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