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
rps_sock_flow_sysctl(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)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
flow_limit_cpu_sysctl(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)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
flow_limit_table_len_sysctl(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)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
set_default_qdisc(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)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
proc_do_dev_weight(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)235 static int proc_do_dev_weight(struct ctl_table *table, int write,
236 void *buffer, size_t *lenp, loff_t *ppos)
237 {
238 static DEFINE_MUTEX(dev_weight_mutex);
239 int ret, weight;
240
241 mutex_lock(&dev_weight_mutex);
242 ret = proc_dointvec(table, write, buffer, lenp, ppos);
243 if (!ret && write) {
244 weight = READ_ONCE(weight_p);
245 WRITE_ONCE(dev_rx_weight, weight * dev_weight_rx_bias);
246 WRITE_ONCE(dev_tx_weight, weight * dev_weight_tx_bias);
247 }
248 mutex_unlock(&dev_weight_mutex);
249
250 return ret;
251 }
252
proc_do_rss_key(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)253 static int proc_do_rss_key(struct ctl_table *table, int write,
254 void *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
proc_dointvec_minmax_bpf_enable(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)266 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write,
267 void *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(current_cred()))) {
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 # ifdef CONFIG_HAVE_EBPF_JIT
292 static int
proc_dointvec_minmax_bpf_restricted(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)293 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write,
294 void *buffer, size_t *lenp, 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 # endif /* CONFIG_HAVE_EBPF_JIT */
302
303 static int
proc_dolongvec_minmax_bpf_restricted(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)304 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
305 void *buffer, size_t *lenp, 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 = &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 #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 = &two,
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 = &three,
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 .procname = "gro_normal_batch",
572 .data = &gro_normal_batch,
573 .maxlen = sizeof(unsigned int),
574 .mode = 0644,
575 .proc_handler = proc_dointvec_minmax,
576 .extra1 = SYSCTL_ONE,
577 },
578 { }
579 };
580
581 static struct ctl_table netns_core_table[] = {
582 {
583 .procname = "somaxconn",
584 .data = &init_net.core.sysctl_somaxconn,
585 .maxlen = sizeof(int),
586 .mode = 0644,
587 .extra1 = SYSCTL_ZERO,
588 .proc_handler = proc_dointvec_minmax
589 },
590 { }
591 };
592
fb_tunnels_only_for_init_net_sysctl_setup(char * str)593 static int __init fb_tunnels_only_for_init_net_sysctl_setup(char *str)
594 {
595 /* fallback tunnels for initns only */
596 if (!strncmp(str, "initns", 6))
597 sysctl_fb_tunnels_only_for_init_net = 1;
598 /* no fallback tunnels anywhere */
599 else if (!strncmp(str, "none", 4))
600 sysctl_fb_tunnels_only_for_init_net = 2;
601
602 return 1;
603 }
604 __setup("fb_tunnels=", fb_tunnels_only_for_init_net_sysctl_setup);
605
sysctl_core_net_init(struct net * net)606 static __net_init int sysctl_core_net_init(struct net *net)
607 {
608 struct ctl_table *tbl;
609
610 tbl = netns_core_table;
611 if (!net_eq(net, &init_net)) {
612 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
613 if (tbl == NULL)
614 goto err_dup;
615
616 tbl[0].data = &net->core.sysctl_somaxconn;
617
618 /* Don't export any sysctls to unprivileged users */
619 if (net->user_ns != &init_user_ns) {
620 tbl[0].procname = NULL;
621 }
622 }
623
624 net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
625 if (net->core.sysctl_hdr == NULL)
626 goto err_reg;
627
628 return 0;
629
630 err_reg:
631 if (tbl != netns_core_table)
632 kfree(tbl);
633 err_dup:
634 return -ENOMEM;
635 }
636
sysctl_core_net_exit(struct net * net)637 static __net_exit void sysctl_core_net_exit(struct net *net)
638 {
639 struct ctl_table *tbl;
640
641 tbl = net->core.sysctl_hdr->ctl_table_arg;
642 unregister_net_sysctl_table(net->core.sysctl_hdr);
643 BUG_ON(tbl == netns_core_table);
644 kfree(tbl);
645 }
646
647 static __net_initdata struct pernet_operations sysctl_core_ops = {
648 .init = sysctl_core_net_init,
649 .exit = sysctl_core_net_exit,
650 };
651
sysctl_core_init(void)652 static __init int sysctl_core_init(void)
653 {
654 register_net_sysctl(&init_net, "net/core", net_core_table);
655 return register_pernet_subsys(&sysctl_core_ops);
656 }
657
658 fs_initcall(sysctl_core_init);
659