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
  33 static int net_msg_warn;        /* Unused, but still a sysctl */
  34
  35 #ifdef CONFIG_RPS
  36 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
  37                                 void __user *buffer, size_t *lenp, loff_t *ppos)
  38 {
  39         unsigned int orig_size, size;
  40         int ret, i;
  41         struct ctl_table tmp = {
  42                 .data = &size,
  43                 .maxlen = sizeof(size),
  44                 .mode = table->mode
  45         };
  46         struct rps_sock_flow_table *orig_sock_table, *sock_table;
  47         static DEFINE_MUTEX(sock_flow_mutex);
  48
  49         mutex_lock(&sock_flow_mutex);
  50
  51         orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
  52                                         lockdep_is_held(&sock_flow_mutex));
  53         size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
  54
  55         ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
  56
  57         if (write) {
  58                 if (size) {
  59                         if (size > 1<<29) {
  60                                 /* Enforce limit to prevent overflow */
  61                                 mutex_unlock(&sock_flow_mutex);
  62                                 return -EINVAL;
  63                         }
  64                         size = roundup_pow_of_two(size);
  65                         if (size != orig_size) {
  66                                 sock_table =
  67                                     vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
  68                                 if (!sock_table) {
  69                                         mutex_unlock(&sock_flow_mutex);
  70                                         return -ENOMEM;
  71                                 }
  72                                 rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
  73                                 sock_table->mask = size - 1;
  74                         } else
  75                                 sock_table = orig_sock_table;
  76
  77                         for (i = 0; i < size; i++)
  78                                 sock_table->ents[i] = RPS_NO_CPU;
  79                 } else
  80                         sock_table = NULL;
  81
  82                 if (sock_table != orig_sock_table) {
  83                         rcu_assign_pointer(rps_sock_flow_table, sock_table);
  84                         if (sock_table) {
  85                                 static_key_slow_inc(&rps_needed);
  86                                 static_key_slow_inc(&rfs_needed);
  87                         }
  88                         if (orig_sock_table) {
  89                                 static_key_slow_dec(&rps_needed);
  90                                 static_key_slow_dec(&rfs_needed);
  91                                 synchronize_rcu();
  92                                 vfree(orig_sock_table);
  93                         }
  94                 }
  95         }
  96
  97         mutex_unlock(&sock_flow_mutex);
  98
  99         return ret;
 100 }
 101 #endif /* CONFIG_RPS */
 102
 103 #ifdef CONFIG_NET_FLOW_LIMIT
 104 static DEFINE_MUTEX(flow_limit_update_mutex);
 105
 106 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
 107                                  void __user *buffer, size_t *lenp,
 108                                  loff_t *ppos)
 109 {
 110         struct sd_flow_limit *cur;
 111         struct softnet_data *sd;
 112         cpumask_var_t mask;
 113         int i, len, ret = 0;
 114
 115         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 116                 return -ENOMEM;
 117
 118         if (write) {
 119                 ret = cpumask_parse_user(buffer, *lenp, mask);
 120                 if (ret)
 121                         goto done;
 122
 123                 mutex_lock(&flow_limit_update_mutex);
 124                 len = sizeof(*cur) + netdev_flow_limit_table_len;
 125                 for_each_possible_cpu(i) {
 126                         sd = &per_cpu(softnet_data, i);
 127                         cur = rcu_dereference_protected(sd->flow_limit,
 128                                      lockdep_is_held(&flow_limit_update_mutex));
 129                         if (cur && !cpumask_test_cpu(i, mask)) {
 130                                 RCU_INIT_POINTER(sd->flow_limit, NULL);
 131                                 synchronize_rcu();
 132                                 kfree(cur);
 133                         } else if (!cur && cpumask_test_cpu(i, mask)) {
 134                                 cur = kzalloc_node(len, GFP_KERNEL,
 135                                                    cpu_to_node(i));
 136                                 if (!cur) {
 137                                         /* not unwinding previous changes */
 138                                         ret = -ENOMEM;
 139                                         goto write_unlock;
 140                                 }
 141                                 cur->num_buckets = netdev_flow_limit_table_len;
 142                                 rcu_assign_pointer(sd->flow_limit, cur);
 143                         }
 144                 }
 145 write_unlock:
 146                 mutex_unlock(&flow_limit_update_mutex);
 147         } else {
 148                 char kbuf[128];
 149
 150                 if (*ppos || !*lenp) {
 151                         *lenp = 0;
 152                         goto done;
 153                 }
 154
 155                 cpumask_clear(mask);
 156                 rcu_read_lock();
 157                 for_each_possible_cpu(i) {
 158                         sd = &per_cpu(softnet_data, i);
 159                         if (rcu_dereference(sd->flow_limit))
 160                                 cpumask_set_cpu(i, mask);
 161                 }
 162                 rcu_read_unlock();
 163
 164                 len = min(sizeof(kbuf) - 1, *lenp);
 165                 len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
 166                 if (!len) {
 167                         *lenp = 0;
 168                         goto done;
 169                 }
 170                 if (len < *lenp)
 171                         kbuf[len++] = '\n';
 172                 if (copy_to_user(buffer, kbuf, len)) {
 173                         ret = -EFAULT;
 174                         goto done;
 175                 }
 176                 *lenp = len;
 177                 *ppos += len;
 178         }
 179
 180 done:
 181         free_cpumask_var(mask);
 182         return ret;
 183 }
 184
 185 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
 186                                        void __user *buffer, size_t *lenp,
 187                                        loff_t *ppos)
 188 {
 189         unsigned int old, *ptr;
 190         int ret;
 191
 192         mutex_lock(&flow_limit_update_mutex);
 193
 194         ptr = table->data;
 195         old = *ptr;
 196         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 197         if (!ret && write && !is_power_of_2(*ptr)) {
 198                 *ptr = old;
 199                 ret = -EINVAL;
 200         }
 201
 202         mutex_unlock(&flow_limit_update_mutex);
 203         return ret;
 204 }
 205 #endif /* CONFIG_NET_FLOW_LIMIT */
 206
 207 #ifdef CONFIG_NET_SCHED
 208 static int set_default_qdisc(struct ctl_table *table, int write,
 209                              void __user *buffer, size_t *lenp, loff_t *ppos)
 210 {
 211         char id[IFNAMSIZ];
 212         struct ctl_table tbl = {
 213                 .data = id,
 214                 .maxlen = IFNAMSIZ,
 215         };
 216         int ret;
 217
 218         qdisc_get_default(id, IFNAMSIZ);
 219
 220         ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
 221         if (write && ret == 0)
 222                 ret = qdisc_set_default(id);
 223         return ret;
 224 }
 225 #endif
 226
 227 static int proc_do_dev_weight(struct ctl_table *table, int write,
 228                            void __user *buffer, size_t *lenp, loff_t *ppos)
 229 {
 230         int ret;
 231
 232         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 233         if (ret != 0)
 234                 return ret;
 235
 236         dev_rx_weight = weight_p * dev_weight_rx_bias;
 237         dev_tx_weight = weight_p * dev_weight_tx_bias;
 238
 239         return ret;
 240 }
 241
 242 static int proc_do_rss_key(struct ctl_table *table, int write,
 243                            void __user *buffer, size_t *lenp, loff_t *ppos)
 244 {
 245         struct ctl_table fake_table;
 246         char buf[NETDEV_RSS_KEY_LEN * 3];
 247
 248         snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
 249         fake_table.data = buf;
 250         fake_table.maxlen = sizeof(buf);
 251         return proc_dostring(&fake_table, write, buffer, lenp, ppos);
 252 }
 253
 254 #ifdef CONFIG_BPF_JIT
 255 static int proc_dointvec_minmax_bpf_enable(struct ctl_table *table, int write,
 256                                            void __user *buffer, size_t *lenp,
 257                                            loff_t *ppos)
 258 {
 259         int ret, jit_enable = *(int *)table->data;
 260         struct ctl_table tmp = *table;
 261
 262         if (write && !capable(CAP_SYS_ADMIN))
 263                 return -EPERM;
 264
 265         tmp.data = &jit_enable;
 266         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
 267         if (write && !ret) {
 268                 if (jit_enable < 2 ||
 269                     (jit_enable == 2 && bpf_dump_raw_ok())) {
 270                         *(int *)table->data = jit_enable;
 271                         if (jit_enable == 2)
 272                                 pr_warn("bpf_jit_enable = 2 was set! NEVER use this in production, only for JIT debugging!\n");
 273                 } else {
 274                         ret = -EPERM;
 275                 }
 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
 292 #endif
 293
 294 static struct ctl_table net_core_table[] = {
 295 #ifdef CONFIG_NET
 296         {
 297                 .procname       = "wmem_max",
 298                 .data           = &sysctl_wmem_max,
 299                 .maxlen         = sizeof(int),
 300                 .mode           = 0644,
 301                 .proc_handler   = proc_dointvec_minmax,
 302                 .extra1         = &min_sndbuf,
 303         },
 304         {
 305                 .procname       = "rmem_max",
 306                 .data           = &sysctl_rmem_max,
 307                 .maxlen         = sizeof(int),
 308                 .mode           = 0644,
 309                 .proc_handler   = proc_dointvec_minmax,
 310                 .extra1         = &min_rcvbuf,
 311         },
 312         {
 313                 .procname       = "wmem_default",
 314                 .data           = &sysctl_wmem_default,
 315                 .maxlen         = sizeof(int),
 316                 .mode           = 0644,
 317                 .proc_handler   = proc_dointvec_minmax,
 318                 .extra1         = &min_sndbuf,
 319         },
 320         {
 321                 .procname       = "rmem_default",
 322                 .data           = &sysctl_rmem_default,
 323                 .maxlen         = sizeof(int),
 324                 .mode           = 0644,
 325                 .proc_handler   = proc_dointvec_minmax,
 326                 .extra1         = &min_rcvbuf,
 327         },
 328         {
 329                 .procname       = "dev_weight",
 330                 .data           = &weight_p,
 331                 .maxlen         = sizeof(int),
 332                 .mode           = 0644,
 333                 .proc_handler   = proc_do_dev_weight,
 334         },
 335         {
 336                 .procname       = "dev_weight_rx_bias",
 337                 .data           = &dev_weight_rx_bias,
 338                 .maxlen         = sizeof(int),
 339                 .mode           = 0644,
 340                 .proc_handler   = proc_do_dev_weight,
 341         },
 342         {
 343                 .procname       = "dev_weight_tx_bias",
 344                 .data           = &dev_weight_tx_bias,
 345                 .maxlen         = sizeof(int),
 346                 .mode           = 0644,
 347                 .proc_handler   = proc_do_dev_weight,
 348         },
 349         {
 350                 .procname       = "netdev_max_backlog",
 351                 .data           = &netdev_max_backlog,
 352                 .maxlen         = sizeof(int),
 353                 .mode           = 0644,
 354                 .proc_handler   = proc_dointvec
 355         },
 356         {
 357                 .procname       = "netdev_rss_key",
 358                 .data           = &netdev_rss_key,
 359                 .maxlen         = sizeof(int),
 360                 .mode           = 0444,
 361                 .proc_handler   = proc_do_rss_key,
 362         },
 363 #ifdef CONFIG_BPF_JIT
 364         {
 365                 .procname       = "bpf_jit_enable",
 366                 .data           = &bpf_jit_enable,
 367                 .maxlen         = sizeof(int),
 368                 .mode           = 0644,
 369                 .proc_handler   = proc_dointvec_minmax_bpf_enable,
 370 # ifdef CONFIG_BPF_JIT_ALWAYS_ON
 371                 .extra1         = &one,
 372                 .extra2         = &one,
 373 # else
 374                 .extra1         = &zero,
 375                 .extra2         = &two,
 376 # endif
 377         },
 378 # ifdef CONFIG_HAVE_EBPF_JIT
 379         {
 380                 .procname       = "bpf_jit_harden",
 381                 .data           = &bpf_jit_harden,
 382                 .maxlen         = sizeof(int),
 383                 .mode           = 0600,
 384                 .proc_handler   = proc_dointvec_minmax_bpf_restricted,
 385                 .extra1         = &zero,
 386                 .extra2         = &two,
 387         },
 388         {
 389                 .procname       = "bpf_jit_kallsyms",
 390                 .data           = &bpf_jit_kallsyms,
 391                 .maxlen         = sizeof(int),
 392                 .mode           = 0600,
 393                 .proc_handler   = proc_dointvec_minmax_bpf_restricted,
 394                 .extra1         = &zero,
 395                 .extra2         = &one,
 396         },
 397 # endif
 398 #endif
 399         {
 400                 .procname       = "netdev_tstamp_prequeue",
 401                 .data           = &netdev_tstamp_prequeue,
 402                 .maxlen         = sizeof(int),
 403                 .mode           = 0644,
 404                 .proc_handler   = proc_dointvec
 405         },
 406         {
 407                 .procname       = "message_cost",
 408                 .data           = &net_ratelimit_state.interval,
 409                 .maxlen         = sizeof(int),
 410                 .mode           = 0644,
 411                 .proc_handler   = proc_dointvec_jiffies,
 412         },
 413         {
 414                 .procname       = "message_burst",
 415                 .data           = &net_ratelimit_state.burst,
 416                 .maxlen         = sizeof(int),
 417                 .mode           = 0644,
 418                 .proc_handler   = proc_dointvec,
 419         },
 420         {
 421                 .procname       = "optmem_max",
 422                 .data           = &sysctl_optmem_max,
 423                 .maxlen         = sizeof(int),
 424                 .mode           = 0644,
 425                 .proc_handler   = proc_dointvec
 426         },
 427         {
 428                 .procname       = "tstamp_allow_data",
 429                 .data           = &sysctl_tstamp_allow_data,
 430                 .maxlen         = sizeof(int),
 431                 .mode           = 0644,
 432                 .proc_handler   = proc_dointvec_minmax,
 433                 .extra1         = &zero,
 434                 .extra2         = &one
 435         },
 436 #ifdef CONFIG_RPS
 437         {
 438                 .procname       = "rps_sock_flow_entries",
 439                 .maxlen         = sizeof(int),
 440                 .mode           = 0644,
 441                 .proc_handler   = rps_sock_flow_sysctl
 442         },
 443 #endif
 444 #ifdef CONFIG_NET_FLOW_LIMIT
 445         {
 446                 .procname       = "flow_limit_cpu_bitmap",
 447                 .mode           = 0644,
 448                 .proc_handler   = flow_limit_cpu_sysctl
 449         },
 450         {
 451                 .procname       = "flow_limit_table_len",
 452                 .data           = &netdev_flow_limit_table_len,
 453                 .maxlen         = sizeof(int),
 454                 .mode           = 0644,
 455                 .proc_handler   = flow_limit_table_len_sysctl
 456         },
 457 #endif /* CONFIG_NET_FLOW_LIMIT */
 458 #ifdef CONFIG_NET_RX_BUSY_POLL
 459         {
 460                 .procname       = "busy_poll",
 461                 .data           = &sysctl_net_busy_poll,
 462                 .maxlen         = sizeof(unsigned int),
 463                 .mode           = 0644,
 464                 .proc_handler   = proc_dointvec_minmax,
 465                 .extra1         = &zero,
 466         },
 467         {
 468                 .procname       = "busy_read",
 469                 .data           = &sysctl_net_busy_read,
 470                 .maxlen         = sizeof(unsigned int),
 471                 .mode           = 0644,
 472                 .proc_handler   = proc_dointvec_minmax,
 473                 .extra1         = &zero,
 474         },
 475 #endif
 476 #ifdef CONFIG_NET_SCHED
 477         {
 478                 .procname       = "default_qdisc",
 479                 .mode           = 0644,
 480                 .maxlen         = IFNAMSIZ,
 481                 .proc_handler   = set_default_qdisc
 482         },
 483 #endif
 484 #endif /* CONFIG_NET */
 485         {
 486                 .procname       = "netdev_budget",
 487                 .data           = &netdev_budget,
 488                 .maxlen         = sizeof(int),
 489                 .mode           = 0644,
 490                 .proc_handler   = proc_dointvec
 491         },
 492         {
 493                 .procname       = "warnings",
 494                 .data           = &net_msg_warn,
 495                 .maxlen         = sizeof(int),
 496                 .mode           = 0644,
 497                 .proc_handler   = proc_dointvec
 498         },
 499         {
 500                 .procname       = "max_skb_frags",
 501                 .data           = &sysctl_max_skb_frags,
 502                 .maxlen         = sizeof(int),
 503                 .mode           = 0644,
 504                 .proc_handler   = proc_dointvec_minmax,
 505                 .extra1         = &one,
 506                 .extra2         = &max_skb_frags,
 507         },
 508         {
 509                 .procname       = "netdev_budget_usecs",
 510                 .data           = &netdev_budget_usecs,
 511                 .maxlen         = sizeof(unsigned int),
 512                 .mode           = 0644,
 513                 .proc_handler   = proc_dointvec_minmax,
 514                 .extra1         = &zero,
 515         },
 516         { }
 517 };
 518
 519 static struct ctl_table netns_core_table[] = {
 520         {
 521                 .procname       = "somaxconn",
 522                 .data           = &init_net.core.sysctl_somaxconn,
 523                 .maxlen         = sizeof(int),
 524                 .mode           = 0644,
 525                 .extra1         = &zero,
 526                 .proc_handler   = proc_dointvec_minmax
 527         },
 528         { }
 529 };
 530
 531 static __net_init int sysctl_core_net_init(struct net *net)
 532 {
 533         struct ctl_table *tbl;
 534
 535         tbl = netns_core_table;
 536         if (!net_eq(net, &init_net)) {
 537                 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
 538                 if (tbl == NULL)
 539                         goto err_dup;
 540
 541                 tbl[0].data = &net->core.sysctl_somaxconn;
 542
 543                 /* Don't export any sysctls to unprivileged users */
 544                 if (net->user_ns != &init_user_ns) {
 545                         tbl[0].procname = NULL;
 546                 }
 547         }
 548
 549         net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
 550         if (net->core.sysctl_hdr == NULL)
 551                 goto err_reg;
 552
 553         return 0;
 554
 555 err_reg:
 556         if (tbl != netns_core_table)
 557                 kfree(tbl);
 558 err_dup:
 559         return -ENOMEM;
 560 }
 561
 562 static __net_exit void sysctl_core_net_exit(struct net *net)
 563 {
 564         struct ctl_table *tbl;
 565
 566         tbl = net->core.sysctl_hdr->ctl_table_arg;
 567         unregister_net_sysctl_table(net->core.sysctl_hdr);
 568         BUG_ON(tbl == netns_core_table);
 569         kfree(tbl);
 570 }
 571
 572 static __net_initdata struct pernet_operations sysctl_core_ops = {
 573         .init = sysctl_core_net_init,
 574         .exit = sysctl_core_net_exit,
 575 };
 576
 577 static __init int sysctl_core_init(void)
 578 {
 579         register_net_sysctl(&init_net, "net/core", net_core_table);
 580         return register_pernet_subsys(&sysctl_core_ops);
 581 }
 582
 583 fs_initcall(sysctl_core_init);