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>
18 #include <linux/kmemleak.h>
22 #include <net/net_ratelimit.h>
23 #include <net/busy_poll.h>
24 #include <net/pkt_sched.h>
28 static int two __maybe_unused = 2;
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;
32 static long long_one __maybe_unused = 1;
33 static long long_max __maybe_unused = LONG_MAX;
35 static int net_msg_warn; /* Unused, but still a sysctl */
38 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
39 void __user *buffer, size_t *lenp, loff_t *ppos)
41 unsigned int orig_size, size;
43 struct ctl_table tmp = {
45 .maxlen = sizeof(size),
48 struct rps_sock_flow_table *orig_sock_table, *sock_table;
49 static DEFINE_MUTEX(sock_flow_mutex);
51 mutex_lock(&sock_flow_mutex);
53 orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
54 lockdep_is_held(&sock_flow_mutex));
55 size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
57 ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
62 /* Enforce limit to prevent overflow */
63 mutex_unlock(&sock_flow_mutex);
66 size = roundup_pow_of_two(size);
67 if (size != orig_size) {
69 vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
71 mutex_unlock(&sock_flow_mutex);
74 rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
75 sock_table->mask = size - 1;
77 sock_table = orig_sock_table;
79 for (i = 0; i < size; i++)
80 sock_table->ents[i] = RPS_NO_CPU;
84 if (sock_table != orig_sock_table) {
85 rcu_assign_pointer(rps_sock_flow_table, sock_table);
87 static_key_slow_inc(&rps_needed);
88 static_key_slow_inc(&rfs_needed);
90 if (orig_sock_table) {
91 static_key_slow_dec(&rps_needed);
92 static_key_slow_dec(&rfs_needed);
94 vfree(orig_sock_table);
99 mutex_unlock(&sock_flow_mutex);
103 #endif /* CONFIG_RPS */
105 #ifdef CONFIG_NET_FLOW_LIMIT
106 static DEFINE_MUTEX(flow_limit_update_mutex);
108 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
109 void __user *buffer, size_t *lenp,
112 struct sd_flow_limit *cur;
113 struct softnet_data *sd;
117 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
121 ret = cpumask_parse_user(buffer, *lenp, mask);
125 mutex_lock(&flow_limit_update_mutex);
126 len = sizeof(*cur) + netdev_flow_limit_table_len;
127 for_each_possible_cpu(i) {
128 sd = &per_cpu(softnet_data, i);
129 cur = rcu_dereference_protected(sd->flow_limit,
130 lockdep_is_held(&flow_limit_update_mutex));
131 if (cur && !cpumask_test_cpu(i, mask)) {
132 RCU_INIT_POINTER(sd->flow_limit, NULL);
135 } else if (!cur && cpumask_test_cpu(i, mask)) {
136 cur = kzalloc_node(len, GFP_KERNEL,
139 /* not unwinding previous changes */
143 cur->num_buckets = netdev_flow_limit_table_len;
144 rcu_assign_pointer(sd->flow_limit, cur);
148 mutex_unlock(&flow_limit_update_mutex);
152 if (*ppos || !*lenp) {
159 for_each_possible_cpu(i) {
160 sd = &per_cpu(softnet_data, i);
161 if (rcu_dereference(sd->flow_limit))
162 cpumask_set_cpu(i, mask);
166 len = min(sizeof(kbuf) - 1, *lenp);
167 len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
174 if (copy_to_user(buffer, kbuf, len)) {
183 free_cpumask_var(mask);
187 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
188 void __user *buffer, size_t *lenp,
191 unsigned int old, *ptr;
194 mutex_lock(&flow_limit_update_mutex);
198 ret = proc_dointvec(table, write, buffer, lenp, ppos);
199 if (!ret && write && !is_power_of_2(*ptr)) {
204 mutex_unlock(&flow_limit_update_mutex);
207 #endif /* CONFIG_NET_FLOW_LIMIT */
209 #ifdef CONFIG_NET_SCHED
210 static int set_default_qdisc(struct ctl_table *table, int write,
211 void __user *buffer, size_t *lenp, loff_t *ppos)
214 struct ctl_table tbl = {
220 qdisc_get_default(id, IFNAMSIZ);
222 ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
223 if (write && ret == 0)
224 ret = qdisc_set_default(id);
229 static int proc_do_dev_weight(struct ctl_table *table, int write,
230 void __user *buffer, size_t *lenp, loff_t *ppos)
234 ret = proc_dointvec(table, write, buffer, lenp, ppos);
238 dev_rx_weight = weight_p * dev_weight_rx_bias;
239 dev_tx_weight = weight_p * dev_weight_tx_bias;
244 static int proc_do_rss_key(struct ctl_table *table, int write,
245 void __user *buffer, size_t *lenp, loff_t *ppos)
247 struct ctl_table fake_table;
248 char buf[NETDEV_RSS_KEY_LEN * 3];
250 snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
251 fake_table.data = buf;
252 fake_table.maxlen = sizeof(buf);
253 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
256 #ifdef CONFIG_BPF_JIT
257 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write,
258 void __user *buffer, size_t *lenp,
261 int ret, jit_enable = *(int *)table->data;
262 struct ctl_table tmp = *table;
264 if (write && !capable(CAP_SYS_ADMIN))
267 tmp.data = &jit_enable;
268 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
270 *(int *)table->data = jit_enable;
272 pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
277 # ifdef CONFIG_HAVE_EBPF_JIT
279 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write,
280 void __user *buffer, size_t *lenp,
283 if (!capable(CAP_SYS_ADMIN))
286 return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
288 # endif /* CONFIG_HAVE_EBPF_JIT */
291 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
292 void __user *buffer, size_t *lenp,
295 if (!capable(CAP_SYS_ADMIN))
298 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
302 static struct ctl_table net_core_table[] = {
305 .procname = "wmem_max",
306 .data = &sysctl_wmem_max,
307 .maxlen = sizeof(int),
309 .proc_handler = proc_dointvec_minmax,
310 .extra1 = &min_sndbuf,
313 .procname = "rmem_max",
314 .data = &sysctl_rmem_max,
315 .maxlen = sizeof(int),
317 .proc_handler = proc_dointvec_minmax,
318 .extra1 = &min_rcvbuf,
321 .procname = "wmem_default",
322 .data = &sysctl_wmem_default,
323 .maxlen = sizeof(int),
325 .proc_handler = proc_dointvec_minmax,
326 .extra1 = &min_sndbuf,
329 .procname = "rmem_default",
330 .data = &sysctl_rmem_default,
331 .maxlen = sizeof(int),
333 .proc_handler = proc_dointvec_minmax,
334 .extra1 = &min_rcvbuf,
337 .procname = "dev_weight",
339 .maxlen = sizeof(int),
341 .proc_handler = proc_do_dev_weight,
344 .procname = "dev_weight_rx_bias",
345 .data = &dev_weight_rx_bias,
346 .maxlen = sizeof(int),
348 .proc_handler = proc_do_dev_weight,
351 .procname = "dev_weight_tx_bias",
352 .data = &dev_weight_tx_bias,
353 .maxlen = sizeof(int),
355 .proc_handler = proc_do_dev_weight,
358 .procname = "netdev_max_backlog",
359 .data = &netdev_max_backlog,
360 .maxlen = sizeof(int),
362 .proc_handler = proc_dointvec
365 .procname = "netdev_rss_key",
366 .data = &netdev_rss_key,
367 .maxlen = sizeof(int),
369 .proc_handler = proc_do_rss_key,
371 #ifdef CONFIG_BPF_JIT
373 .procname = "bpf_jit_enable",
374 .data = &bpf_jit_enable,
375 .maxlen = sizeof(int),
377 .proc_handler = proc_dointvec_minmax_bpf_enable,
378 # ifdef CONFIG_BPF_JIT_ALWAYS_ON
386 # ifdef CONFIG_HAVE_EBPF_JIT
388 .procname = "bpf_jit_harden",
389 .data = &bpf_jit_harden,
390 .maxlen = sizeof(int),
392 .proc_handler = proc_dointvec_minmax_bpf_restricted,
397 .procname = "bpf_jit_kallsyms",
398 .data = &bpf_jit_kallsyms,
399 .maxlen = sizeof(int),
401 .proc_handler = proc_dointvec_minmax_bpf_restricted,
407 .procname = "bpf_jit_limit",
408 .data = &bpf_jit_limit,
409 .maxlen = sizeof(long),
411 .proc_handler = proc_dolongvec_minmax_bpf_restricted,
413 .extra2 = &bpf_jit_limit_max,
417 .procname = "netdev_tstamp_prequeue",
418 .data = &netdev_tstamp_prequeue,
419 .maxlen = sizeof(int),
421 .proc_handler = proc_dointvec
424 .procname = "message_cost",
425 .data = &net_ratelimit_state.interval,
426 .maxlen = sizeof(int),
428 .proc_handler = proc_dointvec_jiffies,
431 .procname = "message_burst",
432 .data = &net_ratelimit_state.burst,
433 .maxlen = sizeof(int),
435 .proc_handler = proc_dointvec,
438 .procname = "optmem_max",
439 .data = &sysctl_optmem_max,
440 .maxlen = sizeof(int),
442 .proc_handler = proc_dointvec
445 .procname = "tstamp_allow_data",
446 .data = &sysctl_tstamp_allow_data,
447 .maxlen = sizeof(int),
449 .proc_handler = proc_dointvec_minmax,
455 .procname = "rps_sock_flow_entries",
456 .maxlen = sizeof(int),
458 .proc_handler = rps_sock_flow_sysctl
461 #ifdef CONFIG_NET_FLOW_LIMIT
463 .procname = "flow_limit_cpu_bitmap",
465 .proc_handler = flow_limit_cpu_sysctl
468 .procname = "flow_limit_table_len",
469 .data = &netdev_flow_limit_table_len,
470 .maxlen = sizeof(int),
472 .proc_handler = flow_limit_table_len_sysctl
474 #endif /* CONFIG_NET_FLOW_LIMIT */
475 #ifdef CONFIG_NET_RX_BUSY_POLL
477 .procname = "busy_poll",
478 .data = &sysctl_net_busy_poll,
479 .maxlen = sizeof(unsigned int),
481 .proc_handler = proc_dointvec_minmax,
485 .procname = "busy_read",
486 .data = &sysctl_net_busy_read,
487 .maxlen = sizeof(unsigned int),
489 .proc_handler = proc_dointvec_minmax,
493 #ifdef CONFIG_NET_SCHED
495 .procname = "default_qdisc",
498 .proc_handler = set_default_qdisc
501 #endif /* CONFIG_NET */
503 .procname = "netdev_budget",
504 .data = &netdev_budget,
505 .maxlen = sizeof(int),
507 .proc_handler = proc_dointvec
510 .procname = "warnings",
511 .data = &net_msg_warn,
512 .maxlen = sizeof(int),
514 .proc_handler = proc_dointvec
517 .procname = "max_skb_frags",
518 .data = &sysctl_max_skb_frags,
519 .maxlen = sizeof(int),
521 .proc_handler = proc_dointvec_minmax,
523 .extra2 = &max_skb_frags,
526 .procname = "netdev_budget_usecs",
527 .data = &netdev_budget_usecs,
528 .maxlen = sizeof(unsigned int),
530 .proc_handler = proc_dointvec_minmax,
536 static struct ctl_table netns_core_table[] = {
538 .procname = "somaxconn",
539 .data = &init_net.core.sysctl_somaxconn,
540 .maxlen = sizeof(int),
543 .proc_handler = proc_dointvec_minmax
548 static __net_init int sysctl_core_net_init(struct net *net)
550 struct ctl_table *tbl;
552 tbl = netns_core_table;
553 if (!net_eq(net, &init_net)) {
554 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
558 tbl[0].data = &net->core.sysctl_somaxconn;
560 /* Don't export any sysctls to unprivileged users */
561 if (net->user_ns != &init_user_ns) {
562 tbl[0].procname = NULL;
566 net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
567 if (net->core.sysctl_hdr == NULL)
573 if (tbl != netns_core_table)
579 static __net_exit void sysctl_core_net_exit(struct net *net)
581 struct ctl_table *tbl;
583 tbl = net->core.sysctl_hdr->ctl_table_arg;
584 unregister_net_sysctl_table(net->core.sysctl_hdr);
585 BUG_ON(tbl == netns_core_table);
589 static __net_initdata struct pernet_operations sysctl_core_ops = {
590 .init = sysctl_core_net_init,
591 .exit = sysctl_core_net_exit,
594 static __init int sysctl_core_init(void)
596 register_net_sysctl(&init_net, "net/core", net_core_table);
597 return register_pernet_subsys(&sysctl_core_ops);
600 fs_initcall(sysctl_core_init);