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
  23 static int one = 1;
  24
  25 #ifdef CONFIG_RPS
  26 static int rps_sock_flow_sysctl(ctl_table *table, int write,
  27                                 void __user *buffer, size_t *lenp, loff_t *ppos)
  28 {
  29         unsigned int orig_size, size;
  30         int ret, i;
  31         ctl_table tmp = {
  32                 .data = &size,
  33                 .maxlen = sizeof(size),
  34                 .mode = table->mode
  35         };
  36         struct rps_sock_flow_table *orig_sock_table, *sock_table;
  37         static DEFINE_MUTEX(sock_flow_mutex);
  38
  39         mutex_lock(&sock_flow_mutex);
  40
  41         orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
  42                                         lockdep_is_held(&sock_flow_mutex));
  43         size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
  44
  45         ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
  46
  47         if (write) {
  48                 if (size) {
  49                         if (size > 1<<30) {
  50                                 /* Enforce limit to prevent overflow */
  51                                 mutex_unlock(&sock_flow_mutex);
  52                                 return -EINVAL;
  53                         }
  54                         size = roundup_pow_of_two(size);
  55                         if (size != orig_size) {
  56                                 sock_table =
  57                                     vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
  58                                 if (!sock_table) {
  59                                         mutex_unlock(&sock_flow_mutex);
  60                                         return -ENOMEM;
  61                                 }
  62
  63                                 sock_table->mask = size - 1;
  64                         } else
  65                                 sock_table = orig_sock_table;
  66
  67                         for (i = 0; i < size; i++)
  68                                 sock_table->ents[i] = RPS_NO_CPU;
  69                 } else
  70                         sock_table = NULL;
  71
  72                 if (sock_table != orig_sock_table) {
  73                         rcu_assign_pointer(rps_sock_flow_table, sock_table);
  74                         if (sock_table)
  75                                 static_key_slow_inc(&rps_needed);
  76                         if (orig_sock_table) {
  77                                 static_key_slow_dec(&rps_needed);
  78                                 synchronize_rcu();
  79                                 vfree(orig_sock_table);
  80                         }
  81                 }
  82         }
  83
  84         mutex_unlock(&sock_flow_mutex);
  85
  86         return ret;
  87 }
  88 #endif /* CONFIG_RPS */
  89
  90 static struct ctl_table net_core_table[] = {
  91 #ifdef CONFIG_NET
  92         {
  93                 .procname       = "wmem_max",
  94                 .data           = &sysctl_wmem_max,
  95                 .maxlen         = sizeof(int),
  96                 .mode           = 0644,
  97                 .proc_handler   = proc_dointvec_minmax,
  98                 .extra1         = &one,
  99         },
 100         {
 101                 .procname       = "rmem_max",
 102                 .data           = &sysctl_rmem_max,
 103                 .maxlen         = sizeof(int),
 104                 .mode           = 0644,
 105                 .proc_handler   = proc_dointvec_minmax,
 106                 .extra1         = &one,
 107         },
 108         {
 109                 .procname       = "wmem_default",
 110                 .data           = &sysctl_wmem_default,
 111                 .maxlen         = sizeof(int),
 112                 .mode           = 0644,
 113                 .proc_handler   = proc_dointvec_minmax,
 114                 .extra1         = &one,
 115         },
 116         {
 117                 .procname       = "rmem_default",
 118                 .data           = &sysctl_rmem_default,
 119                 .maxlen         = sizeof(int),
 120                 .mode           = 0644,
 121                 .proc_handler   = proc_dointvec_minmax,
 122                 .extra1         = &one,
 123         },
 124         {
 125                 .procname       = "dev_weight",
 126                 .data           = &weight_p,
 127                 .maxlen         = sizeof(int),
 128                 .mode           = 0644,
 129                 .proc_handler   = proc_dointvec
 130         },
 131         {
 132                 .procname       = "netdev_max_backlog",
 133                 .data           = &netdev_max_backlog,
 134                 .maxlen         = sizeof(int),
 135                 .mode           = 0644,
 136                 .proc_handler   = proc_dointvec
 137         },
 138 #ifdef CONFIG_BPF_JIT
 139         {
 140                 .procname       = "bpf_jit_enable",
 141                 .data           = &bpf_jit_enable,
 142                 .maxlen         = sizeof(int),
 143                 .mode           = 0644,
 144                 .proc_handler   = proc_dointvec
 145         },
 146 #endif
 147         {
 148                 .procname       = "netdev_tstamp_prequeue",
 149                 .data           = &netdev_tstamp_prequeue,
 150                 .maxlen         = sizeof(int),
 151                 .mode           = 0644,
 152                 .proc_handler   = proc_dointvec
 153         },
 154         {
 155                 .procname       = "message_cost",
 156                 .data           = &net_ratelimit_state.interval,
 157                 .maxlen         = sizeof(int),
 158                 .mode           = 0644,
 159                 .proc_handler   = proc_dointvec_jiffies,
 160         },
 161         {
 162                 .procname       = "message_burst",
 163                 .data           = &net_ratelimit_state.burst,
 164                 .maxlen         = sizeof(int),
 165                 .mode           = 0644,
 166                 .proc_handler   = proc_dointvec,
 167         },
 168         {
 169                 .procname       = "optmem_max",
 170                 .data           = &sysctl_optmem_max,
 171                 .maxlen         = sizeof(int),
 172                 .mode           = 0644,
 173                 .proc_handler   = proc_dointvec
 174         },
 175 #ifdef CONFIG_RPS
 176         {
 177                 .procname       = "rps_sock_flow_entries",
 178                 .maxlen         = sizeof(int),
 179                 .mode           = 0644,
 180                 .proc_handler   = rps_sock_flow_sysctl
 181         },
 182 #endif
 183 #endif /* CONFIG_NET */
 184         {
 185                 .procname       = "netdev_budget",
 186                 .data           = &netdev_budget,
 187                 .maxlen         = sizeof(int),
 188                 .mode           = 0644,
 189                 .proc_handler   = proc_dointvec
 190         },
 191         {
 192                 .procname       = "warnings",
 193                 .data           = &net_msg_warn,
 194                 .maxlen         = sizeof(int),
 195                 .mode           = 0644,
 196                 .proc_handler   = proc_dointvec
 197         },
 198         { }
 199 };
 200
 201 static struct ctl_table netns_core_table[] = {
 202         {
 203                 .procname       = "somaxconn",
 204                 .data           = &init_net.core.sysctl_somaxconn,
 205                 .maxlen         = sizeof(int),
 206                 .mode           = 0644,
 207                 .proc_handler   = proc_dointvec
 208         },
 209         { }
 210 };
 211
 212 static __net_init int sysctl_core_net_init(struct net *net)
 213 {
 214         struct ctl_table *tbl;
 215
 216         net->core.sysctl_somaxconn = SOMAXCONN;
 217
 218         tbl = netns_core_table;
 219         if (!net_eq(net, &init_net)) {
 220                 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
 221                 if (tbl == NULL)
 222                         goto err_dup;
 223
 224                 tbl[0].data = &net->core.sysctl_somaxconn;
 225
 226                 /* Don't export any sysctls to unprivileged users */
 227                 if (net->user_ns != &init_user_ns) {
 228                         tbl[0].procname = NULL;
 229                 }
 230         }
 231
 232         net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
 233         if (net->core.sysctl_hdr == NULL)
 234                 goto err_reg;
 235
 236         return 0;
 237
 238 err_reg:
 239         if (tbl != netns_core_table)
 240                 kfree(tbl);
 241 err_dup:
 242         return -ENOMEM;
 243 }
 244
 245 static __net_exit void sysctl_core_net_exit(struct net *net)
 246 {
 247         struct ctl_table *tbl;
 248
 249         tbl = net->core.sysctl_hdr->ctl_table_arg;
 250         unregister_net_sysctl_table(net->core.sysctl_hdr);
 251         BUG_ON(tbl == netns_core_table);
 252         kfree(tbl);
 253 }
 254
 255 static __net_initdata struct pernet_operations sysctl_core_ops = {
 256         .init = sysctl_core_net_init,
 257         .exit = sysctl_core_net_exit,
 258 };
 259
 260 static __init int sysctl_core_init(void)
 261 {
 262         register_net_sysctl(&init_net, "net/core", net_core_table);
 263         return register_pernet_subsys(&sysctl_core_ops);
 264 }
 265
 266 fs_initcall(sysctl_core_init);