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 20 #include <net/ip.h> 21 #include <net/sock.h> 22 #include <net/net_ratelimit.h> 23 #include <net/busy_poll.h> 24 #include <net/pkt_sched.h> 25 26 #include "dev.h" 27 28 static int int_3600 = 3600; 29 static int min_sndbuf = SOCK_MIN_SNDBUF; 30 static int min_rcvbuf = SOCK_MIN_RCVBUF; 31 static int max_skb_frags = MAX_SKB_FRAGS; 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 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 kvfree_rcu(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); 118 119 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 kfree_rcu(cur); 144 } else if (!cur && cpumask_test_cpu(i, mask)) { 145 cur = kzalloc_node(len, GFP_KERNEL, 146 cpu_to_node(i)); 147 if (!cur) { 148 /* not unwinding previous changes */ 149 ret = -ENOMEM; 150 goto write_unlock; 151 } 152 cur->num_buckets = netdev_flow_limit_table_len; 153 rcu_assign_pointer(sd->flow_limit, cur); 154 } 155 } 156 write_unlock: 157 mutex_unlock(&flow_limit_update_mutex); 158 } else { 159 char kbuf[128]; 160 161 if (*ppos || !*lenp) { 162 *lenp = 0; 163 goto done; 164 } 165 166 cpumask_clear(mask); 167 rcu_read_lock(); 168 for_each_possible_cpu(i) { 169 sd = &per_cpu(softnet_data, i); 170 if (rcu_dereference(sd->flow_limit)) 171 cpumask_set_cpu(i, mask); 172 } 173 rcu_read_unlock(); 174 175 len = min(sizeof(kbuf) - 1, *lenp); 176 len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask)); 177 if (!len) { 178 *lenp = 0; 179 goto done; 180 } 181 if (len < *lenp) 182 kbuf[len++] = '\n'; 183 memcpy(buffer, kbuf, len); 184 *lenp = len; 185 *ppos += len; 186 } 187 188 done: 189 free_cpumask_var(mask); 190 return ret; 191 } 192 193 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write, 194 void *buffer, size_t *lenp, loff_t *ppos) 195 { 196 unsigned int old, *ptr; 197 int ret; 198 199 mutex_lock(&flow_limit_update_mutex); 200 201 ptr = table->data; 202 old = *ptr; 203 ret = proc_dointvec(table, write, buffer, lenp, ppos); 204 if (!ret && write && !is_power_of_2(*ptr)) { 205 *ptr = old; 206 ret = -EINVAL; 207 } 208 209 mutex_unlock(&flow_limit_update_mutex); 210 return ret; 211 } 212 #endif /* CONFIG_NET_FLOW_LIMIT */ 213 214 #ifdef CONFIG_NET_SCHED 215 static int set_default_qdisc(struct ctl_table *table, int write, 216 void *buffer, size_t *lenp, loff_t *ppos) 217 { 218 char id[IFNAMSIZ]; 219 struct ctl_table tbl = { 220 .data = id, 221 .maxlen = IFNAMSIZ, 222 }; 223 int ret; 224 225 qdisc_get_default(id, IFNAMSIZ); 226 227 ret = proc_dostring(&tbl, write, buffer, lenp, ppos); 228 if (write && ret == 0) 229 ret = qdisc_set_default(id); 230 return ret; 231 } 232 #endif 233 234 static int proc_do_dev_weight(struct ctl_table *table, int write, 235 void *buffer, size_t *lenp, loff_t *ppos) 236 { 237 int ret; 238 239 ret = proc_dointvec(table, write, buffer, lenp, ppos); 240 if (ret != 0) 241 return ret; 242 243 dev_rx_weight = weight_p * dev_weight_rx_bias; 244 dev_tx_weight = weight_p * dev_weight_tx_bias; 245 246 return ret; 247 } 248 249 static int proc_do_rss_key(struct ctl_table *table, int write, 250 void *buffer, size_t *lenp, loff_t *ppos) 251 { 252 struct ctl_table fake_table; 253 char buf[NETDEV_RSS_KEY_LEN * 3]; 254 255 snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key); 256 fake_table.data = buf; 257 fake_table.maxlen = sizeof(buf); 258 return proc_dostring(&fake_table, write, buffer, lenp, ppos); 259 } 260 261 #ifdef CONFIG_BPF_JIT 262 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write, 263 void *buffer, size_t *lenp, 264 loff_t *ppos) 265 { 266 int ret, jit_enable = *(int *)table->data; 267 struct ctl_table tmp = *table; 268 269 if (write && !capable(CAP_SYS_ADMIN)) 270 return -EPERM; 271 272 tmp.data = &jit_enable; 273 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); 274 if (write && !ret) { 275 if (jit_enable < 2 || 276 (jit_enable == 2 && bpf_dump_raw_ok(current_cred()))) { 277 *(int *)table->data = jit_enable; 278 if (jit_enable == 2) 279 pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n"); 280 } else { 281 ret = -EPERM; 282 } 283 } 284 return ret; 285 } 286 287 # ifdef CONFIG_HAVE_EBPF_JIT 288 static int 289 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write, 290 void *buffer, size_t *lenp, loff_t *ppos) 291 { 292 if (!capable(CAP_SYS_ADMIN)) 293 return -EPERM; 294 295 return proc_dointvec_minmax(table, write, buffer, lenp, ppos); 296 } 297 # endif /* CONFIG_HAVE_EBPF_JIT */ 298 299 static int 300 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write, 301 void *buffer, size_t *lenp, loff_t *ppos) 302 { 303 if (!capable(CAP_SYS_ADMIN)) 304 return -EPERM; 305 306 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); 307 } 308 #endif 309 310 static struct ctl_table net_core_table[] = { 311 { 312 .procname = "wmem_max", 313 .data = &sysctl_wmem_max, 314 .maxlen = sizeof(int), 315 .mode = 0644, 316 .proc_handler = proc_dointvec_minmax, 317 .extra1 = &min_sndbuf, 318 }, 319 { 320 .procname = "rmem_max", 321 .data = &sysctl_rmem_max, 322 .maxlen = sizeof(int), 323 .mode = 0644, 324 .proc_handler = proc_dointvec_minmax, 325 .extra1 = &min_rcvbuf, 326 }, 327 { 328 .procname = "wmem_default", 329 .data = &sysctl_wmem_default, 330 .maxlen = sizeof(int), 331 .mode = 0644, 332 .proc_handler = proc_dointvec_minmax, 333 .extra1 = &min_sndbuf, 334 }, 335 { 336 .procname = "rmem_default", 337 .data = &sysctl_rmem_default, 338 .maxlen = sizeof(int), 339 .mode = 0644, 340 .proc_handler = proc_dointvec_minmax, 341 .extra1 = &min_rcvbuf, 342 }, 343 { 344 .procname = "dev_weight", 345 .data = &weight_p, 346 .maxlen = sizeof(int), 347 .mode = 0644, 348 .proc_handler = proc_do_dev_weight, 349 }, 350 { 351 .procname = "dev_weight_rx_bias", 352 .data = &dev_weight_rx_bias, 353 .maxlen = sizeof(int), 354 .mode = 0644, 355 .proc_handler = proc_do_dev_weight, 356 }, 357 { 358 .procname = "dev_weight_tx_bias", 359 .data = &dev_weight_tx_bias, 360 .maxlen = sizeof(int), 361 .mode = 0644, 362 .proc_handler = proc_do_dev_weight, 363 }, 364 { 365 .procname = "netdev_max_backlog", 366 .data = &netdev_max_backlog, 367 .maxlen = sizeof(int), 368 .mode = 0644, 369 .proc_handler = proc_dointvec 370 }, 371 { 372 .procname = "netdev_rss_key", 373 .data = &netdev_rss_key, 374 .maxlen = sizeof(int), 375 .mode = 0444, 376 .proc_handler = proc_do_rss_key, 377 }, 378 #ifdef CONFIG_BPF_JIT 379 { 380 .procname = "bpf_jit_enable", 381 .data = &bpf_jit_enable, 382 .maxlen = sizeof(int), 383 .mode = 0644, 384 .proc_handler = proc_dointvec_minmax_bpf_enable, 385 # ifdef CONFIG_BPF_JIT_ALWAYS_ON 386 .extra1 = SYSCTL_ONE, 387 .extra2 = SYSCTL_ONE, 388 # else 389 .extra1 = SYSCTL_ZERO, 390 .extra2 = SYSCTL_TWO, 391 # endif 392 }, 393 # ifdef CONFIG_HAVE_EBPF_JIT 394 { 395 .procname = "bpf_jit_harden", 396 .data = &bpf_jit_harden, 397 .maxlen = sizeof(int), 398 .mode = 0600, 399 .proc_handler = proc_dointvec_minmax_bpf_restricted, 400 .extra1 = SYSCTL_ZERO, 401 .extra2 = SYSCTL_TWO, 402 }, 403 { 404 .procname = "bpf_jit_kallsyms", 405 .data = &bpf_jit_kallsyms, 406 .maxlen = sizeof(int), 407 .mode = 0600, 408 .proc_handler = proc_dointvec_minmax_bpf_restricted, 409 .extra1 = SYSCTL_ZERO, 410 .extra2 = SYSCTL_ONE, 411 }, 412 # endif 413 { 414 .procname = "bpf_jit_limit", 415 .data = &bpf_jit_limit, 416 .maxlen = sizeof(long), 417 .mode = 0600, 418 .proc_handler = proc_dolongvec_minmax_bpf_restricted, 419 .extra1 = SYSCTL_LONG_ONE, 420 .extra2 = &bpf_jit_limit_max, 421 }, 422 #endif 423 { 424 .procname = "netdev_tstamp_prequeue", 425 .data = &netdev_tstamp_prequeue, 426 .maxlen = sizeof(int), 427 .mode = 0644, 428 .proc_handler = proc_dointvec 429 }, 430 { 431 .procname = "message_cost", 432 .data = &net_ratelimit_state.interval, 433 .maxlen = sizeof(int), 434 .mode = 0644, 435 .proc_handler = proc_dointvec_jiffies, 436 }, 437 { 438 .procname = "message_burst", 439 .data = &net_ratelimit_state.burst, 440 .maxlen = sizeof(int), 441 .mode = 0644, 442 .proc_handler = proc_dointvec, 443 }, 444 { 445 .procname = "optmem_max", 446 .data = &sysctl_optmem_max, 447 .maxlen = sizeof(int), 448 .mode = 0644, 449 .proc_handler = proc_dointvec 450 }, 451 { 452 .procname = "tstamp_allow_data", 453 .data = &sysctl_tstamp_allow_data, 454 .maxlen = sizeof(int), 455 .mode = 0644, 456 .proc_handler = proc_dointvec_minmax, 457 .extra1 = SYSCTL_ZERO, 458 .extra2 = SYSCTL_ONE 459 }, 460 #ifdef CONFIG_RPS 461 { 462 .procname = "rps_sock_flow_entries", 463 .maxlen = sizeof(int), 464 .mode = 0644, 465 .proc_handler = rps_sock_flow_sysctl 466 }, 467 #endif 468 #ifdef CONFIG_NET_FLOW_LIMIT 469 { 470 .procname = "flow_limit_cpu_bitmap", 471 .mode = 0644, 472 .proc_handler = flow_limit_cpu_sysctl 473 }, 474 { 475 .procname = "flow_limit_table_len", 476 .data = &netdev_flow_limit_table_len, 477 .maxlen = sizeof(int), 478 .mode = 0644, 479 .proc_handler = flow_limit_table_len_sysctl 480 }, 481 #endif /* CONFIG_NET_FLOW_LIMIT */ 482 #ifdef CONFIG_NET_RX_BUSY_POLL 483 { 484 .procname = "busy_poll", 485 .data = &sysctl_net_busy_poll, 486 .maxlen = sizeof(unsigned int), 487 .mode = 0644, 488 .proc_handler = proc_dointvec_minmax, 489 .extra1 = SYSCTL_ZERO, 490 }, 491 { 492 .procname = "busy_read", 493 .data = &sysctl_net_busy_read, 494 .maxlen = sizeof(unsigned int), 495 .mode = 0644, 496 .proc_handler = proc_dointvec_minmax, 497 .extra1 = SYSCTL_ZERO, 498 }, 499 #endif 500 #ifdef CONFIG_NET_SCHED 501 { 502 .procname = "default_qdisc", 503 .mode = 0644, 504 .maxlen = IFNAMSIZ, 505 .proc_handler = set_default_qdisc 506 }, 507 #endif 508 { 509 .procname = "netdev_budget", 510 .data = &netdev_budget, 511 .maxlen = sizeof(int), 512 .mode = 0644, 513 .proc_handler = proc_dointvec 514 }, 515 { 516 .procname = "warnings", 517 .data = &net_msg_warn, 518 .maxlen = sizeof(int), 519 .mode = 0644, 520 .proc_handler = proc_dointvec 521 }, 522 { 523 .procname = "max_skb_frags", 524 .data = &sysctl_max_skb_frags, 525 .maxlen = sizeof(int), 526 .mode = 0644, 527 .proc_handler = proc_dointvec_minmax, 528 .extra1 = SYSCTL_ONE, 529 .extra2 = &max_skb_frags, 530 }, 531 { 532 .procname = "netdev_budget_usecs", 533 .data = &netdev_budget_usecs, 534 .maxlen = sizeof(unsigned int), 535 .mode = 0644, 536 .proc_handler = proc_dointvec_minmax, 537 .extra1 = SYSCTL_ZERO, 538 }, 539 { 540 .procname = "fb_tunnels_only_for_init_net", 541 .data = &sysctl_fb_tunnels_only_for_init_net, 542 .maxlen = sizeof(int), 543 .mode = 0644, 544 .proc_handler = proc_dointvec_minmax, 545 .extra1 = SYSCTL_ZERO, 546 .extra2 = SYSCTL_TWO, 547 }, 548 { 549 .procname = "devconf_inherit_init_net", 550 .data = &sysctl_devconf_inherit_init_net, 551 .maxlen = sizeof(int), 552 .mode = 0644, 553 .proc_handler = proc_dointvec_minmax, 554 .extra1 = SYSCTL_ZERO, 555 .extra2 = SYSCTL_THREE, 556 }, 557 { 558 .procname = "high_order_alloc_disable", 559 .data = &net_high_order_alloc_disable_key.key, 560 .maxlen = sizeof(net_high_order_alloc_disable_key), 561 .mode = 0644, 562 .proc_handler = proc_do_static_key, 563 }, 564 { 565 .procname = "gro_normal_batch", 566 .data = &gro_normal_batch, 567 .maxlen = sizeof(unsigned int), 568 .mode = 0644, 569 .proc_handler = proc_dointvec_minmax, 570 .extra1 = SYSCTL_ONE, 571 }, 572 { 573 .procname = "netdev_unregister_timeout_secs", 574 .data = &netdev_unregister_timeout_secs, 575 .maxlen = sizeof(unsigned int), 576 .mode = 0644, 577 .proc_handler = proc_dointvec_minmax, 578 .extra1 = SYSCTL_ONE, 579 .extra2 = &int_3600, 580 }, 581 { 582 .procname = "skb_defer_max", 583 .data = &sysctl_skb_defer_max, 584 .maxlen = sizeof(unsigned int), 585 .mode = 0644, 586 .proc_handler = proc_dointvec_minmax, 587 .extra1 = SYSCTL_ZERO, 588 }, 589 { } 590 }; 591 592 static struct ctl_table netns_core_table[] = { 593 { 594 .procname = "somaxconn", 595 .data = &init_net.core.sysctl_somaxconn, 596 .maxlen = sizeof(int), 597 .mode = 0644, 598 .extra1 = SYSCTL_ZERO, 599 .proc_handler = proc_dointvec_minmax 600 }, 601 { 602 .procname = "txrehash", 603 .data = &init_net.core.sysctl_txrehash, 604 .maxlen = sizeof(u8), 605 .mode = 0644, 606 .extra1 = SYSCTL_ZERO, 607 .extra2 = SYSCTL_ONE, 608 .proc_handler = proc_dou8vec_minmax, 609 }, 610 { } 611 }; 612 613 static int __init fb_tunnels_only_for_init_net_sysctl_setup(char *str) 614 { 615 /* fallback tunnels for initns only */ 616 if (!strncmp(str, "initns", 6)) 617 sysctl_fb_tunnels_only_for_init_net = 1; 618 /* no fallback tunnels anywhere */ 619 else if (!strncmp(str, "none", 4)) 620 sysctl_fb_tunnels_only_for_init_net = 2; 621 622 return 1; 623 } 624 __setup("fb_tunnels=", fb_tunnels_only_for_init_net_sysctl_setup); 625 626 static __net_init int sysctl_core_net_init(struct net *net) 627 { 628 struct ctl_table *tbl, *tmp; 629 630 tbl = netns_core_table; 631 if (!net_eq(net, &init_net)) { 632 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL); 633 if (tbl == NULL) 634 goto err_dup; 635 636 for (tmp = tbl; tmp->procname; tmp++) 637 tmp->data += (char *)net - (char *)&init_net; 638 639 /* Don't export any sysctls to unprivileged users */ 640 if (net->user_ns != &init_user_ns) { 641 tbl[0].procname = NULL; 642 } 643 } 644 645 net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl); 646 if (net->core.sysctl_hdr == NULL) 647 goto err_reg; 648 649 return 0; 650 651 err_reg: 652 if (tbl != netns_core_table) 653 kfree(tbl); 654 err_dup: 655 return -ENOMEM; 656 } 657 658 static __net_exit void sysctl_core_net_exit(struct net *net) 659 { 660 struct ctl_table *tbl; 661 662 tbl = net->core.sysctl_hdr->ctl_table_arg; 663 unregister_net_sysctl_table(net->core.sysctl_hdr); 664 BUG_ON(tbl == netns_core_table); 665 kfree(tbl); 666 } 667 668 static __net_initdata struct pernet_operations sysctl_core_ops = { 669 .init = sysctl_core_net_init, 670 .exit = sysctl_core_net_exit, 671 }; 672 673 static __init int sysctl_core_init(void) 674 { 675 register_net_sysctl(&init_net, "net/core", net_core_table); 676 return register_pernet_subsys(&sysctl_core_ops); 677 } 678 679 fs_initcall(sysctl_core_init); 680