xref: /openbmc/linux/net/core/sysctl_net_core.c (revision 62eab49f) 
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 	{
313 		.procname	= "wmem_max",
314 		.data		= &sysctl_wmem_max,
315 		.maxlen		= sizeof(int),
316 		.mode		= 0644,
317 		.proc_handler	= proc_dointvec_minmax,
318 		.extra1		= &min_sndbuf,
319 	},
320 	{
321 		.procname	= "rmem_max",
322 		.data		= &sysctl_rmem_max,
323 		.maxlen		= sizeof(int),
324 		.mode		= 0644,
325 		.proc_handler	= proc_dointvec_minmax,
326 		.extra1		= &min_rcvbuf,
327 	},
328 	{
329 		.procname	= "wmem_default",
330 		.data		= &sysctl_wmem_default,
331 		.maxlen		= sizeof(int),
332 		.mode		= 0644,
333 		.proc_handler	= proc_dointvec_minmax,
334 		.extra1		= &min_sndbuf,
335 	},
336 	{
337 		.procname	= "rmem_default",
338 		.data		= &sysctl_rmem_default,
339 		.maxlen		= sizeof(int),
340 		.mode		= 0644,
341 		.proc_handler	= proc_dointvec_minmax,
342 		.extra1		= &min_rcvbuf,
343 	},
344 	{
345 		.procname	= "dev_weight",
346 		.data		= &weight_p,
347 		.maxlen		= sizeof(int),
348 		.mode		= 0644,
349 		.proc_handler	= proc_do_dev_weight,
350 	},
351 	{
352 		.procname	= "dev_weight_rx_bias",
353 		.data		= &dev_weight_rx_bias,
354 		.maxlen		= sizeof(int),
355 		.mode		= 0644,
356 		.proc_handler	= proc_do_dev_weight,
357 	},
358 	{
359 		.procname	= "dev_weight_tx_bias",
360 		.data		= &dev_weight_tx_bias,
361 		.maxlen		= sizeof(int),
362 		.mode		= 0644,
363 		.proc_handler	= proc_do_dev_weight,
364 	},
365 	{
366 		.procname	= "netdev_max_backlog",
367 		.data		= &netdev_max_backlog,
368 		.maxlen		= sizeof(int),
369 		.mode		= 0644,
370 		.proc_handler	= proc_dointvec
371 	},
372 	{
373 		.procname	= "netdev_rss_key",
374 		.data		= &netdev_rss_key,
375 		.maxlen		= sizeof(int),
376 		.mode		= 0444,
377 		.proc_handler	= proc_do_rss_key,
378 	},
379 #ifdef CONFIG_BPF_JIT
380 	{
381 		.procname	= "bpf_jit_enable",
382 		.data		= &bpf_jit_enable,
383 		.maxlen		= sizeof(int),
384 		.mode		= 0644,
385 		.proc_handler	= proc_dointvec_minmax_bpf_enable,
386 # ifdef CONFIG_BPF_JIT_ALWAYS_ON
387 		.extra1		= SYSCTL_ONE,
388 		.extra2		= SYSCTL_ONE,
389 # else
390 		.extra1		= SYSCTL_ZERO,
391 		.extra2		= &two,
392 # endif
393 	},
394 # ifdef CONFIG_HAVE_EBPF_JIT
395 	{
396 		.procname	= "bpf_jit_harden",
397 		.data		= &bpf_jit_harden,
398 		.maxlen		= sizeof(int),
399 		.mode		= 0600,
400 		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
401 		.extra1		= SYSCTL_ZERO,
402 		.extra2		= &two,
403 	},
404 	{
405 		.procname	= "bpf_jit_kallsyms",
406 		.data		= &bpf_jit_kallsyms,
407 		.maxlen		= sizeof(int),
408 		.mode		= 0600,
409 		.proc_handler	= proc_dointvec_minmax_bpf_restricted,
410 		.extra1		= SYSCTL_ZERO,
411 		.extra2		= SYSCTL_ONE,
412 	},
413 # endif
414 	{
415 		.procname	= "bpf_jit_limit",
416 		.data		= &bpf_jit_limit,
417 		.maxlen		= sizeof(long),
418 		.mode		= 0600,
419 		.proc_handler	= proc_dolongvec_minmax_bpf_restricted,
420 		.extra1		= &long_one,
421 		.extra2		= &long_max,
422 	},
423 #endif
424 	{
425 		.procname	= "netdev_tstamp_prequeue",
426 		.data		= &netdev_tstamp_prequeue,
427 		.maxlen		= sizeof(int),
428 		.mode		= 0644,
429 		.proc_handler	= proc_dointvec
430 	},
431 	{
432 		.procname	= "message_cost",
433 		.data		= &net_ratelimit_state.interval,
434 		.maxlen		= sizeof(int),
435 		.mode		= 0644,
436 		.proc_handler	= proc_dointvec_jiffies,
437 	},
438 	{
439 		.procname	= "message_burst",
440 		.data		= &net_ratelimit_state.burst,
441 		.maxlen		= sizeof(int),
442 		.mode		= 0644,
443 		.proc_handler	= proc_dointvec,
444 	},
445 	{
446 		.procname	= "optmem_max",
447 		.data		= &sysctl_optmem_max,
448 		.maxlen		= sizeof(int),
449 		.mode		= 0644,
450 		.proc_handler	= proc_dointvec
451 	},
452 	{
453 		.procname	= "tstamp_allow_data",
454 		.data		= &sysctl_tstamp_allow_data,
455 		.maxlen		= sizeof(int),
456 		.mode		= 0644,
457 		.proc_handler	= proc_dointvec_minmax,
458 		.extra1		= SYSCTL_ZERO,
459 		.extra2		= SYSCTL_ONE
460 	},
461 #ifdef CONFIG_RPS
462 	{
463 		.procname	= "rps_sock_flow_entries",
464 		.maxlen		= sizeof(int),
465 		.mode		= 0644,
466 		.proc_handler	= rps_sock_flow_sysctl
467 	},
468 #endif
469 #ifdef CONFIG_NET_FLOW_LIMIT
470 	{
471 		.procname	= "flow_limit_cpu_bitmap",
472 		.mode		= 0644,
473 		.proc_handler	= flow_limit_cpu_sysctl
474 	},
475 	{
476 		.procname	= "flow_limit_table_len",
477 		.data		= &netdev_flow_limit_table_len,
478 		.maxlen		= sizeof(int),
479 		.mode		= 0644,
480 		.proc_handler	= flow_limit_table_len_sysctl
481 	},
482 #endif /* CONFIG_NET_FLOW_LIMIT */
483 #ifdef CONFIG_NET_RX_BUSY_POLL
484 	{
485 		.procname	= "busy_poll",
486 		.data		= &sysctl_net_busy_poll,
487 		.maxlen		= sizeof(unsigned int),
488 		.mode		= 0644,
489 		.proc_handler	= proc_dointvec_minmax,
490 		.extra1		= SYSCTL_ZERO,
491 	},
492 	{
493 		.procname	= "busy_read",
494 		.data		= &sysctl_net_busy_read,
495 		.maxlen		= sizeof(unsigned int),
496 		.mode		= 0644,
497 		.proc_handler	= proc_dointvec_minmax,
498 		.extra1		= SYSCTL_ZERO,
499 	},
500 #endif
501 #ifdef CONFIG_NET_SCHED
502 	{
503 		.procname	= "default_qdisc",
504 		.mode		= 0644,
505 		.maxlen		= IFNAMSIZ,
506 		.proc_handler	= set_default_qdisc
507 	},
508 #endif
509 	{
510 		.procname	= "netdev_budget",
511 		.data		= &netdev_budget,
512 		.maxlen		= sizeof(int),
513 		.mode		= 0644,
514 		.proc_handler	= proc_dointvec
515 	},
516 	{
517 		.procname	= "warnings",
518 		.data		= &net_msg_warn,
519 		.maxlen		= sizeof(int),
520 		.mode		= 0644,
521 		.proc_handler	= proc_dointvec
522 	},
523 	{
524 		.procname	= "max_skb_frags",
525 		.data		= &sysctl_max_skb_frags,
526 		.maxlen		= sizeof(int),
527 		.mode		= 0644,
528 		.proc_handler	= proc_dointvec_minmax,
529 		.extra1		= SYSCTL_ONE,
530 		.extra2		= &max_skb_frags,
531 	},
532 	{
533 		.procname	= "netdev_budget_usecs",
534 		.data		= &netdev_budget_usecs,
535 		.maxlen		= sizeof(unsigned int),
536 		.mode		= 0644,
537 		.proc_handler	= proc_dointvec_minmax,
538 		.extra1		= SYSCTL_ZERO,
539 	},
540 	{
541 		.procname	= "fb_tunnels_only_for_init_net",
542 		.data		= &sysctl_fb_tunnels_only_for_init_net,
543 		.maxlen		= sizeof(int),
544 		.mode		= 0644,
545 		.proc_handler	= proc_dointvec_minmax,
546 		.extra1		= SYSCTL_ZERO,
547 		.extra2		= &two,
548 	},
549 	{
550 		.procname	= "devconf_inherit_init_net",
551 		.data		= &sysctl_devconf_inherit_init_net,
552 		.maxlen		= sizeof(int),
553 		.mode		= 0644,
554 		.proc_handler	= proc_dointvec_minmax,
555 		.extra1		= SYSCTL_ZERO,
556 		.extra2		= &three,
557 	},
558 	{
559 		.procname	= "high_order_alloc_disable",
560 		.data		= &net_high_order_alloc_disable_key.key,
561 		.maxlen         = sizeof(net_high_order_alloc_disable_key),
562 		.mode		= 0644,
563 		.proc_handler	= proc_do_static_key,
564 	},
565 	{
566 		.procname	= "gro_normal_batch",
567 		.data		= &gro_normal_batch,
568 		.maxlen		= sizeof(unsigned int),
569 		.mode		= 0644,
570 		.proc_handler	= proc_dointvec_minmax,
571 		.extra1		= SYSCTL_ONE,
572 	},
573 	{ }
574 };
575 
576 static struct ctl_table netns_core_table[] = {
577 	{
578 		.procname	= "somaxconn",
579 		.data		= &init_net.core.sysctl_somaxconn,
580 		.maxlen		= sizeof(int),
581 		.mode		= 0644,
582 		.extra1		= SYSCTL_ZERO,
583 		.proc_handler	= proc_dointvec_minmax
584 	},
585 	{ }
586 };
587 
588 static int __init fb_tunnels_only_for_init_net_sysctl_setup(char *str)
589 {
590 	/* fallback tunnels for initns only */
591 	if (!strncmp(str, "initns", 6))
592 		sysctl_fb_tunnels_only_for_init_net = 1;
593 	/* no fallback tunnels anywhere */
594 	else if (!strncmp(str, "none", 4))
595 		sysctl_fb_tunnels_only_for_init_net = 2;
596 
597 	return 1;
598 }
599 __setup("fb_tunnels=", fb_tunnels_only_for_init_net_sysctl_setup);
600 
601 static __net_init int sysctl_core_net_init(struct net *net)
602 {
603 	struct ctl_table *tbl;
604 
605 	tbl = netns_core_table;
606 	if (!net_eq(net, &init_net)) {
607 		tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
608 		if (tbl == NULL)
609 			goto err_dup;
610 
611 		tbl[0].data = &net->core.sysctl_somaxconn;
612 
613 		/* Don't export any sysctls to unprivileged users */
614 		if (net->user_ns != &init_user_ns) {
615 			tbl[0].procname = NULL;
616 		}
617 	}
618 
619 	net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
620 	if (net->core.sysctl_hdr == NULL)
621 		goto err_reg;
622 
623 	return 0;
624 
625 err_reg:
626 	if (tbl != netns_core_table)
627 		kfree(tbl);
628 err_dup:
629 	return -ENOMEM;
630 }
631 
632 static __net_exit void sysctl_core_net_exit(struct net *net)
633 {
634 	struct ctl_table *tbl;
635 
636 	tbl = net->core.sysctl_hdr->ctl_table_arg;
637 	unregister_net_sysctl_table(net->core.sysctl_hdr);
638 	BUG_ON(tbl == netns_core_table);
639 	kfree(tbl);
640 }
641 
642 static __net_initdata struct pernet_operations sysctl_core_ops = {
643 	.init = sysctl_core_net_init,
644 	.exit = sysctl_core_net_exit,
645 };
646 
647 static __init int sysctl_core_init(void)
648 {
649 	register_net_sysctl(&init_net, "net/core", net_core_table);
650 	return register_pernet_subsys(&sysctl_core_ops);
651 }
652 
653 fs_initcall(sysctl_core_init);
654