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                         if (orig_sock_table) {
  85                                 static_key_slow_dec(&rps_needed);
  86                                 synchronize_rcu();
  87                                 vfree(orig_sock_table);
  88                         }
  89                 }
  90         }
  91
  92         mutex_unlock(&sock_flow_mutex);
  93
  94         return ret;
  95 }
  96 #endif /* CONFIG_RPS */
  97
  98 #ifdef CONFIG_NET_FLOW_LIMIT
  99 static DEFINE_MUTEX(flow_limit_update_mutex);
 100
 101 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
 102                                  void __user *buffer, size_t *lenp,
 103                                  loff_t *ppos)
 104 {
 105         struct sd_flow_limit *cur;
 106         struct softnet_data *sd;
 107         cpumask_var_t mask;
 108         int i, len, ret = 0;
 109
 110         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 111                 return -ENOMEM;
 112
 113         if (write) {
 114                 ret = cpumask_parse_user(buffer, *lenp, mask);
 115                 if (ret)
 116                         goto done;
 117
 118                 mutex_lock(&flow_limit_update_mutex);
 119                 len = sizeof(*cur) + netdev_flow_limit_table_len;
 120                 for_each_possible_cpu(i) {
 121                         sd = &per_cpu(softnet_data, i);
 122                         cur = rcu_dereference_protected(sd->flow_limit,
 123                                      lockdep_is_held(&flow_limit_update_mutex));
 124                         if (cur && !cpumask_test_cpu(i, mask)) {
 125                                 RCU_INIT_POINTER(sd->flow_limit, NULL);
 126                                 synchronize_rcu();
 127                                 kfree(cur);
 128                         } else if (!cur && cpumask_test_cpu(i, mask)) {
 129                                 cur = kzalloc_node(len, GFP_KERNEL,
 130                                                    cpu_to_node(i));
 131                                 if (!cur) {
 132                                         /* not unwinding previous changes */
 133                                         ret = -ENOMEM;
 134                                         goto write_unlock;
 135                                 }
 136                                 cur->num_buckets = netdev_flow_limit_table_len;
 137                                 rcu_assign_pointer(sd->flow_limit, cur);
 138                         }
 139                 }
 140 write_unlock:
 141                 mutex_unlock(&flow_limit_update_mutex);
 142         } else {
 143                 char kbuf[128];
 144
 145                 if (*ppos || !*lenp) {
 146                         *lenp = 0;
 147                         goto done;
 148                 }
 149
 150                 cpumask_clear(mask);
 151                 rcu_read_lock();
 152                 for_each_possible_cpu(i) {
 153                         sd = &per_cpu(softnet_data, i);
 154                         if (rcu_dereference(sd->flow_limit))
 155                                 cpumask_set_cpu(i, mask);
 156                 }
 157                 rcu_read_unlock();
 158
 159                 len = min(sizeof(kbuf) - 1, *lenp);
 160                 len = scnprintf(kbuf, len, "%*pb", cpumask_pr_args(mask));
 161                 if (!len) {
 162                         *lenp = 0;
 163                         goto done;
 164                 }
 165                 if (len < *lenp)
 166                         kbuf[len++] = '\n';
 167                 if (copy_to_user(buffer, kbuf, len)) {
 168                         ret = -EFAULT;
 169                         goto done;
 170                 }
 171                 *lenp = len;
 172                 *ppos += len;
 173         }
 174
 175 done:
 176         free_cpumask_var(mask);
 177         return ret;
 178 }
 179
 180 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
 181                                        void __user *buffer, size_t *lenp,
 182                                        loff_t *ppos)
 183 {
 184         unsigned int old, *ptr;
 185         int ret;
 186
 187         mutex_lock(&flow_limit_update_mutex);
 188
 189         ptr = table->data;
 190         old = *ptr;
 191         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 192         if (!ret && write && !is_power_of_2(*ptr)) {
 193                 *ptr = old;
 194                 ret = -EINVAL;
 195         }
 196
 197         mutex_unlock(&flow_limit_update_mutex);
 198         return ret;
 199 }
 200 #endif /* CONFIG_NET_FLOW_LIMIT */
 201
 202 #ifdef CONFIG_NET_SCHED
 203 static int set_default_qdisc(struct ctl_table *table, int write,
 204                              void __user *buffer, size_t *lenp, loff_t *ppos)
 205 {
 206         char id[IFNAMSIZ];
 207         struct ctl_table tbl = {
 208                 .data = id,
 209                 .maxlen = IFNAMSIZ,
 210         };
 211         int ret;
 212
 213         qdisc_get_default(id, IFNAMSIZ);
 214
 215         ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
 216         if (write && ret == 0)
 217                 ret = qdisc_set_default(id);
 218         return ret;
 219 }
 220 #endif
 221
 222 static int proc_do_rss_key(struct ctl_table *table, int write,
 223                            void __user *buffer, size_t *lenp, loff_t *ppos)
 224 {
 225         struct ctl_table fake_table;
 226         char buf[NETDEV_RSS_KEY_LEN * 3];
 227
 228         snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
 229         fake_table.data = buf;
 230         fake_table.maxlen = sizeof(buf);
 231         return proc_dostring(&fake_table, write, buffer, lenp, ppos);
 232 }
 233
 234 static struct ctl_table net_core_table[] = {
 235 #ifdef CONFIG_NET
 236         {
 237                 .procname       = "wmem_max",
 238                 .data           = &sysctl_wmem_max,
 239                 .maxlen         = sizeof(int),
 240                 .mode           = 0644,
 241                 .proc_handler   = proc_dointvec_minmax,
 242                 .extra1         = &min_sndbuf,
 243         },
 244         {
 245                 .procname       = "rmem_max",
 246                 .data           = &sysctl_rmem_max,
 247                 .maxlen         = sizeof(int),
 248                 .mode           = 0644,
 249                 .proc_handler   = proc_dointvec_minmax,
 250                 .extra1         = &min_rcvbuf,
 251         },
 252         {
 253                 .procname       = "wmem_default",
 254                 .data           = &sysctl_wmem_default,
 255                 .maxlen         = sizeof(int),
 256                 .mode           = 0644,
 257                 .proc_handler   = proc_dointvec_minmax,
 258                 .extra1         = &min_sndbuf,
 259         },
 260         {
 261                 .procname       = "rmem_default",
 262                 .data           = &sysctl_rmem_default,
 263                 .maxlen         = sizeof(int),
 264                 .mode           = 0644,
 265                 .proc_handler   = proc_dointvec_minmax,
 266                 .extra1         = &min_rcvbuf,
 267         },
 268         {
 269                 .procname       = "dev_weight",
 270                 .data           = &weight_p,
 271                 .maxlen         = sizeof(int),
 272                 .mode           = 0644,
 273                 .proc_handler   = proc_dointvec
 274         },
 275         {
 276                 .procname       = "netdev_max_backlog",
 277                 .data           = &netdev_max_backlog,
 278                 .maxlen         = sizeof(int),
 279                 .mode           = 0644,
 280                 .proc_handler   = proc_dointvec
 281         },
 282         {
 283                 .procname       = "netdev_rss_key",
 284                 .data           = &netdev_rss_key,
 285                 .maxlen         = sizeof(int),
 286                 .mode           = 0444,
 287                 .proc_handler   = proc_do_rss_key,
 288         },
 289 #ifdef CONFIG_BPF_JIT
 290         {
 291                 .procname       = "bpf_jit_enable",
 292                 .data           = &bpf_jit_enable,
 293                 .maxlen         = sizeof(int),
 294                 .mode           = 0644,
 295 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
 296                 .proc_handler   = proc_dointvec
 297 #else
 298                 .proc_handler   = proc_dointvec_minmax,
 299                 .extra1         = &one,
 300                 .extra2         = &one,
 301 #endif
 302         },
 303 #endif
 304         {
 305                 .procname       = "netdev_tstamp_prequeue",
 306                 .data           = &netdev_tstamp_prequeue,
 307                 .maxlen         = sizeof(int),
 308                 .mode           = 0644,
 309                 .proc_handler   = proc_dointvec
 310         },
 311         {
 312                 .procname       = "message_cost",
 313                 .data           = &net_ratelimit_state.interval,
 314                 .maxlen         = sizeof(int),
 315                 .mode           = 0644,
 316                 .proc_handler   = proc_dointvec_jiffies,
 317         },
 318         {
 319                 .procname       = "message_burst",
 320                 .data           = &net_ratelimit_state.burst,
 321                 .maxlen         = sizeof(int),
 322                 .mode           = 0644,
 323                 .proc_handler   = proc_dointvec,
 324         },
 325         {
 326                 .procname       = "optmem_max",
 327                 .data           = &sysctl_optmem_max,
 328                 .maxlen         = sizeof(int),
 329                 .mode           = 0644,
 330                 .proc_handler   = proc_dointvec
 331         },
 332         {
 333                 .procname       = "tstamp_allow_data",
 334                 .data           = &sysctl_tstamp_allow_data,
 335                 .maxlen         = sizeof(int),
 336                 .mode           = 0644,
 337                 .proc_handler   = proc_dointvec_minmax,
 338                 .extra1         = &zero,
 339                 .extra2         = &one
 340         },
 341 #ifdef CONFIG_RPS
 342         {
 343                 .procname       = "rps_sock_flow_entries",
 344                 .maxlen         = sizeof(int),
 345                 .mode           = 0644,
 346                 .proc_handler   = rps_sock_flow_sysctl
 347         },
 348 #endif
 349 #ifdef CONFIG_NET_FLOW_LIMIT
 350         {
 351                 .procname       = "flow_limit_cpu_bitmap",
 352                 .mode           = 0644,
 353                 .proc_handler   = flow_limit_cpu_sysctl
 354         },
 355         {
 356                 .procname       = "flow_limit_table_len",
 357                 .data           = &netdev_flow_limit_table_len,
 358                 .maxlen         = sizeof(int),
 359                 .mode           = 0644,
 360                 .proc_handler   = flow_limit_table_len_sysctl
 361         },
 362 #endif /* CONFIG_NET_FLOW_LIMIT */
 363 #ifdef CONFIG_NET_RX_BUSY_POLL
 364         {
 365                 .procname       = "busy_poll",
 366                 .data           = &sysctl_net_busy_poll,
 367                 .maxlen         = sizeof(unsigned int),
 368                 .mode           = 0644,
 369                 .proc_handler   = proc_dointvec_minmax,
 370                 .extra1         = &zero,
 371         },
 372         {
 373                 .procname       = "busy_read",
 374                 .data           = &sysctl_net_busy_read,
 375                 .maxlen         = sizeof(unsigned int),
 376                 .mode           = 0644,
 377                 .proc_handler   = proc_dointvec_minmax,
 378                 .extra1         = &zero,
 379         },
 380 #endif
 381 #ifdef CONFIG_NET_SCHED
 382         {
 383                 .procname       = "default_qdisc",
 384                 .mode           = 0644,
 385                 .maxlen         = IFNAMSIZ,
 386                 .proc_handler   = set_default_qdisc
 387         },
 388 #endif
 389 #endif /* CONFIG_NET */
 390         {
 391                 .procname       = "netdev_budget",
 392                 .data           = &netdev_budget,
 393                 .maxlen         = sizeof(int),
 394                 .mode           = 0644,
 395                 .proc_handler   = proc_dointvec
 396         },
 397         {
 398                 .procname       = "warnings",
 399                 .data           = &net_msg_warn,
 400                 .maxlen         = sizeof(int),
 401                 .mode           = 0644,
 402                 .proc_handler   = proc_dointvec
 403         },
 404         {
 405                 .procname       = "max_skb_frags",
 406                 .data           = &sysctl_max_skb_frags,
 407                 .maxlen         = sizeof(int),
 408                 .mode           = 0644,
 409                 .proc_handler   = proc_dointvec_minmax,
 410                 .extra1         = &one,
 411                 .extra2         = &max_skb_frags,
 412         },
 413         { }
 414 };
 415
 416 static struct ctl_table netns_core_table[] = {
 417         {
 418                 .procname       = "somaxconn",
 419                 .data           = &init_net.core.sysctl_somaxconn,
 420                 .maxlen         = sizeof(int),
 421                 .mode           = 0644,
 422                 .extra1         = &zero,
 423                 .proc_handler   = proc_dointvec_minmax
 424         },
 425         { }
 426 };
 427
 428 static __net_init int sysctl_core_net_init(struct net *net)
 429 {
 430         struct ctl_table *tbl;
 431
 432         tbl = netns_core_table;
 433         if (!net_eq(net, &init_net)) {
 434                 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
 435                 if (tbl == NULL)
 436                         goto err_dup;
 437
 438                 tbl[0].data = &net->core.sysctl_somaxconn;
 439
 440                 /* Don't export any sysctls to unprivileged users */
 441                 if (net->user_ns != &init_user_ns) {
 442                         tbl[0].procname = NULL;
 443                 }
 444         }
 445
 446         net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
 447         if (net->core.sysctl_hdr == NULL)
 448                 goto err_reg;
 449
 450         return 0;
 451
 452 err_reg:
 453         if (tbl != netns_core_table)
 454                 kfree(tbl);
 455 err_dup:
 456         return -ENOMEM;
 457 }
 458
 459 static __net_exit void sysctl_core_net_exit(struct net *net)
 460 {
 461         struct ctl_table *tbl;
 462
 463         tbl = net->core.sysctl_hdr->ctl_table_arg;
 464         unregister_net_sysctl_table(net->core.sysctl_hdr);
 465         BUG_ON(tbl == netns_core_table);
 466         kfree(tbl);
 467 }
 468
 469 static __net_initdata struct pernet_operations sysctl_core_ops = {
 470         .init = sysctl_core_net_init,
 471         .exit = sysctl_core_net_exit,
 472 };
 473
 474 static __init int sysctl_core_init(void)
 475 {
 476         register_net_sysctl(&init_net, "net/core", net_core_table);
 477         return register_pernet_subsys(&sysctl_core_ops);
 478 }
 479
 480 fs_initcall(sysctl_core_init);