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
  18 #include <net/ip.h>
  19 #include <net/sock.h>
  20 #include <net/net_ratelimit.h>
  21
  22 #ifdef CONFIG_RPS
  23 static int rps_sock_flow_sysctl(ctl_table *table, int write,
  24                                 void __user *buffer, size_t *lenp, loff_t *ppos)
  25 {
  26         unsigned int orig_size, size;
  27         int ret, i;
  28         ctl_table tmp = {
  29                 .data = &size,
  30                 .maxlen = sizeof(size),
  31                 .mode = table->mode
  32         };
  33         struct rps_sock_flow_table *orig_sock_table, *sock_table;
  34         static DEFINE_MUTEX(sock_flow_mutex);
  35
  36         mutex_lock(&sock_flow_mutex);
  37
  38         orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
  39                                         lockdep_is_held(&sock_flow_mutex));
  40         size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
  41
  42         ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
  43
  44         if (write) {
  45                 if (size) {
  46                         if (size > 1<<30) {
  47                                 /* Enforce limit to prevent overflow */
  48                                 mutex_unlock(&sock_flow_mutex);
  49                                 return -EINVAL;
  50                         }
  51                         size = roundup_pow_of_two(size);
  52                         if (size != orig_size) {
  53                                 sock_table =
  54                                     vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
  55                                 if (!sock_table) {
  56                                         mutex_unlock(&sock_flow_mutex);
  57                                         return -ENOMEM;
  58                                 }
  59
  60                                 sock_table->mask = size - 1;
  61                         } else
  62                                 sock_table = orig_sock_table;
  63
  64                         for (i = 0; i < size; i++)
  65                                 sock_table->ents[i] = RPS_NO_CPU;
  66                 } else
  67                         sock_table = NULL;
  68
  69                 if (sock_table != orig_sock_table) {
  70                         rcu_assign_pointer(rps_sock_flow_table, sock_table);
  71                         if (sock_table)
  72                                 static_key_slow_inc(&rps_needed);
  73                         if (orig_sock_table) {
  74                                 static_key_slow_dec(&rps_needed);
  75                                 synchronize_rcu();
  76                                 vfree(orig_sock_table);
  77                         }
  78                 }
  79         }
  80
  81         mutex_unlock(&sock_flow_mutex);
  82
  83         return ret;
  84 }
  85 #endif /* CONFIG_RPS */
  86
  87 static struct ctl_table net_core_table[] = {
  88 #ifdef CONFIG_NET
  89         {
  90                 .procname       = "wmem_max",
  91                 .data           = &sysctl_wmem_max,
  92                 .maxlen         = sizeof(int),
  93                 .mode           = 0644,
  94                 .proc_handler   = proc_dointvec
  95         },
  96         {
  97                 .procname       = "rmem_max",
  98                 .data           = &sysctl_rmem_max,
  99                 .maxlen         = sizeof(int),
 100                 .mode           = 0644,
 101                 .proc_handler   = proc_dointvec
 102         },
 103         {
 104                 .procname       = "wmem_default",
 105                 .data           = &sysctl_wmem_default,
 106                 .maxlen         = sizeof(int),
 107                 .mode           = 0644,
 108                 .proc_handler   = proc_dointvec
 109         },
 110         {
 111                 .procname       = "rmem_default",
 112                 .data           = &sysctl_rmem_default,
 113                 .maxlen         = sizeof(int),
 114                 .mode           = 0644,
 115                 .proc_handler   = proc_dointvec
 116         },
 117         {
 118                 .procname       = "dev_weight",
 119                 .data           = &weight_p,
 120                 .maxlen         = sizeof(int),
 121                 .mode           = 0644,
 122                 .proc_handler   = proc_dointvec
 123         },
 124         {
 125                 .procname       = "netdev_max_backlog",
 126                 .data           = &netdev_max_backlog,
 127                 .maxlen         = sizeof(int),
 128                 .mode           = 0644,
 129                 .proc_handler   = proc_dointvec
 130         },
 131 #ifdef CONFIG_BPF_JIT
 132         {
 133                 .procname       = "bpf_jit_enable",
 134                 .data           = &bpf_jit_enable,
 135                 .maxlen         = sizeof(int),
 136                 .mode           = 0644,
 137                 .proc_handler   = proc_dointvec
 138         },
 139 #endif
 140         {
 141                 .procname       = "netdev_tstamp_prequeue",
 142                 .data           = &netdev_tstamp_prequeue,
 143                 .maxlen         = sizeof(int),
 144                 .mode           = 0644,
 145                 .proc_handler   = proc_dointvec
 146         },
 147         {
 148                 .procname       = "message_cost",
 149                 .data           = &net_ratelimit_state.interval,
 150                 .maxlen         = sizeof(int),
 151                 .mode           = 0644,
 152                 .proc_handler   = proc_dointvec_jiffies,
 153         },
 154         {
 155                 .procname       = "message_burst",
 156                 .data           = &net_ratelimit_state.burst,
 157                 .maxlen         = sizeof(int),
 158                 .mode           = 0644,
 159                 .proc_handler   = proc_dointvec,
 160         },
 161         {
 162                 .procname       = "optmem_max",
 163                 .data           = &sysctl_optmem_max,
 164                 .maxlen         = sizeof(int),
 165                 .mode           = 0644,
 166                 .proc_handler   = proc_dointvec
 167         },
 168 #ifdef CONFIG_RPS
 169         {
 170                 .procname       = "rps_sock_flow_entries",
 171                 .maxlen         = sizeof(int),
 172                 .mode           = 0644,
 173                 .proc_handler   = rps_sock_flow_sysctl
 174         },
 175 #endif
 176 #endif /* CONFIG_NET */
 177         {
 178                 .procname       = "netdev_budget",
 179                 .data           = &netdev_budget,
 180                 .maxlen         = sizeof(int),
 181                 .mode           = 0644,
 182                 .proc_handler   = proc_dointvec
 183         },
 184         {
 185                 .procname       = "warnings",
 186                 .data           = &net_msg_warn,
 187                 .maxlen         = sizeof(int),
 188                 .mode           = 0644,
 189                 .proc_handler   = proc_dointvec
 190         },
 191         { }
 192 };
 193
 194 static struct ctl_table netns_core_table[] = {
 195         {
 196                 .procname       = "somaxconn",
 197                 .data           = &init_net.core.sysctl_somaxconn,
 198                 .maxlen         = sizeof(int),
 199                 .mode           = 0644,
 200                 .proc_handler   = proc_dointvec
 201         },
 202         { }
 203 };
 204
 205 __net_initdata struct ctl_path net_core_path[] = {
 206         { .procname = "net", },
 207         { .procname = "core", },
 208         { },
 209 };
 210
 211 static __net_init int sysctl_core_net_init(struct net *net)
 212 {
 213         struct ctl_table *tbl;
 214
 215         net->core.sysctl_somaxconn = SOMAXCONN;
 216
 217         tbl = netns_core_table;
 218         if (!net_eq(net, &init_net)) {
 219                 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
 220                 if (tbl == NULL)
 221                         goto err_dup;
 222
 223                 tbl[0].data = &net->core.sysctl_somaxconn;
 224         }
 225
 226         net->core.sysctl_hdr = register_net_sysctl_table(net,
 227                         net_core_path, tbl);
 228         if (net->core.sysctl_hdr == NULL)
 229                 goto err_reg;
 230
 231         return 0;
 232
 233 err_reg:
 234         if (tbl != netns_core_table)
 235                 kfree(tbl);
 236 err_dup:
 237         return -ENOMEM;
 238 }
 239
 240 static __net_exit void sysctl_core_net_exit(struct net *net)
 241 {
 242         struct ctl_table *tbl;
 243
 244         tbl = net->core.sysctl_hdr->ctl_table_arg;
 245         unregister_net_sysctl_table(net->core.sysctl_hdr);
 246         BUG_ON(tbl == netns_core_table);
 247         kfree(tbl);
 248 }
 249
 250 static __net_initdata struct pernet_operations sysctl_core_ops = {
 251         .init = sysctl_core_net_init,
 252         .exit = sysctl_core_net_exit,
 253 };
 254
 255 static __init int sysctl_core_init(void)
 256 {
 257         static struct ctl_table empty[1];
 258
 259         kmemleak_not_leak(register_sysctl_paths(net_core_path, empty));
 260         register_net_sysctl_rotable(net_core_path, net_core_table);
 261         return register_pernet_subsys(&sysctl_core_ops);
 262 }
 263
 264 fs_initcall(sysctl_core_init);