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 int_3600 = 3600;
28 static int min_sndbuf = SOCK_MIN_SNDBUF;
29 static int min_rcvbuf = SOCK_MIN_RCVBUF;
30 static int max_skb_frags = MAX_SKB_FRAGS;
31 static long long_one __maybe_unused = 1;
32 static long long_max __maybe_unused = LONG_MAX;
33
34 static int net_msg_warn; /* Unused, but still a sysctl */
35
36 int sysctl_fb_tunnels_only_for_init_net __read_mostly = 0;
37 EXPORT_SYMBOL(sysctl_fb_tunnels_only_for_init_net);
38
39 /* 0 - Keep current behavior:
40 * IPv4: inherit all current settings from init_net
41 * IPv6: reset all settings to default
42 * 1 - Both inherit all current settings from init_net
43 * 2 - Both reset all settings to default
44 * 3 - Both inherit all settings from current netns
45 */
46 int sysctl_devconf_inherit_init_net __read_mostly;
47 EXPORT_SYMBOL(sysctl_devconf_inherit_init_net);
48
49 #ifdef CONFIG_RPS
rps_sock_flow_sysctl(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)50 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
51 void *buffer, size_t *lenp, loff_t *ppos)
52 {
53 unsigned int orig_size, size;
54 int ret, i;
55 struct ctl_table tmp = {
56 .data = &size,
57 .maxlen = sizeof(size),
58 .mode = table->mode
59 };
60 struct rps_sock_flow_table *orig_sock_table, *sock_table;
61 static DEFINE_MUTEX(sock_flow_mutex);
62
63 mutex_lock(&sock_flow_mutex);
64
65 orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
66 lockdep_is_held(&sock_flow_mutex));
67 size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
68
69 ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
70
71 if (write) {
72 if (size) {
73 if (size > 1<<29) {
74 /* Enforce limit to prevent overflow */
75 mutex_unlock(&sock_flow_mutex);
76 return -EINVAL;
77 }
78 size = roundup_pow_of_two(size);
79 if (size != orig_size) {
80 sock_table =
81 vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
82 if (!sock_table) {
83 mutex_unlock(&sock_flow_mutex);
84 return -ENOMEM;
85 }
86 rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
87 sock_table->mask = size - 1;
88 } else
89 sock_table = orig_sock_table;
90
91 for (i = 0; i < size; i++)
92 sock_table->ents[i] = RPS_NO_CPU;
93 } else
94 sock_table = NULL;
95
96 if (sock_table != orig_sock_table) {
97 rcu_assign_pointer(rps_sock_flow_table, sock_table);
98 if (sock_table) {
99 static_branch_inc(&rps_needed);
100 static_branch_inc(&rfs_needed);
101 }
102 if (orig_sock_table) {
103 static_branch_dec(&rps_needed);
104 static_branch_dec(&rfs_needed);
105 synchronize_rcu();
106 vfree(orig_sock_table);
107 }
108 }
109 }
110
111 mutex_unlock(&sock_flow_mutex);
112
113 return ret;
114 }
115 #endif /* CONFIG_RPS */
116
117 #ifdef CONFIG_NET_FLOW_LIMIT
118 static DEFINE_MUTEX(flow_limit_update_mutex);
119
flow_limit_cpu_sysctl(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)120 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
121 void *buffer, size_t *lenp, loff_t *ppos)
122 {
123 struct sd_flow_limit *cur;
124 struct softnet_data *sd;
125 cpumask_var_t mask;
126 int i, len, ret = 0;
127
128 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
129 return -ENOMEM;
130
131 if (write) {
132 ret = cpumask_parse(buffer, mask);
133 if (ret)
134 goto done;
135
136 mutex_lock(&flow_limit_update_mutex);
137 len = sizeof(*cur) + netdev_flow_limit_table_len;
138 for_each_possible_cpu(i) {
139 sd = &per_cpu(softnet_data, i);
140 cur = rcu_dereference_protected(sd->flow_limit,
141 lockdep_is_held(&flow_limit_update_mutex));
142 if (cur && !cpumask_test_cpu(i, mask)) {
143 RCU_INIT_POINTER(sd->flow_limit, NULL);
144 synchronize_rcu();
145 kfree(cur);
146 } else if (!cur && cpumask_test_cpu(i, mask)) {
147 cur = kzalloc_node(len, GFP_KERNEL,
148 cpu_to_node(i));
149 if (!cur) {
150 /* not unwinding previous changes */
151 ret = -ENOMEM;
152 goto write_unlock;
153 }
154 cur->num_buckets = netdev_flow_limit_table_len;
155 rcu_assign_pointer(sd->flow_limit, cur);
156 }
157 }
158 write_unlock:
159 mutex_unlock(&flow_limit_update_mutex);
160 } else {
161 char kbuf[128];
162
163 if (*ppos || !*lenp) {
164 *lenp = 0;
165 goto done;
166 }
167
168 cpumask_clear(mask);
169 rcu_read_lock();
170 for_each_possible_cpu(i) {
171 sd = &per_cpu(softnet_data, i);
172 if (rcu_dereference(sd->flow_limit))
173 cpumask_set_cpu(i, mask);
174 }
175 rcu_read_unlock();
176
177 len = min(sizeof(kbuf) - 1, *lenp);
178 len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
179 if (!len) {
180 *lenp = 0;
181 goto done;
182 }
183 if (len < *lenp)
184 kbuf[len++] = '\n';
185 memcpy(buffer, kbuf, len);
186 *lenp = len;
187 *ppos += len;
188 }
189
190 done:
191 free_cpumask_var(mask);
192 return ret;
193 }
194
flow_limit_table_len_sysctl(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)195 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
196 void *buffer, size_t *lenp, loff_t *ppos)
197 {
198 unsigned int old, *ptr;
199 int ret;
200
201 mutex_lock(&flow_limit_update_mutex);
202
203 ptr = table->data;
204 old = *ptr;
205 ret = proc_dointvec(table, write, buffer, lenp, ppos);
206 if (!ret && write && !is_power_of_2(*ptr)) {
207 *ptr = old;
208 ret = -EINVAL;
209 }
210
211 mutex_unlock(&flow_limit_update_mutex);
212 return ret;
213 }
214 #endif /* CONFIG_NET_FLOW_LIMIT */
215
216 #ifdef CONFIG_NET_SCHED
set_default_qdisc(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)217 static int set_default_qdisc(struct ctl_table *table, int write,
218 void *buffer, size_t *lenp, loff_t *ppos)
219 {
220 char id[IFNAMSIZ];
221 struct ctl_table tbl = {
222 .data = id,
223 .maxlen = IFNAMSIZ,
224 };
225 int ret;
226
227 qdisc_get_default(id, IFNAMSIZ);
228
229 ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
230 if (write && ret == 0)
231 ret = qdisc_set_default(id);
232 return ret;
233 }
234 #endif
235
proc_do_dev_weight(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)236 static int proc_do_dev_weight(struct ctl_table *table, int write,
237 void *buffer, size_t *lenp, loff_t *ppos)
238 {
239 static DEFINE_MUTEX(dev_weight_mutex);
240 int ret, weight;
241
242 mutex_lock(&dev_weight_mutex);
243 ret = proc_dointvec(table, write, buffer, lenp, ppos);
244 if (!ret && write) {
245 weight = READ_ONCE(weight_p);
246 WRITE_ONCE(dev_rx_weight, weight * dev_weight_rx_bias);
247 WRITE_ONCE(dev_tx_weight, weight * dev_weight_tx_bias);
248 }
249 mutex_unlock(&dev_weight_mutex);
250
251 return ret;
252 }
253
proc_do_rss_key(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)254 static int proc_do_rss_key(struct ctl_table *table, int write,
255 void *buffer, size_t *lenp, loff_t *ppos)
256 {
257 struct ctl_table fake_table;
258 char buf[NETDEV_RSS_KEY_LEN * 3];
259
260 snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
261 fake_table.data = buf;
262 fake_table.maxlen = sizeof(buf);
263 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
264 }
265
266 #ifdef CONFIG_BPF_JIT
proc_dointvec_minmax_bpf_enable(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)267 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write,
268 void *buffer, size_t *lenp,
269 loff_t *ppos)
270 {
271 int ret, jit_enable = *(int *)table->data;
272 struct ctl_table tmp = *table;
273
274 if (write && !capable(CAP_SYS_ADMIN))
275 return -EPERM;
276
277 tmp.data = &jit_enable;
278 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
279 if (write && !ret) {
280 if (jit_enable < 2 ||
281 (jit_enable == 2 && bpf_dump_raw_ok(current_cred()))) {
282 *(int *)table->data = jit_enable;
283 if (jit_enable == 2)
284 pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
285 } else {
286 ret = -EPERM;
287 }
288 }
289 return ret;
290 }
291
292 # ifdef CONFIG_HAVE_EBPF_JIT
293 static int
proc_dointvec_minmax_bpf_restricted(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)294 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write,
295 void *buffer, size_t *lenp, loff_t *ppos)
296 {
297 if (!capable(CAP_SYS_ADMIN))
298 return -EPERM;
299
300 return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
301 }
302 # endif /* CONFIG_HAVE_EBPF_JIT */
303
304 static int
proc_dolongvec_minmax_bpf_restricted(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)305 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
306 void *buffer, size_t *lenp, loff_t *ppos)
307 {
308 if (!capable(CAP_SYS_ADMIN))
309 return -EPERM;
310
311 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
312 }
313 #endif
314
315 static struct ctl_table net_core_table[] = {
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 = &bpf_jit_limit_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 {
514 .procname = "netdev_budget",
515 .data = &netdev_budget,
516 .maxlen = sizeof(int),
517 .mode = 0644,
518 .proc_handler = proc_dointvec
519 },
520 {
521 .procname = "warnings",
522 .data = &net_msg_warn,
523 .maxlen = sizeof(int),
524 .mode = 0644,
525 .proc_handler = proc_dointvec
526 },
527 {
528 .procname = "max_skb_frags",
529 .data = &sysctl_max_skb_frags,
530 .maxlen = sizeof(int),
531 .mode = 0644,
532 .proc_handler = proc_dointvec_minmax,
533 .extra1 = SYSCTL_ONE,
534 .extra2 = &max_skb_frags,
535 },
536 {
537 .procname = "netdev_budget_usecs",
538 .data = &netdev_budget_usecs,
539 .maxlen = sizeof(unsigned int),
540 .mode = 0644,
541 .proc_handler = proc_dointvec_minmax,
542 .extra1 = SYSCTL_ZERO,
543 },
544 {
545 .procname = "fb_tunnels_only_for_init_net",
546 .data = &sysctl_fb_tunnels_only_for_init_net,
547 .maxlen = sizeof(int),
548 .mode = 0644,
549 .proc_handler = proc_dointvec_minmax,
550 .extra1 = SYSCTL_ZERO,
551 .extra2 = &two,
552 },
553 {
554 .procname = "devconf_inherit_init_net",
555 .data = &sysctl_devconf_inherit_init_net,
556 .maxlen = sizeof(int),
557 .mode = 0644,
558 .proc_handler = proc_dointvec_minmax,
559 .extra1 = SYSCTL_ZERO,
560 .extra2 = &three,
561 },
562 {
563 .procname = "high_order_alloc_disable",
564 .data = &net_high_order_alloc_disable_key.key,
565 .maxlen = sizeof(net_high_order_alloc_disable_key),
566 .mode = 0644,
567 .proc_handler = proc_do_static_key,
568 },
569 {
570 .procname = "gro_normal_batch",
571 .data = &gro_normal_batch,
572 .maxlen = sizeof(unsigned int),
573 .mode = 0644,
574 .proc_handler = proc_dointvec_minmax,
575 .extra1 = SYSCTL_ONE,
576 },
577 {
578 .procname = "netdev_unregister_timeout_secs",
579 .data = &netdev_unregister_timeout_secs,
580 .maxlen = sizeof(unsigned int),
581 .mode = 0644,
582 .proc_handler = proc_dointvec_minmax,
583 .extra1 = SYSCTL_ONE,
584 .extra2 = &int_3600,
585 },
586 { }
587 };
588
589 static struct ctl_table netns_core_table[] = {
590 {
591 .procname = "somaxconn",
592 .data = &init_net.core.sysctl_somaxconn,
593 .maxlen = sizeof(int),
594 .mode = 0644,
595 .extra1 = SYSCTL_ZERO,
596 .proc_handler = proc_dointvec_minmax
597 },
598 { }
599 };
600
fb_tunnels_only_for_init_net_sysctl_setup(char * str)601 static int __init fb_tunnels_only_for_init_net_sysctl_setup(char *str)
602 {
603 /* fallback tunnels for initns only */
604 if (!strncmp(str, "initns", 6))
605 sysctl_fb_tunnels_only_for_init_net = 1;
606 /* no fallback tunnels anywhere */
607 else if (!strncmp(str, "none", 4))
608 sysctl_fb_tunnels_only_for_init_net = 2;
609
610 return 1;
611 }
612 __setup("fb_tunnels=", fb_tunnels_only_for_init_net_sysctl_setup);
613
sysctl_core_net_init(struct net * net)614 static __net_init int sysctl_core_net_init(struct net *net)
615 {
616 struct ctl_table *tbl;
617
618 tbl = netns_core_table;
619 if (!net_eq(net, &init_net)) {
620 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
621 if (tbl == NULL)
622 goto err_dup;
623
624 tbl[0].data = &net->core.sysctl_somaxconn;
625
626 /* Don't export any sysctls to unprivileged users */
627 if (net->user_ns != &init_user_ns) {
628 tbl[0].procname = NULL;
629 }
630 }
631
632 net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
633 if (net->core.sysctl_hdr == NULL)
634 goto err_reg;
635
636 return 0;
637
638 err_reg:
639 if (tbl != netns_core_table)
640 kfree(tbl);
641 err_dup:
642 return -ENOMEM;
643 }
644
sysctl_core_net_exit(struct net * net)645 static __net_exit void sysctl_core_net_exit(struct net *net)
646 {
647 struct ctl_table *tbl;
648
649 tbl = net->core.sysctl_hdr->ctl_table_arg;
650 unregister_net_sysctl_table(net->core.sysctl_hdr);
651 BUG_ON(tbl == netns_core_table);
652 kfree(tbl);
653 }
654
655 static __net_initdata struct pernet_operations sysctl_core_ops = {
656 .init = sysctl_core_net_init,
657 .exit = sysctl_core_net_exit,
658 };
659
sysctl_core_init(void)660 static __init int sysctl_core_init(void)
661 {
662 register_net_sysctl(&init_net, "net/core", net_core_table);
663 return register_pernet_subsys(&sysctl_core_ops);
664 }
665
666 fs_initcall(sysctl_core_init);
667