xref: /openbmc/linux/net/core/sysctl_net_core.c (revision 11a163f2) 
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 = 2;
26 static int three = 3;
27 static int min_sndbuf = SOCK_MIN_SNDBUF;
28 static int min_rcvbuf = SOCK_MIN_RCVBUF;
29 static int max_skb_frags = MAX_SKB_FRAGS;
30 static long long_one __maybe_unused = 1;
31 static long long_max __maybe_unused = LONG_MAX;
32 
33 static int net_msg_warn;	/* Unused, but still a sysctl */
34 
35 int sysctl_fb_tunnels_only_for_init_net __read_mostly = 0;
36 EXPORT_SYMBOL(sysctl_fb_tunnels_only_for_init_net);
37 
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
44  */
45 int sysctl_devconf_inherit_init_net __read_mostly;
46 EXPORT_SYMBOL(sysctl_devconf_inherit_init_net);
47 
48 #ifdef CONFIG_RPS
49 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
50 				void *buffer, size_t *lenp, loff_t *ppos)
51 {
52 	unsigned int orig_size, size;
53 	int ret, i;
54 	struct ctl_table tmp = {
55 		.data = &size,
56 		.maxlen = sizeof(size),
57 		.mode = table->mode
58 	};
59 	struct rps_sock_flow_table *orig_sock_table, *sock_table;
60 	static DEFINE_MUTEX(sock_flow_mutex);
61 
62 	mutex_lock(&sock_flow_mutex);
63 
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;
67 
68 	ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
69 
70 	if (write) {
71 		if (size) {
72 			if (size > 1<<29) {
73 				/* Enforce limit to prevent overflow */
74 				mutex_unlock(&sock_flow_mutex);
75 				return -EINVAL;
76 			}
77 			size = roundup_pow_of_two(size);
78 			if (size != orig_size) {
79 				sock_table =
80 				    vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
81 				if (!sock_table) {
82 					mutex_unlock(&sock_flow_mutex);
83 					return -ENOMEM;
84 				}
85 				rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
86 				sock_table->mask = size - 1;
87 			} else
88 				sock_table = orig_sock_table;
89 
90 			for (i = 0; i < size; i++)
91 				sock_table->ents[i] = RPS_NO_CPU;
92 		} else
93 			sock_table = NULL;
94 
95 		if (sock_table != orig_sock_table) {
96 			rcu_assign_pointer(rps_sock_flow_table, sock_table);
97 			if (sock_table) {
98 				static_branch_inc(&rps_needed);
99 				static_branch_inc(&rfs_needed);
100 			}
101 			if (orig_sock_table) {
102 				static_branch_dec(&rps_needed);
103 				static_branch_dec(&rfs_needed);
104 				synchronize_rcu();
105 				vfree(orig_sock_table);
106 			}
107 		}
108 	}
109 
110 	mutex_unlock(&sock_flow_mutex);
111 
112 	return ret;
113 }
114 #endif /* CONFIG_RPS */
115 
116 #ifdef CONFIG_NET_FLOW_LIMIT
117 static DEFINE_MUTEX(flow_limit_update_mutex);
118 
119 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
120 				 void *buffer, size_t *lenp, loff_t *ppos)
121 {
122 	struct sd_flow_limit *cur;
123 	struct softnet_data *sd;
124 	cpumask_var_t mask;
125 	int i, len, ret = 0;
126 
127 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
128 		return -ENOMEM;
129 
130 	if (write) {
131 		ret = cpumask_parse(buffer, mask);
132 		if (ret)
133 			goto done;
134 
135 		mutex_lock(&flow_limit_update_mutex);
136 		len = sizeof(*cur) + netdev_flow_limit_table_len;
137 		for_each_possible_cpu(i) {
138 			sd = &per_cpu(softnet_data, i);
139 			cur = rcu_dereference_protected(sd->flow_limit,
140 				     lockdep_is_held(&flow_limit_update_mutex));
141 			if (cur && !cpumask_test_cpu(i, mask)) {
142 				RCU_INIT_POINTER(sd->flow_limit, NULL);
143 				synchronize_rcu();
144 				kfree(cur);
145 			} else if (!cur && cpumask_test_cpu(i, mask)) {
146 				cur = kzalloc_node(len, GFP_KERNEL,
147 						   cpu_to_node(i));
148 				if (!cur) {
149 					/* not unwinding previous changes */
150 					ret = -ENOMEM;
151 					goto write_unlock;
152 				}
153 				cur->num_buckets = netdev_flow_limit_table_len;
154 				rcu_assign_pointer(sd->flow_limit, cur);
155 			}
156 		}
157 write_unlock:
158 		mutex_unlock(&flow_limit_update_mutex);
159 	} else {
160 		char kbuf[128];
161 
162 		if (*ppos || !*lenp) {
163 			*lenp = 0;
164 			goto done;
165 		}
166 
167 		cpumask_clear(mask);
168 		rcu_read_lock();
169 		for_each_possible_cpu(i) {
170 			sd = &per_cpu(softnet_data, i);
171 			if (rcu_dereference(sd->flow_limit))
172 				cpumask_set_cpu(i, mask);
173 		}
174 		rcu_read_unlock();
175 
176 		len = min(sizeof(kbuf) - 1, *lenp);
177 		len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
178 		if (!len) {
179 			*lenp = 0;
180 			goto done;
181 		}
182 		if (len < *lenp)
183 			kbuf[len++] = '\n';
184 		memcpy(buffer, kbuf, len);
185 		*lenp = len;
186 		*ppos += len;
187 	}
188 
189 done:
190 	free_cpumask_var(mask);
191 	return ret;
192 }
193 
194 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
195 				       void *buffer, size_t *lenp, loff_t *ppos)
196 {
197 	unsigned int old, *ptr;
198 	int ret;
199 
200 	mutex_lock(&flow_limit_update_mutex);
201 
202 	ptr = table->data;
203 	old = *ptr;
204 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
205 	if (!ret && write && !is_power_of_2(*ptr)) {
206 		*ptr = old;
207 		ret = -EINVAL;
208 	}
209 
210 	mutex_unlock(&flow_limit_update_mutex);
211 	return ret;
212 }
213 #endif /* CONFIG_NET_FLOW_LIMIT */
214 
215 #ifdef CONFIG_NET_SCHED
216 static int set_default_qdisc(struct ctl_table *table, int write,
217 			     void *buffer, size_t *lenp, loff_t *ppos)
218 {
219 	char id[IFNAMSIZ];
220 	struct ctl_table tbl = {
221 		.data = id,
222 		.maxlen = IFNAMSIZ,
223 	};
224 	int ret;
225 
226 	qdisc_get_default(id, IFNAMSIZ);
227 
228 	ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
229 	if (write && ret == 0)
230 		ret = qdisc_set_default(id);
231 	return ret;
232 }
233 #endif
234 
235 static int proc_do_dev_weight(struct ctl_table *table, int write,
236 			   void *buffer, size_t *lenp, loff_t *ppos)
237 {
238 	int ret;
239 
240 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
241 	if (ret != 0)
242 		return ret;
243 
244 	dev_rx_weight = weight_p * dev_weight_rx_bias;
245 	dev_tx_weight = weight_p * dev_weight_tx_bias;
246 
247 	return ret;
248 }
249 
250 static int proc_do_rss_key(struct ctl_table *table, int write,
251 			   void *buffer, size_t *lenp, loff_t *ppos)
252 {
253 	struct ctl_table fake_table;
254 	char buf[NETDEV_RSS_KEY_LEN * 3];
255 
256 	snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
257 	fake_table.data = buf;
258 	fake_table.maxlen = sizeof(buf);
259 	return proc_dostring(&fake_table, write, buffer, lenp, ppos);
260 }
261 
262 #ifdef CONFIG_BPF_JIT
263 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write,
264 					   void *buffer, size_t *lenp,
265 					   loff_t *ppos)
266 {
267 	int ret, jit_enable = *(int *)table->data;
268 	struct ctl_table tmp = *table;
269 
270 	if (write && !capable(CAP_SYS_ADMIN))
271 		return -EPERM;
272 
273 	tmp.data = &jit_enable;
274 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
275 	if (write && !ret) {
276 		if (jit_enable < 2 ||
277 		    (jit_enable == 2 && bpf_dump_raw_ok(current_cred()))) {
278 			*(int *)table->data = jit_enable;
279 			if (jit_enable == 2)
280 				pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
281 		} else {
282 			ret = -EPERM;
283 		}
284 	}
285 	return ret;
286 }
287 
288 # ifdef CONFIG_HAVE_EBPF_JIT
289 static int
290 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write,
291 				    void *buffer, size_t *lenp, loff_t *ppos)
292 {
293 	if (!capable(CAP_SYS_ADMIN))
294 		return -EPERM;
295 
296 	return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
297 }
298 # endif /* CONFIG_HAVE_EBPF_JIT */
299 
300 static int
301 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
302 				     void *buffer, size_t *lenp, loff_t *ppos)
303 {
304 	if (!capable(CAP_SYS_ADMIN))
305 		return -EPERM;
306 
307 	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
308 }
309 #endif
310 
311 static struct ctl_table net_core_table[] = {
312 #ifdef CONFIG_NET
313 	{
314 		.procname	= "wmem_max",
315 		.data		= &sysctl_wmem_max,
316 		.maxlen		= sizeof(int),
317 		.mode		= 0644,
318 		.proc_handler	= proc_dointvec_minmax,
319 		.extra1		= &min_sndbuf,
320 	},
321 	{
322 		.procname	= "rmem_max",
323 		.data		= &sysctl_rmem_max,
324 		.maxlen		= sizeof(int),
325 		.mode		= 0644,
326 		.proc_handler	= proc_dointvec_minmax,
327 		.extra1		= &min_rcvbuf,
328 	},
329 	{
330 		.procname	= "wmem_default",
331 		.data		= &sysctl_wmem_default,
332 		.maxlen		= sizeof(int),
333 		.mode		= 0644,
334 		.proc_handler	= proc_dointvec_minmax,
335 		.extra1		= &min_sndbuf,
336 	},
337 	{
338 		.procname	= "rmem_default",
339 		.data		= &sysctl_rmem_default,
340 		.maxlen		= sizeof(int),
341 		.mode		= 0644,
342 		.proc_handler	= proc_dointvec_minmax,
343 		.extra1		= &min_rcvbuf,
344 	},
345 	{
346 		.procname	= "dev_weight",
347 		.data		= &weight_p,
348 		.maxlen		= sizeof(int),
349 		.mode		= 0644,
350 		.proc_handler	= proc_do_dev_weight,
351 	},
352 	{
353 		.procname	= "dev_weight_rx_bias",
354 		.data		= &dev_weight_rx_bias,
355 		.maxlen		= sizeof(int),
356 		.mode		= 0644,
357 		.proc_handler	= proc_do_dev_weight,
358 	},
359 	{
360 		.procname	= "dev_weight_tx_bias",
361 		.data		= &dev_weight_tx_bias,
362 		.maxlen		= sizeof(int),
363 		.mode		= 0644,
364 		.proc_handler	= proc_do_dev_weight,
365 	},
366 	{
367 		.procname	= "netdev_max_backlog",
368 		.data		= &netdev_max_backlog,
369 		.maxlen		= sizeof(int),
370 		.mode		= 0644,
371 		.proc_handler	= proc_dointvec
372 	},
373 	{
374 		.procname	= "netdev_rss_key",
375 		.data		= &netdev_rss_key,
376 		.maxlen		= sizeof(int),
377 		.mode		= 0444,
378 		.proc_handler	= proc_do_rss_key,
379 	},
380 #ifdef CONFIG_BPF_JIT
381 	{
382 		.procname	= "bpf_jit_enable",
383 		.data		= &bpf_jit_enable,
384 		.maxlen		= sizeof(int),
385 		.mode		= 0644,
386 		.proc_handler	= proc_dointvec_minmax_bpf_enable,
387 # ifdef CONFIG_BPF_JIT_ALWAYS_ON
388 		.extra1		= SYSCTL_ONE,
389 		.extra2		= SYSCTL_ONE,
390 # else
391 		.extra1		= SYSCTL_ZERO,
392 		.extra2		= &two,
393 # endif
394 	},
395 # ifdef CONFIG_HAVE_EBPF_JIT
396 	{
397 		.procname	= "bpf_jit_harden",
398 		.data		= &bpf_jit_harden,
399 		.maxlen		= sizeof(int),
400 		.mode		= 0600,
401 		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
402 		.extra1		= SYSCTL_ZERO,
403 		.extra2		= &two,
404 	},
405 	{
406 		.procname	= "bpf_jit_kallsyms",
407 		.data		= &bpf_jit_kallsyms,
408 		.maxlen		= sizeof(int),
409 		.mode		= 0600,
410 		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
411 		.extra1		= SYSCTL_ZERO,
412 		.extra2		= SYSCTL_ONE,
413 	},
414 # endif
415 	{
416 		.procname	= "bpf_jit_limit",
417 		.data		= &bpf_jit_limit,
418 		.maxlen		= sizeof(long),
419 		.mode		= 0600,
420 		.proc_handler	= proc_dolongvec_minmax_bpf_restricted,
421 		.extra1		= &long_one,
422 		.extra2		= &long_max,
423 	},
424 #endif
425 	{
426 		.procname	= "netdev_tstamp_prequeue",
427 		.data		= &netdev_tstamp_prequeue,
428 		.maxlen		= sizeof(int),
429 		.mode		= 0644,
430 		.proc_handler	= proc_dointvec
431 	},
432 	{
433 		.procname	= "message_cost",
434 		.data		= &net_ratelimit_state.interval,
435 		.maxlen		= sizeof(int),
436 		.mode		= 0644,
437 		.proc_handler	= proc_dointvec_jiffies,
438 	},
439 	{
440 		.procname	= "message_burst",
441 		.data		= &net_ratelimit_state.burst,
442 		.maxlen		= sizeof(int),
443 		.mode		= 0644,
444 		.proc_handler	= proc_dointvec,
445 	},
446 	{
447 		.procname	= "optmem_max",
448 		.data		= &sysctl_optmem_max,
449 		.maxlen		= sizeof(int),
450 		.mode		= 0644,
451 		.proc_handler	= proc_dointvec
452 	},
453 	{
454 		.procname	= "tstamp_allow_data",
455 		.data		= &sysctl_tstamp_allow_data,
456 		.maxlen		= sizeof(int),
457 		.mode		= 0644,
458 		.proc_handler	= proc_dointvec_minmax,
459 		.extra1		= SYSCTL_ZERO,
460 		.extra2		= SYSCTL_ONE
461 	},
462 #ifdef CONFIG_RPS
463 	{
464 		.procname	= "rps_sock_flow_entries",
465 		.maxlen		= sizeof(int),
466 		.mode		= 0644,
467 		.proc_handler	= rps_sock_flow_sysctl
468 	},
469 #endif
470 #ifdef CONFIG_NET_FLOW_LIMIT
471 	{
472 		.procname	= "flow_limit_cpu_bitmap",
473 		.mode		= 0644,
474 		.proc_handler	= flow_limit_cpu_sysctl
475 	},
476 	{
477 		.procname	= "flow_limit_table_len",
478 		.data		= &netdev_flow_limit_table_len,
479 		.maxlen		= sizeof(int),
480 		.mode		= 0644,
481 		.proc_handler	= flow_limit_table_len_sysctl
482 	},
483 #endif /* CONFIG_NET_FLOW_LIMIT */
484 #ifdef CONFIG_NET_RX_BUSY_POLL
485 	{
486 		.procname	= "busy_poll",
487 		.data		= &sysctl_net_busy_poll,
488 		.maxlen		= sizeof(unsigned int),
489 		.mode		= 0644,
490 		.proc_handler	= proc_dointvec_minmax,
491 		.extra1		= SYSCTL_ZERO,
492 	},
493 	{
494 		.procname	= "busy_read",
495 		.data		= &sysctl_net_busy_read,
496 		.maxlen		= sizeof(unsigned int),
497 		.mode		= 0644,
498 		.proc_handler	= proc_dointvec_minmax,
499 		.extra1		= SYSCTL_ZERO,
500 	},
501 #endif
502 #ifdef CONFIG_NET_SCHED
503 	{
504 		.procname	= "default_qdisc",
505 		.mode		= 0644,
506 		.maxlen		= IFNAMSIZ,
507 		.proc_handler	= set_default_qdisc
508 	},
509 #endif
510 #endif /* CONFIG_NET */
511 	{
512 		.procname	= "netdev_budget",
513 		.data		= &netdev_budget,
514 		.maxlen		= sizeof(int),
515 		.mode		= 0644,
516 		.proc_handler	= proc_dointvec
517 	},
518 	{
519 		.procname	= "warnings",
520 		.data		= &net_msg_warn,
521 		.maxlen		= sizeof(int),
522 		.mode		= 0644,
523 		.proc_handler	= proc_dointvec
524 	},
525 	{
526 		.procname	= "max_skb_frags",
527 		.data		= &sysctl_max_skb_frags,
528 		.maxlen		= sizeof(int),
529 		.mode		= 0644,
530 		.proc_handler	= proc_dointvec_minmax,
531 		.extra1		= SYSCTL_ONE,
532 		.extra2		= &max_skb_frags,
533 	},
534 	{
535 		.procname	= "netdev_budget_usecs",
536 		.data		= &netdev_budget_usecs,
537 		.maxlen		= sizeof(unsigned int),
538 		.mode		= 0644,
539 		.proc_handler	= proc_dointvec_minmax,
540 		.extra1		= SYSCTL_ZERO,
541 	},
542 	{
543 		.procname	= "fb_tunnels_only_for_init_net",
544 		.data		= &sysctl_fb_tunnels_only_for_init_net,
545 		.maxlen		= sizeof(int),
546 		.mode		= 0644,
547 		.proc_handler	= proc_dointvec_minmax,
548 		.extra1		= SYSCTL_ZERO,
549 		.extra2		= &two,
550 	},
551 	{
552 		.procname	= "devconf_inherit_init_net",
553 		.data		= &sysctl_devconf_inherit_init_net,
554 		.maxlen		= sizeof(int),
555 		.mode		= 0644,
556 		.proc_handler	= proc_dointvec_minmax,
557 		.extra1		= SYSCTL_ZERO,
558 		.extra2		= &three,
559 	},
560 	{
561 		.procname	= "high_order_alloc_disable",
562 		.data		= &net_high_order_alloc_disable_key.key,
563 		.maxlen         = sizeof(net_high_order_alloc_disable_key),
564 		.mode		= 0644,
565 		.proc_handler	= proc_do_static_key,
566 	},
567 	{
568 		.procname	= "gro_normal_batch",
569 		.data		= &gro_normal_batch,
570 		.maxlen		= sizeof(unsigned int),
571 		.mode		= 0644,
572 		.proc_handler	= proc_dointvec_minmax,
573 		.extra1		= SYSCTL_ONE,
574 	},
575 	{ }
576 };
577 
578 static struct ctl_table netns_core_table[] = {
579 	{
580 		.procname	= "somaxconn",
581 		.data		= &init_net.core.sysctl_somaxconn,
582 		.maxlen		= sizeof(int),
583 		.mode		= 0644,
584 		.extra1		= SYSCTL_ZERO,
585 		.proc_handler	= proc_dointvec_minmax
586 	},
587 	{ }
588 };
589 
590 static int __init fb_tunnels_only_for_init_net_sysctl_setup(char *str)
591 {
592 	/* fallback tunnels for initns only */
593 	if (!strncmp(str, "initns", 6))
594 		sysctl_fb_tunnels_only_for_init_net = 1;
595 	/* no fallback tunnels anywhere */
596 	else if (!strncmp(str, "none", 4))
597 		sysctl_fb_tunnels_only_for_init_net = 2;
598 
599 	return 1;
600 }
601 __setup("fb_tunnels=", fb_tunnels_only_for_init_net_sysctl_setup);
602 
603 static __net_init int sysctl_core_net_init(struct net *net)
604 {
605 	struct ctl_table *tbl;
606 
607 	tbl = netns_core_table;
608 	if (!net_eq(net, &init_net)) {
609 		tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
610 		if (tbl == NULL)
611 			goto err_dup;
612 
613 		tbl[0].data = &net->core.sysctl_somaxconn;
614 
615 		/* Don't export any sysctls to unprivileged users */
616 		if (net->user_ns != &init_user_ns) {
617 			tbl[0].procname = NULL;
618 		}
619 	}
620 
621 	net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
622 	if (net->core.sysctl_hdr == NULL)
623 		goto err_reg;
624 
625 	return 0;
626 
627 err_reg:
628 	if (tbl != netns_core_table)
629 		kfree(tbl);
630 err_dup:
631 	return -ENOMEM;
632 }
633 
634 static __net_exit void sysctl_core_net_exit(struct net *net)
635 {
636 	struct ctl_table *tbl;
637 
638 	tbl = net->core.sysctl_hdr->ctl_table_arg;
639 	unregister_net_sysctl_table(net->core.sysctl_hdr);
640 	BUG_ON(tbl == netns_core_table);
641 	kfree(tbl);
642 }
643 
644 static __net_initdata struct pernet_operations sysctl_core_ops = {
645 	.init = sysctl_core_net_init,
646 	.exit = sysctl_core_net_exit,
647 };
648 
649 static __init int sysctl_core_init(void)
650 {
651 	register_net_sysctl(&init_net, "net/core", net_core_table);
652 	return register_pernet_subsys(&sysctl_core_ops);
653 }
654 
655 fs_initcall(sysctl_core_init);
656