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
rps_sock_flow_sysctl(struct ctl_table * table,int write,void __user * buffer,size_t * lenp,loff_t * ppos)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
flow_limit_cpu_sysctl(struct ctl_table * table,int write,void __user * buffer,size_t * lenp,loff_t * ppos)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
flow_limit_table_len_sysctl(struct ctl_table * table,int write,void __user * buffer,size_t * lenp,loff_t * ppos)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
set_default_qdisc(struct ctl_table * table,int write,void __user * buffer,size_t * lenp,loff_t * ppos)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
proc_do_dev_weight(struct ctl_table * table,int write,void __user * buffer,size_t * lenp,loff_t * ppos)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 static DEFINE_MUTEX(dev_weight_mutex);
242 int ret, weight;
243
244 mutex_lock(&dev_weight_mutex);
245 ret = proc_dointvec(table, write, buffer, lenp, ppos);
246 if (!ret && write) {
247 weight = READ_ONCE(weight_p);
248 WRITE_ONCE(dev_rx_weight, weight * dev_weight_rx_bias);
249 WRITE_ONCE(dev_tx_weight, weight * dev_weight_tx_bias);
250 }
251 mutex_unlock(&dev_weight_mutex);
252
253 return ret;
254 }
255
proc_do_rss_key(struct ctl_table * table,int write,void __user * buffer,size_t * lenp,loff_t * ppos)256 static int proc_do_rss_key(struct ctl_table *table, int write,
257 void __user *buffer, size_t *lenp, loff_t *ppos)
258 {
259 struct ctl_table fake_table;
260 char buf[NETDEV_RSS_KEY_LEN * 3];
261
262 snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
263 fake_table.data = buf;
264 fake_table.maxlen = sizeof(buf);
265 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
266 }
267
268 #ifdef CONFIG_BPF_JIT
proc_dointvec_minmax_bpf_enable(struct ctl_table * table,int write,void __user * buffer,size_t * lenp,loff_t * ppos)269 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write,
270 void __user *buffer, size_t *lenp,
271 loff_t *ppos)
272 {
273 int ret, jit_enable = *(int *)table->data;
274 struct ctl_table tmp = *table;
275
276 if (write && !capable(CAP_SYS_ADMIN))
277 return -EPERM;
278
279 tmp.data = &jit_enable;
280 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
281 if (write && !ret) {
282 if (jit_enable < 2 ||
283 (jit_enable == 2 && bpf_dump_raw_ok(current_cred()))) {
284 *(int *)table->data = jit_enable;
285 if (jit_enable == 2)
286 pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
287 } else {
288 ret = -EPERM;
289 }
290 }
291 return ret;
292 }
293
294 # ifdef CONFIG_HAVE_EBPF_JIT
295 static int
proc_dointvec_minmax_bpf_restricted(struct ctl_table * table,int write,void __user * buffer,size_t * lenp,loff_t * ppos)296 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write,
297 void __user *buffer, size_t *lenp,
298 loff_t *ppos)
299 {
300 if (!capable(CAP_SYS_ADMIN))
301 return -EPERM;
302
303 return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
304 }
305 # endif /* CONFIG_HAVE_EBPF_JIT */
306
307 static int
proc_dolongvec_minmax_bpf_restricted(struct ctl_table * table,int write,void __user * buffer,size_t * lenp,loff_t * ppos)308 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
309 void __user *buffer, size_t *lenp,
310 loff_t *ppos)
311 {
312 if (!capable(CAP_SYS_ADMIN))
313 return -EPERM;
314
315 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
316 }
317 #endif
318
319 static struct ctl_table net_core_table[] = {
320 #ifdef CONFIG_NET
321 {
322 .procname = "wmem_max",
323 .data = &sysctl_wmem_max,
324 .maxlen = sizeof(int),
325 .mode = 0644,
326 .proc_handler = proc_dointvec_minmax,
327 .extra1 = &min_sndbuf,
328 },
329 {
330 .procname = "rmem_max",
331 .data = &sysctl_rmem_max,
332 .maxlen = sizeof(int),
333 .mode = 0644,
334 .proc_handler = proc_dointvec_minmax,
335 .extra1 = &min_rcvbuf,
336 },
337 {
338 .procname = "wmem_default",
339 .data = &sysctl_wmem_default,
340 .maxlen = sizeof(int),
341 .mode = 0644,
342 .proc_handler = proc_dointvec_minmax,
343 .extra1 = &min_sndbuf,
344 },
345 {
346 .procname = "rmem_default",
347 .data = &sysctl_rmem_default,
348 .maxlen = sizeof(int),
349 .mode = 0644,
350 .proc_handler = proc_dointvec_minmax,
351 .extra1 = &min_rcvbuf,
352 },
353 {
354 .procname = "dev_weight",
355 .data = &weight_p,
356 .maxlen = sizeof(int),
357 .mode = 0644,
358 .proc_handler = proc_do_dev_weight,
359 },
360 {
361 .procname = "dev_weight_rx_bias",
362 .data = &dev_weight_rx_bias,
363 .maxlen = sizeof(int),
364 .mode = 0644,
365 .proc_handler = proc_do_dev_weight,
366 },
367 {
368 .procname = "dev_weight_tx_bias",
369 .data = &dev_weight_tx_bias,
370 .maxlen = sizeof(int),
371 .mode = 0644,
372 .proc_handler = proc_do_dev_weight,
373 },
374 {
375 .procname = "netdev_max_backlog",
376 .data = &netdev_max_backlog,
377 .maxlen = sizeof(int),
378 .mode = 0644,
379 .proc_handler = proc_dointvec
380 },
381 {
382 .procname = "netdev_rss_key",
383 .data = &netdev_rss_key,
384 .maxlen = sizeof(int),
385 .mode = 0444,
386 .proc_handler = proc_do_rss_key,
387 },
388 #ifdef CONFIG_BPF_JIT
389 {
390 .procname = "bpf_jit_enable",
391 .data = &bpf_jit_enable,
392 .maxlen = sizeof(int),
393 .mode = 0644,
394 .proc_handler = proc_dointvec_minmax_bpf_enable,
395 # ifdef CONFIG_BPF_JIT_ALWAYS_ON
396 .extra1 = SYSCTL_ONE,
397 .extra2 = SYSCTL_ONE,
398 # else
399 .extra1 = SYSCTL_ZERO,
400 .extra2 = &two,
401 # endif
402 },
403 # ifdef CONFIG_HAVE_EBPF_JIT
404 {
405 .procname = "bpf_jit_harden",
406 .data = &bpf_jit_harden,
407 .maxlen = sizeof(int),
408 .mode = 0600,
409 .proc_handler = proc_dointvec_minmax_bpf_restricted,
410 .extra1 = SYSCTL_ZERO,
411 .extra2 = &two,
412 },
413 {
414 .procname = "bpf_jit_kallsyms",
415 .data = &bpf_jit_kallsyms,
416 .maxlen = sizeof(int),
417 .mode = 0600,
418 .proc_handler = proc_dointvec_minmax_bpf_restricted,
419 .extra1 = SYSCTL_ZERO,
420 .extra2 = SYSCTL_ONE,
421 },
422 # endif
423 {
424 .procname = "bpf_jit_limit",
425 .data = &bpf_jit_limit,
426 .maxlen = sizeof(long),
427 .mode = 0600,
428 .proc_handler = proc_dolongvec_minmax_bpf_restricted,
429 .extra1 = &long_one,
430 .extra2 = &bpf_jit_limit_max,
431 },
432 #endif
433 {
434 .procname = "netdev_tstamp_prequeue",
435 .data = &netdev_tstamp_prequeue,
436 .maxlen = sizeof(int),
437 .mode = 0644,
438 .proc_handler = proc_dointvec
439 },
440 {
441 .procname = "message_cost",
442 .data = &net_ratelimit_state.interval,
443 .maxlen = sizeof(int),
444 .mode = 0644,
445 .proc_handler = proc_dointvec_jiffies,
446 },
447 {
448 .procname = "message_burst",
449 .data = &net_ratelimit_state.burst,
450 .maxlen = sizeof(int),
451 .mode = 0644,
452 .proc_handler = proc_dointvec,
453 },
454 {
455 .procname = "optmem_max",
456 .data = &sysctl_optmem_max,
457 .maxlen = sizeof(int),
458 .mode = 0644,
459 .proc_handler = proc_dointvec
460 },
461 {
462 .procname = "tstamp_allow_data",
463 .data = &sysctl_tstamp_allow_data,
464 .maxlen = sizeof(int),
465 .mode = 0644,
466 .proc_handler = proc_dointvec_minmax,
467 .extra1 = SYSCTL_ZERO,
468 .extra2 = SYSCTL_ONE
469 },
470 #ifdef CONFIG_RPS
471 {
472 .procname = "rps_sock_flow_entries",
473 .maxlen = sizeof(int),
474 .mode = 0644,
475 .proc_handler = rps_sock_flow_sysctl
476 },
477 #endif
478 #ifdef CONFIG_NET_FLOW_LIMIT
479 {
480 .procname = "flow_limit_cpu_bitmap",
481 .mode = 0644,
482 .proc_handler = flow_limit_cpu_sysctl
483 },
484 {
485 .procname = "flow_limit_table_len",
486 .data = &netdev_flow_limit_table_len,
487 .maxlen = sizeof(int),
488 .mode = 0644,
489 .proc_handler = flow_limit_table_len_sysctl
490 },
491 #endif /* CONFIG_NET_FLOW_LIMIT */
492 #ifdef CONFIG_NET_RX_BUSY_POLL
493 {
494 .procname = "busy_poll",
495 .data = &sysctl_net_busy_poll,
496 .maxlen = sizeof(unsigned int),
497 .mode = 0644,
498 .proc_handler = proc_dointvec_minmax,
499 .extra1 = SYSCTL_ZERO,
500 },
501 {
502 .procname = "busy_read",
503 .data = &sysctl_net_busy_read,
504 .maxlen = sizeof(unsigned int),
505 .mode = 0644,
506 .proc_handler = proc_dointvec_minmax,
507 .extra1 = SYSCTL_ZERO,
508 },
509 #endif
510 #ifdef CONFIG_NET_SCHED
511 {
512 .procname = "default_qdisc",
513 .mode = 0644,
514 .maxlen = IFNAMSIZ,
515 .proc_handler = set_default_qdisc
516 },
517 #endif
518 #endif /* CONFIG_NET */
519 {
520 .procname = "netdev_budget",
521 .data = &netdev_budget,
522 .maxlen = sizeof(int),
523 .mode = 0644,
524 .proc_handler = proc_dointvec
525 },
526 {
527 .procname = "warnings",
528 .data = &net_msg_warn,
529 .maxlen = sizeof(int),
530 .mode = 0644,
531 .proc_handler = proc_dointvec
532 },
533 {
534 .procname = "max_skb_frags",
535 .data = &sysctl_max_skb_frags,
536 .maxlen = sizeof(int),
537 .mode = 0644,
538 .proc_handler = proc_dointvec_minmax,
539 .extra1 = SYSCTL_ONE,
540 .extra2 = &max_skb_frags,
541 },
542 {
543 .procname = "netdev_budget_usecs",
544 .data = &netdev_budget_usecs,
545 .maxlen = sizeof(unsigned int),
546 .mode = 0644,
547 .proc_handler = proc_dointvec_minmax,
548 .extra1 = SYSCTL_ZERO,
549 },
550 {
551 .procname = "fb_tunnels_only_for_init_net",
552 .data = &sysctl_fb_tunnels_only_for_init_net,
553 .maxlen = sizeof(int),
554 .mode = 0644,
555 .proc_handler = proc_dointvec_minmax,
556 .extra1 = SYSCTL_ZERO,
557 .extra2 = SYSCTL_ONE,
558 },
559 {
560 .procname = "devconf_inherit_init_net",
561 .data = &sysctl_devconf_inherit_init_net,
562 .maxlen = sizeof(int),
563 .mode = 0644,
564 .proc_handler = proc_dointvec_minmax,
565 .extra1 = SYSCTL_ZERO,
566 .extra2 = &two,
567 },
568 {
569 .procname = "high_order_alloc_disable",
570 .data = &net_high_order_alloc_disable_key.key,
571 .maxlen = sizeof(net_high_order_alloc_disable_key),
572 .mode = 0644,
573 .proc_handler = proc_do_static_key,
574 },
575 {
576 .procname = "gro_normal_batch",
577 .data = &gro_normal_batch,
578 .maxlen = sizeof(unsigned int),
579 .mode = 0644,
580 .proc_handler = proc_dointvec_minmax,
581 .extra1 = SYSCTL_ONE,
582 },
583 { }
584 };
585
586 static struct ctl_table netns_core_table[] = {
587 {
588 .procname = "somaxconn",
589 .data = &init_net.core.sysctl_somaxconn,
590 .maxlen = sizeof(int),
591 .mode = 0644,
592 .extra1 = SYSCTL_ZERO,
593 .proc_handler = proc_dointvec_minmax
594 },
595 { }
596 };
597
sysctl_core_net_init(struct net * net)598 static __net_init int sysctl_core_net_init(struct net *net)
599 {
600 struct ctl_table *tbl;
601
602 tbl = netns_core_table;
603 if (!net_eq(net, &init_net)) {
604 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
605 if (tbl == NULL)
606 goto err_dup;
607
608 tbl[0].data = &net->core.sysctl_somaxconn;
609
610 /* Don't export any sysctls to unprivileged users */
611 if (net->user_ns != &init_user_ns) {
612 tbl[0].procname = NULL;
613 }
614 }
615
616 net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
617 if (net->core.sysctl_hdr == NULL)
618 goto err_reg;
619
620 return 0;
621
622 err_reg:
623 if (tbl != netns_core_table)
624 kfree(tbl);
625 err_dup:
626 return -ENOMEM;
627 }
628
sysctl_core_net_exit(struct net * net)629 static __net_exit void sysctl_core_net_exit(struct net *net)
630 {
631 struct ctl_table *tbl;
632
633 tbl = net->core.sysctl_hdr->ctl_table_arg;
634 unregister_net_sysctl_table(net->core.sysctl_hdr);
635 BUG_ON(tbl == netns_core_table);
636 kfree(tbl);
637 }
638
639 static __net_initdata struct pernet_operations sysctl_core_ops = {
640 .init = sysctl_core_net_init,
641 .exit = sysctl_core_net_exit,
642 };
643
sysctl_core_init(void)644 static __init int sysctl_core_init(void)
645 {
646 register_net_sysctl(&init_net, "net/core", net_core_table);
647 return register_pernet_subsys(&sysctl_core_ops);
648 }
649
650 fs_initcall(sysctl_core_init);
651