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