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 #include <linux/kmemleak.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 static int zero = 0;
  27 static int one = 1;
  28 static int two __maybe_unused = 2;
  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_one __maybe_unused = 1;
  33 static long long_max __maybe_unused = LONG_MAX;
  34
  35 static int net_msg_warn;        /* Unused, but still a sysctl */
  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         static DEFINE_MUTEX(dev_weight_mutex);
 233         int ret, weight;
 234
 235         mutex_lock(&dev_weight_mutex);
 236         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 237         if (!ret && write) {
 238                 weight = READ_ONCE(weight_p);
 239                 WRITE_ONCE(dev_rx_weight, weight * dev_weight_rx_bias);
 240                 WRITE_ONCE(dev_tx_weight, weight * dev_weight_tx_bias);
 241         }
 242         mutex_unlock(&dev_weight_mutex);
 243
 244         return ret;
 245 }
 246
 247 static int proc_do_rss_key(struct ctl_table *table, int write,
 248                            void __user *buffer, size_t *lenp, loff_t *ppos)
 249 {
 250         struct ctl_table fake_table;
 251         char buf[NETDEV_RSS_KEY_LEN * 3];
 252
 253         snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
 254         fake_table.data = buf;
 255         fake_table.maxlen = sizeof(buf);
 256         return proc_dostring(&fake_table, write, buffer, lenp, ppos);
 257 }
 258
 259 #ifdef CONFIG_BPF_JIT
 260 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write,
 261                                            void __user *buffer, size_t *lenp,
 262                                            loff_t *ppos)
 263 {
 264         int ret, jit_enable = *(int *)table->data;
 265         struct ctl_table tmp = *table;
 266
 267         if (write && !capable(CAP_SYS_ADMIN))
 268                 return -EPERM;
 269
 270         tmp.data = &jit_enable;
 271         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
 272         if (write && !ret) {
 273                 *(int *)table->data = jit_enable;
 274                 if (jit_enable == 2)
 275                         pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
 276         }
 277         return ret;
 278 }
 279
 280 # ifdef CONFIG_HAVE_EBPF_JIT
 281 static int
 282 proc_dointvec_minmax_bpf_restricted(struct ctl_table *table, int write,
 283                                     void __user *buffer, size_t *lenp,
 284                                     loff_t *ppos)
 285 {
 286         if (!capable(CAP_SYS_ADMIN))
 287                 return -EPERM;
 288
 289         return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
 290 }
 291 # endif /* CONFIG_HAVE_EBPF_JIT */
 292
 293 static int
 294 proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
 295                                      void __user *buffer, size_t *lenp,
 296                                      loff_t *ppos)
 297 {
 298         if (!capable(CAP_SYS_ADMIN))
 299                 return -EPERM;
 300
 301         return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
 302 }
 303 #endif
 304
 305 static struct ctl_table net_core_table[] = {
 306 #ifdef CONFIG_NET
 307         {
 308                 .procname       = "wmem_max",
 309                 .data           = &sysctl_wmem_max,
 310                 .maxlen         = sizeof(int),
 311                 .mode           = 0644,
 312                 .proc_handler   = proc_dointvec_minmax,
 313                 .extra1         = &min_sndbuf,
 314         },
 315         {
 316                 .procname       = "rmem_max",
 317                 .data           = &sysctl_rmem_max,
 318                 .maxlen         = sizeof(int),
 319                 .mode           = 0644,
 320                 .proc_handler   = proc_dointvec_minmax,
 321                 .extra1         = &min_rcvbuf,
 322         },
 323         {
 324                 .procname       = "wmem_default",
 325                 .data           = &sysctl_wmem_default,
 326                 .maxlen         = sizeof(int),
 327                 .mode           = 0644,
 328                 .proc_handler   = proc_dointvec_minmax,
 329                 .extra1         = &min_sndbuf,
 330         },
 331         {
 332                 .procname       = "rmem_default",
 333                 .data           = &sysctl_rmem_default,
 334                 .maxlen         = sizeof(int),
 335                 .mode           = 0644,
 336                 .proc_handler   = proc_dointvec_minmax,
 337                 .extra1         = &min_rcvbuf,
 338         },
 339         {
 340                 .procname       = "dev_weight",
 341                 .data           = &weight_p,
 342                 .maxlen         = sizeof(int),
 343                 .mode           = 0644,
 344                 .proc_handler   = proc_do_dev_weight,
 345         },
 346         {
 347                 .procname       = "dev_weight_rx_bias",
 348                 .data           = &dev_weight_rx_bias,
 349                 .maxlen         = sizeof(int),
 350                 .mode           = 0644,
 351                 .proc_handler   = proc_do_dev_weight,
 352         },
 353         {
 354                 .procname       = "dev_weight_tx_bias",
 355                 .data           = &dev_weight_tx_bias,
 356                 .maxlen         = sizeof(int),
 357                 .mode           = 0644,
 358                 .proc_handler   = proc_do_dev_weight,
 359         },
 360         {
 361                 .procname       = "netdev_max_backlog",
 362                 .data           = &netdev_max_backlog,
 363                 .maxlen         = sizeof(int),
 364                 .mode           = 0644,
 365                 .proc_handler   = proc_dointvec
 366         },
 367         {
 368                 .procname       = "netdev_rss_key",
 369                 .data           = &netdev_rss_key,
 370                 .maxlen         = sizeof(int),
 371                 .mode           = 0444,
 372                 .proc_handler   = proc_do_rss_key,
 373         },
 374 #ifdef CONFIG_BPF_JIT
 375         {
 376                 .procname       = "bpf_jit_enable",
 377                 .data           = &bpf_jit_enable,
 378                 .maxlen         = sizeof(int),
 379                 .mode           = 0644,
 380                 .proc_handler   = proc_dointvec_minmax_bpf_enable,
 381 # ifdef CONFIG_BPF_JIT_ALWAYS_ON
 382                 .extra1         = &one,
 383                 .extra2         = &one,
 384 # else
 385                 .extra1         = &zero,
 386                 .extra2         = &two,
 387 # endif
 388         },
 389 # ifdef CONFIG_HAVE_EBPF_JIT
 390         {
 391                 .procname       = "bpf_jit_harden",
 392                 .data           = &bpf_jit_harden,
 393                 .maxlen         = sizeof(int),
 394                 .mode           = 0600,
 395                 .proc_handler   = proc_dointvec_minmax_bpf_restricted,
 396                 .extra1         = &zero,
 397                 .extra2         = &two,
 398         },
 399         {
 400                 .procname       = "bpf_jit_kallsyms",
 401                 .data           = &bpf_jit_kallsyms,
 402                 .maxlen         = sizeof(int),
 403                 .mode           = 0600,
 404                 .proc_handler   = proc_dointvec_minmax_bpf_restricted,
 405                 .extra1         = &zero,
 406                 .extra2         = &one,
 407         },
 408 # endif
 409         {
 410                 .procname       = "bpf_jit_limit",
 411                 .data           = &bpf_jit_limit,
 412                 .maxlen         = sizeof(long),
 413                 .mode           = 0600,
 414                 .proc_handler   = proc_dolongvec_minmax_bpf_restricted,
 415                 .extra1         = &long_one,
 416                 .extra2         = &bpf_jit_limit_max,
 417         },
 418 #endif
 419         {
 420                 .procname       = "netdev_tstamp_prequeue",
 421                 .data           = &netdev_tstamp_prequeue,
 422                 .maxlen         = sizeof(int),
 423                 .mode           = 0644,
 424                 .proc_handler   = proc_dointvec
 425         },
 426         {
 427                 .procname       = "message_cost",
 428                 .data           = &net_ratelimit_state.interval,
 429                 .maxlen         = sizeof(int),
 430                 .mode           = 0644,
 431                 .proc_handler   = proc_dointvec_jiffies,
 432         },
 433         {
 434                 .procname       = "message_burst",
 435                 .data           = &net_ratelimit_state.burst,
 436                 .maxlen         = sizeof(int),
 437                 .mode           = 0644,
 438                 .proc_handler   = proc_dointvec,
 439         },
 440         {
 441                 .procname       = "optmem_max",
 442                 .data           = &sysctl_optmem_max,
 443                 .maxlen         = sizeof(int),
 444                 .mode           = 0644,
 445                 .proc_handler   = proc_dointvec
 446         },
 447         {
 448                 .procname       = "tstamp_allow_data",
 449                 .data           = &sysctl_tstamp_allow_data,
 450                 .maxlen         = sizeof(int),
 451                 .mode           = 0644,
 452                 .proc_handler   = proc_dointvec_minmax,
 453                 .extra1         = &zero,
 454                 .extra2         = &one
 455         },
 456 #ifdef CONFIG_RPS
 457         {
 458                 .procname       = "rps_sock_flow_entries",
 459                 .maxlen         = sizeof(int),
 460                 .mode           = 0644,
 461                 .proc_handler   = rps_sock_flow_sysctl
 462         },
 463 #endif
 464 #ifdef CONFIG_NET_FLOW_LIMIT
 465         {
 466                 .procname       = "flow_limit_cpu_bitmap",
 467                 .mode           = 0644,
 468                 .proc_handler   = flow_limit_cpu_sysctl
 469         },
 470         {
 471                 .procname       = "flow_limit_table_len",
 472                 .data           = &netdev_flow_limit_table_len,
 473                 .maxlen         = sizeof(int),
 474                 .mode           = 0644,
 475                 .proc_handler   = flow_limit_table_len_sysctl
 476         },
 477 #endif /* CONFIG_NET_FLOW_LIMIT */
 478 #ifdef CONFIG_NET_RX_BUSY_POLL
 479         {
 480                 .procname       = "busy_poll",
 481                 .data           = &sysctl_net_busy_poll,
 482                 .maxlen         = sizeof(unsigned int),
 483                 .mode           = 0644,
 484                 .proc_handler   = proc_dointvec_minmax,
 485                 .extra1         = &zero,
 486         },
 487         {
 488                 .procname       = "busy_read",
 489                 .data           = &sysctl_net_busy_read,
 490                 .maxlen         = sizeof(unsigned int),
 491                 .mode           = 0644,
 492                 .proc_handler   = proc_dointvec_minmax,
 493                 .extra1         = &zero,
 494         },
 495 #endif
 496 #ifdef CONFIG_NET_SCHED
 497         {
 498                 .procname       = "default_qdisc",
 499                 .mode           = 0644,
 500                 .maxlen         = IFNAMSIZ,
 501                 .proc_handler   = set_default_qdisc
 502         },
 503 #endif
 504 #endif /* CONFIG_NET */
 505         {
 506                 .procname       = "netdev_budget",
 507                 .data           = &netdev_budget,
 508                 .maxlen         = sizeof(int),
 509                 .mode           = 0644,
 510                 .proc_handler   = proc_dointvec
 511         },
 512         {
 513                 .procname       = "warnings",
 514                 .data           = &net_msg_warn,
 515                 .maxlen         = sizeof(int),
 516                 .mode           = 0644,
 517                 .proc_handler   = proc_dointvec
 518         },
 519         {
 520                 .procname       = "max_skb_frags",
 521                 .data           = &sysctl_max_skb_frags,
 522                 .maxlen         = sizeof(int),
 523                 .mode           = 0644,
 524                 .proc_handler   = proc_dointvec_minmax,
 525                 .extra1         = &one,
 526                 .extra2         = &max_skb_frags,
 527         },
 528         {
 529                 .procname       = "netdev_budget_usecs",
 530                 .data           = &netdev_budget_usecs,
 531                 .maxlen         = sizeof(unsigned int),
 532                 .mode           = 0644,
 533                 .proc_handler   = proc_dointvec_minmax,
 534                 .extra1         = &zero,
 535         },
 536         { }
 537 };
 538
 539 static struct ctl_table netns_core_table[] = {
 540         {
 541                 .procname       = "somaxconn",
 542                 .data           = &init_net.core.sysctl_somaxconn,
 543                 .maxlen         = sizeof(int),
 544                 .mode           = 0644,
 545                 .extra1         = &zero,
 546                 .proc_handler   = proc_dointvec_minmax
 547         },
 548         { }
 549 };
 550
 551 static __net_init int sysctl_core_net_init(struct net *net)
 552 {
 553         struct ctl_table *tbl;
 554
 555         tbl = netns_core_table;
 556         if (!net_eq(net, &init_net)) {
 557                 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
 558                 if (tbl == NULL)
 559                         goto err_dup;
 560
 561                 tbl[0].data = &net->core.sysctl_somaxconn;
 562
 563                 /* Don't export any sysctls to unprivileged users */
 564                 if (net->user_ns != &init_user_ns) {
 565                         tbl[0].procname = NULL;
 566                 }
 567         }
 568
 569         net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
 570         if (net->core.sysctl_hdr == NULL)
 571                 goto err_reg;
 572
 573         return 0;
 574
 575 err_reg:
 576         if (tbl != netns_core_table)
 577                 kfree(tbl);
 578 err_dup:
 579         return -ENOMEM;
 580 }
 581
 582 static __net_exit void sysctl_core_net_exit(struct net *net)
 583 {
 584         struct ctl_table *tbl;
 585
 586         tbl = net->core.sysctl_hdr->ctl_table_arg;
 587         unregister_net_sysctl_table(net->core.sysctl_hdr);
 588         BUG_ON(tbl == netns_core_table);
 589         kfree(tbl);
 590 }
 591
 592 static __net_initdata struct pernet_operations sysctl_core_ops = {
 593         .init = sysctl_core_net_init,
 594         .exit = sysctl_core_net_exit,
 595 };
 596
 597 static __init int sysctl_core_init(void)
 598 {
 599         register_net_sysctl(&init_net, "net/core", net_core_table);
 600         return register_pernet_subsys(&sysctl_core_ops);
 601 }
 602
 603 fs_initcall(sysctl_core_init);