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/ll_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                 if (*ppos || !*lenp) {
 136                         *lenp = 0;
 137                         goto done;
 138                 }
 139
 140                 cpumask_clear(mask);
 141                 rcu_read_lock();
 142                 for_each_possible_cpu(i) {
 143                         sd = &per_cpu(softnet_data, i);
 144                         if (rcu_dereference(sd->flow_limit))
 145                                 cpumask_set_cpu(i, mask);
 146                 }
 147                 rcu_read_unlock();
 148
 149                 len = cpumask_scnprintf(buffer, *lenp, mask);
 150                 *lenp = len + 1;
 151                 *ppos += len + 1;
 152         }
 153
 154 done:
 155         free_cpumask_var(mask);
 156         return ret;
 157 }
 158
 159 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
 160                                        void __user *buffer, size_t *lenp,
 161                                        loff_t *ppos)
 162 {
 163         unsigned int old, *ptr;
 164         int ret;
 165
 166         mutex_lock(&flow_limit_update_mutex);
 167
 168         ptr = table->data;
 169         old = *ptr;
 170         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 171         if (!ret && write && !is_power_of_2(*ptr)) {
 172                 *ptr = old;
 173                 ret = -EINVAL;
 174         }
 175
 176         mutex_unlock(&flow_limit_update_mutex);
 177         return ret;
 178 }
 179 #endif /* CONFIG_NET_FLOW_LIMIT */
 180
 181 static struct ctl_table net_core_table[] = {
 182 #ifdef CONFIG_NET
 183         {
 184                 .procname       = "wmem_max",
 185                 .data           = &sysctl_wmem_max,
 186                 .maxlen         = sizeof(int),
 187                 .mode           = 0644,
 188                 .proc_handler   = proc_dointvec_minmax,
 189                 .extra1         = &one,
 190         },
 191         {
 192                 .procname       = "rmem_max",
 193                 .data           = &sysctl_rmem_max,
 194                 .maxlen         = sizeof(int),
 195                 .mode           = 0644,
 196                 .proc_handler   = proc_dointvec_minmax,
 197                 .extra1         = &one,
 198         },
 199         {
 200                 .procname       = "wmem_default",
 201                 .data           = &sysctl_wmem_default,
 202                 .maxlen         = sizeof(int),
 203                 .mode           = 0644,
 204                 .proc_handler   = proc_dointvec_minmax,
 205                 .extra1         = &one,
 206         },
 207         {
 208                 .procname       = "rmem_default",
 209                 .data           = &sysctl_rmem_default,
 210                 .maxlen         = sizeof(int),
 211                 .mode           = 0644,
 212                 .proc_handler   = proc_dointvec_minmax,
 213                 .extra1         = &one,
 214         },
 215         {
 216                 .procname       = "dev_weight",
 217                 .data           = &weight_p,
 218                 .maxlen         = sizeof(int),
 219                 .mode           = 0644,
 220                 .proc_handler   = proc_dointvec
 221         },
 222         {
 223                 .procname       = "netdev_max_backlog",
 224                 .data           = &netdev_max_backlog,
 225                 .maxlen         = sizeof(int),
 226                 .mode           = 0644,
 227                 .proc_handler   = proc_dointvec
 228         },
 229 #ifdef CONFIG_BPF_JIT
 230         {
 231                 .procname       = "bpf_jit_enable",
 232                 .data           = &bpf_jit_enable,
 233                 .maxlen         = sizeof(int),
 234                 .mode           = 0644,
 235                 .proc_handler   = proc_dointvec
 236         },
 237 #endif
 238         {
 239                 .procname       = "netdev_tstamp_prequeue",
 240                 .data           = &netdev_tstamp_prequeue,
 241                 .maxlen         = sizeof(int),
 242                 .mode           = 0644,
 243                 .proc_handler   = proc_dointvec
 244         },
 245         {
 246                 .procname       = "message_cost",
 247                 .data           = &net_ratelimit_state.interval,
 248                 .maxlen         = sizeof(int),
 249                 .mode           = 0644,
 250                 .proc_handler   = proc_dointvec_jiffies,
 251         },
 252         {
 253                 .procname       = "message_burst",
 254                 .data           = &net_ratelimit_state.burst,
 255                 .maxlen         = sizeof(int),
 256                 .mode           = 0644,
 257                 .proc_handler   = proc_dointvec,
 258         },
 259         {
 260                 .procname       = "optmem_max",
 261                 .data           = &sysctl_optmem_max,
 262                 .maxlen         = sizeof(int),
 263                 .mode           = 0644,
 264                 .proc_handler   = proc_dointvec
 265         },
 266 #ifdef CONFIG_RPS
 267         {
 268                 .procname       = "rps_sock_flow_entries",
 269                 .maxlen         = sizeof(int),
 270                 .mode           = 0644,
 271                 .proc_handler   = rps_sock_flow_sysctl
 272         },
 273 #endif
 274 #ifdef CONFIG_NET_FLOW_LIMIT
 275         {
 276                 .procname       = "flow_limit_cpu_bitmap",
 277                 .mode           = 0644,
 278                 .proc_handler   = flow_limit_cpu_sysctl
 279         },
 280         {
 281                 .procname       = "flow_limit_table_len",
 282                 .data           = &netdev_flow_limit_table_len,
 283                 .maxlen         = sizeof(int),
 284                 .mode           = 0644,
 285                 .proc_handler   = flow_limit_table_len_sysctl
 286         },
 287 #endif /* CONFIG_NET_FLOW_LIMIT */
 288 #ifdef CONFIG_NET_LL_RX_POLL
 289         {
 290                 .procname       = "low_latency_poll",
 291                 .data           = &sysctl_net_ll_poll,
 292                 .maxlen         = sizeof(unsigned long),
 293                 .mode           = 0644,
 294                 .proc_handler   = proc_doulongvec_minmax
 295         },
 296 #endif
 297 #endif /* CONFIG_NET */
 298         {
 299                 .procname       = "netdev_budget",
 300                 .data           = &netdev_budget,
 301                 .maxlen         = sizeof(int),
 302                 .mode           = 0644,
 303                 .proc_handler   = proc_dointvec
 304         },
 305         {
 306                 .procname       = "warnings",
 307                 .data           = &net_msg_warn,
 308                 .maxlen         = sizeof(int),
 309                 .mode           = 0644,
 310                 .proc_handler   = proc_dointvec
 311         },
 312         { }
 313 };
 314
 315 static struct ctl_table netns_core_table[] = {
 316         {
 317                 .procname       = "somaxconn",
 318                 .data           = &init_net.core.sysctl_somaxconn,
 319                 .maxlen         = sizeof(int),
 320                 .mode           = 0644,
 321                 .proc_handler   = proc_dointvec
 322         },
 323         { }
 324 };
 325
 326 static __net_init int sysctl_core_net_init(struct net *net)
 327 {
 328         struct ctl_table *tbl;
 329
 330         net->core.sysctl_somaxconn = SOMAXCONN;
 331
 332         tbl = netns_core_table;
 333         if (!net_eq(net, &init_net)) {
 334                 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
 335                 if (tbl == NULL)
 336                         goto err_dup;
 337
 338                 tbl[0].data = &net->core.sysctl_somaxconn;
 339
 340                 /* Don't export any sysctls to unprivileged users */
 341                 if (net->user_ns != &init_user_ns) {
 342                         tbl[0].procname = NULL;
 343                 }
 344         }
 345
 346         net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
 347         if (net->core.sysctl_hdr == NULL)
 348                 goto err_reg;
 349
 350         return 0;
 351
 352 err_reg:
 353         if (tbl != netns_core_table)
 354                 kfree(tbl);
 355 err_dup:
 356         return -ENOMEM;
 357 }
 358
 359 static __net_exit void sysctl_core_net_exit(struct net *net)
 360 {
 361         struct ctl_table *tbl;
 362
 363         tbl = net->core.sysctl_hdr->ctl_table_arg;
 364         unregister_net_sysctl_table(net->core.sysctl_hdr);
 365         BUG_ON(tbl == netns_core_table);
 366         kfree(tbl);
 367 }
 368
 369 static __net_initdata struct pernet_operations sysctl_core_ops = {
 370         .init = sysctl_core_net_init,
 371         .exit = sysctl_core_net_exit,
 372 };
 373
 374 static __init int sysctl_core_init(void)
 375 {
 376         register_net_sysctl(&init_net, "net/core", net_core_table);
 377         return register_pernet_subsys(&sysctl_core_ops);
 378 }
 379
 380 fs_initcall(sysctl_core_init);