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