net/core/sysctl_net_core.c

   1 /* -*- linux-c -*-
   2  * sysctl_net_core.c: sysctl interface to net core subsystem.
   3  *
   4  * Begun April 1, 1996, Mike Shaver.
   5  * Added /proc/sys/net/core directory entry (empty =) ). [MS]
   6  */
   7
   8 #include <linux/mm.h>
   9 #include <linux/sysctl.h>
  10 #include <linux/module.h>
  11 #include <linux/socket.h>
  12 #include <linux/netdevice.h>
  13 #include <linux/ratelimit.h>
  14 #include <linux/vmalloc.h>
  15 #include <linux/init.h>
  16 #include <linux/slab.h>
  17 #include <linux/kmemleak.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 min_sndbuf = SOCK_MIN_SNDBUF;
  28 static int min_rcvbuf = SOCK_MIN_RCVBUF;
  29 static int max_skb_frags = MAX_SKB_FRAGS;
  30
  31 static int net_msg_warn;        /* Unused, but still a sysctl */
  32
  33 #ifdef CONFIG_RPS
  34 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
  35                                 void __user *buffer, size_t *lenp, loff_t *ppos)
  36 {
  37         unsigned int orig_size, size;
  38         int ret, i;
  39         struct ctl_table tmp = {
  40                 .data = &size,
  41                 .maxlen = sizeof(size),
  42                 .mode = table->mode
  43         };
  44         struct rps_sock_flow_table *orig_sock_table, *sock_table;
  45         static DEFINE_MUTEX(sock_flow_mutex);
  46
  47         mutex_lock(&sock_flow_mutex);
  48
  49         orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
  50                                         lockdep_is_held(&sock_flow_mutex));
  51         size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
  52
  53         ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
  54
  55         if (write) {
  56                 if (size) {
  57                         if (size > 1<<29) {
  58                                 /* Enforce limit to prevent overflow */
  59                                 mutex_unlock(&sock_flow_mutex);
  60                                 return -EINVAL;
  61                         }
  62                         size = roundup_pow_of_two(size);
  63                         if (size != orig_size) {
  64                                 sock_table =
  65                                     vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
  66                                 if (!sock_table) {
  67                                         mutex_unlock(&sock_flow_mutex);
  68                                         return -ENOMEM;
  69                                 }
  70                                 rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
  71                                 sock_table->mask = size - 1;
  72                         } else
  73                                 sock_table = orig_sock_table;
  74
  75                         for (i = 0; i < size; i++)
  76                                 sock_table->ents[i] = RPS_NO_CPU;
  77                 } else
  78                         sock_table = NULL;
  79
  80                 if (sock_table != orig_sock_table) {
  81                         rcu_assign_pointer(rps_sock_flow_table, sock_table);
  82                         if (sock_table) {
  83                                 static_key_slow_inc(&rps_needed);
  84                                 static_key_slow_inc(&rfs_needed);
  85                         }
  86                         if (orig_sock_table) {
  87                                 static_key_slow_dec(&rps_needed);
  88                                 static_key_slow_dec(&rfs_needed);
  89                                 synchronize_rcu();
  90                                 vfree(orig_sock_table);
  91                         }
  92                 }
  93         }
  94
  95         mutex_unlock(&sock_flow_mutex);
  96
  97         return ret;
  98 }
  99 #endif /* CONFIG_RPS */
 100
 101 #ifdef CONFIG_NET_FLOW_LIMIT
 102 static DEFINE_MUTEX(flow_limit_update_mutex);
 103
 104 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
 105                                  void __user *buffer, size_t *lenp,
 106                                  loff_t *ppos)
 107 {
 108         struct sd_flow_limit *cur;
 109         struct softnet_data *sd;
 110         cpumask_var_t mask;
 111         int i, len, ret = 0;
 112
 113         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 114                 return -ENOMEM;
 115
 116         if (write) {
 117                 ret = cpumask_parse_user(buffer, *lenp, mask);
 118                 if (ret)
 119                         goto done;
 120
 121                 mutex_lock(&flow_limit_update_mutex);
 122                 len = sizeof(*cur) + netdev_flow_limit_table_len;
 123                 for_each_possible_cpu(i) {
 124                         sd = &per_cpu(softnet_data, i);
 125                         cur = rcu_dereference_protected(sd->flow_limit,
 126                                      lockdep_is_held(&flow_limit_update_mutex));
 127                         if (cur && !cpumask_test_cpu(i, mask)) {
 128                                 RCU_INIT_POINTER(sd->flow_limit, NULL);
 129                                 synchronize_rcu();
 130                                 kfree(cur);
 131                         } else if (!cur && cpumask_test_cpu(i, mask)) {
 132                                 cur = kzalloc_node(len, GFP_KERNEL,
 133                                                    cpu_to_node(i));
 134                                 if (!cur) {
 135                                         /* not unwinding previous changes */
 136                                         ret = -ENOMEM;
 137                                         goto write_unlock;
 138                                 }
 139                                 cur->num_buckets = netdev_flow_limit_table_len;
 140                                 rcu_assign_pointer(sd->flow_limit, cur);
 141                         }
 142                 }
 143 write_unlock:
 144                 mutex_unlock(&flow_limit_update_mutex);
 145         } else {
 146                 char kbuf[128];
 147
 148                 if (*ppos || !*lenp) {
 149                         *lenp = 0;
 150                         goto done;
 151                 }
 152
 153                 cpumask_clear(mask);
 154                 rcu_read_lock();
 155                 for_each_possible_cpu(i) {
 156                         sd = &per_cpu(softnet_data, i);
 157                         if (rcu_dereference(sd->flow_limit))
 158                                 cpumask_set_cpu(i, mask);
 159                 }
 160                 rcu_read_unlock();
 161
 162                 len = min(sizeof(kbuf) - 1, *lenp);
 163                 len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
 164                 if (!len) {
 165                         *lenp = 0;
 166                         goto done;
 167                 }
 168                 if (len < *lenp)
 169                         kbuf[len++] = '\n';
 170                 if (copy_to_user(buffer, kbuf, len)) {
 171                         ret = -EFAULT;
 172                         goto done;
 173                 }
 174                 *lenp = len;
 175                 *ppos += len;
 176         }
 177
 178 done:
 179         free_cpumask_var(mask);
 180         return ret;
 181 }
 182
 183 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
 184                                        void __user *buffer, size_t *lenp,
 185                                        loff_t *ppos)
 186 {
 187         unsigned int old, *ptr;
 188         int ret;
 189
 190         mutex_lock(&flow_limit_update_mutex);
 191
 192         ptr = table->data;
 193         old = *ptr;
 194         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 195         if (!ret && write && !is_power_of_2(*ptr)) {
 196                 *ptr = old;
 197                 ret = -EINVAL;
 198         }
 199
 200         mutex_unlock(&flow_limit_update_mutex);
 201         return ret;
 202 }
 203 #endif /* CONFIG_NET_FLOW_LIMIT */
 204
 205 #ifdef CONFIG_NET_SCHED
 206 static int set_default_qdisc(struct ctl_table *table, int write,
 207                              void __user *buffer, size_t *lenp, loff_t *ppos)
 208 {
 209         char id[IFNAMSIZ];
 210         struct ctl_table tbl = {
 211                 .data = id,
 212                 .maxlen = IFNAMSIZ,
 213         };
 214         int ret;
 215
 216         qdisc_get_default(id, IFNAMSIZ);
 217
 218         ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
 219         if (write && ret == 0)
 220                 ret = qdisc_set_default(id);
 221         return ret;
 222 }
 223 #endif
 224
 225 static int proc_do_dev_weight(struct ctl_table *table, int write,
 226                            void __user *buffer, size_t *lenp, loff_t *ppos)
 227 {
 228         int ret;
 229
 230         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 231         if (ret != 0)
 232                 return ret;
 233
 234         dev_rx_weight = weight_p * dev_weight_rx_bias;
 235         dev_tx_weight = weight_p * dev_weight_tx_bias;
 236
 237         return ret;
 238 }
 239
 240 static int proc_do_rss_key(struct ctl_table *table, int write,
 241                            void __user *buffer, size_t *lenp, loff_t *ppos)
 242 {
 243         struct ctl_table fake_table;
 244         char buf[NETDEV_RSS_KEY_LEN * 3];
 245
 246         snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
 247         fake_table.data = buf;
 248         fake_table.maxlen = sizeof(buf);
 249         return proc_dostring(&fake_table, write, buffer, lenp, ppos);
 250 }
 251
 252 static struct ctl_table net_core_table[] = {
 253 #ifdef CONFIG_NET
 254         {
 255                 .procname       = "wmem_max",
 256                 .data           = &sysctl_wmem_max,
 257                 .maxlen         = sizeof(int),
 258                 .mode           = 0644,
 259                 .proc_handler   = proc_dointvec_minmax,
 260                 .extra1         = &min_sndbuf,
 261         },
 262         {
 263                 .procname       = "rmem_max",
 264                 .data           = &sysctl_rmem_max,
 265                 .maxlen         = sizeof(int),
 266                 .mode           = 0644,
 267                 .proc_handler   = proc_dointvec_minmax,
 268                 .extra1         = &min_rcvbuf,
 269         },
 270         {
 271                 .procname       = "wmem_default",
 272                 .data           = &sysctl_wmem_default,
 273                 .maxlen         = sizeof(int),
 274                 .mode           = 0644,
 275                 .proc_handler   = proc_dointvec_minmax,
 276                 .extra1         = &min_sndbuf,
 277         },
 278         {
 279                 .procname       = "rmem_default",
 280                 .data           = &sysctl_rmem_default,
 281                 .maxlen         = sizeof(int),
 282                 .mode           = 0644,
 283                 .proc_handler   = proc_dointvec_minmax,
 284                 .extra1         = &min_rcvbuf,
 285         },
 286         {
 287                 .procname       = "dev_weight",
 288                 .data           = &weight_p,
 289                 .maxlen         = sizeof(int),
 290                 .mode           = 0644,
 291                 .proc_handler   = proc_do_dev_weight,
 292         },
 293         {
 294                 .procname       = "dev_weight_rx_bias",
 295                 .data           = &dev_weight_rx_bias,
 296                 .maxlen         = sizeof(int),
 297                 .mode           = 0644,
 298                 .proc_handler   = proc_do_dev_weight,
 299         },
 300         {
 301                 .procname       = "dev_weight_tx_bias",
 302                 .data           = &dev_weight_tx_bias,
 303                 .maxlen         = sizeof(int),
 304                 .mode           = 0644,
 305                 .proc_handler   = proc_do_dev_weight,
 306         },
 307         {
 308                 .procname       = "netdev_max_backlog",
 309                 .data           = &netdev_max_backlog,
 310                 .maxlen         = sizeof(int),
 311                 .mode           = 0644,
 312                 .proc_handler   = proc_dointvec
 313         },
 314         {
 315                 .procname       = "netdev_rss_key",
 316                 .data           = &netdev_rss_key,
 317                 .maxlen         = sizeof(int),
 318                 .mode           = 0444,
 319                 .proc_handler   = proc_do_rss_key,
 320         },
 321 #ifdef CONFIG_BPF_JIT
 322         {
 323                 .procname       = "bpf_jit_enable",
 324                 .data           = &bpf_jit_enable,
 325                 .maxlen         = sizeof(int),
 326                 .mode           = 0644,
 327                 .proc_handler   = proc_dointvec
 328         },
 329 # ifdef CONFIG_HAVE_EBPF_JIT
 330         {
 331                 .procname       = "bpf_jit_harden",
 332                 .data           = &bpf_jit_harden,
 333                 .maxlen         = sizeof(int),
 334                 .mode           = 0600,
 335                 .proc_handler   = proc_dointvec,
 336         },
 337         {
 338                 .procname       = "bpf_jit_kallsyms",
 339                 .data           = &bpf_jit_kallsyms,
 340                 .maxlen         = sizeof(int),
 341                 .mode           = 0600,
 342                 .proc_handler   = proc_dointvec,
 343         },
 344 # endif
 345 #endif
 346         {
 347                 .procname       = "netdev_tstamp_prequeue",
 348                 .data           = &netdev_tstamp_prequeue,
 349                 .maxlen         = sizeof(int),
 350                 .mode           = 0644,
 351                 .proc_handler   = proc_dointvec
 352         },
 353         {
 354                 .procname       = "message_cost",
 355                 .data           = &net_ratelimit_state.interval,
 356                 .maxlen         = sizeof(int),
 357                 .mode           = 0644,
 358                 .proc_handler   = proc_dointvec_jiffies,
 359         },
 360         {
 361                 .procname       = "message_burst",
 362                 .data           = &net_ratelimit_state.burst,
 363                 .maxlen         = sizeof(int),
 364                 .mode           = 0644,
 365                 .proc_handler   = proc_dointvec,
 366         },
 367         {
 368                 .procname       = "optmem_max",
 369                 .data           = &sysctl_optmem_max,
 370                 .maxlen         = sizeof(int),
 371                 .mode           = 0644,
 372                 .proc_handler   = proc_dointvec
 373         },
 374         {
 375                 .procname       = "tstamp_allow_data",
 376                 .data           = &sysctl_tstamp_allow_data,
 377                 .maxlen         = sizeof(int),
 378                 .mode           = 0644,
 379                 .proc_handler   = proc_dointvec_minmax,
 380                 .extra1         = &zero,
 381                 .extra2         = &one
 382         },
 383 #ifdef CONFIG_RPS
 384         {
 385                 .procname       = "rps_sock_flow_entries",
 386                 .maxlen         = sizeof(int),
 387                 .mode           = 0644,
 388                 .proc_handler   = rps_sock_flow_sysctl
 389         },
 390 #endif
 391 #ifdef CONFIG_NET_FLOW_LIMIT
 392         {
 393                 .procname       = "flow_limit_cpu_bitmap",
 394                 .mode           = 0644,
 395                 .proc_handler   = flow_limit_cpu_sysctl
 396         },
 397         {
 398                 .procname       = "flow_limit_table_len",
 399                 .data           = &netdev_flow_limit_table_len,
 400                 .maxlen         = sizeof(int),
 401                 .mode           = 0644,
 402                 .proc_handler   = flow_limit_table_len_sysctl
 403         },
 404 #endif /* CONFIG_NET_FLOW_LIMIT */
 405 #ifdef CONFIG_NET_RX_BUSY_POLL
 406         {
 407                 .procname       = "busy_poll",
 408                 .data           = &sysctl_net_busy_poll,
 409                 .maxlen         = sizeof(unsigned int),
 410                 .mode           = 0644,
 411                 .proc_handler   = proc_dointvec_minmax,
 412                 .extra1         = &zero,
 413         },
 414         {
 415                 .procname       = "busy_read",
 416                 .data           = &sysctl_net_busy_read,
 417                 .maxlen         = sizeof(unsigned int),
 418                 .mode           = 0644,
 419                 .proc_handler   = proc_dointvec_minmax,
 420                 .extra1         = &zero,
 421         },
 422 #endif
 423 #ifdef CONFIG_NET_SCHED
 424         {
 425                 .procname       = "default_qdisc",
 426                 .mode           = 0644,
 427                 .maxlen         = IFNAMSIZ,
 428                 .proc_handler   = set_default_qdisc
 429         },
 430 #endif
 431 #endif /* CONFIG_NET */
 432         {
 433                 .procname       = "netdev_budget",
 434                 .data           = &netdev_budget,
 435                 .maxlen         = sizeof(int),
 436                 .mode           = 0644,
 437                 .proc_handler   = proc_dointvec
 438         },
 439         {
 440                 .procname       = "warnings",
 441                 .data           = &net_msg_warn,
 442                 .maxlen         = sizeof(int),
 443                 .mode           = 0644,
 444                 .proc_handler   = proc_dointvec
 445         },
 446         {
 447                 .procname       = "max_skb_frags",
 448                 .data           = &sysctl_max_skb_frags,
 449                 .maxlen         = sizeof(int),
 450                 .mode           = 0644,
 451                 .proc_handler   = proc_dointvec_minmax,
 452                 .extra1         = &one,
 453                 .extra2         = &max_skb_frags,
 454         },
 455         {
 456                 .procname       = "netdev_budget_usecs",
 457                 .data           = &netdev_budget_usecs,
 458                 .maxlen         = sizeof(unsigned int),
 459                 .mode           = 0644,
 460                 .proc_handler   = proc_dointvec_minmax,
 461                 .extra1         = &zero,
 462         },
 463         { }
 464 };
 465
 466 static struct ctl_table netns_core_table[] = {
 467         {
 468                 .procname       = "somaxconn",
 469                 .data           = &init_net.core.sysctl_somaxconn,
 470                 .maxlen         = sizeof(int),
 471                 .mode           = 0644,
 472                 .extra1         = &zero,
 473                 .proc_handler   = proc_dointvec_minmax
 474         },
 475         { }
 476 };
 477
 478 static __net_init int sysctl_core_net_init(struct net *net)
 479 {
 480         struct ctl_table *tbl;
 481
 482         tbl = netns_core_table;
 483         if (!net_eq(net, &init_net)) {
 484                 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
 485                 if (tbl == NULL)
 486                         goto err_dup;
 487
 488                 tbl[0].data = &net->core.sysctl_somaxconn;
 489
 490                 /* Don't export any sysctls to unprivileged users */
 491                 if (net->user_ns != &init_user_ns) {
 492                         tbl[0].procname = NULL;
 493                 }
 494         }
 495
 496         net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
 497         if (net->core.sysctl_hdr == NULL)
 498                 goto err_reg;
 499
 500         return 0;
 501
 502 err_reg:
 503         if (tbl != netns_core_table)
 504                 kfree(tbl);
 505 err_dup:
 506         return -ENOMEM;
 507 }
 508
 509 static __net_exit void sysctl_core_net_exit(struct net *net)
 510 {
 511         struct ctl_table *tbl;
 512
 513         tbl = net->core.sysctl_hdr->ctl_table_arg;
 514         unregister_net_sysctl_table(net->core.sysctl_hdr);
 515         BUG_ON(tbl == netns_core_table);
 516         kfree(tbl);
 517 }
 518
 519 static __net_initdata struct pernet_operations sysctl_core_ops = {
 520         .init = sysctl_core_net_init,
 521         .exit = sysctl_core_net_exit,
 522 };
 523
 524 static __init int sysctl_core_init(void)
 525 {
 526         register_net_sysctl(&init_net, "net/core", net_core_table);
 527         return register_pernet_subsys(&sysctl_core_ops);
 528 }
 529
 530 fs_initcall(sysctl_core_init);