limit,
416 .maxlen = sizeof(long),
418 .proc_handler = proc_dolongvec_minmax_bpf_restricted,
424 .procname = "netdev_tstamp_prequeue",
425 .data = &netdev_tstamp_prequeue,
426 .maxlen = sizeof(int),
428 .proc_handler = proc_dointvec
431 .procname = "message_cost",
432 .data = &net_ratelimit_state.interval,
433 .maxlen = sizeof(int),
435 .proc_handler = proc_dointvec_jiffies,
438 .procname = "message_burst",
439 .data = &net_ratelimit_state.burst,
440 .maxlen = sizeof(int),
442 .proc_handler = proc_dointvec,
445 .procname = "optmem_max",
446 .data = &sysctl_optmem_max,
447 .maxlen = sizeof(int),
449 .proc_handler = proc_dointvec
452 .procname = "tstamp_allow_data",
453 .data = &sysctl_tstamp_allow_data,
454 .maxlen = sizeof(int),
456 .proc_handler = proc_dointvec_minmax,
457 .extra1 = SYSCTL_ZERO,
462 .procname = "rps_sock_flow_entries",
463 .maxlen = sizeof(int),
465 .proc_handler = rps_sock_flow_sysctl
468 #ifdef CONFIG_NET_FLOW_LIMIT
470 .procname = "flow_limit_cpu_bitmap",
472 .proc_handler = flow_limit_cpu_sysctl
475 .procname = "flow_limit_table_len",
476 .data = &netdev_flow_limit_table_len,
477 .maxlen = sizeof(int),
479 .proc_handler = flow_limit_table_len_sysctl
481 #endif /* CONFIG_NET_FLOW_LIMIT */
482 #ifdef CONFIG_NET_RX_BUSY_POLL
484 .procname = "busy_poll",
485 .data = &sysctl_net_busy_poll,
486 .maxlen = sizeof(unsigned int),
488 .proc_handler = proc_dointvec_minmax,
489 .extra1 = SYSCTL_ZERO,
492 .procname = "busy_read",
493 .data = &sysctl_net_busy_read,
494 .maxlen = sizeof(unsigned int),
496 .proc_handler = proc_dointvec_minmax,
497 .extra1 = SYSCTL_ZERO,
500 #ifdef CONFIG_NET_SCHED
502 .procname = "default_qdisc",
505 .proc_handler = set_default_qdisc
508 #endif /* CONFIG_NET */
510 .procname = "netdev_budget",
511 .data = &netdev_budget,
512 .maxlen = sizeof(int),
514 .proc_handler = proc_dointvec
517 .procname = "warnings",
518 .data = &net_msg_warn,
519 .maxlen = sizeof(int),
521 .proc_handler = proc_dointvec
524 .procname = "max_skb_frags",
525 .data = &sysctl_max_skb_frags,
526 .maxlen = sizeof(int),
528 .proc_handler = proc_dointvec_minmax,
529 .extra1 = SYSCTL_ONE,
530 .extra2 = &max_skb_frags,
533 .procname = "netdev_budget_usecs",
534 .data = &netdev_budget_usecs,
535 .maxlen = sizeof(unsigned int),
537 .proc_handler = proc_dointvec_minmax,
538 .extra1 = SYSCTL_ZERO,
541 .procname = "fb_tunnels_only_for_init_net",
542 .data = &sysctl_fb_tunnels_only_for_init_net,
543 .maxlen = sizeof(int),
545 .proc_handler = proc_dointvec_minmax,
546 .extra1 = SYSCTL_ZERO,
547 .extra2 = SYSCTL_ONE,
550 .procname = "devconf_inherit_init_net",
551 .data = &sysctl_devconf_inherit_init_net,
552 .maxlen = sizeof(int),
554 .proc_handler = proc_dointvec_minmax,
555 .extra1 = SYSCTL_ZERO,
559 .procname = "high_order_alloc_disable",
560 .data = &net_high_order_alloc_disable_key.key,
561 .maxlen = sizeof(net_high_order_alloc_disable_key),
563 .proc_handler = proc_do_static_key,
566 .procname = "gro_normal_batch",
567 .data = &gro_normal_batch,
568 .maxlen = sizeof(unsigned int),
570 .proc_handler = proc_dointvec_minmax,
571 .extra1 = SYSCTL_ONE,
576 static struct ctl_table netns_core_table[] = {
578 .procname = "somaxconn",
579 .data = &init_net.core.sysctl_somaxconn,
580 .maxlen = sizeof(int),
582 .extra1 = SYSCTL_ZERO,
583 .proc_handler = proc_dointvec_minmax
588 static __net_init int sysctl_core_net_init(struct net *net)
590 struct ctl_table *tbl;
592 tbl = netns_core_table;
593 if (!net_eq(net, &init_net)) {
594 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
598 tbl[0].data = &net->core.sysctl_somaxconn;
600 /* Don't export any sysctls to unprivileged users */
601 if (net->user_ns != &init_user_ns) {
602 tbl[0].procname = NULL;
606 net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
607 if (net->core.sysctl_hdr == NULL)
613 if (tbl != netns_core_table)
619 static __net_exit void sysctl_core_net_exit(struct net *net)
621 struct ctl_table *tbl;
623 tbl = net->core.sysctl_hdr->ctl_table_arg;
624 unregister_net_sysctl_table(net->core.sysctl_hdr);
625 BUG_ON(tbl == netns_core_table);
629 static __net_initdata struct pernet_operations sysctl_core_ops = {
630 .init = sysctl_core_net_init,
631 .exit = sysctl_core_net_exit,
634 static __init int sysctl_core_init(void)
636 register_net_sysctl(&init_net, "net/core", net_core_table);
637 return register_pernet_subsys(&sysctl_core_ops);
640 fs_initcall(sysctl_core_init);