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