1 // SPDX-License-Identifier: GPL-2.0 2 /* -*- linux-c -*- 3 * sysctl_net_core.c: sysctl interface to net core subsystem. 4 * 5 * Begun April 1, 1996, Mike Shaver. 6 * Added /proc/sys/net/core directory entry (empty =) ). [MS] 7 */ 8 9 #include <linux/mm.h> 10 #include <linux/sysctl.h> 11 #include <linux/module.h> 12 #include <linux/socket.h> 13 #include <linux/netdevice.h> 14 #include <linux/ratelimit.h> 15 #include <linux/vmalloc.h> 16 #include <linux/init.h> 17 #include <linux/slab.h> 18 19 #include <net/ip.h> 20 #include <net/sock.h> 21 #include <net/net_ratelimit.h> 22 #include <net/busy_poll.h> 23 #include <net/pkt_sched.h> 24 25 static int two __maybe_unused = 2; 26 static int min_sndbuf = SOCK_MIN_SNDBUF; 27 static int min_rcvbuf = SOCK_MIN_RCVBUF; 28 static int max_skb_frags = MAX_SKB_FRAGS; 29 static long long_one __maybe_unused = 1; 30 static long long_max __maybe_unused = LONG_MAX; 31 32 static int net_msg_warn; /* Unused, but still a sysctl */ 33 34 int sysctl_fb_tunnels_only_for_init_net __read_mostly = 0; 35 EXPORT_SYMBOL(sysctl_fb_tunnels_only_for_init_net); 36 37 /* 0 - Keep current behavior: 38 * IPv4: inherit all current settings from init_net 39 * IPv6: reset all settings to default 40 * 1 - Both inherit all current settings from init_net 41 * 2 - Both reset all settings to default 42 */ 43 int sysctl_devconf_inherit_init_net __read_mostly; 44 EXPORT_SYMBOL(sysctl_devconf_inherit_init_net); 45 46 #ifdef CONFIG_RPS 47 static int rps_sock_flow_sysctl(struct ctl_table *table, int write, 48 void __user *buffer, size_t *lenp, loff_t *ppos) 49 { 50 unsigned int orig_size, size; 51 int ret, i; 52 struct ctl_table tmp = { 53 .data = &size, 54 .maxlen = sizeof(size), 55 .mode = table->mode 56 }; 57 struct rps_sock_flow_table *orig_sock_table, *sock_table; 58 static DEFINE_MUTEX(sock_flow_mutex); 59 60 mutex_lock(&sock_flow_mutex); 61 62 orig_sock_table = rcu_dereference_protected(rps_sock_flow_table, 63 lockdep_is_held(&sock_flow_mutex)); 64 size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0; 65 66 ret = proc_dointvec(&tmp, write, buffer, lenp, ppos); 67 68 if (write) { 69 if (size) { 70 if (size > 1<<29) { 71 /* Enforce limit to prevent overflow */ 72 mutex_unlock(&sock_flow_mutex); 73 return -EINVAL; 74 } 75 size = roundup_pow_of_two(size); 76 if (size != orig_size) { 77 sock_table = 78 vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size)); 79 if (!sock_table) { 80 mutex_unlock(&sock_flow_mutex); 81 return -ENOMEM; 82 } 83 rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1; 84 sock_table->mask = size - 1; 85 } else 86 sock_table = orig_sock_table; 87 88 for (i = 0; i < size; i++) 89 sock_table->ents[i] = RPS_NO_CPU; 90 } else 91 sock_table = NULL; 92 93 if (sock_table != orig_sock_table) { 94 rcu_assign_pointer(rps_sock_flow_table, sock_table); 95 if (sock_table) { 96 static_branch_inc(&rps_needed); 97 static_branch_inc(&rfs_needed); 98 } 99 if (orig_sock_table) { 100 static_branch_dec(&rps_needed); 101 static_branch_dec(&rfs_needed); 102 synchronize_rcu(); 103 vfree(orig_sock_table); 104 } 105 } 106 } 107 108 mutex_unlock(&sock_flow_mutex); 109 110 return ret; 111 } 112 #endif /* CONFIG_RPS */ 113 114 #ifdef CONFIG_NET_FLOW_LIMIT 115 static DEFINE_MUTEX(flow_limit_update_mutex); 116 117 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write, 118 void __user *buffer, size_t *lenp, 119 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_user(buffer, *lenp, 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 synchronize_rcu(); 143 kfree(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 if (copy_to_user(buffer, kbuf, len)) { 184 ret = -EFAULT; 185 goto done; 186 } 187 *lenp = len; 188 *ppos += len; 189 } 190 191 done: 192 free_cpumask_var(mask); 193 return ret; 194 } 195 196 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write, 197 void __user *buffer, size_t *lenp, 198 loff_t *ppos) 199 { 200 unsigned int old, *ptr; 201 int ret; 202 203 mutex_lock(&flow_limit_update_mutex); 204 205 ptr = table->data; 206 old = *ptr; 207 ret = proc_dointvec(table, write, buffer, lenp, ppos); 208 if (!ret && write && !is_power_of_2(*ptr)) { 209 *ptr = old; 210 ret = -EINVAL; 211 } 212 213 mutex_unlock(&flow_limit_update_mutex); 214 return ret; 215 } 216 #endif /* CONFIG_NET_FLOW_LIMIT */ 217 218 #ifdef CONFIG_NET_SCHED 219 static int set_default_qdisc(struct ctl_table *table, int write, 220 void __user *buffer, size_t *lenp, loff_t *ppos) 221 { 222 char id[IFNAMSIZ]; 223 struct ctl_table tbl = { 224 .data = id, 225 .maxlen = IFNAMSIZ, 226 }; 227 int ret; 228 229 qdisc_get_default(id, IFNAMSIZ); 230 231 ret = proc_dostring(&tbl, write, buffer, lenp, ppos); 232 if (write && ret == 0) 233 ret = qdisc_set_default(id); 234 return ret; 235 } 236 #endif 237 238 static int proc_do_dev_weight(struct ctl_table *table, int write, 239 void __user *buffer, size_t *lenp, loff_t *ppos) 240 { 241 int ret; 242 243 ret = proc_dointvec(table, write, buffer, lenp, ppos); 244 if (ret != 0) 245 return ret; 246 247 dev_rx_weight = weight_p * dev_weight_rx_bias; 248 dev_tx_weight = weight_p * dev_weight_tx_bias; 249 250 return ret; 251 } 252 253 static int proc_do_rss_key(struct ctl_table *table, int write, 254 void __user *buffer, size_t *lenp, loff_t *ppos) 255 { 256 struct ctl_table fake_table; 257 char buf[NETDEV_RSS_KEY_LEN * 3]; 258 259 snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key); 260 fake_table.data = buf; 261 fake_table.maxlen = sizeof(buf); 262 return proc_dostring(&fake_table, write, buffer, lenp, ppos); 263 } 264 265 #ifdef CONFIG_BPF_JIT 266 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write, 267 void __user *buffer, size_t *lenp, 268 loff_t *ppos) 269 { 270 int ret, jit_enable = *(int *)table->data; 271 struct ctl_table tmp = *table; 272 273 if (write && !capable(CAP_SYS_ADMIN)) 274 return -EPERM; 275 276 tmp.data = &jit_enable; 277 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); 278 if (write && !ret) { 279 if (jit_enable < 2 || 280 (jit_enable == 2 && bpf_dump_raw_ok())) { 281 *(int *)table->data = jit_enable; 282 if (jit_enable == 2) 283 pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n"); 284 } else { 285 ret = -EPERM; 286 } 287 } 288 return ret; 289 } 290 291 # ifdef CONFIG_HAVE_EBPF_JIT 292 static int 293 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write, 294 void __user *buffer, size_t *lenp, 295 loff_t *ppos) 296 { 297 if (!capable(CAP_SYS_ADMIN)) 298 return -EPERM; 299 300 return proc_dointvec_minmax(table, write, buffer, lenp, ppos); 301 } 302 # endif /* CONFIG_HAVE_EBPF_JIT */ 303 304 static int 305 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write, 306 void __user *buffer, size_t *lenp, 307 loff_t *ppos) 308 { 309 if (!capable(CAP_SYS_ADMIN)) 310 return -EPERM; 311 312 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); 313 } 314 #endif 315 316 static struct ctl_table net_core_table[] = { 317 #ifdef CONFIG_NET 318 { 319 .procname = "wmem_max", 320 .data = &sysctl_wmem_max, 321 .maxlen = sizeof(int), 322 .mode = 0644, 323 .proc_handler = proc_dointvec_minmax, 324 .extra1 = &min_sndbuf, 325 }, 326 { 327 .procname = "rmem_max", 328 .data = &sysctl_rmem_max, 329 .maxlen = sizeof(int), 330 .mode = 0644, 331 .proc_handler = proc_dointvec_minmax, 332 .extra1 = &min_rcvbuf, 333 }, 334 { 335 .procname = "wmem_default", 336 .data = &sysctl_wmem_default, 337 .maxlen = sizeof(int), 338 .mode = 0644, 339 .proc_handler = proc_dointvec_minmax, 340 .extra1 = &min_sndbuf, 341 }, 342 { 343 .procname = "rmem_default", 344 .data = &sysctl_rmem_default, 345 .maxlen = sizeof(int), 346 .mode = 0644, 347 .proc_handler = proc_dointvec_minmax, 348 .extra1 = &min_rcvbuf, 349 }, 350 { 351 .procname = "dev_weight", 352 .data = &weight_p, 353 .maxlen = sizeof(int), 354 .mode = 0644, 355 .proc_handler = proc_do_dev_weight, 356 }, 357 { 358 .procname = "dev_weight_rx_bias", 359 .data = &dev_weight_rx_bias, 360 .maxlen = sizeof(int), 361 .mode = 0644, 362 .proc_handler = proc_do_dev_weight, 363 }, 364 { 365 .procname = "dev_weight_tx_bias", 366 .data = &dev_weight_tx_bias, 367 .maxlen = sizeof(int), 368 .mode = 0644, 369 .proc_handler = proc_do_dev_weight, 370 }, 371 { 372 .procname = "netdev_max_backlog", 373 .data = &netdev_max_backlog, 374 .maxlen = sizeof(int), 375 .mode = 0644, 376 .proc_handler = proc_dointvec 377 }, 378 { 379 .procname = "netdev_rss_key", 380 .data = &netdev_rss_key, 381 .maxlen = sizeof(int), 382 .mode = 0444, 383 .proc_handler = proc_do_rss_key, 384 }, 385 #ifdef CONFIG_BPF_JIT 386 { 387 .procname = "bpf_jit_enable", 388 .data = &bpf_jit_enable, 389 .maxlen = sizeof(int), 390 .mode = 0644, 391 .proc_handler = proc_dointvec_minmax_bpf_enable, 392 # ifdef CONFIG_BPF_JIT_ALWAYS_ON 393 .extra1 = SYSCTL_ONE, 394 .extra2 = SYSCTL_ONE, 395 # else 396 .extra1 = SYSCTL_ZERO, 397 .extra2 = &two, 398 # endif 399 }, 400 # ifdef CONFIG_HAVE_EBPF_JIT 401 { 402 .procname = "bpf_jit_harden", 403 .data = &bpf_jit_harden, 404 .maxlen = sizeof(int), 405 .mode = 0600, 406 .proc_handler = proc_dointvec_minmax_bpf_restricted, 407 .extra1 = SYSCTL_ZERO, 408 .extra2 = &two, 409 }, 410 { 411 .procname = "bpf_jit_kallsyms", 412 .data = &bpf_jit_kallsyms, 413 .maxlen = sizeof(int), 414 .mode = 0600, 415 .proc_handler = proc_dointvec_minmax_bpf_restricted, 416 .extra1 = SYSCTL_ZERO, 417 .extra2 = SYSCTL_ONE, 418 }, 419 # endif 420 { 421 .procname = "bpf_jit_limit", 422 .data = &bpf_jit_limit, 423 .maxlen = sizeof(long), 424 .mode = 0600, 425 .proc_handler = proc_dolongvec_minmax_bpf_restricted, 426 .extra1 = &long_one, 427 .extra2 = &long_max, 428 }, 429 #endif 430 { 431 .procname = "netdev_tstamp_prequeue", 432 .data = &netdev_tstamp_prequeue, 433 .maxlen = sizeof(int), 434 .mode = 0644, 435 .proc_handler = proc_dointvec 436 }, 437 { 438 .procname = "message_cost", 439 .data = &net_ratelimit_state.interval, 440 .maxlen = sizeof(int), 441 .mode = 0644, 442 .proc_handler = proc_dointvec_jiffies, 443 }, 444 { 445 .procname = "message_burst", 446 .data = &net_ratelimit_state.burst, 447 .maxlen = sizeof(int), 448 .mode = 0644, 449 .proc_handler = proc_dointvec, 450 }, 451 { 452 .procname = "optmem_max", 453 .data = &sysctl_optmem_max, 454 .maxlen = sizeof(int), 455 .mode = 0644, 456 .proc_handler = proc_dointvec 457 }, 458 { 459 .procname = "tstamp_allow_data", 460 .data = &sysctl_tstamp_allow_data, 461 .maxlen = sizeof(int), 462 .mode = 0644, 463 .proc_handler = proc_dointvec_minmax, 464 .extra1 = SYSCTL_ZERO, 465 .extra2 = SYSCTL_ONE 466 }, 467 #ifdef CONFIG_RPS 468 { 469 .procname = "rps_sock_flow_entries", 470 .maxlen = sizeof(int), 471 .mode = 0644, 472 .proc_handler = rps_sock_flow_sysctl 473 }, 474 #endif 475 #ifdef CONFIG_NET_FLOW_LIMIT 476 { 477 .procname = "flow_limit_cpu_bitmap", 478 .mode = 0644, 479 .proc_handler = flow_limit_cpu_sysctl 480 }, 481 { 482 .procname = "flow_limit_table_len", 483 .data = &netdev_flow_limit_table_len, 484 .maxlen = sizeof(int), 485 .mode = 0644, 486 .proc_handler = flow_limit_table_len_sysctl 487 }, 488 #endif /* CONFIG_NET_FLOW_LIMIT */ 489 #ifdef CONFIG_NET_RX_BUSY_POLL 490 { 491 .procname = "busy_poll", 492 .data = &sysctl_net_busy_poll, 493 .maxlen = sizeof(unsigned int), 494 .mode = 0644, 495 .proc_handler = proc_dointvec_minmax, 496 .extra1 = SYSCTL_ZERO, 497 }, 498 { 499 .procname = "busy_read", 500 .data = &sysctl_net_busy_read, 501 .maxlen = sizeof(unsigned int), 502 .mode = 0644, 503 .proc_handler = proc_dointvec_minmax, 504 .extra1 = SYSCTL_ZERO, 505 }, 506 #endif 507 #ifdef CONFIG_NET_SCHED 508 { 509 .procname = "default_qdisc", 510 .mode = 0644, 511 .maxlen = IFNAMSIZ, 512 .proc_handler = set_default_qdisc 513 }, 514 #endif 515 #endif /* CONFIG_NET */ 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_ONE, 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 = &two, 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 }; 582 583 static struct ctl_table netns_core_table[] = { 584 { 585 .procname = "somaxconn", 586 .data = &init_net.core.sysctl_somaxconn, 587 .maxlen = sizeof(int), 588 .mode = 0644, 589 .extra1 = SYSCTL_ZERO, 590 .proc_handler = proc_dointvec_minmax 591 }, 592 { } 593 }; 594 595 static __net_init int sysctl_core_net_init(struct net *net) 596 { 597 struct ctl_table *tbl; 598 599 tbl = netns_core_table; 600 if (!net_eq(net, &init_net)) { 601 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL); 602 if (tbl == NULL) 603 goto err_dup; 604 605 tbl[0].data = &net->core.sysctl_somaxconn; 606 607 /* Don't export any sysctls to unprivileged users */ 608 if (net->user_ns != &init_user_ns) { 609 tbl[0].procname = NULL; 610 } 611 } 612 613 net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl); 614 if (net->core.sysctl_hdr == NULL) 615 goto err_reg; 616 617 return 0; 618 619 err_reg: 620 if (tbl != netns_core_table) 621 kfree(tbl); 622 err_dup: 623 return -ENOMEM; 624 } 625 626 static __net_exit void sysctl_core_net_exit(struct net *net) 627 { 628 struct ctl_table *tbl; 629 630 tbl = net->core.sysctl_hdr->ctl_table_arg; 631 unregister_net_sysctl_table(net->core.sysctl_hdr); 632 BUG_ON(tbl == netns_core_table); 633 kfree(tbl); 634 } 635 636 static __net_initdata struct pernet_operations sysctl_core_ops = { 637 .init = sysctl_core_net_init, 638 .exit = sysctl_core_net_exit, 639 }; 640 641 static __init int sysctl_core_init(void) 642 { 643 register_net_sysctl(&init_net, "net/core", net_core_table); 644 return register_pernet_subsys(&sysctl_core_ops); 645 } 646 647 fs_initcall(sysctl_core_init); 648