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>
22 #include <net/net_ratelimit.h>
23 #include <net/busy_poll.h>
24 #include <net/pkt_sched.h>
28 static int int_3600 = 3600;
29 static int min_sndbuf = SOCK_MIN_SNDBUF;
30 static int min_rcvbuf = SOCK_MIN_RCVBUF;
31 static int max_skb_frags = MAX_SKB_FRAGS;
33 static int net_msg_warn; /* Unused, but still a sysctl */
35 int sysctl_fb_tunnels_only_for_init_net __read_mostly = 0;
36 EXPORT_SYMBOL(sysctl_fb_tunnels_only_for_init_net);
38 /* 0 - Keep current behavior:
39 * IPv4: inherit all current settings from init_net
40 * IPv6: reset all settings to default
41 * 1 - Both inherit all current settings from init_net
42 * 2 - Both reset all settings to default
43 * 3 - Both inherit all settings from current netns
45 int sysctl_devconf_inherit_init_net __read_mostly;
46 EXPORT_SYMBOL(sysctl_devconf_inherit_init_net);
49 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
50 void *buffer, size_t *lenp, loff_t *ppos)
52 unsigned int orig_size, size;
54 struct ctl_table tmp = {
56 .maxlen = sizeof(size),
59 struct rps_sock_flow_table *orig_sock_table, *sock_table;
60 static DEFINE_MUTEX(sock_flow_mutex);
62 mutex_lock(&sock_flow_mutex);
64 orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
65 lockdep_is_held(&sock_flow_mutex));
66 size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
68 ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
73 /* Enforce limit to prevent overflow */
74 mutex_unlock(&sock_flow_mutex);
77 size = roundup_pow_of_two(size);
78 if (size != orig_size) {
80 vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
82 mutex_unlock(&sock_flow_mutex);
85 rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
86 sock_table->mask = size - 1;
88 sock_table = orig_sock_table;
90 for (i = 0; i < size; i++)
91 sock_table->ents[i] = RPS_NO_CPU;
95 if (sock_table != orig_sock_table) {
96 rcu_assign_pointer(rps_sock_flow_table, sock_table);
98 static_branch_inc(&rps_needed);
99 static_branch_inc(&rfs_needed);
101 if (orig_sock_table) {
102 static_branch_dec(&rps_needed);
103 static_branch_dec(&rfs_needed);
104 kvfree_rcu(orig_sock_table);
109 mutex_unlock(&sock_flow_mutex);
113 #endif /* CONFIG_RPS */
115 #ifdef CONFIG_NET_FLOW_LIMIT
116 static DEFINE_MUTEX(flow_limit_update_mutex);
118 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
119 void *buffer, size_t *lenp, loff_t *ppos)
121 struct sd_flow_limit *cur;
122 struct softnet_data *sd;
126 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
130 ret = cpumask_parse(buffer, 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);
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)
236 static DEFINE_MUTEX(dev_weight_mutex);
239 mutex_lock(&dev_weight_mutex);
240 ret = proc_dointvec(table, write, buffer, lenp, ppos);
242 weight = READ_ONCE(weight_p);
243 WRITE_ONCE(dev_rx_weight, weight * dev_weight_rx_bias);
244 WRITE_ONCE(dev_tx_weight, weight * dev_weight_tx_bias);
246 mutex_unlock(&dev_weight_mutex);
251 static int proc_do_rss_key(struct ctl_table *table, int write,
252 void *buffer, size_t *lenp, loff_t *ppos)
254 struct ctl_table fake_table;
255 char buf[NETDEV_RSS_KEY_LEN * 3];
257 snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
258 fake_table.data = buf;
259 fake_table.maxlen = sizeof(buf);
260 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
263 #ifdef CONFIG_BPF_JIT
264 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write,
265 void *buffer, size_t *lenp,
268 int ret, jit_enable = *(int *)table->data;
269 int min = *(int *)table->extra1;
270 int max = *(int *)table->extra2;
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(current_cred()))) {
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");
289 if (write && ret && min == max)
290 pr_info_once("CONFIG_BPF_JIT_ALWAYS_ON is enabled, bpf_jit_enable is permanently set to 1.\n");
295 # ifdef CONFIG_HAVE_EBPF_JIT
297 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write,
298 void *buffer, size_t *lenp, loff_t *ppos)
300 if (!capable(CAP_SYS_ADMIN))
303 return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
305 # endif /* CONFIG_HAVE_EBPF_JIT */
308 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
309 void *buffer, size_t *lenp, loff_t *ppos)
311 if (!capable(CAP_SYS_ADMIN))
314 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
318 static struct ctl_table net_core_table[] = {
320 .procname = "wmem_max",
321 .data = &sysctl_wmem_max,
322 .maxlen = sizeof(int),
324 .proc_handler = proc_dointvec_minmax,
325 .extra1 = &min_sndbuf,
328 .procname = "rmem_max",
329 .data = &sysctl_rmem_max,
330 .maxlen = sizeof(int),
332 .proc_handler = proc_dointvec_minmax,
333 .extra1 = &min_rcvbuf,
336 .procname = "wmem_default",
337 .data = &sysctl_wmem_default,
338 .maxlen = sizeof(int),
340 .proc_handler = proc_dointvec_minmax,
341 .extra1 = &min_sndbuf,
344 .procname = "rmem_default",
345 .data = &sysctl_rmem_default,
346 .maxlen = sizeof(int),
348 .proc_handler = proc_dointvec_minmax,
349 .extra1 = &min_rcvbuf,
352 .procname = "dev_weight",
354 .maxlen = sizeof(int),
356 .proc_handler = proc_do_dev_weight,
359 .procname = "dev_weight_rx_bias",
360 .data = &dev_weight_rx_bias,
361 .maxlen = sizeof(int),
363 .proc_handler = proc_do_dev_weight,
366 .procname = "dev_weight_tx_bias",
367 .data = &dev_weight_tx_bias,
368 .maxlen = sizeof(int),
370 .proc_handler = proc_do_dev_weight,
373 .procname = "netdev_max_backlog",
374 .data = &netdev_max_backlog,
375 .maxlen = sizeof(int),
377 .proc_handler = proc_dointvec
380 .procname = "netdev_rss_key",
381 .data = &netdev_rss_key,
382 .maxlen = sizeof(int),
384 .proc_handler = proc_do_rss_key,
386 #ifdef CONFIG_BPF_JIT
388 .procname = "bpf_jit_enable",
389 .data = &bpf_jit_enable,
390 .maxlen = sizeof(int),
392 .proc_handler = proc_dointvec_minmax_bpf_enable,
393 # ifdef CONFIG_BPF_JIT_ALWAYS_ON
394 .extra1 = SYSCTL_ONE,
395 .extra2 = SYSCTL_ONE,
397 .extra1 = SYSCTL_ZERO,
398 .extra2 = SYSCTL_TWO,
401 # ifdef CONFIG_HAVE_EBPF_JIT
403 .procname = "bpf_jit_harden",
404 .data = &bpf_jit_harden,
405 .maxlen = sizeof(int),
407 .proc_handler = proc_dointvec_minmax_bpf_restricted,
408 .extra1 = SYSCTL_ZERO,
409 .extra2 = SYSCTL_TWO,
412 .procname = "bpf_jit_kallsyms",
413 .data = &bpf_jit_kallsyms,
414 .maxlen = sizeof(int),
416 .proc_handler = proc_dointvec_minmax_bpf_restricted,
417 .extra1 = SYSCTL_ZERO,
418 .extra2 = SYSCTL_ONE,
422 .procname = "bpf_jit_limit",
423 .data = &bpf_jit_limit,
424 .maxlen = sizeof(long),
426 .proc_handler = proc_dolongvec_minmax_bpf_restricted,
427 .extra1 = SYSCTL_LONG_ONE,
428 .extra2 = &bpf_jit_limit_max,
432 .procname = "netdev_tstamp_prequeue",
433 .data = &netdev_tstamp_prequeue,
434 .maxlen = sizeof(int),
436 .proc_handler = proc_dointvec
439 .procname = "message_cost",
440 .data = &net_ratelimit_state.interval,
441 .maxlen = sizeof(int),
443 .proc_handler = proc_dointvec_jiffies,
446 .procname = "message_burst",
447 .data = &net_ratelimit_state.burst,
448 .maxlen = sizeof(int),
450 .proc_handler = proc_dointvec,
453 .procname = "optmem_max",
454 .data = &sysctl_optmem_max,
455 .maxlen = sizeof(int),
457 .proc_handler = proc_dointvec
460 .procname = "tstamp_allow_data",
461 .data = &sysctl_tstamp_allow_data,
462 .maxlen = sizeof(int),
464 .proc_handler = proc_dointvec_minmax,
465 .extra1 = SYSCTL_ZERO,
470 .procname = "rps_sock_flow_entries",
471 .maxlen = sizeof(int),
473 .proc_handler = rps_sock_flow_sysctl
476 #ifdef CONFIG_NET_FLOW_LIMIT
478 .procname = "flow_limit_cpu_bitmap",
480 .proc_handler = flow_limit_cpu_sysctl
483 .procname = "flow_limit_table_len",
484 .data = &netdev_flow_limit_table_len,
485 .maxlen = sizeof(int),
487 .proc_handler = flow_limit_table_len_sysctl
489 #endif /* CONFIG_NET_FLOW_LIMIT */
490 #ifdef CONFIG_NET_RX_BUSY_POLL
492 .procname = "busy_poll",
493 .data = &sysctl_net_busy_poll,
494 .maxlen = sizeof(unsigned int),
496 .proc_handler = proc_dointvec_minmax,
497 .extra1 = SYSCTL_ZERO,
500 .procname = "busy_read",
501 .data = &sysctl_net_busy_read,
502 .maxlen = sizeof(unsigned int),
504 .proc_handler = proc_dointvec_minmax,
505 .extra1 = SYSCTL_ZERO,
508 #ifdef CONFIG_NET_SCHED
510 .procname = "default_qdisc",
513 .proc_handler = set_default_qdisc
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_TWO,
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,
563 .extra2 = SYSCTL_THREE,
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,
581 .procname = "netdev_unregister_timeout_secs",
582 .data = &netdev_unregister_timeout_secs,
583 .maxlen = sizeof(unsigned int),
585 .proc_handler = proc_dointvec_minmax,
586 .extra1 = SYSCTL_ONE,
590 .procname = "skb_defer_max",
591 .data = &sysctl_skb_defer_max,
592 .maxlen = sizeof(unsigned int),
594 .proc_handler = proc_dointvec_minmax,
595 .extra1 = SYSCTL_ZERO,
600 static struct ctl_table netns_core_table[] = {
602 .procname = "somaxconn",
603 .data = &init_net.core.sysctl_somaxconn,
604 .maxlen = sizeof(int),
606 .extra1 = SYSCTL_ZERO,
607 .proc_handler = proc_dointvec_minmax
610 .procname = "txrehash",
611 .data = &init_net.core.sysctl_txrehash,
612 .maxlen = sizeof(u8),
614 .extra1 = SYSCTL_ZERO,
615 .extra2 = SYSCTL_ONE,
616 .proc_handler = proc_dou8vec_minmax,
621 static int __init fb_tunnels_only_for_init_net_sysctl_setup(char *str)
623 /* fallback tunnels for initns only */
624 if (!strncmp(str, "initns", 6))
625 sysctl_fb_tunnels_only_for_init_net = 1;
626 /* no fallback tunnels anywhere */
627 else if (!strncmp(str, "none", 4))
628 sysctl_fb_tunnels_only_for_init_net = 2;
632 __setup("fb_tunnels=", fb_tunnels_only_for_init_net_sysctl_setup);
634 static __net_init int sysctl_core_net_init(struct net *net)
636 struct ctl_table *tbl, *tmp;
638 tbl = netns_core_table;
639 if (!net_eq(net, &init_net)) {
640 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
644 for (tmp = tbl; tmp->procname; tmp++)
645 tmp->data += (char *)net - (char *)&init_net;
647 /* Don't export any sysctls to unprivileged users */
648 if (net->user_ns != &init_user_ns) {
649 tbl[0].procname = NULL;
653 net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
654 if (net->core.sysctl_hdr == NULL)
660 if (tbl != netns_core_table)
666 static __net_exit void sysctl_core_net_exit(struct net *net)
668 struct ctl_table *tbl;
670 tbl = net->core.sysctl_hdr->ctl_table_arg;
671 unregister_net_sysctl_table(net->core.sysctl_hdr);
672 BUG_ON(tbl == netns_core_table);
676 static __net_initdata struct pernet_operations sysctl_core_ops = {
677 .init = sysctl_core_net_init,
678 .exit = sysctl_core_net_exit,
681 static __init int sysctl_core_init(void)
683 register_net_sysctl(&init_net, "net/core", net_core_table);
684 return register_pernet_subsys(&sysctl_core_ops);
687 fs_initcall(sysctl_core_init);