Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
tjh.dev
kernel
os
linux
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/proto_memory.h>
28#include <net/rps.h>
29
30#include "dev.h"
31
32static int int_3600 = 3600;
33static int min_sndbuf = SOCK_MIN_SNDBUF;
34static int min_rcvbuf = SOCK_MIN_RCVBUF;
35static int max_skb_frags = MAX_SKB_FRAGS;
36static int min_mem_pcpu_rsv = SK_MEMORY_PCPU_RESERVE;
37
38static int net_msg_warn; /* Unused, but still a sysctl */
39
40int sysctl_fb_tunnels_only_for_init_net __read_mostly = 0;
41EXPORT_SYMBOL(sysctl_fb_tunnels_only_for_init_net);
42
43/* 0 - Keep current behavior:
44 * IPv4: inherit all current settings from init_net
45 * IPv6: reset all settings to default
46 * 1 - Both inherit all current settings from init_net
47 * 2 - Both reset all settings to default
48 * 3 - Both inherit all settings from current netns
49 */
50int sysctl_devconf_inherit_init_net __read_mostly;
51EXPORT_SYMBOL(sysctl_devconf_inherit_init_net);
52
53#if IS_ENABLED(CONFIG_NET_FLOW_LIMIT) || IS_ENABLED(CONFIG_RPS)
54static int dump_cpumask(void *buffer, size_t *lenp, loff_t *ppos,
55 struct cpumask *mask)
56{
57 char *kbuf;
58 int len;
59
60 if (*ppos || !*lenp) {
61 *lenp = 0;
62 return 0;
63 }
64
65 /* CPUs are displayed as a hex bitmap + a comma between each groups of 8
66 * nibbles (except the last one which has a newline instead).
67 * Guesstimate the buffer size at the group granularity level.
68 */
69 len = min(DIV_ROUND_UP(nr_cpumask_bits, 32) * (8 + 1), *lenp);
70 kbuf = kmalloc(len, GFP_KERNEL);
71 if (!kbuf) {
72 *lenp = 0;
73 return -ENOMEM;
74 }
75
76 len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
77 if (!len) {
78 *lenp = 0;
79 goto free_buf;
80 }
81
82 /* scnprintf writes a trailing null char not counted in the returned
83 * length, override it with a newline.
84 */
85 kbuf[len++] = '\n';
86 memcpy(buffer, kbuf, len);
87 *lenp = len;
88 *ppos += len;
89
90free_buf:
91 kfree(kbuf);
92 return 0;
93}
94#endif
95
96#ifdef CONFIG_RPS
97
98static struct cpumask *rps_default_mask_cow_alloc(struct net *net)
99{
100 struct cpumask *rps_default_mask;
101
102 if (net->core.rps_default_mask)
103 return net->core.rps_default_mask;
104
105 rps_default_mask = kzalloc(cpumask_size(), GFP_KERNEL);
106 if (!rps_default_mask)
107 return NULL;
108
109 /* pairs with READ_ONCE in rx_queue_default_mask() */
110 WRITE_ONCE(net->core.rps_default_mask, rps_default_mask);
111 return rps_default_mask;
112}
113
114static int rps_default_mask_sysctl(const struct ctl_table *table, int write,
115 void *buffer, size_t *lenp, loff_t *ppos)
116{
117 struct net *net = (struct net *)table->data;
118 int err = 0;
119
120 rtnl_lock();
121 if (write) {
122 struct cpumask *rps_default_mask = rps_default_mask_cow_alloc(net);
123
124 err = -ENOMEM;
125 if (!rps_default_mask)
126 goto done;
127
128 err = cpumask_parse(buffer, rps_default_mask);
129 if (err)
130 goto done;
131
132 err = rps_cpumask_housekeeping(rps_default_mask);
133 if (err)
134 goto done;
135 } else {
136 err = dump_cpumask(buffer, lenp, ppos,
137 net->core.rps_default_mask ? : cpu_none_mask);
138 }
139
140done:
141 rtnl_unlock();
142 return err;
143}
144
145static int rps_sock_flow_sysctl(const struct ctl_table *table, int write,
146 void *buffer, size_t *lenp, loff_t *ppos)
147{
148 unsigned int orig_size, size;
149 int ret, i;
150 struct ctl_table tmp = {
151 .data = &size,
152 .maxlen = sizeof(size),
153 .mode = table->mode
154 };
155 struct rps_sock_flow_table *orig_sock_table, *sock_table;
156 static DEFINE_MUTEX(sock_flow_mutex);
157
158 mutex_lock(&sock_flow_mutex);
159
160 orig_sock_table = rcu_dereference_protected(
161 net_hotdata.rps_sock_flow_table,
162 lockdep_is_held(&sock_flow_mutex));
163 size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
164
165 ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
166
167 if (write) {
168 if (size) {
169 if (size > 1<<29) {
170 /* Enforce limit to prevent overflow */
171 mutex_unlock(&sock_flow_mutex);
172 return -EINVAL;
173 }
174 size = roundup_pow_of_two(size);
175 if (size != orig_size) {
176 sock_table =
177 vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
178 if (!sock_table) {
179 mutex_unlock(&sock_flow_mutex);
180 return -ENOMEM;
181 }
182 net_hotdata.rps_cpu_mask =
183 roundup_pow_of_two(nr_cpu_ids) - 1;
184 sock_table->mask = size - 1;
185 } else
186 sock_table = orig_sock_table;
187
188 for (i = 0; i < size; i++)
189 sock_table->ents[i] = RPS_NO_CPU;
190 } else
191 sock_table = NULL;
192
193 if (sock_table != orig_sock_table) {
194 rcu_assign_pointer(net_hotdata.rps_sock_flow_table,
195 sock_table);
196 if (sock_table) {
197 static_branch_inc(&rps_needed);
198 static_branch_inc(&rfs_needed);
199 }
200 if (orig_sock_table) {
201 static_branch_dec(&rps_needed);
202 static_branch_dec(&rfs_needed);
203 kvfree_rcu_mightsleep(orig_sock_table);
204 }
205 }
206 }
207
208 mutex_unlock(&sock_flow_mutex);
209
210 return ret;
211}
212#endif /* CONFIG_RPS */
213
214#ifdef CONFIG_NET_FLOW_LIMIT
215static DEFINE_MUTEX(flow_limit_update_mutex);
216
217static int flow_limit_cpu_sysctl(const struct ctl_table *table, int write,
218 void *buffer, size_t *lenp, loff_t *ppos)
219{
220 struct sd_flow_limit *cur;
221 struct softnet_data *sd;
222 cpumask_var_t mask;
223 int i, len, ret = 0;
224
225 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
226 return -ENOMEM;
227
228 if (write) {
229 ret = cpumask_parse(buffer, mask);
230 if (ret)
231 goto done;
232
233 mutex_lock(&flow_limit_update_mutex);
234 len = sizeof(*cur) + netdev_flow_limit_table_len;
235 for_each_possible_cpu(i) {
236 sd = &per_cpu(softnet_data, i);
237 cur = rcu_dereference_protected(sd->flow_limit,
238 lockdep_is_held(&flow_limit_update_mutex));
239 if (cur && !cpumask_test_cpu(i, mask)) {
240 RCU_INIT_POINTER(sd->flow_limit, NULL);
241 kfree_rcu_mightsleep(cur);
242 } else if (!cur && cpumask_test_cpu(i, mask)) {
243 cur = kzalloc_node(len, GFP_KERNEL,
244 cpu_to_node(i));
245 if (!cur) {
246 /* not unwinding previous changes */
247 ret = -ENOMEM;
248 goto write_unlock;
249 }
250 cur->num_buckets = netdev_flow_limit_table_len;
251 rcu_assign_pointer(sd->flow_limit, cur);
252 }
253 }
254write_unlock:
255 mutex_unlock(&flow_limit_update_mutex);
256 } else {
257 cpumask_clear(mask);
258 rcu_read_lock();
259 for_each_possible_cpu(i) {
260 sd = &per_cpu(softnet_data, i);
261 if (rcu_dereference(sd->flow_limit))
262 cpumask_set_cpu(i, mask);
263 }
264 rcu_read_unlock();
265
266 ret = dump_cpumask(buffer, lenp, ppos, mask);
267 }
268
269done:
270 free_cpumask_var(mask);
271 return ret;
272}
273
274static int flow_limit_table_len_sysctl(const struct ctl_table *table, int write,
275 void *buffer, size_t *lenp, loff_t *ppos)
276{
277 unsigned int old, *ptr;
278 int ret;
279
280 mutex_lock(&flow_limit_update_mutex);
281
282 ptr = table->data;
283 old = *ptr;
284 ret = proc_dointvec(table, write, buffer, lenp, ppos);
285 if (!ret && write && !is_power_of_2(*ptr)) {
286 *ptr = old;
287 ret = -EINVAL;
288 }
289
290 mutex_unlock(&flow_limit_update_mutex);
291 return ret;
292}
293#endif /* CONFIG_NET_FLOW_LIMIT */
294
295#ifdef CONFIG_NET_SCHED
296static int set_default_qdisc(const struct ctl_table *table, int write,
297 void *buffer, size_t *lenp, loff_t *ppos)
298{
299 char id[IFNAMSIZ];
300 struct ctl_table tbl = {
301 .data = id,
302 .maxlen = IFNAMSIZ,
303 };
304 int ret;
305
306 qdisc_get_default(id, IFNAMSIZ);
307
308 ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
309 if (write && ret == 0)
310 ret = qdisc_set_default(id);
311 return ret;
312}
313#endif
314
315static int proc_do_dev_weight(const struct ctl_table *table, int write,
316 void *buffer, size_t *lenp, loff_t *ppos)
317{
318 static DEFINE_MUTEX(dev_weight_mutex);
319 int ret, weight;
320
321 mutex_lock(&dev_weight_mutex);
322 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
323 if (!ret && write) {
324 weight = READ_ONCE(weight_p);
325 WRITE_ONCE(net_hotdata.dev_rx_weight, weight * dev_weight_rx_bias);
326 WRITE_ONCE(net_hotdata.dev_tx_weight, weight * dev_weight_tx_bias);
327 }
328 mutex_unlock(&dev_weight_mutex);
329
330 return ret;
331}
332
333static int proc_do_rss_key(const struct ctl_table *table, int write,
334 void *buffer, size_t *lenp, loff_t *ppos)
335{
336 struct ctl_table fake_table;
337 char buf[NETDEV_RSS_KEY_LEN * 3];
338
339 snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
340 fake_table.data = buf;
341 fake_table.maxlen = sizeof(buf);
342 return proc_dostring(&fake_table, write, buffer, lenp, ppos);
343}
344
345#ifdef CONFIG_BPF_JIT
346static int proc_dointvec_minmax_bpf_enable(const struct ctl_table *table, int write,
347 void *buffer, size_t *lenp,
348 loff_t *ppos)
349{
350 int ret, jit_enable = *(int *)table->data;
351 int min = *(int *)table->extra1;
352 int max = *(int *)table->extra2;
353 struct ctl_table tmp = *table;
354
355 if (write && !capable(CAP_SYS_ADMIN))
356 return -EPERM;
357
358 tmp.data = &jit_enable;
359 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
360 if (write && !ret) {
361 if (jit_enable < 2 ||
362 (jit_enable == 2 && bpf_dump_raw_ok(current_cred()))) {
363 *(int *)table->data = jit_enable;
364 if (jit_enable == 2)
365 pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
366 } else {
367 ret = -EPERM;
368 }
369 }
370
371 if (write && ret && min == max)
372 pr_info_once("CONFIG_BPF_JIT_ALWAYS_ON is enabled, bpf_jit_enable is permanently set to 1.\n");
373
374 return ret;
375}
376
377# ifdef CONFIG_HAVE_EBPF_JIT
378static int
379proc_dointvec_minmax_bpf_restricted(const struct ctl_table *table, int write,
380 void *buffer, size_t *lenp, loff_t *ppos)
381{
382 if (!capable(CAP_SYS_ADMIN))
383 return -EPERM;
384
385 return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
386}
387# endif /* CONFIG_HAVE_EBPF_JIT */
388
389static int
390proc_dolongvec_minmax_bpf_restricted(const struct ctl_table *table, int write,
391 void *buffer, size_t *lenp, loff_t *ppos)
392{
393 if (!capable(CAP_SYS_ADMIN))
394 return -EPERM;
395
396 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
397}
398#endif
399
400static struct ctl_table net_core_table[] = {
401 {
402 .procname = "mem_pcpu_rsv",
403 .data = &net_hotdata.sysctl_mem_pcpu_rsv,
404 .maxlen = sizeof(int),
405 .mode = 0644,
406 .proc_handler = proc_dointvec_minmax,
407 .extra1 = &min_mem_pcpu_rsv,
408 },
409 {
410 .procname = "dev_weight",
411 .data = &weight_p,
412 .maxlen = sizeof(int),
413 .mode = 0644,
414 .proc_handler = proc_do_dev_weight,
415 .extra1 = SYSCTL_ONE,
416 },
417 {
418 .procname = "dev_weight_rx_bias",
419 .data = &dev_weight_rx_bias,
420 .maxlen = sizeof(int),
421 .mode = 0644,
422 .proc_handler = proc_do_dev_weight,
423 .extra1 = SYSCTL_ONE,
424 },
425 {
426 .procname = "dev_weight_tx_bias",
427 .data = &dev_weight_tx_bias,
428 .maxlen = sizeof(int),
429 .mode = 0644,
430 .proc_handler = proc_do_dev_weight,
431 .extra1 = SYSCTL_ONE,
432 },
433 {
434 .procname = "netdev_max_backlog",
435 .data = &net_hotdata.max_backlog,
436 .maxlen = sizeof(int),
437 .mode = 0644,
438 .proc_handler = proc_dointvec
439 },
440 {
441 .procname = "netdev_rss_key",
442 .data = &netdev_rss_key,
443 .maxlen = sizeof(int),
444 .mode = 0444,
445 .proc_handler = proc_do_rss_key,
446 },
447#ifdef CONFIG_BPF_JIT
448 {
449 .procname = "bpf_jit_enable",
450 .data = &bpf_jit_enable,
451 .maxlen = sizeof(int),
452 .mode = 0644,
453 .proc_handler = proc_dointvec_minmax_bpf_enable,
454# ifdef CONFIG_BPF_JIT_ALWAYS_ON
455 .extra1 = SYSCTL_ONE,
456 .extra2 = SYSCTL_ONE,
457# else
458 .extra1 = SYSCTL_ZERO,
459 .extra2 = SYSCTL_TWO,
460# endif
461 },
462# ifdef CONFIG_HAVE_EBPF_JIT
463 {
464 .procname = "bpf_jit_harden",
465 .data = &bpf_jit_harden,
466 .maxlen = sizeof(int),
467 .mode = 0600,
468 .proc_handler = proc_dointvec_minmax_bpf_restricted,
469 .extra1 = SYSCTL_ZERO,
470 .extra2 = SYSCTL_TWO,
471 },
472 {
473 .procname = "bpf_jit_kallsyms",
474 .data = &bpf_jit_kallsyms,
475 .maxlen = sizeof(int),
476 .mode = 0600,
477 .proc_handler = proc_dointvec_minmax_bpf_restricted,
478 .extra1 = SYSCTL_ZERO,
479 .extra2 = SYSCTL_ONE,
480 },
481# endif
482 {
483 .procname = "bpf_jit_limit",
484 .data = &bpf_jit_limit,
485 .maxlen = sizeof(long),
486 .mode = 0600,
487 .proc_handler = proc_dolongvec_minmax_bpf_restricted,
488 .extra1 = SYSCTL_LONG_ONE,
489 .extra2 = &bpf_jit_limit_max,
490 },
491#endif
492 {
493 .procname = "netdev_tstamp_prequeue",
494 .data = &net_hotdata.tstamp_prequeue,
495 .maxlen = sizeof(int),
496 .mode = 0644,
497 .proc_handler = proc_dointvec
498 },
499 {
500 .procname = "message_cost",
501 .data = &net_ratelimit_state.interval,
502 .maxlen = sizeof(int),
503 .mode = 0644,
504 .proc_handler = proc_dointvec_jiffies,
505 },
506 {
507 .procname = "message_burst",
508 .data = &net_ratelimit_state.burst,
509 .maxlen = sizeof(int),
510 .mode = 0644,
511 .proc_handler = proc_dointvec,
512 },
513#ifdef CONFIG_RPS
514 {
515 .procname = "rps_sock_flow_entries",
516 .maxlen = sizeof(int),
517 .mode = 0644,
518 .proc_handler = rps_sock_flow_sysctl
519 },
520#endif
521#ifdef CONFIG_NET_FLOW_LIMIT
522 {
523 .procname = "flow_limit_cpu_bitmap",
524 .mode = 0644,
525 .proc_handler = flow_limit_cpu_sysctl
526 },
527 {
528 .procname = "flow_limit_table_len",
529 .data = &netdev_flow_limit_table_len,
530 .maxlen = sizeof(int),
531 .mode = 0644,
532 .proc_handler = flow_limit_table_len_sysctl
533 },
534#endif /* CONFIG_NET_FLOW_LIMIT */
535#ifdef CONFIG_NET_RX_BUSY_POLL
536 {
537 .procname = "busy_poll",
538 .data = &sysctl_net_busy_poll,
539 .maxlen = sizeof(unsigned int),
540 .mode = 0644,
541 .proc_handler = proc_dointvec_minmax,
542 .extra1 = SYSCTL_ZERO,
543 },
544 {
545 .procname = "busy_read",
546 .data = &sysctl_net_busy_read,
547 .maxlen = sizeof(unsigned int),
548 .mode = 0644,
549 .proc_handler = proc_dointvec_minmax,
550 .extra1 = SYSCTL_ZERO,
551 },
552#endif
553#ifdef CONFIG_NET_SCHED
554 {
555 .procname = "default_qdisc",
556 .mode = 0644,
557 .maxlen = IFNAMSIZ,
558 .proc_handler = set_default_qdisc
559 },
560#endif
561 {
562 .procname = "netdev_budget",
563 .data = &net_hotdata.netdev_budget,
564 .maxlen = sizeof(int),
565 .mode = 0644,
566 .proc_handler = proc_dointvec
567 },
568 {
569 .procname = "warnings",
570 .data = &net_msg_warn,
571 .maxlen = sizeof(int),
572 .mode = 0644,
573 .proc_handler = proc_dointvec
574 },
575 {
576 .procname = "max_skb_frags",
577 .data = &net_hotdata.sysctl_max_skb_frags,
578 .maxlen = sizeof(int),
579 .mode = 0644,
580 .proc_handler = proc_dointvec_minmax,
581 .extra1 = SYSCTL_ONE,
582 .extra2 = &max_skb_frags,
583 },
584 {
585 .procname = "netdev_budget_usecs",
586 .data = &net_hotdata.netdev_budget_usecs,
587 .maxlen = sizeof(unsigned int),
588 .mode = 0644,
589 .proc_handler = proc_dointvec_minmax,
590 .extra1 = SYSCTL_ZERO,
591 },
592 {
593 .procname = "fb_tunnels_only_for_init_net",
594 .data = &sysctl_fb_tunnels_only_for_init_net,
595 .maxlen = sizeof(int),
596 .mode = 0644,
597 .proc_handler = proc_dointvec_minmax,
598 .extra1 = SYSCTL_ZERO,
599 .extra2 = SYSCTL_TWO,
600 },
601 {
602 .procname = "devconf_inherit_init_net",
603 .data = &sysctl_devconf_inherit_init_net,
604 .maxlen = sizeof(int),
605 .mode = 0644,
606 .proc_handler = proc_dointvec_minmax,
607 .extra1 = SYSCTL_ZERO,
608 .extra2 = SYSCTL_THREE,
609 },
610 {
611 .procname = "high_order_alloc_disable",
612 .data = &net_high_order_alloc_disable_key.key,
613 .maxlen = sizeof(net_high_order_alloc_disable_key),
614 .mode = 0644,
615 .proc_handler = proc_do_static_key,
616 },
617 {
618 .procname = "gro_normal_batch",
619 .data = &net_hotdata.gro_normal_batch,
620 .maxlen = sizeof(unsigned int),
621 .mode = 0644,
622 .proc_handler = proc_dointvec_minmax,
623 .extra1 = SYSCTL_ONE,
624 },
625 {
626 .procname = "netdev_unregister_timeout_secs",
627 .data = &netdev_unregister_timeout_secs,
628 .maxlen = sizeof(unsigned int),
629 .mode = 0644,
630 .proc_handler = proc_dointvec_minmax,
631 .extra1 = SYSCTL_ONE,
632 .extra2 = &int_3600,
633 },
634 {
635 .procname = "skb_defer_max",
636 .data = &net_hotdata.sysctl_skb_defer_max,
637 .maxlen = sizeof(unsigned int),
638 .mode = 0644,
639 .proc_handler = proc_dointvec_minmax,
640 .extra1 = SYSCTL_ZERO,
641 },
642};
643
644static struct ctl_table netns_core_table[] = {
645#if IS_ENABLED(CONFIG_RPS)
646 {
647 .procname = "rps_default_mask",
648 .data = &init_net,
649 .mode = 0644,
650 .proc_handler = rps_default_mask_sysctl
651 },
652#endif
653 {
654 .procname = "somaxconn",
655 .data = &init_net.core.sysctl_somaxconn,
656 .maxlen = sizeof(int),
657 .mode = 0644,
658 .extra1 = SYSCTL_ZERO,
659 .proc_handler = proc_dointvec_minmax
660 },
661 {
662 .procname = "optmem_max",
663 .data = &init_net.core.sysctl_optmem_max,
664 .maxlen = sizeof(int),
665 .mode = 0644,
666 .extra1 = SYSCTL_ZERO,
667 .proc_handler = proc_dointvec_minmax
668 },
669 {
670 .procname = "txrehash",
671 .data = &init_net.core.sysctl_txrehash,
672 .maxlen = sizeof(u8),
673 .mode = 0644,
674 .extra1 = SYSCTL_ZERO,
675 .extra2 = SYSCTL_ONE,
676 .proc_handler = proc_dou8vec_minmax,
677 },
678 {
679 .procname = "tstamp_allow_data",
680 .data = &init_net.core.sysctl_tstamp_allow_data,
681 .maxlen = sizeof(u8),
682 .mode = 0644,
683 .proc_handler = proc_dou8vec_minmax,
684 .extra1 = SYSCTL_ZERO,
685 .extra2 = SYSCTL_ONE
686 },
687 /* sysctl_core_net_init() will set the values after this
688 * to readonly in network namespaces
689 */
690 {
691 .procname = "wmem_max",
692 .data = &sysctl_wmem_max,
693 .maxlen = sizeof(int),
694 .mode = 0644,
695 .proc_handler = proc_dointvec_minmax,
696 .extra1 = &min_sndbuf,
697 },
698 {
699 .procname = "rmem_max",
700 .data = &sysctl_rmem_max,
701 .maxlen = sizeof(int),
702 .mode = 0644,
703 .proc_handler = proc_dointvec_minmax,
704 .extra1 = &min_rcvbuf,
705 },
706 {
707 .procname = "wmem_default",
708 .data = &sysctl_wmem_default,
709 .maxlen = sizeof(int),
710 .mode = 0644,
711 .proc_handler = proc_dointvec_minmax,
712 .extra1 = &min_sndbuf,
713 },
714 {
715 .procname = "rmem_default",
716 .data = &sysctl_rmem_default,
717 .maxlen = sizeof(int),
718 .mode = 0644,
719 .proc_handler = proc_dointvec_minmax,
720 .extra1 = &min_rcvbuf,
721 },
722};
723
724static int __init fb_tunnels_only_for_init_net_sysctl_setup(char *str)
725{
726 /* fallback tunnels for initns only */
727 if (!strncmp(str, "initns", 6))
728 sysctl_fb_tunnels_only_for_init_net = 1;
729 /* no fallback tunnels anywhere */
730 else if (!strncmp(str, "none", 4))
731 sysctl_fb_tunnels_only_for_init_net = 2;
732
733 return 1;
734}
735__setup("fb_tunnels=", fb_tunnels_only_for_init_net_sysctl_setup);
736
737static __net_init int sysctl_core_net_init(struct net *net)
738{
739 size_t table_size = ARRAY_SIZE(netns_core_table);
740 struct ctl_table *tbl;
741
742 tbl = netns_core_table;
743 if (!net_eq(net, &init_net)) {
744 int i;
745 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
746 if (tbl == NULL)
747 goto err_dup;
748
749 for (i = 0; i < table_size; ++i) {
750 if (tbl[i].data == &sysctl_wmem_max)
751 break;
752
753 tbl[i].data += (char *)net - (char *)&init_net;
754 }
755 for (; i < table_size; ++i)
756 tbl[i].mode &= ~0222;
757 }
758
759 net->core.sysctl_hdr = register_net_sysctl_sz(net, "net/core", tbl, table_size);
760 if (net->core.sysctl_hdr == NULL)
761 goto err_reg;
762
763 return 0;
764
765err_reg:
766 if (tbl != netns_core_table)
767 kfree(tbl);
768err_dup:
769 return -ENOMEM;
770}
771
772static __net_exit void sysctl_core_net_exit(struct net *net)
773{
774 const struct ctl_table *tbl;
775
776 tbl = net->core.sysctl_hdr->ctl_table_arg;
777 unregister_net_sysctl_table(net->core.sysctl_hdr);
778 BUG_ON(tbl == netns_core_table);
779#if IS_ENABLED(CONFIG_RPS)
780 kfree(net->core.rps_default_mask);
781#endif
782 kfree(tbl);
783}
784
785static __net_initdata struct pernet_operations sysctl_core_ops = {
786 .init = sysctl_core_net_init,
787 .exit = sysctl_core_net_exit,
788};
789
790static __init int sysctl_core_init(void)
791{
792 register_net_sysctl(&init_net, "net/core", net_core_table);
793 return register_pernet_subsys(&sysctl_core_ops);
794}
795
796fs_initcall(sysctl_core_init);