xref: /openbmc/linux/net/core/sysctl_net_core.c (revision b830f94f)
1 // SPDX-License-Identifier: GPL-2.0
2 /* -*- linux-c -*-
3  * sysctl_net_core.c: sysctl interface to net core subsystem.
4  *
5  * Begun April 1, 1996, Mike Shaver.
6  * Added /proc/sys/net/core directory entry (empty =) ). [MS]
7  */
8 
9 #include <linux/mm.h>
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 
19 #include <net/ip.h>
20 #include <net/sock.h>
21 #include <net/net_ratelimit.h>
22 #include <net/busy_poll.h>
23 #include <net/pkt_sched.h>
24 
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;
31 
32 static int net_msg_warn;	/* Unused, but still a sysctl */
33 
34 int sysctl_fb_tunnels_only_for_init_net __read_mostly = 0;
35 EXPORT_SYMBOL(sysctl_fb_tunnels_only_for_init_net);
36 
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
42  */
43 int sysctl_devconf_inherit_init_net __read_mostly;
44 EXPORT_SYMBOL(sysctl_devconf_inherit_init_net);
45 
46 #ifdef CONFIG_RPS
47 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
48 				void __user *buffer, size_t *lenp, loff_t *ppos)
49 {
50 	unsigned int orig_size, size;
51 	int ret, i;
52 	struct ctl_table tmp = {
53 		.data = &size,
54 		.maxlen = sizeof(size),
55 		.mode = table->mode
56 	};
57 	struct rps_sock_flow_table *orig_sock_table, *sock_table;
58 	static DEFINE_MUTEX(sock_flow_mutex);
59 
60 	mutex_lock(&sock_flow_mutex);
61 
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;
65 
66 	ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
67 
68 	if (write) {
69 		if (size) {
70 			if (size > 1<<29) {
71 				/* Enforce limit to prevent overflow */
72 				mutex_unlock(&sock_flow_mutex);
73 				return -EINVAL;
74 			}
75 			size = roundup_pow_of_two(size);
76 			if (size != orig_size) {
77 				sock_table =
78 				    vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
79 				if (!sock_table) {
80 					mutex_unlock(&sock_flow_mutex);
81 					return -ENOMEM;
82 				}
83 				rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
84 				sock_table->mask = size - 1;
85 			} else
86 				sock_table = orig_sock_table;
87 
88 			for (i = 0; i < size; i++)
89 				sock_table->ents[i] = RPS_NO_CPU;
90 		} else
91 			sock_table = NULL;
92 
93 		if (sock_table != orig_sock_table) {
94 			rcu_assign_pointer(rps_sock_flow_table, sock_table);
95 			if (sock_table) {
96 				static_branch_inc(&rps_needed);
97 				static_branch_inc(&rfs_needed);
98 			}
99 			if (orig_sock_table) {
100 				static_branch_dec(&rps_needed);
101 				static_branch_dec(&rfs_needed);
102 				synchronize_rcu();
103 				vfree(orig_sock_table);
104 			}
105 		}
106 	}
107 
108 	mutex_unlock(&sock_flow_mutex);
109 
110 	return ret;
111 }
112 #endif /* CONFIG_RPS */
113 
114 #ifdef CONFIG_NET_FLOW_LIMIT
115 static DEFINE_MUTEX(flow_limit_update_mutex);
116 
117 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
118 				 void __user *buffer, size_t *lenp,
119 				 loff_t *ppos)
120 {
121 	struct sd_flow_limit *cur;
122 	struct softnet_data *sd;
123 	cpumask_var_t mask;
124 	int i, len, ret = 0;
125 
126 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
127 		return -ENOMEM;
128 
129 	if (write) {
130 		ret = cpumask_parse_user(buffer, *lenp, mask);
131 		if (ret)
132 			goto done;
133 
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);
142 				synchronize_rcu();
143 				kfree(cur);
144 			} else if (!cur && cpumask_test_cpu(i, mask)) {
145 				cur = kzalloc_node(len, GFP_KERNEL,
146 						   cpu_to_node(i));
147 				if (!cur) {
148 					/* not unwinding previous changes */
149 					ret = -ENOMEM;
150 					goto write_unlock;
151 				}
152 				cur->num_buckets = netdev_flow_limit_table_len;
153 				rcu_assign_pointer(sd->flow_limit, cur);
154 			}
155 		}
156 write_unlock:
157 		mutex_unlock(&flow_limit_update_mutex);
158 	} else {
159 		char kbuf[128];
160 
161 		if (*ppos || !*lenp) {
162 			*lenp = 0;
163 			goto done;
164 		}
165 
166 		cpumask_clear(mask);
167 		rcu_read_lock();
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);
172 		}
173 		rcu_read_unlock();
174 
175 		len = min(sizeof(kbuf) - 1, *lenp);
176 		len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
177 		if (!len) {
178 			*lenp = 0;
179 			goto done;
180 		}
181 		if (len < *lenp)
182 			kbuf[len++] = '\n';
183 		if (copy_to_user(buffer, kbuf, len)) {
184 			ret = -EFAULT;
185 			goto done;
186 		}
187 		*lenp = len;
188 		*ppos += len;
189 	}
190 
191 done:
192 	free_cpumask_var(mask);
193 	return ret;
194 }
195 
196 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
197 				       void __user *buffer, size_t *lenp,
198 				       loff_t *ppos)
199 {
200 	unsigned int old, *ptr;
201 	int ret;
202 
203 	mutex_lock(&flow_limit_update_mutex);
204 
205 	ptr = table->data;
206 	old = *ptr;
207 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
208 	if (!ret && write && !is_power_of_2(*ptr)) {
209 		*ptr = old;
210 		ret = -EINVAL;
211 	}
212 
213 	mutex_unlock(&flow_limit_update_mutex);
214 	return ret;
215 }
216 #endif /* CONFIG_NET_FLOW_LIMIT */
217 
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)
221 {
222 	char id[IFNAMSIZ];
223 	struct ctl_table tbl = {
224 		.data = id,
225 		.maxlen = IFNAMSIZ,
226 	};
227 	int ret;
228 
229 	qdisc_get_default(id, IFNAMSIZ);
230 
231 	ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
232 	if (write && ret == 0)
233 		ret = qdisc_set_default(id);
234 	return ret;
235 }
236 #endif
237 
238 static int proc_do_dev_weight(struct ctl_table *table, int write,
239 			   void __user *buffer, size_t *lenp, loff_t *ppos)
240 {
241 	int ret;
242 
243 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
244 	if (ret != 0)
245 		return ret;
246 
247 	dev_rx_weight = weight_p * dev_weight_rx_bias;
248 	dev_tx_weight = weight_p * dev_weight_tx_bias;
249 
250 	return ret;
251 }
252 
253 static int proc_do_rss_key(struct ctl_table *table, int write,
254 			   void __user *buffer, size_t *lenp, loff_t *ppos)
255 {
256 	struct ctl_table fake_table;
257 	char buf[NETDEV_RSS_KEY_LEN * 3];
258 
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);
263 }
264 
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,
268 					   loff_t *ppos)
269 {
270 	int ret, jit_enable = *(int *)table->data;
271 	struct ctl_table tmp = *table;
272 
273 	if (write && !capable(CAP_SYS_ADMIN))
274 		return -EPERM;
275 
276 	tmp.data = &jit_enable;
277 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
278 	if (write && !ret) {
279 		if (jit_enable < 2 ||
280 		    (jit_enable == 2 && bpf_dump_raw_ok())) {
281 			*(int *)table->data = jit_enable;
282 			if (jit_enable == 2)
283 				pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
284 		} else {
285 			ret = -EPERM;
286 		}
287 	}
288 	return ret;
289 }
290 
291 static int
292 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write,
293 				    void __user *buffer, size_t *lenp,
294 				    loff_t *ppos)
295 {
296 	if (!capable(CAP_SYS_ADMIN))
297 		return -EPERM;
298 
299 	return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
300 }
301 
302 static int
303 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
304 				     void __user *buffer, size_t *lenp,
305 				     loff_t *ppos)
306 {
307 	if (!capable(CAP_SYS_ADMIN))
308 		return -EPERM;
309 
310 	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
311 }
312 #endif
313 
314 static struct ctl_table net_core_table[] = {
315 #ifdef CONFIG_NET
316 	{
317 		.procname	= "wmem_max",
318 		.data		= &sysctl_wmem_max,
319 		.maxlen		= sizeof(int),
320 		.mode		= 0644,
321 		.proc_handler	= proc_dointvec_minmax,
322 		.extra1		= &min_sndbuf,
323 	},
324 	{
325 		.procname	= "rmem_max",
326 		.data		= &sysctl_rmem_max,
327 		.maxlen		= sizeof(int),
328 		.mode		= 0644,
329 		.proc_handler	= proc_dointvec_minmax,
330 		.extra1		= &min_rcvbuf,
331 	},
332 	{
333 		.procname	= "wmem_default",
334 		.data		= &sysctl_wmem_default,
335 		.maxlen		= sizeof(int),
336 		.mode		= 0644,
337 		.proc_handler	= proc_dointvec_minmax,
338 		.extra1		= &min_sndbuf,
339 	},
340 	{
341 		.procname	= "rmem_default",
342 		.data		= &sysctl_rmem_default,
343 		.maxlen		= sizeof(int),
344 		.mode		= 0644,
345 		.proc_handler	= proc_dointvec_minmax,
346 		.extra1		= &min_rcvbuf,
347 	},
348 	{
349 		.procname	= "dev_weight",
350 		.data		= &weight_p,
351 		.maxlen		= sizeof(int),
352 		.mode		= 0644,
353 		.proc_handler	= proc_do_dev_weight,
354 	},
355 	{
356 		.procname	= "dev_weight_rx_bias",
357 		.data		= &dev_weight_rx_bias,
358 		.maxlen		= sizeof(int),
359 		.mode		= 0644,
360 		.proc_handler	= proc_do_dev_weight,
361 	},
362 	{
363 		.procname	= "dev_weight_tx_bias",
364 		.data		= &dev_weight_tx_bias,
365 		.maxlen		= sizeof(int),
366 		.mode		= 0644,
367 		.proc_handler	= proc_do_dev_weight,
368 	},
369 	{
370 		.procname	= "netdev_max_backlog",
371 		.data		= &netdev_max_backlog,
372 		.maxlen		= sizeof(int),
373 		.mode		= 0644,
374 		.proc_handler	= proc_dointvec
375 	},
376 	{
377 		.procname	= "netdev_rss_key",
378 		.data		= &netdev_rss_key,
379 		.maxlen		= sizeof(int),
380 		.mode		= 0444,
381 		.proc_handler	= proc_do_rss_key,
382 	},
383 #ifdef CONFIG_BPF_JIT
384 	{
385 		.procname	= "bpf_jit_enable",
386 		.data		= &bpf_jit_enable,
387 		.maxlen		= sizeof(int),
388 		.mode		= 0644,
389 		.proc_handler	= proc_dointvec_minmax_bpf_enable,
390 # ifdef CONFIG_BPF_JIT_ALWAYS_ON
391 		.extra1		= SYSCTL_ONE,
392 		.extra2		= SYSCTL_ONE,
393 # else
394 		.extra1		= SYSCTL_ZERO,
395 		.extra2		= &two,
396 # endif
397 	},
398 # ifdef CONFIG_HAVE_EBPF_JIT
399 	{
400 		.procname	= "bpf_jit_harden",
401 		.data		= &bpf_jit_harden,
402 		.maxlen		= sizeof(int),
403 		.mode		= 0600,
404 		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
405 		.extra1		= SYSCTL_ZERO,
406 		.extra2		= &two,
407 	},
408 	{
409 		.procname	= "bpf_jit_kallsyms",
410 		.data		= &bpf_jit_kallsyms,
411 		.maxlen		= sizeof(int),
412 		.mode		= 0600,
413 		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
414 		.extra1		= SYSCTL_ZERO,
415 		.extra2		= SYSCTL_ONE,
416 	},
417 # endif
418 	{
419 		.procname	= "bpf_jit_limit",
420 		.data		= &bpf_jit_limit,
421 		.maxlen		= sizeof(long),
422 		.mode		= 0600,
423 		.proc_handler	= proc_dolongvec_minmax_bpf_restricted,
424 		.extra1		= &long_one,
425 		.extra2		= &long_max,
426 	},
427 #endif
428 	{
429 		.procname	= "netdev_tstamp_prequeue",
430 		.data		= &netdev_tstamp_prequeue,
431 		.maxlen		= sizeof(int),
432 		.mode		= 0644,
433 		.proc_handler	= proc_dointvec
434 	},
435 	{
436 		.procname	= "message_cost",
437 		.data		= &net_ratelimit_state.interval,
438 		.maxlen		= sizeof(int),
439 		.mode		= 0644,
440 		.proc_handler	= proc_dointvec_jiffies,
441 	},
442 	{
443 		.procname	= "message_burst",
444 		.data		= &net_ratelimit_state.burst,
445 		.maxlen		= sizeof(int),
446 		.mode		= 0644,
447 		.proc_handler	= proc_dointvec,
448 	},
449 	{
450 		.procname	= "optmem_max",
451 		.data		= &sysctl_optmem_max,
452 		.maxlen		= sizeof(int),
453 		.mode		= 0644,
454 		.proc_handler	= proc_dointvec
455 	},
456 	{
457 		.procname	= "tstamp_allow_data",
458 		.data		= &sysctl_tstamp_allow_data,
459 		.maxlen		= sizeof(int),
460 		.mode		= 0644,
461 		.proc_handler	= proc_dointvec_minmax,
462 		.extra1		= SYSCTL_ZERO,
463 		.extra2		= SYSCTL_ONE
464 	},
465 #ifdef CONFIG_RPS
466 	{
467 		.procname	= "rps_sock_flow_entries",
468 		.maxlen		= sizeof(int),
469 		.mode		= 0644,
470 		.proc_handler	= rps_sock_flow_sysctl
471 	},
472 #endif
473 #ifdef CONFIG_NET_FLOW_LIMIT
474 	{
475 		.procname	= "flow_limit_cpu_bitmap",
476 		.mode		= 0644,
477 		.proc_handler	= flow_limit_cpu_sysctl
478 	},
479 	{
480 		.procname	= "flow_limit_table_len",
481 		.data		= &netdev_flow_limit_table_len,
482 		.maxlen		= sizeof(int),
483 		.mode		= 0644,
484 		.proc_handler	= flow_limit_table_len_sysctl
485 	},
486 #endif /* CONFIG_NET_FLOW_LIMIT */
487 #ifdef CONFIG_NET_RX_BUSY_POLL
488 	{
489 		.procname	= "busy_poll",
490 		.data		= &sysctl_net_busy_poll,
491 		.maxlen		= sizeof(unsigned int),
492 		.mode		= 0644,
493 		.proc_handler	= proc_dointvec_minmax,
494 		.extra1		= SYSCTL_ZERO,
495 	},
496 	{
497 		.procname	= "busy_read",
498 		.data		= &sysctl_net_busy_read,
499 		.maxlen		= sizeof(unsigned int),
500 		.mode		= 0644,
501 		.proc_handler	= proc_dointvec_minmax,
502 		.extra1		= SYSCTL_ZERO,
503 	},
504 #endif
505 #ifdef CONFIG_NET_SCHED
506 	{
507 		.procname	= "default_qdisc",
508 		.mode		= 0644,
509 		.maxlen		= IFNAMSIZ,
510 		.proc_handler	= set_default_qdisc
511 	},
512 #endif
513 #endif /* CONFIG_NET */
514 	{
515 		.procname	= "netdev_budget",
516 		.data		= &netdev_budget,
517 		.maxlen		= sizeof(int),
518 		.mode		= 0644,
519 		.proc_handler	= proc_dointvec
520 	},
521 	{
522 		.procname	= "warnings",
523 		.data		= &net_msg_warn,
524 		.maxlen		= sizeof(int),
525 		.mode		= 0644,
526 		.proc_handler	= proc_dointvec
527 	},
528 	{
529 		.procname	= "max_skb_frags",
530 		.data		= &sysctl_max_skb_frags,
531 		.maxlen		= sizeof(int),
532 		.mode		= 0644,
533 		.proc_handler	= proc_dointvec_minmax,
534 		.extra1		= SYSCTL_ONE,
535 		.extra2		= &max_skb_frags,
536 	},
537 	{
538 		.procname	= "netdev_budget_usecs",
539 		.data		= &netdev_budget_usecs,
540 		.maxlen		= sizeof(unsigned int),
541 		.mode		= 0644,
542 		.proc_handler	= proc_dointvec_minmax,
543 		.extra1		= SYSCTL_ZERO,
544 	},
545 	{
546 		.procname	= "fb_tunnels_only_for_init_net",
547 		.data		= &sysctl_fb_tunnels_only_for_init_net,
548 		.maxlen		= sizeof(int),
549 		.mode		= 0644,
550 		.proc_handler	= proc_dointvec_minmax,
551 		.extra1		= SYSCTL_ZERO,
552 		.extra2		= SYSCTL_ONE,
553 	},
554 	{
555 		.procname	= "devconf_inherit_init_net",
556 		.data		= &sysctl_devconf_inherit_init_net,
557 		.maxlen		= sizeof(int),
558 		.mode		= 0644,
559 		.proc_handler	= proc_dointvec_minmax,
560 		.extra1		= SYSCTL_ZERO,
561 		.extra2		= &two,
562 	},
563 	{
564 		.procname	= "high_order_alloc_disable",
565 		.data		= &net_high_order_alloc_disable_key.key,
566 		.maxlen         = sizeof(net_high_order_alloc_disable_key),
567 		.mode		= 0644,
568 		.proc_handler	= proc_do_static_key,
569 	},
570 	{ }
571 };
572 
573 static struct ctl_table netns_core_table[] = {
574 	{
575 		.procname	= "somaxconn",
576 		.data		= &init_net.core.sysctl_somaxconn,
577 		.maxlen		= sizeof(int),
578 		.mode		= 0644,
579 		.extra1		= SYSCTL_ZERO,
580 		.proc_handler	= proc_dointvec_minmax
581 	},
582 	{ }
583 };
584 
585 static __net_init int sysctl_core_net_init(struct net *net)
586 {
587 	struct ctl_table *tbl;
588 
589 	tbl = netns_core_table;
590 	if (!net_eq(net, &init_net)) {
591 		tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
592 		if (tbl == NULL)
593 			goto err_dup;
594 
595 		tbl[0].data = &net->core.sysctl_somaxconn;
596 
597 		/* Don't export any sysctls to unprivileged users */
598 		if (net->user_ns != &init_user_ns) {
599 			tbl[0].procname = NULL;
600 		}
601 	}
602 
603 	net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
604 	if (net->core.sysctl_hdr == NULL)
605 		goto err_reg;
606 
607 	return 0;
608 
609 err_reg:
610 	if (tbl != netns_core_table)
611 		kfree(tbl);
612 err_dup:
613 	return -ENOMEM;
614 }
615 
616 static __net_exit void sysctl_core_net_exit(struct net *net)
617 {
618 	struct ctl_table *tbl;
619 
620 	tbl = net->core.sysctl_hdr->ctl_table_arg;
621 	unregister_net_sysctl_table(net->core.sysctl_hdr);
622 	BUG_ON(tbl == netns_core_table);
623 	kfree(tbl);
624 }
625 
626 static __net_initdata struct pernet_operations sysctl_core_ops = {
627 	.init = sysctl_core_net_init,
628 	.exit = sysctl_core_net_exit,
629 };
630 
631 static __init int sysctl_core_init(void)
632 {
633 	register_net_sysctl(&init_net, "net/core", net_core_table);
634 	return register_pernet_subsys(&sysctl_core_ops);
635 }
636 
637 fs_initcall(sysctl_core_init);
638