Loading...
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
25static int two __maybe_unused = 2;
26static int three = 3;
27static int min_sndbuf = SOCK_MIN_SNDBUF;
28static int min_rcvbuf = SOCK_MIN_RCVBUF;
29static int max_skb_frags = MAX_SKB_FRAGS;
30static long long_one __maybe_unused = 1;
31static long long_max __maybe_unused = LONG_MAX;
32
33static int net_msg_warn; /* Unused, but still a sysctl */
34
35int sysctl_fb_tunnels_only_for_init_net __read_mostly = 0;
36EXPORT_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 */
45int sysctl_devconf_inherit_init_net __read_mostly;
46EXPORT_SYMBOL(sysctl_devconf_inherit_init_net);
47
48#ifdef CONFIG_RPS
49static 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
117static DEFINE_MUTEX(flow_limit_update_mutex);
118
119static 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 }
157write_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
189done:
190 free_cpumask_var(mask);
191 return ret;
192}
193
194static 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
216static 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
235static 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
250static 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
263static 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
289static int
290proc_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
300static int
301proc_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
311static 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 = SYSCTL_ONE,
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
578static 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
590static __net_init int sysctl_core_net_init(struct net *net)
591{
592 struct ctl_table *tbl;
593
594 tbl = netns_core_table;
595 if (!net_eq(net, &init_net)) {
596 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
597 if (tbl == NULL)
598 goto err_dup;
599
600 tbl[0].data = &net->core.sysctl_somaxconn;
601
602 /* Don't export any sysctls to unprivileged users */
603 if (net->user_ns != &init_user_ns) {
604 tbl[0].procname = NULL;
605 }
606 }
607
608 net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
609 if (net->core.sysctl_hdr == NULL)
610 goto err_reg;
611
612 return 0;
613
614err_reg:
615 if (tbl != netns_core_table)
616 kfree(tbl);
617err_dup:
618 return -ENOMEM;
619}
620
621static __net_exit void sysctl_core_net_exit(struct net *net)
622{
623 struct ctl_table *tbl;
624
625 tbl = net->core.sysctl_hdr->ctl_table_arg;
626 unregister_net_sysctl_table(net->core.sysctl_hdr);
627 BUG_ON(tbl == netns_core_table);
628 kfree(tbl);
629}
630
631static __net_initdata struct pernet_operations sysctl_core_ops = {
632 .init = sysctl_core_net_init,
633 .exit = sysctl_core_net_exit,
634};
635
636static __init int sysctl_core_init(void)
637{
638 register_net_sysctl(&init_net, "net/core", net_core_table);
639 return register_pernet_subsys(&sysctl_core_ops);
640}
641
642fs_initcall(sysctl_core_init);
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/filter.h>
10#include <linux/mm.h>
11#include <linux/sysctl.h>
12#include <linux/module.h>
13#include <linux/socket.h>
14#include <linux/netdevice.h>
15#include <linux/ratelimit.h>
16#include <linux/vmalloc.h>
17#include <linux/init.h>
18#include <linux/slab.h>
19#include <linux/sched/isolation.h>
20
21#include <net/ip.h>
22#include <net/sock.h>
23#include <net/net_ratelimit.h>
24#include <net/busy_poll.h>
25#include <net/pkt_sched.h>
26#include <net/hotdata.h>
27#include <net/rps.h>
28
29#include "dev.h"
30
31static int int_3600 = 3600;
32static int min_sndbuf = SOCK_MIN_SNDBUF;
33static int min_rcvbuf = SOCK_MIN_RCVBUF;
34static int max_skb_frags = MAX_SKB_FRAGS;
35static int min_mem_pcpu_rsv = SK_MEMORY_PCPU_RESERVE;
36
37static int net_msg_warn; /* Unused, but still a sysctl */
38
39int sysctl_fb_tunnels_only_for_init_net __read_mostly = 0;
40EXPORT_SYMBOL(sysctl_fb_tunnels_only_for_init_net);
41
42/* 0 - Keep current behavior:
43 * IPv4: inherit all current settings from init_net
44 * IPv6: reset all settings to default
45 * 1 - Both inherit all current settings from init_net
46 * 2 - Both reset all settings to default
47 * 3 - Both inherit all settings from current netns
48 */
49int sysctl_devconf_inherit_init_net __read_mostly;
50EXPORT_SYMBOL(sysctl_devconf_inherit_init_net);
51
52#if IS_ENABLED(CONFIG_NET_FLOW_LIMIT) || IS_ENABLED(CONFIG_RPS)
53static void dump_cpumask(void *buffer, size_t *lenp, loff_t *ppos,
54 struct cpumask *mask)
55{
56 char kbuf[128];
57 int len;
58
59 if (*ppos || !*lenp) {
60 *lenp = 0;
61 return;
62 }
63
64 len = min(sizeof(kbuf) - 1, *lenp);
65 len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
66 if (!len) {
67 *lenp = 0;
68 return;
69 }
70
71 if (len < *lenp)
72 kbuf[len++] = '\n';
73 memcpy(buffer, kbuf, len);
74 *lenp = len;
75 *ppos += len;
76}
77#endif
78
79#ifdef CONFIG_RPS
80
81static struct cpumask *rps_default_mask_cow_alloc(struct net *net)
82{
83 struct cpumask *rps_default_mask;
84
85 if (net->core.rps_default_mask)
86 return net->core.rps_default_mask;
87
88 rps_default_mask = kzalloc(cpumask_size(), GFP_KERNEL);
89 if (!rps_default_mask)
90 return NULL;
91
92 /* pairs with READ_ONCE in rx_queue_default_mask() */
93 WRITE_ONCE(net->core.rps_default_mask, rps_default_mask);
94 return rps_default_mask;
95}
96
97static int rps_default_mask_sysctl(struct ctl_table *table, int write,
98 void *buffer, size_t *lenp, loff_t *ppos)
99{
100 struct net *net = (struct net *)table->data;
101 int err = 0;
102
103 rtnl_lock();
104 if (write) {
105 struct cpumask *rps_default_mask = rps_default_mask_cow_alloc(net);
106
107 err = -ENOMEM;
108 if (!rps_default_mask)
109 goto done;
110
111 err = cpumask_parse(buffer, rps_default_mask);
112 if (err)
113 goto done;
114
115 err = rps_cpumask_housekeeping(rps_default_mask);
116 if (err)
117 goto done;
118 } else {
119 dump_cpumask(buffer, lenp, ppos,
120 net->core.rps_default_mask ? : cpu_none_mask);
121 }
122
123done:
124 rtnl_unlock();
125 return err;
126}
127
128static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
129 void *buffer, size_t *lenp, loff_t *ppos)
130{
131 unsigned int orig_size, size;
132 int ret, i;
133 struct ctl_table tmp = {
134 .data = &size,
135 .maxlen = sizeof(size),
136 .mode = table->mode
137 };
138 struct rps_sock_flow_table *orig_sock_table, *sock_table;
139 static DEFINE_MUTEX(sock_flow_mutex);
140
141 mutex_lock(&sock_flow_mutex);
142
143 orig_sock_table = rcu_dereference_protected(
144 net_hotdata.rps_sock_flow_table,
145 lockdep_is_held(&sock_flow_mutex));
146 size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
147
148 ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
149
150 if (write) {
151 if (size) {
152 if (size > 1<<29) {
153 /* Enforce limit to prevent overflow */
154 mutex_unlock(&sock_flow_mutex);
155 return -EINVAL;
156 }
157 size = roundup_pow_of_two(size);
158 if (size != orig_size) {
159 sock_table =
160 vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
161 if (!sock_table) {
162 mutex_unlock(&sock_flow_mutex);
163 return -ENOMEM;
164 }
165 net_hotdata.rps_cpu_mask =
166 roundup_pow_of_two(nr_cpu_ids) - 1;
167 sock_table->mask = size - 1;
168 } else
169 sock_table = orig_sock_table;
170
171 for (i = 0; i < size; i++)
172 sock_table->ents[i] = RPS_NO_CPU;
173 } else
174 sock_table = NULL;
175
176 if (sock_table != orig_sock_table) {
177 rcu_assign_pointer(net_hotdata.rps_sock_flow_table,
178 sock_table);
179 if (sock_table) {
180 static_branch_inc(&rps_needed);
181 static_branch_inc(&rfs_needed);
182 }
183 if (orig_sock_table) {
184 static_branch_dec(&rps_needed);
185 static_branch_dec(&rfs_needed);
186 kvfree_rcu_mightsleep(orig_sock_table);
187 }
188 }
189 }
190
191 mutex_unlock(&sock_flow_mutex);
192
193 return ret;
194}
195#endif /* CONFIG_RPS */
196
197#ifdef CONFIG_NET_FLOW_LIMIT
198static DEFINE_MUTEX(flow_limit_update_mutex);
199
200static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
201 void *buffer, size_t *lenp, loff_t *ppos)
202{
203 struct sd_flow_limit *cur;
204 struct softnet_data *sd;
205 cpumask_var_t mask;
206 int i, len, ret = 0;
207
208 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
209 return -ENOMEM;
210
211 if (write) {
212 ret = cpumask_parse(buffer, mask);
213 if (ret)
214 goto done;
215
216 mutex_lock(&flow_limit_update_mutex);
217 len = sizeof(*cur) + netdev_flow_limit_table_len;
218 for_each_possible_cpu(i) {
219 sd = &per_cpu(softnet_data, i);
220 cur = rcu_dereference_protected(sd->flow_limit,
221 lockdep_is_held(&flow_limit_update_mutex));
222 if (cur && !cpumask_test_cpu(i, mask)) {
223 RCU_INIT_POINTER(sd->flow_limit, NULL);
224 kfree_rcu_mightsleep(cur);
225 } else if (!cur && cpumask_test_cpu(i, mask)) {
226 cur = kzalloc_node(len, GFP_KERNEL,
227 cpu_to_node(i));
228 if (!cur) {
229 /* not unwinding previous changes */
230 ret = -ENOMEM;
231 goto write_unlock;
232 }
233 cur->num_buckets = netdev_flow_limit_table_len;
234 rcu_assign_pointer(sd->flow_limit, cur);
235 }
236 }
237write_unlock:
238 mutex_unlock(&flow_limit_update_mutex);
239 } else {
240 cpumask_clear(mask);
241 rcu_read_lock();
242 for_each_possible_cpu(i) {
243 sd = &per_cpu(softnet_data, i);
244 if (rcu_dereference(sd->flow_limit))
245 cpumask_set_cpu(i, mask);
246 }
247 rcu_read_unlock();
248
249 dump_cpumask(buffer, lenp, ppos, mask);
250 }
251
252done:
253 free_cpumask_var(mask);
254 return ret;
255}
256
257static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
258 void *buffer, size_t *lenp, loff_t *ppos)
259{
260 unsigned int old, *ptr;
261 int ret;
262
263 mutex_lock(&flow_limit_update_mutex);
264
265 ptr = table->data;
266 old = *ptr;
267 ret = proc_dointvec(table, write, buffer, lenp, ppos);
268 if (!ret && write && !is_power_of_2(*ptr)) {
269 *ptr = old;
270 ret = -EINVAL;
271 }
272
273 mutex_unlock(&flow_limit_update_mutex);
274 return ret;
275}
276#endif /* CONFIG_NET_FLOW_LIMIT */
277
278#ifdef CONFIG_NET_SCHED
279static int set_default_qdisc(struct ctl_table *table, int write,
280 void *buffer, size_t *lenp, loff_t *ppos)
281{
282 char id[IFNAMSIZ];
283 struct ctl_table tbl = {
284 .data = id,
285 .maxlen = IFNAMSIZ,
286 };
287 int ret;
288
289 qdisc_get_default(id, IFNAMSIZ);
290
291 ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
292 if (write && ret == 0)
293 ret = qdisc_set_default(id);
294 return ret;
295}
296#endif
297
298static int proc_do_dev_weight(struct ctl_table *table, int write,
299 void *buffer, size_t *lenp, loff_t *ppos)
300{
301 static DEFINE_MUTEX(dev_weight_mutex);
302 int ret, weight;
303
304 mutex_lock(&dev_weight_mutex);
305 ret = proc_dointvec(table, write, buffer, lenp, ppos);
306 if (!ret && write) {
307 weight = READ_ONCE(weight_p);
308 WRITE_ONCE(net_hotdata.dev_rx_weight, weight * dev_weight_rx_bias);
309 WRITE_ONCE(net_hotdata.dev_tx_weight, weight * dev_weight_tx_bias);
310 }
311 mutex_unlock(&dev_weight_mutex);
312
313 return ret;
314}
315
316static int proc_do_rss_key(struct ctl_table *table, int write,
317 void *buffer, size_t *lenp, loff_t *ppos)
318{
319 struct ctl_table fake_table;
320 char buf[NETDEV_RSS_KEY_LEN * 3];
321
322 snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
323 fake_table.data = buf;
324 fake_table.maxlen = sizeof(buf);
325 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
326}
327
328#ifdef CONFIG_BPF_JIT
329static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write,
330 void *buffer, size_t *lenp,
331 loff_t *ppos)
332{
333 int ret, jit_enable = *(int *)table->data;
334 int min = *(int *)table->extra1;
335 int max = *(int *)table->extra2;
336 struct ctl_table tmp = *table;
337
338 if (write && !capable(CAP_SYS_ADMIN))
339 return -EPERM;
340
341 tmp.data = &jit_enable;
342 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
343 if (write && !ret) {
344 if (jit_enable < 2 ||
345 (jit_enable == 2 && bpf_dump_raw_ok(current_cred()))) {
346 *(int *)table->data = jit_enable;
347 if (jit_enable == 2)
348 pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
349 } else {
350 ret = -EPERM;
351 }
352 }
353
354 if (write && ret && min == max)
355 pr_info_once("CONFIG_BPF_JIT_ALWAYS_ON is enabled, bpf_jit_enable is permanently set to 1.\n");
356
357 return ret;
358}
359
360# ifdef CONFIG_HAVE_EBPF_JIT
361static int
362proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write,
363 void *buffer, size_t *lenp, loff_t *ppos)
364{
365 if (!capable(CAP_SYS_ADMIN))
366 return -EPERM;
367
368 return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
369}
370# endif /* CONFIG_HAVE_EBPF_JIT */
371
372static int
373proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
374 void *buffer, size_t *lenp, loff_t *ppos)
375{
376 if (!capable(CAP_SYS_ADMIN))
377 return -EPERM;
378
379 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
380}
381#endif
382
383static struct ctl_table net_core_table[] = {
384 {
385 .procname = "wmem_max",
386 .data = &sysctl_wmem_max,
387 .maxlen = sizeof(int),
388 .mode = 0644,
389 .proc_handler = proc_dointvec_minmax,
390 .extra1 = &min_sndbuf,
391 },
392 {
393 .procname = "rmem_max",
394 .data = &sysctl_rmem_max,
395 .maxlen = sizeof(int),
396 .mode = 0644,
397 .proc_handler = proc_dointvec_minmax,
398 .extra1 = &min_rcvbuf,
399 },
400 {
401 .procname = "wmem_default",
402 .data = &sysctl_wmem_default,
403 .maxlen = sizeof(int),
404 .mode = 0644,
405 .proc_handler = proc_dointvec_minmax,
406 .extra1 = &min_sndbuf,
407 },
408 {
409 .procname = "rmem_default",
410 .data = &sysctl_rmem_default,
411 .maxlen = sizeof(int),
412 .mode = 0644,
413 .proc_handler = proc_dointvec_minmax,
414 .extra1 = &min_rcvbuf,
415 },
416 {
417 .procname = "mem_pcpu_rsv",
418 .data = &sysctl_mem_pcpu_rsv,
419 .maxlen = sizeof(int),
420 .mode = 0644,
421 .proc_handler = proc_dointvec_minmax,
422 .extra1 = &min_mem_pcpu_rsv,
423 },
424 {
425 .procname = "dev_weight",
426 .data = &weight_p,
427 .maxlen = sizeof(int),
428 .mode = 0644,
429 .proc_handler = proc_do_dev_weight,
430 },
431 {
432 .procname = "dev_weight_rx_bias",
433 .data = &dev_weight_rx_bias,
434 .maxlen = sizeof(int),
435 .mode = 0644,
436 .proc_handler = proc_do_dev_weight,
437 },
438 {
439 .procname = "dev_weight_tx_bias",
440 .data = &dev_weight_tx_bias,
441 .maxlen = sizeof(int),
442 .mode = 0644,
443 .proc_handler = proc_do_dev_weight,
444 },
445 {
446 .procname = "netdev_max_backlog",
447 .data = &net_hotdata.max_backlog,
448 .maxlen = sizeof(int),
449 .mode = 0644,
450 .proc_handler = proc_dointvec
451 },
452 {
453 .procname = "netdev_rss_key",
454 .data = &netdev_rss_key,
455 .maxlen = sizeof(int),
456 .mode = 0444,
457 .proc_handler = proc_do_rss_key,
458 },
459#ifdef CONFIG_BPF_JIT
460 {
461 .procname = "bpf_jit_enable",
462 .data = &bpf_jit_enable,
463 .maxlen = sizeof(int),
464 .mode = 0644,
465 .proc_handler = proc_dointvec_minmax_bpf_enable,
466# ifdef CONFIG_BPF_JIT_ALWAYS_ON
467 .extra1 = SYSCTL_ONE,
468 .extra2 = SYSCTL_ONE,
469# else
470 .extra1 = SYSCTL_ZERO,
471 .extra2 = SYSCTL_TWO,
472# endif
473 },
474# ifdef CONFIG_HAVE_EBPF_JIT
475 {
476 .procname = "bpf_jit_harden",
477 .data = &bpf_jit_harden,
478 .maxlen = sizeof(int),
479 .mode = 0600,
480 .proc_handler = proc_dointvec_minmax_bpf_restricted,
481 .extra1 = SYSCTL_ZERO,
482 .extra2 = SYSCTL_TWO,
483 },
484 {
485 .procname = "bpf_jit_kallsyms",
486 .data = &bpf_jit_kallsyms,
487 .maxlen = sizeof(int),
488 .mode = 0600,
489 .proc_handler = proc_dointvec_minmax_bpf_restricted,
490 .extra1 = SYSCTL_ZERO,
491 .extra2 = SYSCTL_ONE,
492 },
493# endif
494 {
495 .procname = "bpf_jit_limit",
496 .data = &bpf_jit_limit,
497 .maxlen = sizeof(long),
498 .mode = 0600,
499 .proc_handler = proc_dolongvec_minmax_bpf_restricted,
500 .extra1 = SYSCTL_LONG_ONE,
501 .extra2 = &bpf_jit_limit_max,
502 },
503#endif
504 {
505 .procname = "netdev_tstamp_prequeue",
506 .data = &net_hotdata.tstamp_prequeue,
507 .maxlen = sizeof(int),
508 .mode = 0644,
509 .proc_handler = proc_dointvec
510 },
511 {
512 .procname = "message_cost",
513 .data = &net_ratelimit_state.interval,
514 .maxlen = sizeof(int),
515 .mode = 0644,
516 .proc_handler = proc_dointvec_jiffies,
517 },
518 {
519 .procname = "message_burst",
520 .data = &net_ratelimit_state.burst,
521 .maxlen = sizeof(int),
522 .mode = 0644,
523 .proc_handler = proc_dointvec,
524 },
525 {
526 .procname = "tstamp_allow_data",
527 .data = &sysctl_tstamp_allow_data,
528 .maxlen = sizeof(int),
529 .mode = 0644,
530 .proc_handler = proc_dointvec_minmax,
531 .extra1 = SYSCTL_ZERO,
532 .extra2 = SYSCTL_ONE
533 },
534#ifdef CONFIG_RPS
535 {
536 .procname = "rps_sock_flow_entries",
537 .maxlen = sizeof(int),
538 .mode = 0644,
539 .proc_handler = rps_sock_flow_sysctl
540 },
541#endif
542#ifdef CONFIG_NET_FLOW_LIMIT
543 {
544 .procname = "flow_limit_cpu_bitmap",
545 .mode = 0644,
546 .proc_handler = flow_limit_cpu_sysctl
547 },
548 {
549 .procname = "flow_limit_table_len",
550 .data = &netdev_flow_limit_table_len,
551 .maxlen = sizeof(int),
552 .mode = 0644,
553 .proc_handler = flow_limit_table_len_sysctl
554 },
555#endif /* CONFIG_NET_FLOW_LIMIT */
556#ifdef CONFIG_NET_RX_BUSY_POLL
557 {
558 .procname = "busy_poll",
559 .data = &sysctl_net_busy_poll,
560 .maxlen = sizeof(unsigned int),
561 .mode = 0644,
562 .proc_handler = proc_dointvec_minmax,
563 .extra1 = SYSCTL_ZERO,
564 },
565 {
566 .procname = "busy_read",
567 .data = &sysctl_net_busy_read,
568 .maxlen = sizeof(unsigned int),
569 .mode = 0644,
570 .proc_handler = proc_dointvec_minmax,
571 .extra1 = SYSCTL_ZERO,
572 },
573#endif
574#ifdef CONFIG_NET_SCHED
575 {
576 .procname = "default_qdisc",
577 .mode = 0644,
578 .maxlen = IFNAMSIZ,
579 .proc_handler = set_default_qdisc
580 },
581#endif
582 {
583 .procname = "netdev_budget",
584 .data = &net_hotdata.netdev_budget,
585 .maxlen = sizeof(int),
586 .mode = 0644,
587 .proc_handler = proc_dointvec
588 },
589 {
590 .procname = "warnings",
591 .data = &net_msg_warn,
592 .maxlen = sizeof(int),
593 .mode = 0644,
594 .proc_handler = proc_dointvec
595 },
596 {
597 .procname = "max_skb_frags",
598 .data = &sysctl_max_skb_frags,
599 .maxlen = sizeof(int),
600 .mode = 0644,
601 .proc_handler = proc_dointvec_minmax,
602 .extra1 = SYSCTL_ONE,
603 .extra2 = &max_skb_frags,
604 },
605 {
606 .procname = "netdev_budget_usecs",
607 .data = &net_hotdata.netdev_budget_usecs,
608 .maxlen = sizeof(unsigned int),
609 .mode = 0644,
610 .proc_handler = proc_dointvec_minmax,
611 .extra1 = SYSCTL_ZERO,
612 },
613 {
614 .procname = "fb_tunnels_only_for_init_net",
615 .data = &sysctl_fb_tunnels_only_for_init_net,
616 .maxlen = sizeof(int),
617 .mode = 0644,
618 .proc_handler = proc_dointvec_minmax,
619 .extra1 = SYSCTL_ZERO,
620 .extra2 = SYSCTL_TWO,
621 },
622 {
623 .procname = "devconf_inherit_init_net",
624 .data = &sysctl_devconf_inherit_init_net,
625 .maxlen = sizeof(int),
626 .mode = 0644,
627 .proc_handler = proc_dointvec_minmax,
628 .extra1 = SYSCTL_ZERO,
629 .extra2 = SYSCTL_THREE,
630 },
631 {
632 .procname = "high_order_alloc_disable",
633 .data = &net_high_order_alloc_disable_key.key,
634 .maxlen = sizeof(net_high_order_alloc_disable_key),
635 .mode = 0644,
636 .proc_handler = proc_do_static_key,
637 },
638 {
639 .procname = "gro_normal_batch",
640 .data = &net_hotdata.gro_normal_batch,
641 .maxlen = sizeof(unsigned int),
642 .mode = 0644,
643 .proc_handler = proc_dointvec_minmax,
644 .extra1 = SYSCTL_ONE,
645 },
646 {
647 .procname = "netdev_unregister_timeout_secs",
648 .data = &netdev_unregister_timeout_secs,
649 .maxlen = sizeof(unsigned int),
650 .mode = 0644,
651 .proc_handler = proc_dointvec_minmax,
652 .extra1 = SYSCTL_ONE,
653 .extra2 = &int_3600,
654 },
655 {
656 .procname = "skb_defer_max",
657 .data = &sysctl_skb_defer_max,
658 .maxlen = sizeof(unsigned int),
659 .mode = 0644,
660 .proc_handler = proc_dointvec_minmax,
661 .extra1 = SYSCTL_ZERO,
662 },
663 { }
664};
665
666static struct ctl_table netns_core_table[] = {
667#if IS_ENABLED(CONFIG_RPS)
668 {
669 .procname = "rps_default_mask",
670 .data = &init_net,
671 .mode = 0644,
672 .proc_handler = rps_default_mask_sysctl
673 },
674#endif
675 {
676 .procname = "somaxconn",
677 .data = &init_net.core.sysctl_somaxconn,
678 .maxlen = sizeof(int),
679 .mode = 0644,
680 .extra1 = SYSCTL_ZERO,
681 .proc_handler = proc_dointvec_minmax
682 },
683 {
684 .procname = "optmem_max",
685 .data = &init_net.core.sysctl_optmem_max,
686 .maxlen = sizeof(int),
687 .mode = 0644,
688 .extra1 = SYSCTL_ZERO,
689 .proc_handler = proc_dointvec_minmax
690 },
691 {
692 .procname = "txrehash",
693 .data = &init_net.core.sysctl_txrehash,
694 .maxlen = sizeof(u8),
695 .mode = 0644,
696 .extra1 = SYSCTL_ZERO,
697 .extra2 = SYSCTL_ONE,
698 .proc_handler = proc_dou8vec_minmax,
699 },
700 { }
701};
702
703static int __init fb_tunnels_only_for_init_net_sysctl_setup(char *str)
704{
705 /* fallback tunnels for initns only */
706 if (!strncmp(str, "initns", 6))
707 sysctl_fb_tunnels_only_for_init_net = 1;
708 /* no fallback tunnels anywhere */
709 else if (!strncmp(str, "none", 4))
710 sysctl_fb_tunnels_only_for_init_net = 2;
711
712 return 1;
713}
714__setup("fb_tunnels=", fb_tunnels_only_for_init_net_sysctl_setup);
715
716static __net_init int sysctl_core_net_init(struct net *net)
717{
718 struct ctl_table *tbl, *tmp;
719
720 tbl = netns_core_table;
721 if (!net_eq(net, &init_net)) {
722 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
723 if (tbl == NULL)
724 goto err_dup;
725
726 for (tmp = tbl; tmp->procname; tmp++)
727 tmp->data += (char *)net - (char *)&init_net;
728 }
729
730 net->core.sysctl_hdr = register_net_sysctl_sz(net, "net/core", tbl,
731 ARRAY_SIZE(netns_core_table));
732 if (net->core.sysctl_hdr == NULL)
733 goto err_reg;
734
735 return 0;
736
737err_reg:
738 if (tbl != netns_core_table)
739 kfree(tbl);
740err_dup:
741 return -ENOMEM;
742}
743
744static __net_exit void sysctl_core_net_exit(struct net *net)
745{
746 struct ctl_table *tbl;
747
748 tbl = net->core.sysctl_hdr->ctl_table_arg;
749 unregister_net_sysctl_table(net->core.sysctl_hdr);
750 BUG_ON(tbl == netns_core_table);
751#if IS_ENABLED(CONFIG_RPS)
752 kfree(net->core.rps_default_mask);
753#endif
754 kfree(tbl);
755}
756
757static __net_initdata struct pernet_operations sysctl_core_ops = {
758 .init = sysctl_core_net_init,
759 .exit = sysctl_core_net_exit,
760};
761
762static __init int sysctl_core_init(void)
763{
764 register_net_sysctl(&init_net, "net/core", net_core_table);
765 return register_pernet_subsys(&sysctl_core_ops);
766}
767
768fs_initcall(sysctl_core_init);