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 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 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 kvfree_rcu(orig_sock_table); 105 } 106 } 107 } 108 109 mutex_unlock(&sock_flow_mutex); 110 111 return ret; 112 } 113 #endif /* CONFIG_RPS */ 114 115 #ifdef CONFIG_NET_FLOW_LIMIT 116 static DEFINE_MUTEX(flow_limit_update_mutex); 117 118 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write, 119 void *buffer, size_t *lenp, loff_t *ppos) 120 { 121 struct sd_flow_limit *cur; 122 struct softnet_data *sd; 123 cpumask_var_t mask; 124 int i, len, ret = 0; 125 126 if (!alloc_cpumask_var(&mask, GFP_KERNEL)) 127 return -ENOMEM; 128 129 if (write) { 130 ret = cpumask_parse(buffer, mask); 131 if (ret) 132 goto done; 133 134 mutex_lock(&flow_limit_update_mutex); 135 len = sizeof(*cur) + netdev_flow_limit_table_len; 136 for_each_possible_cpu(i) { 137 sd = &per_cpu(softnet_data, i); 138 cur = rcu_dereference_protected(sd->flow_limit, 139 lockdep_is_held(&flow_limit_update_mutex)); 140 if (cur && !cpumask_test_cpu(i, mask)) { 141 RCU_INIT_POINTER(sd->flow_limit, NULL); 142 kfree_rcu(cur); 143 } else if (!cur && cpumask_test_cpu(i, mask)) { 144 cur = kzalloc_node(len, GFP_KERNEL, 145 cpu_to_node(i)); 146 if (!cur) { 147 /* not unwinding previous changes */ 148 ret = -ENOMEM; 149 goto write_unlock; 150 } 151 cur->num_buckets = netdev_flow_limit_table_len; 152 rcu_assign_pointer(sd->flow_limit, cur); 153 } 154 } 155 write_unlock: 156 mutex_unlock(&flow_limit_update_mutex); 157 } else { 158 char kbuf[128]; 159 160 if (*ppos || !*lenp) { 161 *lenp = 0; 162 goto done; 163 } 164 165 cpumask_clear(mask); 166 rcu_read_lock(); 167 for_each_possible_cpu(i) { 168 sd = &per_cpu(softnet_data, i); 169 if (rcu_dereference(sd->flow_limit)) 170 cpumask_set_cpu(i, mask); 171 } 172 rcu_read_unlock(); 173 174 len = min(sizeof(kbuf) - 1, *lenp); 175 len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask)); 176 if (!len) { 177 *lenp = 0; 178 goto done; 179 } 180 if (len < *lenp) 181 kbuf[len++] = '\n'; 182 memcpy(buffer, kbuf, len); 183 *lenp = len; 184 *ppos += len; 185 } 186 187 done: 188 free_cpumask_var(mask); 189 return ret; 190 } 191 192 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write, 193 void *buffer, size_t *lenp, loff_t *ppos) 194 { 195 unsigned int old, *ptr; 196 int ret; 197 198 mutex_lock(&flow_limit_update_mutex); 199 200 ptr = table->data; 201 old = *ptr; 202 ret = proc_dointvec(table, write, buffer, lenp, ppos); 203 if (!ret && write && !is_power_of_2(*ptr)) { 204 *ptr = old; 205 ret = -EINVAL; 206 } 207 208 mutex_unlock(&flow_limit_update_mutex); 209 return ret; 210 } 211 #endif /* CONFIG_NET_FLOW_LIMIT */ 212 213 #ifdef CONFIG_NET_SCHED 214 static int set_default_qdisc(struct ctl_table *table, int write, 215 void *buffer, size_t *lenp, loff_t *ppos) 216 { 217 char id[IFNAMSIZ]; 218 struct ctl_table tbl = { 219 .data = id, 220 .maxlen = IFNAMSIZ, 221 }; 222 int ret; 223 224 qdisc_get_default(id, IFNAMSIZ); 225 226 ret = proc_dostring(&tbl, write, buffer, lenp, ppos); 227 if (write && ret == 0) 228 ret = qdisc_set_default(id); 229 return ret; 230 } 231 #endif 232 233 static int proc_do_dev_weight(struct ctl_table *table, int write, 234 void *buffer, size_t *lenp, loff_t *ppos) 235 { 236 static DEFINE_MUTEX(dev_weight_mutex); 237 int ret, weight; 238 239 mutex_lock(&dev_weight_mutex); 240 ret = proc_dointvec(table, write, buffer, lenp, ppos); 241 if (!ret && write) { 242 weight = READ_ONCE(weight_p); 243 WRITE_ONCE(dev_rx_weight, weight * dev_weight_rx_bias); 244 WRITE_ONCE(dev_tx_weight, weight * dev_weight_tx_bias); 245 } 246 mutex_unlock(&dev_weight_mutex); 247 248 return ret; 249 } 250 251 static int proc_do_rss_key(struct ctl_table *table, int write, 252 void *buffer, size_t *lenp, loff_t *ppos) 253 { 254 struct ctl_table fake_table; 255 char buf[NETDEV_RSS_KEY_LEN * 3]; 256 257 snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key); 258 fake_table.data = buf; 259 fake_table.maxlen = sizeof(buf); 260 return proc_dostring(&fake_table, write, buffer, lenp, ppos); 261 } 262 263 #ifdef CONFIG_BPF_JIT 264 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write, 265 void *buffer, size_t *lenp, 266 loff_t *ppos) 267 { 268 int ret, jit_enable = *(int *)table->data; 269 int min = *(int *)table->extra1; 270 int max = *(int *)table->extra2; 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 } 288 289 if (write && ret && min == max) 290 pr_info_once("CONFIG_BPF_JIT_ALWAYS_ON is enabled, bpf_jit_enable is permanently set to 1.\n"); 291 292 return ret; 293 } 294 295 # ifdef CONFIG_HAVE_EBPF_JIT 296 static int 297 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write, 298 void *buffer, size_t *lenp, loff_t *ppos) 299 { 300 if (!capable(CAP_SYS_ADMIN)) 301 return -EPERM; 302 303 return proc_dointvec_minmax(table, write, buffer, lenp, ppos); 304 } 305 # endif /* CONFIG_HAVE_EBPF_JIT */ 306 307 static int 308 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write, 309 void *buffer, size_t *lenp, loff_t *ppos) 310 { 311 if (!capable(CAP_SYS_ADMIN)) 312 return -EPERM; 313 314 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); 315 } 316 #endif 317 318 static struct ctl_table net_core_table[] = { 319 { 320 .procname = "wmem_max", 321 .data = &sysctl_wmem_max, 322 .maxlen = sizeof(int), 323 .mode = 0644, 324 .proc_handler = proc_dointvec_minmax, 325 .extra1 = &min_sndbuf, 326 }, 327 { 328 .procname = "rmem_max", 329 .data = &sysctl_rmem_max, 330 .maxlen = sizeof(int), 331 .mode = 0644, 332 .proc_handler = proc_dointvec_minmax, 333 .extra1 = &min_rcvbuf, 334 }, 335 { 336 .procname = "wmem_default", 337 .data = &sysctl_wmem_default, 338 .maxlen = sizeof(int), 339 .mode = 0644, 340 .proc_handler = proc_dointvec_minmax, 341 .extra1 = &min_sndbuf, 342 }, 343 { 344 .procname = "rmem_default", 345 .data = &sysctl_rmem_default, 346 .maxlen = sizeof(int), 347 .mode = 0644, 348 .proc_handler = proc_dointvec_minmax, 349 .extra1 = &min_rcvbuf, 350 }, 351 { 352 .procname = "dev_weight", 353 .data = &weight_p, 354 .maxlen = sizeof(int), 355 .mode = 0644, 356 .proc_handler = proc_do_dev_weight, 357 }, 358 { 359 .procname = "dev_weight_rx_bias", 360 .data = &dev_weight_rx_bias, 361 .maxlen = sizeof(int), 362 .mode = 0644, 363 .proc_handler = proc_do_dev_weight, 364 }, 365 { 366 .procname = "dev_weight_tx_bias", 367 .data = &dev_weight_tx_bias, 368 .maxlen = sizeof(int), 369 .mode = 0644, 370 .proc_handler = proc_do_dev_weight, 371 }, 372 { 373 .procname = "netdev_max_backlog", 374 .data = &netdev_max_backlog, 375 .maxlen = sizeof(int), 376 .mode = 0644, 377 .proc_handler = proc_dointvec 378 }, 379 { 380 .procname = "netdev_rss_key", 381 .data = &netdev_rss_key, 382 .maxlen = sizeof(int), 383 .mode = 0444, 384 .proc_handler = proc_do_rss_key, 385 }, 386 #ifdef CONFIG_BPF_JIT 387 { 388 .procname = "bpf_jit_enable", 389 .data = &bpf_jit_enable, 390 .maxlen = sizeof(int), 391 .mode = 0644, 392 .proc_handler = proc_dointvec_minmax_bpf_enable, 393 # ifdef CONFIG_BPF_JIT_ALWAYS_ON 394 .extra1 = SYSCTL_ONE, 395 .extra2 = SYSCTL_ONE, 396 # else 397 .extra1 = SYSCTL_ZERO, 398 .extra2 = SYSCTL_TWO, 399 # endif 400 }, 401 # ifdef CONFIG_HAVE_EBPF_JIT 402 { 403 .procname = "bpf_jit_harden", 404 .data = &bpf_jit_harden, 405 .maxlen = sizeof(int), 406 .mode = 0600, 407 .proc_handler = proc_dointvec_minmax_bpf_restricted, 408 .extra1 = SYSCTL_ZERO, 409 .extra2 = SYSCTL_TWO, 410 }, 411 { 412 .procname = "bpf_jit_kallsyms", 413 .data = &bpf_jit_kallsyms, 414 .maxlen = sizeof(int), 415 .mode = 0600, 416 .proc_handler = proc_dointvec_minmax_bpf_restricted, 417 .extra1 = SYSCTL_ZERO, 418 .extra2 = SYSCTL_ONE, 419 }, 420 # endif 421 { 422 .procname = "bpf_jit_limit", 423 .data = &bpf_jit_limit, 424 .maxlen = sizeof(long), 425 .mode = 0600, 426 .proc_handler = proc_dolongvec_minmax_bpf_restricted, 427 .extra1 = SYSCTL_LONG_ONE, 428 .extra2 = &bpf_jit_limit_max, 429 }, 430 #endif 431 { 432 .procname = "netdev_tstamp_prequeue", 433 .data = &netdev_tstamp_prequeue, 434 .maxlen = sizeof(int), 435 .mode = 0644, 436 .proc_handler = proc_dointvec 437 }, 438 { 439 .procname = "message_cost", 440 .data = &net_ratelimit_state.interval, 441 .maxlen = sizeof(int), 442 .mode = 0644, 443 .proc_handler = proc_dointvec_jiffies, 444 }, 445 { 446 .procname = "message_burst", 447 .data = &net_ratelimit_state.burst, 448 .maxlen = sizeof(int), 449 .mode = 0644, 450 .proc_handler = proc_dointvec, 451 }, 452 { 453 .procname = "optmem_max", 454 .data = &sysctl_optmem_max, 455 .maxlen = sizeof(int), 456 .mode = 0644, 457 .proc_handler = proc_dointvec 458 }, 459 { 460 .procname = "tstamp_allow_data", 461 .data = &sysctl_tstamp_allow_data, 462 .maxlen = sizeof(int), 463 .mode = 0644, 464 .proc_handler = proc_dointvec_minmax, 465 .extra1 = SYSCTL_ZERO, 466 .extra2 = SYSCTL_ONE 467 }, 468 #ifdef CONFIG_RPS 469 { 470 .procname = "rps_sock_flow_entries", 471 .maxlen = sizeof(int), 472 .mode = 0644, 473 .proc_handler = rps_sock_flow_sysctl 474 }, 475 #endif 476 #ifdef CONFIG_NET_FLOW_LIMIT 477 { 478 .procname = "flow_limit_cpu_bitmap", 479 .mode = 0644, 480 .proc_handler = flow_limit_cpu_sysctl 481 }, 482 { 483 .procname = "flow_limit_table_len", 484 .data = &netdev_flow_limit_table_len, 485 .maxlen = sizeof(int), 486 .mode = 0644, 487 .proc_handler = flow_limit_table_len_sysctl 488 }, 489 #endif /* CONFIG_NET_FLOW_LIMIT */ 490 #ifdef CONFIG_NET_RX_BUSY_POLL 491 { 492 .procname = "busy_poll", 493 .data = &sysctl_net_busy_poll, 494 .maxlen = sizeof(unsigned int), 495 .mode = 0644, 496 .proc_handler = proc_dointvec_minmax, 497 .extra1 = SYSCTL_ZERO, 498 }, 499 { 500 .procname = "busy_read", 501 .data = &sysctl_net_busy_read, 502 .maxlen = sizeof(unsigned int), 503 .mode = 0644, 504 .proc_handler = proc_dointvec_minmax, 505 .extra1 = SYSCTL_ZERO, 506 }, 507 #endif 508 #ifdef CONFIG_NET_SCHED 509 { 510 .procname = "default_qdisc", 511 .mode = 0644, 512 .maxlen = IFNAMSIZ, 513 .proc_handler = set_default_qdisc 514 }, 515 #endif 516 { 517 .procname = "netdev_budget", 518 .data = &netdev_budget, 519 .maxlen = sizeof(int), 520 .mode = 0644, 521 .proc_handler = proc_dointvec 522 }, 523 { 524 .procname = "warnings", 525 .data = &net_msg_warn, 526 .maxlen = sizeof(int), 527 .mode = 0644, 528 .proc_handler = proc_dointvec 529 }, 530 { 531 .procname = "max_skb_frags", 532 .data = &sysctl_max_skb_frags, 533 .maxlen = sizeof(int), 534 .mode = 0644, 535 .proc_handler = proc_dointvec_minmax, 536 .extra1 = SYSCTL_ONE, 537 .extra2 = &max_skb_frags, 538 }, 539 { 540 .procname = "netdev_budget_usecs", 541 .data = &netdev_budget_usecs, 542 .maxlen = sizeof(unsigned int), 543 .mode = 0644, 544 .proc_handler = proc_dointvec_minmax, 545 .extra1 = SYSCTL_ZERO, 546 }, 547 { 548 .procname = "fb_tunnels_only_for_init_net", 549 .data = &sysctl_fb_tunnels_only_for_init_net, 550 .maxlen = sizeof(int), 551 .mode = 0644, 552 .proc_handler = proc_dointvec_minmax, 553 .extra1 = SYSCTL_ZERO, 554 .extra2 = SYSCTL_TWO, 555 }, 556 { 557 .procname = "devconf_inherit_init_net", 558 .data = &sysctl_devconf_inherit_init_net, 559 .maxlen = sizeof(int), 560 .mode = 0644, 561 .proc_handler = proc_dointvec_minmax, 562 .extra1 = SYSCTL_ZERO, 563 .extra2 = SYSCTL_THREE, 564 }, 565 { 566 .procname = "high_order_alloc_disable", 567 .data = &net_high_order_alloc_disable_key.key, 568 .maxlen = sizeof(net_high_order_alloc_disable_key), 569 .mode = 0644, 570 .proc_handler = proc_do_static_key, 571 }, 572 { 573 .procname = "gro_normal_batch", 574 .data = &gro_normal_batch, 575 .maxlen = sizeof(unsigned int), 576 .mode = 0644, 577 .proc_handler = proc_dointvec_minmax, 578 .extra1 = SYSCTL_ONE, 579 }, 580 { 581 .procname = "netdev_unregister_timeout_secs", 582 .data = &netdev_unregister_timeout_secs, 583 .maxlen = sizeof(unsigned int), 584 .mode = 0644, 585 .proc_handler = proc_dointvec_minmax, 586 .extra1 = SYSCTL_ONE, 587 .extra2 = &int_3600, 588 }, 589 { 590 .procname = "skb_defer_max", 591 .data = &sysctl_skb_defer_max, 592 .maxlen = sizeof(unsigned int), 593 .mode = 0644, 594 .proc_handler = proc_dointvec_minmax, 595 .extra1 = SYSCTL_ZERO, 596 }, 597 { } 598 }; 599 600 static struct ctl_table netns_core_table[] = { 601 { 602 .procname = "somaxconn", 603 .data = &init_net.core.sysctl_somaxconn, 604 .maxlen = sizeof(int), 605 .mode = 0644, 606 .extra1 = SYSCTL_ZERO, 607 .proc_handler = proc_dointvec_minmax 608 }, 609 { 610 .procname = "txrehash", 611 .data = &init_net.core.sysctl_txrehash, 612 .maxlen = sizeof(u8), 613 .mode = 0644, 614 .extra1 = SYSCTL_ZERO, 615 .extra2 = SYSCTL_ONE, 616 .proc_handler = proc_dou8vec_minmax, 617 }, 618 { } 619 }; 620 621 static int __init fb_tunnels_only_for_init_net_sysctl_setup(char *str) 622 { 623 /* fallback tunnels for initns only */ 624 if (!strncmp(str, "initns", 6)) 625 sysctl_fb_tunnels_only_for_init_net = 1; 626 /* no fallback tunnels anywhere */ 627 else if (!strncmp(str, "none", 4)) 628 sysctl_fb_tunnels_only_for_init_net = 2; 629 630 return 1; 631 } 632 __setup("fb_tunnels=", fb_tunnels_only_for_init_net_sysctl_setup); 633 634 static __net_init int sysctl_core_net_init(struct net *net) 635 { 636 struct ctl_table *tbl, *tmp; 637 638 tbl = netns_core_table; 639 if (!net_eq(net, &init_net)) { 640 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL); 641 if (tbl == NULL) 642 goto err_dup; 643 644 for (tmp = tbl; tmp->procname; tmp++) 645 tmp->data += (char *)net - (char *)&init_net; 646 647 /* Don't export any sysctls to unprivileged users */ 648 if (net->user_ns != &init_user_ns) { 649 tbl[0].procname = NULL; 650 } 651 } 652 653 net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl); 654 if (net->core.sysctl_hdr == NULL) 655 goto err_reg; 656 657 return 0; 658 659 err_reg: 660 if (tbl != netns_core_table) 661 kfree(tbl); 662 err_dup: 663 return -ENOMEM; 664 } 665 666 static __net_exit void sysctl_core_net_exit(struct net *net) 667 { 668 struct ctl_table *tbl; 669 670 tbl = net->core.sysctl_hdr->ctl_table_arg; 671 unregister_net_sysctl_table(net->core.sysctl_hdr); 672 BUG_ON(tbl == netns_core_table); 673 kfree(tbl); 674 } 675 676 static __net_initdata struct pernet_operations sysctl_core_ops = { 677 .init = sysctl_core_net_init, 678 .exit = sysctl_core_net_exit, 679 }; 680 681 static __init int sysctl_core_init(void) 682 { 683 register_net_sysctl(&init_net, "net/core", net_core_table); 684 return register_pernet_subsys(&sysctl_core_ops); 685 } 686 687 fs_initcall(sysctl_core_init); 688