net/core/sysctl_net_core.c

   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 zero = 0;
  26 static int one = 1;
  27 static int two __maybe_unused = 2;
  28 static int min_sndbuf = SOCK_MIN_SNDBUF;
  29 static int min_rcvbuf = SOCK_MIN_RCVBUF;
  30 static int max_skb_frags = MAX_SKB_FRAGS;
  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 #ifdef CONFIG_RPS
  38 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
  39                                 void __user *buffer, size_t *lenp, loff_t *ppos)
  40 {
  41         unsigned int orig_size, size;
  42         int ret, i;
  43         struct ctl_table tmp = {
  44                 .data = &size,
  45                 .maxlen = sizeof(size),
  46                 .mode = table->mode
  47         };
  48         struct rps_sock_flow_table *orig_sock_table, *sock_table;
  49         static DEFINE_MUTEX(sock_flow_mutex);
  50
  51         mutex_lock(&sock_flow_mutex);
  52
  53         orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
  54                                         lockdep_is_held(&sock_flow_mutex));
  55         size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
  56
  57         ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
  58
  59         if (write) {
  60                 if (size) {
  61                         if (size > 1<<29) {
  62                                 /* Enforce limit to prevent overflow */
  63                                 mutex_unlock(&sock_flow_mutex);
  64                                 return -EINVAL;
  65                         }
  66                         size = roundup_pow_of_two(size);
  67                         if (size != orig_size) {
  68                                 sock_table =
  69                                     vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
  70                                 if (!sock_table) {
  71                                         mutex_unlock(&sock_flow_mutex);
  72                                         return -ENOMEM;
  73                                 }
  74                                 rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
  75                                 sock_table->mask = size - 1;
  76                         } else
  77                                 sock_table = orig_sock_table;
  78
  79                         for (i = 0; i < size; i++)
  80                                 sock_table->ents[i] = RPS_NO_CPU;
  81                 } else
  82                         sock_table = NULL;
  83
  84                 if (sock_table != orig_sock_table) {
  85                         rcu_assign_pointer(rps_sock_flow_table, sock_table);
  86                         if (sock_table) {
  87                                 static_key_slow_inc(&rps_needed);
  88                                 static_key_slow_inc(&rfs_needed);
  89                         }
  90                         if (orig_sock_table) {
  91                                 static_key_slow_dec(&rps_needed);
  92                                 static_key_slow_dec(&rfs_needed);
  93                                 synchronize_rcu();
  94                                 vfree(orig_sock_table);
  95                         }
  96                 }
  97         }
  98
  99         mutex_unlock(&sock_flow_mutex);
 100
 101         return ret;
 102 }
 103 #endif /* CONFIG_RPS */
 104
 105 #ifdef CONFIG_NET_FLOW_LIMIT
 106 static DEFINE_MUTEX(flow_limit_update_mutex);
 107
 108 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
 109                                  void __user *buffer, size_t *lenp,
 110                                  loff_t *ppos)
 111 {
 112         struct sd_flow_limit *cur;
 113         struct softnet_data *sd;
 114         cpumask_var_t mask;
 115         int i, len, ret = 0;
 116
 117         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 118                 return -ENOMEM;
 119
 120         if (write) {
 121                 ret = cpumask_parse_user(buffer, *lenp, mask);
 122                 if (ret)
 123                         goto done;
 124
 125                 mutex_lock(&flow_limit_update_mutex);
 126                 len = sizeof(*cur) + netdev_flow_limit_table_len;
 127                 for_each_possible_cpu(i) {
 128                         sd = &per_cpu(softnet_data, i);
 129                         cur = rcu_dereference_protected(sd->flow_limit,
 130                                      lockdep_is_held(&flow_limit_update_mutex));
 131                         if (cur && !cpumask_test_cpu(i, mask)) {
 132                                 RCU_INIT_POINTER(sd->flow_limit, NULL);
 133                                 synchronize_rcu();
 134                                 kfree(cur);
 135                         } else if (!cur && cpumask_test_cpu(i, mask)) {
 136                                 cur = kzalloc_node(len, GFP_KERNEL,
 137                                                    cpu_to_node(i));
 138                                 if (!cur) {
 139                                         /* not unwinding previous changes */
 140                                         ret = -ENOMEM;
 141                                         goto write_unlock;
 142                                 }
 143                                 cur->num_buckets = netdev_flow_limit_table_len;
 144                                 rcu_assign_pointer(sd->flow_limit, cur);
 145                         }
 146                 }
 147 write_unlock:
 148                 mutex_unlock(&flow_limit_update_mutex);
 149         } else {
 150                 char kbuf[128];
 151
 152                 if (*ppos || !*lenp) {
 153                         *lenp = 0;
 154                         goto done;
 155                 }
 156
 157                 cpumask_clear(mask);
 158                 rcu_read_lock();
 159                 for_each_possible_cpu(i) {
 160                         sd = &per_cpu(softnet_data, i);
 161                         if (rcu_dereference(sd->flow_limit))
 162                                 cpumask_set_cpu(i, mask);
 163                 }
 164                 rcu_read_unlock();
 165
 166                 len = min(sizeof(kbuf) - 1, *lenp);
 167                 len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
 168                 if (!len) {
 169                         *lenp = 0;
 170                         goto done;
 171                 }
 172                 if (len < *lenp)
 173                         kbuf[len++] = '\n';
 174                 if (copy_to_user(buffer, kbuf, len)) {
 175                         ret = -EFAULT;
 176                         goto done;
 177                 }
 178                 *lenp = len;
 179                 *ppos += len;
 180         }
 181
 182 done:
 183         free_cpumask_var(mask);
 184         return ret;
 185 }
 186
 187 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
 188                                        void __user *buffer, size_t *lenp,
 189                                        loff_t *ppos)
 190 {
 191         unsigned int old, *ptr;
 192         int ret;
 193
 194         mutex_lock(&flow_limit_update_mutex);
 195
 196         ptr = table->data;
 197         old = *ptr;
 198         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 199         if (!ret && write && !is_power_of_2(*ptr)) {
 200                 *ptr = old;
 201                 ret = -EINVAL;
 202         }
 203
 204         mutex_unlock(&flow_limit_update_mutex);
 205         return ret;
 206 }
 207 #endif /* CONFIG_NET_FLOW_LIMIT */
 208
 209 #ifdef CONFIG_NET_SCHED
 210 static int set_default_qdisc(struct ctl_table *table, int write,
 211                              void __user *buffer, size_t *lenp, loff_t *ppos)
 212 {
 213         char id[IFNAMSIZ];
 214         struct ctl_table tbl = {
 215                 .data = id,
 216                 .maxlen = IFNAMSIZ,
 217         };
 218         int ret;
 219
 220         qdisc_get_default(id, IFNAMSIZ);
 221
 222         ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
 223         if (write && ret == 0)
 224                 ret = qdisc_set_default(id);
 225         return ret;
 226 }
 227 #endif
 228
 229 static int proc_do_dev_weight(struct ctl_table *table, int write,
 230                            void __user *buffer, size_t *lenp, loff_t *ppos)
 231 {
 232         int ret;
 233
 234         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 235         if (ret != 0)
 236                 return ret;
 237
 238         dev_rx_weight = weight_p * dev_weight_rx_bias;
 239         dev_tx_weight = weight_p * dev_weight_tx_bias;
 240
 241         return ret;
 242 }
 243
 244 static int proc_do_rss_key(struct ctl_table *table, int write,
 245                            void __user *buffer, size_t *lenp, loff_t *ppos)
 246 {
 247         struct ctl_table fake_table;
 248         char buf[NETDEV_RSS_KEY_LEN * 3];
 249
 250         snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
 251         fake_table.data = buf;
 252         fake_table.maxlen = sizeof(buf);
 253         return proc_dostring(&fake_table, write, buffer, lenp, ppos);
 254 }
 255
 256 #ifdef CONFIG_BPF_JIT
 257 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write,
 258                                            void __user *buffer, size_t *lenp,
 259                                            loff_t *ppos)
 260 {
 261         int ret, jit_enable = *(int *)table->data;
 262         struct ctl_table tmp = *table;
 263
 264         if (write && !capable(CAP_SYS_ADMIN))
 265                 return -EPERM;
 266
 267         tmp.data = &jit_enable;
 268         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
 269         if (write && !ret) {
 270                 if (jit_enable < 2 ||
 271                     (jit_enable == 2 && bpf_dump_raw_ok())) {
 272                         *(int *)table->data = jit_enable;
 273                         if (jit_enable == 2)
 274                                 pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
 275                 } else {
 276                         ret = -EPERM;
 277                 }
 278         }
 279         return ret;
 280 }
 281
 282 # ifdef CONFIG_HAVE_EBPF_JIT
 283 static int
 284 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write,
 285                                     void __user *buffer, size_t *lenp,
 286                                     loff_t *ppos)
 287 {
 288         if (!capable(CAP_SYS_ADMIN))
 289                 return -EPERM;
 290
 291         return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
 292 }
 293 # endif
 294 #endif
 295
 296 static struct ctl_table net_core_table[] = {
 297 #ifdef CONFIG_NET
 298         {
 299                 .procname       = "wmem_max",
 300                 .data           = &sysctl_wmem_max,
 301                 .maxlen         = sizeof(int),
 302                 .mode           = 0644,
 303                 .proc_handler   = proc_dointvec_minmax,
 304                 .extra1         = &min_sndbuf,
 305         },
 306         {
 307                 .procname       = "rmem_max",
 308                 .data           = &sysctl_rmem_max,
 309                 .maxlen         = sizeof(int),
 310                 .mode           = 0644,
 311                 .proc_handler   = proc_dointvec_minmax,
 312                 .extra1         = &min_rcvbuf,
 313         },
 314         {
 315                 .procname       = "wmem_default",
 316                 .data           = &sysctl_wmem_default,
 317                 .maxlen         = sizeof(int),
 318                 .mode           = 0644,
 319                 .proc_handler   = proc_dointvec_minmax,
 320                 .extra1         = &min_sndbuf,
 321         },
 322         {
 323                 .procname       = "rmem_default",
 324                 .data           = &sysctl_rmem_default,
 325                 .maxlen         = sizeof(int),
 326                 .mode           = 0644,
 327                 .proc_handler   = proc_dointvec_minmax,
 328                 .extra1         = &min_rcvbuf,
 329         },
 330         {
 331                 .procname       = "dev_weight",
 332                 .data           = &weight_p,
 333                 .maxlen         = sizeof(int),
 334                 .mode           = 0644,
 335                 .proc_handler   = proc_do_dev_weight,
 336         },
 337         {
 338                 .procname       = "dev_weight_rx_bias",
 339                 .data           = &dev_weight_rx_bias,
 340                 .maxlen         = sizeof(int),
 341                 .mode           = 0644,
 342                 .proc_handler   = proc_do_dev_weight,
 343         },
 344         {
 345                 .procname       = "dev_weight_tx_bias",
 346                 .data           = &dev_weight_tx_bias,
 347                 .maxlen         = sizeof(int),
 348                 .mode           = 0644,
 349                 .proc_handler   = proc_do_dev_weight,
 350         },
 351         {
 352                 .procname       = "netdev_max_backlog",
 353                 .data           = &netdev_max_backlog,
 354                 .maxlen         = sizeof(int),
 355                 .mode           = 0644,
 356                 .proc_handler   = proc_dointvec
 357         },
 358         {
 359                 .procname       = "netdev_rss_key",
 360                 .data           = &netdev_rss_key,
 361                 .maxlen         = sizeof(int),
 362                 .mode           = 0444,
 363                 .proc_handler   = proc_do_rss_key,
 364         },
 365 #ifdef CONFIG_BPF_JIT
 366         {
 367                 .procname       = "bpf_jit_enable",
 368                 .data           = &bpf_jit_enable,
 369                 .maxlen         = sizeof(int),
 370                 .mode           = 0644,
 371                 .proc_handler   = proc_dointvec_minmax_bpf_enable,
 372 # ifdef CONFIG_BPF_JIT_ALWAYS_ON
 373                 .extra1         = &one,
 374                 .extra2         = &one,
 375 # else
 376                 .extra1         = &zero,
 377                 .extra2         = &two,
 378 # endif
 379         },
 380 # ifdef CONFIG_HAVE_EBPF_JIT
 381         {
 382                 .procname       = "bpf_jit_harden",
 383                 .data           = &bpf_jit_harden,
 384                 .maxlen         = sizeof(int),
 385                 .mode           = 0600,
 386                 .proc_handler   = proc_dointvec_minmax_bpf_restricted,
 387                 .extra1         = &zero,
 388                 .extra2         = &two,
 389         },
 390         {
 391                 .procname       = "bpf_jit_kallsyms",
 392                 .data           = &bpf_jit_kallsyms,
 393                 .maxlen         = sizeof(int),
 394                 .mode           = 0600,
 395                 .proc_handler   = proc_dointvec_minmax_bpf_restricted,
 396                 .extra1         = &zero,
 397                 .extra2         = &one,
 398         },
 399 # endif
 400 #endif
 401         {
 402                 .procname       = "netdev_tstamp_prequeue",
 403                 .data           = &netdev_tstamp_prequeue,
 404                 .maxlen         = sizeof(int),
 405                 .mode           = 0644,
 406                 .proc_handler   = proc_dointvec
 407         },
 408         {
 409                 .procname       = "message_cost",
 410                 .data           = &net_ratelimit_state.interval,
 411                 .maxlen         = sizeof(int),
 412                 .mode           = 0644,
 413                 .proc_handler   = proc_dointvec_jiffies,
 414         },
 415         {
 416                 .procname       = "message_burst",
 417                 .data           = &net_ratelimit_state.burst,
 418                 .maxlen         = sizeof(int),
 419                 .mode           = 0644,
 420                 .proc_handler   = proc_dointvec,
 421         },
 422         {
 423                 .procname       = "optmem_max",
 424                 .data           = &sysctl_optmem_max,
 425                 .maxlen         = sizeof(int),
 426                 .mode           = 0644,
 427                 .proc_handler   = proc_dointvec
 428         },
 429         {
 430                 .procname       = "tstamp_allow_data",
 431                 .data           = &sysctl_tstamp_allow_data,
 432                 .maxlen         = sizeof(int),
 433                 .mode           = 0644,
 434                 .proc_handler   = proc_dointvec_minmax,
 435                 .extra1         = &zero,
 436                 .extra2         = &one
 437         },
 438 #ifdef CONFIG_RPS
 439         {
 440                 .procname       = "rps_sock_flow_entries",
 441                 .maxlen         = sizeof(int),
 442                 .mode           = 0644,
 443                 .proc_handler   = rps_sock_flow_sysctl
 444         },
 445 #endif
 446 #ifdef CONFIG_NET_FLOW_LIMIT
 447         {
 448                 .procname       = "flow_limit_cpu_bitmap",
 449                 .mode           = 0644,
 450                 .proc_handler   = flow_limit_cpu_sysctl
 451         },
 452         {
 453                 .procname       = "flow_limit_table_len",
 454                 .data           = &netdev_flow_limit_table_len,
 455                 .maxlen         = sizeof(int),
 456                 .mode           = 0644,
 457                 .proc_handler   = flow_limit_table_len_sysctl
 458         },
 459 #endif /* CONFIG_NET_FLOW_LIMIT */
 460 #ifdef CONFIG_NET_RX_BUSY_POLL
 461         {
 462                 .procname       = "busy_poll",
 463                 .data           = &sysctl_net_busy_poll,
 464                 .maxlen         = sizeof(unsigned int),
 465                 .mode           = 0644,
 466                 .proc_handler   = proc_dointvec_minmax,
 467                 .extra1         = &zero,
 468         },
 469         {
 470                 .procname       = "busy_read",
 471                 .data           = &sysctl_net_busy_read,
 472                 .maxlen         = sizeof(unsigned int),
 473                 .mode           = 0644,
 474                 .proc_handler   = proc_dointvec_minmax,
 475                 .extra1         = &zero,
 476         },
 477 #endif
 478 #ifdef CONFIG_NET_SCHED
 479         {
 480                 .procname       = "default_qdisc",
 481                 .mode           = 0644,
 482                 .maxlen         = IFNAMSIZ,
 483                 .proc_handler   = set_default_qdisc
 484         },
 485 #endif
 486 #endif /* CONFIG_NET */
 487         {
 488                 .procname       = "netdev_budget",
 489                 .data           = &netdev_budget,
 490                 .maxlen         = sizeof(int),
 491                 .mode           = 0644,
 492                 .proc_handler   = proc_dointvec
 493         },
 494         {
 495                 .procname       = "warnings",
 496                 .data           = &net_msg_warn,
 497                 .maxlen         = sizeof(int),
 498                 .mode           = 0644,
 499                 .proc_handler   = proc_dointvec
 500         },
 501         {
 502                 .procname       = "max_skb_frags",
 503                 .data           = &sysctl_max_skb_frags,
 504                 .maxlen         = sizeof(int),
 505                 .mode           = 0644,
 506                 .proc_handler   = proc_dointvec_minmax,
 507                 .extra1         = &one,
 508                 .extra2         = &max_skb_frags,
 509         },
 510         {
 511                 .procname       = "netdev_budget_usecs",
 512                 .data           = &netdev_budget_usecs,
 513                 .maxlen         = sizeof(unsigned int),
 514                 .mode           = 0644,
 515                 .proc_handler   = proc_dointvec_minmax,
 516                 .extra1         = &zero,
 517         },
 518         {
 519                 .procname       = "fb_tunnels_only_for_init_net",
 520                 .data           = &sysctl_fb_tunnels_only_for_init_net,
 521                 .maxlen         = sizeof(int),
 522                 .mode           = 0644,
 523                 .proc_handler   = proc_dointvec_minmax,
 524                 .extra1         = &zero,
 525                 .extra2         = &one,
 526         },
 527         { }
 528 };
 529
 530 static struct ctl_table netns_core_table[] = {
 531         {
 532                 .procname       = "somaxconn",
 533                 .data           = &init_net.core.sysctl_somaxconn,
 534                 .maxlen         = sizeof(int),
 535                 .mode           = 0644,
 536                 .extra1         = &zero,
 537                 .proc_handler   = proc_dointvec_minmax
 538         },
 539         { }
 540 };
 541
 542 static __net_init int sysctl_core_net_init(struct net *net)
 543 {
 544         struct ctl_table *tbl;
 545
 546         tbl = netns_core_table;
 547         if (!net_eq(net, &init_net)) {
 548                 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
 549                 if (tbl == NULL)
 550                         goto err_dup;
 551
 552                 tbl[0].data = &net->core.sysctl_somaxconn;
 553
 554                 /* Don't export any sysctls to unprivileged users */
 555                 if (net->user_ns != &init_user_ns) {
 556                         tbl[0].procname = NULL;
 557                 }
 558         }
 559
 560         net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
 561         if (net->core.sysctl_hdr == NULL)
 562                 goto err_reg;
 563
 564         return 0;
 565
 566 err_reg:
 567         if (tbl != netns_core_table)
 568                 kfree(tbl);
 569 err_dup:
 570         return -ENOMEM;
 571 }
 572
 573 static __net_exit void sysctl_core_net_exit(struct net *net)
 574 {
 575         struct ctl_table *tbl;
 576
 577         tbl = net->core.sysctl_hdr->ctl_table_arg;
 578         unregister_net_sysctl_table(net->core.sysctl_hdr);
 579         BUG_ON(tbl == netns_core_table);
 580         kfree(tbl);
 581 }
 582
 583 static __net_initdata struct pernet_operations sysctl_core_ops = {
 584         .init = sysctl_core_net_init,
 585         .exit = sysctl_core_net_exit,
 586 };
 587
 588 static __init int sysctl_core_init(void)
 589 {
 590         register_net_sysctl(&init_net, "net/core", net_core_table);
 591         return register_pernet_subsys(&sysctl_core_ops);
 592 }
 593
 594 fs_initcall(sysctl_core_init);