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 static long long_one __maybe_unused = 1;
  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 #ifdef CONFIG_RPS
  40 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
  41                                 void __user *buffer, size_t *lenp, loff_t *ppos)
  42 {
  43         unsigned int orig_size, size;
  44         int ret, i;
  45         struct ctl_table tmp = {
  46                 .data = &size,
  47                 .maxlen = sizeof(size),
  48                 .mode = table->mode
  49         };
  50         struct rps_sock_flow_table *orig_sock_table, *sock_table;
  51         static DEFINE_MUTEX(sock_flow_mutex);
  52
  53         mutex_lock(&sock_flow_mutex);
  54
  55         orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
  56                                         lockdep_is_held(&sock_flow_mutex));
  57         size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
  58
  59         ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
  60
  61         if (write) {
  62                 if (size) {
  63                         if (size > 1<<29) {
  64                                 /* Enforce limit to prevent overflow */
  65                                 mutex_unlock(&sock_flow_mutex);
  66                                 return -EINVAL;
  67                         }
  68                         size = roundup_pow_of_two(size);
  69                         if (size != orig_size) {
  70                                 sock_table =
  71                                     vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
  72                                 if (!sock_table) {
  73                                         mutex_unlock(&sock_flow_mutex);
  74                                         return -ENOMEM;
  75                                 }
  76                                 rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
  77                                 sock_table->mask = size - 1;
  78                         } else
  79                                 sock_table = orig_sock_table;
  80
  81                         for (i = 0; i < size; i++)
  82                                 sock_table->ents[i] = RPS_NO_CPU;
  83                 } else
  84                         sock_table = NULL;
  85
  86                 if (sock_table != orig_sock_table) {
  87                         rcu_assign_pointer(rps_sock_flow_table, sock_table);
  88                         if (sock_table) {
  89                                 static_key_slow_inc(&rps_needed);
  90                                 static_key_slow_inc(&rfs_needed);
  91                         }
  92                         if (orig_sock_table) {
  93                                 static_key_slow_dec(&rps_needed);
  94                                 static_key_slow_dec(&rfs_needed);
  95                                 synchronize_rcu();
  96                                 vfree(orig_sock_table);
  97                         }
  98                 }
  99         }
 100
 101         mutex_unlock(&sock_flow_mutex);
 102
 103         return ret;
 104 }
 105 #endif /* CONFIG_RPS */
 106
 107 #ifdef CONFIG_NET_FLOW_LIMIT
 108 static DEFINE_MUTEX(flow_limit_update_mutex);
 109
 110 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
 111                                  void __user *buffer, size_t *lenp,
 112                                  loff_t *ppos)
 113 {
 114         struct sd_flow_limit *cur;
 115         struct softnet_data *sd;
 116         cpumask_var_t mask;
 117         int i, len, ret = 0;
 118
 119         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 120                 return -ENOMEM;
 121
 122         if (write) {
 123                 ret = cpumask_parse_user(buffer, *lenp, mask);
 124                 if (ret)
 125                         goto done;
 126
 127                 mutex_lock(&flow_limit_update_mutex);
 128                 len = sizeof(*cur) + netdev_flow_limit_table_len;
 129                 for_each_possible_cpu(i) {
 130                         sd = &per_cpu(softnet_data, i);
 131                         cur = rcu_dereference_protected(sd->flow_limit,
 132                                      lockdep_is_held(&flow_limit_update_mutex));
 133                         if (cur && !cpumask_test_cpu(i, mask)) {
 134                                 RCU_INIT_POINTER(sd->flow_limit, NULL);
 135                                 synchronize_rcu();
 136                                 kfree(cur);
 137                         } else if (!cur && cpumask_test_cpu(i, mask)) {
 138                                 cur = kzalloc_node(len, GFP_KERNEL,
 139                                                    cpu_to_node(i));
 140                                 if (!cur) {
 141                                         /* not unwinding previous changes */
 142                                         ret = -ENOMEM;
 143                                         goto write_unlock;
 144                                 }
 145                                 cur->num_buckets = netdev_flow_limit_table_len;
 146                                 rcu_assign_pointer(sd->flow_limit, cur);
 147                         }
 148                 }
 149 write_unlock:
 150                 mutex_unlock(&flow_limit_update_mutex);
 151         } else {
 152                 char kbuf[128];
 153
 154                 if (*ppos || !*lenp) {
 155                         *lenp = 0;
 156                         goto done;
 157                 }
 158
 159                 cpumask_clear(mask);
 160                 rcu_read_lock();
 161                 for_each_possible_cpu(i) {
 162                         sd = &per_cpu(softnet_data, i);
 163                         if (rcu_dereference(sd->flow_limit))
 164                                 cpumask_set_cpu(i, mask);
 165                 }
 166                 rcu_read_unlock();
 167
 168                 len = min(sizeof(kbuf) - 1, *lenp);
 169                 len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
 170                 if (!len) {
 171                         *lenp = 0;
 172                         goto done;
 173                 }
 174                 if (len < *lenp)
 175                         kbuf[len++] = '\n';
 176                 if (copy_to_user(buffer, kbuf, len)) {
 177                         ret = -EFAULT;
 178                         goto done;
 179                 }
 180                 *lenp = len;
 181                 *ppos += len;
 182         }
 183
 184 done:
 185         free_cpumask_var(mask);
 186         return ret;
 187 }
 188
 189 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
 190                                        void __user *buffer, size_t *lenp,
 191                                        loff_t *ppos)
 192 {
 193         unsigned int old, *ptr;
 194         int ret;
 195
 196         mutex_lock(&flow_limit_update_mutex);
 197
 198         ptr = table->data;
 199         old = *ptr;
 200         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 201         if (!ret && write && !is_power_of_2(*ptr)) {
 202                 *ptr = old;
 203                 ret = -EINVAL;
 204         }
 205
 206         mutex_unlock(&flow_limit_update_mutex);
 207         return ret;
 208 }
 209 #endif /* CONFIG_NET_FLOW_LIMIT */
 210
 211 #ifdef CONFIG_NET_SCHED
 212 static int set_default_qdisc(struct ctl_table *table, int write,
 213                              void __user *buffer, size_t *lenp, loff_t *ppos)
 214 {
 215         char id[IFNAMSIZ];
 216         struct ctl_table tbl = {
 217                 .data = id,
 218                 .maxlen = IFNAMSIZ,
 219         };
 220         int ret;
 221
 222         qdisc_get_default(id, IFNAMSIZ);
 223
 224         ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
 225         if (write && ret == 0)
 226                 ret = qdisc_set_default(id);
 227         return ret;
 228 }
 229 #endif
 230
 231 static int proc_do_dev_weight(struct ctl_table *table, int write,
 232                            void __user *buffer, size_t *lenp, loff_t *ppos)
 233 {
 234         static DEFINE_MUTEX(dev_weight_mutex);
 235         int ret, weight;
 236
 237         mutex_lock(&dev_weight_mutex);
 238         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 239         if (!ret && write) {
 240                 weight = READ_ONCE(weight_p);
 241                 WRITE_ONCE(dev_rx_weight, weight * dev_weight_rx_bias);
 242                 WRITE_ONCE(dev_tx_weight, weight * dev_weight_tx_bias);
 243         }
 244         mutex_unlock(&dev_weight_mutex);
 245
 246         return ret;
 247 }
 248
 249 static int proc_do_rss_key(struct ctl_table *table, int write,
 250                            void __user *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 __user *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 __user *buffer, size_t *lenp,
 291                                     loff_t *ppos)
 292 {
 293         if (!capable(CAP_SYS_ADMIN))
 294                 return -EPERM;
 295
 296         return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
 297 }
 298 # endif /* CONFIG_HAVE_EBPF_JIT */
 299
 300 static int
 301 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
 302                                      void __user *buffer, size_t *lenp,
 303                                      loff_t *ppos)
 304 {
 305         if (!capable(CAP_SYS_ADMIN))
 306                 return -EPERM;
 307
 308         return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
 309 }
 310 #endif
 311
 312 static struct ctl_table net_core_table[] = {
 313 #ifdef CONFIG_NET
 314         {
 315                 .procname       = "wmem_max",
 316                 .data           = &sysctl_wmem_max,
 317                 .maxlen         = sizeof(int),
 318                 .mode           = 0644,
 319                 .proc_handler   = proc_dointvec_minmax,
 320                 .extra1         = &min_sndbuf,
 321         },
 322         {
 323                 .procname       = "rmem_max",
 324                 .data           = &sysctl_rmem_max,
 325                 .maxlen         = sizeof(int),
 326                 .mode           = 0644,
 327                 .proc_handler   = proc_dointvec_minmax,
 328                 .extra1         = &min_rcvbuf,
 329         },
 330         {
 331                 .procname       = "wmem_default",
 332                 .data           = &sysctl_wmem_default,
 333                 .maxlen         = sizeof(int),
 334                 .mode           = 0644,
 335                 .proc_handler   = proc_dointvec_minmax,
 336                 .extra1         = &min_sndbuf,
 337         },
 338         {
 339                 .procname       = "rmem_default",
 340                 .data           = &sysctl_rmem_default,
 341                 .maxlen         = sizeof(int),
 342                 .mode           = 0644,
 343                 .proc_handler   = proc_dointvec_minmax,
 344                 .extra1         = &min_rcvbuf,
 345         },
 346         {
 347                 .procname       = "dev_weight",
 348                 .data           = &weight_p,
 349                 .maxlen         = sizeof(int),
 350                 .mode           = 0644,
 351                 .proc_handler   = proc_do_dev_weight,
 352         },
 353         {
 354                 .procname       = "dev_weight_rx_bias",
 355                 .data           = &dev_weight_rx_bias,
 356                 .maxlen         = sizeof(int),
 357                 .mode           = 0644,
 358                 .proc_handler   = proc_do_dev_weight,
 359         },
 360         {
 361                 .procname       = "dev_weight_tx_bias",
 362                 .data           = &dev_weight_tx_bias,
 363                 .maxlen         = sizeof(int),
 364                 .mode           = 0644,
 365                 .proc_handler   = proc_do_dev_weight,
 366         },
 367         {
 368                 .procname       = "netdev_max_backlog",
 369                 .data           = &netdev_max_backlog,
 370                 .maxlen         = sizeof(int),
 371                 .mode           = 0644,
 372                 .proc_handler   = proc_dointvec
 373         },
 374         {
 375                 .procname       = "netdev_rss_key",
 376                 .data           = &netdev_rss_key,
 377                 .maxlen         = sizeof(int),
 378                 .mode           = 0444,
 379                 .proc_handler   = proc_do_rss_key,
 380         },
 381 #ifdef CONFIG_BPF_JIT
 382         {
 383                 .procname       = "bpf_jit_enable",
 384                 .data           = &bpf_jit_enable,
 385                 .maxlen         = sizeof(int),
 386                 .mode           = 0644,
 387                 .proc_handler   = proc_dointvec_minmax_bpf_enable,
 388 # ifdef CONFIG_BPF_JIT_ALWAYS_ON
 389                 .extra1         = &one,
 390                 .extra2         = &one,
 391 # else
 392                 .extra1         = &zero,
 393                 .extra2         = &two,
 394 # endif
 395         },
 396 # ifdef CONFIG_HAVE_EBPF_JIT
 397         {
 398                 .procname       = "bpf_jit_harden",
 399                 .data           = &bpf_jit_harden,
 400                 .maxlen         = sizeof(int),
 401                 .mode           = 0600,
 402                 .proc_handler   = proc_dointvec_minmax_bpf_restricted,
 403                 .extra1         = &zero,
 404                 .extra2         = &two,
 405         },
 406         {
 407                 .procname       = "bpf_jit_kallsyms",
 408                 .data           = &bpf_jit_kallsyms,
 409                 .maxlen         = sizeof(int),
 410                 .mode           = 0600,
 411                 .proc_handler   = proc_dointvec_minmax_bpf_restricted,
 412                 .extra1         = &zero,
 413                 .extra2         = &one,
 414         },
 415 # endif
 416         {
 417                 .procname       = "bpf_jit_limit",
 418                 .data           = &bpf_jit_limit,
 419                 .maxlen         = sizeof(long),
 420                 .mode           = 0600,
 421                 .proc_handler   = proc_dolongvec_minmax_bpf_restricted,
 422                 .extra1         = &long_one,
 423                 .extra2         = &bpf_jit_limit_max,
 424         },
 425 #endif
 426         {
 427                 .procname       = "netdev_tstamp_prequeue",
 428                 .data           = &netdev_tstamp_prequeue,
 429                 .maxlen         = sizeof(int),
 430                 .mode           = 0644,
 431                 .proc_handler   = proc_dointvec
 432         },
 433         {
 434                 .procname       = "message_cost",
 435                 .data           = &net_ratelimit_state.interval,
 436                 .maxlen         = sizeof(int),
 437                 .mode           = 0644,
 438                 .proc_handler   = proc_dointvec_jiffies,
 439         },
 440         {
 441                 .procname       = "message_burst",
 442                 .data           = &net_ratelimit_state.burst,
 443                 .maxlen         = sizeof(int),
 444                 .mode           = 0644,
 445                 .proc_handler   = proc_dointvec,
 446         },
 447         {
 448                 .procname       = "optmem_max",
 449                 .data           = &sysctl_optmem_max,
 450                 .maxlen         = sizeof(int),
 451                 .mode           = 0644,
 452                 .proc_handler   = proc_dointvec
 453         },
 454         {
 455                 .procname       = "tstamp_allow_data",
 456                 .data           = &sysctl_tstamp_allow_data,
 457                 .maxlen         = sizeof(int),
 458                 .mode           = 0644,
 459                 .proc_handler   = proc_dointvec_minmax,
 460                 .extra1         = &zero,
 461                 .extra2         = &one
 462         },
 463 #ifdef CONFIG_RPS
 464         {
 465                 .procname       = "rps_sock_flow_entries",
 466                 .maxlen         = sizeof(int),
 467                 .mode           = 0644,
 468                 .proc_handler   = rps_sock_flow_sysctl
 469         },
 470 #endif
 471 #ifdef CONFIG_NET_FLOW_LIMIT
 472         {
 473                 .procname       = "flow_limit_cpu_bitmap",
 474                 .mode           = 0644,
 475                 .proc_handler   = flow_limit_cpu_sysctl
 476         },
 477         {
 478                 .procname       = "flow_limit_table_len",
 479                 .data           = &netdev_flow_limit_table_len,
 480                 .maxlen         = sizeof(int),
 481                 .mode           = 0644,
 482                 .proc_handler   = flow_limit_table_len_sysctl
 483         },
 484 #endif /* CONFIG_NET_FLOW_LIMIT */
 485 #ifdef CONFIG_NET_RX_BUSY_POLL
 486         {
 487                 .procname       = "busy_poll",
 488                 .data           = &sysctl_net_busy_poll,
 489                 .maxlen         = sizeof(unsigned int),
 490                 .mode           = 0644,
 491                 .proc_handler   = proc_dointvec_minmax,
 492                 .extra1         = &zero,
 493         },
 494         {
 495                 .procname       = "busy_read",
 496                 .data           = &sysctl_net_busy_read,
 497                 .maxlen         = sizeof(unsigned int),
 498                 .mode           = 0644,
 499                 .proc_handler   = proc_dointvec_minmax,
 500                 .extra1         = &zero,
 501         },
 502 #endif
 503 #ifdef CONFIG_NET_SCHED
 504         {
 505                 .procname       = "default_qdisc",
 506                 .mode           = 0644,
 507                 .maxlen         = IFNAMSIZ,
 508                 .proc_handler   = set_default_qdisc
 509         },
 510 #endif
 511 #endif /* CONFIG_NET */
 512         {
 513                 .procname       = "netdev_budget",
 514                 .data           = &netdev_budget,
 515                 .maxlen         = sizeof(int),
 516                 .mode           = 0644,
 517                 .proc_handler   = proc_dointvec
 518         },
 519         {
 520                 .procname       = "warnings",
 521                 .data           = &net_msg_warn,
 522                 .maxlen         = sizeof(int),
 523                 .mode           = 0644,
 524                 .proc_handler   = proc_dointvec
 525         },
 526         {
 527                 .procname       = "max_skb_frags",
 528                 .data           = &sysctl_max_skb_frags,
 529                 .maxlen         = sizeof(int),
 530                 .mode           = 0644,
 531                 .proc_handler   = proc_dointvec_minmax,
 532                 .extra1         = &one,
 533                 .extra2         = &max_skb_frags,
 534         },
 535         {
 536                 .procname       = "netdev_budget_usecs",
 537                 .data           = &netdev_budget_usecs,
 538                 .maxlen         = sizeof(unsigned int),
 539                 .mode           = 0644,
 540                 .proc_handler   = proc_dointvec_minmax,
 541                 .extra1         = &zero,
 542         },
 543         {
 544                 .procname       = "fb_tunnels_only_for_init_net",
 545                 .data           = &sysctl_fb_tunnels_only_for_init_net,
 546                 .maxlen         = sizeof(int),
 547                 .mode           = 0644,
 548                 .proc_handler   = proc_dointvec_minmax,
 549                 .extra1         = &zero,
 550                 .extra2         = &one,
 551         },
 552         { }
 553 };
 554
 555 static struct ctl_table netns_core_table[] = {
 556         {
 557                 .procname       = "somaxconn",
 558                 .data           = &init_net.core.sysctl_somaxconn,
 559                 .maxlen         = sizeof(int),
 560                 .mode           = 0644,
 561                 .extra1         = &zero,
 562                 .proc_handler   = proc_dointvec_minmax
 563         },
 564         { }
 565 };
 566
 567 static __net_init int sysctl_core_net_init(struct net *net)
 568 {
 569         struct ctl_table *tbl;
 570
 571         tbl = netns_core_table;
 572         if (!net_eq(net, &init_net)) {
 573                 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
 574                 if (tbl == NULL)
 575                         goto err_dup;
 576
 577                 tbl[0].data = &net->core.sysctl_somaxconn;
 578
 579                 /* Don't export any sysctls to unprivileged users */
 580                 if (net->user_ns != &init_user_ns) {
 581                         tbl[0].procname = NULL;
 582                 }
 583         }
 584
 585         net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
 586         if (net->core.sysctl_hdr == NULL)
 587                 goto err_reg;
 588
 589         return 0;
 590
 591 err_reg:
 592         if (tbl != netns_core_table)
 593                 kfree(tbl);
 594 err_dup:
 595         return -ENOMEM;
 596 }
 597
 598 static __net_exit void sysctl_core_net_exit(struct net *net)
 599 {
 600         struct ctl_table *tbl;
 601
 602         tbl = net->core.sysctl_hdr->ctl_table_arg;
 603         unregister_net_sysctl_table(net->core.sysctl_hdr);
 604         BUG_ON(tbl == netns_core_table);
 605         kfree(tbl);
 606 }
 607
 608 static __net_initdata struct pernet_operations sysctl_core_ops = {
 609         .init = sysctl_core_net_init,
 610         .exit = sysctl_core_net_exit,
 611 };
 612
 613 static __init int sysctl_core_init(void)
 614 {
 615         register_net_sysctl(&init_net, "net/core", net_core_table);
 616         return register_pernet_subsys(&sysctl_core_ops);
 617 }
 618
 619 fs_initcall(sysctl_core_init);