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
  24 static int zero = 0;
  25 static int one = 1;
  26 static int ushort_max = USHRT_MAX;
  27
  28 #ifdef CONFIG_RPS
  29 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
  30                                 void __user *buffer, size_t *lenp, loff_t *ppos)
  31 {
  32         unsigned int orig_size, size;
  33         int ret, i;
  34         struct ctl_table tmp = {
  35                 .data = &size,
  36                 .maxlen = sizeof(size),
  37                 .mode = table->mode
  38         };
  39         struct rps_sock_flow_table *orig_sock_table, *sock_table;
  40         static DEFINE_MUTEX(sock_flow_mutex);
  41
  42         mutex_lock(&sock_flow_mutex);
  43
  44         orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
  45                                         lockdep_is_held(&sock_flow_mutex));
  46         size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
  47
  48         ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
  49
  50         if (write) {
  51                 if (size) {
  52                         if (size > 1<<30) {
  53                                 /* Enforce limit to prevent overflow */
  54                                 mutex_unlock(&sock_flow_mutex);
  55                                 return -EINVAL;
  56                         }
  57                         size = roundup_pow_of_two(size);
  58                         if (size != orig_size) {
  59                                 sock_table =
  60                                     vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
  61                                 if (!sock_table) {
  62                                         mutex_unlock(&sock_flow_mutex);
  63                                         return -ENOMEM;
  64                                 }
  65
  66                                 sock_table->mask = size - 1;
  67                         } else
  68                                 sock_table = orig_sock_table;
  69
  70                         for (i = 0; i < size; i++)
  71                                 sock_table->ents[i] = RPS_NO_CPU;
  72                 } else
  73                         sock_table = NULL;
  74
  75                 if (sock_table != orig_sock_table) {
  76                         rcu_assign_pointer(rps_sock_flow_table, sock_table);
  77                         if (sock_table)
  78                                 static_key_slow_inc(&rps_needed);
  79                         if (orig_sock_table) {
  80                                 static_key_slow_dec(&rps_needed);
  81                                 synchronize_rcu();
  82                                 vfree(orig_sock_table);
  83                         }
  84                 }
  85         }
  86
  87         mutex_unlock(&sock_flow_mutex);
  88
  89         return ret;
  90 }
  91 #endif /* CONFIG_RPS */
  92
  93 #ifdef CONFIG_NET_FLOW_LIMIT
  94 static DEFINE_MUTEX(flow_limit_update_mutex);
  95
  96 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
  97                                  void __user *buffer, size_t *lenp,
  98                                  loff_t *ppos)
  99 {
 100         struct sd_flow_limit *cur;
 101         struct softnet_data *sd;
 102         cpumask_var_t mask;
 103         int i, len, ret = 0;
 104
 105         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 106                 return -ENOMEM;
 107
 108         if (write) {
 109                 ret = cpumask_parse_user(buffer, *lenp, mask);
 110                 if (ret)
 111                         goto done;
 112
 113                 mutex_lock(&flow_limit_update_mutex);
 114                 len = sizeof(*cur) + netdev_flow_limit_table_len;
 115                 for_each_possible_cpu(i) {
 116                         sd = &per_cpu(softnet_data, i);
 117                         cur = rcu_dereference_protected(sd->flow_limit,
 118                                      lockdep_is_held(&flow_limit_update_mutex));
 119                         if (cur && !cpumask_test_cpu(i, mask)) {
 120                                 RCU_INIT_POINTER(sd->flow_limit, NULL);
 121                                 synchronize_rcu();
 122                                 kfree(cur);
 123                         } else if (!cur && cpumask_test_cpu(i, mask)) {
 124                                 cur = kzalloc(len, GFP_KERNEL);
 125                                 if (!cur) {
 126                                         /* not unwinding previous changes */
 127                                         ret = -ENOMEM;
 128                                         goto write_unlock;
 129                                 }
 130                                 cur->num_buckets = netdev_flow_limit_table_len;
 131                                 rcu_assign_pointer(sd->flow_limit, cur);
 132                         }
 133                 }
 134 write_unlock:
 135                 mutex_unlock(&flow_limit_update_mutex);
 136         } else {
 137                 char kbuf[128];
 138
 139                 if (*ppos || !*lenp) {
 140                         *lenp = 0;
 141                         goto done;
 142                 }
 143
 144                 cpumask_clear(mask);
 145                 rcu_read_lock();
 146                 for_each_possible_cpu(i) {
 147                         sd = &per_cpu(softnet_data, i);
 148                         if (rcu_dereference(sd->flow_limit))
 149                                 cpumask_set_cpu(i, mask);
 150                 }
 151                 rcu_read_unlock();
 152
 153                 len = min(sizeof(kbuf) - 1, *lenp);
 154                 len = cpumask_scnprintf(kbuf, len, mask);
 155                 if (!len) {
 156                         *lenp = 0;
 157                         goto done;
 158                 }
 159                 if (len < *lenp)
 160                         kbuf[len++] = '\n';
 161                 if (copy_to_user(buffer, kbuf, len)) {
 162                         ret = -EFAULT;
 163                         goto done;
 164                 }
 165                 *lenp = len;
 166                 *ppos += len;
 167         }
 168
 169 done:
 170         free_cpumask_var(mask);
 171         return ret;
 172 }
 173
 174 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
 175                                        void __user *buffer, size_t *lenp,
 176                                        loff_t *ppos)
 177 {
 178         unsigned int old, *ptr;
 179         int ret;
 180
 181         mutex_lock(&flow_limit_update_mutex);
 182
 183         ptr = table->data;
 184         old = *ptr;
 185         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 186         if (!ret && write && !is_power_of_2(*ptr)) {
 187                 *ptr = old;
 188                 ret = -EINVAL;
 189         }
 190
 191         mutex_unlock(&flow_limit_update_mutex);
 192         return ret;
 193 }
 194 #endif /* CONFIG_NET_FLOW_LIMIT */
 195
 196 static struct ctl_table net_core_table[] = {
 197 #ifdef CONFIG_NET
 198         {
 199                 .procname       = "wmem_max",
 200                 .data           = &sysctl_wmem_max,
 201                 .maxlen         = sizeof(int),
 202                 .mode           = 0644,
 203                 .proc_handler   = proc_dointvec_minmax,
 204                 .extra1         = &one,
 205         },
 206         {
 207                 .procname       = "rmem_max",
 208                 .data           = &sysctl_rmem_max,
 209                 .maxlen         = sizeof(int),
 210                 .mode           = 0644,
 211                 .proc_handler   = proc_dointvec_minmax,
 212                 .extra1         = &one,
 213         },
 214         {
 215                 .procname       = "wmem_default",
 216                 .data           = &sysctl_wmem_default,
 217                 .maxlen         = sizeof(int),
 218                 .mode           = 0644,
 219                 .proc_handler   = proc_dointvec_minmax,
 220                 .extra1         = &one,
 221         },
 222         {
 223                 .procname       = "rmem_default",
 224                 .data           = &sysctl_rmem_default,
 225                 .maxlen         = sizeof(int),
 226                 .mode           = 0644,
 227                 .proc_handler   = proc_dointvec_minmax,
 228                 .extra1         = &one,
 229         },
 230         {
 231                 .procname       = "dev_weight",
 232                 .data           = &weight_p,
 233                 .maxlen         = sizeof(int),
 234                 .mode           = 0644,
 235                 .proc_handler   = proc_dointvec
 236         },
 237         {
 238                 .procname       = "netdev_max_backlog",
 239                 .data           = &netdev_max_backlog,
 240                 .maxlen         = sizeof(int),
 241                 .mode           = 0644,
 242                 .proc_handler   = proc_dointvec
 243         },
 244 #ifdef CONFIG_BPF_JIT
 245         {
 246                 .procname       = "bpf_jit_enable",
 247                 .data           = &bpf_jit_enable,
 248                 .maxlen         = sizeof(int),
 249                 .mode           = 0644,
 250                 .proc_handler   = proc_dointvec
 251         },
 252 #endif
 253         {
 254                 .procname       = "netdev_tstamp_prequeue",
 255                 .data           = &netdev_tstamp_prequeue,
 256                 .maxlen         = sizeof(int),
 257                 .mode           = 0644,
 258                 .proc_handler   = proc_dointvec
 259         },
 260         {
 261                 .procname       = "message_cost",
 262                 .data           = &net_ratelimit_state.interval,
 263                 .maxlen         = sizeof(int),
 264                 .mode           = 0644,
 265                 .proc_handler   = proc_dointvec_jiffies,
 266         },
 267         {
 268                 .procname       = "message_burst",
 269                 .data           = &net_ratelimit_state.burst,
 270                 .maxlen         = sizeof(int),
 271                 .mode           = 0644,
 272                 .proc_handler   = proc_dointvec,
 273         },
 274         {
 275                 .procname       = "optmem_max",
 276                 .data           = &sysctl_optmem_max,
 277                 .maxlen         = sizeof(int),
 278                 .mode           = 0644,
 279                 .proc_handler   = proc_dointvec
 280         },
 281 #ifdef CONFIG_RPS
 282         {
 283                 .procname       = "rps_sock_flow_entries",
 284                 .maxlen         = sizeof(int),
 285                 .mode           = 0644,
 286                 .proc_handler   = rps_sock_flow_sysctl
 287         },
 288 #endif
 289 #ifdef CONFIG_NET_FLOW_LIMIT
 290         {
 291                 .procname       = "flow_limit_cpu_bitmap",
 292                 .mode           = 0644,
 293                 .proc_handler   = flow_limit_cpu_sysctl
 294         },
 295         {
 296                 .procname       = "flow_limit_table_len",
 297                 .data           = &netdev_flow_limit_table_len,
 298                 .maxlen         = sizeof(int),
 299                 .mode           = 0644,
 300                 .proc_handler   = flow_limit_table_len_sysctl
 301         },
 302 #endif /* CONFIG_NET_FLOW_LIMIT */
 303 #ifdef CONFIG_NET_RX_BUSY_POLL
 304         {
 305                 .procname       = "busy_poll",
 306                 .data           = &sysctl_net_busy_poll,
 307                 .maxlen         = sizeof(unsigned int),
 308                 .mode           = 0644,
 309                 .proc_handler   = proc_dointvec
 310         },
 311         {
 312                 .procname       = "busy_read",
 313                 .data           = &sysctl_net_busy_read,
 314                 .maxlen         = sizeof(unsigned int),
 315                 .mode           = 0644,
 316                 .proc_handler   = proc_dointvec
 317         },
 318 #
 319 #endif
 320 #endif /* CONFIG_NET */
 321         {
 322                 .procname       = "netdev_budget",
 323                 .data           = &netdev_budget,
 324                 .maxlen         = sizeof(int),
 325                 .mode           = 0644,
 326                 .proc_handler   = proc_dointvec
 327         },
 328         {
 329                 .procname       = "warnings",
 330                 .data           = &net_msg_warn,
 331                 .maxlen         = sizeof(int),
 332                 .mode           = 0644,
 333                 .proc_handler   = proc_dointvec
 334         },
 335         { }
 336 };
 337
 338 static struct ctl_table netns_core_table[] = {
 339         {
 340                 .procname       = "somaxconn",
 341                 .data           = &init_net.core.sysctl_somaxconn,
 342                 .maxlen         = sizeof(int),
 343                 .mode           = 0644,
 344                 .extra1         = &zero,
 345                 .extra2         = &ushort_max,
 346                 .proc_handler   = proc_dointvec_minmax
 347         },
 348         { }
 349 };
 350
 351 static __net_init int sysctl_core_net_init(struct net *net)
 352 {
 353         struct ctl_table *tbl;
 354
 355         net->core.sysctl_somaxconn = SOMAXCONN;
 356
 357         tbl = netns_core_table;
 358         if (!net_eq(net, &init_net)) {
 359                 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
 360                 if (tbl == NULL)
 361                         goto err_dup;
 362
 363                 tbl[0].data = &net->core.sysctl_somaxconn;
 364
 365                 /* Don't export any sysctls to unprivileged users */
 366                 if (net->user_ns != &init_user_ns) {
 367                         tbl[0].procname = NULL;
 368                 }
 369         }
 370
 371         net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
 372         if (net->core.sysctl_hdr == NULL)
 373                 goto err_reg;
 374
 375         return 0;
 376
 377 err_reg:
 378         if (tbl != netns_core_table)
 379                 kfree(tbl);
 380 err_dup:
 381         return -ENOMEM;
 382 }
 383
 384 static __net_exit void sysctl_core_net_exit(struct net *net)
 385 {
 386         struct ctl_table *tbl;
 387
 388         tbl = net->core.sysctl_hdr->ctl_table_arg;
 389         unregister_net_sysctl_table(net->core.sysctl_hdr);
 390         BUG_ON(tbl == netns_core_table);
 391         kfree(tbl);
 392 }
 393
 394 static __net_initdata struct pernet_operations sysctl_core_ops = {
 395         .init = sysctl_core_net_init,
 396         .exit = sysctl_core_net_exit,
 397 };
 398
 399 static __init int sysctl_core_init(void)
 400 {
 401         register_net_sysctl(&init_net, "net/core", net_core_table);
 402         return register_pernet_subsys(&sysctl_core_ops);
 403 }
 404
 405 fs_initcall(sysctl_core_init);